EDI TO R I A L
Help wanted, scientists need apply
T
he world is continuously being transformed by
science and technology (S&T), but to deliver eq-
uitable benefits to the public, scientists must be
embedded in influential sectors of society—policy,
diplomacy, journalism, law, business, education,
and more. This means injecting PhD-level experts
at every stage of research and development, from
ideation, investigation, and investment to manufacture,
deployment, regulation, and after-market evaluation.
At first glance, it seems unlikely that scientists would
be welcome outside of science itself. Trust in science was
severely politicized and eroded during the pandemic, so
why would scientists be invited into nonscience spheres
of influence? Moreover, familiarizing graduate students
with “nonresearch” pursuits and
demonstrating how they can apply
their skills to ensure that S&T ad-
dresses societal challenges are not
standard in PhD science curricula.
“…scientists must
Although trust in science has
declined in the United States, trust
be embedded in
in scientists has not. A S&T Ac-
tion Committee survey comprising
influential sectors
48% Republicans, 38% Democrats,
and 13% Independent Americans
of society…”
showed that nearly 80% of respon-
dents across political affiliations
found politicization and distrust of science to be con-
cerning, and 80% wanted scientists to contribute more
to shaping public policy. In fact, The Pew Research Cen-
ter found that public confidence in scientists increased
in the US, from 76 to 87%, during the pandemic. Why?
Perhaps, as O’Connor and Weatherall suggest in The
Misinformation Age, it is because the public accords
scientists broad trust as experts in evidence-based de-
liberations and finding solutions to complex problems.
Scientists generally know little about the nonscience
parts of governance, communication, and the economic
and procedural systems that affect whether and how
science discoveries advance through development and
into the public domain. Here, academia, funding agen-
cies, and professional societies and foundations bear
crucial responsibility. Organizations that fund research
and training should require that universities provide
PHOTO: UNIVERSITY OF CALIFORNIA, SAN FRANCISCO
substantial career exploration courses to graduate
trainees. Optional 1-day anecdotal presentations are
insufficient, and deferring such offerings to the post-
doc stage is too late. In parallel, professional societies
and advocacy organizations should offer internships or
fellowships that place newly minted science PhDs into
influential outside-of-science environments.
SCIENCE science.org
Such programs do work. For example, in 2014, the
US National Institutes of Health awarded pilot grants
to 17 universities to devise training programs aimed
at furnishing graduate trainees with working knowl-
edge of careers in which their scientific expertise
would bring great value. The University of California,
San Francisco, created Motivating Informed Decisions,
three 4-hour workshops followed by a dozen biweekly Keith R. Yamamoto
“peer mentoring” meetings, in which each member of
is a professor in
six- to eight-person teams defines and progressively
the Department
interrogates potential career goals. Together with the
of Cellular
university’s Internships for Career Exploration pro-
gram, which supports 10-week internships in diverse and Molecular
career settings, graduate students have reported confi- Pharmacology
dence in embarking on diverse ca- at the University
reer paths. Some internship hosts of California, San
were so impressed by the partici- Francisco (UCSF),
pating students that they offered
their interns jobs.
and vice chancellor
for Science Policy
Another example is the 50-year-
old Science and Technology Policy
and Strategy and
director of Precision
Fellows program of the American
Association for the Advancement
Medicine, UCSF,
San Franciso, CA,
of Science (AAAS, the publisher
of Science). Members of Congress
USA. He is also the
president of AAAS,
and heads of federal agencies bol- Washington, DC,
ster their staffs with these PhD fel- USA. yamamoto@
lows, and the outcomes speak for themselves: Among ucsf.edu
4000 graduates of this program, 50% remained in pol-
icy, 25% returned to science careers, and 25% moved
to other fields. Several states have now established
similar programs to populate state government offices
with scientists.
Clearly, the need for supporting career exploration
is not limited to the US. Royal Society programs in
the United Kingdom, for example, provide senior re-
searchers and their graduate students and postdocs
opportunities to learn about, and contribute to, law
and the judiciary, business, management, finance, in-
surance, and other disciplines.
Globally, we’re blessed with a robust cohort of tal-
ented S&T doctoral trainees who go on to make break-
through contributions in academia or the private
sector. But in today’s complex world, we also need
scientists resolving problems and rendering decisions
across society to ensure that S&T adds value to lives
and livelihoods, sustains a safe environment, and con-
tributes to security and economic well-being for all. We
must illuminate those career pathways for all trainees.
–Keith R. Yamamoto
10.1126/science.ado4539
9 FEBRUARY 2024 • VOL 383 ISSUE 6683 57 1
 NEWS 142%
Estimated increase by 2050 of annual new cancer cases in low-
income countries, compared with 2022’s level, according to
the World Health Organization. Causes include increased tobacco
and alcohol use, obesity, and air pollution.

pollsters excluded scientists in 60 countries

IN BRIEF Edited by Jeffrey Brainard
and territories—including China, where
evidence indicates the pandemic started—
CLIMATE POLICY deemed “not free.”) On average, the
168 respondents, most of them virologists

EU mulls bigger carbon emissions cut and epidemiologists, assigned a 77% likeli-
hood to the natural origin scenario and a
21% likelihood to a lab accident. Still, one

T
he European Union should aim to reduce net carbon emis- in five said there was at least a 50% chance
sions by 90% by 2040, one of the most aggressive targets of that COVID-19 resulted from a lab mishap.
any bloc or nation, the European Commission proposed this The results quickly triggered a fierce
week. The EU has already committed to cuts of 55% by 2030 debate on social media about the study’s
methodology and seemed unlikely to settle
and zero net emissions by 2050 to help avoid the worst effects the question among scientists: Only 12%
of climate change. Some commentators say the 2040 milestone of respondents said no further studies into
proposed by the Commission does not go far enough, noting that the pandemic’s origin are necessary.
the 90% target was the minimum recommended by an EU scientific
advisory panel. The proposal also omits a politically sensitive state- CERN maps path for big collider
ment in an earlier, leaked draft that suggested it would be possible | The future of the
PA R T I C L E P H YS I C S
to reduce some kinds of carbon emissions from agriculture by 30% European laboratory CERN became
by 2040, compared with 2015 levels. The Commission also released clearer this week as officials updated
their plan to build two enormous parti-
a new carbon management strategy, including plans for scaling up cle colliders in a tunnel that would cross
carbon capture and storage. The proposed target requires approval the border of Switzerland and France,
by the European Parliament; some members have voiced opposition. circling beneath both countries. The
tunnel’s path has been decided, yielding
a circumference of 91 kilometers, slightly
shorter than the initial goal of 100 kilo-
crown of spiky leaves stuck out perpendicu- meters, lab officials said. The first of the
Rare fossil reveals weird early tree larly from the trunk. Scientists named the colliders, to be built by the mid-2040s,
| The earliest trees, from
PA L E O N T O L O GY tree Sanfordiacaulis densifolia, after the would be an electron-positron collider
nearly 400 million years ago, are known owner of the New Brunswick quarry where to produce copious amounts of a known
mostly from fossils of their trunks; their they found five specimens. The fossils, particle, the Higgs boson, and study it in
leaves and canopy shapes have remained a among the few showing trees with attached detail. To generate new types of particles,
mystery. A newly reported, 350-million-year- leaves and branches, probably were pre- CERN would then build a proton-proton
old tree found in Canada provides a vivid served when a landslide buried them in a collider with an energy seven times that
answer for one such primordial species: As lake. These trees stood at least 2.6 meters of CERN’s current Large Hadron Collider,
if having a perpetual bad hair day, a thick tall, with each of their more than 200 leaves which enabled researchers to discover the
extending about 1.7 meters, the research Higgs boson in 2012. The tunnel and first
team reported last week in Current Biology. collider would cost roughly €15 billion.
1m
CERN hopes to decide by 2028 whether
to proceed with it.
COVID-19 poll backs natural cause
| In a global survey
P U B L I C H E A LT H
released last week, most scientists said the U.S. seeks plan for wolf recovery
COVID-19 pandemic probably began when | The U.S. Fish and
C O N S E R VAT I O N
a natural virus jumped from an animal Wildlife Service (FWS) will create its
to a human, not because of an accident first national conservation plan for
in a research lab studying coronaviruses. gray wolves, which may help harmonize
Organized by the Global Catastrophic Risk regional efforts to preserve the controver-
Institute, a U.S. think tank, the poll set sial species. Its populations—and people’s
out to gauge opinions among a larger and willingness to save them—vary widely
more diverse set of researchers than is across the lower 48 states. FWS protects
usually quoted about the controversy. (The gray wolves in several northeastern states

572 9 FEBRUARY 2024 • VOL 383 ISSUE 6683 science.org SCIENCE

 ENERGY

Simpler heating method raises hopes for laser fusion

R
esearchers this week reported that a simpler way to use
lasers to spark nuclear fusion may make it more viable
as an energy source. Fusion, which promises carbon-
free energy generation, works by forcing nuclei of heavy
hydrogen isotopes to merge—the process that powers
the Sun. In 2022, the powerful laser at Lawrence Livermore
National Laboratory’s National Ignition Facility (NIF) achieved
a world first by crushing a pellet of fuel and sparking a fusion
reaction that generated more energy than the lasers put in. Now,
a team using the University of Rochester’s Omega laser system
(pictured) says it can achieve equivalent energy output more
cost effectively by using an alternative approach, “direct drive.” It
eliminates a step used by NIF that converts laser light into x-rays
to trigger the fusion reaction—a change that allows the fuel
targets to be simpler and cheaper, the research team reports this
week in Nature Physics.

under the U.S. Endangered Species Act
because of small populations. The preda-
tors are more plentiful in western states,
but hunters are killing growing numbers
in Idaho and Montana as permitted by
new laws there. In contrast, Colorado’s
wildlife managers began to reintroduce
wolves in December 2023, as required by
a state ballot initiative. The inconsistent
picture has forestalled FWS efforts to
delist certain populations because judges
have ruled the agency did not consider the
species’ recovery nationwide. FWS aims to
PHOTO: EUGENE KOWALUK/UNIVERSITY OF ROCHESTER LABORATORY FOR LASER ENERGETICS
Infectious Diseases found, but they did
develop an infection resembling a human
one. For example, the virus replicated
primarily in the jejunum, a part of
the small intestine, as in humans. The
animals also failed to develop a durable
immune response, making them suscep-
tible to a repeat infection, the researchers
write in Nature Microbiology.
Burned scroll muses on pleasure
A R C H A E O L O GY | Using a combination of
IN OTHER NEWS
UP AND AWAY? A German company,
Messer, last month submitted winning
bids totaling $364 million in an auction of
the U.S. government’s helium reserve and
production facilities, which supply 9% of
global demand. Scientists remain worried
about how the sale will affect access and
cost for the gas, needed for instruments
such as nuclear magnetic resonance
spectroscopy systems.

finish its plan by the end of 2025.
GI virus gets monkey model
B I O M E D I C I N E | Rhesus macaques are
susceptible to infections by human
noroviruses, researchers report this week,
providing the first animal model for
studies to develop drugs and vaccines for
the viruses. Infamous for ruining holiday
cruises and large catered events, noro-
viruses cause an estimated 700 million
gastrointestinal infections each year,
prompting diarrhea and vomiting. Most
people recover within days, but infec-
tions can become chronic or fatal in the
elderly and immunocompromised. Rhesus
monkeys given oral norovirus doses didn’t
experience those symptoms, a team from
the U.S. National Institute of Allergy and
x-rays and artificial intelligence, scien-
tists have reached a new milestone in
efforts to read hundreds of 2000-year-old
papyrus scrolls incinerated by the erup-
tion of Italy’s Mount Vesuvius in 79 C.E.
Physically unfurling the scrolls, found
in what was a library near Pompeii,
would destroy them. This week, a trio
of researchers won the $700,000 grand
prize offered by the Vesuvius Challenge
organization after they digitally peered
inside a scroll and recovered 11 columns
of text containing more than 2000
characters written in ancient Greek. (In
October 2023, two of the team members
identified only a single word—porphyras,
or purple.) Preliminary analysis has
revealed the text contains musings about
pleasure and may have been penned by
an Epicurean philosopher, Philodemus.
NEWBORN SCREENING A U.S. advisory
panel last week added Krabbe disease,
a rare but fatal brain disorder, to the rec-
ommended screening list for U.S. infants.
The 10-3 approval reverses a split vote in
2023 by the same group. Increasing evi-
dence shows that stem cell transplants
can help (although not cure) affected
children if given early.
SNOW LEOPARD COUNT India has com-
pleted its first systematic census of the
elusive snow leopard, tallying 718 across
areas that included prime habitat in
the Himalayas. The Wildlife Institute of
India and its partners based the num-
ber on sampling and camera traps. The
International Union for Conservation of
Nature classifies the species, native to
Central and South Asia, as vulnerable.

SCIENCE science.org 9 FEBRUARY 2024 • VOL 383 ISSUE 6683 573

NE WS
IN DEP TH
In altermagnets, neighboring atoms are rotated and their magnetic spins are flipped.
PHYSICS
Researchers discover new kind of magnetism
More than 200 materials could be “altermagnets,” predicted just a few years ago
By Zack Savitsky
F
or thousands of years, people have
been drawn to the apparent magic of
magnets. Ancient Greek philosophers
believed dark rocks called lodestones
had souls because of their ability to
move iron flakes.
Physicists now know that magnetic mate-
rials glean their power from the behavior of
the atoms inside them. But magnetism still
holds secrets. Researchers have recently
found signs of a wholly new class of magne-
tism, one with characteristics of each of the
two conventional kinds, ferromagnetism
and antiferromagnetism.
More than 200 materials should exhibit
the newfound phenomenon, according to
theoretical predictions, and physicists are
closing in on direct experimental evidence
for it, which could lead to more efficient
electronic devices. Already they have found
a handful of materials that seem to exhibit
this “fundamentally new type of magne-
tism,” says Paul McClarty, a physicist at the
Léon Brillouin Laboratory. “It’s expanding
our understanding of the ways that matter
can work.”
Inside solid materials, atoms are sur-
rounded by electrons that all have a prop-
574 9 FEBRUARY 2024 • VOL 383 ISSUE 6683
erty called spin, which endows each atom
with its own tiny magnetic field. The total
spin for each atom is represented by an ar-
row that can point in different directions.
In ferromagnets, all the spins inside the
material are aligned, resulting in a net
magnetic field. In addition to sticking pho-
tos to the fridge, ferromagnets are useful
because their spins can easily be flipped
around by applying another magnetic field,
creating distinct states that can be used as
computer memory. This technique birthed
the emerging technology of spintronics, in
which information is encoded via electron
spin rather than charge.
In the 1930s, scientists realized it’s much
more common for the spins of neighbor-
ing atoms to point in opposite directions
so their net magnetization cancels out. Be-
cause the staggered arrangement is much
more stable than the uniform one, these
antiferromagnets are nearly impossible to
magnetize with applied magnetic or elec-
tric fields. When French physicist Louis
Néel won a Nobel Prize in 1970 for his
pioneering work on antiferromagnetism,
he described the phenomenon as “interest-
ing but useless.” Nevertheless, the concept
has proved handy: During World War II,
electric coils were used to make ship hulls
behave like antiferromagnets and evade
magnet-seeking mines.
More recently, scientists have begun to
devise strategies for building spintronic
devices out of antiferromagnets. Although
their rigid spins are harder to manipulate,
they can in principle flip 1000 times faster
than those in ferromagnets, allowing for
more energy efficient information storage
and processing.
A few years ago, Libor Šmejkal, a physi-
cist at the Johannes Gutenberg University
of Mainz, was hunting for a possible anti-
ferromagnetic spintronic material. He stum-
bled across a compound called ruthenium
dioxide that seemed promising—but odd.
His calculations suggested it should have
no net magnetization, like a normal anti-
ferromagnet. But he also predicted that when
subjected to an electric current, the material
would behave like a ferromagnet: Magnetic
forces in the material would deflect the elec-
trons in the current, leading to a strong volt-
age in the perpendicular direction. In 2020,
a team in China experimentally confirmed
ruthenium dioxide’s paradoxical properties.
The following year, Šmejkal and col-
leagues laid out a proposal explaining
how materials like ruthenium dioxide
could be part ferromagnet and part anti-
science.org SCIENCE

ferromagnet. They called them alter-
magnets. In most materials, electron spin
arrows align with the orientation of their
host atoms within the crystal lattice. But in
some materials, spin arrows can rotate in-
dependently of the atoms, and Šmejkal and
colleagues considered one in which every
other atom was rotated by 90° and its spin
flipped by 180°.
Altermagnets would combine the most
prized features of ferromagnets and anti-
ferromagnets. With zero net magnetiza-
tion, they are graced with the stability
and fast spin-flipping speeds of an anti-
ferromagnet. But the spins in an alter-
magnet, like those in a ferromagnet, can
be readily ushered into distinct up and
down states, allowing for easier memory
writing. “You can have your cake and
eat it, too,” says Jairo Sinova, another
physicist in the Mainz group. Whereas
ferromagnetic spins are typically flipped
with magnetic fields, spins in an alter-
magnet could be manipulated by applying
currents in different directions.
Theorists were quick to accept Šmejkal’s
description because of its mathematical ele-
gance, but many are surprised the phenom-
enon went unnoticed for so long. “It’s one
of those theoretical constructs which are
unquestionable,” says Igor Mazin, a physi-
cist at George Mason University. “Yet it has
never been discussed before.”
More than 200 materials are predicted
to be altermagnetic—more than double the
number of known ferromagnetic materials.
Researchers are now beginning to look for
the property by shining laser light on a ma-
terial to coax it to eject electrons. By mea-
suring the properties of those electrons,
scientists can look for a hallmark of alter-
magnetism: energy levels that fall within
two distinct bands, reflecting both spin-up
Magnetism’s new twist
The properties of most magnetic materials depend on whether each atom’s magnetic field—denoted by its
spin—is pointing up (pink) or down (blue). In altermagnets, the atoms and their spins rotate independently,
giving them properties of both ferromagnets and antiferromagnets.
Ferromagnetic Antiferromagnetic
GRAPHIC: A. MASTIN/SCIENCE
Atom Electron spins
SCIENCE science.org
and spin-down electrons. (Antiferromagnets
also have spin-up and spin-down electrons,
but they sit at the same energy levels.)
Last month, a team in South Korea found
the predicted split in electron energies in
the material manganese telluride. Two ad-
ditional recent studies identify similar sig-
nals in manganese telluride and ruthenium
dioxide, and also attempt to tie the energy
bands to specific spin polarities. “Crystal-
clear proof is really hard to achieve experi-
mentally,” says Suyoung Lee, a Ph.D. student
at Seoul National University who led one of
the latest studies. “But I would say that we
now have sufficient experimental evidence
… that altermagnetism is really a thing.”
McClarty says the new experiments are
“consistent with altermagnetism,” but only
reveal the spin behavior through a slice of
the material’s magnetic landscape. Until ex-
perimentalists capture the behavior across
an entire 3D structure, “I wouldn’t hang up
my coat,” he says. Also, before altermagnets
can be exploited in electronic devices, sci-
entists must learn to synthesize materials
that have a consistent altermagnetic orien-
tation rather than a patchwork of shifting
configurations.
Mazin says confirmation of materials’
altermagnetic makeup is all but certain.
“There’s no way in nature that they wouldn’t
be,” he says. He sees the verification effort
akin to “an experiment that proves two
times two is four.”
But for Lee, the hunt promises other
payoffs: an opportunity to explore emer-
gent complex phenomena that may lead
to practical applications. “I think this is
the starting point for a whole new field of
altermagnetism,” she says. “I am happy to
be part of it.” j
Zack Savitsky is a science journalist in Berlin.
Altermagnetic
Rotated atom
N E WS
CLIMATE CHANGE
NASA mission
will zoom in on
ocean life and
bright hazes
PACE satellite will study
climate effects of
carbon-storing plankton
and reflective particles
By Paul Voosen
I
t isn’t easy seeing green. The ocean’s mix
of plankton, algae, and bacteria absorbs
vast amounts of carbon dioxide while
producing 50% of Earth’s oxygen. But
for decades, Earth-observing satellites
could not tease apart the many species
making up the green goop. That hampered
attempts to study how the floating plants
influence climate—and how global warm-
ing is affecting this foundational compo-
nent of the climate system.
The $964 million Plankton, Aerosol,
Cloud, Ocean Ecosystem (PACE) satellite is
about to change that. After surviving pan-
demic delays and four cancellation attempts
by former President Donald Trump’s ad-
ministration, it was due to launch this week
from Cape Canaveral, Florida, aboard a Fal-
con 9 rocket. Once operational, it will probe
the plants floating in the ocean as well as
the hazes of fine particles above it, another
major influence on climate. “We’re not go-
ing to be limited technologically anymore,”
says Collin Roesler, an optical oceano-
grapher at Bowdoin College. “We’re going to
be limited by our ideas.”
The satellite’s primary instrument is
NASA’s first “hyperspectral” imager to fly
on a major geoscience mission. Rather
than collecting reflected light in just a
few discrete channels, like the red, green,
and blue cones of a human eye, the in-
strument divides light into more than
200 channels, including many shades of
green, which can distinguish different pig-
ments used by phytoplankton depending
on their species and habitat. “Right now,
what we have on orbit is a box of cray-
ons with only eight crayons,” says Jeremy
Werdell, the mission’s principal investiga-
tor at NASA’s Goddard Space Flight Center.
9 FEBRUARY 2024 • VOL 383 ISSUE 6683 575
NE WS | I N D E P T H
“With PACE, we are getting a box of cray-
ons with 200 different colors.”
By studying the mix of phytoplankton
species, PACE should help answer a key
climate question, Roesler says. Large phyto-
plankton, which dominate in colder waters,
are bulwarks of carbon storage. “These re-
ally big cells get eaten by big zooplankton
and then they poop really big poop,” she
says. The carbon in the waste is more likely
to reach the ocean floor and stay safely se-
questered. In the warmer waters at lower
latitudes, smaller phytoplankton thrive—
but the carbon in their cells tends to be
gobbled up by microbes and ultimately
emitted back into the atmosphere. Mod-
els have suggested these smaller plankton
could expand toward the poles with global
warming; PACE could see whether these
fears are true.
PACE’s sharp color vision should also
help it track human impacts on lakes, riv-
ers, and coastlines, by separating natu-
rally occurring species from bacteria or
dissolved organic matter associated with
NASA’s latest satellite will scan the planet every 2 days, tracking shifts in ocean plankton and particles in the sky.
sewage or fertilizer runoff. And it will ex-
plore the role that eddies, kilometers-wide
swirls of turbulent water, play in driving
the movements of plankton species and
the nutrients they depend on.
The satellite will restore NASA’s ability
to study another key factor in climate: the
small aerosol particles and clouds that re-
flect some 90% of the Sun’s light into space.
NASA is still reeling from the loss of the
$424 million Glory satellite, which crashed
soon after launch in 2011. It carried an ad-
vanced polarimeter designed to capture po-
larized light, in which the waves all vibrate
in the same plane. Sunlight bouncing cha-
otically off the ground and clouds tends to
576 9 FEBRUARY 2024 • VOL 383 ISSUE 6683
lose any polarization. But because aerosol
particles are discrete and compact, they re-
flect polarized light and pop out vividly.
Brian Cairns, PACE’s deputy project sci-
entist at NASA’s Goddard Institute for Space
Studies, who also led development of Glory’s
doomed instrument, says the new satellite
carries two “petite but powerful” polari-
meters, which “get almost all of what you
wanted in a much more compact package.”
By identifying the type, size, and abun-
dance of particles, the two polarimeters
should fill out the inventory of aerosols, in-
cluding elusive ones such as sea spray over
the ocean. The instruments should also
show how aerosols influence the growth
and life span of clouds, Cairns says, which
can affect the rate of warming. Researchers
are currently debating the extent to which
cleaner shipping fuels have suppressed the
formation of reflective clouds over the ocean,
thereby adding to global warming (Science,
4 August 2023, p. 467).
James Hansen, the famed former NASA
climate scientist who is now at Columbia
University, welcomes the new attention to
aerosols. Hansen recently published work
suggesting that global warming is accel-
erating as pollution wanes. “As the world
realizes the climate predicament that we
face, they will rue not having a monitor of
aerosol climate forcing.”
The polarimeters, built with off-the-shelf
components, may only last a few years,
but PACE’s main instrument could run for
a decade or more. It’s been a long wait for
Roesler, who has spent her career preparing
for this intimate green view from space. “The
achievement of PACE is that it brings the ob-
serving capabilities almost to what I can do
in the lab. That’s really astounding.” j
CLIMATE CHANGE
Florida coral
restoration in
hot water
After unprecedented heat
wave killed transplanted
coral, reef experts are
charting a new strategy
By Warren Cornwall
F
our years ago, the U.S. National Oce-
anic and Atmospheric Administra-
tion (NOAA) unveiled a $100 million
coral moonshot. Over 2 decades,
nearly half a million hand-reared
coral colonies would be planted on
seven ailing reefs in southern Florida, in
a bid to revive them. Misson: Iconic Reefs
represented “one of the largest ever invest-
ments in coral restoration,” Pat Montanio,
then head of the agency’s habitat conserva-
tion program, said at the time.
Today, the project looks as ailing as
the coral it was meant to save. A record-
breaking underwater heat wave that swept
the Caribbean and southern Florida in
2023 killed most of the transplanted colo-
nies. Elkhorn coral, with its sprawling, flat
branches, was to be the cornerstone of the
initiative’s first phase, creating a foundation
on which other corals could grow. Instead it
proved to be one of the most heat sensitive
species. Many elkhorn corals, both wild and
hand planted, are dead, their blanched skel-
etons coated in algae. “The picture is not
good,” says Jennifer Moore, a coral recovery
coordinator at NOAA who helped lead a
January meeting in Marathon, Florida, not
far from one of the Iconic Reefs, to assess
the damage and plot a path forward.
The results are provoking a crisis of con-
fidence in decades-old methods of reef res-
toration pioneered in Florida and adopted
around the world. “Is it responsible to grow
these species and then just put them back
out to die?” asks Ian Enochs, a marine bio-
logist who heads NOAA’s reef monitoring
program in the Atlantic Ocean and Carib-
bean Sea. “We need to have a shift in how
we conduct restoration.”
When the project began, Florida’s reefs
were already in desperate need of help. A
string of diseases, coupled with water pol-
lution, human disturbance, and successive
heat waves, had all taken a toll. Elkhorn
science.org SCIENCE
 and staghorn corals—two species in the cess story, and found no living elkhorn or He and some others think sexual repro-
Acropora genus—were among the worst staghorn. Similar damage was observed duction—mixing eggs and sperm gathered
hit, having declined more than 97% since through much of the Caribbean at both from spawning coral to grow new colonies—
the early 1980s. natural and restored reefs. “It’s been dev- could yield genetic combinations that would
The rescue effort, like nearly all reef res- astating everywhere,” says Margaret Miller, make some of the offspring more heat toler-
torations in Florida, relied largely on break- a coral ecologist and research director ant. The strategy, which a few Florida-based
ing off chunks of existing coral and coaxing for Secore International, a Miami-based labs have already begun to use, could also
them to build new colonies—all clones of the nonprofit that conducts coral restoration make it easier to treat the new colonies with
original coral. Restoration groups can swiftly research and collaborates with groups heat-resistant algal symbionts, which many
make thousands of new corals each year by throughout the Caribbean region. corals absorb during their larval stage.
suspending coral fragments in open water The destruction has led Nedimyer to Erinn Muller, an ecologist leading coral re-
nurseries until they are large enough to at- conclude that, at least for the Acropora spe- search at the Mote Marine Laboratory, says
tach to the sea floor. cies in Florida, it’s foolhardy to continue to some elkhorn and staghorn corals sexually
Elkhorn and staghorn have long been rely on clonally produced coral. “One of the bred in the lab there survived the hot water
favorites. “They’re ridiculously easy to things I learned is we don’t have any stag- conditions last summer. At an offshore nurs-
propagate,” says Ken Nedimyer, technical horn or elkhorn corals that can handle the ery near the southwestern tip of the Keys,
director of the nonprofit Reef Renewal. But conditions in the middle and lower Keys,” they were the sole survivors of 24,000 coral
their decline means the corals off Florida he says. “It’s a game changer.” fragments. “Seeing this unique sign of resil-
retain very little of the genetic di- ience within the sexually produced
versity that might have enabled offspring allows us a unique oppor-
new colonies to cope with the off- Unbearably hot tunity to dig into what is going on,”
the-charts hot water they faced Corals in Florida and the Caribbean faced a record-setting underwater Muller says.
last year. heat wave in 2023. By 15 September, many areas had endured more than Miller, however, warns that the
Ocean temperatures rose to un- 20 weeks with ocean temperatures 1°C above the average during the strategy risks introducing new ge-
precedented levels starting in July hottest month of the year, causing a mass coral die-off. netic problems such as inbreeding,
2023, fueled by calm, hot weather because the population of wild coral is
tied to El Niño and given an added Degree heating weeks so depleted. Importing sperm or eggs
boost by climate change. Until last 0 4 8 12 16 >20 from corals elsewhere could help. She
year, NOAA’s scale of heat wave was part of a 2019 experiment that
threats to coral reached its high- produced elkhorn coral with eggs
est level when water temperatures from the southern Caribbean island
CREDITS: (GRAPHIC) M. HERSHER/SCIENCE; (DATA) NATIONAL OCEANIC AND ATMOSPHERIC ADMINISTRATION CORAL REEF WATCH; (PHOTO) JASON GULLEY

hit 12 “degree heating weeks”—the of Curaçao fertilized by Florida-origin

equivalent of 12 weeks of tempera-
tures 1°C above the average during
the hottest month of the year. But
last year, when sea-surface tempera-
tures reached as high as 32.8°C, the
agency added three new categories,
topping out at more than 20 degree
heating weeks. “This is analogous to
a Category 5 hurricane,” says Derek
Manzello, a coral reef ecologist and
head of NOAA’s Coral Reef Watch,
which devised the warning system.
In the Keys, reefs endured up to 22
degree heating weeks last year. Many
species bleached, turning bone white
as the coral polyps expelled symbiotic
algae from their bodies. Some died
straight away, sheets of tissue peeling
from their skeletons. Elkhorn and
staghorn suffered the biggest losses.
Artificially planted Acropora
fared little better. There is no final
estimate of casualties at the seven
Iconic Reefs; a team is currently
visiting them to take stock. But a
survey in August 2023 showed 30%
of the planted staghorn and 45%
of the planted elkhorn was already
dead, and about 90% was bleach-
ing. In late September, Nedimyer
visited Pickles Reef, a focus of
planting he had viewed as a suc-
sperm. Today, the offspring are stuck
Florida in a laboratory, she says, because state
officials worry about introducing out-
The side coral genes into Florida waters.
Bahamas To her, the current situation warrants
Mission:
Iconic Reefs such experiments. “I definitely think
additional risky measures are now
called for, unless we just want to give
up,” Miller says.
Phanor Montoya-Maya, a marine
Cuba biologist and head of the restora-
tion program at the Coral Restora-
0 250 tion Foundation, cautions that sexual
km production alone won’t create ma-
ture corals quickly enough to prop
up the reefs because it can take years
for larvae to reach reproductive age.
To breed corals at the scale required
would also take more lab facilities.
“The answer is not to stop asexual
propagation,” he says.
Mission: Iconic Reefs isn’t over.
This year it will plant bulky brain cor-
als, produced asexually, which grow
more slowly but also weathered the
heat wave better, Moore says. It will
also continue to plant some elkhorn
and staghorn to test approaches that
might be more successful. “We have
to be a little more intentional,” Moore
In southern Florida, scientists documented widespread bleaching says. “We don’t just want to put them
and coral death. out to slaughter.” j

SCIENCE science.org 9 FEBRUARY 2024 • VOL 383 ISSUE 6683 577


ECOLOGY
Pollution disrupts nighttime
pollination by degrading scents
Nitrate radicals, a common pollutant, break down the cues
nocturnal insects follow to find nectar sources
By Elizabeth Pennisi
U
nder the cover of darkness, count-
less moths and other insects furiously
dart around woodlands and deserts,
seeking nectar from night-blooming
plants—and pollinating them in the
process. But the scents the insects
home in on have grown fainter. Nitrate radi-
cals, a common pollutant, break them down
before they can travel far, a research team
reports on p. 607. The team thinks the olfac-
tory disruption goes as far back as the Indus-
trial Revolution 200 years ago.
The research, involving field studies, wind
tunnel experiments, and the latest atmo-
spheric models, has worrisome implications.
For pollinators and the crops and ecosystems
dependent on their skills, nitrate radicals
are “potentially a very significant threat,”
says Robbie Girling, an applied ecologist at
the University of Southern Queensland. The
same probably goes for other pollutants. The
work demonstrates “the ‘fragility’ of these
[scents],” particularly at night, adds Maryse
Vanderplanck, a chemical ecologist at CNRS,
the French national research agency.
She, Girling, and others previously
showed that ozone, another pollutant, in-
terferes with the ability of bees and other
daytime pollinators to find flowers. But
impacts at night are especially concerning.
“For every bee and butterfly we see by day
visiting flowers, [we] know that many more
578 9 FEBRUARY 2024 • VOL 383 ISSUE 6683
may be visiting after nightfall,” says David
Wagner, an entomologist at the University
of Connecticut. Nocturnal insects including
moths such as hawk moths (Hyles lineata)
and tobacco hornworms (Manduca sexta)
depend on the pleasant scents of garde-
nias, honeysuckles, lilacs, jasmine, and
most night bloomers to guide them long
distances to nectar sources. Some crops,
too, appear to lure pollinators at night, says
Alison Scott-Brown, a plant biologist at the
University of Cambridge.
Jeff Riffell, a neuroscientist at the Univer-
sity of Washington (UW), and his colleagues
were the first to identify a possible pollut-
ant challenge for nighttime pollinators. In
2014, they showed that certain airborne
chemicals made it more difficult for tobacco
hornworms to find flowers. But early stud-
ies tended to examine high concentrations
of pollutants in rather artificial situations.
So, in 2016, Riffell teamed up with UW at-
mospheric chemist Joel Thornton and their
graduate student Jeremy Chan to probe the
issue under more natural conditions.
In a sagebrush patch about 280 kilometers
from Seattle, Chan recorded pollinator visits
to pale evening primrose (Oenothera pal-
lida). He logged a diverse array of nighttime
visitors, including bees, antlions, and moths.
Underscoring their importance, the plants
produced few to no seeds when Chan cov-
ered blooms for the duration of flowering
with a mesh bag to prevent pollination.
The sweet smell of evening blooms attracts a hawk
moth, a major nighttime pollinator.
The researchers suspected nitrate radi-
cals, produced when nitrogen oxides from
exhaust and industrial smokestacks re-
act with ozone, might also interfere with
pollination, as they build up at night and
are highly reactive. Chan, now a chemical
ecologist at the University of Naples, did lab
studies showing the pollutant breaks down
compounds called monoterpenes, which
are common in many plant scents. He then
synthesized the degraded odors and, in
wind tunnel experiments and field studies,
showed they are less effective than the orig-
inal scents at luring moths. “This paper had
it all,” says Clinton Francis, a sensory eco-
logist at California Polytechnic State Uni-
versity (Cal Poly). “It was really thorough
and very impressive.”
The field studies showed that on aver-
age, moths visited each flower twice a
night. But when Chan released the de-
graded scent into the sagebrush patch,
the visits dropped to about one every two
nights. He also placed moths in a wind
tunnel together with paper cones emitting
the degraded scent. “Some moths couldn’t
smell the ‘flowers’ at all,” Riffell says. “It’s
almost like the moths had COVID.”
Chan worked with Thornton to under-
stand where, globally, nighttime levels of
this pollutant are likely high enough to dis-
rupt pollination, finding hot spots across
the Northern Hemisphere. The group also
calculated that, beginning in preindustrial
times, nitrogen radicals have shortened the
distance scents travel by a factor of five, on
average. In and near big cities, pollution
degrades floral scents so quickly they are
probably ineffective, the team reports.
“I hope [the work] can help elevate chem-
ical pollution to the same recognition that
we give artificial light,” says Sarah Jennings,
a chemical ecologist at Cal Poly, referring to
another major disrupter of insect behavior.
Riffell agrees, adding that nitrogen radicals
and other pollutants may also disrupt the
chemical cues that many animals use for
mating and other purposes.
Wagner is more circumspect, suggesting
pollinators face greater threats, such as in-
vasive plants. But Ilaria Negri, a zoologist at
the Catholic University of the Sacred Heart,
says pollution is another significant blow
to insects already hammered by improper
use of pesticides, diseases, climate change,
and development. “Pollinators [are] literally
starving,” she says. “That pollinators are un-
able to find already scarce food due to pol-
lution is challenging for many species.” And
that can only be bad news for the plants
that depend on them. j
science.org SCIENCE

PUBLIC HEALTH
Does fluoride in drinking water risk IQ loss?
A U.S. federal court is examining a controversial link between fluoride and neurotoxicity
By Erik Stokstad
A
long-simmering scientific battle took
on new life last week, as experts clashed
in a San Francisco courtroom over
whether the U.S. Environmental Pro-
tection Agency (EPA) should ban fluo-
ridation of drinking water to protect
fetuses and children from the risk of neuro-
developmental problems.
The case, being heard in a federal dis-
trict court, “is precedent setting,” says Lynn
Bergeson, a managing partner of Bergeson &
Campbell who focuses on chemi-
cal toxicity. Rarely have judges had
to “manage the enormity of this
record of scientific evidence. …
That’s why there’s a lot of attention
focused on this right now.”
Adding fluoride, a common
mineral, to drinking water lessens
tooth decay in children and adults
by 25%. The Centers for Disease
Control and Prevention calls fluori-
dation, which began in 1946 in the
United States and is decided by lo-
cal water districts, one of 10 “great
public health achievements” of the
20th century. But from the get-
go, some activist groups worried
about potential harm. And over
the past few decades, studies of
laboratory animals and of commu- Fluoride is added to drinking water at a facility in Burlington, Vermont.
nities where drinking water natu-
rally contains fluoride have hinted that high
levels might affect brain development.
The current case has put the spotlight on
an unpublished assessment by the federal
government’s National Toxicology Program
(NTP). It reported “moderate confidence”
that drinking water containing fluoride at
levels at least twice as high as those recom-
mended by the federal government is associ-
ated with lower IQ in children. The Fluoride
Action Network (FAN) and other groups
argue that such data indicate EPA should
be regulating fluoride under the Toxic Sub-
stances Control Act (TSCA).
Last week’s testimony marked the resump-
tion, after a yearslong pause, of litigation that
PHOTO: ALDEN PELLETT/AP
began in 2017. Over 9 days, Judge Edward
Chen of the U.S. District Court for the North-
ern District of California is scheduled to hear
the views of seven experts on a range of tech-
nical topics, including whether there is a safe
threshold for fluoride exposure.
SCIENCE science.org
During opening arguments on 31 Janu-
ary, the attorney for the plaintiffs, Michael
Connett, highlighted babies fed formula
made with tap water as a “critical vulner-
able group being exposed to the highest dose
of fluoride of any age group in the popula-
tion. That is a major cause for concern.” He
argued that if EPA was aware of a different
compound that posed a similar potential
threat to newborns, the agency wouldn’t
hesitate to regulate it. “I don’t think we need
to speculate about what EPA would do with
that scenario,” he said.
EPA and others argue there is little strong
evidence that the current recommended con-
centration of fluoride in drinking water—
0.7 milligrams per liter—poses a threat. (That
level is set to avoid discoloration of teeth in
children.) “The dose makes the poison,” said
Paul Caintic, a Department of Justice attor-
ney. “Given the current state of science, the
court cannot conclude that community water
fluoridation presents an unreasonable risk.”
The fluoride lawsuit is the first to reach
trial under a 2016 TSCA provision allowing
citizens to ask a court to assess a chemical’s
risk. FAN turned to that strategy after EPA
said its request that the agency ban fluorida-
tion lacked a “scientifically defensible basis.”
The trial began in June 2020 but Chen sus-
pended it months later to wait for publication
of the NTP report. That assessment, based on
work conducted by NTP experts from 2016 to
2019, initially classified fluoride as a cogni-
tive developmental hazard. But NTP removed
9 FEBRUARY 2024 • VOL 383 ISSUE 6683
N E WS | I N D E P T H
that determination after two reviews by the
National Academies of Sciences, Engineer-
ing, and Medicine, in 2020 and 2021, found
the evidence was lacking. One flaw, the re-
views said, was that the assessment didn’t fo-
cus on how people respond to various doses
of fluoride.
The report has yet to be formally re-
leased, and FAN charges that the repeated
reviews have been designed to delay it—a
critique supported by some scientists and
former NTP leaders. According to emails
released to FAN under the Freedom of In-
formation Act, a senior official
at the Department of Health and
Human Services intervened in
2022 to halt publication just days
before the report was scheduled
to be released. But in March 2023,
as a result of litigation, NTP did
release a 393-page draft. And
that May, NTP’s science advisory
board signed off on the report
and sent it to NTP leadership.
Dental health advocates say ac-
tivists are already spinning the re-
port to their advantage by claiming
NTP found no safe dose (a claim
not directly stated in the report).
The American Dental Association
has strongly urged NTP to add a
disclaimer to the report highlight-
ing its scientific limitations.
In the meantime, Chen has al-
lowed the lawsuit to resume. He has sug-
gested that EPA must regulate fluoride if
FAN proves that fluoride poses an “unrea-
sonable risk” to pregnant women and chil-
dren. Under TSCA, EPA cannot consider a
chemical’s benefits, such as oral health.
Bergeson describes Chen as “a very dis-
ciplined jurist and a very demanding task-
master,” adding that “presumably the data
will be looked at correctly and weighed
appropriately.”
Lynn Goldman, an epidemiologist at
George Washington University and former
EPA official, doubts that Chen will order EPA
to ban fluoride in drinking water. “I can’t
imagine a court coming back and saying, ‘You
did a risk assessment wrong, and here’s how
you should have done it,’” she says. But the
court might order EPA to consider fluoride as
a toxic substance under TSCA and make it a
high priority for a new evaluation to establish
safe exposure limits. j
579
NE WS
FEATURES
A THOUSAND YEARS OF
SOLITUDE
How did the first human settlers
of the Canary Islands survive a
millennium of isolation?
M
ore than 1000 years ago, a
young man stood on the north-
ern shore of the island now
known as El Hierro. Across the
wave-swept Atlantic Ocean,
he could see the silhouettes
of other islands, a volcanic
peak on one soaring toward
the clouds only 90 kilometers
away. Yet, for him, those islands were as un-
reachable as the Moon.
His body betrayed the rigors of life on
his arid volcanic outcrop. His molars were
580 9 FEBRUARY 2024 • VOL 383 ISSUE 6683
By Warren Cornwall,
in the Canary Islands
worn almost to the gums from grinding fi-
brous wild fern roots. His ancestors here
had farmed wheat, but he and his con-
temporaries grew only barley and raised
livestock such as goats. His genes held
evidence that his parents were closely re-
lated, like many of the roughly 1000 people
on the island, who had not mingled with
outsiders for centuries. Also like many of
his fellow islanders, he bore signs of an old
head injury, likely sustained in a fight.
“This population faced a lot of challenges,”
says archaeologist Jonathan Santana
of the University of Las Palmas de Gran Ca-
naria (ULPGC). “Survival on this island was
a challenge every day.”
Yet the first Canarians, who arrived
from North Africa roughly 1800 years ago,
survived and even thrived on this arid,
windswept archipelago for 1000 years.
They numbered in the tens of thousands
when Europeans arrived at the start of the
science.org SCIENCE

Archaeobotanist Jacob
Morales found ancient
granaries high on a cliff on
Gran Canaria.
14th century. Not long after, conquest and
genocide had largely erased them as a peo-
ple. But their DNA lives on in many island-
ers today, and traces of their lives remain, in
granaries, cliff dwellings, ceramic figurines,
and hundreds of human remains like those
PHOTO: W. CORNWALL/SCIENCE
of the man on El Hierro—all remarkably
well preserved by the dry climate.
By applying the latest archaeological
tools to this trove of material, Santana and
other home-grown archaeologists are un-
earthing their stories, shedding light on
puzzles that have mystified archaeologists
SCIENCE science.org
since the 19th century. For instance, how
did people with no apparent seafaring skills
reach and survive on the archipelago? Why
did their crops and cultures differ from
island to island despite their common ori-
gin? The answers offer insights into how
human societies cope with—and respond
to—challenging environments, says Scott
Fitzpatrick, a University of Oregon archaeo-
logist who studies island cultures. “The Ca-
naries have been sort of an enigma.”
THE CANARY archipelago sits like a small curv-
ing tilde 100 kilometers off the North Afri-
can coast (see map, p. 582). Its major islands
range in size from El Hierro, the smallest at
269 square kilometers, to
Tenerife, more than seven
times larger, roughly the
size of Maui. The islands’
volcanoes, some of which
are still active, reach up
to 3718 meters. Vegeta-
tion ranges from parched
cacti to shrubland with
wild olives and juniper.
On a few islands, ever-
green forests of pine and
laurel thrive on moisture
brought by northeasterly Ancient people on El Hierro, the smallest
trade winds. of the Canary Islands, wore down
When the first Euro- their teeth by eating tough wild plants.
peans arrived in the 14th
century, they found all seven major islands
occupied, with up to tens of thousands of
people each on Gran Canaria and neighbor-
ing Tenerife. What the Europeans didn’t find
was any evidence of seaworthy craft. The is-
land dwellers “have no ships or other means
to get from one [island] to the other, unless
they swim,” wrote Nicoloso da Recco, a Geno-
ese navigator who visited the islands in 1341
on behalf of the Portuguese monarchy.
European archaeologists were fasci-
nated with the early Canarians. The French
thought the first settlers were Cro-Magnon,
like prehistoric people in France; German
archaeologists thought they must have been
Aryan; the Spanish thought they were Stone
Age relatives of the same North Africans who
settled the Iberian Peninsula. More recently,
archaeologists from the Canaries themselves
have begun to lead investigations. They
have tapped into evidence including an-
cient bones, the genes of living people, and
1000-year-old grains of barley.
By analyzing ancient DNA from
radiocarbon-dated bones, archaeologists
in the past 15 to 20 years have found that
the first islanders had the strongest genetic
ties to the Amazigh cultures of northwest-
ern Africa, also known as Berbers. Rock in-
scriptions on the islands also echo Amazigh
alphabets. Two thousand years ago, North
N E WS
Africa was home to a variety of Amazigh so-
cieties, from pastoral herders in rural tribes
to kingdoms with large urban centers heav-
ily influenced by Roman culture.
In the latest genetic study, published in
Nature Communications in 2023, a team led
by geneticist Rosa Fregel of the University
of La Laguna sequenced the whole genomes
of 49 ancient individuals spanning all ma-
jor islands and dated from about 300 to
1500 C.E. Her team found that distinct ge-
netic groups, each tied to Amazigh popula-
tions, had settled the western and eastern
clusters of islands.
But DNA doesn’t answer how the first
settlers crossed the watery gulf from the
mainland. Peter Mitchell,
an archaeologist at the
University of Oxford, dis-
misses speculation that
they were deported to
the islands by Romans
expanding into Africa
in the first century C.E.
Archaeological and writ-
ten evidence shows the
Romans knew of the
islands and briefly op-
erated a dye factory on
a tiny outcrop there, ex-
tracting a prized purple
hue from a marine snail.
But Mitchell says the Ro-
mans had no practice of wholesale deporta-
tions of communities, usually opting to kill
or enslave troublesome groups.
Instead, Mitchell suggests the Amazigh
settlers took to the sea to escape conflicts
sparked by a drying climate or the Roman
expansion. Some islands might have seemed
like an oasis compared with the desert from
which those settlers came. “People might
think going over the Atlantic to get here,
‘This is quite worth it. We’ve never seen so
many trees,’” Mitchell says.
Santana hypothesizes that people’s sea-
faring skills declined over time because they
had little incentive to traverse the ocean for
trade. The islands’ volcanic rock contains
no valuable metal ore, and the dry ground
yielded little food to spare. “It wasn’t worth
it,” Santana says.
A DARK CAVE entrance beckons from the
middle of a 140-meter-tall cliff flanking
a narrow valley on Gran Canaria. Jacob
Morales, an archaeobotanist at ULPGC,
crouches at the end of a ledge, the wind
whipping his face. Below him lies a land-
scape of stunted shrubs, dry grasses, and
precipitous, rocky ridges that Spanish phi-
losopher and poet Miguel de Unamuno
once described as “una tempestad petri-
ficada” (a petrified storm). A 2-meter gap
9 FEBRUARY 2024 • VOL 383 ISSUE 6683 581
NE WS | F E AT U R E S
of blank sheer rock separates Morales from
the cavern.
Morales doesn’t plan to cross that blank
wall today. But in 2011, he choked back his fear
of heights and made his way up to the cave
with the aid of ropes set by a rock climber.
Researchers at the local archaeological mu-
seum wondered why he bothered, he recalls.
Other archaeologists had entered the cave as
early as the 1980s and toted away the obvi-
ous artifacts—bits of basketry and pottery.
But Morales was hunting smaller quarry. Sift-
ing through piles of sediment, he picked out
tiny fragments of seeds representing a rich
variety of foods: barley, durum wheat, lentils,
fava beans, and figs. Radiocarbon dating re-
vealed the seeds were as much as 1000 years
Extent of
Canary Roman Empire
Islands 117 C.E.
La Palma
Tenerife
La Gomera
Gran Canaria
El Hierro
0 75
Atlantic Ocean
km
old; all predated the arrival of Europeans.
“It’s amazing preservation,” he says.
Morales’s discoveries showed that some
of the ancient Canarians stored food in
cliffside granaries. They also indicated
that each island’s society proceeded down
its own path. People on Gran Canaria—but
only on that island—chipped small alcoves
from volcanic rock, lined them with plaster,
and added wooden doors to protect their
food. Each of the alcoves, which were found
in more than 50 sites, contained a medley
of different seeds, suggesting to Morales
that they belonged to separate families who
stashed diverse crops in the same space.
582 9 FEBRUARY 2024 • VOL 383 ISSUE 6683
The granaries and their variety of foods
also suggest that Gran Canaria, where cen-
tral highlands capture up to 1 meter a year of
rainfall compared with 10 to 20 centimeters
elsewhere, was one of the most productive
islands in the archipelago. Even today, the
mesa topping the cliff above the granary “is
the best place for cultivating barley in this
area,” Morales says.
The barley grown there today traces
its lineage back to the ancient granaries.
Morales and Swedish researchers analyzed
DNA in barley seeds found in dwellings and
granaries from three islands dating between
1000 and 600 years ago. Each island had
a different strain descended from a single
shared variety, which itself separated from
The lonely islands
People from North Africa settled the Canary
Islands, an arid volcanic archipelago, in the early
centuries C.E. when the Roman Empire was at
its largest. Evidence suggests little or no contact
with the mainland for the next 1000 years.
Lanzarote
Fuerteventura
Morocco
African barley about 2000 years ago, accord-
ing to genetic estimates. All the strains were
genetically distinct from today’s European
and North African barley. The team also ana-
lyzed other crops; seed by seed, they pieced
together evidence that a similar “packet” of
domesticated African crops arrived with the
first settlers of each island.
The inhabitants of some other islands
were less fortunate than those on Gran Ca-
naria. On those islands, many crops vanish
in more recent layers of sediment, perhaps
because of the harsh environment or cli-
matic shifts. On Fuerteventura, for example,
signs of barley, wheat, and lentils end after
the eighth century. European colonists in
the 15th century noted the island’s residents
lived chiefly on livestock meat and seafood.
On El Hierro, wheat vanishes from the re-
cord, but barley remains. “At the beginning
there is exchange of seeds between islands,”
Morales says. “Then there is isolation.”
The genetic differentiation in barley is
mirrored in the archipelago’s people, paint-
ing a picture of profound separation from
both other islands and the mainland. For
example, the genetic split between people
colonizing the western and eastern islands
persisted even in remains dated to hun-
dreds of years after the islands were settled.
“If migration were something that was su-
per common between islands, we shouldn’t
see two clusters [persist],” Fregel says.
Still, some archaeologists do see signs
of new arrivals, at least on Gran Canaria,
in changes in how the dead were treated.
The first burials, around 300 C.E., were in-
terred communally in caves, says Verónica
Alberto-Barroso, an archaeologist at ULPGC.
Around 600 C.E., some people began inter-
ring bodies in individual open-air tombs
of piled rocks, a trend that lasted about
500 years. Then around 1200 C.E., stone-
lined pits and graves grew popular.
Alberto-Barroso and her colleagues also note
simultaneous shifts in where people lived on
the island, and the kinds of homes and ce-
ramics they made. They conclude new North
African groups with new cultural practices
might have arrived during these periods.
But the changes could simply mark
cultural evolution, Santana notes. Other
islands don’t show such cultural shifts
and some display even stronger signals of
genetic isolation. Fregel found that in an-
cient human remains from four islands—
Fuerteventura, Lanzarote, La Gomera, and
El Hierro—the chromosomes inherited
from each parent were closely related, a
sign of a small, inbred population. At least
one individual on each island had chromo-
somes more similar than if their parents
had been first cousins.
On El Hierro, Fregel analyzed shared ge-
netic variants in four individuals who died
between the 13th and 15th centuries. None
of these people were directly related, but
the similarities in their DNA suggested a re-
cent common ancestor—a sign of a popula-
tion bottleneck. The population apparently
crashed around the ninth century, well after
the islands were settled. “It was super dras-
tic,” Fregel says.
The apparent ninth century crisis coin-
cides with the onset of warmer than aver-
age global temperatures, in a time known as
the Medieval Warm Period. Fregel wonders
whether the warmth might have brought
climate change and famine. But ancient
science.org SCIENCE
 DNA from 34 people on more populous
Tenerife and Gran Canaria shows no signs
of a population collapse.
“To me that’s the most interesting thing,”
Fregel says. “We imagined that all the is-
lands would be the same. We are realizing
that each island had a different scenario.”
THE PEOPLE OF all the islands do exhibit some
similar signs of hardship. In Santana’s base-
ment lab on Gran Canaria, the skull of the
unnamed young man from El Hierro rests on
a table in a plastic tray alongside nine others.
Many skulls bear small dents, some the size
of nickels, others of silver dollars—the marks
of blows. “The rate of interpersonal violence
is very, very high,” Santana says.
On Gran Canaria, Alberto-Barroso and
colleagues reported in 2018 that 27% of
347 adult skulls collected from burial caves
show signs of trauma, usually long before
the person died. Roughly one-third of male
skulls were damaged, and nearly 20% of
the female skulls. Most of the injuries were
from something like a club or stone on the
left side of the front of the skull, consistent
with face-to-face fighting. The injury rate is
far higher than in other ancient burial sites,
including ones in the Iberian Peninsula,
New Guinea, and the Solomon Islands, the
researchers reported.
Other islands are even more extreme,
according to unpublished research by
Hemmamuthé Goudiaby, an archaeologist
at Archaïos, an archaeological company. For
instance, of 82 skulls from El Hierro, 50% of
male skulls and 28% of female skulls show
signs of trauma. “These populations that seem
to have a higher incidence of interpersonal
violence are also the ones that live on islands
with less available resources,” Goudiaby notes.
He and Santana suggest ritualized
violence—orchestrated fights or duels—
might have served as a way to address con-
flicts in communities where food was in
short supply and there were few options to
move away.
It’s easy to begin to see the lives of an-
cient Canarians as ones of deprivation. But
Morales cautions against applying modern
sentiments to a different time and place, or

Archaeologists have found

more than 50 ancient
granaries in the cliffs of
Gran Canaria, some
still containing seeds of
barley, wheat, and figs.
CREDITS: (PHOTO) ERNESTO MARTIN; (GRAPHIC) M. HERSHER/SCIENCE

Chronicle of a vanished people

After the first settlers reached the Canary Islands, they were apparently mostly isolated from the wider world as well as from other islands of the archipelago
for a millennium. Scientists are tracing how the Indigenous islanders sustained themselves and comparing the trajectory of societies on each island.

Exploration Single tombs Graves European arrival Conquest begins

Expedition by Mauritanian King Juba II Some Gran Canarians are Gran Canarians are Italian Lancelotto Spanish military
lands on the archipelago, according to buried in individual open-air increasingly buried Malocello settles expedition invades
Roman author Pliny the Elder. tombs of piled rocks. in the ground. on the islands. Lanzarote.

0 200 400 600 800 1000 1200 1400

Settlement Granaries El Hierro crisis Largest art Conquest complete

Domesticated plants and animals including First evidence of Gran Genetics suggest Geometric mural, perhaps Final defeat of
barley, wheat, goats, and pigs signal the Canarians storing grain population crash a calendar, painted on a Indigenous people
arrival of humans on larger islands. in caves on this island. cave wall on Gran Canaria on Tenerife

SCIENCE science.org 9 FEBRUARY 2024 • VOL 383 ISSUE 6683 583

 NE WS | F E AT U R E S
to viewing the Indigenous people as trapped.
“I don’t agree with this view. I imagine the
lives of those people as profoundly linked
to the natural world, and especially to the
plants,” he says.
He stresses that early Canarian societies
showed remarkable adaptability. The first
settlers almost certainly came from places
with metallurgy, yet, faced with a world
without metal ore, they reinvented tools of
stone, wood, and bone. And they enjoyed
culturally rich lives, as shown by intricately
woven tapestries, clay figurines, stone etch-
ings, and—on the wall of at least one Gran
Canaria cave—elaborate geometric paint-
ings that may represent a calendar system.
Then, the Europeans arrived. The vio-
lence of traditional Canarian life pales in
comparison. The first known contact
happened in the early 1300s, when
Italian navigator Lancelotto Malocello
settled on the island now known as
Lanzarote. In 1402, soldiers of Spain’s
Castilian monarchy landed on Lan-
zarote in force, beginning a century
of conquest that ended in 1496 with
Spanish victory over Indigenous war-
riors on Tenerife. The Indigenous
people were killed, enslaved, forcibly
assimilated, or deported, save a scat-
tered few. On Gran Canaria, an Indige-
nous population estimated at between
10,000 and 60,000 was slashed to as
few as 2000.
The genocide was almost complete.
No Indigenous Canarian community
survives today. Only scraps of lan-
guage, such as names for places, foods,
or famous leaders, remain. And yet,
traces of the early settlers live on. “We
584 9 FEBRUARY 2024 • VOL 383 ISSUE 6683
Geneticist Rosa Fregel collects a sample from the remains of a
woman who lived on Gran Canaria between 600 and 900 C.E.
were told all of the Indigenous people dis-
appeared,” Santana says. “It’s not true. We
know that because of the DNA results.”
In a string of papers beginning in the
early 2000s, Fregel and colleagues found
that, on average, between 15% and 20% of
current islanders’ DNA comes from Indige-
nous sources. Local newspapers trumpeted
the findings with pride, as they did the dis-
covery by Morales and colleagues that to-
day’s Canarian barley is genetically similar
to the ancient island variety.
But Fregel is wary of using genetics to
construct identity. She notes that her sur-
vey of male Y chromosomes, which turned
up less than 10% Indigenous DNA, received
a chillier reception. The dominance of Eu-
ropean genetics in the male chromosomes
Barley grown today on Gran Canaria is related to seeds
that Jacob Morales found in ancient granaries on the island.
suggests most Indigenous men were killed
or kept at the margins of society and left
fewer offspring, whereas women were
more likely to have children through inter-
marriage and sexual violence.
“Your genome is telling you a story
about your ancestors. But your identity is
defined by your experience, your family,
your friends,” says Fregel, who was born
and raised on Tenerife. When she tested
her DNA, it was about 20% Indigenous. “I
wouldn’t feel any less Canarian if my DNA
was from a whole other region.”
Still, her findings and those from Santana
and Morales have bolstered a modern-day
Canarian identity that emerged as the coun-
try moved beyond the nationalist conformity
enforced under Spanish dictator Francisco
Franco, who died in 1975. A year later,
when Morales was born, his uncle paid
his mother the equivalent of 60 euros
to give Morales a middle name evoking
an ancient Canarian leader: Bentejui.
Morales says his work can help
“build memory, and thus identity, for
the people who had very little histori-
cal documentation.” In one example,
Canarian chefs and beermakers have
reached out to the scientists about
using barley with DNA traced to the
ancient Canarian stock. “So much
knowledge was really lost when the
Spanish conquered the islands,” says
Jenny Hagenblad, a plant evolution
geneticist at Linköping University
who works with Morales. “It’s pretty
cool that we can break through this
loss of information and we can, in a
sense, eat the same things. It is a very
living link to the past.” j
science.org SCIENCE
 INSIGHTS
PERSPECTIVES
MATERIALS SCIENCE
Stabilizing 3D-printed metal alloys
A design strategy overcomes the strength-ductility trade-off in alloy manufacturing
By Lai-Chang Zhang1 and Jincheng Wang1,2
T
hree-dimensional (3D) printing, also
known as additive manufacturing,
continues to reshape industries, in-
cluding metals production. Among its
advantages are decreasing the time and
costs for creating intricate metal parts
and increasing customization. However, the
technology still faces challenges in achieving
uniform mechanical properties in 3D-printed
metallic alloys. On page 639 of this issue,
586 9 FEBRUARY 2024 • VOL 383 ISSUE 6683
Zhang et al. (1) report a design strategy for
printing a robust titanium alloy. The authors
show that the addition of molybdenum (Mo)
to the powder metal mixture enhances phase
stability and improves the strength, ductility,
and uniformity of tensile properties of the
3D-printed alloy. The approach could poten-
tially be applied to other powder mixtures
and enable the customization of different al-
loys with enhanced properties.
In the layer-by-layer 3D printing process
(typically with a high cooling rate of ~103 to
10 µm
108 K/s), a substantial thermal gradient forms
near the edge and bottom of the melt pool,
where metal powder has been melted by a
laser beam. The thermal gradient induces
epitaxial grain growth along the interface
between the freshly melted material and
the underlying solid material, with grains
growing toward the melt pool center. Cycles
of heating and partial remelting during the
printing of multiple layers ultimately results
in the formation of large columnar grains
and heterogeneously distributed phases,
science.org SCIENCE
A titanium alloy powder decorated with molybdenum
particles in this scanning electron microscope image
exhibits uniform tensile properties when 3D printed.
both of which are undesirable because they
can lead to nonuniform (anisotropic) and
compromised mechanical properties (2, 3).
Titanium alloys are among the strongest
metallic materials. In engineering applica-
tions at ambient temperature, a suitable
titanium alloy typically exhibits a tensile
elongation (the maximum stretch or defor-
mation that a material can withstand before
breaking) ranging from ~10 to ~25%, which
reflects good material reliability. Although
greater elongation (ductility) facilitates eas-
ier formability and holds priority in certain
applications, an increased strength within
this elongation range is preferred for endur-
ing mechanical loads. In both conventional
and additive manufacturing techniques for
processing metallic materials, a trade-off
between strength and ductility has been
prevalent (4, 5).
There are several strategies generally
used to improve the strength and ductility
of 3D-printed alloys. These include optimiz-
ing alloy design, process control, fine-grain
boundary strength, and grain microstruc-
ture modification. They also encompass
inhibiting undesirable (brittle) phases, in-
troducing second phases, and adding post-
processing treatments. Recent research that
addresses the problems of columnar crystals
and undesirable phases has focused on the
in situ incorporation of elements to modify
microstructures and phase compositions.
This approach also promotes the formation
of equiaxed crystals—structures with grains
of roughly equal dimensions along longitu-
dinal and transverse axes. This distinctive
in situ alloying ability of 3D printing offers
a promising avenue to overcome the en-
during compromise between strength and
ductility, particularly in techniques such as
powder bed fusion and directed energy de-
position (5–8). The powder bed fusion pro-
cess uses thermal power to selectively fuse
regions of a metal powder applied as a thin
layer on a platform. The directed energy de-
position method involves the coaxial feed
of metal powder by inert gas from a nozzle
while simultaneously undergoing melting
by a focused energy source before being de-
posited onto a surface.
There has been exploration of grain
morphologies and mechanical properties
when diverse elements are added to alloys
for 3D printing (7–9). For example, nanoc-
1
Centre for Advanced Materials and Manufacturing,
School of Engineering, Edith Cowan University, Perth, WA,
Australia. 2School of Engineering, The University of Western
Australia, Perth, WA, Australia. Email: l.zhang@ecu.edu.au;
lczhangimr@gmail.com
SCIENCE science.org
eramic zirconium hydride particles that
were incorporated into nonprintable alu-
minum alloys yielded materials that were
printable and noncracking, with a refined
equiaxed grain microstructure and tensile
properties comparable to those of wrought
counterparts (10). However, for titanium al-
loys, commercially available grain refiners
often have limited effectiveness on grain
structure. The refinement mechanism of
titanium alloys, especially the columnar-to-
equiaxed transition during solidification in
3D printing, has been studied extensively,
but efficiency limitations persist. Attempts
to overcome this obstacle include variations
in processing parameters (5), high-intensity
ultrasound application (11), the introduc-
tion of a desired heterostructure through
alloy design (7), the addition of solutes as
“The approach could…enable the
customization of different alloys
with enhanced properties.”
grain refiners for heterogeneous nucleation
sites (6), and the incorporation of solutes
with a high supercooling capacity, such
as b-eutectoid stabilizer elements [copper
(Cu), iron (Fe), chromium (Cr), cobalt (Co),
and nickel (Ni)], which have limited solubil-
ity in titanium (12).
Instead of using b-eutectoid stabilizer
elements, which may lead to the forma-
tion of brittle intermetallic eutectoids in
titanium alloys, Zhang et al. used Mo from
the b-isomorphous family [which includes
niobium (Nb), tantalum (Ta), and vana-
dium (V)] for in situ alloying of a titanium-
aluminum-molybdenum-vanadium-chro-
mium alloy (Ti-5Al-5Mo-5V-3Cr+5Mo). In
situ alloying precisely delivered Mo into
the melt pool, acting as seed nuclei for
crystal formation and refinement during
each layer scan. The Mo additive facilitated
the transition from large columnar grains
to fine equiaxed and narrow columnar
grain structures. Mo also stabilized the de-
sired b phase and inhibited the formation
of phase heterogeneities during thermal
cycling. The strategy yielded exceptional
uniformity with a concurrent ~926-MPa
yield strength and ~26% ductility.
Earlier studies described the use of ad-
ditive elements that were distinct from the
constituents of the parent prealloyed pow-
der. However, in the case of Ti-5Al-5Mo-
5V-3Cr+xMo (where x = 5 wt %), Zhang et
al. added extra Mo to the parent powder,
achieving a total Mo weight percentage of
10%. This minimized chemical complexity,
achieved columnar-to-equiaxed transition
in grain morphology, enhanced desired
phase stability, and achieved isotropy in the
printed metal. Notably, using the prealloyed
powder alone with the same composition
did not yield comparable printing results.
This is unsurprising given earlier observa-
tions. For example, Ti-Nb alloys processed
through in situ alloying yielded distinct
microstructures and properties compared
with using prealloyed powder alone (13).
Similarly, a 3D-printed alloy with preal-
loyed Ti-6Al-4V-5Cu powder showed acicu-
lar a´ grains within the columnar b grains,
resulting in deteriorated ductility (14).
Zhang et al. did, however, identify undis-
solved Mo particles in the microstructure,
and their potential effects are unknown.
Indeed, the random presence of undissolved
particles in the in situ alloying strategy has
raised concerns related to mechanical and
corrosion properties (15). For example, the
full melting of in situ alloyed additive par-
ticles may require higher energy, and over-
heating could lead to an altered microstruc-
ture and poorer mechanical properties.
Also, the dynamic fatigue and corrosion
performance resulting from undissolved ad-
ditive Mo particles are unknown. Although
postprinting heat treatment can eliminate
undissolved particles, it might alter the mi-
crostructure, potentially jeopardizing the
mechanical properties (9).
Zhang et al.’s proposed design strategy
opens an avenue for exploring different
metal powder feedstocks, diverse printable
alloy systems, and different 3D printing
techniques as well as for advancing mul-
timaterial printing. It also enables the in-
hibition of columnar grain formation and
prevents undesirable phase heterogeneities.
These problems arise as the result of varied
heat profiles, which are influenced by print-
ing parameters for each powder. The strat-
egy also overcomes the strength-ductility
trade-off in the as-printed state, minimizing
the need for postprinting treatment. These
advantages will undoubtedly trigger ripples
in the realm of 3D printing. j
RE FE REN CES A ND N OT ES
1. J. Zhang et al., Science 383, 639 (2024).
2. J. Wang, R. Zhu, Y. Liu, L. Zhang, Adv. Powder Mater. 2,
100137 (2023).
3. D. Gu et al., Science 372, eabg1487 (2021).
4. T. Yang et al., Science 362, 933 (2018).
5. H. Y. Ma et al., J. Mater. Sci. Technol. 183, 32 (2024).
6. B. Vrancken, L. Thijs, J.-P. Kruth, J. Van Humbeeck, Acta
Mater. 68, 150 (2014).
7. T. Zhang et al., Science 374, 478 (2021).
8. P. Kürnsteiner et al., Nature 582, 515 (2020).
9. J. Wang et al., J. Mater. Sci. Technol. 61, 221 (2021).
10. J. H. Martin et al., Nature 549, 365 (2017).
11. C. J. Todaro et al., Nat. Commun. 11, 142 (2020).
12. M. S. K. K. Y. Nartu et al., Nat. Commun. 14, 3288 (2023).
13. J. C. Wang et al., J. Mater. Sci. Technol. 105, 1 (2022).
14. K. Li et al., Acta Mater. 256, 119112 (2023).
15. J. C. Wang et al., Mater. Sci. Eng. A 760, 214 (2019).
10.1126/science.adn6566
9 FEBRUARY 2024 • VOL 383 ISSUE 6683 587
I NS I GHTS | P E R S P E C T I V E S
PHYSIOLOGY
Fibroblasts enable penile erection
Perivascular fibroblasts may underlie erectile dysfunction
By Ji-Kan Ryu1 and Gou Young Koh2
S
exual well-being in men depends on
the ability to attain penile erection,
which can be compromised by ag-
ing and chronic conditions, including
diabetes and atherosclerosis (1). Pe-
nile erection results from acetylcho-
line- and nitric oxide–induced dilation of the
supplying arteries and sponge-like corpora
cavernosa tissue and relaxation of smooth
muscle (2). The extent and duration of pe-
nile erection is determined by the balance
between these vasodilators and the vasocon-
strictor norepinephrine (3, 4). On page 604 of
this issue, Linck Guimaraes et al. (5) report
that perivascular fibroblasts expressing the
high-affinity amino acid transporter SLC1A3
(solute carrier family 1, member 3) are essen-
tial for penile erection in mice because they
reduce norepinephrine availability, thereby
promoting dilation of the corpora cavernosa.
The number of SLC1A3+ perivascular fibro-
blasts decreased in aged mice, which reduced
penile blood flow. Because the mechanisms of
penile erection are similar in mice and hu-
mans (4), these findings may be relevant to
erectile dysfunction in aged men.
Linck Guimaraes et al. report that in their
transgenic mouse model, SLC1A3+ fibroblast
proliferation increased by inhibiting Notch
signaling during recurrent erection, which
led, in extreme cases, to ischemic priapism
(a condition characterized by persistent and
painful erection with high rigidity of the
corpora cavernosa). By contrast, continuous
Notch activation reduced SLC1A3+ fibroblast
numbers and penile blood flow. These
fibroblasts are mainly distributed along
blood vessels and smooth muscles in the
corpora cavernosa. The authors show that
SLC1A3+ fibroblasts take up norepinephrine
through SLC6A2 and degrade it through the
activity of intercellular monoamine oxidase
A, which reduces norepinephrine availability.
Most treatments for erectile dysfunction
promote nitric oxide–mediated relaxation
of vascular smooth muscle in the corpora
cavernosa (6). Sildenafil, tadalafil, vardenafil,
and other inhibitors of phosphodiesterase
type 5 (PDE5) augment the smooth muscle–
1
Department of Urology and Medical Research Center for
Controlling Intercellular Communications, Inha University
School of Medicine, Incheon, Republic of Korea. 2Center for
Vascular Research, Institute for Basic Science, Daejeon,
Republic of Korea. Email: gykoh@kaist.ac.kr
588 9 FEBRUARY 2024 • VOL 383 ISSUE 6683
relaxing effects of nitric oxide by blocking
the degradation of the downstream signaling
messenger cyclic guanosine monophosphate
(cGMP) into GMP (6, 7). However, despite
their widespread use, PDE5 inhibitors fail
to restore an erection in up to 30% of men,
leaving limited treatment options (7).
In search of effective alternatives, clinical
trials using melanocortin or dopamine
receptor agonists to target central pathways
or soluble guanylate cyclase activators, Rho-
kinase inhibitors, or potassium channel
openers to target peripheral pathways have
not shown sufficient benefit to gain approval
for this use. This failure is largely due to lack
of greater efficacy or fewer side effects than
already approved PDE5 inhibitors (8).
Although Linck Guimaraes et al. did
not examine humans, their study reveals
a new therapeutic paradigm of creating
conditions that increase norepinephrine
uptake or decrease Notch signaling in
penile perivascular fibroblasts, which could
New therapeutic prospects for intractable erectile dysfunction
Fibroblasts expressing solute carrier family 1, member 3 (SLC1A3+) take up norepinephrine through the
SLC6A2 transporter, which results in its degradation and promotes penile erection in mice. Thus, various
potential approaches for treating refractory erectile dysfunction in humans are revealed.
Cross section of human penis
Corpora
cavernosa
Cavernous
artery
Corpus
spongiosum
Urethra
Vascular
smooth
muscle
Endothelial cell
SLC1A3+
fibroblast
translate into treating erectile dysfunction
in patients who are unresponsive to PDE5
inhibitors. Thus, multiple strategies could
be explored by further preclinical testing
and, if promising, evaluated clinically. One
approach would be to inhibit Notch signaling
in SLC1A3+ fibroblasts of the corpora
cavernosa (see the figure) with available
Notch inhibitors. A potential limiting factor
is that local administration of such inhibitors
would also affect endothelial cells, smooth
muscle cells, neurons, and other cell types in
which Notch signaling is involved in penile
erection but differs mechanistically from
Notch signaling in fibroblasts.
Other approaches could include increasing
expression of the norepinephrine transporter
SLC6A2 in SLC1A3+ perivascular fibroblasts
of the corpora cavernosa. However, the
poorly understood regulatory mechanisms
of SLC6A2 expression must be characterized.
Another approach would be to implant
autologous SLC1A3+ fibroblasts in the
corpora cavernosa. However, still unknown is
whether the number of implanted SLC1A3+
fibroblasts that become associated with
the corpora cavernosa would be adequate
to have sustained SLC6A2 norepinephrine
uptake activity and be functionally sufficient
to correct erectile dysfunction. Furthermore,
a neural or mechanical stimulator that
1 Inducing SLC6A2 expression
SLC6A2
transporter
2 Notch inhibition
Proliferation
Inhibited Notch
3 Implantation of autologous
SLC1A3+ fibroblasts
4 Mechanical stimulation
Proliferation
science.org SCIENCE
expands the number of perivascular SLC1A3+
fibroblasts by increasing erection frequency
could be developed. Efficacy would depend
on determining the yet-to-be-identified
stimulation parameters for promoting the
proliferation of SLC1A3+ fibroblasts that
express SLC6A2.
These new approaches will require rigorous
preclinical and clinical testing to translate
observations made in transgenic mice into
therapies that are safe and effective in men.
Additional studies would also be necessary
to characterize the heterogeneity of corpora
cavernosa fibroblasts in diverse patients with
erectile dysfunction, where penile fibrosis,
endothelial cell injury, smooth muscle
transdifferentiation, and other abnormalities
can be contributing factors (9, 10). These
challenges are, nonetheless, balanced by
the drawbacks of available alternatives that
involve invasive procedures such as injection
of vasodilators into the corpora cavernosa or
surgical implantation of a penile prosthesis.
Possible new interventions should not
overshadow evidence that regular physical
activity can improve erectile function through
increased expression and activity of nitric
oxide synthase, strengthened endothelial
cell function, and reduced stress and anxiety
(11). Exercise and lifestyle changes could
increase the number of perivascular SLC1A3+
fibroblasts and their expression of SLC6A2.
These changes are a natural consequence of
frequent penile erections but are suppressed
when erections are infrequent. Although
erections could be necessary to maintain
perivascular SLC1A3+ fibroblasts, it is unclear
whether this requires regular sexual activity
or if nocturnal penile tumescence is sufficient.
If sedentary behavior and low sexual activity
leads to fewer of these fibroblasts and
greater corpora cavernosa sensitivity to
norepinephrine, then regular exercise, sexual
activity, and other lifestyle changes should be
helpful for sexual dysfunction. j
RE F ER E NC ES AND NOTES
1. I. Goldstein et al., Int. J. Clin. Pract. 72, e13078 (2018).
2. K. J. Hurt et al., Proc. Natl. Acad. Sci. U.S.A. 99, 4061
(2002).
3. S. M. MacDonald, A. L. Burnett, Urol. Clin. North Am. 48,
513 (2021).
4. K. E. Andersson, Pharmacol. Rev. 63, 811 (2011).
5. E. Linck Guimaraes et al., Science 383, eade8064 (2024).
6. I. Goldstein et al., N. Engl. J. Med. 338, 1397 (1998).
7. A. Samidurai et al., Annu. Rev. Pharmacol. Toxicol. 63, 585
(2023).
8. J. K. Ryu et al., Transl. Androl. Urol. 6, 207 (2017).
9. L. Zhao et al., Nat. Commun. 13, 4302 (2022).
10. D. Fang et al., Front. Endocrinol. 13, 874915 (2022).
11. M. S. Allen, Nat. Rev. Urol. 16, 553 (2019).
AC KN OW LE D GMENTS
The authors receive support from the Republic of Korea
Ministry of Science and Technology to the Medical Research
Center (NRF-2021R1A5A2031612 to J.-K.R.) and the Institute
of Basic Science (IBS-R025-D1-2015 to G.Y.K.).
10.1126/science.adn5182
SCIENCE science.org
PLANT SCIENCE
Time for growth
Plants measure the duration of metabolic activity
to promote rapid growth in long days
By Christopher R. Buckley and
Michael J. Haydon
D
aylength varies across the year and
provides a cue for plants to respond
to changing seasons. Photoperiod-
ism, the ability to measure day-
length, is critical for fitness and
survival because of its influence on
plant physiology and behavior. The timing
of numerous developmental processes is
controlled by daylength, including flowering,
growth, dormancy, and senescence (1). Thus,
photoperiodism has important implications
for the regional adaptation of crops and
migration of plant populations in response
to climate change (2). Seasonal control of
flowering occurs by a well-defined molecular
pathway that measures photoperiod as the
duration of light, as detected by photorecep-
tors. On page 605 of this issue, Wang et al.
(3) describe a new mechanism of daylength
measurement that detects the duration of
photosynthetic output and is required for
rapid growth during long summer days. The
study demonstrates that plants use multiple,
independent mechanisms to measure day-
length that can differentially influence dis-
tinct developmental processes.
The transcription factor CONSTANS (CO)
plays a key role in photoperiodic flowering.
It activates the expression of FLOWERING
LOCUS T (FT), which encodes a mobile pro-
tein that promotes flowering in the shoot
apical meristem (1). Daily oscillations of
CO transcription are controlled by the cir-
cadian clock, a biological timekeeper that
is set by light signals at dawn in a process
called entrainment. In Arabidopsis thali-
ana, flowering is induced in long days be-
cause CO protein is stabilized by light in
the afternoon. By contrast, in flowering
plants such as rice, CO activity is promoted
by a short photoperiod.
The CO-FT module also controls other pho-
toperiodic processes, including tuberization
in potatoes, bud dormancy in perennials, and
seed size in annuals (4–6). However, shifting
daylength leads to substantial changes in the
profile of transcribed genes, indicating that
there might be many more processes under
photoperiodic control (7). Because relatively
School of BioSciences, University of Melbourne, Parkville,
VIC, Australia. Email: m.haydon@unimelb.edu.au
few genes are regulated by CO, this suggests
that the other genes are controlled by CO-
independent mechanisms.
Metabolic signals have been implicated in
biological timekeeping in several ways (8).
Sugars produced from photosynthesis con-
tribute to entrainment of the Arabidopsis
circadian clock in a manner different from
light, acting as a “metabolic dawn” (9). Snf1
RELATED PROTEIN KINASE 1 (SnRK1) and
TARGET OF RAPAMYCIN (TOR) are ma-
jor metabolic signaling complexes in plants
that influence the pace of circadian rhythms
(10, 11), and metabolic signals such as reac-
tive oxygen species modify circadian gene
expression in a time-of-day–specific manner
(12, 13).
Metabolic signals also measure and com-
municate changes in daylength. A metabolic
daylength measuring system has been de-
scribed that is specifically required to con-
trol growth in the short days of winter (14).
Expression of the circadian-regulated gene
PHLOEM PROTEIN 2-A13 (PP2-A13) is in-
duced after dusk in short days and is re-
quired to maintain winter growth but has
no effect on growth in long days. The day-
length-dependent regulation of PP2-A13
expression is controlled by the timing of
photosynthetic activity. Supplying sugar
can suppress its expression in short days,
whereas inhibiting photosynthesis can ac-
tivate its expression in long days.
Wang et al. describe a distinct mecha-
nism of metabolic daylength measurement
that operates in long days and is required
to promote rapid growth in these condi-
tions. In Arabidopsis, this mechanism
functions in parallel to, but independently
of, canonical measurement of absolute day-
length by photoreceptors and the CO-FT
module (see the figure). MYO-INOSITOL-
1-PHOSPHATE SYNTHASE 1 (MIPS1) has
been identified as a necessary functional
component of this pathway. MIPS1 expres-
sion is induced specifically during long
days and, opposite to pp2-a13 mutants,
mips1-2 mutants grow slowly in long days
but are unaffected in short days. By modi-
fying the timing, duration, and intensity of
light, Wang et al. observed that MIPS1 ex-
pression and rapid growth depend on pho-
tosynthetic activity in the afternoons of
long summer days. Because photoreceptors
are activated at much lower light quantity
9 FEBRUARY 2024 • VOL 383 ISSUE 6683 589
I NS I GHTS | P E R S P E C T I V E S
than is required for net carbon fixation by
photosynthesis, plants simultaneously detect
a different daylength by these two mecha-
nisms. Thus, the authors propose the con-
cept of a photosynthetic period.
How the photosynthetic period is mea-
sured is unclear but could involve either
metabolic signaling mechanisms that are
already implicated in timekeeping or oth-
erwise unknown pathways (8). Moreover, it
is not clear whether the mechanisms that
control MIPS1 and PP2-A13 expression are
the same or distinct. The enzymatic func-
tion of MIPS1 is required for the metabolic
daylength measuring system because myo-
inositol can rescue the mutant phenotype
but is not sufficient because alone it can-
not induce rapid growth in short days (3).
Although other factors are apparently re-
quired, the discovery of MIPS1 as a marker
and essential component of metabolic
daylength measurement presents an excit-
ing opportunity to identify upstream and
downstream components of the pathway.
The existence of parallel photoperiodic
pathways that detect different signals
raises interesting questions about how pro-
cesses are regulated in natural conditions,
which can be highly variable and unpredict-
able. Furthermore, it now seems likely that
there are numerous mechanisms that could
measure different aspects of daylength.
Because photoperiodic flowering and sea-
sonal growth are controlled by independent
pathways, this opens up the possibility of
specifically targeting photoperiodic growth
for crop improvement. Although long
days induce both flowering and growth in
Arabidopsis and temperate crops such as
wheat and barley, tropical crops such as rice
and soybean preferentially flower in short
days. The existence of independent path-
ways could facilitate different photoperiod
Multiple parallel daylength measurement systems in plants
Absolute photoperiod is the time between dawn and dusk, is detected by photoreceptors, and controls
timing of flowering. Photosynthetic period is the duration of light required to drive net carbon fixation, does
not precisely overlap with absolute photoperiod, and controls seasonal growth.
Dawn
Light
intensity
Photosynthetic period
590 9 FEBRUARY 2024 • VOL 383 ISSUE 6683
responsiveness for these processes in differ-
ent plant species. Photoperiod-insensitive
flowering has been frequently selected for
during the domestication and expansion of
crops into different spatial and temporal
environments (15). It therefore should be
possible to engineer photoperiod-insensi-
tive growth in crops, which could be a use-
ful trait for rapid biomass production in
short growing seasons, especially as climate
change shifts present environmental op-
tima (2). Furthermore, defining critical time
windows to trigger photoperiodic growth
could enable energy-efficient regimes to
maximize growth in artificial cropping en-
vironments. Therefore, future progress in
understanding metabolic daylength mea-
surement in crops presents opportunities
for agriculture and aligns with the emerg-
ing concept of chronoculture (15). j
R E F E R E N C ES A N D N OT ES
1. J. M. Gendron, D. Staiger, Annu. Rev. Plant Biol. 74, 481
(2023).
2. L. L. Sloat et al., Nat. Commun. 11, 1243 (2020).
3. Q. Wang, W. Liu, C. C. Leung, D. A. Tarté, J. M. Gendron,
Science 383, eadg9196 (2024).
4. C. Navarro et al., Nature 478, 119 (2011).
5. H. Böhlenius et al., Science 312, 1040 (2006).
6 B. Yu et al., Nat. Plants 9, 343 (2023).
7. C. C. Leung et al., PLOS Biol. 21, e3002283 (2023).
8. C. R. Buckley et al., Curr. Opin. Plant Biol. 73, 102333
(2023).
9. M. J. Haydon et al., Nature 502, 689 (2013).
10. A. Frank et al., Curr. Biol. 28, 2597 (2018).
11. N. Zhang et al., Proc. Natl. Acad. Sci. U.S.A. 116, 25395
(2019).
12. A. G. Lai et al., Proc. Natl. Acad. Sci. U.S.A. 109, 17129
(2012).
13. Á. Román et al., Proc. Natl. Acad. Sci. U.S.A. 118,
e2020646118 (2021).
14. W. Liu et al., Dev. Cell 56, 2501 (2021).
15. G. Steed et al., Science 372, eabc9141 (2021).
AC K N OW L E D G M E N TS
M.J.H. receives research funding from the Grains Research
and Development Corporation.
10.1126/science.adn5189
Absolute photoperiod Dusk
Cloud cover
CANCER
Transforming
lung cancer
types
Lung cancer cells can
escape targeted therapy
by switching oncogenic
drivers and cell identity
By Anton Berns
T
here are two main classes of lung
cancer: non–small-cell lung cancer,
which is predominantly the lung
adenocarcinoma (LUAD) subtype
that originates from alveolar type 2
(AT2) cells, and small cell lung can-
cer (SCLC), for which pulmonary neuro-
endocrine cells (PNECs) are the main cell
of origin. SCLC has the poorest prognosis,
killing over 200,000 patients each year
worldwide. Remarkably, ~10% of patients
that present with LUAD driven by mutated
epidermal growth factor receptor (EGFR)
and receiving EGFR targeted therapy be-
come refractory to the treatment through
histological transformation (HT) into
SCLC (1). A similar conversion to neuroen-
docrine tumors is occasionally observed in
patients with prostate adenocarcinoma re-
ceiving androgen deprivation therapy (2).
The drivers of these HTs and the transition
stages involved have remained an enigma.
On page 603 of this issue, Gardner et al.
(3) identified the consecutive stages LUAD
cells go through in HT to SCLC in mice,
which helps to uncover this difficult-to-
study conversion in human cancer.
Gardner et al. generated ERPMT mouse
strains with various configurations of condi-
tional alleles of the retinoblastoma Rb (R) and
Trp53 (P) genes, which are both implicated
in nearly all SCLCs, with an oncogenic Myc-
Thr58 Ala (MycT58A; M) and a doxycycline-
¢
inducible EGFR-Leu858 Arg (EGFRL858R; E)
¢
oncogene, supplemented with a cell-trace (T)
marker. Cell-type–specific Cre drivers were
used to switch on these alleles in the various
lung cell lineages, enabling the authors to as-
sess the contribution of these alleles to LUAD
and SCLC as well as their involvement in HT
from LUAD to SCLC (T-SCLC). Single-cell
RNA sequencing was used to characterize the
Division of Molecular Genetics, Netherlands Cancer
Institute, Amsterdam, Netherlands. Email: a.berns@nki.nl
science.org SCIENCE
 cell types emerging during this process (see Lung cancer cell lineages
the figure). Cell tolerance to specific oncogenic drivers controls the conversion of lung adenocarcinoma
It is well recognized that a limited set of (LUAD) to small cell lung cancer (SCLC). In mice, multiple cell types can give rise to tumors
oncogenic drivers are usually associated with that resemble SCLC (T-SCLC). Loss of tumor suppressors Rb and Trp53 (RP) is required,
particular tumors. This is also the case for but strong EGFR signaling is incompatible with SCLC development, independent of the cell
transformation of AT2 cells into LUAD and of origin. Oncogenic MYC is a potent driver of T-SCLC development. PI3K signaling (loss of
PNECs into SCLC. However, Gardner et al. Pten) and loss of Rb facilitate AT2 cell transition to T-SCLC and to tumors with basal cell–like
now show that AT2 cells and PNECs do not features. Oncogenic MYC without EGFR or PI3K signaling is not tolerated by AT2 cells. EGFR
tolerate particular oncogenic drivers and nei- signaling blocks T-SCLC development from both AT2 and pulmonary endocrine cells. Cancer
ther do the neuroendocrine tumors that are
Cell of origin LUAD Residual disease Lung cancer type
derived from them. The HT from LUAD to
T-SCLC is notable in this respect: This trans- Toxic
AT2 cell MYC MYC MYC
formation not only necessitates the acquisi- LUAD
+EGFR –EGFR +EGFR
tion of new oncogenic driver lesions (e.g.,
MYC –EGFR MYC
loss of the tumor suppressors RB and p53, –RP
T-SCLC
gain of MYC hyperactivation) that are largely
MYC –Pten
absent in LUAD, it also requires the down- Basal cell–like
regulation of an oncogenic driver that is in- Basal Pulmonary MYC T-SCLC +
?
dispensable for LUAD (mutant EGFR). The cell–like neuroendocrine cell –RP LCNEC
–Pten
authors found that the underlying lineage
(MYC)
switch that accompanies this HT appears to SCLC
–RP
proceed through a bottleneck consisting of
phenotypically different cell types including AT2, alveolar type 2; EGFR, epidermal growth factor receptor; LCNEC, large cell neuroendocrine carcinoma; PI3K, phosphatidylinositol 3-kinase;
undifferentiated basal cell–like cells express- Pten, phosphatase and tensin homolog; Rb, retinoblastoma

ing target genes of the transcription factors

MYC and SOX2, reminiscent of epigenetic re-
programming. Apparently, dedifferentiation
to a progenitor basal cell–like intermediate
is required for this HT. This aligns well with
the observation of Gardner et al. that keratin
5 (KRT5)–expressing lung basal cells—which
already exhibit progenitor features—can be
easily transformed into T-SCLC by inactiva-
tion of RB and p53.
The sensitivity of a cell lineage to specific
oncogenic drivers is not absolute but depends
on the state the cell is in. For example, AT2
cells do not tolerate expression of MYCT58A
without concomitant EGFRL858R or phospha-
tidylinositol 3-kinase (PI3K) pathway activa-
tion [such as loss of phosphatase and tensin
homolog (Pten)], whereas PNECs and even
SCLCs do not tolerate EGFRL858R either with
or without MYCT58A. This lineage-specific in-
tolerance is a widely observed phenomenon.
To drive tumorigenesis, both the nature and
flux (strength) of oncogenic signaling have to
be in tune with the specific gene expression
configuration of a cell. Otherwise, oncogenic
signals might become toxic to cells, e.g., by
inducing replication stress.
High MYC expression is not only toxic to
AT2 cells, it is also detrimental for mouse
embryo fibroblasts in vitro, causing apopto-
sis (4). This can be relieved by coexpression
of oncogenic (mutant) KRAS, similar to the
blunting of MYCT58A toxicity by EGFRL858R or
Pten loss in AT2 cells. Although MYCT58A ex-
GRAPHIC: A. FISHER/SCIENCE
cogenic KRAS-induced LUAD development
is impaired by Myc haploinsufficiency (6).
Additionally, AT2 cells do not tolerate well
oncogenic KRAS signaling on its own. The
activation of oncogenic KRAS in AT2 cells in
vivo leads to a widespread senescence-like ar-
rest with only a small fraction of AT2 cells
that are able to express oncogenic KRAS and
develop into LUADs in mice (7).
The intolerance of PNECs to the expres-
sion of EGFRL858R is even more striking be-
cause Gardner et al. found that its expression
in ERPMT mice blocks SCLC development
from PNECs. The presence of EGFRL858R is
also unable to force the reverse HT from
SCLC to LUAD. This aligns with previous
work showing that established mouse SCLC
cell lines respond to forced oncogenic KRAS
expression by entering into crisis with the
spurious emergence of cells with a LUAD-like
phenotype (8). In this context, the observa-
tion that inhibition of the mitogen-activated
protein kinase (MAPK) pathway, which is
downstream of KRAS, with trametinib does
not relieve the inhibitory effect of EGFRL858R
expression on PNECs is unexpected. This
suggests that critical EGFR signaling beyond
the MAPK pathway is not compatible with
the PNEC lineage. It is unlikely that this en-
tails PI3K signaling because its activation ac-
celerates SCLC development.
The observation that fibroblast growth
factor (FGF), also a strong inducer of MAPK
tant impaired SCLC development induced
by loss of RB and p53 in PNECs, whereas
it promoted development of T-SCLC when
targeted to KRT14+ bronchial epithelial cells
(12). These seemingly discordant observa-
tions might be reconciled by postulating that
different cells of origin, as a result of a differ-
ent epigenome, require or tolerate dissimilar
oncogenic driver configurations for T-SCLC
development. This could also help explain
the actual diversity seen in lung neuroendo-
crine tumors (13) with their diverse prolifera-
tion and metastatic characteristics (14). This
can be tested experimentally using the ap-
proaches described by Gardner et al.
The predominant driver role of MYC in
SCLC and T-SCLC, as well as the depen-
dence of LUADs on MYC, again emphasizes
the promise of effective MYC inhibitors for
treating lung tumors and other cancers (5,
15). The study by Gardner et al. is in many
respects highly illuminating, showing how
thorough mouse studies can reveal new
cancer biology. j
RE F E REN C ES AN D N OT ES
1. S. Sivakumar et al., Cancer Discov. 13, 1572 (2023).
2. J. O. Mori et al., Prostate 82, 1005 (2022).
3. E. Gardner et al., Science 383, eadj1415 (2024).
4. G. I. Evan et al., Cell 69, 119 (1992).
5. L. Soucek et al., Nature 455, 679 (2008).
6. N. M. Sodir et al., Nat. Commun. 13, 6782 (2022).
7. C. Guerra et al., Cancer Cell 4, 111 (2003).
8. J. Calbo et al., Cancer Cell 19, 244 (2011).
9. D. W. Shia et al., Oncogene 42, 434 (2023).

10. M. C. Kwon et al., Genes Dev. 29, 1587 (2015).

pression appears toxic to AT2 cells, EGFRL858R-
induced LUAD from AT2 cells is accelerated
by MYCT58A. Indeed, it appears that LUAD
development even requires MYC and can
be eliminated by repressing MYC’s function
(5), consistent with the observation that on-
signaling, enhances SCLC growth (9) and po- 11. K. Ishioka et al., Cancer Res. 81, 3916 (2021).
tentiates its metastatic dissemination (10), as 12. G. Ferone et al., Cell Rep. 30, 3837 (2020).
well as facilitating the HT of LUAD to SCLC 13. J. W. Park et al., Science 362, 91 (2018).
14. D. Yang et al., Cancer Discov. 10, 1317 (2018).
(11), adds to the confusion, although this 15. T. E. Speltz et al., Nat. Biotechnol. 41, 541 (2023).
could also relate to signaling flux. However,
an oncogenic FGF receptor 2 (FGFR2) mu- 10.1126/science.adn5218

SCIENCE science.org 9 FEBRUARY 2024 • VOL 383 ISSUE 6683 591

I NS I GHTS
P OLICY FORUM
EDUCATION
Teach Indigenous knowledge alongside science
Evidence supports the teaching of Indigenous knowledge alongside sciences in the classroom
By Amanda Black1,2 and Jason M. Tylianakis2,3
C
onflict has grown around Indigenous
knowledge in education policy. There
has been growing acceptance of the
value of Indigenous knowledge for
promoting ecological resilience,
transformational approaches in
stewardship, and cultural renewal within
global fora such as the Intergovernmental
Panel on Climate Change. However, despite
increasing acceptance at a strategic high level
in science-informed policy, there is often a
lack of wider acceptance, application, and
policy protections of Indigenous knowledge
transmission in more local settings,
including opposition by some scientists.
We argue that Indigenous knowledge can
complement and enhance science teachings,
benefitting students and society in a time of
considerable global challenges. We do not
argue that Indigenous knowledge should usurp
the role of, or be called, science. But to step
from “not science” to “therefore not as (or at
all) valuable and worthy of learning” is a non
sequitur, based on personal values and not a
scientifically defensible position.
The current state of global systems in an
uncertain risk landscape creates an urgent
need for many knowledges and approaches
to build resilience and prosperity of com-
munities. One attempt to provide policy
protections and opportunities for Indig-
enous knowledge is the Aotearoa–New
Zealand government’s decision to ensure
that Indigenous knowledge (Mātauranga
Māori) has equal value with other bodies
of knowledge in the school curriculum, af-
ter lengthy advocacy from Māori educators
to honor the Treaty of Waitangi, Aotearoa–
New Zealand’s founding document. This
policy has precipitated a battle of rhetoric
among researchers and scientists; initial
condemnation of the policy by a group
of academics argued that unlike science,
Indigenous knowledge is inadequately
equipped to provide empirical evidence
of universal truths (1), which resulted in
1Faulty of Agriculture and Life Sciences, Lincoln University,
Lincoln, New Zealand. 2Bioprotection Aotearoa,
Canterbury, New Zealand. 3School of Biological Sciences,
University of Canterbury, Christchurch, New Zealand.
Email: amanda.black@lincoln.ac.nz
592 9 FEBRUARY 2024 • VOL 383 ISSUE 6683
a signed open letter by 2000 academics
and public figures in support of the policy
that includes Mātauranga Māori. Since
then, the Royal Society of New Zealand–Te
Apārangi (the premiere advocacy and advi-
sory body for science and humanities) has
been drawn into the debate, which contin-
ues from both sides. Those that support this
policy have largely argued on the grounds
of ethical responsibilities and moral view-
points, whereas those that oppose it cite di-
lution or, at worst, abandonment of science
that will lead to poor societal outcomes (1).
The considerable research effort on innova-
tion, Indigenous knowledge’s relationship
with science, and its pedagogy have not (to
our knowledge) been synthesized to address
this discussion, which is also pertinent to
efforts beyond New Zealand [such as recent
investment into Indigenous knowledge by
the US National Science Foundation (2)].
We suggest that many of the arguments
used to “defend” science by presenting
Indigenous knowledge as inferior are
themselves rooted in logical fallacies. We
also argue that the treatment of all Indigenous
knowledge as myth is at odds with the
literature, which emphasizes a continuum
from empirical and science-like aspects of
Indigenous knowledge to philosophical and
metaphysical ones (3). Teaching sociocultural
themes of a Māori worldview is already
encouraged in curriculum guidelines,
suggesting that objections are not to having
these aspects taught at school but rather to
giving them value in the context of knowledge.
Yet school curricula already include a range of
subjects across the arts and humanities that
do not meet criteria of science, and it would
be senseless to argue that they do not have
“equal value” with science.
Moreover, we argue that there is a cost to
rejecting Indigenous knowledge, in that
framing it with simplistic caricatures
misses the potential for complementarity
between science and Indigenous
knowledge. Concomitantly, we highlight
learning benefits that emerge when
students are well versed in multiple
knowledge systems. Last, we provide
evidence that science innovation may be
stifled if mainstream science is granted
sole dominion over knowledge generation.
COMPLEMENTARY EXPLORATION
Indigenous knowledge is often generated
empirically and drawn from local context,
complementing and challenging scientifi-
cally derived universal “truths.” These em-
pirical aspects of Indigenous knowledge
(which have been called the “know-how
versions of knowledge”) that emphasize
the method or workability [(3), p. 103]
align most closely with science but still
differ sufficiently from science in their
context and underpinning worldview that
many (but not all) scholars argue that sci-
ence and Indigenous knowledge are not
the same (3).
In addition to the more “science-
like” empirical components and their
framings that we discuss below, there are
components of Indigenous knowledge
that are metaphysical or philosophical
and entirely unrelated to science. These
latter components alone have been
emphasized by those who challenge the
validity of Indigenous knowledge, with
their criticisms likely fueled by recent
examples of Creationism being taught in
schools in place of evolutionary theory. Yet
this framing of Indigenous knowledge as
entirely “myth” is an “appeal to extremes”
argument, which would benefit from
a more mature and nuanced view of
Indigenous knowledge.
A key issue is that Western culture
discretizes knowledge generation into
disciplines, in which science cannot be
contaminated by nonscience, whereas such
divisions between methods of inquiry are
often absent from Indigenous knowledge
systems. However, we believe this only
becomes a problem when teaching
Indigenous knowledge “as” science, but
not “alongside” it, as articulated in the
Aotearoa–New Zealand policy.
By analogy, philosophy is not science
and includes inquiries around existence,
knowledge, and the self. Both the
questions asked and the methods of
enquiry frequently depart from what could
be called science. Moreover, philosophy
can reflect on the nature of science in a
way that science is unable to reflect on
itself. For this reason, students who study
philosophy alongside science can improve
science.org SCIENCE
 their learning both of science concepts and codeveloped conservation solutions. We the basis that they are corrected as part of
concepts about science (4). Attributing hope that viewing Indigenous knowledge the scientific process, in which knowledge
greater “value” to science than philosophy as complementary to science, without is updated as new information becomes
(or other nonscience endeavors such as art replacing nor being science, may lead to available.
or the humanities) would make little sense more nuanced and fruitful conversations Yet although Indigenous knowledge
and be an opinion based on values rather around policy in this space and to is also well known to be dynamic and
than scientific evidence. maximizing the benefits of such policy. continuously updated (7), critics do not
Similarly, we argue that teaching Yet despite all of this, the false dichotomy afford it an equal right to correct itself. For
Indigenous knowledge alongside science between the validity of Indigenous example, “pity the moas were all eaten”
should not seek to usurp science (in the knowledge and science-generated (1) is commonly used rhetoric to imply
way that, for example, creationism seeks knowledge persists and is frequently based the failure of Māori knowledge around
to undermine evolutionary theory because on a straw person. Science and Indigenous conservation of a giant endemic New
they are incompatible with one another), knowledge systems comprise distinct Zealand bird in the 15th century. Yet this
but rather it “provokes science, and can perspectives of understanding the world reasoning mistakenly conflates the validity
act as a mirror for science to see itself
more clearly, reflected in a philosophically
different form of knowledge” [(5), p. 87].
A parallel understanding of science and
Indigenous knowledge systems would
be complementary, emphasizing their
similarities and cultural differences; the
separation versus connection of empirical
and philosophical subjects would be one
example of those differences. Another
example specific to Aotearoa–New
Zealand would be that Te Ao Māori (Māori
worldview) uses an intergenerational lens
inclusive of the observer that gives cultural
integrity to questions and generated
outcomes, whereas the scientific method
strives to be disconnected from that which
it observes.
The timescales of knowledge genera-
tion are also complementary. For example,
short-duration scientific research funding
cycles can create institutional barriers to
long-term data acquisition and study of
large-scale (such as environmental) prob-
lems. By contrast, Indigenous knowledge
can and has contributed empirically gener-
ated, intergenerational knowledge, making
it an increasingly valuable tool in environ-
mental management, particularly around
rare but increasingly frequent natural
events such as large-scale deadly bush fires
that plague Australia and parts of North
America. For at least 40,000 years, Indig- Taxonomic plant identification is taught alongside Indigenous knowledge of the use of these plants to
enous Australians have been managing the Indigenous students from various tribes around Tamaki Mākarau (Auckland) New Zealand.
landscape, leaving a deep human imprint,
one that has been nearly erased from living because they differ in methodologies, of present-day Indigenous knowledge with
memory. However, in parts of Australia, lo- philosophies, worldview, and modes of 15th-century knowledge and decision-
cal authorities, scientists, and Indigenous transmission. The knowledge produced making. By comparison, this extinction was
communities are now coming together to through traditional science methods has two centuries before British colonization
revisit Indigenous fire management and resulted in many game-changing outcomes, would produce such mass environmental
reframing science through Indigenous such as the eradication of smallpox and devastation in its colonies that the Western
knowledge to better understand these the production of life-saving vaccines. conservation paradigm would be born. In
modern environmental dilemmas (6). However, it has also proven itself wrong fact, evidence has shown some present-day
This example highlights how (for example, phlogiston, aether, and Indigenous managed lands to have much
PHOTO: DR. LAUREN WALLER

knowledge and its cultural context have
a place in education because local context
matters, particularly when Indigenous
communities with their knowledge drive
questions or request the support of
science tools such as genomics to generate
phrenology) and produced catastrophic
outcomes for humanity (such as the atomic
bomb), while failing thus far to solve the
most pressing challenges of our time
(such as climate change). As scientists,
we accept such scientific shortcomings on
higher biodiversity than some Western
Conservation managed lands (8), and
this can likely be attributed in part to the
nuanced relationships that are encoded
within Indigenous knowledge. Thus, the
argument that Indigenous knowledge

SCIENCE science.org 9 FEBRUARY 2024 • VOL 383 ISSUE 6683 593

I NS I GHTS | P O L I C Y F O RU M
only includes historical, precolonization
learnings, whereas mainstream science
can continuously learn from its mistakes,
is both a straw person and a circular
argument because it defines Indigenous
knowledge using the exact criteria
(outdatedness) for which it is criticized.
Some societies, such as many
Indigenous groups, lack traditional
written communication and thus
transmit knowledge within memorable
framings, such as stories or myths, to
ensure their longevity (9). A superficial
interpretation of these framings is often
used to depict Indigenous knowledge as
purely metaphysical—an example of the
appeal to extremes fallacy—justifying its
displacement by science (9). Yet this false
dichotomy ignores evidence that, like
Indigenous knowledge, science also uses
abstractions and stories (such as models)
to facilitate knowledge transmission and
illustrate concepts or key messages. For
example, both simulation and statistical
models can require simplifications that are
known to be false. Bohr’s model of the atom
and Newtonian physics are still widely
taught in schools as easily understood
approximations, despite their limitations
with respect to quantum mechanics. By
analogy, dismissing Indigenous narratives
on their verbatim interpretation risks
missing considerable opportunity to
learn from the knowledge and experience
encoded within them (10, 11).
The argument “True science is evidence-
based not tradition-based” (1) ignores
considerable research demonstrating that
false representations of both science and
Indigenous knowledge have unnecessarily
polarized this debate. We argue that it
would be more fruitful to undertake it in
an informed and nuanced way.
KNOWLEDGE TRANSMISSION
In addition to a suite of known benefits to
Indigenous students (12), we see the poten-
tial for all students to benefit from expo-
sure to Indigenous knowledge, alongside a
science curriculum, as a way of fostering
sustainability and environmental integ-
rity (13). For science learning, connecting
science with student values and fostering
understanding of the role of social and cul-
tural context can lead to the production of
ethically sourced scientific knowledge (14).
In addition, the generation and transmis-
sion of Indigenous knowledge are both
closely connected to practice: experiencing
and doing. Such experiential learning is
known to benefit learning in general (14),
and the broader range of contexts pro-
vided by place-based Indigenous knowl-
edge allows students to connect learning
594 9 FEBRUARY 2024 • VOL 383 ISSUE 6683
with their local environment, which may
result in more affective and authentic en-
gagement, leading to greater acceptance
and uptake of new knowledge (14). Given
the societal and environmental issues fac-
ing the planet, providing an intercultural
understanding that leads to a more bal-
anced and connected worldview can result
in positive outcomes, including effective
science education.
INNOVATION DRAWS FROM DIVERSITY
Innovation, like evolution, draws from
diversity, so that diversity of knowledge
sources and transfer among them are
known to positively influence innovation
(15). This value is exemplified by the move
toward cross-disciplinarity, in which sci-
ence can draw on inductive fields of re-
search for hypothesis generation. Given
this value of diversity, global challenges
faced by humanity could benefit from in-
clusive science and maintenance of knowl-
edge diversity more generally rather than
insisting on assimilation into a single cul-
ture of knowledge generation. One path to
preventing the extinction of Indigenous
knowledge is its dissemination in class-
rooms, under Indigenous governance and
management (supported by the Interna-
tional Bill of Rights and, specifically in
New Zealand, the Treaty of Waitangi Act
1975 and the Waitangi Tribunal). Not only
will this help to protect Indigenous knowl-
Learning about the biological structures of plants
used in tradtional practices
edge holders and their culture, it has the
potential to generate innovation more
broadly.
EVIDENCE, NOT CARICATURES
Indigenous knowledge can complement
science-generated knowledge in the peda-
gogy landscape by providing acceptance
and understanding and by contributing
to the addressing of global challenges. We
urge both education policy analysts and
scientists engaging in this debate to draw
on evidence rather than caricatures of
Indigenous knowledge and a partisan ap-
proach to knowledge generation. Knowl-
edge is produced in many traditions. The
scientific method is one of those, Indig-
enous approaches are others, and these
are not necessarily mutually exclusive.
We need to respect Indigenous knowledge
for its inherent value and the philosophi-
cal reflections it can provide science to
improve outcomes, irrespective of how
Indigenous knowledge is contextualized.
Much of our time as researchers is spent
challenging scientifically derived univer-
sal truths through work in local contexts,
and Indigenous knowledge does the same
but with a higher degree of connectivity
between the researcher and what is “re-
searched.” Arguably, the ignorance toward
Indigenous knowledge and its application
is only slightly greater than ignorance to
science methodology. We think this is the
strongest rationale for teaching them both
in schools. j
R E F E R E N C ES A N D N OT ES
1. R. Dawkins, “Why I’m sticking up for science,” The
Spectator, https://www.spectator.co.uk/article/why-
im-sticking-up-for-science (2023).
2. J. Tollefson, Nature 621, 454 (2023).
3. C. Mika, J. World Philos. 7, 101 (2022).
4. M. Monk, J. Osborne, Sci. Educ. 81, 405 (1998).
5. G. Stewart, New Zealand J. Teach. Work 19, 84 (2022).
6. M. S. Fletcher, T. Hall, A. N. Alexandra, Ambio 50, 138
(2021).
7. R. Tang, M. C. Gavin, Conserv. Soc. 14, 57 (2016).
8. S. T. Garnett et al., Nat. Sustain. 1, 369 (2018).
9. G. Snively, J. Corsiglia, Sci. Educ. 85, 6 (2001).
10. S. M. Sidik, Nature 601, 285 (2022).
11. C. H. Trisos, J. Auerbach, M. Katti, Nat. Ecol. Evol. 5, 1205
(2021).
12. F. Duckworth, M. Gibson, S. Macfarlane, A. Macfarlane,
Alternative 17, 3 (2021).
13. R. Zidny, J. Sjöström, I. Eilks, Sci. Educ. 29, 145 (2020).
14. S. Jose, P. G. Patrick, C. Moseley, Int. J. Sci. Educ. 7, 269
(2016).
15. J. De Beer, J. New Gener. Sci. 14, 34 (2016).
AC K N OW L E D G M E N TS
The authors acknowledge colleagues S. Lambert, S. Nissen,
P. Dearden, P. Hulme, and two anonymous referees who
provided feedback on various drafts; S. Scott for refer-
ence checking; and M. Duley, F. Bulman, and Z. Marshall for
images provided. Funding was provided by the New Zealand
government, Tertiary Education Commission Centres of
Research Excellence Fund for Bioprotection Aotearoa con-
tract 2021–2028, and Biological Heritage National Science
Challenge 1920-44-021 A.
10.1126/science. adi9606
science.org SCIENCE

B O OKS et al .
NUCLEAR WEAPONS
In search of nuclear stewards
A journalist probes the culture and convictions of
researchers at US national labs
By Jacob Darwin Hamblin
N
uclear weapons have been the highest
strategic priority of the United States
for nearly 80 years. Of the more than
2000 nuclear detonations that oc-
curred in the half-century after 1945,
about half were US weapons tests. To-
day, the country’s scientists work to maintain
the viability of the arsenal—plan-
ning subcritical explosions, using
computers to make complex cal-
culations, and engaging in myr-
iad projects to act as “stockpile
stewards.” In Countdown, science
journalist Sarah Scoles tells the
stories of the people on the front
lines of nuclear research and de-
velopment at sites managed by
the National Nuclear Security
Administration (NNSA), focusing
specifically on the national labs at Bold Type Books, 2024. to fusion power for electricity
Los Alamos, Lawrence Livermore,
Sandia, and Savannah River.
Much of the work conducted in these
labs is secret “mission science” related to
weapons, but scientists at national labs also
publish research in academic journals. Scoles
PHOTO: CORBIS VIA GETTY IMAGES
probes their feelings about the future of nu-
clear weapons and the roles they are playing
in it. Some believe in deterrence or think they
The reviewer is at the School of History, Philosophy, and
Religion, Oregon State University, Corvallis, OR 97331, USA.
Email: jacob.hamblin@oregonstate.edu
SCIENCE science.org
can do social good at the margins of weap-
ons work. A few simply find nuclear science
exciting and seem to have few qualms about
its implications. Others offer rationalizations:
One noted that if she could not live in a world
without nuclear weapons, she wanted to en-
sure their safety, security, and reliability.
Scoles sees dual-use science as fundamen-
tal to the moral universe of the national labs
and as NNSA’s strategy for keep-
ing highly qualified people on
the payroll. Some employees do
not work on weapons at all, and
most imagine themselves contrib-
uting to nonmilitary domains. At
Lawrence Livermore, the National
Ignition Facility focuses on fu-
sion reactions, for example. It
Countdown: exists because of thermonuclear
The Blinding Future weapons, but the experiments
of Nuclear Weapons conducted there might also lead
Sarah Scoles
272 pp. generation. Similarly, its National
Atmospheric Release Advisory
Center models potential casualties from ex-
plosions or radiological attacks, but it also ad-
vises about civilian reactor accidents. Many
of the lab’s experts pursue technologies that
may prevent the diversion of nuclear materi-
als away from peaceful uses or detect a for-
eign nation’s clandestine nuclear test. Others
perform nuclear forensics to help determine
the source of bomb materials.
If the lines separating nonmilitary work
from the national labs’ weapons mission
A worker surveys equipment at the National Ignition
Facility at Lawrence Livermore Laboratory.
seem hazy, this is because they prefer it that
way, not least because the ambiguity helps
bring young people into the fold. An uncom-
fortably large proportion of NNSA employees
are at retirement age, and the agency is keen
to recruit. It scouts new talent from univer-
sities across the country, convenes nuclear
“bootcamps,” provides paid internships, and
offers myriad funding opportunities. Some
recruits enjoy these opportunities and then
move on, while others stay to establish ca-
reers in the national lab system, becoming
stewards of the US arsenal. As with military
funding in general, the grant money and ca-
reer opportunities are hard to resist.
At the heart of Countdown is a recent
policy change by the United States to resume
large-scale manufacture of plutonium pits—
the part of a warhead that causes a nuclear
explosion. Such pits were once produced at
the Rocky Flats plant in Colorado, but pro-
duction was halted there permanently when
the Federal Bureau of Investigation found the
site in violation of safety and environmental
regulations during a 1989 raid. The US is
now on track to produce them again at Los
Alamos and Savannah River.
It is possible that the old pits work fine,
but testing them would open a can of worms.
The US and other countries agreed to limit
nuclear tests to underground ones in 1963,
and the 1996 Comprehensive Test Ban Treaty
banned nuclear explosions altogether. (The
US signed the treaty but never ratified it.)
Officially, pit production will ensure that
the bombs are reliable, but skeptics argue
that new pit production will enable modern-
ized weapon designs. Whether it is necessary
or not, it will mean jobs and research dollars
for the production facilities and labs.
Since 2006, Los Alamos has been operated
by a profit-driven conglomerate called Los
Alamos National Security, and Scoles sees it
subtly changing the lab’s priorities. Some sci-
entists have departed in frustration, manage-
ment fees have skyrocketed, and corporate
culture seems intent on advocating—in the
name of national security—a weapons mod-
ernization program that also happens to be
quite profitable.
Countdown offers readers an accessible
introduction to some of the frontiers of US
nuclear research, from quantum computing
to laser-induced fusion. But its great strength
is in letting scientists who do this work—
many of them women—speak for themselves,
as they wrestle with moral quandaries and
imagine the consequences of their steward-
ship of the US nuclear arsenal. j
10.1126/science.adn2591
9 FEBRUARY 2024 • VOL 383 ISSUE 6683 595
I NS I GHTS | B O O K S
SCIENCE LIVES
A Black spaceflight legacy
An aspiring Cold War–era astronaut and those who followed
in his footsteps take center stage in a new documentary
By Matthew Shindell
T
he story of Ed Dwight—a young Air
Force captain who might have been
America’s first Black astronaut in
the 1960s—has rarely been treated
in much depth and has never been
adequately incorporated into our un-
derstanding of the early history of human
spaceflight. Rather, it has emerged multiple
times in newspaper articles and documen-
tary films, especially in the past decade, as
an untold story in the history of the Cold
War space race. National Geographic’s new
documentary, The Space Race, is the latest to
tell Dwight’s story. Although it covers some
familiar territory, it manages to do
so in new ways and for the first time
examines Dwight’s legacy in the con-
text of the ongoing contributions of
African Americans to spaceflight.
This is the story as Dwight tells
it: His astronaut candidacy resulted
from a promise President John F.
Kennedy made to the National
Urban League’s Whitney Young Jr.
Young asked Kennedy to deliver a
Black astronaut for the civil rights
movement—a symbol for Black
equality and a role model for Black
students who might pursue military
or engineering careers. At the behest
of the Kennedy White House, Air
Force Chief of Staff General Curtis
LeMay enrolled Dwight, already an
accomplished pilot, in the Aerospace
Research Pilot School, the program
that had trained some members of
NASA’s first astronaut class. Despite resis-
tance, as Dwight remembers it, from the
school’s commandant, Colonel Chuck Yeager,
Dwight did well in the program.
But Dwight was never named to NASA’s
astronaut corps. NASA promoted his candi-
dacy to the Black press and sent him around
the country to talk about space exploration
to Black audiences. Whether Dwight would
have only been used to deliver pro-NASA
propaganda or whether he would have been
sent to space had Kennedy not been assas-
sinated in 1963 is unclear. What is certain is
that Kennedy’s death put an end to Dwight’s
The reviewer is a space history curator at the Smithsonian’s
National Air and Space Museum, Washington, DC, USA.
Email: shindellm@si.edu
596 9 FEBRUARY 2024 • VOL 383 ISSUE 6683
hopes. He resigned from the Air Force in
1966, feeling defeated. Despite all the fanfare
NASA and the White House had drummed
up around Dwight, his space journey ended
unceremoniously.
It must be noted that historian Richard
Paul has previously scrutinized Dwight’s
story and found it to be less clear-cut than
Dwight’s retelling; however, none dispute
that Dwight was used as a PR corrective
for the fact that NASA’s all-white astronaut
corps did not reflect the progress of the civil
rights movement (1).
In 2009, President Barack Obama named
highly accomplished astronaut Major
General Charles F. Bolden Jr. as NASA
Astronaut Charles Bolden is photographed on the shuttle Columbia.
Administrator, making him the first Black
man appointed to this position. During
his confirmation hearing, Bolden drew the
Senate’s attention to Dwight, whom he had
invited to be in the audience, and credited
him for paving the way for African American
astronauts. Dwight’s story forms the back-
bone of this documentary, but it is Bolden
who connects Dwight’s story to the more re-
cent history of human spaceflight.
Rather than privileging the perspectives
of historians, the history of Black contribu-
tions to NASA is here navigated by Dwight
and Bolden, whose voices are joined by the
first Black astronaut, Guion Bluford; the
third Black astronaut, Frederick Gregory;
and more recent astronauts, such as Leland
The Space Race
Lisa Cortés and Diego Hurtado
de Mendoza, directors
National Geographic, 91 min,
premiering 12 February 2024.
Melvin and Victor Glover. This reliance on
individuals’ memories over archival history
has its drawbacks, as noted above, but it
allows the importance of Dwight’s story to
come through more clearly.
Although their selection to the astronaut
corps in 1978 came little more than a decade
after Dwight’s resignation, neither Bluford
nor Gregory knew that Dwight had once
believed himself to be a contender for the
title of first Black astronaut. Nor did they
know the story of Air Force pilot Robert
Lawrence Jr., who died in a training accident
after being selected for the Air Force’s clas-
sified space project, the Manned Orbiting
Laboratory. They learned of these stories
only after their time in space.
In their own successes, as well
as those of younger astronauts who
have followed in their footsteps,
Bluford and Gregory are able to
frame Dwight’s and Lawrence’s expe-
riences as milestones on the road to
equality in space. It is clear to them
that this journey is not yet com-
plete—a fact that is brought home
poignantly by the experience of
Glover, who was on the International
Space Station during the trial of
George Floyd’s murderer, police offi-
cer Derek Chauvin. Glover, who had
brought with him to space a painting
of Floyd, struggled emotionally dur-
ing the trial. But he was able to find
fellowship in a support network of
Black astronauts past and present—a
group that we learn has come to call
itself “the Afronauts.”
Human spaceflight is famously aspira-
tional—not only do we want astronauts to
represent the best of humanity, we also call
on them and the technologies that send
them to space to show us what we are ca-
pable of and who we can be. As The Space
Race reminds us, the possibilities of space
are always constrained by our present hopes,
struggles, and prejudices. But occasionally,
space does show us a glimpse of how we
might transcend our limitations. j
RE FE REN CES A ND N OT ES
1. R. Paul, S. Moss, We Could Not Fail: The First African
Americans in the Space Program (Univ. of Texas Press,
2015).
10.1126/science.adn2127
science.org SCIENCE
 LET TERS
Edited by Jennifer Sills
Protect native fish in
China’s Yellow River
Since 2021, Chinese President Xi Jinping
has underscored the importance of wetland
restoration to safeguard the ecosystems of
the Yellow River (1), China’s second-longest
river and a cradle of its cultural heritage (2).
The administration’s “Outline of ecological
protection and high-quality development in
the Yellow River Basin” identifies measures
to preserve vital wildlife habitats along the
river (1). However, the plan focuses solely
on the flora and fauna in the adjacent ter-
restrial ecosystems and overlooks the river’s
critically endangered indigenous fish.
The Yellow River basin has undergone
a stark decline in fish diversity. During
the 1980s, more than 120 fish species lived
in the river’s main stream, but only 47
remained by 2013 (3). This rapid reduc-
tion has pushed iconic species such as the
Chinese paddlefish and bronze gudgeon to
the brink of extinction (4). The “Red list of
Chinese vertebrates” highlights the precari-
ous state of 24 native fish species in the
Yellow River, with 4 classified as Critically
Endangered, 10 as Endangered, and 10 as
Vulnerable (5). Even species that were once
common are now at risk of extinction.
Overfishing is the primary threat to the
Yellow River’s fish (3), but other factors have
also contributed to the decline. Discharge
from industrial activity along the lower and
middle reaches of the Yellow River has led
to deteriorating water quality (6). Invasive
nonnative fish species and changes in the
genetic purity of native fish also pose a
598 9 FEBRUARY 2024 • VOL 383 ISSUE 6683
threat. In 2013, more than 25 nonnative spe-
cies were found in the Yellow River, includ-
ing sunfish, Brazilian turtles, and crocodiles
(3). The Yellow River carp has almost disap-
peared (4) as a result of poor water quality
and the incursion of nonnative carp since
the late 1990s (6).
Climate change has also contributed to
the decline of fish populations. Between
1919 and 2018, and especially since 1986,
the Yellow River’s water flow has decreased,
with a staggering 64% reduction in the
volume of water that reaches the sea (7).
Primarily driven by climate change, this
reduction in flow exacerbates the challenges
faced by the river’s ecosystems (8).
A comprehensive fishing ban might help
to preserve fish diversity in the Yellow River
basin, but the river’s extensive network of
reservoirs presents a formidable challenge
(9). Spanning three major terrains, the
Yellow River features 219 interconnected
reservoir facilities that cannot be easily
altered (9). Even with a comprehensive
fishing ban, numerous fish species might
remain confined to specific areas, offer-
ing minimal assistance in terms of genetic
exchange and reproduction (10). However,
local fishing bans could offer protective
measures for fish species in specific areas.
In addition, China should implement strin-
gent environmental policies, control inva-
sive species, and conduct climate-adaptive
ecological restoration.
By signing the Kunming-Montreal Global
Biodiversity Framework (11) China has com-
mitted to fortifying biodiversity preserva-
tion efforts. In addition to southern China’s
well-known biodiversity hotspots, the coun-
try must protect northern regions like the
Yellow River basin. The wetland restoration
Decreased water flow in China’s
Yellow River, along with other threats,
has led to a decline in native fish.
plan, combined with efforts to restore the
Yellow River’s fish species, would represent
a good start.
Xu Guo1, Qianhui Lin2, Xiangcheng Zheng3, Shuo
Wang2, Qiao Li4, Chao Shi2*
1
Geotechnical and Structural Engineering
Research Center, Shandong University, Jinan
250061, China. 2College of Marine Science and
Biological Engineering, Qingdao University of
Science and Technology, Qingdao 266042, China.
3
School of Mathematics, Shandong University,
Jinan 250100, China. 4School of Engineering
Medicine, Beihang University, Beijing 100083,
China.
*Corresponding author.
Email: chaoshi@qust.edu.cn
RE FE REN CES A ND N OT ES
1. Central People’s Government of the People’s Republic
of China, “Outline of ecological protection and high-
quality development plan in the Yellow River Basin”
(2023); https://www.gov.cn/zhengce/2021-10/08/
content_5641438.htm [in Chinese].
2. A. Lawler, Science 325, 8 (2009).
3. X. Shan et al., Mar. Coast. Fish. 5, 1 (2013).
4. W. Tang, D. He, J. Lake Sci. 25, 4 (2013).
5. Z. Jiang et al., Biodivers. Sci. 24, 5 (2016).
6. C. Liu, J. Xia, Hydrol. Process. 18, 12 (2004).
7. Y. Ru et al., Mar. Sci. 30, 3 (2006).
8. W. W. Immerzeel, L. P. Vanbeek, M. F. Bierkens, Science
328, 6 (2010).
9. L. Ran, X. X. Lishan, Z. Xin, X. Yang, Glob. Planet. Change
100, 1 (2013).
10. L. Ran, X. X. Lu, Hydrol. Process. 26, 8 (2012).
11. Convention on Biological Diversity, “Kunming-Montreal
Global Biodiversity Framework” (2023); https://www.
cbd.int/doc/decisions/cop-15/cop-15-dec-04-en.pdf.
10.1126/science.adn7432
India’s services-driven
growth excludes women
In the past two decades, India has been
developing its business services sector,
especially information and communication
technology, to drive economic growth (1, 2).
With annual growth of 8.2%, high-skilled
science.org SCIENCE

service jobs have contributed to economy-
wide surge in productivity (2, 3). However,
there are not enough qualified individuals
to fill available positions (4). The low rep-
resentation of women in these industries
exacerbates labor shortages (4).
Women workforce participation in
India has stagnated at less than 30% (5,
6). Of the women who do work, 85% have
jobs in agriculture or industries such as
textiles, construction, and food (4–6). As
India’s business services sector has grown
and high-paying job opportunities have
increased, women have been excluded.
Jobs in the business services sector
require information and communication
technology skills. Women in India are
less likely than men to be able to afford
a cell phone or computer (7). Only 30%
of women know how to use the Internet
to find necessary information, oper-
ate mobile money applications, or use
e-services (7). Because women are less
likely to use technology in their daily
lives, they are at a disadvantage in labor
and financial markets (8).
To address the gender gap in techno-
logical skills, India should provide digital
skills training for women, tailored to
the job requirements of the information
and communication technology industry.
The nationwide Digital India program,
which aims to improve access to technol-
ogy throughout the country (9), could
add such training efforts to its goals. The
training should be supplemented with
support groups for women in the tech-
nology sector and online job-matching
services that include peer networks [e.g.,
(10)]. These efforts, implemented at the
local level, could help women overcome
the obstacles that prevent them from
accessing and using technology. For exam-
ple, support groups for 4.3 million women
learning digital skills across municipali-
ties in Kerala have led to computer use in
village offices, the introduction of a digital
accounting system, and the deployment
of digital technology for information and
public services (10).
Policy efforts directed toward closing
digital gender gaps would have far-reaching
benefits. Women would gain increased
control over financial resources, providing
them with access to additional training
programs and opportunities. Giving women
financial independence would create cas-
cading effects at the household level as they
pass that stability and access to knowledge
to their children (11, 12).
Isha Gupta
Department of Economics, Ramanujan College,
University of Delhi, Delhi, India.
Email: isha.gupta106@gmail.com
SCIENCE science.org
R EFER ENCES AN D N OT ES
1. B. Eichengreen, P. Gupta, “The service sector as India’s
road to economic growth,” Working Paper w16757,
National Bureau of Economic Research (2011).
2. K. L. Krishna et al., “Productivity dynamics in India’s
service sector: An industry-level perspective,” Working
Paper 261, Centre for Development Economics (2016).
3. D. W. Jorgenson, M. P. Timmer, Scand. J. Econ. 113, 1 (2011).
4. B. Bhandari et al., “Skilling India–No time to lose,”
Technical Report, National Council of Applied Economic
Research (2018).
5. Employment Statistics in Focus, “Female labour uti-
lization in India,” Ministry of Labour and Employment
(2023).
6. E. Fletcher, R. Pande, C. M. T. Moore, “Women and work
in India: Descriptive evidence and a review of potential
policies,” CID Working Papers 339 (2017).
7. “The Mobile Gender Gap Report 2023,” Global System
for Mobile Communications (2023); https://www.gsma.
com/r/wp-content/uploads/2023/07/The-Mobile-
Gender-Gap-Report-2023.pdf.
8. “Bridging the digital gender divide: Include, upskill,
innovate,” Organization for Economic Co-operation and
Development (2018).
9. Ministry of Electronics and Information Technology,
“Digital India” (2023); https://csc.gov.in/digitalIndia.
10. A. Gurumurthy, N. Chami, “Digital India through a
gender lens” (Research Report, Heinrich Bohl Stiftung,
2018).
11. T. Suri, W. Jack, Science 354, 1288 (2016).
12. P. Choudhuri, “India’s employment challenges and the
demand for skills,” Working Paper 121, National Council
of Applied Economic Research (2021).
10.1126/science.adn7738
Oil pollution threatens
Persian Gulf marine life
The Persian Gulf is a shallow sea located
between Iran, Iraq, Bahrain, Kuwait,
Oman, Saudi Arabia, Qatar, the United
Arab Emirates, and Yemen. These nine
countries possess 50% of the world’s oil
reserves (1) and produce about 25% of the
oil consumed globally (2). The extensive
oil and industrial activities, including
numerous oil spills between 1980 and
2020 (1, 2), have led to substantial pollu-
tion and severely damaged marine life in
the Persian Gulf (3). Given that preserving
the quality of the marine environment is
essential to the social fabric of life and to
the economic success in the region (2),
governments must prioritize environ-
mental protection and provide support
to environmental organizations in the
region.
Aromatic hydrocarbons are commonly
found in oil spills. These pollutants have
high melting and boiling points but low
solubility and vapor pressure, and they
are toxic and mutagenic (4). The high
levels of ultraviolet radiation and elevated
temperature in the Persian Gulf can sub-
stantially increase the toxicity of hydro-
carbons (1), which subsequently cause
damage to coral reefs, fish, phytoplankton,
and zooplankton (1, 5).
Aromatic hydrocarbons harm corals and
rocky organisms by damaging tissues (4),
I N SI G H T S
causing bleaching (6), and preventing lar-
val settlement (1). Fish and other aquatic
organisms can absorb pollutants in water
directly and through the food they con-
sume (7). The Gulf War and oil pollution in
1991 caused a reduction in the Saudi prawn
population from 4000 t to 25 t, as well as
a decrease to 1/10 the abundance of eggs
and larvae within a year (2). On Kharg
Island, where much of Iran’s crude oil is
exported (1), elevated levels of polycyclic
aromatic hydrocarbons and their deriva-
tives were found in the liver and muscle
tissues of three edible fish species in 2018
(7). Hydrocarbon pollutants also threaten
phytoplankton and zooplankton (4), which
can then harm higher-level species in the
food chain.
All governments in the region must pri-
oritize the establishment and enforcement
of effective management practices, sus-
tainable methods, and strict regulations
to mitigate oil pollution and safeguard the
marine environment in the Persian Gulf
(2). It is imperative that these govern-
ments remove any constraints that hinder
the continuous monitoring and assess-
ment of pollution in industrial zones. By
implementing monitoring programs, gov-
ernments can aid in the recovery of dam-
aged ecosystems and ensure the safety of
seafood (2, 3).
In addition to national government
action, regional management programs
should be established under international
supervision. Independent research knowl-
edge should be shared transparently,
and problems caused by regional and
local pollution should be identified. The
Regional Organization for the Protection
of the Marine Environment (8), the sole
Sea Forum in the Persian Gulf, plays a
crucial role in managing and conserving
the area and could facilitate collabora-
tions among oil-producing countries in
the region (2).
Hossein Yarahmadi
Department of Chemical Engineering, Sirjan
University of Technology, Sirjan, Iran, Email:
hyarahmadi61@gmail.com
RE FE REN CES A ND N OT ES
1. M. Oladi, M. R. Shokri, J. Hazard. Mater. 409, 124993
(2021).
2. A. M. Freije, J. Assoc. Arab Univ. Basic Appl. Sci. 17, 90
(2015).
3. S. Uddin, et al., Mar. Pollut. Bull. 173, 112913 (2021).
4. F. Khaksar et al., Mar. Pollut. Bull. 140, 35 (2019).
5. S. L. Coles, Atoll Res. Bull. 507, 1 (2003).
6. “Bleaching; a threat to coral reefs,” IRNA (2021); www.
irna.ir/news/84519319, [in Farsi].
7. A. Ranjbar Jafarabadi et al., Chemosphere 215, 835
(2018).
8. UN Environment Programme, “ROPME sea area”
(2024); https://www.unep.org/explore-topics/
oceans-seas/what-we-do/working-regional-seas/
regional-seas-programmes/ropme-sea-area.
10.1126/science.adn5624
9 FEBRUARY 2024 • VOL 383 ISSUE 6683 599
 RESEARCH
IN S CIENCE JOU R NA L S
Edited by Michael Funk

NEUROSCIENCE

Moving mitochondria in neurons

M
itochondria must be trafficked over long distances in neurons to support energy-intensive processes. Izquierdo-Villalba et al. found
that mitochondrial trafficking in neurons required the interaction of the G protein Gαq with the mitochondrial protein Alex3. Gαq
enhanced the binding of Alex3 to proteins that mediate anterograde trafficking of mitochondria. Expression of a constitutively active
form of Gαq in neurons from mice with a central nervous system deficiency in Alex3 revealed that Gαq was necessary for the function-
ing of Alex3 in dendritic arborization and in mitochondrial trafficking and distribution. —WW Sci. Signal. (2024) 10.1126/scisignal.abq1007

Neuronal mitochondria (light blue) are moved away from the cell body in a mechanism that involves G protein signaling.

PLANT SCIENCE
Oil production
in citrus plants
Secretory cavities serve as
reservoirs for a variety of
secondary metabolites, such
as essential oils and chemicals
that defend against pathogens
and herbivores. Wang et al.
describe a molecular pathway for
the development of the secre-
tory cavity (oil glands) in Citrus
species. Using genetic mapping
and genome editing, the authors
identified two transcription fac-
tors involved in the development
of the secretory cavity and cis-
regulatory elements within a leaf
shape gene that enhance gene
activation. A regulatory cascade
activates the MYC5 gene, which
regulates oil gland sheath cell dif-
ferentiation and cavity formation.
This work lays a foundation for
understanding secretory cavity
development and the synthesis of
essential oils. —AWa
Science p. 659, 10.1126/science.adl2953
QUANTUM GASES
Seeing second sound
In superfluids, heat propagates
like a wave in a phenomenon
dubbed “second sound.” However,
observing this propagation
directly is tricky. Yan et al. used a
quantum gas of strongly interact-
ing fermionic lithium atoms held
in a box potential to visualize
second sound. The researchers
used radiofrequency spectros-
copy to map out local changes
to the temperature. Above the
superfluid transition, heat propa-
gated diffusively, but below the
transition, wave-like propagation
characteristic of second sound
was observed. —JS
Science p. 629, 10.1126/science.adg3430
POLLINATION
Pollution disrupts scent
Chemical pollutants not only
reduce animal survival and
reproduction, but can also
disrupt their senses, changing
their behavior and interactions
with other species. Common
air pollutants such as ozone
degrade floral scents, poten-
tially affecting insects’ ability
to locate and pollinate flowers.
Chan et al. tested the effects
of ozone and nitrate radicals
(NO3) on nocturnal hawkmoth
pollination of a desert plant.
Both pollutants decreased con-
centrations of monoterpenes,
but NO3 had a stronger effect.
Degradation of scent molecules
by NO3 led to a decline in hawk-
moth visitations in wind tunnel
and field experiments, with
predicted reductions in fruit set
and plant fitness, with predicted
reductions in fruit set and plant
fitness. A global model suggests

600 9 FEBRUARY 2024 • VOL 383 ISSUE 6683 science.org SCIENCE

 that many urban areas have suf-
ficient pollution to significantly
reduce the distances at which
pollinators can sense flowers.
—BEL
Science p. 607, 10.1126/science.adi0858
MASS SPECTROMETRY
A collision course
breaks the mirror
Because proteins, carbohy-
drates, and nucleic acids occur
in just one of two mirror-image
geometries in nature, it is likewise
essential for chemists to make
just one of the mirror images, or
enantiomers, of many pharma-
ceutical compounds. Currently,
the most common means of
distinguishing enantiomers
is by separating them with
chromatography. Zhou et al.
report a technique to determine
enantiomeric ratios by mass
spectrometry alone. The method
relies on alternating current
excitations of the ions to induce
rotational trajectories in which
each enantiomer experiences
distinct collision behavior, leading
to sufficient separation for high-
total enzymes in tobacco to
IN OTHER JOURNALS Edited by Caroline Ash
demonstrate a minimal pathway
and Jesse Smith
to a key intermediate product:
baccatin III. —MAF
Science p. 622, 10.1126/science.adj3484
Lichens attached to trees make
a substantial, and often overlooked,
GEOLOGY contribution to carbon cycling
in boreal forests.
Free hydrogen
Hydrogen is an attractive alter-
native to traditional fossil fuels
because it can be used to pro-
duce energy without generating
carbon dioxide as a by-product.
However, natural sources of
hydrogen are rare and producing
it is energy intensive. Truche et
al. found a large natural source
of hydrogen gas emitting from
deep within the Bulqizë chromite
mine. This large hydrogen flux is
likely from long-term accumula-
tion within a faulted reservoir.
Places with similar geology
should be good targets for
finding other natural sources of
hydrogen. —BG
Science p. 618, 10.1126/science.adk9099
LICHENS
GENE THERAPY
A CRISPR fix for Cloaked in life

resolution quantification. —JSY
Science p. 612, 10.1126/science.adj8342
BIOSYNTHESIS
Enzymes for
paclitaxel production
Chemical total synthesis of
paclitaxel, an important anti-
cancer drug originally derived
from yew bark, has been an
important challenge for organic
chemists, but the material used
for drug production comes from
isolation of the natural product
or semisynthesis from precur-
sors. It is therefore important
to fully understand the biosyn-
thetic pathway in plants. Jiang
et al. identified a cytochrome
P450 enzyme that generates a
PHOTO: DAVID AUBREY/SCIENCE SOURCE
M
dysfunctional T cells ost static outdoor surfaces are blanketed by micro-
organisms and, in boreal regions, noticeably by the
Familial hemophagocytic
symbiotic algal and fungal assemblages of lichens.
lymphohistiocytosis (FHL) is a
Lichens can photosynthesize at freezing tempera-
spectrum of inherited, poten-
tures and grow when covered by snow, but little is
tially fatal inflammatory diseases
known about their contribution to forest photosynthesis or
caused by loss-of-function
mutations in the natural killer
respiration. Matvienko et al. examined the role of epiphytic
lichens in the carbon cycle in Siberian forests, where they are
cell and T cell cytotoxic machin-
estimated to amount to 3 kilograms per hectare of forest. In
ery. Although FHL can be treated
the laboratory, hydrated lichens were shown to emit methane
with hematopoietic stem cell
for several weeks, and carbon dioxide flux was found to be
transplantation, mortality is
species specific. The findings raise many questions about the
still high, and more effective
roles of lichen diversity and abundance, season, drought, and
therapeutic approaches are
aspect on forest carbon budgets. —CA
needed. Li et al. developed an
adeno-associated virus–based Forests (2024) 10.3390/f15010107
CRISPR-Cas9 system combined
with nonhomologous end joining
inhibition to repair T cells from a
perforin-deficient Epstein-Barr MICROBIOME to bacteria. Ayala et al. inves-
virus–triggered mouse model tigated whether commensal
of FHL. Repaired T cells were
Tonic signaling bacteria within the intestine elicit
able to prevent and cure disease for tolerance IFN responses that can affect

key chemical feature in the final
products, a four-membered
oxetane ring. The authors
found that a parallel epoxida-
tion is off pathway and not an
essential intermediate. Along
the way, they characterized
another oxidation enzyme and
ultimately reconstituted nine
in this model. Moreover, this
CRISPR-Cas9 system could be
used to mend T cells from pedi-
atric patients with FHL diseases.
Gene repair of autologous T cells
is thus a potentially promising
treatment for human FHL. —STS
Sci. Immunol. (2024)
10.1126/sciimmunol.adi0042
Type I interferons (IFNs) are a
type of chemical messenger
known as cytokines that the
immune system uses to initiate
immune responses. IFNs are
well recognized for their ability
to protect against viral infec-
tions, but they also have a less
well understood role in response
immune tolerance, a safeguard-
ing process that limits excessive
and damaging immune reactions.
Within the intestinal lumen, com-
mensal bacteria induced a subtle
type I IFN response referred to as
tonic signaling. The tonic IFN sig-
nals promoted dendritic cells to
release IL-27, which in turn shaped

SCIENCE science.org 9 FEBRUARY 2024 • VOL 383 ISSUE 6683 601


ALSO IN SCIENCE JOURNALS
CANCER
Changing lung
cancer subtypes
Acquired drug resistance to cer-
tain targeted cancer therapies
can lead to histological trans-
formation (HT) from one tumor
type to a more aggressive type.
HT is a poorly understood pro-
cess, so Gardner et al. designed
experimental models to study
the molecular events that under-
lie the conversion of mutant
epidermal growth factor recep-
tor (EGFR) lung adenocarcinoma
(LUAD) to neuroendocrine
small cell lung cancer (SCLC)
(see the Perspective by Berns).
Switching certain genes on and
off in either alveolar type 2 cells
(the precursors of LUAD) or in
pulmonary neuroendocrine cells
(precursors of SCLC) led to HT
that mimicked that observed in
human cancer. Neuroendocrine
SCLC transformation was only
observed when tumor cells
reworked their oncogenic driver
program from EGFR to the Myc
oncogene. —PNK
Science p. 603, 10.1126/science.adj1415;
see also p. 590, 10.1126/science.adn5218
REPRODUCTION
Helpful fibroblasts
The corpora cavernosa are
masses of vascular tissue that
can fill with blood and thereby
enlarge upon stimulation, creat-
ing the structure needed for
penile erection. By studying the
underlying mechanism for this
process in mice, Guimaraes et
al. determined that perivascu-
lar fibroblasts in the corpora
cavernosa play a key role in
erection physiology (see the
Perspective by Ryu and Koh).
Norepinephrine is a vasocon-
strictor that restricts penile
blood flow at baseline, whereas
vasodilators released by sexual
arousal counteract its effects,
allowing an erection to take
place. Recurrent erectile activity
down-regulates Notch signaling,
thereby increasing the numbers
of perivascular fibroblasts, and
SCIENCE science.org
Edited by Michael Funk
these fibroblasts then suppress
vasoconstrictive norepinephrine
signaling. Conversely, aging is
associated with a decrease in
these fibroblasts, contributing
to the risk of erectile dysfunc-
tion. —YN
Science p. 604, 10.1126/science.ade8064;
see also p. 588, 10.1126/science.adn5182
PLANT SCIENCE
Photoperiodic growth
Plants are highly responsive
to photoperiodic cues, and
in many parts of the world,
daylength varies considerably
throughout the year. Complex
signaling networks have been
identified that regulate the
timing of flowering in response
to daylength. Wang et al. found
that a parallel mechanism exists
to regulate vegetative growth
(see the Perspective by Buckley
and Haydon). Whereas flowering
time primarily depends on light
signaling, vegetative growth is
controlled by photosynthetic
and metabolic cues that change
according to photoperiod. The
authors found that photosyn-
thetic control of vegetative
growth is partially dependent on
the production of myo-inositol,
a precursor for many other mol-
ecules involved in plant growth.
—MRS
Science p. 605, 10.1126/science.adg9196;
see also p. 589, 10.1126/science.adn5189
HEALTH ECONOMICS
Understanding
providers’ motives
Diarrhea is a leading cause
of child mortality in India. It
becomes deadly when excre-
tions exacerbate severe
dehydration and loss of
electrolytes. Most health care
providers in India know that oral
rehydration salts (ORS) are an
inexpensive, lifesaving treat-
ment for child diarrhea, yet they
are widely underused. Wagner
et al. undertook randomized
controlled trials involving stan-
dardized patients (actors trained
to seek care for a child’s diar-
rhea) who visited 2282 private
health care providers in India.
Trials were designed to identify
three barriers driving underuti-
lization: assuming patients lack
interest in ORS, incentives to
prescribe more lucrative (but
inappropriate) medicines, and
incentives to sell non-ORS medi-
cines in stock when ORS are
unavailable. The dominant bar-
rier was assuming that patients
were uninterested, showing that
simple interventions could save
many lives. —EEU
Science p. 606, 10.1126/science.adj9986
METALLURGY
Printing preferable parts
Laser powder bed fusion
provides the opportunity to
make custom-built metal
structures, but these objects
can have undesirable variability
in mechanical properties. Zhang
et al. addressed the origin of
this issue, unwanted metastable
phases and columnar-shaped
crystals, by adding molybdenum
nanoparticles to a common alu-
minum alloy (see the Perspective
by Zhang and Wang). The
nanoparticles both encouraged
the growth of symmetric grains
and suppressed the formation of
unwanted phases. The resulting
samples made from laser pow-
der bed fusion have much better
mechanical properties, show-
ing the promise of this design
strategy. —BG
Science p. 639, 10.1126/science.adj0141;
see also p. 586, 10.1126/science.adn6566
SOLID-STATE PHYSICS
Mixing the right
ingredients
One of the recipes for realizing
topological superconductivity
calls for interfacing a topological
insulator with a superconductor.
In a variant of that approach, Yi
et al. grew a heterostructure con-
sisting of layers of a magnetic
topological insulator, (Bi,Sb)2Te3
doped with chromium, and
9 FEBRUARY 2024 • VOL 383 ISSUE 6683
R E S E A RC H
antiferromagnetic iron tellu-
ride. Neither of these materials
is superconducting, but iron
telluride is a parent compound
for a family of iron-based
superconductors. Interfacing
the layers led to the appear-
ance of superconductivity in the
presence of ferromagnetism
and topological band structure.
This combination of proper-
ties makes the heterostructure
a promising, although not yet
proven, platform for observing
chiral topological superconduc-
tivity. —JS
Science p. 634, 10.1126/science.adk1270
PLANT REPRODUCTION
Parental coordination
During reproduction, flowering
plants undergo two fertilization
events to form the embryo and
the nutritive endosperm. After
gametes fuse, they must then
resume cell cycle progression
to allow them to divide and
develop. Simonini et al. found
that a protein called RBR1 holds
the female endosperm precursor
cell in stasis, preventing full DNA
replication. During fertilization,
the sperm cell delivers a D-type
cyclin to the female cell to reacti-
vate the cell cycle. These results
give insight into the communica-
tion between parental cells to
enable coordination of subse-
quent development. —MRS
Science p. 646, 10.1126/science.adj4996
BIOGEOGRAPHY
Evolution of landscapes
and plants
Madagascar is a hotspot of
biodiversity and unique species,
both because of its long isolation
from the mainland and its high
rates of speciation. Liu et al.
hypothesized that landscape
evolution contributed to plant
diversification by creating
barriers and habitat heteroge-
neity, providing opportunities
for allopatric speciation. They
tested this hypothesis by model-
ing topographic dynamics on
602-B
R ES E ARCH | I N S C I E N C E J O U R NA L S

Madagascar over the past 45 improved B cell production. The

million years and found high
plant species richness and
diversification rates along the
landward-shifting continen-
tal rift escarpment, with low
north-south connectivity and
high rates of local endemism.
This work shows how relatively
small-scale, transient landscape
evolution can shape long-term
biodiversity patterns. —BEL
Science p. 653, 10.1126/science.adi0833
approach also overcame the T
cell differentiation block seen
with RAG1 mutations in artificial
thymic organoids. These findings
support further study of HDR-
mediated gene editing for the
correction of RAG1 mutations.
—MLN
Sci. Transl. Med. (2024)
10.1126/scitranslmed.adh8162

PALEOCLIMATE
Clay and climate
How were atmospheric carbon
dioxide levels regulated to allow
habitable surface conditions
since the Archean, and what
were temperatures really like
over that interval? Isson and
Rauzi present an ensemble of
oxygen isotope measurements
from shale, iron oxide, carbon-
ate, and chert over the past two
billion years. These data show
that the Proterozoic climate was
temperate until temperatures
began to fall during the early
Paleozoic, around 500 million
years ago. This drop in tempera-
ture occurred in concert with a
decline in clay authigenesis and
the rise of siliceous life, both of
which would have contributed to
declining levels of atmospheric
carbon dioxide. —HJS
Science p. 666, 10.1126/science.adg1366

GENE THERAPY
Knock it out (and in)
Gene therapy is a promising
treatment for patients with muta-
tions in recombination activating
gene 1 (RAG1), a cause of severe
combined immunodeficiency,
but there are safety concerns
due to the resultant constitutive
RAG1 expression. Castiello et al.
used homology-directed repair
(HDR)–mediated gene editing
to correct RAG1 while preserv-
ing its physiological regulation.
A combined knock-out and
knock-in gene-editing strategy
corrected RAG1 in patient hema-
topoietic stem and progenitor
cells (HSPCs), leading to rescue
of protein expression. When
transplanted into humanized
mice, these edited HSPCs led to

602-C 9 FEBRUARY 2024 • VOL 383 ISSUE 6683 science.org SCIENCE

R ES EA RCH | I N O T H E R J O U R NA L S

UPCYCLING

Nitrogen stitches
for plastic mixtures

N
otwithstanding the well-known 1 through 7
labels in their characteristic triangles, separating
plastic waste mixtures into their pure constitu-
ents is often impractical. There is particular
interest in finding ways of better processing mix-
tures of polyethylene and polypropylene because they
are both so abundant and are often used in tandem.
Vialon et al. report that a nitrogen-rich azidotriazine
molecule can decompose into reactive nitrenes at the
melting temperatures of the polyolefin mixtures,
stitching them together. The resultant extruded plas-
tic has much improved ductility and creep resistance.
—JSY
J. Am. Chem. Soc. (2024) 10.1021/jacs.3c12303

A simple azidotriazine-based grafting agent enables direct

upcycling of blends of polyethylene and polypropylene waste
into high-performance materials.

regulatory T cell responses to
maintain a tolerogenic intestinal
milieu. —PNK
J. Exp. Med. (2024)
10.1084/jem.20230063
BIOENGINEERING
Microtubule tracks
on a chip
Microtubules are cytoskeletal
structures in cells that form the
mitotic spindle to allow chromo-
some separation during cell
division. These filamentous
structures also form tracks for
transporting molecules within
cells. Because studies in cells
had previously revealed mecha-
nisms that control microtubule
subunit nucleation and branching,
Zaferani et al. explored meth-
ods to build synthetic circuits
of microtubular biopolymers.
The authors used extracts of
meiotic eggs from the African
clawed toad (Xenopus laevis)
to make patterns by control-
ling nucleation and physically
affecting growth. For example, a
chip decorated with microtubular
curves and bifurcations could
be used to form “microtubule
diodes.” Microtubule chips could
guide molecular motors or even
combine with other components
to make cytoskeletal switches or
logic gates. —LBR
Proc. Natl. Acad. Sci. U.S.A. (2024)
10.1073/pnas.2315992121
SCIENCE EDUCATION
Hidden effects of science
competitions
Science competitions are meant
to encourage student interest in
science and science careers, and
evaluations typically focus on
positive outcomes and success-
ful competitors. Garrecht et al.
took the opposite approach, using
a latent change score model
to examine the effects of early
elimination on participants in
the German Biology Olympiad.
Results showed that most partici-
pants’ biology-related study and
career task value remained stable
from the first to the second round
of the competition, and their
expectancy of success in biology
and biology careers developed
positively. However, for partici-
pants placing high importance
on advancing in the competi-
tion, elimination interfered with
the development of study and
career expectations. The authors
encourage science competitions
to address participants’ values
about science and science
careers, especially in earlier
rounds, and to develop innovative
ways of providing feedback to
improve post-elimination perfor-
mance. —MMc
J. Res. Sci. Teach. (2023)
10.1002/tea.21901
CARBON EMISSIONS
Trawling for problems
It is well known that trawling, a
fisheries practice in which nets
are dragged across the sea
floor, has disastrous impacts on
marine ecosystems. Recent work
has revealed that disturbance of
seafloor sediments also releases
carbon that contributes to cli-
mate-altering emissions. Atwood
et al. used data-based predic-
tions of global trawling activity
from 1996 to 2020, in conjunc-
tion with a series of models, and
estimated that about half the car-
bon stirred up was released into
the atmosphere over the follow-
ing 7 to 9 years. This translates
into the equivalent of about 10%
of the global emissions resulting
from land-use change in 2020.
The authors also predicted that
trawling-related carbon release
could lead to marine pH changes
in some heavily trawled regions.
Limiting trawling will not only
protect marine ecosystems, but
could also be part of the solution
for reducing carbon emissions.
—SNV
Front. Mar. Sci. (2024)
10.3389/fmars.2023.1125137
PROTEIN CRYSTALS
Crystalline immunity
In the laboratory, many proteins
can be coaxed into forming small,
delicate crystals. Natural protein
crystallization is a much rarer
phenomenon, but there are a
few examples in which functional
crystals form within cells or tis-
sues. Heyndrickx et al. studied the
proteins Ym1 and Ym2 from mice,
which are produced by certain
immune cells, and found that
administering a crystalline form
of Ym1, but not the soluble form,
stimulated both innate and adap-
tive immune responses. These
proteins are highly expressed
in response to allergens and
readily form crystals in vivo, but
the physiological mechanisms
and evolutionary origins of this
phenomenon need more study.
—MAF
eLife (2023)
10.7554/eLife.90676

602 9 FEBRUARY 2024 • VOL 383 ISSUE 6683 science.org SCIENCE

RES EARCH

◥ gated whether the number of fibroblasts is

RESEARCH ARTICLE SUMMARY
REPRODUCTION
Corpora cavernosa fibroblasts mediate
penile erection
Eduardo Linck Guimaraes, David Oliveira Dias, Wing Fung Hau, Anais Julien, Daniel Holl,
Maria Garcia-Collado, Soniya Savant, Evelina Vågesjö, Mia Phillipson,
Lars Jakobsson, Christian Göritz*
INTRODUCTION: Penile erection, a physiolog-
ical process crucial for sexual function, relies
on the intricate regulation of blood flow
within the sponge-like vascular bed of the
corpora cavernosa (CC). In the flaccid penis,
sympathetic release of the vasoconstrictor
norepinephrine maintains vascular smooth
muscle cells tonically contracted, restricting
penile blood flow. Upon sexual arousal,
nitric oxide and acetylcholine are released
from parasympathetic nerves, mediating vaso-
dilation through the relaxation of vascular
smooth muscle cells. The incoming blood
fills the CC, leading to penile erection. De-
spite the recognized importance of endo-
thelial and vascular smooth muscle cells in
the erectile process, the vast population
of fibroblasts in the CC has been largely
overlooked.
Notch
activity
Norepinephrine
availability
NE NO
Flaccid
Vasodilation
RATIONALE: Fibroblasts constitute the largest
cell population in the human CC, but their
physiological functions remain largely unex-
plored. Our aim was to elucidate the contri-
bution of CC fibroblasts to the regulation of
penile blood flow. Characterization of CC fi-
broblasts through single-cell gene expres-
sion profiling and histological analysis in
cleared tissue revealed their heterogeneity
and integration in the erectile tissue. Using
genetic targeting and optogenetic-induced
fibroblast depolarization, we found that fi-
broblasts actively participate in the regu-
lation of penile blood flow. Furthermore, by
chronically altering erection frequency using
chemogenetic modulation of brain regions re-
sponsible for arousal, we addressed the impact
of erection recurrency on fibroblast number
and blood flow regulation. Finally, we investi-
Erection
frequency
Fibroblast
number
NE NO
Erected
altered by aging and how a reduction in fibro-
blast number affects penile blood flow.
RESULTS: Our study revealed that fibroblasts
in the CC play a pivotal role in supporting vaso-
dilation by modulating norepinephrine avail-
ability. The efficacy of this process depends
on the number of fibroblasts, which is reg-
ulated by erectile activity. Penile erection tem-
porarily alters the spatial arrangement of
cells throughout the CC, leading to down-
regulation of Notch signaling in fibroblasts.
Inhibition of Notch signaling in fibroblasts
leads to a substantial increase in fibroblast
numbers, which can cause long-lasting erec-
tions characteristic of priapism. Constitutively
active Notch signaling decreases fibroblast
numbers and lowers penile blood perfusion.
Boosting the frequency of erections reduces
Notch signaling, increasing fibroblast num-
bers and promoting vasodilation. Conversely,
a reduction in erection recurrency increases
Notch signaling, decreasing the number of
fibroblasts and diminishing penile blood per-
fusion. Aging, one of the major risk factors
for erectile dysfunction, reduces the number
of penile fibroblasts and limits penile blood
perfusion. A reduction of penile fibroblasts in
young animals mimics the penile blood flow
phenotype of aged animals.
CONCLUSION: Fibroblasts, previously regarded
as static and homogeneous cells, are emerg-
ing as a dynamic and heterogeneous cell pop-
ulation. We discovered that CC fibroblasts
act as key blood flow regulators, shifting the
balance between vasodilators and the vaso-
constrictor norepinephrine toward vasodila-
tion. Notch signaling serves as a central hub
coordinating fibroblast turnover, norepineph-
rine sensitivity, and, ultimately, the erectile
process. The dynamic regulation of fibroblast
numbers coupled to erection recurrency un-
derscores the plasticity of erectile function.
This positive feedback loop may exacerbate
erectile dysfunction in chronic conditions
such as aging or diabetes. Indeed, our obser-
vations in aged animals suggest a potential
link between reduced fibroblast number and
erectile dysfunction, highlighting the clin-
ical relevance of understanding the cellular
mechanisms of erection. Overall, this study
provides a mechanism for modulation of penile
erection and establishes a foundation for fur-

vSMC contraction NE scavenging

ther research in the field of sexual health.
▪
stimulation via free NE via fibroblasts
The list of author affiliations is available in the full article online.
Ä

*Corresponding author. Email: christian.goeritz@ki.se

Fibroblasts mediate erectile activity–dependent modulation of penile blood flow. Notch signaling in
Cite this article as E. Linck Guimaraes et al., Science 383,
CC fibroblasts decreases upon erection. Increased erection frequency leads to proliferation and higher eade8064 (2024). DOI: 10.1126/science.ade8064
number of penile fibroblasts, which in turn reduces the availability of the vasoconstrictor norepinephrine (NE),
supporting penile erection. A chronically low erection frequency reduces fibroblast numbers and lowers READ THE FULL ARTICLE AT
penile blood flow. NO, nitric oxide; vSMC, vascular smooth muscle cell. https://doi.org/10.1126/science.ade8064

Guimaraes et al., Science 383, 604 (2024) 9 February 2024 1 of 1

RES EARCH

◥ ture, whereas cells from cluster 3 localized

RESEARCH ARTICLE mainly within the tunica albuginea, the largely
avascular outer region of the penis (fig. S1, H
REPRODUCTION to L, and fig. S2, H to J). Intravital confocal
and multiphoton imaging of mice injected
Corpora cavernosa fibroblasts mediate intravenously with fluorescein isothiocyanate
(FITC)–labeled dextran highlighted the high
penile erection density of tdTomato+ cells and their proximity
to circulating blood throughout the CC under
Eduardo Linck Guimaraes1, David Oliveira Dias1, Wing Fung Hau1, Anais Julien1, Daniel Holl1, physiological conditions (Fig. 1, H to J, and
Maria Garcia-Collado2, Soniya Savant1, Evelina Vågesjö3, Mia Phillipson3, movie S1).
Lars Jakobsson2, Christian Göritz1* Surface rendering of optically cleared tissue
and immunohistochemical analyses confirmed
Penile erection is mediated by the corpora cavernosa, a trabecular-like vascular bed that enlarges proximity to the vasculature and clear distinc-
upon vasodilation, but its regulation is not completely understood. Here, we show that perivascular tion of tdTomato+ fibroblasts from transgelin+
fibroblasts in the corpora cavernosa support vasodilation by reducing norepinephrine availability. (TAGLN) vSMCs and podocalyxin+ (PODXL)
The effect on penile blood flow depends on the number of fibroblasts, which is regulated by erectile endothelial cells in the CC (Fig. 1, K to M, and
activity. Erection dynamically alters the positional arrangement of fibroblasts, temporarily down- fig. S3, A to L). Compared with both vSMCs and
regulating Notch signaling. Inhibition of Notch increases fibroblast numbers and consequently raises endothelial cells, the number of tdTomato+ fibro-
penile blood flow. Continuous Notch activation lowers fibroblast numbers and reduces penile blood perfusion. blasts was three times higher in the CC (Fig. 1N).
Recurrent erections stimulate fibroblast proliferation and limit vasoconstriction, whereas aging reduces Altogether, we identified two SLC1A3-expressing
the number of fibroblasts and lowers penile blood flow. Our findings reveal adaptive, erectile perivascular fibroblast subsets in the CC.
activity-dependent modulation of penile blood flow by fibroblasts.
CC fibroblasts mediate vasodilation

I
n addition to reproduction, sexual activity
has several physiological and mental benefits
that affect the health of adult humans. For
men, sexual well-being largely depends on
the ability to attain penile erections, which
is compromised by aging and other risk fac-
tors (1, 2).
Penile erection is mediated through vasodila-
tion of the corpora cavernosa (CC), a trabecular-
like vascular bed that extends from the base of
the penis throughout the penile body into the
glans penis (3). In the flaccid penis, sympa-
thetic release of norepinephrine maintains
arterial and trabecular vascular smooth mus-
cle cells (vSMCs) tonically contracted, restrict-
ing penile blood flow to a basal level. Upon
sexual arousal, nitric oxide (NO) and acetylcho-
line are released from parasympathetic nerves
and mediate vasodilation through relaxation
of vSMCs along the central arteries and within
the CC. The incoming blood fills the CC, lead-
ing to an expansion of the organ. Shear stress–
induced endothelial NO maintains the erection
through continuous relaxation of vSMCs in the
CC (4). The balance between vasodilators and
the vasoconstrictor norepinephrine is deci-
sive for the establishment and maintenance
of penile erection.
Recently, fibroblasts have been recognized
in the human CC (5, 6), but whether they have
a role in blood flow regulation has not been
addressed.
1 populations, which can be identified by the
Department of Cell and Molecular Biology, Karolinska
Institutet, 171 77 Stockholm, Sweden. 2Department of
Medical Biochemistry and Biophysics, Division of Vascular
Biology, Karolinska Institutet, 171 77 Stockholm, Sweden.
3
Department of Medical Cell Biology, Division of Integrative
Physiology, Uppsala University, 751 23 Uppsala, Sweden.
*Corresponding author. Email: christian.goeritz@ki.se
The CC contains two subsets of
perivascular fibroblasts
Here, we identified a large population of cells
throughout the C C in mice that express the
transporter solute carrier family 1, member 3
(SLC1A3, synonym GLAST; gene name Slc1a3)
(fig. S1, A to C), a marker that has been pre-
viously used to label perivascular cells in the
central nervous system (7–9). To mark SLC1A3+
cells, we used Slc1a3-CreERT2 transgenic mice
(10) carrying a Rosa26-tdTomato reporter allele
(fig. S1D). Tamoxifen-induced genetic recombi-
nation in adult Slc1a3-CreERT2;Rosa26-tdTomato
mice led to tdTomato expression throughout
the CC (Fig. 1, A and B, and fig. S1, E to G). We
rarely found tdTomato+ cells close to the dorsal
vein and arteries (Fig. 1B and fig. S1, H to L).
To characterize the gene expression profile of
SLC1A3+ cells from the adult mouse penis, we
isolated tdTomato+ cells from recombined Slc1a3-
CreERT2;Rosa26-tdTomato mice by fluorescent-
activated cell sorting (FACS) and performed
single-cell RNA sequencing (fig. S2A). Sorted
cells expressed Slc1a3 and fibroblasts markers
such as Col1a1, Pdgfra, and S100a4, but little to
no vSMC and endothelial cell markers (Fig. 1,
C and D, and fig. S2, B to D). We confirmed the
expression of the fibroblast markers platelet-
derived growth factor receptor a (PDGFRa)
and S100 calcium binding protein A4 (S100A4)
in tdTomato+ cells at the protein level (Fig. 1,
E to G). Using unbiased clustering, we showed
that SLC1A3+ fibroblasts encompass three sub-
expression of Dnah5 for cluster 1, Cxcl12 for
cluster 2, and Gja1 for cluster 3 (Fig. 1, C and D,
and fig. S2, E to G). Cells from clusters 1 and 2
distributed throughout the subtunica and tra-
becular regions of the CC close to the vascula-
Optogenetic-induced membrane depolariza-
tion of mural cells has been used to study their
involvement in blood flow regulation in the
brain (11, 12). We used this strategy to investi-
gate the potential role of SLC1A3+ fibroblasts
in penile blood flow modulation. For compar-
ison, we applied the same approach to vSMCs,
which are known to constrict blood vessels
upon optogenetic-induced membrane depolar-
ization (13). To this end, we generated double-
transgenic mice expressing the light-gated cation
channel channelrhodopsin-2 (ChR2) fused to
tdTomato fluorescent protein (14) specifically
in penile fibroblasts (Slc1a3-CreERT2;Rosa26-
ChR2-tdTomato) or vSMCs (Tagln-CreERT2;
Rosa26-ChR2-tdTomato) after tamoxifen-induced
recombination (fig. S4A). As control, we used
Cre– animals from the same lines, which also
received tamoxifen without leading to ChR2-
tdTomato expression. Blood flow was assessed
noninvasively using high-resolution laser spec-
kle contrast imaging, which detects red blood
cell movement (fig. S4B) (15). Focal exposure of
the flaccid glans penis of Tagln-CreERT2;Rosa26-
ChR2-tdTomato mice to blue light (473 nm,
4.5 mW, 100 Hz, 8 ms pulse width, 8 s on/3 s off)
activated vSMCs and resulted in immediate
blood flow decrease in the exposed area. Ter-
mination of light exposure rapidly reestablished
basal blood flow (Fig. 1O). These results are
consistent with a previous report showing that
ChR2-induced depolarization of vSMCs led to
calcium-mediated vasoconstriction (13), vali-
dating our approach. By contrast, exposure of
the flaccid glans penis or penile body of Slc1a3-
CreERT2;Rosa26-ChR2-tdTomato mice with the
same regimen of blue light activated fibroblasts
and induced a steady rise in blood flow com-
pared with light-exposed control mice. After
termination of photoactivation, blood flow

Guimaraes et al., Science 383, eade8064 (2024) 9 February 2024 1 of 11

RES EARCH | R E S E A R C H A R T I C L E
Fig. 1. Perivascular fibroblasts of the CC mediate vasodilation in the
penis. (A) Schematics depicting anatomical structures of the mouse penis.
(B) Transverse section of the adult mouse penis showing vSMCs (TAGLN+
in green), endothelium (PODXL+ in cyan), and recombined perivascular
fibroblasts (tdTomato+ in red) in the C C of Slc1a3-CreERT2;R26R-tdTomato
mice. (C) Uniform manifold approximation and projection (UMAP) representation of
sorted tdTomato+ cells from Slc1a3-CreERT2;R26R-tdTomato adult mouse penis.
(D) Violin plots depicting the expression of lineage markers of fibroblasts, vSMCs,
and endothelial cells. (E to G) TdTomato+ cells expressing the fibroblast markers
S100A4 (E) and PDGFRa (F). (G) Merged image showing coexpression of TdTomato,
S100A4, and PDGFRa. (H and I) Intravital microscopy time-lapse images showing
the penile vasculature (H) and the high density of tdTomato+ cells in the native
C C and limited presence around the dorsal vasculature (DV) (I). (J) Intravital
multiphoton microscopy image showing the proximity of tdTomato+ fibroblasts
to circulating blood (FITC-dextran, i.v.). (K to M) Surface rendering and three-
dimensional (3D) visualization of the cleared C C showing tdTomato+ perivascular
Guimaraes et al., Science 383, eade8064 (2024) 9 February 2024
cells [(K), in red], vSMCs [(L), TAGLN+ cells in green and endothelial cells
(PODXL+) cells in cyan], and (M) the proximity between the three cell types.
(N) Quantification of tdTomato+ cells, vSMCs, and endothelial cells in the C C.
(O and P) Basal blood flow of the glans penis in response to optogenetic-induced
depolarization of vSMCs [(O), Tagln;R26R-ChR2 mice (n = 3)] and perivascular
fibroblasts [(P), Slc1a3;R26R-ChR2 mice (n = 3)]. R26R-ChR2 mice were used as
a control (both n = 3). Blue light (473 nm) was delivered focally on the glans
penis (4.5 mW, 100 Hz, 8 ms pulse width, 8 s on/3 s off). (Q) Changes in FITC-
dextran lumen cross-sectional area in response to intravital two-photon
optogenetic activation of tdTomato+ fibroblasts in the C C of Slc1a3;R26R-ChR2
(n = 4) and R26R-ChR2 control mice (n = 3). Data are shown as mean ± SEM. For
(N), one-way ANOVA followed by Holm-Šidák multiple-comparisons test was
used. For (O) to (Q), two-way ANOVA was used. DA, dorsal artery; TA, tunica
albuginea; U, urethra; Trb, trabecula. Cell nuclei are labeled with DAPI. Scale
bars: (B), 200 mm; (E) to (G) and (J), 25 mm; (H) and (I), 100 mm; (K) and
(L), 50 mm; (M), 10 mm.
2 of 11
RES EARCH | R E S E A R C H A R T I C L E

declined slowly (Fig. 1P and fig. S4C). To im-
prove spatial precision in photoactivation and
gain higher resolution, we performed multi-
photon imaging and optogenetic stimulation
of CC fibroblasts in the penile body (fig. S4D).
Focal optogenetic activation of ChR2-tdTomato+
fibroblasts in the trabecular and subtunica
regions of the CC increased vasodilation in the
exposed area compared with control mice with-
out ChR2 expression, as determined by the FITC-
dextran–filled trabecular lumen area (Fig. 1Q
and fig. S4, E and F). Compared with the slow
and steady increase in blood flow mediated by
fibroblast depolarization, exposure of the flac-
cid glans penis to NO or acetylcholine, the
main parasympathetic vasodilatory effectors,
caused a strong and immediate increase in
blood flow (fig. S5, A and B). The distinct ki-
netics suggest that optogenetic-induced mem-
brane depolarization of fibroblasts does not
cause parasympathetic stimulation but rather
continuously reduces basal vasoconstriction.
CC fibroblasts mediate vasodilation
by reduction of norepinephrine availability
In the flaccid penis, blood flow is maintained
at a basal level through the presence of the
vasoconstrictor norepinephrine (16). A steady
reduction of norepinephrine levels in response
to continuous fibroblast depolarization could
explain the observed increase in blood flow.
Indeed, we found that both tdTomato+ fibro-
blast subsets in the CC expressed the trans-
porter solute carrier family 6, member 2 (SLC6A2)
(fig. S6, A to D), which actively transports nor-
epinephrine into cells, preventing receptor bind-
ing (17). In addition, tdTomato+ cells in the CC
expressed monoamine oxidase A (MAO-A), an
essential enzyme for norepinephrine degrada-
tion (18) (fig. S6, E and F).
SLC6A2-mediated norepinephrine uptake
requires cell membrane localization of the
transporter, which is induced by membrane
depolarization (19, 20). Fibroblasts isolated
from the CC maintained SLC6A2 expression
in primary cultures (Fig. 2A) and, in response
to depolarization, significantly (P < 0.0001)
increased plasma membrane localization of
SLC6A2 (fig. S7, A to G). Consistent with this,
optogenetic-induced depolarization of pri-
mary cultured CC fibroblasts stimulated the
uptake of the fluorescent norepinephrine ana-
log FFN270 (21) (Fig. 2, B and C). Application
of nisoxetine (NES), a specific SLC6A2 inhibitor
(17), prevented depolarization-induced FFN270
uptake (Fig. 2, B and C). To investigate whether
SLC6A2 plays a role in fibroblast-mediated vaso-
dilation in vivo, we repeated the optogenetic-
induced depolarization of penile fibroblasts
in the presence of NES. NES injection into the
glans penis completely abrogated the rise in
blood flow induced by fibroblast photoactiva-
tion (Fig. 2, D and E, and fig. S8A). To di-
rectly test the regulation of norepinephrine
availability by SLC1A3+ fibroblasts and its im-
pact on blood flow, we pretreated the penis
with a NO donor to induce robust blood flow
and assessed the vasoconstriction potency of
the norepinephrine analog phenylephrine,
which is also transported by SLC6A2 (22).
After stabilization of NO-induced blood flow,
the penis was simultaneously exposed to blue
light and phenylephrine. In control animals
with no ChR2 expression, phenylephrine in-
duced immediate vasoconstriction, which per-
sisted for the entire duration of the experiment.
By contrast, animals with ChR2-expressing fibro-
blasts showed a steady increase in blood flow
upon light exposure, reflecting reduced bio-
availability of phenylephrine. This effect was
completely abolished in the presence of NES
(Fig. 2F and fig. S8B), suggesting that penile
fibroblasts facilitate vasodilation by SLC6A2-
mediated uptake of norepinephrine.
Penile erection dynamically alters the spatial
arrangement of fibroblasts and vSMCs
Next, we used optical tissue clearing and com-
pared the cellular arrangement of SLC1A3+
fibroblasts, vSMCs, and endothelial cells in the

Fig. 2. Penile fibroblasts mediate

vasodilation through the
norepinephrine transporter
SLC6A2. (A) Expression of the
norepinephrine transporter
SLC6A2 in primary cultures of
tdTomato+ penile fibroblasts.
(B and C) Representative images
(B) and quantification (C) of
the uptake of the norepinephrine
fluorescent analog FFN270 by
tdTomato+ penile fibroblasts
isolated from Slc1a3;R26R-tdTomato
(n = 3) and Slc1a3;R26R-ChR2
(n = 3) mice and Slc1a3;R26R-ChR2
fibroblasts treated with the
SLC6A2 inhibitor NES (Slc1a3;R26R-
ChR2+NES, n = 3) after exposure
to blue light (470 nm) in vitro.
(D) TdTomato+ fibroblasts in the
adult mouse CC express the
norepinephrine transporter SLC6A2.
(E) Penile blood flow in response
to focal illumination of the glans
penis with blue light in R26R-ChR2
(control; n = 3), Slc1a3;R26R-ChR2
(n = 4), and (Slc1a3;R26R-ChR2+NES)
(n = 4) mice, represented as
percentage of baseline blood flow. (F) Penile blood flow in response to focal as mean ± SEM. For (C), one-way ANOVA followed by Holm-Šidák multiple-
illumination of the glans penis with blue light and administration of the vasoconstrictor comparisons test was used. For (E) and (F), two-way ANOVA with Bonferroni’s
agent phenylephrine in R26R-ChR2 (n = 4), Slc1a3;R26R-ChR2 (n = 5), and Slc1a3; multiple-comparisons post hoc test was used. P values represent comparison of
R26R-ChR2+NES (n = 4) mice after NO-induced vasodilatation. For (E) and (F), blue the last time point between different groups. Cell nuclei are labeled with DAPI.
light: 473 nm, 4.5 mW, 100 Hz, 8 ms pulse width, 8 s on/3 s off). Data are shown Scale bars: (A) and (B), 10 mm; (D), 25 mm.

Guimaraes et al., Science 383, eade8064 (2024) 9 February 2024 3 of 11

RES EARCH | R E S E A R C H A R T I C L E

CC of the flaccid and erected penis (Fig. 3, A
to F, and fig. S9, A to F). In the flaccid state,
tdTomato+ fibroblasts and endothelial cells
showed a high cellular density, prominent in
the subtunica area along the outer rim of the
CC (Fig. 3, A and C, and fig. S9, A, C, and D).
vSMC fibers polarized in various directions
and had large contact areas with fibroblasts
(Fig. 3, A, C, and D). Erection was induced by
cavernous nerve stimulation (23) and, com-
pared with the flaccid state, the dilated penis
exhibited a largely expanded trabecular lu-
men, whereas the tunica albuginea was com-
pressed (Fig. 3B and fig. S9B). Fibroblasts,
vSMC fibers, and endothelial cells appeared
more elongated and aligned radially (Fig. 3,
B, E, and F, and fig. S9, B, E, and F). Although
the contact area between endothelial cells and
fibroblasts did not change (P = 0.52) (fig. S9G),
the contact between vSMCs and fibroblasts was
significantly (P = 0.021) reduced during erec-
tion (Fig. 3G), indicating that penile erection
dynamically alters proximity between perivas-
cular cell types.
Notch signaling is transiently active
in CC fibroblasts
The Notch signaling pathway orchestrates
cell turnover and blood flow through dynamic
changes in cell-to-cell contact in the bone
(24, 25) and could serve as a readout for the
transition between flaccid and erect penile
states. We found expression of the Notch1 and
Notch2 receptors, the Notch ligand Jagged
1 (Jag1), as well as the Notch downstream ef-
fector recombination signal binding protein
for immunoglobulin kappa J region (RBP-Jk,
synonym CBF1; gene name Rbpj), in Cxcl12+
(cluster 2) fibroblasts throughout the CC (fig.
S10, A to P), indicating that CC Cxcl12+ fibro-
blasts can both receive and transmit Notch
signals. Active Notch signaling can be assessed
at single-cell resolution in vivo using trans-
genic CBF-H2B-Venus reporter mice. In these
mice, binding of the Notch intracellular domain
(NICD), RBP-Jk, and mastermind-like (MAML)
DNA binding complex to cis-regulatory regions
leads to expression of the nuclear localized Venus
reporter protein (fig. S11A) (26). We detected
recently active or ongoing Notch signaling in
SLC1A3+ fibroblasts, endothelial cells, and vSMCs
throughout the CC (Fig. 3H and fig. S11, B to R).
When analyzing the penis of adult Slc1a3-
CreERT2;Rosa26-tdTomato;CBF-H2B-Venus mice
and assessing Venus mRNA expression in
combination with Venus protein expression,
we found that 10 ± 2.15% of tdTomato+ cells

Fig. 3. Penile erection causes

dynamic changes in cell-to-cell
contact and reduces Notch
signaling in fibroblasts of the
CC. (A and B) Cellular arrange-
ment of vSMCs (TAGLN+ in
green) and fibroblasts (tdTomato+
in red) in the flaccid (A) and
erected (B) penis (optically
cleared 200-mm-thick transverse
sections). (C to F) Representative
images showing the spatial
arrangement of fibroblasts (C) in
relation to vSMCs (D) in the
flaccid CC and fibroblasts (E) and
vSMCs (F) in the erected CC. Yellow
represents areas of fibroblast-
vSMC contact. (G) Quantification
of tdTomato+ fibroblasts associated
with vSMCs in the CC in the
flaccid and erected state (flaccid,
n = 4; erected, n = 4). (H) Expression
of the Notch reporter CBF:H2B-
Venus (white arrowheads) in
SLC1A3+ CC fibroblasts.
(I) Expression of H2B-Venus
mRNA and protein in tdTomato+
fibroblasts as percentage of the
total tdTomato+ fibroblast popula-
tion in the CC. (J and K) H2B-
Venus mRNA (J) or Hes1 mRNA (K)
expression in tdTomato+ fibroblasts
of control and cavernous nerve
(CN)–stimulated penises, repre-
sented as the number of mRNA
puncta in tdTomato+Venus mRNA+
or tdTomato+ cells, respectively.
Data are shown as mean ± SEM.
For (G), (J), and (K), Mann-Whitney
test was used. Cell nuclei are
labeled with DAPI. Scale bars: (A)
and (B), 200 mm; (C) to (F) and
(H), 25 mm.

Guimaraes et al., Science 383, eade8064 (2024) 9 February 2024 4 of 11

RES EARCH | R E S E A R C H A R T I C L E
expressed the Venus protein, 16 ± 0.9% ex-
pressed the Venus mRNA, whereas only 2 ±
0.4% expressed both (Fig. 3I). These results
indicate that Notch signaling is transiently
activated in fibroblasts of the CC.
Frequent erections lead to decreased Notch
signaling in CC fibroblasts
To investigate whether penile erection modu-
lates Notch signaling in fibroblasts of the CC,
we performed cavernous nerve stimulation in
recombined Slc1a3-CreERT2;Rosa26-tdTomato;
CBF-H2B-Venus mice. After 1 hour of repeated
stimulation of the cavernous nerve (fig. S12A),
Venus mRNA expression was significantly (P =
0.019) reduced in tdTomato+ fibroblasts of the
CC (Fig. 3J and fig. S12B). Consistent with this,
we also detected down-regulation of Notch2,
Jag1, and the downstream Notch effector Hes1
in tdTomato+ fibroblasts after cavernous nerve
stimulation, whereas genes unrelated to Notch
signaling remained unaltered (Fig. 3K and fig.
S12, C to K). Our results show that Notch sig-
naling is dynamically regulated in fibroblasts
of the CC and reduced upon penile erection.
Inhibition of Notch signaling in CC
fibroblasts increases their cell number
and raises penile blood flow
To investigate whether modulation of Notch
signaling affects penile blood flow, we exper-
imentally reduced Notch signaling in fibro-
blasts by conditional deletion of RBP-Jk, an
obligatory DNA-binding protein for canoni-
cal Notch signaling (27), using Slc1a3-CreERT2;
Rosa26-tdTomato mice homozygous for con-
ditional Rbp-Jk null alleles (RBP-Jkfl/fl) (28). In
a complementary experiment, we constitutive-
ly activated Notch signaling in penile fibro-
blasts using Slc1a3-CreERT2;Rosa26-tdTomato
mice crossed to transgenic Rosa26Notch mice
(29), which allow Cre-mediated expression of
NICD upon tamoxifen administration (fig. S13A).
Two months after RBP-Jk deletion, basal blood
flow in the penile body and glans penis were
significantly (P = 0.0003 and P = 0.004, re-
spectively) increased compared with control
animals with unaltered Notch signaling. In
turn, NICD animals with constitutively active
Notch signaling had low basal blood flow com-
pared with RBP-Jk fl/fl animals (Fig. 4, A and
B, and fig. S13, B and C). Tissue analyses re-
vealed that deletion of RBP-Jk significantly (P =
0.0063) increased the number of tdTomato+
fibroblasts in the CC, whereas NICD expres-
sion reduced the number of tdTomato+ cells
compared with control mice (Fig. 4, C and D).
The number, cell surface, and cell volume of en-
dothelial cells and vSMCs were not altered by
Notch signaling modulation in fibroblasts (fig.
S13, D to K). Brain-specific deletion of RBP-Jk
in SLC1A3-expressing perivascular cells (fig.
S14, A to E) did not alter basal penile blood
flow nor response to phenylephrine (fig. S14,
Guimaraes et al., Science 383, eade8064 (2024) 9 February 2024
F to H). Altogether, these results show that
reduced Notch signaling in fibroblasts of the
CC increased their number and led to increased
blood perfusion. In turn, constitutive activation
of Notch signaling reduced the number of fibro-
blasts in the CC and decreased blood perfusion.
Modulation of Notch signaling did not change
the average expression level of SLC6A2 in
tdTomato+ fibroblasts (fig. S15, A and B) but
did alter the overall abundance of SLC6A2 in the
CC through the change in fibroblast numbers
(Fig. 4E and fig. S15C).
The number of CC fibroblasts regulates
penile blood flow
To test whether the number of SLC1A3+ fibro-
blasts in the CC affects norepinephrine-mediated
blood flow regulation, we deleted RBP-Jk or in-
duced constitutive expression of NICD in penile
fibroblasts. Two months after recombination,
we exposed the penis to phenylephrine after
pretreatment with a NO donor. In control mice
with unaltered Notch signaling, phenylephrine
sharply reduced penile blood flow, efficiently
counteracting NO-induced vasodilation for
the duration of the experiment. By contrast,
phenylephrine-induced blood flow reduction
was less effective in RBP-Jkfl/fl animals, which
present an increased number of CC fibroblasts.
Indeed, we noted that the vasoconstrictor effect
of phenylephrine was completely abolished
after 5 min, and blood flow steadily increased
thereafter. The effectiveness of phenylephrine
in reducing blood flow was completely reestab-
lished in RBP-Jkfl/fl mice in the presence of
the SLC6A2 inhibitor NES. In NICD-expressing
animals, which present fewer CC fibroblasts,
phenylephrine induced immediate, long-lasting,
vasoconstriction (Fig. 4F). Norepinephrine in-
duces vSMC contraction through Rho-associated
protein kinase (ROCK) activation and consequent
phosphorylation of myosin phosphatase target
subunit-1 (MYPT-1) and C-kinase potentiated
protein phosphatase-1 inhibitor (CPI-17) (30–33).
We observed that the levels of MYPT-1 and CPI-17
phosphorylation were decreased in RBP-Jk fl/fl
mice but increased in the NICD group and
were inversely correlated with the number of
tdTomato+ cells (fig. S16, A to F), further indi-
cating that CC fibroblasts can limit norepi-
nephrine availability. Consistent with these
findings, we observed that a subset of RBP-Jk
deleted animals presented atypical long-lasting
erections, featuring a largely increased glans
penis and reduced blood perfusion, character-
istic of ischemic priapism (Fig. 4, G to I) (34).
This phenotype was consistently observed across
several experiments from 2 months after RBP-Jk
deletion. Because the efficiency of tamoxifen-
inducible Cre-mediated recombination can vary
(8), it may explain why this phenotype only
appeared in a subset of mice. Indeed, animals
presenting priapism contained a higher num-
ber of tdTomato+ fibroblasts in the CC compared
with nonpriapic animals of the same genotype
and age (Fig. 4J), indicating that penile fibro-
blasts can effectively diminish vasoconstriction.
The frequency of erections determines
fibroblast proliferation
Because inhibition of Notch signaling increases
the fibroblast cell population (Fig. 4, C and D),
and repeated stimulation of erection can ac-
tively modulate Notch signaling (Fig. 3, J and
K, and fig. S12, A to G), we tested whether the
frequency of erections affects the number of
fibroblasts in the CC and penile blood flow.
For that, we induced or inhibited penile erec-
tion in a physiological context by chemoge-
netic activation or inhibition of brain neurons
in areas known to control erection, such as the
medial preoptic area (mPOA) and paraventric-
ular nucleus of the hypothalamus (PVN) (Fig. 5A),
(35–37). Neurons in the mPOA project to oxytocin
(OXT)–expressing neurons in the PVN (38) that
are vital for sexual behavior and induce pe-
nile erection when activated by glutamate and
analogs (37, 39, 40). Using viral constructs, we
expressed hM3D(Gq)-mCherry or hM4D(Gi)-
mCherry, an excitatory or an inhibitory designer
receptor exclusively activated by designer drugs
(eDREADD or iDREADD, respectively), or the
reporter gene mCherry alone (mock), in calcium/
calmodulin–dependent protein kinase IIa–positive
(CaMKIIa+) neurons of the mPOA and PVN
(Fig. 5A and fig. S17, A to C). CaMKIIa+ neurons
showed a similar level of mCherry expression
in animals transduced with mock, eDREADD,
and iDREADD viral constructs (fig. S17D). Neu-
rons expressing mCherry or OXT were found
intermingled in the same region but did not
overlap (fig. S17, B and E to G). We confirmed
that, compared with the mock-transduced
group, activation of eDREADD or iDREADD
by intraperitoneal injection of the DREADD
ligand clozapine N-oxide (CNO) increased or
decreased neuronal activity in mCherry+ and
OXT+ neurons, respectively, as shown by c-fos
expression (fig. S17, E to K). A single injec-
tion of CNO induced a fourfold increase or
a fourfold decrease in the number of erec-
tions per hour in freely moving animals ex-
pressing eDREADD or iDREADD, compared
with mock-transduced animals, respective-
ly (Fig. 5B). To determine whether chronic
increase or decrease in the frequency of erec-
tions modulates the number of fibroblasts in
the CC, we administrated CNO daily (5 mg/kg
in saline) for 4 weeks to recombined Slc1a3-
CreERT2;Rosa26-tdTomato mice transduced
with eDREADD, iDREADD, or mock viral con-
structs in the mPOA and PVN while supplying
the thymidine analog 5-ethynyl-2′-deoxyuridine
(EdU) in the drinking water (Fig. 5A). We ob-
served that boosting the frequency of penile
erections significantly increased the genera-
tion of tdTomato+ fibroblasts, as measured by
EdU incorporation, as well as the total number
5 of 11
RES EARCH | R E S E A R C H A R T I C L E

Fig. 4. Modulation of Notch signaling in A B C P<0.0001

Slc1a3;Rbp-Jκfl/fl Control Slc1a3;NICD P=0.0001 P=0.0063
SLC1A3+ fibroblasts alters their cell 250 1200
P=0.004

No. tdTomato+ cells / section

number and penile blood flow response P=0.006
P=0.069
to phenylephrine. (A and B) Representative 200

Perfusion units
images (A) and quantification (B) of glans 800
150
penis (dashed lines) basal blood flow in
Slc1a3;RBP-Jkfl/fl (n = 6), control (n = 9), 100
400
and Slc1a3;NICD (n = 8) animals. (C and
D) Quantification (C) and representative 50

images (D) of tdTomato+ cells in the CC of 0 0

Slc1a3;RBP-Jkfl/fl (n = 4), control (n = 5), and

fl
fl

l
fl/
D

D
fl/

tro

tro
κ
IC

IC
κ

on

on
-J
-J
Slc1a3;NICD (n = 5) animals 8 weeks after

;N

;N
bp
Blood flow

bp

C
a3

a3
;R
;R

c1

c1
recombination. (E) Quantification of SLC6A2

a3
a3

Sl

Sl
c1
c1
expression in the CC of Slc1a3;RBP-Jkfl/fl

Sl
Sl
D Slc1a3;Rbp-Jκ fl/fl
Control Slc1a3;NICD
(n = 4), control (n = 4), and Slc1a3;NICD
(n = 4) animals. (F) Penile blood flow
presented as percentage of baseline (NO
donor–induced blood flow) upon exposure
of the glans penis to the vasoconstrictor
agent phenylephrine (indicated by black
arrowhead) in control (n = 8), Slc1a3;RBP-
Jkfl/fl (n = 6), Slc1a3;NICD (n = 8), and
Slc1a3;RBP-Jkfl/fl animals pretreated with
NES (Slc1a3;RBP-Jkfl/fl+NES) (n = 3).
(G) Representative images of control and tdTomato DAPI tdTomato DAPI tdTomato DAPI
Slc1a3;RBP-Jkfl/fl animals 2 months after Control
recombination, demonstrating the presence E F Slc1a3;Rbp-Jκfl/fl
2.5 P=0.024
of a flaccid penis and long-lasting penile Phenylephrine Slc1a3;NICD
SLC6A2 intensity / section (a.u.)

P=0.043 120 Slc1a3;Rbp-Jκfl/fl+NES

erection characteristic of priapism, respec- 2.0
(% NO-induced blood flow)

tively. (H and I) Representative images (H) 100

and quantification (I) of blood flow in the 1.5 P=0.55

P<0.0001
P<0.0001
P<0.0001
80
Blood flow

flaccid (control; n = 9) and priapic (Slc1a3;

RBP-Jk fl/fl; n = 5) glans penis (dashed 1.0 60

lines). (J) Number of tdTomato+ cells in the 40

NSNS
CC of nonpriapic Slc1a3;RBP-Jk fl/fl (n = 5)

NS
0.5
and priapic Slc1a3;RBP-Jkfl/fl (n = 5) 20

animals. Data are shown as mean ± SEM. 0

0
0 2 4 6 8 10
l

For (B), (C), and (E), one-way ANOVA

κ fl/f

D
tro

IC

minutes
on
-J

;N
bp

followed by Holm-Šidák multiple-compari-

C

a3
;R

c1
a3

sons test was used. For (F), two-way I J 1800 P=0.0054

Sl

P=0.045
c1

150
Sl

ANOVA followed by Bonferroni’s multiple-

No. tdTomato+ cells / section

comparisons test was used. P values
G H
Blood flow (% control)

Flaccid Priapic
represent comparison of the last time point Flaccid Priapic 100 1200
between different groups. For (I) and (J),
two-sided unpaired Student’s t test was
used. Cell nuclei are labeled with DAPI. 50 600
Scale bar in (D), 100 mm.

0 0
c

l
l
id

/f
κ fl/f

sm κ fl
pi
c

ia
ac

pi p-J
-J

Blood flow
Pr
Fl

bp

ria b
+p ;R
; R

a3
a3

c1
c1

Sl
Sl

of tdTomato+ fibroblasts in the CC, compared animals and more pronounced in iDREADD in fibroblasts of the CC were reduced or raised
with mock-transduced animals (P < 0.0001 animals (Fig. 5F). The number, cell surface, in response to chronic increase or decrease in
and P = 0.005, respectively). Chronic decrease and cell volume of endothelial cells (fig. S18, A erection frequency, respectively (fig. S19, A to
in the frequency of erections reduced fibro- to D) and vSMCs (fig. S18, E to H) did not dif- F). Altogether, these results suggest that an
blast proliferation and resulted in lower fibro- fer among eDREADD, iDREADD, and mock- increased frequency of penile erections low-
blast numbers (Fig. 5, C to E). Consistent with transduced animals. We observed a decrease ers the sensitivity to norepinephrine through
the altered numbers of tdTomato+ fibroblasts in basal blood flow in the iDREADD group induction of proliferation in fibroblasts of
in the CC, phenylephrine-induced vasocon- compared with eDREADD animals (fig. S18I). the CC. Inversely, chronic reduction of erec-
striction was less prominent in eDREADD The expression levels of Hes1, Notch2, and Jag1 tion frequency limits fibroblast proliferation,

Guimaraes et al., Science 383, eade8064 (2024) 9 February 2024 6 of 11

RES EARCH | R E S E A R C H A R T I C L E

P=0.019
A B C 12 P=0.015
D P=0.005 P<0.0001
10 P=0.001 1500
P<0.0001 P<0.0001
AAV8:CaMKII -hM3D(Gq)-mCherry (eDREADD) P<0.0001 P<0.0001

No. tdTomato+ cells / section

AAV8:CaMKII -hM4D(Gi)-mCherry (iDREADD)
8

% tdTomato+EdU+ cells
/ total tdTomato+ cells
AAV8:CaMKII -mCherry (Mock)

No. of erections / hour

8 1000
6

PVN 4
4 500
Virus injection Daily injection of CNO
+
TAM EdU in water
2

5 days 1 month 1 month 1 month

0 0 0

D
D

D
k
k

k
D
D

D
oc

D
oc

oc

D
EA
EA

EA

EA

EA

EA
M
M

M
R
R

R
iD
eD

eD

iD

eD

iD
E F Phenylephrine
Slc1a3-CreERT2;R26R-tdTomato
Mock eDREADD iDREADD Mock

Blood flow (% NO-induced blood flow)

100
eDREADD
iDREADD
80

60

P = 0. 0 2
P=0.003
P>0.9
40

EdU tdTomato DAPI EdU tdTomato DAPI EdU tdTomato DAPI 20

0 5 10 15
minutes

Fig. 5. Erection frequency modulates fibroblast proliferation and
sensitivity to norepinephrine. (A) Experimental design and schematic
representation of viral injections with AAV8:CaMKIIa-hM3D(Gq)-mCherry
(eDREADD), AAV8:CaMKIIa-hM4D(Gi)-mCherry (iDREADD) and AAV8:CaMKIIa-
mCherry (mock) in the mPOA and PVN of Slc1a3-CreERT2;R26R-tdTomato
mice. (B) Number of erections per hour in freely moving animals transduced
with mock viral constructs (n = 12), eDREADD (n = 3), or iDREADD (n = 8) after
CNO injection. (C and D) Percentage of tdTomato+EdU+ cells (C) and total
number of tdTomato+ cells (D) in the CC of mock-transduced (n = 11), eDREADD
(n = 7), and iDREADD (n = 8) animals after daily injection of CNO for 30 days.
(E) Representative immunofluorescence images of proliferating (EdU+)
tdTomato+ cells (white arrowheads). (F) Quantification of penile blood flow of
mock-transduced (n = 9), eDREADD (n = 5), and iDREADD (n = 8) animals
after phenylephrine exposure (arrowhead), represented as percentage of
baseline (NO donor-induced blood flow). Data are shown as mean ± SEM. For
(B), (C), and (D), one-way ANOVA followed by Holm-Šidák multiple-comparisons
test was used. For (F), two-way ANOVA followed by Bonferroni’s multiple-
comparisons test was used. P value on the right side refers to the comparison
between groups for the last time point. Cell nuclei are labeled with DAPI.
Scale bar in (E), 50 mm.

diminishing their number and increasing sen-
sitivity to norepinephrine.
Aging reduces the number of fibroblasts in
the CC and lowers penile blood flow
Aging is accompanied by a reduction in sex-
ual activity and an increased risk for erectile
dysfunction (41). To test whether aging alters
the number of fibroblasts, we compared aged
(14 months old) with young (2 months old) ani-
mals. We found that the aged penis contained
a significantly lower number of tdTomato+
fibroblasts in the CC (P = 0.043) (Fig. 6, A to
C) and presented lower basal penile blood flow
(Fig. 6D), whereas the number of endothelial
cells and vSMCs remained similar between
young and aged mice (fig. S20, A to D).
Because aging can affect systemic blood flow
(42, 43), we could not ascertain whether the
reduced basal penile blood flow observed in
aged mice is a consequence of lower num-
bers of fibroblasts in the CC or if it is related to
systemic blood flow changes. To test this directly,
we experimentally reduced the number of fibro- higher sensitivity to phenylephrine compared
blasts in young animals by using Slc1a3-CreERT2; with control animals (fig. S22, G to I). Local ab-
Rosa26-tdTomato mice heterozygous for con- lation of fibroblasts did not change the number
ditional inducible diphtheria toxin recep- of recombined perivascular cells in the brain (fig.
tor (iDTR) alleles. In these mice, tamoxifen- S22, J and K). Altogether, these results show that
induced recombination leads to iDTR expression aging reduces the number of CC fibroblasts,
in SLC1A3+ fibroblasts, enabling their ablation which increases the sensitivity to phenylephrine
upon diphtheria toxin (DT) administration (44) and lowers basal penile blood flow, suggesting
(fig. S21, A and B). As control, we used Slc1a3- that fibroblasts might contribute to aging-
CreERT2;Rosa26-tdTomato mice, which followed related erectile dysfunction.
the same injection scheme and received the
same dose of tamoxifen and DT as iDTR mice. Discussion
Although endothelial cells and vSMCs remained In this study, we show that perivascular fibro-
unaffected (fig. S21, C to J), we achieved a similar blasts in the CC modulate penile blood flow by
relative reduction in the number of tdTomato+ shifting the balance between vasodilators and
fibroblasts in the CC of iDTR mice upon sys- the vasoconstrictor norepinephrine toward
temic (Fig. 6, E to G) or local (fig. S22, A to C) vasodilation. Sympathetic projections into pe-
DT administration, as observed between aged ripheral tissues release norepinephrine not only
and young animals (Fig. 6, A to C). The re- through nerve terminals, but also in a more
duction of fibroblasts in the CC of young diffuse manner by varicosities (45). Because of
animals led to a comparable decrease in basal this ubiquitous release of norepinephrine in
penile blood flow as observed in aged animals the flaccid penis, it is vital that a rapid and
(Fig. 6, D and H, and fig. S22, D to F) and a efficient mechanism of clearance is in place to

Guimaraes et al., Science 383, eade8064 (2024) 9 February 2024 7 of 11

RES EARCH | R E S E A R C H A R T I C L E

Fig. 6. Aging decreases SLC1A3+ Slc1a3-CreERT2;R26R-tdTomato

P=0.043
fibroblasts in the CC and A Young B Aged C 1.2 D 1.5
reduces basal blood flow, which P=0.012

No. tdTomato+ cells / section

(relative to young animals)

1.0

(relative to young animals)

can be recapitulated by depleting
SLC1A3+ fibroblasts in young

Basal blood flow

0.8 1.0
animals. (A and B) Representative
immunofluorescence images 0.6
of the CC of young (2 months old)
0.4 0.5
(A) and aged (14 months old)
T2
(B) Slc1a3-CreER ;R26R-tdTomato 0.2
mice. (C) Quantification of tdTomato+
0 0
cells in the CC of young (n = 4)

ng

ng
ed

ed
u
and aged (n = 3) animals. (D) Basal

u
Ag

Ag
tdTomato DAPI tdTomato DAPI

Yo

Yo
penile blood flow in aged (n = 4) Slc1a3-CreERT2;R26R-tdTomato
P=0.019
relative to young (n = 4) animals. E Control F iDTR H
G 1.2 1.5
(E and F) Representative

(relative to control animals)

No. tdTomato+ cells / section
P=0.004

(relative to control animals)

immunofluorescence images of 1.0
tdTomato+ cells in the CC of

Basal blood flow

0.8 1.0
Slc1a3-CreERT2;R26R-tdTomato
T2
(control) (E) and Slc1a3-CreER ; 0.6
R26R-tdTomato;Rosa26-iDTR
0.4 0.5
(iDTR) (F) animals. (G) Quantification
+
of tdTomato cells in the CC of 0.2
Slc1a3-CreERT2;R26R-tdTomato;
0 0
R26R-iDTR (iDTR, n = 4) relative to

l
TR
l

tro
tro

TR
T2
Slc1a3-CreER ;R26R-tdTomato

iD
on
tdTomato DAPI tdTomato DAPI

on

iD

C
C
(control, n = 3) animals. (H) Basal
penile blood flow in iDTR animals
(n = 5) relative to control (n = 4). Data are shown as mean ± SEM. For (C), (D), (G), and (H), two-sided unpaired Student’s t test was used. Cell nuclei are labeled
with DAPI. Scale bars in (A), (B), (E), and (F), 100 mm.

facilitate an erection. Fibroblasts, previously
regarded as static and homogeneous cells,
are emerging as a dynamic and heterogeneous
cell population (46). Although fibroblasts con-
stitute the largest cell population in the hu-
man CC (5, 6), their physiological functions
remain largely unexplored. We demonstrate
here that perivascular fibroblasts in the mouse
CC regulate norepinephrine availability through
SLC6A2. Membrane depolarization of fibro-
blasts stimulates norepinephrine uptake by
inducing SLC6A2 translocation to the cell mem-
brane. Additionally, the number of fibroblasts
in the CC can fine-tune blood flow regulation.
Increased erection frequency stimulates pro-
liferation, resulting in a higher number of CC
fibroblasts, elevated basal blood flow, and
reduced norepinephrine sensitivity. Consistent
with this, frequent erections, such as noctur-
nal penile tumescence, are positively correlated
with sexual potency, and an increase in erec-
tile activity by daily intake of phosphodiester-
ase inhibitors benefits patients with erectile
dysfunction (47, 48). Excessive penile fibroblasts,
however, can limit norepinephrine availability,
leading to prolonged, non-arousal-dependent
erections, a hallmark of priapism (34). Con-
versely, inflammation-induced fibrosis acti-
vates fibroblasts aberrantly, impeding erections
through fibrotic tissue formation (49). Indeed,
CC fibroblasts in patients with erectile dys-
function undergo gene expression changes,
adopting a myofibroblast-like phenotype (5).
Moreover, we demonstrate that decreased erec-
tion frequency results in reduced fibroblast
proliferation and cell count, thereby enhancing
norepinephrine sensitivity and potentially
influencing erection thresholds. In chronic
conditions such as aging or diabetes, this pos-
itive feedback loop may exacerbate erectile
dysfunction.
Aging, which is marked by reduced sexual
activity, often coincides with a higher inci-
dence of erectile dysfunction (50). We found
that aged animals presented fewer CC fibro-
blasts compared with younger animals. When
we experimentally reduced fibroblast num-
bers in young mice to mimic conditions of an
aged penis, we observed a corresponding de-
crease in blood flow. This suggests that a grad-
ual, chronic reduction in penile fibroblasts
might suffice to affect penile blood flow and
aggravate erectile dysfunction, especially in
conjunction with other risk factors. Fibroblasts
in the human CC express Notch ligands and re-
ceptors, allowing for vital cell-to-cell commu-
nication with all surrounding vascular and
perivascular cell types (5). We demonstrate
that Notch signaling in CC fibroblasts is dy-
namically modulated by erections, consequent-
ly influencing the number of cells present in
the penile vasculature. Thus, Notch signaling
serves as a central hub coordinating fibroblast
turnover, norepinephrine sensitivity, and, ulti-
mately, penile blood flow and the erectile pro-
cess. Furthermore, risk factors associated with
erectile dysfunction, such as aging and diabetes,
have been shown to influence Notch signaling
in the vasculature of other organs (25, 51). This
suggests that in a chronic context, disruption
of this pathway could play a role in exacerbat-
ing the adverse effects of these risk factors,
ultimately disrupting the erectile process. We
uncovered that a decrease in Notch signaling
(temporarily occurring during erection) in-
creases fibroblast numbers, whereas overacti-
vation of the Notch pathway lowers fibroblast
numbers. Therefore, conditions that limit erec-
tion frequency or directly enhance Notch signal-
ing in CC fibroblasts would lead to a reduction
in the fibroblast population over time.
Current therapies for the treatment of erec-
tile dysfunction mainly focus on increasing
vSMC relaxation by phosphodiesterase inhib-
itors, amplifying the effect of NO. Neverthe-
less, these therapies fail to restore erection in
up to 30% of patients, leaving few alternative
treatments available (52). Factors detrimen-
tal for erection lead to a reduction in perivas-
cular cells, paralleled by a decrease in vSMCs
and endothelial cells (53–55), hinting that
reestablishment of the fibroblast population
might be beneficial for erectile function. Our
results show that Notch-dependent dynamic
changes in fibroblast cell number affect the
fine-tuned regulation of penile blood flow, sug-
gesting that modulation of Notch signaling
in penile fibroblasts might open new therapeutic
alternatives for treatment of erectile dysfunction.

Guimaraes et al., Science 383, eade8064 (2024) 9 February 2024 8 of 11

RES EARCH | R E S E A R C H A R T I C L E
Methods summary
Transgenic mice
All animal procedures were performed in ac-
cordance with the Swedish and European
Union guidelines and approved by the insti-
tutional ethical committee (Stockholms Norra
Djurförsöksetiska Nämnd). Mouse colonies
were maintained on a 12:12 hour light:dark
cycle with free access to food and water. All
animals were in a C57BL/6 genetic background
except for the CBF:H2B-Venus mouse line,
which was in a BALB/cJ background. All ex-
periments were performed with animals aged
8 to 12 weeks unless stated otherwise. The
Notch reporter transgenic mouse line CBF:H2B-
Venus has been previously described [(26);
Tg(Cp-HIST1H2BB/Venus)47Hadj/J, obtained
from the Jackson Laboratory, JAX stock:
020942)]. Slc1a3-CreERT2 (10) transgenic mice
were crossed to the R26R-tdTomato Cre-reporter
line (B6.Cg-Gt(ROSA)26Sortm14(CAG-tdTomato)Hze/J,
obtained from Jackson Laboratory, JAX stock:
007914) to generate Slc1a3-CreERT2;R26R-tdTomato
mice. This line was further crossed to RBP-Jk fl/fl
(28), R26R-NICD (also known as transgenic
Rosa26Notch) (29) or R26R-iDTR (C57BL/6-
Gt(ROSA)26Sortm1(HBEGF)Awai/J, obtained from
Jackson Laboratory, JAX stock: 007900) mice
to generate Slc1a3-CreERT2;R26R-tdTomato;
RBP-Jk fl/fl, Slc1a3-CreERT2;R26R-tdTomato;R26R-
NICD or Slc1a3-CreERT2;R26R-tdTomato;R26R-
iDTR mouse lines, respectively. Slc1a3-CreERT2
and Tagln-CreERT2 (56) transgenic mice were
crossed to the R26R-ChR2-tdTomato line (B6.
Cg-Gt(ROSA)26Sortm27.1(CAG-COP4*H134R/tdTomato)Hze/
J, obtained from Jackson Laboratory, JAX stock:
012567) to generate Slc1a3-CreERT2;R26R-ChR2-
tdTomato or Tagln-CreERT2;R26R-ChR2-tdTomato
mice, respectively.
Blood flow analysis
Anesthesia was induced using isoflurane until
mice were unconscious and reduced to 1 to
1.2% isoflurane. Mice were placed on a heating
pad (37°C) for the duration of the blood flow
experiment. For measurements of basal blood
flow, the penis was exposed by gradually press-
ing the animal’s abdomen until the glans penis
surfaced, followed by organ immobilization.
For recordings of penile body blood flow, a
small abdominal incision was performed, and
the penile body was carefully exposed. Blood
flow of the glans penis or penile body was
recorded 2 to 5 min after organ positioning
with a focal distance of 10 ± 2 cm using the
blood perfusion imager PeriCam PSI System
(Perimed AB, Stockholm, Sweden). For experi-
ments testing the effect of phenylephrine on
penile blood flow, the mice were injected with
2 ml of the NO donor sodium nitropusside
(SNP, 25 mM, Abcam) in the glans penis to
induce a rise in blood flow. After 8 to 10 min,
blood flow levels reached a plateau and stabi-
lized, at which time 1 ml of phenylephrine (5 mM)
Guimaraes et al., Science 383, eade8064 (2024) 9 February 2024
was injected. In experiments in which the
SLC6A2 inhibitor NES was used, 1 ml of NES
(10 mM, Abcam) was injected in the glans penis
10 min before blood flow recording.
For optogenetic stimulation, a multimode
optical fiber (200 mm and 0.39 numerical aper-
ture; ThorLabs) supported by a stereotaxic ap-
paratus and positioned at 1- to 2-mm distance
directly above the glans penis or penile body
was used for light delivery. The following pa-
rameters were used: 4.5 mW power, 100 Hz
frequency, and 8 ms pulse width for a period
of 8 s on and 3 s off.
In vivo multiphoton optogenetic stimulation
of penile fibroblasts
Slc1a3-CreERT2;R26R-ChR2-tdTomato (referred
to as Slc1a3;R26R-ChR2) received tamoxifen
(as described in the materials and methods
section of the supplementary materials) to in-
duce ChR2-tdTomato expression. CreWT litter-
mates (referred to as control) received the same
tamoxifen regimen, but no ChR2-tdTomato ex-
pression was induced.
After a period of at least 30 days after recom-
bination, animals were anesthetized by 4%
isoflurane until unconscious, followed by 2%
isoflurane, and placed in a supine position on
a heating pad (37°C). Next, animals received a
tail vein injection of 100 ml of a mixture of
FITC-dextran (10 mg/ml, 70 kD, 46945; Sigma-
Aldrich) and Cascade Blue-dextran (5 mg/ml,
10 kD, D1976; Life Technologies). The penile
body was exposed by a midline abdominal in-
cision and carefully placed in a 1-mm-deep
groove of a metallic cube (dimensions, 1 × 1 ×
1 cm) and then stabilized. Next, the penis was
placed under a 25×/1.0 water immersion ob-
jective (507704, Leica Microsystems) of a TCS
SP8 multiphoton microscope (Leica Microsys-
tems). First, the penis was screened by two-
photon microscopy using laser at 1000 nm
(<20 mW power) to identify regions of interest
containing ChR2-tdTomato-expressing fibro-
blasts and trabecular lumens filled with FITC-
labeled dextran in the mouse CC. This laser
wavelength and power avoid ChR2 activation
and unspecific heat-induced vessel dilation,
respectively (57). The same procedure was
used with CreWT animals. Once the region of
interest was identified, ChR2 activation and
time-lapse recording of cross-sectional lumen
area over time was performed using the fol-
lowing parameters: 825 nm laser exposure,
1.2 ms dwell time, and 200 Hz bidirectional
scanning speed. Changes in FITC-dextran–
filled lumen cross-sectional area were quan-
tified using Imaris software (version 9.6.0,
Oxford Instruments).
Chemogenetic activation of CaMKIIa+ PVN and
mPOA neurons and behavioral analysis
Animals were anesthetized with 4% isoflurane
until unconscious, followed by 2% isoflurane
during surgery. Six microinjections (three
injections/hemisphere; 200 nl per injection)
of adeno-associated viruses (AAVs) expressing
hM3D(Gq)-mCherry (eDREADD), hM4D(Gi)-
mCherry (iDREADD) or mCherry alone (Mock),
under the CaMKIIa promoter (AAV8-CaMKIIa-
hM3D(Gq)-mCherry, titer 1 × 1012 vg/ml; AAV8-
CaMKIIa-hM4D(Gi)-mCherry, titer 1 × 1012 vg/ml;
AAV8-CaMKIIa-mCherry, titer 1 × 1012 vg/ml,
Duke University School of Medicine Viral Vec-
tor Core, Durham, NC, USA) were targeted to
the PVN. Six additional microinjections (three
injections/hemisphere, 200 nl per injection)
were targeted to the mPOA of the hypothala-
mus. Four weeks after viral transduction with
AAV8-CaMKIIa-hM3D(Gq)-mCherry (eDREADD
group), AAV8-CaMKIIa-hM4D(Gi)-mCherry
(iDREADD group) or AAV8-CaMKIIa-mCherry
(mock group), animals were injected with
CNO (CNO dihydrochloride, R&D Systems, 6329;
5 mg/kg in saline, intraperitoneal injection)
and returned to their home cage. Animals were
placed in a transparent cylindrical Plexiglas
box (20 cm diameter, 20 cm height) and allowed
to freely explore the box for 10 min before record-
ing using a digital camera. The number of penile
erections was manually counted offline at low
speed by an observer blinded to the experimen-
tal group and plotted as number of erections/
hour. Erections were scored when the animal
engaged in intensive grooming of genital area
for more than 5 s and hip constriction was ob-
served (58). The behavioral test was repeated
once a week for 3 consecutive weeks.
Statistical analysis
No statistical methods were used to predeter-
mine sample size. Group size for each experi-
ment was determined based on preliminary
experiments. Data are presented as mean ±
SEM. Individual data points are plotted in
the graphs for every experiment. Sample size
(n) represents biological replicates. Exclusion
of biological replicates was performed only
when a technical issue was detected. Datasets
were tested for Gaussian distribution followed
by the appropriate statistical test. P values were
calculated using two-tailed unpaired Student’s
t test, two-tailed Mann-Whitney test, one-way
ANOVA followed by Holm-Šidák multiple-
comparisons test, or two-way ANOVA followed
by Bonferroni’s multiple-comparisons test. Cor-
relation was calculated using the Spearman
correlation coefficient. Differences were con-
sidered statistically significant at P < 0.05. All
data were analyzed using GraphPad Prism
version 9.5.1 software for MacOS.
For a detailed description of the materials
and methods, please see the supplementary
materials.
REFERENCES AND NOTES
1. F. A. Yafi et al., Erectile dysfunction. Nat. Rev. Dis. Primers 2,
16003 (2016). doi: 10.1038/nrdp.2016.3; pmid: 27188339
9 of 11
RES EARCH | R E S E A R C H A R T I C L E
2. R. Shamloul, H. Ghanem, Erectile dysfunction. Lancet 381,
153–165 (2013). doi: 10.1016/S0140-6736(12)60520-0;
pmid: 23040455
3. G. R. Cunha et al., Reproductive tract biology: Of mice
and men. Differentiation 110, 49–63 (2019). doi: 10.1016/
j.diff.2019.07.004; pmid: 31622789
4. I. Sáenz de Tejada et al., Physiology of erectile function. J. Sex.
Med. 1, 254–265 (2004). doi: 10.1111/j.1743-6109.04038.x;
pmid: 16422955
5. L. Zhao et al., Single-cell transcriptome atlas of the human
corpus cavernosum. Nat. Commun. 13, 4302 (2022).
doi: 10.1038/s41467-022-31950-9; pmid: 35879305
6. D. Fang et al., Single-cell RNA sequencing of human corpus
cavernosum reveals cellular heterogeneity landscapes in
erectile dysfunction. Front. Endocrinol. (Lausanne) 13, 874915
(2022). doi: 10.3389/fendo.2022.874915; pmid: 35518933
7. D. O. Dias et al., Pericyte-derived fibrotic scarring is
conserved across diverse central nervous system lesions.
Nat. Commun. 12, 5501 (2021). doi: 10.1038/s41467-021-
25585-5; pmid: 34535655
8. D. O. Dias et al., Reducing pericyte-derived scarring promotes
recovery after spinal cord injury. Cell 173, 153–165.e22 (2018).
doi: 10.1016/j.cell.2018.02.004; pmid: 29502968
9. C. Göritz et al., A pericyte origin of spinal cord scar tissue.
Science 333, 238–242 (2011). doi: 10.1126/science.1203165;
pmid: 21737741
10. M. Slezak et al., Transgenic mice for conditional gene
manipulation in astroglial cells. Glia 55, 1565–1576 (2007).
doi: 10.1002/glia.20570; pmid: 17823970
11. R. A. Hill et al., Regional blood flow in the normal and ischemic
brain is controlled by arteriolar smooth muscle cell
contractility and not by capillary pericytes. Neuron 87, 95–110
(2015). doi: 10.1016/j.neuron.2015.06.001; pmid: 26119027
12. D. A. Hartmann et al., Brain capillary pericytes exert a
substantial but slow influence on blood flow. Nat. Neurosci. 24,
633–645 (2021). doi: 10.1038/s41593-020-00793-2;
pmid: 33603231
13. Y. Wu et al., Optogenetic approach for functional assays of the
cardiovascular system by light activation of the vascular
smooth muscle. Vascul. Pharmacol. 71, 192–200 (2015).
doi: 10.1016/j.vph.2015.03.006; pmid: 25869510
14. L. Madisen et al., A toolbox of Cre-dependent optogenetic
transgenic mice for light-induced activation and silencing.
Nat. Neurosci. 15, 793–802 (2012). doi: 10.1038/nn.3078;
pmid: 22446880
15. D. A. Boas, A. K. Dunn, Laser speckle contrast imaging in
biomedical optics. J. Biomed. Opt. 15, 011109 (2010).
doi: 10.1117/1.3285504; pmid: 20210435
16. K. E. Andersson, P. Hedlund, P. Alm, Sympathetic pathways
and adrenergic innervation of the penis. Int. J. Impot. Res. 12,
S5–S12 (2000). doi: 10.1038/sj.ijir.3900513
17. M. H. Cheng, I. Bahar, Monoamine transporters: Structure,
intrinsic dynamics and allosteric regulation. Nat. Struct. Mol.
Biol. 26, 545–556 (2019). doi: 10.1038/s41594-019-0253-7;
pmid: 31270469
18. D. N. Jones, M. A. Raghanti, The role of monoamine oxidase
enzymes in the pathophysiology of neurological disorders.
J. Chem. Neuroanat. 114, 101957 (2021). doi: 10.1016/
j.jchemneu.2021.101957; pmid: 33836221
19. V. Savchenko, U. Sung, R. D. Blakely, Cell surface trafficking of
the antidepressant-sensitive norepinephrine transporter
revealed with an ectodomain antibody. Mol. Cell. Neurosci. 24,
1131–1150 (2003). doi: 10.1016/S1044-7431(03)00235-5;
pmid: 14697674
20. U. Sung, F. Binda, V. Savchenko, W. A. Owens, L. C. Daws, Ca2+
dependent surface trafficking of norepinephrine transporters
depends on threonine 30 and Ca2+ calmodulin kinases.
J. Chem. Neuroanat. 83-84, 19–35 (2017). doi: 10.1016/
j.jchemneu.2016.12.003; pmid: 28017803
21. M. Dunn et al., Designing a norepinephrine optical tracer for
imaging individual noradrenergic synapses and their
activity in vivo. Nat. Commun. 9, 2838 (2018). doi: 10.1038/
s41467-018-05075-x; pmid: 30026491
22. A. Schlessinger et al., Structure-based discovery of
prescription drugs that interact with the norepinephrine
transporter,
NET. Proc. Natl. Acad. Sci. U.S.A. 108, 15810–15815 (2011).
doi: 10.1073/pnas.1106030108; pmid: 21885739
23. M. H. Kwon et al., Penile erection induces angiogenic, survival,
and antifibrotic signals: Molecular events associated with
penile erection induced by cavernous nerve stimulation in
mice. Int. J. Urol. 23, 614–622 (2016). doi: 10.1111/iju.13105;
pmid: 27109455
Guimaraes et al., Science 383, eade8064 (2024) 9 February 2024
24. S. K. Ramasamy, A. P. Kusumbe, L. Wang, R. H. Adams,
Endothelial Notch activity promotes angiogenesis and
osteogenesis in bone. Nature 507, 376–380 (2014).
doi: 10.1038/nature13146; pmid: 24647000
25. S. K. Ramasamy et al., Blood flow controls bone vascular
function and osteogenesis. Nat. Commun. 7, 13601 (2016).
doi: 10.1038/ncomms13601; pmid: 27922003
26. S. Nowotschin, P. Xenopoulos, N. Schrode, A. K. Hadjantonakis,
A bright single-cell resolution live imaging reporter of
Notch signaling in the mouse. BMC Dev. Biol. 13, 15 (2013).
doi: 10.1186/1471-213X-13-15; pmid: 23617465
27. S. J. Bray, Notch signalling in context. Nat. Rev. Mol.
Cell Biol. 17, 722–735 (2016). doi: 10.1038/nrm.2016.94;
pmid: 27507209
28. K. Tanigaki et al., Notch-RBP-J signaling is involved in cell fate
determination of marginal zone B cells. Nat. Immunol. 3,
443–450 (2002). doi: 10.1038/ni793; pmid: 11967543
29. L. C. Murtaugh, B. Z. Stanger, K. M. Kwan, D. A. Melton,
Notch signaling controls multiple steps of pancreatic
differentiation. Proc. Natl. Acad. Sci. U.S.A. 100, 14920–14925
(2003). doi: 10.1073/pnas.2436557100; pmid: 14657333
30. J. R. Sladek Jr., Age-dependent differences in catecholamine
distribution within cat reticular formation. Exp. Neurol. 38,
520–524 (1973). doi: 10.1016/0014-4886(73)90173-8;
pmid: 4696116
31. A. Morelli et al., Atorvastatin ameliorates sildenafil-induced
penile erections in experimental diabetes by inhibiting
diabetes-induced RhoA/Rho-kinase signaling hyperactivation.
J. Sex. Med. 6, 91–106 (2009). doi: 10.1111/j.1743-
6109.2008.01057.x; pmid: 19170840
32. S. H. Choo et al., Effects of eupatilin on the contractility of
corpus cavernosal smooth muscle through nitric oxide-
independent pathways. Andrology 5, 1016–1022 (2017).
doi: 10.1111/andr.12397; pmid: 28719725
33. Z. Liu, R. A. Khalil, Evolving mechanisms of vascular smooth
muscle contraction highlight key targets in vascular disease.
Biochem. Pharmacol. 153, 91–122 (2018). doi: 10.1016/
j.bcp.2018.02.012; pmid: 29452094
34. M. Hudnall, A. B. Reed-Maldonado, T. F. Lue, Advances
in the understanding of priapism. Transl. Androl. Urol. 6,
199–206 (2017). doi: 10.21037/tau.2017.01.18;
pmid: 28540227
35. F. Giuliano et al., Stimulation of the medial preoptic area of the
hypothalamus in the rat elicits increases in intracavernous
pressure. Neurosci. Lett. 209, 1–4 (1996). doi: 10.1016/
0304-3940(96)12594-5; pmid: 8734895
36. F. Giuliano, O. Rampin, Central neural regulation of penile
erection. Neurosci. Biobehav. Rev. 24, 517–533 (2000).
doi: 10.1016/S0149-7634(00)00020-8; pmid: 10880818
37. K. K. Chen, L. S. Chang, Effect of excitatory amino acid
receptor agonists on penile erection after administration into
paraventricular nucleus of hypothalamus in the rat. Urology 62,
575–580 (2003). doi: 10.1016/S0090-4295(03)00411-4;
pmid: 12946780
38. O. Rampin, Mode of action of a new oral treatment for
erectile dysfunction: Apomorphine SL. BJU Int. 88 (Suppl 3),
22–24 (2001). doi: 10.1046/j.1464-4096.2001.00125.x;
pmid: 11578275
39. A. Argiolas, M. R. Melis, The role of oxytocin and the
paraventricular nucleus in the sexual behaviour of male
mammals. Physiol. Behav. 83, 309–317 (2004). doi: 10.1016/
j.physbeh.2004.08.019; pmid: 15488547
40. S. S. Daftary, C. Boudaba, K. Szabó, J. G. Tasker,
Noradrenergic excitation of magnocellular neurons in the
rat hypothalamic paraventricular nucleus via intranuclear
glutamatergic circuits. J. Neurosci. 18, 10619–10628
(1998). doi: 10.1523/JNEUROSCI.18-24-10619.1998;
pmid: 9852597
41. H. A. Feldman, I. Goldstein, D. G. Hatzichristou, R. J. Krane,
J. B. McKinlay, Impotence and its medical and psychosocial
correlates: Results of the Massachusetts Male Aging Study.
J. Urol. 151, 54–61 (1994). doi: 10.1016/S0022-5347(17)34871-1;
pmid: 8254833
42. M. R. Lowerison et al., Aging-related cerebral microvascular
changes visualized using ultrasound localization microscopy in
the living mouse. Sci. Rep. 12, 619 (2022). doi: 10.1038/
s41598-021-04712-8; pmid: 35022482
43. H. M. Kang, I. Sohn, J. Jung, J. W. Jeong, C. Park, Age-related
changes in pial arterial structure and blood flow in mice.
Neurobiol. Aging 37, 161–170 (2016). doi: 10.1016/
j.neurobiolaging.2015.09.008; pmid: 26460142
44. T. Buch et al., A Cre-inducible diphtheria toxin receptor
mediates cell lineage ablation after toxin administration.
Nat. Methods 2, 419–426 (2005). doi: 10.1038/nmeth762;
pmid: 15908920
45. Y. Sheng, L. Zhu, The crosstalk between autonomic nervous
system and blood vessels. Int. J. Physiol. Pathophysiol.
Pharmacol. 10, 17–28 (2018). pmid: 29593847
46. U. Lendahl, L. Muhl, C. Betsholtz, Identification,
discrimination and heterogeneity of fibroblasts. Nat.
Commun. 13, 3409 (2022). doi: 10.1038/s41467-022-
30633-9; pmid: 35701396
47. A. Bannowsky, H. Schulze, C. van der Horst, S. Hautmann,
K. P. Jünemann, Recovery of erectile function after nerve-
sparing radical prostatectomy: Improvement with nightly low-
dose sildenafil. BJU Int. 101, 1279–1283 (2008). doi: 10.1111/
j.1464-410X.2008.07515.x; pmid: 18284406
48. O. Yaman, Z. Tokatli, E. Ozdiler, K. Anafarta, Effect of aging on
quality of nocturnal erections: Evaluation with NPTR testing.
Int. J. Impot. Res. 16, 150–153 (2004). doi: 10.1038/
sj.ijir.3901199; pmid: 14973521
49. N. F. Gonzalez-Cadavid, Mechanisms of penile fibrosis.
J. Sex. Med. 6, 353–362 (2009). doi: 10.1111/
j.1743-6109.2008.01195.x; pmid: 19267860
50. G. R. Cunningham et al., Association of sex hormones with
sexual function, vitality, and physical function of symptomatic
older men with low testosterone levels at baseline in the
testosterone trials. J. Clin. Endocrinol. Metab. 100, 1146–1155
(2015). doi: 10.1210/jc.2014-3818; pmid: 25548978
51. J. Yang, Z. Liu, Mechanistic pathogenesis of endothelial
dysfunction in diabetic nephropathy and retinopathy. Front.
Endocrinol. (Lausanne) 13, 816400 (2022). doi: 10.3389/
fendo.2022.816400; pmid: 35692405
52. N. E. Munk, J. S. Knudsen, S. Comerma-Steffensen,
U. Simonsen, Systematic review of oral combination therapy
for erectile dysfunction when phosphodiesterase type 5
inhibitor monotherapy fails. Sex. Med. Rev. 7, 430–441 (2019).
doi: 10.1016/j.sxmr.2018.11.007; pmid: 30711478
53. G. N. Yin et al., The pericyte as a cellular regulator of
penile erection and a novel therapeutic target for erectile
dysfunction. Sci. Rep. 5, 10891 (2015). doi: 10.1038/srep10891;
pmid: 26044953
54. N. M. Nguyen et al., Inhibition of proNGF and p75NTR pathway
restores erectile function through dual angiogenic and
neurotrophic effects in the diabetic mouse. J. Sex. Med. 16,
351–364 (2019). doi: 10.1016/j.jsxm.2019.01.013;
pmid: 30846111
55. K. M. Song et al., Vasohibin-1 rescues erectile function through
up-regulation of angiogenic factors in the diabetic mice.
Sci. Rep. 11, 1114 (2021). doi: 10.1038/s41598-020-80925-7;
pmid: 33441910
56. S. Kühbandner et al., Temporally controlled somatic
mutagenesis in smooth muscle. Genesis 28, 15–22 (2000).
doi: 10.1002/1526-968X(200009)28:1<15::AID-GENE20>
3.0.CO;2-C; pmid: 11020712
57. L. Tong et al., Imaging and optogenetic modulation
of vascular mural cells in the live brain. Nat. Protoc. 16,
472–496 (2021). doi: 10.1038/s41596-020-00425-w;
pmid: 33299155
58. B. D. Sachs, Erection evoked in male rats by airborne scent
from estrous females. Physiol. Behav. 62, 921–924 (1997).
doi: 10.1016/S0031-9384(97)00307-7; pmid: 9284518
AC KNOWLED GME NTS
We thank U. Lendahl, J. Frisén, G. Silberberg, and members of the
Göritz laboratory for valuable comments on the manuscript;
U. Lendahl for providing Tagln-CreERT2 mice; M. Hagemann-Jensen
and C. Ziegenhain for single-cell gene expression profiling; the
single-cell core facility at Flemingsberg campus (SICOF),
Karolinska Institutet, for single-cell sequencing services; and
T. Häneke for help with the graphical summary. Funding: C.G. is a
Hållsten Academy and a Knut and Alice Wallenberg Academy
Fellow. E.L.G. and A.J. were supported by a Wenner Gren
postdoctoral stipend. D.O.D. was supported by a postdoctoral
fellowship from STRATNEURO at Karolinska Institutet. Research in
the C.G. laboratory was supported by the Knut and Alice
Wallenberg Foundation (grant 2018.0159), the Swedish Research
Council (grants 2018-02798 and 2022-01244), the Swedish Brain
Foundation (grants FO2020-0093 and FO2022-0206), the Ming
Wai Lau Centre for Reparative Medicine, the Wings for Life
Foundation (grant WFL-SE-10/20), the Swedish Cancer Foundation
(grant 21-1768Pj01H), the Government of Canada’s New Frontiers
in Research Fund (grant NFRFT-2020-00238), and the strategic
network for stem cells and regenerative medicine (STRATREGEN)
at Karolinska Institutet. Experiments conducted in the L.J. and M.P.
laboratories were supported by the Swedish Research Council
10 of 11
RES EARCH | R E S E A R C H A R T I C L E
(grant 2021-01210 to L.J. and grants 2018-02552 and 2021-06014
to M.P.), the Swedish Brain Foundation (grant FO190101 to L.J.),
the Swedish Heart and Lung Foundation (grant 20220354 to
L.J.), and the Knut and Alice Wallenberg Foundation (grant
2019.0441 to M.P.). Author contributions: E.L.G., D.O.D., W.F.H.,
A.J., D.H., S.S., M.G.C., and E.V. performed experiments and/or
analyses. E.L.G. and C.G. designed experiments. D.O.D. designed
chemogenetics experiments. C.G. supervised experiments. L.J.
supervised intravital multiphoton microscopy experiments. M.P.
supervised confocal intravital microscopy experiments and initially
advised on blood flow measurements. C.G. conceptualized and
Guimaraes et al., Science 383, eade8064 (2024) 9 February 2024
coordinated the study. E.L.G., D.O.D., and C.G. wrote the SUPPLEMENTARY MATERIALS
manuscript. Competing interests: The authors declare no science.org/doi/10.1126/science.ade8064
competing interests. Data and materials availability: The scRNA- Materials and Methods
seq data reported in this manuscript are available at the National Figs. S1 to S22
Center for Biotechnology Information Gene Expression Omnibus References (59–68)
under accession number GSE249526. All other data are available in MDAR Reproducibility Checklist
the manuscript or supplementary materials. License information: Movie S1
Copyright © 2024 the authors, some rights reserved; exclusive
licensee American Association for the Advancement of Science. No Submitted 9 September 2022; resubmitted 28 September 2023
claim to original US government works. https://www.science.org/ Accepted 19 December 2023
about/science-licenses-journal-article-reuse 10.1126/science.ade8064
11 of 11
RES EARCH

◥ that photoperiodic flowering and growth are

RESEARCH ARTICLE SUMMARY genetically separable and that the photoperiod
measurement system governing flowering is not
PLANT SCIENCE controlling photoperiodic growth. Further exper-
iments showed that MIPS1 expression and
Plants distinguish different photoperiods function are regulated by a circadian clock–
controlled metabolic daylength measurement
to independently control seasonal flowering system. By changing light intensities over the
course of a day while maintaining the integrated
and growth intensity, we demonstrated that photoperiodic
growth and MIPS1 function are controlled by
Qingqing Wang, Wei Liu, Chun Chung Leung, Daniel A. Tarté, Joshua M. Gendron* the photosynthetic period and that flowering
is controlled by a wholly different photoperiod.

CONCLUSION: Our results show that plants can

INTRODUCTION: Daylength, or photoperiod, is
a stable indicator of the season, and organisms
can measure photoperiods to predict seasonal
changes in climate. In plants, flowering and
growth are often regulated by photoperiod,
and the photoperiodic flowering pathway is
well understood. By contrast, much less is known
about how photoperiod regulates growth, includ-
ing the measurement system and genes that
are required.
RATIONALE: Studies of photoperiodic flower-
ing have benefitted from mutants with obvious
defects in seasonal flowering time, and genes
whose expression is controlled by photoperiod
and can be tracked under different daylengths.
We hypothesized that similar tools would allow
for increased understanding of the genes and
measurement systems that participate in photo-
periodic growth, potentially revealing a different
mechanism than is used for seasonal flowering.
RESULTS: Arabidopsis thaliana grows faster in
long days, so we mined transcriptomic data for
genes that are induced in long days and required
for proper vegetative growth. We identified MYO-
INOSITOL-1-PHOSPHATE SYNTHASE 1 (MIPS1),
which encodes a gene necessary for plants to
produce myo-inositol, a sugar required for a
variety of important cellular processes that
control growth. We then showed that MIPS1
expression is induced during long but not
short days and that a mips1 mutant plant has
growth defects in long but not short days.
Because the flowering photoperiod measure-
ment system has been characterized in plants,
we tested whether our growth mutants were
in the same pathway. We found that the mips1
mutant has no flowering defect, and photo-
periodic flowering mutants have no growth
defect, results that were confirmed by double
mutants with mips1 and photoperiodic flower-
ing mutantions. These experiments showed
measure two different photoperiods in natural
day cycles. They can detect an absolute photo-
period with low light–detecting photoreceptor
systems to control flowering time. In parallel,
they can measure the photosynthetic period as
a metabolic daylength to control growth. This
allows plants to independently coordinate
seasonal developmental processes. This work
opens the possibility that several photoperiod
measurement systems can operate in parallel
in organisms to precisely regulate a variety of
seasonal processes.
▪
Department of Molecular, Cellular, and Developmental
Biology, Yale University, New Haven, CT 06511, USA.
*Corresponding author. Email: joshua.gendron@yale.edu
Cite this article as Q. Wang et al., Science 383, eadg9196
(2024). DOI: 10.1126/science.adg9196
READ THE FULL ARTICLE AT
https://doi.org/10.1126/science.adg9196

Flowering Leaf growth

Long day Short day Long day Short day
Light intensity

Light intensity

Compensation
Photoreceptor
point
detection
threshold

0 Absolute 24 0 24 0 Photosynthetic 24 0 24
photoperiod hours hours period hours hours

Responsive Responsive Responsive Responsive

Low MIPS1
High FT induces Low FT
flowering

High MIPS1
supports leaf growth

Plants detect two different daylengths to control seasonal flowering and
growth. In natural day cycles, photoperiodic flowering and photoperiodic growth
operate through different measuring systems. Photoperiodic flowering detects
an “absolute” photoperiod that is sensed by photoreceptors activated at low
light intensities. For flowering during long-day seasons, the photoperiod extends
into the light-sensitive portion of the day, which activates expression of the
florigen gene (FT) to promote flowering. Photoperiodic growth detects the
photosynthetic period, which is defined as the duration of time that light is
above the photosynthetic compensation point. For growth during long-day
seasons, the photosynthetic period extends into the light-sensitive portion of
the day, activating MYO-INOSITOL-1-PHOSPHATE SYNTHASE 1 (MIPS1) expres-
sion to support rapid growth. Hence, parallel photoperiod measurement
systems allow different photoperiodic processes to be coordinated indepen-
dently across the year.

Wang et al., Science 383, 605 (2024) 9 February 2024 1 of 1

RES EARCH

◥ those seen in the microarray, RNA-seq, and

RESEARCH ARTICLE qRT-PCR experiments (Fig. 1, A and B and fig. S1A),
with the 16L:8D-specific peak around dusk and
PLANT SCIENCE a common peak at ZT4 in both 8L:16D and
16L:8D.
Plants distinguish different photoperiods Previously, mips1 mutant plants were re-
ported to have growth defects that are de-
to independently control seasonal flowering pendent on the daily light integral (the
amount of photosynthetically active light
and growth passing through a defined area in 24 hours)
and possibly photoperiod, although these de-
Qingqing Wang, Wei Liu, Chun Chung Leung, Daniel A. Tarté, Joshua M. Gendron* fects varied on the basis of experimental con-
ditions (11, 17). We wanted to distinguish the
Plants measure daylength (photoperiod) to regulate seasonal growth and flowering. Photoperiodic role, if any, that MIPS1 has on photoperiodic
flowering has been well studied, but less is known about photoperiodic growth. By using a mutant growth. Thus, we designed a multifactorial
with defects in photoperiodic growth, we identified a seasonal growth regulation pathway that functions growth experiment that allowed us to distin-
in long days in parallel to the canonical long-day photoperiod flowering mechanism. This is achieved guish between intensity- and photoperiod-
by using distinct mechanisms to detect different photoperiods: The flowering pathway measures specific effects on growth. We germinated and
photoperiod as the duration of light intensity, whereas the growth pathway measures photoperiod as the grew wild-type and mips1-2 (SALK_023626)
duration of photosynthetic activity (photosynthetic period). Plants can then independently control mutant plants in two light intensities (100 and
expression of genes required for flowering or growth. This demonstrates that seasonal flowering and 200 mmol/m2/s) and two photoperiods (16L:8D
growth are dissociable, allowing them to be coordinated independently across seasons. and 8L:16D) and measured preflowering vege-
tative fresh rosette weight as a proxy for

P
hotoperiod (daylength) is a stable indi-
cator of the season, and many organisms
have evolved photoperiod measuring sys-
tems to predict abiotic and biotic changes
associated with a given season (1). Plants
have served as a preeminent study system
for photoperiodism because of their propensity
to flower in specific seasons, and a pathway
for seasonal flowering is known. Red and
blue light are sensed by photoreceptors that
control the stability of the CONSTANS (CO)
transcription factor, which in turn activates
expression of the florigen gene, FLOWERING
LOCUS T (FT), triggering the transition to
flowering (2–5). Despite intense focus on
flowering, research from more than 100 years
ago shows that both flowering and growth are
under the control of photoperiod in plants but
are separable and can be timed differentially
throughout the year. For instance, plants often
grow rapidly in long days but can flower faster
in short days (6). However, little is known about
the molecular determinants that support photo-
periodic growth, including the photoperiod
measuring systems, cellular signaling path-
ways, and genes required for photoperiod-
dependent growth. This deficiency is partially
due to a lack of genetic tools and markers for
growth that are akin to the tools available for
the CO-FT pathway.
We analyzed published RNA sequencing
(RNA-seq) (7) and microarray data (8) and
identified genes that are expressed higher
in long days when growth is promoted in
Arabidopsis thaliana. We cross-referenced these
against the literature to identify genes that cause
Department of Molecular, Cellular, and Developmental
Biology, Yale University, New Haven, CT 06511, USA.
*
Corresponding author. Email: joshua.gendron@yale.edu
growth defects when mutated (8–10). One of the
identified genes, MYO-INOSITOL-1-PHOSPHATE
SYNTHASE 1 (MIPS1), is required for produc-
tion of myo-inositol, which serves as the chem-
ical precursor for a wide variety of signaling
molecules, vitamins, and membrane components
required for cell division, cell expansion, and
maintenance of cell viability (11–14). Addition-
ally, MIPS1 participates in pathogen defense,
another cellular process that is known to alter
growth (15, 16). We first plotted the expression
of MIPS1 from microarray and RNA-seq exper-
iments and verified the expression pattern using
quantitative reverse transcription polymerase
chain reaction (qRT-PCR) (Fig. 1, A and B, and
fig. S1A). MIPS1 expression is qualitatively and
quantitatively different in 16 hours light to
8 hours dark (16L:8D) versus 8L:16D. In 8L:16D,
MIPS1 expression is monophasic, with a peak in
expression at ZT4 (4 hours after dawn). In
16L:8D, MIPS1 expression is biphasic, contain-
ing the ZT4 peak seen in 8L:16D and a second
photoperiod-specific peak phased to dusk at
ZT16. This results in a relative daily expression
integral of 16L:8D/8L:16D [rDEI16L:8D/8L:16D =
sum of 24 hours of expression in condition
one/sum of 24h of expression in condition two
(9)] of 1.44 in the microarray experiment, 2.77 in
the RNA-seq experiment, and 1.70 in qRT-PCR.
These results indicate that plants grown in 16L:8D
have higher MIPS1 expression owing to a
photoperiod-specific peak in expression at ZT16.
Next, we generated transgenic plants ex-
pressing the Luciferase gene under the control of
the MIPS1 promoter (MIPS1promoter::Luciferase)
(Fig. 1C). We first measured the luminescence
from the MIPS1promoter::Luciferase plants under
8L:16D and 16L:8D conditions (Fig. 1C and fig.
S1B). The expression patterns generated by
the luciferase reporter were consistent with
growth. We collected samples over the vege-
tative growth phase of the plant (Fig. 1D and
fig. S2, A to C).
The mips1-2 mutant was unable to generate
fresh weight at the same rate as that of the
wild type in the 16L:8D growth conditions,
but there was no observable fresh weight
defect in the 8L:16D growth conditions. To
determine the role of MIPS1 in photoperiod-
dependent growth, independent of the daily light
integral, we compared the plants grown in 16L:8D
in 100 mmol/m2/s white light (16L100:8D) and
8L200:16D on day 28 (Fig. 1E). In these condi-
tions, the daily light integral was the same, but
the wild-type plants grown in 16L100:8D gen-
erated 2.6 times the mass of the 8L200:16D-grown
plants, confirming the role of photoperiod in
Arabidopsis growth.
By dividing the fresh weight by the daily light
integral and then calculating a ratio between
two conditions, we can calculate relative fresh
weight gain per available photon. This is a
simple measure of how well each plant can
use available light for growth. The ratio was
2.6 when comparing plants in 16L100:8D and
8L200:16D, showing that in 16L100:8D, the
plants used each available photon 2.6 times more
effectively for growth than in 8L200:16D (Fig. 1F
and fig. S2D). Conversely, the mips1-2 mutant
plants grown in 16L100:8D generated 0.83 times
the mass of the 8L200:16D-grown plants, demon-
strating that the mips1-2 mutant lost the ability
to generate rosette fresh weight in a photo-
periodic manner.
To determine the role of MIPS1 in light
intensity–dependent growth, we compared
rosette fresh weight when light intensity was
doubled in 16L:8D or 8L:16D (Fig. 1G and
fig. S2E). Wild-type plants in 16L200:8D were
3.5 or four times larger than in 16L100:8D at
days 16 and 20, respectively, meaning that

Wang et al., Science 383, eadg9196 (2024) 9 February 2024 1 of 9

RES EARCH | R E S E A R C H A R T I C L E

Fig. 1. MIPS1 is involved in A B C

MIPS1promoter::LUC
photoperiod-dependent growth. 1.50 16L:8D

Normalized intensity
8L:16D
(A) RNA-seq and (B) qRT-PCR of
1.25
MIPS1 expression from 12-day-old
plants grown in 16L:8D (red) and

TMM
1.00
8L:16D (blue). Shading indicates
the dark period. TMM expression, 0.75

trimmed mean of M values. IPP2 0 4 8 12

Time (hours)
16 20 24

was used as an internal control D

for (B). Error bars represent SD.
*P ≤ 0.05; **P ≤ 0.01; ns, not
significant (Welch’s t test). n = 3
samples; samples include
multiple plants. (C) Normalized
MIPS1promoter::Luciferase trace
from plants grown in 16L:8D (red)
and 8L:16D (blue). (D) (Top) Shoot
fresh weight with (bottom) repre-
sentative images of wild-type (Col),
mips1-2, co-9, and ft-10 mutants
grown in 16L:8D or 8L:16D with
100- or 200-mmol/m2/s light
intensity, as indicated. Error bars
represent SD (n = 5 plants). Scale
bar, 1 cm. Ages of the plants in
E F G
16L200:8D, 16L100:8D, 8L200:16D,
and 8L100:16D were 20, 24, 56,
and 70 days, respectively, as
indicated. (E) Shoot fresh weight
of wild type and mips1-2 mutant
from 28-day-old plants grown in
16L:8D or 8L:16D conditions
with a light intensity of 100 or
200 mmol/m2/s, as indicated. H I
7L:17D 8L:16D 9L:15D 10L:14D
n = 5 plants. (F) Ratio of growth 1.50
2L:12D 14L:10D 16L:8D
12L:12D
Normalized intensity

per available photon of 28-day-old

wild-type and mips1-2 mutant 1.25

plants grown in 16L100:8D com-

1.00
pared with that of those grown in
8L200:16D. Growth per available 0.75
photon = shoot fresh weight/days 0 4 8 12 16 20 24
Time (hours)
of growth/(light intensity × number
of hours of light). (G) Ratio of
J
growth per available photon of
wild-type and mips1-2 plants was
compared for those grown in
16L200:8D and 16L100:8D, as well
as 8L200:16D and 8L100:16D, at
indicated age. Growth per
available photon = shoot fresh
weight changed/days of growth/
(light intensity × number of
hours of light). (H) Normalized MIPS1promoter::Luciferase trace data in different photoperiods. The data from (C) [16L:8D (magenta) and 8L:16D (dark blue)] were
replotted here for comparison. (I) Shoot fresh weight of wild-type and mips1-2 mutant plants at day 28 grown in different photoperiods with a light intensity of 200 mmol/m2/s.
The green box notes that plants grown in 16L:8D started to flower before the measurement. Error bars represent SD. *P ≤ 0.05; **P ≤ 0.01; ****P ≤ 0.0001 (Welch’s t test).
n = 5 plants. (J) Leaf number at 1-cm bolt length. The plants were grown in 16L:8D or 8L:16D with a light intensity of 100 or 200 mmol/m2/s, as indicated. Error
bars represent SD (n = 19 to 30 plants). In (E) and (J), different letters indicate significant differences as determined by one-way analysis of variance (ANOVA), followed
by Dunnett’s T3 multiple comparison test. P ≤ 0.05. Raw data and calculations of physiology experiments are presented in table S2.

they can generate up to twice as much fresh development, they can generate up to three the mips1-2 mutant lost the ability to generate
weight per available photon. Plants in 8L200:16D times as much weight per additional photon. rosette fresh weight in a light intensity–
were 6.1, 4.8, or 1.5 times larger than plants This effective use of the increased photons for dependent manner, but this ability was un-
grown in 8L100:16D, which shows that early in growth decreased as the plants aged. In 16L:8D, affected in 8L:16D. This indicates that wild-type

Wang et al., Science 383, eadg9196 (2024) 9 February 2024 2 of 9

RES EARCH | R E S E A R C H A R T I C L E
plants alter light intensity–dependent growth
responses based on the photoperiod, and that
MIPS1 plays a role in these photoperiod-
specific growth processes.
The mips1-2 mutant displays cell division,
cell expansion, and cell viability defects (11).
The cell viability defect can be seen as leaf
lesions and is caused by overactivation of sali-
cylic acid (SA) production (11). The lesions were
present when we grew the mips1-2 mutant in
16L:8D (Fig. 1D and fig. S2C), and we wanted
to test whether the observed lesions or SA
overproduction caused the observed long-day
specific fresh weight defect of the mips1-2
mutant. We therefore crossed mips1-2 with
a lesion-defect mutant that does not accumu-
late SA owing to mutation of the SALICYLIC
ACID INDUCTION DEFICIENT 2 (SID2) gene
(18, 19). We then observed the formation of
lesions and measured fresh weight in 8L:16D
and 16L:8D (fig. S2F). In 8L:16D, we observed
no lesions and no defects in fresh weight in any
mutant combination. In 16L:8D, as expected,
the mips1-2 sid2 double mutant had no lesions,
confirming the role of SA in mips1-2 lesion
formation. Furthermore, the mips1-2 and mips1-2
sid2 mutants had similar fresh weights (43 and
47 mg, respectively) and were statistically dif-
ferent from the wild type. This shows that the
lesion defect of mips1-2 can be restored by
blocking SA production but that this does not
restore mips1-2 to wild-type fresh weight levels.
The mips1-2 sid2 double mutant was 19-mg
(29%) smaller on average than the sid2 single
mutant (66 mg), but this difference did not
reach the threshold for statistical significance
in these experiments (P = 0.102). The lesions seen
in the mips1-2 mutant were caused by higher
SA levels, and growth was not fully rescued
when SA levels are reduced with the addition
of the sid2 mutation. However, the lack of sig-
nificant difference between the double mutant
and sid2 single mutant means that SA may be
one participant in the fresh weight defect of
the mips1-2 mutant, perhaps reflecting the di-
verse roles of myo-inositol in supporting growth
in long days.
We next wanted to find the minimum daylength
required to induce MIPS1 expression and growth.
We grew the MIPS1promoter::Luciferase reporter
line in photoperiods ranging from 7L:17D to
16L:8D and monitored luminescence (Fig. 1H
and fig. S3). The long photoperiod–specific peak
in MIPS1 expression appeared at the 12L:12D
photoperiod. We next performed growth ex-
periments with wild-type and mips1-2 mutant
plants in the same photoperiods (Fig. 1I and
fig. S4). Wild-type plants grew very little in
photoperiods less than 12L:12D but grew rapid-
ly in photoperiods of 12L:12D or longer. Addi-
tionally, the mips1-2 mutant showed clear growth
defects in photoperiods 12L:12D or longer com-
pared with those in photoperiods less than
12L:12D. This suggests that the critical photo-
Wang et al., Science 383, eadg9196 (2024) 9 February 2024
period for growth is near the equinox and
matches the critical photoperiod for MIPS1
expression and function. Furthermore, the
data suggest that the photoperiod-specific
peak in MIPS1 expression is relevant for its role
in growth, whereas the peak at ZT4, which is
present in all photoperiods tested, plays no
observable role in photoperiodic growth.
MIPS1 is required for leaf growth
We measured flowering time using two metrics:
number of days to reach a 1-cm floral bolt and
number of leaves at the time when the floral
bolt reaches 1 cm (Fig. 1J and fig. S5A). De-
spite vegetative growth defects in 16L:8D,
the mips1-2 mutant had no measurable defect
in flowering time in 16L:8D or 8L:16D. Iden-
tical leaf-count numbers between wild type
and mips1-2 mutants, in combination with
the rosette fresh weight data, indicates that
MIPS1 has no role in flowering or leaf ini-
tiation rate during the normal vegetative
growth period but is required for generating
leaf mass in 16L:8D.
Postflowering development was also assessed
by elongation rate of the inflorescence stem
(days from 1- to 10-cm inflorescence stem
length), anthesis (first observable flower open-
ing), and seed yield (total mass of dry seeds) (fig.
S5, B and C). In wild-type plants, the elongation
rate of the stem was lower in 16L:8D than in
8L:16D, but it was not affected by intensity
in either photoperiod, whereas anthesis was
faster in 16L:8D than in 8L:16D. The mips1-2
mutant delayed both developmental processes
in 16L:8D but not in 8L:16D, except for anthesis
in 16L100:8D, where mips1-2 was not different
from the wild type.
We found that wild-type seed yield is greater
in 16L:8D than in 8L:16D and that doubling
light intensity increased yield in 16L:8D but
not in 8L:16D (fig. S5D). In 16L:8D, the mips1-2
mutant had lower yield compared with that of
the wild type in both light intensities. Con-
versely, we detected no yield difference be-
tween the wild type and mips1-2 in 8L:16D at
both light intensities. Thus, we conclude that
MIPS1 is required for long day–specific pre-
flowering vegetative growth, postflowering
bolt elongation, and reproductive development
but has no effect on flowering time or leaf
organogenesis.
MIPS1 expression is independent of CO
The photoperiod-controlled expression peak of
MIPS1 is phased to the same time of day (ZT16)
as the expression peak for FT in 16L:8D (2),
making CO a candidate regulator of MIPS1
and photoperiodic growth. We assayed photo-
periodic growth in co-9 (20) and ft-10 (21)
mutants, which have compromised photo-
periodic flowering, and saw no defects in
fresh weight (Fig. 1, D and J, and figs. S2A
and S5, A to C). The co-9 and ft-10 mutants
had reduced seed yield in 16L:8D, and ft-10
had reduced seed yield in 8L100:16D (fig. S5D).
We tested MIPS1 expression with qRT-PCR on
samples from wild-type and co-9 mutant plants
collected at ZT12 in 16L:8D and 12L:12D, a time
at which FT is normally induced. We found that
FT expression was reduced in co-9 mutants,
whereas MIPS1 expression was not altered
(Fig. 2A). We next crossed the co-9 mutant to
the MIPS1promoter::Luciferase reporter and grew
the plants in 16L:8D and 8L:16D for imaging
(Fig. 2B and fig. S6A). The expression pattern
was similar between the wild-type and co-9 mu-
tant plants, supporting the qRT-PCR results.
Next, we generated a co-9 mips1-2 double
mutant. The double mutant showed no addi-
tional growth defects beyond those of the
mips1-2 mutant alone in the 16L100:8D con-
dition (Fig. 2C). The co-9 mips1-2 double
mutant required more days to bolt in the
16L100:8D condition than the co-9 mutant,
although leaf numbers are the same at time of
bolting (Fig. 2C and fig. S6B). This suggests
that MIPS1 can play a role in the rate of plant
development in the absence of CO.
The metabolic daylength measurement system
regulates MIPS1
The metabolic daylength measurement system
was shown to induce expression of genes in
short-day photoperiods, but it remains to be
seen if it can also cause induction of gene ex-
pression in long-day photoperiods (9). This
system relies on light sensing by photo-
synthesis and the circadian clock–controlled
balance between starch and sucrose to con-
trol gene expression in a photoperiodic manner.
First, we determined whether photosynthesis
is required for induction of MIPS1 in 16L:8D.
Using qRT-PCR, we tested expression of MIPS1
at ZT12 from plants grown in 16L:8D and then
placed in darkness or kept in light but treated
with a chemical inhibitor of photosystem II
called 3-(3,4-dichlorophenyl)-1,1-dimethylurea
(DCMU) (Fig. 3A). In both treatments, MIPS1
expression decreased, suggesting that the tran-
sition from light to dark is communicated
through photosynthesis to control induction
of MIPS1 in 16L:8D.
To determine if the light-dark transition is
communicated as a metabolic change, we treated
wild-type plants with sucrose in 8L:16D and
measured MIPS1 expression at ZT12, when
the plants were in the dark and MIPS1 ex-
pression is normally low (Fig. 3B). Sucrose
treatment induced expression of MIPS1, sug-
gesting that metabolism plays an important
role in controlling MIPS1 photoperiodic ex-
pression. Previously, sucrose levels have been
measured in 16L:8D and 8L:16D in Arabidopsis
(22), which allowed us to perform correlation
analyses between sucrose levels and MIPS1
expression. We compared sucrose levels with
MIPS1 expression data from our qRT-PCR and
3 of 9
RES EARCH | R E S E A R C H A R T I C L E

A grew our MIPS1promoter::Luciferase reporter in

constant light conditions and found that MIPS1
had a circadian rhythm [relative amplitude
error (RAE) = 0.14] with increasing expres-
sion starting around ZT12 and peaking near
dawn, matching that of the critical photo-
period experiment (Fig. 3C). We crossed the
MIPS1promoter::Luciferase reporter to a plant
with a mutation in the evening complex gene
LUX ARRYTHMO (lux-4) that causes circa-
B dian clock arrhythmicity (23). In constant
Col MIPS1promoter::LUC Col MIPS1promoter::LUC light, the lux-4 mutation caused arrythmia
co-9 co-9
1.00 16L:8D 1.00 8L:16D (RAE 1.13) of MIPS1 expression, showing that
the circadian clock controls the light-sensitive
Normalized
intensity

0.75 0.75
period for MIPS1 expression (Fig. 3C). We next
0.50 0.50
tested whether lux-4 regulates the expression
0.25 0.25 of MIPS1 in 16L:8D or 8L:16D (fig. S8). In
0 0 16L:8D, the first photoperiodic peak remained
similar between wild-type and the lux-4 mutant,
0 4 8 12 16 20 24 0 4 8 12 16 20 24
Time (hours) Time (hours) but the second peak had a phase advance of
~2.7 hours in the lux-4 mutant. There was no
C change in the phase of the first peak in 8L:16D.
Disabling the circadian clock with an arrhyth-
mic mutant can ablate rhythms under constant
conditions, but when diurnal cycles are imposed
on a clock mutant, 24-hour rhythms are often
restored in the other clock components that
are not mutated. Together, these results show
that the circadian clock and lux-4 are required
for proper MIPS1 expression when days are long.
Sucrose can induce MIPS1 expression, and
photoperiod-controlled sucrose levels rely on
the regulated sequestration of sucrose as the
storage sugar, starch (10, 22, 24–30). Further-
more, starch production is necessary for proper
function of the metabolic daylength measure-
ment system. The starchless mutant, phospho-
glucomutase (pgm-1), is unable to produce
starch or regulate sucrose levels in a photo-
periodic manner, causing a malfunction of
daylength measurement (9, 31). We measured
expression of MIPS1 in the pgm-1 mutant in
16L:8D and 8L:16D with qRT-PCR (Fig. 3D).
In 16L:8D, MIPS1 expression had a greater
amplitude at the ZT4 peak of expression and
Fig. 2. MIPS1 expression and function are distinct from CO-FT. (A) qRT-PCR of FT and MIPS1 from a less-pronounced peak at ZT16 in pgm-1 as
12-day-old wild type (Col) and co-9 mutant grown in 16L:8D (orange) and 12L:12D (green) conditions. Samples compared with the wild type. In 8L:16D, MIPS1
were harvested at ZT12. n = 3 samples; samples include multiple plants. Error bars represent SD. ***P ≤ 0.001; expression had a larger amplitude of expression
****P ≤ 0.0001; ns, not significant (Welch’s t test). (B) Normalized traces of MIPS1promoter::Luciferase from during the light period in pgm-1, showing an
wild-type and co-9 mutant plants grown under 16L:8D and 8L:16D. (C) (Left) Shoot fresh weight with (middle) inability to maintain low levels of MIPS1 in short
representative images and (right) leaf number at 1-cm bolting of wild type, mips1-2, co-9, and mips1-2 co-9 days. This was confirmed by crossing the pgm-1
mutants grown in 16L:8D or 8L:16D condition with a light intensity of 100 mmol/m2/s. Ages of the mutant to the MIPS1promoter::Luciferase reporter,
plants in 16L100:8D and 8L100:16D were 24 and 70 days, respectively. Compact letter display indicates which showed similar expression defects as in
significant differences as determined by one-way ANOVA followed by Dunnett’s T3 multiple comparison the pgm-1 mutant (Fig. 3D). We tested the
test; P ≤ 0.05. Error bars represent SD (n = 5 plants for fresh weight; n = 19 to 30 plants for leaf count). Scale bar, genetic relationship between PGM and MIPS1
1 cm. The data for flowering time from wild type, mips1-2, and co-9 in the 8L100:16D condition were replotted by crossing the mips1-2 mutant with the
from Fig. 1J for comparison. Raw data are presented in table S2. pgm-1 mutant and assaying rosette fresh
weight in 16L:8D and 8L:16D (Fig. 3E and
RNA-seq experiments (Fig. 1, A and B). In both In photoperiodic systems, the circadian clock fig. S9A). The pgm-1 mips1-2 double mutant
cases, we found a positive correlation between sets transient light-sensitive periods to allow showed greater defects in rosette fresh weight
sucrose levels and expression of MIPS1 (fig. S7, for daylength measurement. Our critical photo- in 8L:16D than the pgm-1 single mutant. This
A to C), although the correlation with qRT-PCR period experiment suggests that the light- is consistent with the observation that MIPS1
data did not fall below the cutoff of P < 0.05, sensitive phase for MIPS1 expression occurs after expression is improperly induced in 8L:16D in
having a P value of 0.0502. ZT12 (Fig. 1, H and I, and figs. S3 and S4). We the pgm-1 mutant. In 16L:8D, the pgm-1 mutant

Wang et al., Science 383, eadg9196 (2024) 9 February 2024 4 of 9

RES EARCH | R E S E A R C H A R T I C L E
A ZT12 B ZT16
Fig. 3. A metabolic daylength measurement system controls photoperiodic
expression of MIPS1. (A) qRT-PCR of MIPS1 from 12-day-old wild-type (Col)
plants grown in 16L:8D treated with or without darkness or DCMU at ZT0.
Samples were harvested at ZT12. (B) qRT-PCR of MIPS1 from 12-day-old plants
grown in 8L:16D treated with or without 90-mM sorbitol or sucrose at ZT12.
Samples were harvested at ZT16. (C) MIPS1promoter::Luciferase traces from wild-
type (black) or lux-4 mutant (red) plants grown under constant light conditions.
(D) qRT-PCR of MIPS1 and FT from 12-day-old wild-type and pgm-1 mutant plants
grown under 16L:8D and 8L:16D conditions, and MIPS1promoter::Luciferase traces from
wild-type (black) and pgm (blue) mutant plants grown under 16L:8D or 8L:16D.
UBQ10 was used as an internal control. Error bars represent SD. *P ≤ 0.05;
Wang et al., Science 383, eadg9196 (2024) 9 February 2024
**P ≤ 0.01; ***P ≤ 0.001; ns, not significant (Welch’s t test). n = 3 samples.
(E) Representative images and shoot fresh weight of wild type and mips1-2, pgm-1,
and pgm-1 mips1-2 mutants grown in 16L:8D and 8L:16D with a light intensity of
200 mmol/m2/s. Ages of the plants in 16L200:8D and 8L200:16D were 20 and
56 days, respectively. Scale bars, 1 cm. (F) MIPS1promoter::Luciferase traces from wild-
type (black) and lux-4 (red) and pgm-1 (blue) mutant plants grown under 16L:8D and
then transferred to double dusk (8L:4D:8L:4D) on day 11. A.U., arbitrary units.
Compact letter display in (A), (B), and (E) indicates significant differences
as determined by one-way ANOVA followed by Dunnett’s T3 multiple comparison
test; P ≤ 0.05. Error bars represent SD (n = 5 plants). Raw data and calculations for
physiology experiments are presented in table S2.
5 of 9
RES EARCH | R E S E A R C H A R T I C L E
partially suppressed the lesion and growth de-
fects of the mips1-2 mutant, again suggesting a
genetic interaction between MIPS1 and PGM.
Our critical photoperiod and circadian clock
experiments (Figs. 1, H and I, and 3C) suggest
that MIPS1 expression responds to light dif-
ferently before and after ZT12. We tested this
directly using a “double dusk” experiment
(8L:4D:8L:4D) with the MIPS1promoter::Luciferase
reporter (Fig. 3F) (9). MIPS1 showed a repeating
daily pattern. At both “dawns” in each 24-hour
period, there were transient peaks in MIPS1
expression; however, across the first 8-hour
light period (starting at ZT0), expression de-
creased, whereas during the second 8-hour
light period (starting at ZT12), expression in-
creased. We used this experiment to test if the
circadian clock or starch is required for this dif-
ferential response to light across a 24-hour pe-
riod. We again used the MIPS1promoter::Luciferase
crossed to lux-4 or pgm-1 and performed double
dusk experiments (Fig. 3F). The wild-type pat-
tern of MIPS1 expression was ablated in both
mutants, demonstrating that the circadian clock
and starch are required for proper seasonal
expression of MIPS1 and confirming that MIPS1
expression is controlled by a metabolic day-
length measurement system.
Myo-inositol is not sufficient to promote
growth on its own (11). We confirmed this by
generating two transgenic lines that overexpress
MIPS1 (fig. S9B), growing the plants in 8L200:16D,
and measuring their fresh weight (fig. S9C). As
expected, the MIPS1-OX lines did not have greater
fresh weight on average than the wild-type plants.
In support of this, myo-inositol supplementation
in 8L200:16D also did not promote growth (fig.
S9D). However, in 16L200:8D, myo-inositol sup-
plementation rescued the mips1-2 mutant phe-
notypes, supporting the idea that the enzymatic
function of MIPS1 is required for growth in
long days (fig. S9D). These results suggest that
MIPS1 and myo-inositol production are required
for rapid long-day growth but are not sufficient
to promote rapid growth in short days. This
indicates that there are other photoperiod-
controlled processes that must be coactivated
with MIPS1 to promote growth. Alternatively,
there may be a short-day growth-restrictive
mechanism that must be suppressed for MIPS1
to promote growth.
Our results show that the CO-FT mechanism
has little or no effect on MIPS1 expression or
photoperiodic growth under the conditions
tested here. The pgm-1 mutant shows defects
in flowering time, but published results sug-
gest that FT is not misregulated in the pgm-1
mutant in 12L:12D (32). To further test
this, we measured FT expression in the pgm-1
mutant in 8L:16D and 16L:8D (Fig. 3D). FT
expression in the pgm-1 mutant was the same
as in the wild type in 8L:16D and 16L:8D, sug-
gesting that the metabolic daylength and CO
photoperiod measuring systems control distinct
Wang et al., Science 383, eadg9196 (2024) 9 February 2024
target genes to regulate growth and flowering,
respectively.
The photosynthetic period controls
plant growth
The CO-FT and metabolic daylength measure-
ment system/MIPS1 photoperiod regulons func-
tion in parallel with respect to the molecular
components that comprise each mechanism.
We used the mips1-2 growth defects to test if
the two systems measure the same photo-
period. The CO-FT system uses blue- and red-
light photoreceptors to measure the duration
of the day, even at very low light levels, a near-
“absolute” daylength (33). Our results indicate
that Arabidopsis may also be able to measure
the photosynthetic period in parallel to the
absolute photoperiod. To test this, we started
by reducing light intensity to 40 mmol/m2/s,
which is near the light compensation point
(light intensity at which the rate of photo-
synthesis matches the rate of cellular respira-
tion) for Arabidopsis (30). At the compensation
point, red and blue light photoreceptors remain
active, but photosynthesis is unable to produce
more sugar than is used by respiration, leaving
little extra usable sugar for processes such as
growth. In 16L40:8D, the mips1-2 mutant showed
no defects in growth, flowering, or seed yield
compared with the wild type, indicating that
MIPS1 is not required for growth during long
days when light is near the compensation point
(Fig. 4, A and B, and fig. S10, A to C). Conversely,
wild-type and mips1-2 mutant plants flowered at
approximately 35 days in 16L40:8D, which is
greatly accelerated compared with the flowering
time in 8L100:16D (~82 days) or 8L200:16D
(~65 days) (figs. S5A and S10B). Because
16L40:8D has a lower daily light integral than
these two conditions, this suggests that photo-
periodic flowering, which is under the control
of photoreceptors that are active at 40 mmol/m2/s,
functions normally.
We found that photoperiodic plant growth
is promoted and the mips1-2 phenotype is
apparent in 16L100:8D and 16L200:8D, but
growth is restricted and the mips1-2 mutant
phenotype is absent in 8L100:16D, 8L200:16D,
and 16L40:8D. These results suggest that in
the middle 8 hours of the day, ZT8 to ZT16,
plant growth is more sensitive to light above the
compensation point than in the first 8 hours of
the day, ZT0 to ZT8. We tested this directly by
growing the wild type and mips1-2 mutant in
two long-day lighting regimes: one with high
light (190 mmol/m2/s) from ZT0 to ZT8 and
low light (10 mmol/m2/s) from ZT8 to ZT16
(8L190:8L10:8D) and one with low light (10 mmol/
m2/s) from ZT0 to ZT8 and high light (190
mmol/m2/s) from ZT8 to ZT16 (8L10:8L190:8D)
(Fig. 4, C and D, and fig. S10D). In these ex-
periments, the daily light integral and daylength
are held constant, but the time of day when the
compensation point is crossed and photosyn-
thates are available for growth is altered. Rather
than a “night break” experiment, this may be
better termed a “fast break” experiment be-
cause we are interrupting a low-light and low-
photosynthate long day with light above the
compensation point to produce photosyn-
thates during a specific period. The mips1-2
plants grown in 8L190:8L10:8D showed no
altered phenotype compared with the wild-
type plants when examined at 36 days (time
of flowering); additionally, they showed no
lesions but had generated slightly less fresh
weight at the 28-day time point. The mips1-2
plants grown in 8L10:8L190:8D showed growth
and cell death defects at 28 days, similar to
plants grown in 16L200:8D (Figs. 1D and 4C),
and we were not able to collect fresh weight
data at 36 days because they had flowered at
28 days. In addition, the wild-type plants grew
nearly 70% larger in 8L10:8L190:8D, suggesting
that light above the compensation during ZT8
to ZT16 has a greater effect on growth than
light during ZT0 to ZT8. This indicates that
the light-sensitive time of day for growth
occurs between ZT8 and ZT16 when MIPS1
is induced in long-day photoperiods, although
if MIPS1 expression is elevated between ZT0
and ZT8 as in the pgm-1 mutant (Fig. 3, D
and E), growth can also be supported to some
extent.
We also tracked the expression of MIPS1 using
the MIPS1promoter::Luciferase reporter line under
similar conditions. We grew the reporter plants
in 8L140:8L10:8D and 8L10:8L140:8D (fig. S10E).
We found that MIPS1 expression was induced
when the light was 140 mmol/m2/s but was not
induced when the light was 10 mmol/m2/s.
To track the circadian clock, we included a
CCA1promoter::Luciferase reporter, which had a small
change in phase but retained early-day phasing
under both conditions (fig. S10E) (34). This sug-
gests that the cause of the MIPS1 phase shift
cannot be explained by a clock phase shift, but
rather that it responds to light above the com-
pensation point. When comparing 8L140:8L10:8D
with 8L10:8L140:8D, amplitude differences can
be hard to gauge because they were performed
in separate experiments. Thus, we performed
a single experiment in which we flipped between
these two light regimes (8L10:8L140:8D:8L140:8L10:8D)
(fig. S10F). In this condition, MIPS1 increased
to a greater extent when 140 mmol/m2/s light
was provided between ZT8 and ZT16. This pat-
tern was ablated in the pgm-1 mutant back-
ground, showing that proper starch production
is required for this differential response (fig.
S10F). Conversely, the pattern was retained in
the lux-4 mutant, similar to what was seen in
standard 8L:16D and 16L:8D conditions (figs.
S10F and S8). Again, the CCA1promoter::Luciferase
reporter maintained its phase in 8L10:8L140:
8D:8L140:8L10:8D condition (fig. S10F). This
corroborates the idea that plants are re-
sponding to photosynthetic light differentially
6 of 9
RES EARCH | R E S E A R C H A R T I C L E

A B F

D E

Fig. 4. Photosynthetic period controls MIPS1-dependent growth. (A) Shoot
fresh weight with representative image and (B) leaf number at 1-cm bolting of
wild-type (Col) and mips1-2 mutant plants grown in 16L:8D with a light intensity
of 40 mmol/m2/s. Age of the plants in 16L40:8D was 28 days. No significant
difference (ns) was detected by Welch’s t test. Scale bar, 1 cm. n = 5 plants in (A)
and 23 plants in (B). (C) Representative images, shoot fresh weight, and (D) leaf
number at 1-cm bolting of wild-type and mips1-2 mutant plants grown in
8L190:8L10:8D and 8L10:8L190:8D conditions. Ages of the plants were 28 or
36 days, as indicated. n = 5 plants in (C) and 20 to 27 plants in (D). (E) Representative
images and fresh weight of wild-type and mips1-2 mutant plants grown in 8L:16D
with light intensity of 200 mmol/m2/s and treated with 30-mM sorbitol or sucrose.
Plants were geminated and grew on 1/2 MS plates in 12L:12D for 7 days, then
transferred to the indicated media and grown in 8L:16D for 24 days. n = 5 plants.
In (C), (D), and (E), error bars represent SD. *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; ns,
not significant (Welch’s t test). Raw data and calculations for physiology
experiments are presented in table S2. (F) Model for two distinct photoperiod
measurement mechanisms controlling photoperiodic flowering and leaf growth. Within a
natural diurnal light cycle are multiple photoperiods, between which plants can
distinguish. The classic model of photoperiod flowering is shown on the left and the
model for photoperiodic growth on the right. 24-hour periods are shown with low light in
yellow, high light (above the compensation point) in orange, and dark in gray. For
photoperiodic flowering, low-light daylengths are measured to control flowering, and the
light-sensitive portion of the day can be determined by a night break experiment.
Light or dark is sensed by photoreceptors that control CO stability, which in turn
activates FT expression to trigger flowering. For photoperiodic growth, high light, or
photosynthetic period, is measured to trigger growth. Fast-break experiments show the
high light–sensitive portions of the day. Light above the compensation point is sensed
by the photosynthetic complex and measured by the metabolic daylength
measurement system to control MIPS1 expression, which is required with other
components to support rapid leaf growth in long days.

across the day by sequestering sugars as starch
and that this results in photoperiodic gene
expression.
Conversely, flowering time in 8L190:8L10:8D
was earlier than in the 8L200:16D condition,
suggesting that the flowering pathway is
still measuring the absolute daylength. The
8L10:8L190:8D growth condition induced flow-
ering earlier than in 8L190:8L10:8D (Fig. 4D
and fig. S10D), suggesting that high light from
ZT8 to ZT16 can accelerate long-day flowering
as well. Together, these results confirm the idea
that photoperiodic flowering is controlled by
low light, a near-absolute daylength, but growth
is controlled by the photosynthetic period, and
that these two photoperiods are dissociable
(Fig. 4F).
The fast-break experiments indicated that
plants are likely responding differentially to
the presence or absence of photosynthates
from ZT8 to ZT16. To test this, we performed
sucrose supplementation experiments in a
short-day growth condition, which should
aberrantly promote long-day–like growth and
thus require the sucrose-induced expression of
MIPS1 (Fig. 3B). We grew wild-type and mips1-2
mutant plants in 8L200:16D, where growth is
normally restricted and MIPS1 is not required,
in the presence of sucrose or a sorbitol control,
which cannot be hydrolyzed (Fig. 4E). Sucrose
promoted rapid growth in the wild-type plants,
and this growth was diminished in the mips1-2
mutant. This shows that plants are detecting

Wang et al., Science 383, eadg9196 (2024) 9 February 2024 7 of 9

RES EARCH | R E S E A R C H A R T I C L E
the photosynthetic period as a metabolic day-
length and that the promotion of growth by
photosynthetic sugars currently requires MIPS1.
Discussion
This work provides a new view of seasonal
timekeeping in plants by showing that photo-
periodic growth is controlled by a wholly
separate molecular mechanism from photo-
periodic flowering, despite often occurring in
the same season (Fig. 4F). We accomplished
this by identifying one of likely multiple long-
day photoperiodic growth pathways, from photo-
period detection to gene expression. This
photoperiodic growth pathway relies on the
measurement of daylength as the duration
of photosynthetic activity (9, 35, 36).
Starch, a storage sugar whose synthesis and
degradation are known to be controlled by the
circadian clock and photoperiod (10, 27, 29,
30, 37), is required for proper expression of
MIPS1 and seasonal growth. Starch degra-
dation can happen in the afternoon of long
days, providing a possible mechanism for
increasing cellular sugar concentrations in
a photoperiodic manner (28, 35) (fig. S11).
Identification of genes involved in starch
degradation might help to explain this daytime
degradation and whether they influence MIPS1
expression. There are multiple sugar sensors in
plants (38), and sucrose controls the cellular
levels of other sugars as well as ROS production
(39, 40). It will be important to determine
which of these play a role in regulating MIPS1
expression with the MIPS1promoter::Luciferase
and mips1-2 mutant phenotype.
Classic flowering mutants, such as ft or co,
with easily observable defects in flowering time
precipitated the discovery of the photoperiodic
flowering mechanism. In a similar way, our
studies relied on the easily observed growth
defects of the mips1 mutant to uncover a
mechanism for photoperiodic growth. The mips1
mutant is small and forms lesions exclusively
in long days, allowing us to easily detect the
mips1 phenotype as an indicator for long-day
photoperiod. Additionally, MIPS1 is expressed
with an easily distinguishable, photoperiod-
specific expression pattern, similar to that of FT.
This expression pattern correlates with active
times of rosette expansion (41). Myo-inositol,
the product of MIPS1, is an important precur-
sor to a variety of important cellular building
blocks needed for membrane and cell wall
synthesis (12, 13, 17, 42, 43). Additionally, it
has been implicated in the regulation of the
transport of growth hormone auxin, the pro-
duction of auxin conjugates, and plant defense,
all closely tied to plant growth. We show that
MIPS1 is required for growth in 16L:8D but is not
sufficient for growth in 8L:16D. Additionally, its
effects on growth are independent of its role
in lesion prevention in 16L:8D but could be
affected by its role in SA production. This leaves
Wang et al., Science 383, eadg9196 (2024) 9 February 2024
a few possibilities for the role of myo-inositol in
long-day growth. It could be coactivated with
another growth-promoting factor working to-
gether to promote growth, or alternately, it
could be coactivated with a growth-promoting
factor that produces a growth-restricting by-
product that is inactivated by myo-inositol.
Alternatively, there may be a growth-restrictive
mechanism that is activated in short days
that prevents myo-inositol from promoting
growth.
The fast-break experiments demonstrated that
growth is truly a photoperiodic process. How-
ever, the type of photoperiod that is detected
for growth, the photosynthetic period, is dif-
ferent from that detected for flowering, which
is a near-absolute photoperiod perceived by
low-light photoreceptors (33). This suggests that
different types of daylengths enact distinct
gene expression networks and is the simplest
explanation for the disconnect between photo-
periodic growth and flowering in short-day
plants that was recognized a century ago (6).
Together, our data also show that natural
diurnal light cycles do not constitute a single
photoperiod, but rather are composed of many
different photoperiod types that can be differ-
entially detected to independently coordinate
seasonal developmental processes.
MATERIALS AND METHODS
Arabidopsis materials
The Arabidopsis seeds of wild type and mips1-
2 (SALK_023626), mips1-3 (SAIL_676_D08),
mips1-4 (CS851587), pgm-1 (CS210), co-9
(CS870084), and ft-10 (CS9869) mutants were
obtained from ABRC. Lux-4 mutant seeds
were obtained from Dr. Dmitri Nusinow (23).
sid2 mutant seed was obtained from Dr. Raynaud
(11). The mips1-2 mutant was also crossed to co-9,
pgm-1, and sid2 mutant plants, respectively,
and the co-9 mips1-2, pgm-1 mips1-2, and
sid2 mips1-2 double mutants were identified
by phenotyping and genotyping. The pgm-1 allele
was genotyped as described in (44). The lux-4
allele was genotyped by PCR, followed by
sequencing. The sid2 allele was genotyped by
PCR, followed by sequencing (11). The primers
used for genotyping are listed in table S1.
Arabidopsis growth conditions
A. thaliana seeds from wild type and mutants
were surface sterilized for 20 min in 70%
ethanol with 0.1% Triton X-100, then sown
on freshly poured 1/2 MS plates, pH 5.7 (Cassion
Laboratories, cat. # MSP01), and 0.8% bacterio-
logical agar (AmericanBio, cat. # AB01185) with-
out sucrose. The seeds were stratified in the dark
for two days at 4°C, then transferred into 22°C
12L:12D illuminated by white fluorescent lamps
at 150 mmol/m2/s for seven days. The seven-day-
old seedlings were then transferred to different
photoperiods for given experiments, as indi-
cated. For sucrose treatment, seedlings trans-
ferred to culture vessels (vented, PhytoTech)
contained 1/2 MS with 30mM Sorbitol or
Sucrose. For soil-grown plants, after two days
of stratification, seeds were germinated and
grown in Premier Horticulture PRO-MIX BX
at 22°C in 16L:8D or 8L:16D with a light in-
tensity of 10/40/100/190/200 mmol/m2/s, as
indicated. For Myo-inositol (Sigma-Aldrich)
treatment, the plant was sprayed with 50-mg/
mL myo-inositol solution (dissolved in water)
or water every two days (11).
Calculation of pairwise correlation between
sucrose content and MIPS1 expression level
Mean sucrose content in 16L:8D and 8L:16D
in Arabidopsis leaves was obtained from Lu et al.
(22). Pearson’s correlation test was performed
with the R function cor.test to assess the sta-
tistical significance of the correlation between
the sucrose content and MIPS1 expression level
(45). Regression was performed with the ggplot2
package geom-smooth [method = (lm)] (46).
Plasmid construction
For the MIPS1 overexpression constructs, the
MIPS1 coding sequence was obtained with PCR
by using wild-type complementary DNA as the
template, inserted into pENTR/D-TOPO and
then transferred into PB7-HFN destination
vectors by using LR recombination (47). To
generate the MIPS1promoter::Luciferase cons-
truct, a 1537-bp promoter sequence upstream
of the MIPS1 coding sequence, including a
5′ untranslated region, was obtained with
PCR and inserted into pDONR207 vectors
(Invitrogen) and then transferred into the
pFLASH destination vectors to drive the lucif-
erase (48). The destination vectors were then
transformed into agrobacterium GV3101 to gen-
erate transgenic plants. The primers used for
cloning are listed in table S1.
Luciferase imaging, analysis, and
data normalization
Seven-day-old seedlings grown in 12L:12D at
22°C were transferred onto a 10×10-grid fresh-
ly poured 100-mm square MS plates with
or without added sugars as indicated for a
given experiments. Seedlings were then treated
with 5-mM D-luciferin (Cayman Chemical
Company, cat. # 115144-35-9), dissolved in
0.01% TritonX-100, and imaged at 22°C under
the indicated conditions. For Fig. 2B and figs.
S8 and S10E, data was normalized with min-max
scaling as described in (9). For Fig. 1, C and H,
and fig. S3, the values of MIPS1 daily expres-
sion were normalized to ZT4, the time point
that showed the same expression levels in tested
photoperiods. The mean and standard devia-
tion of these rate values were calculated with
the last four days of imaging under the same
light conditions. For Fig. 3D, the intensity of
MIPS1promoter::LUC was normalized to ZT0
of day 10.
8 of 9
RES EARCH | R E S E A R C H A R T I C L E
qRT-PCR
For qRT-PCR experiments, RNA extraction was
15.
performed with two different methods. For Fig.
1, A and B, total RNA was extracted from
Arabidopsis seedlings grown in indicated con-
16.
ditions by using TRIzolTM reagent (Thermo-
Fisher, cat. # 15596026); for the remaining Figs.
2A and 3, A, B, D, and fig. S9B, extraction was
17.
performed with RNeasy Plant Mini Kit (QIAGEN
cat. # 74904). In both methods, the RNA was
subsequently treated with deoxyribonuclease
(QIAGEN, cat. # 79254). The subsequent reverse
18.
transcription and conditions for qRT-PCR reac-
tions were described previously with minor
modifications (34). Briefly, 400 ng of total
19.
RNA were used for reverse transcription with
iScript Reverse Transcription Supermix for qRT-
PCR (Bio-Rad, cat. # 1708841). iTaq Universal
20.
SYBR Green Supermix was used for the qRT-PCR
reaction (Bio-Rad, cat. # 1725121). IPP2 (AT3G02780)
or UBQ10 (AT4G05320) was used as an internal
21.
control as indicated. The relative expression
represents means of 2(−DDCT) from three bio-
logical replicates. The primers used are listed
22.
in table S1.
RE FE RENCES AND N OT ES
23.
1. R. J. Nelson, D. L. Denlinger, D. E. Somers, Eds., ProQuest,
Photoperiodism: The biological calendar (Oxford Univ. Press,
2009). doi: 10.1093/acprof:oso/9780195335903.001.0001
2. M. J. Yanovsky, S. A. Kay, Molecular basis of seasonal time 24.
measurement in Arabidopsis. Nature 419, 308–312 (2002).
doi: 10.1038/nature00996; pmid: 12239570
3. J. Putterill, F. Robson, K. Lee, R. Simon, G. Coupland, The
CONSTANS gene of Arabidopsis promotes flowering and
encodes a protein showing similarities to zinc finger 25.
transcription factors. Cell 80, 847–857 (1995). doi: 10.1016/
0092-8674(95)90288-0; pmid: 7697715
4. P. Suárez-López et al., CONSTANS mediates between the circadian 26.
clock and the control of flowering in Arabidopsis. Nature 410,
1116–1120 (2001). doi: 10.1038/35074138; pmid: 11323677
5. M. Osnato, I. Cota, P. Nebhnani, U. Cereijo, S. Pelaz,
Photoperiod Control of Plant Growth: Flowering Time Genes 27.
Beyond Flowering. Front. Plant Sci. 12, 805635 (2022).
doi: 10.3389/fpls.2021.805635; pmid: 35222453
6. D. Vince-Prue, Photoperiodism in plants, European plant
biology series (McGraw-Hill, 1975), pp. doi: XIV. 28.
7. C. C. Leung, D. A. Tarté, L. S. Oliver, Q. Wang, J. M. Gendron,
Systematic characterization of photoperiodic gene expression
patterns reveals diverse seasonal transcriptional systems in
29.
Arabidopsis. PLOS Biol. 21, e3002283 (2023). doi: 10.1371/
journal.pbio.3002283; pmid: 37699055
8. T. C. Mockler et al., The DIURNAL project: DIURNAL and circadian
expression profiling, model-based pattern matching, and promoter
analysis. Cold Spring Harb. Symp. Quant. Biol. 72, 353–363 (2007).
30.
doi: 10.1101/sqb.2007.72.006; pmid: 18419293
9. W. Liu et al., A metabolic daylength measurement system mediates
winter photoperiodism in plants. Dev. Cell 56, 2501–2515.e5
(2021). doi: 10.1016/j.devcel.2021.07.016; pmid: 34407427
31.
10. R. Sulpice et al., Starch as a major integrator in the regulation
of plant growth. Proc. Natl. Acad. Sci. U.S.A. 106, 10348–10353
(2009). doi: 10.1073/pnas.0903478106; pmid: 19506259
11. P. H. Meng et al., Crosstalks between myo-inositol
metabolism, programmed cell death and basal immunity in
Arabidopsis. PLOS ONE 4, e7364 (2009). doi: 10.1371/ 32.
journal.pone.0007364; pmid: 19812700
12. J. L. Donahue et al., The Arabidopsis thaliana Myo-inositol
1-phosphate synthase1 gene is required for Myo-inositol synthesis
and suppression of cell death. Plant Cell 22, 888–903 (2010).
doi: 10.1105/tpc.109.071779; pmid: 20215587 33.
13. H. Chen, L. Xiong, myo-Inositol-1-phosphate synthase is
required for polar auxin transport and organ development.
J. Biol. Chem. 285, 24238–24247 (2010). doi: 10.1074/ 34.
jbc.M110.123661; pmid: 20516080
14. L. Ma et al., Arabidopsis FHY3 and FAR1 Regulate Light-Induced
myo-Inositol Biosynthesis and Oxidative Stress Responses by
Wang et al., Science 383, eadg9196 (2024) 9 February 2024
Transcriptional Activation of MIPS1. Mol. Plant 9, 541–557 (2016).
doi: 10.1016/j.molp.2015.12.013; pmid: 26714049
A. M. Murphy, B. Otto, C. A. Brearley, J. P. Carr, D. E. Hanke, A
role for inositol hexakisphosphate in the maintenance of basal
resistance to plant pathogens. Plant J. 56, 638–652 (2008).
doi: 10.1111/j.1365-313X.2008.03629.x; pmid: 18643983
S. Chaouch, G. Noctor, Myo-inositol abolishes salicylic acid-
dependent cell death and pathogen defence responses triggered
by peroxisomal hydrogen peroxide. New Phytol. 188, 711–718
(2010). doi: 10.1111/j.1469-8137.2010.03453.x; pmid: 20807338
Y. Luo et al., D-myo-inositol-3-phosphate affects
phosphatidylinositol-mediated endomembrane function in
Arabidopsis and is essential for auxin-regulated
embryogenesis. Plant Cell 23, 1352–1372 (2011).
doi: 10.1105/tpc.111.083337; pmid: 21505066
C. Nawrath, J. P. Métraux, Salicylic acid induction-deficient
mutants of Arabidopsis express PR-2 and PR-5 and
accumulate high levels of camalexin after pathogen
inoculation. Plant Cell 11, 1393–1404 (1999). pmid: 10449575
M. C. Wildermuth, J. Dewdney, G. Wu, F. M. Ausubel,
Isochorismate synthase is required to synthesize salicylic acid
for plant defence. Nature 414, 562–565 (2001).
doi: 10.1038/35107108; pmid: 11734859
S. Balasubramanian, S. Sureshkumar, J. Lempe, D. Weigel,
Potent induction of Arabidopsis thaliana flowering by elevated
growth temperature. PLOS Genet. 2, e106 (2006).
doi: 10.1371/journal.pgen.0020106; pmid: 16839183
S. K. Yoo et al., CONSTANS activates SUPPRESSOR OF
OVEREXPRESSION OF CONSTANS 1 through FLOWERING LOCUS
T to promote flowering in Arabidopsis. Plant Physiol. 139, 770–778
(2005). doi: 10.1104/pp.105.066928; pmid: 16183837
Y. Lu, J. P. Gehan, T. D. Sharkey, Daylength and circadian effects on
starch degradation and maltose metabolism. Plant Physiol. 138,
2280–2291 (2005). doi: 10.1104/pp.105.061903; pmid: 16055686
S. P. Hazen et al., LUX ARRHYTHMO encodes a Myb domain
protein essential for circadian rhythms. Proc. Natl. Acad.
Sci. U.S.A. 102, 10387–10392 (2005). doi: 10.1073/
pnas.0503029102; pmid: 16006522
Y. Gibon et al., Adjustment of growth, starch turnover, protein
content and central metabolism to a decrease of the carbon
supply when Arabidopsis is grown in very short photoperiods.
Plant Cell Environ. 32, 859–874 (2009). doi: 10.1111/
j.1365-3040.2009.01965.x; pmid: 19236606
M. Stitt, S. C. Zeeman, Starch turnover: Pathways, regulation
and role in growth. Curr. Opin. Plant Biol. 15, 282–292 (2012).
doi: 10.1016/j.pbi.2012.03.016; pmid: 22541711
S. K. Pal et al., Diurnal changes of polysome loading track
sucrose content in the rosette of wild-type Arabidopsis and the
starchless pgm mutant. Plant Physiol. 162, 1246–1265 (2013).
doi: 10.1104/pp.112.212258; pmid: 23674104
R. Sulpice et al., Arabidopsis coordinates the diurnal regulation
of carbon allocation and growth across a wide range of
photoperiods. Mol. Plant 7, 137–155 (2014). doi: 10.1093/mp/
sst127; pmid: 24121291
O. Fernandez et al., Leaf Starch Turnover Occurs in Long Days
and in Falling Light at the End of the Day. Plant Physiol. 174,
2199–2212 (2017). doi: 10.1104/pp.17.00601; pmid: 28663333
V. Mengin et al., Photosynthate partitioning to starch in
Arabidopsis thaliana is insensitive to light intensity but
sensitive to photoperiod due to a restriction on growth in the
light in short photoperiods. Plant Cell Environ. 40, 2608–2627
(2017). doi: 10.1111/pce.13000; pmid: 28628949
T. A. Moraes et al., Response of the Circadian Clock and Diel
Starch Turnover to One Day of Low Light or Low CO2.
Plant Physiol. 179, 1457–1478 (2019). doi: 10.1104/pp.18.01418;
pmid: 30670603
T. Caspar, S. C. Huber, C. Somerville, Alterations in growth,
photosynthesis, and respiration in a starchless mutant of
Arabidopsis thaliana (L.) deficient in chloroplast
phosphoglucomutase activity. Plant Physiol. 79, 11–17 (1985).
doi: 10.1104/pp.79.1.11; pmid: 16664354
A. Flis et al., Photoperiod-dependent changes in the
phase of core clock transcripts and global transcriptional
outputs at dawn and dusk in Arabidopsis. Plant Cell
Environ. 39, 1955–1981 (2016). doi: 10.1111/pce.12754;
pmid: 27075884
F. Valverde et al., Photoreceptor regulation of CONSTANS
protein in photoperiodic flowering. Science 303, 1003–1006
(2004). doi: 10.1126/science.1091761; pmid: 14963328
C. M. Lee et al., Decoys Untangle Complicated Redundancy and
Reveal Targets of Circadian Clock F-Box Proteins.
Plant Physiol. 177, 1170–1186 (2018). doi: 10.1104/
pp.18.00331; pmid: 29794020
35. J. M. Gendron, C. C. Leung, W. Liu, Energy as a seasonal signal
for growth and reproduction. Curr. Opin. Plant Biol. 63, 102092
(2021). doi: 10.1016/j.pbi.2021.102092; pmid: 34461431
36. J. M. Gendron, D. Staiger, New Horizons in Plant
Photoperiodism. Annu. Rev. Plant Biol. 74, 481–509 (2023).
doi: 10.1146/annurev-arplant-070522-055628; pmid: 36854481
37. A. Graf, A. Schlereth, M. Stitt, A. M. Smith, Circadian control of
carbohydrate availability for growth in Arabidopsis plants at
night. Proc. Natl. Acad. Sci. U.S.A. 107, 9458–9463 (2010).
doi: 10.1073/pnas.0914299107; pmid: 20439704
38. A. Choudhary, A. Kumar, N. Kaur, H. Kaur, Molecular cues of
sugar signaling in plants. Physiol. Plant. 174, e13630 (2022).
doi: 10.1111/ppl.13630; pmid: 35049040
39. X. Li et al., A reactive oxygen species Ca2+ signalling pathway
identified from a chemical screen for modifiers of sugar-
activated circadian gene expression. New Phytol. 236,
1027–1041 (2022). doi: 10.1111/nph.18380; pmid: 35842791
40. Á. Román et al., Superoxide is promoted by sucrose and
affects amplitude of circadian rhythms in the evening.
Proc. Natl. Acad. Sci. U.S.A. 118, e2020646118 (2021).
doi: 10.1073/pnas.2020646118; pmid: 33674383
41. F. Apelt, D. Breuer, Z. Nikoloski, M. Stitt, F. Kragler,
Phytotyping(4D) : A light-field imaging system for non-invasive
and accurate monitoring of spatio-temporal plant growth.
Plant J. 82, 693–706 (2015). doi: 10.1111/tpj.12833;
pmid: 25801304
42. Q. Bruggeman et al., Chloroplast Activity and
3‘phosphadenosine 5’phosphate Signaling Regulate
Programmed Cell Death in Arabidopsis. Plant Physiol. 170,
1745–1756 (2016). doi: 10.1104/pp.15.01872; pmid: 26747283
43. Q. Bruggeman et al., Involvement of Arabidopsis
Hexokinase1 in Cell Death Mediated by Myo-Inositol
Accumulation. Plant Cell 27, 1801–1814 (2015). doi: 10.1105/
tpc.15.00068; pmid: 26048869
44. K. M. Veley, S. Marshburn, C. E. Clure, E. S. Haswell,
Mechanosensitive channels protect plastids from hypoosmotic
stress during normal plant growth. Curr. Biol. 22, 408–413
(2012). doi: 10.1016/j.cub.2012.01.027; pmid: 22326022
45. R Core Team, R: A Language and Environment for Statistical
Computing (2018); https://www.R-project.org/.
46. H. Wickham, ggplot2: Elegant Graphics for Data Analysis
(Springer International Publishing, 2016).
47. H. Huang et al., Identification of Evening Complex Associated
Proteins in Arabidopsis by Affinity Purification and Mass
Spectrometry. Mol. Cell. Proteomics 15, 201–217 (2016).
doi: 10.1074/mcp.M115.054064; pmid: 26545401
48. J. M. Gendron et al., Arabidopsis circadian clock protein,
TOC1, is a DNA-binding transcription factor. Proc. Natl. Acad.
Sci. U.S.A. 109, 3167–3172 (2012). doi: 10.1073/
pnas.1200355109; pmid: 22315425
AC KNOWLED GME NTS
We thank C. Adamchek for technical support. We also thank
S. Pariseau and J. Pengsavath for administrative support.
Additionally, we thank C. Bolick, N. Guzzo, and the staff at Marsh
Botanical Gardens for their support in maintaining plant growth
spaces. We also thank A. Saffer, M.-W. Li, L. Oliver, H. Lowrey,
D. Clark, and S. Bahmanyar for insightful discussions and critical
reading of the manuscript. Funding: J.M.G. was supported by
the National Institutes of Health (R35 GM128670). W.L. and Q.W.
were supported by the Forest B.H. and Elizabeth D.W. Brown Fund
Fellowship. D.A.T. was supported by the National Institutes
of Health (T32GM007223-44). Author contributions: Q.W., C.C.L.,
W.L., and J.M.G. designed the experiments. Q.W., C.C.L., W.L.,
and D.A.T. performed the experiments and experimental analyses.
Q.W. and J.M.G. wrote the article. Competing interests: The
authors claim no competing interests. Data and materials
availability: All data are available in the manuscript or the
supplementary materials. All materials are freely available upon
request. License information: Copyright © 2024 the authors, some
rights reserved; exclusive licensee American Association for the
Advancement of Science. No claim to original US government works.
https://www.science.org/about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adg9196
Figs. S1 to S11
Tables S1 and S2
MDAR Reproducibility Checklist
Submitted 1 February 2023; resubmitted 17 July 2023
Accepted 12 December 2023
10.1126/science.adg9196
9 of 9
RES EARCH
◥ ing the function of the mutant EGFR gene and
RESEARCH ARTICLE SUMMARY
CANCER
Lineage-specific intolerance to oncogenic drivers
restricts histological transformation
Eric E. Gardner*, Ethan M. Earlie, Kate Li, Jerin Thomas, Melissa J. Hubisz, Benjamin D. Stein,
Chen Zhang, Lewis C. Cantley†, Ashley M. Laughney*, Harold Varmus*
INTRODUCTION: What determines the behavior endocrine cancer indistinguishable from small
and appearance of different cancers? Several cell lung cancer (SCLC), which is therapeuti-
factors may be involved, including the kind of cally recalcitrant and connotes poor prognosis.
normal cell from which a cancer arises (the cell These events are considered off-target mecha-
of origin) and the set of specific genes (espe- nisms of resistance because the original onco-
cially tumor suppressors and proto-oncogenes) genic driver pathway is thought to be no longer
that are altered during the conversion of a nor- critical for tumor cell proliferation. Thus, a
mal cell to a malignant one. The variations in distinct, separate driver program has emerged.
appearance and behavior and the affected genes
often correlate with a specific cell of origin, sug- RATIONALE: The complex process of HT is dif-
gesting that different cell lineages show different ficult to characterize rigorously in human sub-
responses to various genetic changes. Context is jects. But the possibility of modeling HT in the
key—not every oncogene is broadly oncogenic. lungs of experimental mice offers an opportu-
Some types of cancers, such as adenocarcinomas nity to study the relationships among cancer
of the lung and the prostate, undergo dramatic phenotypes, cells of origin, and genetic alter-
changes in appearance and behavior and dis- ations. To that end, we proposed to use genetic
play an altered genetic profile when they devel- engineering to mimic the events observed in
op resistance to targeted therapies. In lung human lung HT. We planned to generate LUAD
cancers, this histological transformation (HT) driven by a mutationally activated epidermal
can result in the conversion of lung adenocar- growth factor receptor (EGFR) gene and then
cinoma (LUAD) into an aggressive type of neuro- to transform those tumors into SCLCs by block-
Histological
Transformation (HT)
EGFR MYC
Lung Small Cell
Adenocarcinoma Lung Cancer
Residual
(LUAD) (SCLC)
disease
Proximal Mapk Myc
airway
program program
Distal
airway
Basal “stem-like”
intermediate
Pulmonary
Alveolar Type II
Pneumocyte Neuroendocrine
(AT2) Cell (PNEC)
Histological transformation through a stem-like intermediate. Gardner et al. developed a genetically
engineered mouse model of histological transformation (HT) of EGFR-driven lung adenocarcinoma (LUAD) to
small cell lung cancer (SCLC) and closely examined this phenomenon with single-cell RNA-sequencing. They
found that the intermediate state between LUAD and SCLC was stem-like and most closely resembled a
pulmonary basal cell.
Gardner et al., Science 383, 603 (2024) 9 February 2024
introducing changes in other genes, such as the
oncogene Myc and the tumor suppressors Rb1
and Trp53. We would then be able to study HT in
detail by performing single-cell RNA-sequencing
during this process.
RESULTS: We were able to reconstruct pulmo-
nary HT in a mouse model. We then charac-
terized the transcriptional programs throughout
three stages: during growth of EGFR-driven
LUAD; in the limited disease that remained
after reduction of oncogenic EGFR; and during
the transformation of tumor cells into neuro-
endocrine SCLC. By following the expression of
single oncogenes in different lung-cell lineages,
we demonstrated that the process of HT is reg-
ulated by the tolerance of different cell types to
distinct oncogenic drivers. Thus, whereas most
lung cells are resistant to transformation by Myc,
neuroendocrine cells are highly sensitive to
its oncogenic effects; however, their fitness
is impaired by mutated EGFR. Conversely,
lung alveolar epithelial cells grow excessively
in response to mutated EGFR but are impaired
by Myc alone.
During the process of HT, undifferentiated
stem-like cells accumulate, and some become
neuroendocrine in character as SCLC emerges.
This process appears to require the loss of two
tumor suppressor genes, Trp53 and Rb1, that
are almost universally inactivated in human
SCLC. Furthermore, we found that loss of yet
another tumor suppressor gene, Pten, allowed
Myc to transform the AT2 lineage. However,
the additional loss of Rb1 was required for
transformation to a neuroendocrine pheno-
type. These results suggest that both Myc
and Rb1 are critical regulators of neuro-
endocrine HT.
CONCLUSION: By developing this novel experi-
mental model for pulmonary HT, we have re-
capitulated the complex sequence of events
that depends on varied responses of different
cell lineages to changes in a few known cancer
genes. These findings will enable efforts to study
the responses of various cell types to changes
in other genes implicated in carcinogenesis
and to explore strategies for treating aggre-
ssive cancers such as SCLC with agents that
target MYC for which approved, targeted ther-
apeutics are not yet available.
▪
The list of author affiliations is available in the full article online.
*Corresponding author. Email: eeg2001@med.cornell.edu
(E.E.G); ashley.laughney@gmail.com (A.M.L.); varmus@
med.cornell.edu (H.V.)
†Present address: Department of Cell Biology and Dana-Farber
Cancer Institute, Harvard Medical School, Boston, MA.
Cite this article as E. E. Gardner et al., Science 383, eadj1415
(2024). DOI: 10.1126/science.adj1415
READ THE FULL ARTICLE AT
https://doi.org/10.1126/science.adj1415
1 of 1
RES EARCH

◥ the pulmonary neuroendocrine lineage is. We

RESEARCH ARTICLE addressed these questions by combining mod-
els of genetically engineered lung tumorigenesis,
CANCER lineage tracing, and chronologic single-cell
RNA-sequencing (scRNA-seq) to map cell type–
Lineage-specific intolerance to oncogenic drivers specific oncogenic driver programs and their
contributions to HT.
restricts histological transformation Results
Eric E. Gardner1*, Ethan M. Earlie1,2,3, Kate Li1, Jerin Thomas1, Melissa J. Hubisz1,2,3,4, Histologically distinct lung tumors in an isogenic
Benjamin D. Stein1,5, Chen Zhang6, Lewis C. Cantley1,5,†, Ashley M. Laughney1,2,3*, Harold Varmus1* mouse model
To chronicle HT, we generated a new genet-
Lung adenocarcinoma (LUAD) and small cell lung cancer (SCLC) are thought to originate from different ically engineered mouse model (GEMM) that
epithelial cell types in the lung. Intriguingly, LUAD can histologically transform into SCLC after treatment combines the conditional expression of Myc,
with targeted therapies. In this study, we designed models to follow the conversion of LUAD to SCLC and rtTA3, and tdTomato (lox-stop-lox alleles) with
found that the barrier to histological transformation converges on tolerance to Myc, which we implicate the loss of tumor suppressors Rb1 and Trp53
as a lineage-specific driver of the pulmonary neuroendocrine cell. Histological transformations are (floxed alleles) and a doxycycline (DOX)–inducible,
frequently accompanied by activation of the Akt pathway. Manipulating this pathway permitted tolerance oncogenic EGFR transgene (20). The model (here-
to Myc as an oncogenic driver, producing rare, stem-like cells that transcriptionally resemble the after ERPMT) allows for two classes of manip-
pulmonary basal lineage. These findings suggest that histological transformation may require the plasticity ulations: (i) control of the cell of origin through
inherent to the basal stem cell, enabling tolerance to previously incompatible oncogenic driver programs. lineage-restricted expression of Cre recombinase
after intratracheal infection by using adenoviral

H
istological transformation (HT) is a poor-
ly understood process whereby a cancer’s
initial histology is altered and presents as
a new histologic type of cancer. These
changes are presumed to be under the
selective pressure of an oncogene-targeted the-
rapy. Best described in the context of epidermal
growth factor receptor (EGFR) inhibition in
lung adenocarcinoma (LUAD) (1–3) and andro-
gen receptor inhibition of prostate adeno-
carcinoma (4, 5), these transformations most
commonly lead to neuroendocrine and/or
squamous differentiation (6). Considered an
off-target form of acquired resistance, the
cancer’s proliferative character is no longer
dependent on the original oncogenic driver
pathway. In the earliest reports of HT, it was
unclear whether the recurrent tumors were
independent, primary small cell lung cancer
(SCLC), subclones selected for under the pres-
sure of targeted therapy, or products of di-
rect conversion of LUAD into SCLC. Supporting
evidence for direct conversion relies on the
continued presence of a LUAD driver oncogene
within a cancer that is histologically SCLC.
Such oncogenic mutations in genes that
commonly drive the formation of LUADs have
rarely been encountered in genome sequenc-
ing studies of primary, treatment-naïve SCLC
(7, 8). Most posttransformation samples
1
Meyer Cancer Center, Weill Cornell Medicine, New York, NY.
2
Department of Physiology, Biophysics, and Systems Biology,
Weill Cornell Medicine, New York, NY. 3Institute for
Computational Biomedicine, Weill Cornell Medicine, New
York, NY. 4Bioinformatics Facility, Institute of Biotechnology,
Cornell University, Ithaca, NY. 5Department of Medicine, Weill
Cornell Medicine, New York, NY. 6Department of Pathology
and Laboratory Medicine, Weill Cornell Medicine, New York, NY.
*Corresponding author. Email: eeg2001@med.cornell.edu (E.E.G);
ashley.laughney@gmail.com (A.M.L.); varmus@med.cornell.edu (H.V.)
†Present address: Department of Cell Biology and Dana-Farber Cancer
Institute, Harvard Medical School, Boston, MA.
have shared, clonally related mutations present
in the LUAD before undergoing HT; however,
protein expression of the LUAD driver is cons-
picuously absent after transformation (9). Fur-
thermore, loss of the retinoblastoma (RB) tumor
suppressor is observed in cases of HT in both
lung and prostate cancers (4, 10), and patients
with EGFR-driven LUADs harboring inactiva-
tion of TP53 and RB1 are at an especially high
risk for developing SCLC after targeted therapy
(1, 11).
Primary LUAD and SCLC are thought to
develop from distinct cell types in the lung: the
alveolar type II (AT2) cell and the pulmonary
neuroendocrine cell (PNEC), respectively. Much
is known about surfactant-producing AT2 cells
as a cell of origin for LUAD, for which the EGF
signaling pathway is an established mitogenic
program (12–14). Hence, activating mutations
and amplifications of genes that encode pro-
teins in the mitogen-activated protein kinase
(MAPK) pathway, including RAS and EGFR,
are common in LUAD (15). By contrast, SCLC
is thought to arise predominantly from PNECs
(8, 16, 17). PNECs are rare and are found near
anatomic branchpoints of the large airway;
they function as sentinels for inhaled patho-
gens and environmental changes, signaling
to other cells through both electrochemical
innervation and secretory function (17–19).
Because of their scarcity, much less is known
about PNEC biology, including what signal-
ing events drive their proliferation.
We hypothesized that the complexity of HT
may be simplified to represent a mechanism
by which a cell can change its oncogenic driver
program. We set out to address several fun-
damental unknowns: (i) how an adenocarcinoma
transforms to a high-grade neuroendocrine can-
cer, (ii) what the intermediate steps in HT are,
and (iii) what the oncogenic driver program of
vectors with cell type–specific promoters (21),
and (ii) oncogenic EGFRL858R expression con-
trolled in a DOX-dependent manner (Fig. 1A).
Tumors initiated in the AT2 or PNEC lineages
in ERPMT mice produced histologically dis-
tinct lung tumors with opposing dependencies
on the presence or absence of DOX. If expression
of Cre was initiated in the AT2 lineage and the
mice received DOX, the ERPMT model devel-
oped an aggressive LUAD. By contrast, no mice
succumbed to disease in this timeframe if they
were off DOX. In mice receiving DOX, we
observed multifocal, glandular lesions consist-
ent with LUAD throughout all lobes of the
airway, demonstrating homogenous expression
of tdTomato (tdTom) and lack of the neuro-
endocrine marker synaptophysin (Fig. 1B).
Conversely, if expression of Cre was initiated
in the PNEC lineage, then the cohort off DOX
developed an aggressive neuroendocrine SCLC
and the mice on DOX appeared healthy at a
time when 100% of the cohort off DOX was
moribund (Fig. 1C).
LUAD and SCLC tumorigenesis developed
with similar penetrance and latency in the
ERPMT model. To compare these models
transcriptionally, we isolated tdTom+ cells in
the on-DOX group from the AT2 lineage (here-
after labeled LUAD; red) and the off-DOX group
from the PNEC lineage (hereafter labeled SCLC;
blue) at 8 weeks on study and performed scRNA-
seq. As expected, these tumor cells were tran-
scriptionally distinct and closely resembled their
precursors (Fig. 1D and fig. S1, A and B). LUAD
cells exhibited mixed AT2 and AT1 character,
which is consistent with prior observations in
the regenerating airway (22) and in mouse
(12) and human tumors (23). Compared with
LUAD, SCLC had lower expression of com-
ponents of the major histocompatibility complex
(MHC) class II antigen pathway (fig. S1C), which
is consistent with MHC class II deficiency in

Gardner et al., Science 383, eadj1415 (2024) 9 February 2024 1 of 11

RES EARCH | R E S E A R C H A R T I C L E

in the SCLC model (fig. S1F) (30). Taken to-

A D
gether, we generated a mouse model capable
of forming two histologically distinct lung
cancers that are united by the process of HT—
prompting us to ask whether the LUAD tumors
in these mice could be encouraged to transform
to SCLC.

EGFR removal and the emergence of

neuroendocrine character
We speculated that the ERPMT model could
be used to understand conversion of LUAD to
SCLC, provided adequate selective pressure was
B applied against the LUAD driver. After the
development of late-stage LUAD, we random-
ized cohorts to one of three arms: remain on
DOX (ERPMT on), come off DOX for 1 month
and then restart DOX (ERPMT on > off > on),
or come off DOX and remain off for the dura-
tion of the study (ERPMT on > off). Terminal
lung cancers developed at statistically differ-
ent rates across these perturbations (Fig. 2A).
If DOX was permanently removed, the result-
ing lung tumors were consistent with SCLC,
but if DOX was restarted (ERPMT on > off >
on), tumors were consistent with LUAD, albeit
C with fewer, larger lesions than were found in
mice continually on DOX (Fig. 2B).
To better understand the proliferative nature
of residual disease that emerges upon EGFR
withdrawal, we randomized ERPMT mice de-
veloping LUAD to come off or stay on DOX
after three daily pulses of the S-phase label 5-
ethynyl-2′-deoxyuridine (EdU). Three weeks
later we labeled again, now with a different
S-phase analog, bromodeoxyuridine (BrdU),
thus labeling cells with a proliferative history
that continued to cycle after tumor regression
(fig. S2, A and B). Tumors that remained on
Fig. 1. A GEMM model to generate distinct histologic subtypes of lung cancer. (A) Nomenclature DOX were EdU+/BrdU+, whereas off DOX,
used for the ERPMT model, with abbreviated alleles bolded and underlined and general strategy to use residual tdTom+ cells were present as single
promoter-restricted adenovirus to initiate tumorigenesis in specific airway cells. (B) Survival and histologic cells and were EdU+/BrdU−. This result sug-
appearance of the ERPMT model initiated in alveolar type II (AT2) cells (red cartoon cells) or (C) pulmonary gested that residual cells did not continue to
neuroendocrine (PNEC) cells (blue cartoon cells) when mice are on DOX-containing (dashed line; n = 10 cycle after DOX removal (fig. S2, C and D).
mice per group) or control diet (solid line; n = 10 mice per group). Representative sagittal hematoxylin and Manipulating transcription of EGFR with
eosin (H&E) and tdTomato (tdTom) IHC lung sections alongside high-powered H&E and synaptophysin DOX is a convenient and effective experimen-
(Syp) IHC (in boxes on right; scalebars, 100 mm) from the LUAD model initiated from AT2 cells on DOX tal approach, but it does not precisely recapitu-
(above) or the SCLC model initiated from PNEC cells off DOX (below). (D) Mean imputed expression of late the clinical scenario in which patients are
AT2 and PNEC lineage markers (table S1) in single cells isolated from LUAD (red; n = 5394 cells) or SCLC treated with inhibitors of the EGFR kinase
(blue; n = 4371 cells) ERPMT models with overlaid kernel density estimates (KDEs) reflecting cell density; domain. To address this difference, we com-
tdTom+ tumor cells sorted and pooled from n = 3 mice at 8 weeks after infection. pared inhibiting the kinase activity of EGFR
using osimertinib with the effect of removing
DOX, thus gradually extinguishing EGFR ex-
primary SCLC (24, 25) and AT2-specific Notch2, Hes1, and Sftpc that were differentially pression. After 8 weeks of LUAD development,
expression of MHC class II molecules (26). accessible in LUAD as compared with SCLC the frequency of tdTom+ cells in the airway of
Instead, most SCLC tumor cells expressed fac- (28, 29). By contrast, the SCLC model had greater ERPMT mice ranged from ~5 to 35%. However,
tors that suppress Notch signaling, including accessibility of neuroendocrine genes, includ- this was reduced to <1% in all animals after
the inhibitory ligand Dll3 and transcription ing Insm1, Chga, and Ascl1 (fig. S1E). Fur- 1 month of treatment with osimertinib (osi) (31)
factor Hes6 (27). Conversely, the LUAD model thermore, differentially accessible peaks in the or the removal of DOX (Fig. 2C). We isolated
was enriched for Notch-stimulating factors, MAPK-driven LUAD were significantly enriched tdTom+ cells from these groups and found
which is in keeping with this model presenting for activator protein 1 (AP-1) motifs such as Jun that genetic—as opposed to pharmacologic—
as a nonneuroendocrine, alveolar-derived LUAD and Fos, whereas several basic helix-loop-helix suppression of EGFR led to a greater increase
(fig. S1D). Additionally, bulk ATAC-sequencing (bHLH) regulatory elements involved in neuro- in Ascl1-expressing cells, the definitive lineage
revealed regions of chromatin that mapped to genesis (Tcf21, Tcf4, and Ascl1) were enriched marker of the PNEC (32) (fig. S3, A and B).

Gardner et al., Science 383, eadj1415 (2024) 9 February 2024 2 of 11

RES EARCH | R E S E A R C H A R T I C L E
A B
D E
Fig. 2. Tracing the origins and transitions between LUAD and SCLC in vivo.
(A) Survival of AT2-derived ERPMT model on DOX (red), after DOX removal at
8 weeks (teal), and after restarting DOX diet 1 month after initial DOX
removal (yellow; n = 10 mice per group); ***P < 0.001, ****P < 0.0001.
(B) Representative sagittal lung H&E sections from moribund mice in each group
in (A). (C) tdTom+ burden in the airway after 1 month of EGFR inhibition through
DOX removal (pink; n = 9) or daily treatment with osimertinib (10 mg/kg;
orally days 1 to 5 of 7; white; n = 9) as compared with 8 weeks time point before
DOX removal (red; n = 12). DOX removal or osimertinib treatment (on DOX)
were initiated at 8 weeks after the development of extensive ERPMT-derived
LUAD tumorigenesis. For clarity, comparisons are effectively an 8-week pre-
treatment cohort to ~12-week posttreatment cohorts (1 month of treatment);
***P < 0.001, ****P < 0.0001. (D) Outline of single-cell samples sequenced at
Gardner et al., Science 383, eadj1415 (2024) 9 February 2024
C
distinct time points along the transition between LUAD and SCLC after DOX
perturbations [as described in (A)]. (E) Force-directed layout of cell states captured
along the transition from an AT2 cell to ERPMT LUAD and lastly toward a
neuroendocrine fate as compared with de novo SCLC tumorigenesis (transparent
blue); colored by sample [annotated in (D); n = 15,828 cells pooled from
three mice per sample]. FA, Force Atlas. (F) Principal components analysis
(PCA) projection of the lung epithelial lineage probability space (see materials
and methods) for all tumor-epithelial cells. Individual cells are colored by their
max lineage probability, and archetypes are overlaid as colored nodes (fig. S4,
C to E). (G) Flow plot showing relative abundance of cells assigned to their
nearest lineage archetype, ordered by sampling time. (H) Heatmap of scaled
imputed expression of highly variable transition genes (HVGs) (see materials
and methods), top ranked macrostate genes, and lineage markers along terminal
3 of 11
RES EARCH | R E S E A R C H A R T I C L E
cell-state probabilities computed with CellRank for all cells in (E). For each
gene, expression was smoothed with a generalized additive model (GAM) as
described in CellRank. The top HVGs correlated with the bottleneck macrostate
(table S2) and select lineage-specific markers are labeled on the right. Above
Cells expressing both Ascl1 and EGFR were
extremely rare, suggesting that a dual-positive
state is either short-lived or unviable. Further-
more, independent of Ascl1 expression, residual
tumor cells retained expression of AT2 lineage
markers and lacked high expression of a pro-
liferative program characteristic of de novo
SCLC (fig. S3C). These data provided the ratio-
nale to explore HT in our ERPMT model by
extinguishing oncogenic EGFR transcription-
ally through the withdrawal of DOX.
An undifferentiated, stem-like state emerges
during HT
To generate conditions likely to favor HT, we
removed DOX from ERPMT mice with late-
stage LUAD and transcriptionally profiled single
tdTom+ cells from pools of mice before, during,
and after EGFR removal (Fig. 2D and materials
and methods). We observed a continuous cell-
state transition from a normal AT2 cell, through
EGFR-driven LUAD, and finally toward trans-
formed SCLC at later time points. A minority of
tumor cells isolated from samples that were off
DOX formed a bottleneck adjacent to the
de novo LUAD model before breaking out
toward transformed SCLC (Fig. 2E). Five extreme
phenotypic states (“archetypes”; see materials
and methods) (33) were identified that corre-
sponded to mature lung epithelial lineages,
including AT1, AT2, PNEC, secretory, and basal
cell types; however, one archetype exhibited
features of a highly undifferentiated state
(Fig. 2F and fig. S4, A to C) and conspicuously
mapped to the bottleneck between de novo
LUAD and transformed SCLC (Fig. 2, F and G,
and fig. S4, B to D). This undifferentiated ar-
chetype was not associated with any lung
epithelial lineage but instead expressed rela-
tively modest levels of basal–stem cell programs,
as well as Myc and Sox2 target genes (fig. S4C).
It retained features of the original AT2 lineage
and had not yet begun to express neuroendocrine
markers, including Ascl1 (fig. S4E). Tumor cells
belonging to this undifferentiated state were
a minority in primary LUAD, expanded in the
residual LUAD (1 month off DOX), and dimin-
ished in the transforming (2 months off DOX)
and transformed (>3 months off DOX) popula-
tions, which were composed largely of neuro-
endocrine tumor cells (Fig. 2G).
To model the transition from an AT2 cell of
origin to an EGFR-driven LUAD, and lastly to
a transformed neuroendocrine state, we applied
CellRank (34) for single-cell fate mapping. Four
stable cellular phenotypes (“macrostates”)
were identified along this trajectory, capturing
Gardner et al., Science 383, eadj1415 (2024) 9 February 2024
the ranked heatmap are two KDEs showing relative sample abundance (lower)
and total cell frequency (upper, gray). Above the KDEs are scaled imputed gene
expression trends for the oncogenic drivers Myc and EGFR, modeled using a
GAM along the terminal probability.
normal AT2, de novo LUAD, the HT bottleneck,
and then a transformed SCLC population, re-
spectively (Fig. 2H, below heatmap). As the
LUAD oncogenic driver was removed, MAPK
pathway activity (35) expectedly decreased along
this trajectory; conversely, we observed a time-
dependent increase in Myc transcriptional output
(table S2). Along this trajectory, a highly specific
bottleneck to transformation emerged that
was both stem-like (Tm4sf1) (36, 37) and highly
proliferative (38) and exhibited features of neuro-
nal differentiation (Creb) (39–41) (Fig. 2H) and
Myc downstream signaling (fig. S4F). Tran-
scription factor regulatory modules (“regu-
lons”) that characterize this bottleneck were
likewise associated with neuronal plasticity
(Creb5), airway stemness (Sox9), and basal-
cell function (Trp63) (fig. S4G). Thus, on the
path of HT, as levels of EGFR transcript wane,
there may be selection for a cell state most fit
to be driven by high levels of Myc (Fig. 2H
and fig. S4F), and cells that break through this
bottleneck may be rapidly transformed by Myc
toward a neuroendocrine fate (Fig. 2H).
AT2 and PNEC lineage driver differences
To directly compare SCLC transformation effi-
ciencies between AT2 and PNEC cells, we in-
fected RPMT mice (which differ from ERPMT
mice by lack of the oncogenic EGFR transgene)
with equal titers of Ad5.Spc-Cre (for AT2) or
Ad5.Cgrp-Cre (for PNEC). Initially, the tdTom+
frequency was greater in the airway of mice
infected with Ad5.Spc-Cre, which is consistent
with a higher baseline frequency of AT2 cells
in the lung. However, this was short-lived and
was followed by exponential expansion of the
Ad5.Cgrp-Cre group (Fig. 3A). At 8 weeks after
infection, macroscopic disease was clearly visi-
ble in the PNEC-derived RPMT model but not
in the comparator (Fig. 3B).
To confirm whether our results with RPMT
mice reflect the differential tolerance of cell
lineages to the two oncogenic drivers, Myc and
EGFR, we performed experiments in which prod-
uction of a tamoxifen-activated Cre is governed
by cell-specific promoters active in neuroendocrine
(Ascl1CreERT2) or AT2 (SpcCreERT2) lineages, in the
absence of Cre-susceptible loci for Rb1 and Trp53.
After tamoxifen administration to activate CreERT2,
we found that Myc expanded the airway Ascl1+
population and that EGFR led to an eventual
decline. Conversely, EGFR expanded the AT2
lineage, whereas Myc was detrimental as com-
pared with wild-type (WT) controls (Fig. 3C).
In a larger cohort of mice, we observed that
Myc expression alone from the Ascl1+ line-
age was sufficient to produce a lethal, fully
penetrant phenotype, whereas EGFR was not.
By contrast, EGFR expression alone was suffi-
cient to transform the AT2 lineage, but Myc
was not (Fig. 3D). These data strongly support
cell lineage–specific differences in the tolerance
of oncogenic drivers Myc and EGFR in the lung.
The pulmonary neuroendocrine cell is refractory
to transformation by oncogenic EGFR
The phrase “terminal neuroendocrine” has been
used to describe the eventual histology that the
ERPMT model transitions toward following
EGFR withdrawal; however, it was unclear
whether this was a terminal state or if we could
study HT in reverse by converting a SCLC tu-
mor toward a LUAD state. As in earlier experi-
ments, we initiated tumorigenesis from PNECs
in the ERPMT model and followed cohorts of
mice that were on or off DOX. There was a
significant delay in the lethality of the model on
DOX (fig. S5A); however, in the lungs of mice
on DOX (where EGFR protein should be pro-
duced), we observed low-to-absent EGFR using
immunohistochemistry (IHC) in tdTom+ tumor
regions, which were histologically consistent
with SCLC (fig. S5B). Immunofluorescence for
EGFRL858R and Ascl1 showed patchy regions
of EGFR positivity that were excluded from
larger areas of Ascl1 positivity (fig. S5C), with the
Ascl1-positive/EGFR-negative SCLC component
of these tumors being the most common in
all mice examined. Taken together, these data
support the conclusions that cells in the PNEC
lineage resist transformation toward an EGFR-
driven LUAD state, just as cells in the AT2
lineage cannot be easily transformed to SCLC,
even though the latter form of HT can occasion-
ally occur under select conditions.
Oncogenic EGFR led to a gradual elimination
of PNECs over months, without evidence of acute
intoxication (Fig. 3C). It was unclear whether
elevated signaling through the MAPK pathway
(through EGFR) was a disfavored situation for
the PNEC or if EGFR incompatibility arose
through some other mechanism. If excessive
MAPK signaling suppressed PNEC prolifera-
tion, then inhibition of this pathway should
increase Ascl1+ cells. To test this, we traced
Ascl1+ cells and randomized mice to receive a
diet formulated with or without the MAPK
kinase inhibitor (MEKi) trametinib to suppress
Mek>Erk signaling (fig. S6A). We terminated
the study after 3 months on MEKi because of
toxicity, with adult mice experiencing weight
loss approaching our protocol limits (fig. S6B).
However, no significant differences were
4 of 11
RES EARCH | R E S E A R C H A R T I C L E

A B Ad5.Cgrp-Cre C D
Ascl1CreERT2 Ascl1CreERT2
RPMT

tdTomato % (viable)
AT2 Myc 100
10-1 EGFR

Survival (%)
**
WT 75
** 50
101 10-3
EGFR
PNEC 25
Myc
-5 0
10
tdTomato % (viable)

0
10 0 2 4 6 0 100 200 300
Weeks (post-TAM) Days (post-TAM)
** Ad5.Spc-Cre
-1
10 SpcCreERT2 SpcCreERT2

tdTomato % (viable)
102
100 Myc
EGFR

Survival (%)
-2 0 75
10 10 WT

****
50
10-2
-3 Myc 25
10 EGFR
10-4 0
ks

ks

ks

ks

0 2 4 6 0 100 200 300

ee

ee

ee

ee
w

Weeks (post-TAM) Days (post-TAM)

2

Fig. 3. Cell of origin and oncogenic driver incompatibility. (A) Frequency of
tdTom+ cells in the airway of RPMT mice after infection with equivalent titers
of adenovirus (~106 plaque-forming units per mouse) delivered using Ad5.Cgrp-Cre
(blue) or Ad5.Spc-Cre (red) over a period of 8 weeks; n = 4 mice per time point;
**P < 0.01. (B) Comparative histology of RPMT (no Tg.TetO-EGFRL858R transgene) mice
8 weeks after infection with Ad5.Cgrp-Cre (blue outline) or Ad5.Spc-Cre (red outline).
(C) Lineage-tracing oncogenic MycT58A (down-pointing triangles) or EGFRL858R
(circles) on AT2 (SpcCreERT2; red) or PNEC (Ascl1CreERT2; blue) cells in the airway over
time; n = 3 mice per time point. Control traces (tdTomato only; WT) are shown as gray
squares; **P < 0.01, ****P < 0.0001. (D) Long-term survival for cohorts shown in
(C); (top) Ascl1CreERT2 > EGFRL858R (n = 13) or MycT58A (n = 15) and (bottom) SpcCreERT2 >
EGFRL858R (n = 14) or MycT58A (n = 12). Mice having a single copy of Rosa26LSL-tdTom
and Rosa26LSL-rtTA3 were maintained on DOX chow throughout studies
investigating lineage-trace allele-mediated expression of Tg.TetO-EGFRL858R.

observed in the abundance of tdTom+ cells
between the two cohorts (fig. S6C). Moreover,
despite achieving chronic suppression of phospho-
Erk signaling (fig. S6D), we did not observe any
change in the location or proliferative status of
tdTom+ cells in mice treated with the MEKi diet
(fig. S6E), suggesting that physiologic Mek>Erk
signaling was not suppressing proliferation
of PNECs.
Myc is sufficient to transform the PNEC
Lineage tracing demonstrated a clear differ-
ence in the sensitivity of the AT2 and PNEC
cell types to oncogenic transformation by Myc
(Fig. 3); however, a hallmark of SCLC is in-
activation of the RB1 and TP53 tumor sup-
pressors, in addition to heightened expression
of a Myc family member and its transcription-
al targets (7). Inspecting the bronchioles of
mice expressing Myc driven by the Ascl1CreERT2
lineage trace (Ascl1>Myc) 1 month after label-
ing revealed clusters of proliferative tdTom+
cells that had not yet invaded surrounding
tissues, which is consistent with carcinoma
in situ (fig. S7A). Instead, all Ascl1>Myc mice
were dying of cancer localized to the thyroid
(fig. S7B). To date, we have been unsuccessful
in activating the Ascl1CreERT2 allele specifically
in the lungs, while sparing the trachea and
thyroid. We therefore isolated tdTom+ cells
from the airways of four distinct genotypes
of mice combining Ascl1 lineage–traced Myc
with loss of Rb1, Trp53, or both tumor sup-
pressor genes. We expanded cells ex vivo using
organotypic culture conditions and engrafted
equivalent cell numbers into the flanks of
athymic mice. Combined loss of Rb1 and
Trp53 accelerated the growth of these Myc-
driven tumors, but all genotypes were suffi-
cient to form transplantable cancers. Moreover,
all tumors had a similar histologic appearance
consistent with high-grade neuroendocrine can-
cer (fig. S7C). These data suggest that PNECs
could be transformed by Myc (alone) if ex-
panded ex vivo, but this experiment did not
demonstrate that Myc was required for tumor
maintenance.
To test whether PNEC-derived tumors are
dependent on Myc, we sorted tdTom+, rtTA3-
expressing PNECs from the airway and infected
cells with lentiviruses containing tetracycline
promoter–driven Myc constructs. Coupling rtTA3
expression to the lineage trace (Ascl1CreERT2)
removed the likelihood of infecting lineage-
negative cells present as contaminants. Although
these cells were sparse (fig. S8), we could con-
sistently generate organoid cultures from as
few as 50 cells when DOX was present in the
culture to drive the transcription of Myc. Re-
moval of DOX resulted in near complete growth
suppression of organoids expressing MycWT
as compared with MycT58A (fig. S9, A to C), a
long-lived version of Myc (42) (fig. S9C). En-
grafting PNECs expressing inducible MycWT into
athymic mice demonstrated DOX-dependent
tumor growth. Removing DOX from tumor-
bearing mice dramatically reduced tumor vol-
umes over the course of several weeks (fig. S9D),
with residual fibrotic tumor tissue compris-
ing noncycling cells (fig. S9E). Consistent with
their rapid proliferation, Myc-driven PNEC
organoids were sensitive to compounds that
exacerbated replication stress, such as topo-
isomerase inhibitors (etoposide), as well as
inhibitors of enzymes required for cell-cycle
progression, including Cdk4/6 (palbociclib) and
Wee1 (adavosertib) (fig. S9, F and G). Addition-
ally, direct inhibition of the Myc-Max interface
using the small molecule MYCi975 (43) pro-
vided similar reduction in organoid growth as
compared with reduction resulting from DOX
removal (fig. S9F).
AT2 cells are refractory to transformation
by Myc
Although Myc alone may be sufficient to drive
transformation and expansion of PNECs, our
earlier results suggested that it was insuffi-
cient to transform the AT2 lineage (Fig. 3C).
To further investigate what underlies this
bottleneck, we established AT2 organoid cul-
tures using recently published methods (44) from
lineage-traced, WT, or MycT58A-expressing
cells. These organoids expanded rapidly as
compared with their WT counterparts but were
unsustainable beyond three passages (fig. S10A).
In early time points after the expression of
MycT58A in AT2 cells in vivo, we noted incor-
poration of EdU in tdTom+ cells. However,

Gardner et al., Science 383, eadj1415 (2024) 9 February 2024 5 of 11

RES EARCH | R E S E A R C H A R T I C L E
1 year after the initiation of the trace, tdTom+
cells in the airway failed to incorporate EdU,
suggesting that they were no longer prolifer-
ative or were eliminated (fig. S10B). Ex vivo, AT2
organoids expressing MycT58A demonstrated
increased DNA damage sensing, replication
stress, and markers of programmed cell death
as compared with WT (fig. S10C). It is unlikely
that this resulted from an excess of Myc pro-
tein, because levels were lower in AT2 cells
compared with levels tolerated in PNEC organ-
oids (45) (fig. S10D). Lastly, consistent with
the observation that Ras signaling through
phosphatidylinositol-3-kinase (PI3K) relieves
Myc-induced apoptosis (46), we likewise observed
that Myc significantly accelerated oncogenic
EGFR-driven LUAD, implying that enhanced
signaling via the EGFR>Ras>Mek>Erk path-
way can relieve intolerance to Myc in the AT2
lineage (fig. S10, E and F).
Although our data suggest that the different
oncogenes driving AT2 and PNEC lineages are
in stark contrast, it remained unclear whether
other epithelial airway cells can transform in
response to Myc alone. To address this, we
performed a generalized, conditional trace
using an Nkx2.1CreERT2 allele that will drive Cre-
mediated recombination in a broad range of
cell types derived from the anterior foregut
endoderm, including the trachea, pituitary,
thyroid, and most of the lung (47). At early
time points, lineage-labeled tissues within the
thyroid (both Ascl1+ and Ascl1–) expanded
after Myc expression as compared with tissues
in control mice (fig. S11A); however, at later time
points there was outgrowth of tdTom+/Ascl1+
cells in bronchioles not observed in WT controls
(fig. S11B). Together, these data suggest that the
Ascl1+ PNEC is distinct in its tolerance to Myc,
but we had not yet explained how intolerance
to Myc could be overcome in the AT2 lineage
during HT.
Deletion of Pten removes a barrier to
Myc transformation
Genes up-regulated in tumor cells as they
escaped the bottleneck to HT in our ERPMT
model (termed “breakout”; fig. S12A) were
notably associated with PI3K signaling (fig.
S12, B and C) as compared with cells found
in the bottleneck and not yet adopting neuro-
endocrine fate (48). Thus, we asked whe-
ther increasing PI3K-dependent Akt signaling
through deletion of Pten would enable
MycT58A-driven transformation in an AT2 line-
age trace. Notably, we observed fully penetrant
Myc-driven transformation in an AT2 cell (using
SpcCreERT2) when one copy of Pten was inacti-
vated (Spc>Pten;Myc; Fig. 4A). At the median
period when animals in the Spc>Pten;Myc cohort
were moribund, lungs from Spc>Myc mice
(PtenWT/WT) showed no evidence of macroscopic
disease (Fig. 4B). Similar observations were
made when combining Myc expression with
Gardner et al., Science 383, eadj1415 (2024) 9 February 2024
a conditionally active, mutant PI3K allele
(E545K) or inactivating both copies of Pten
(PtenFl/Fl; Fig. 4B). Examining Spc>Pten;Myc
animals 3 months after recombination reinforced
the observation that lesions were variable in
their frequency, size, and histologic appearance
(fig. S13). However, they were Ascl1-negative
and glandular in appearance, suggesting that
they were adenocarcinoma-like and excluding
the likelihood of SCLC, squamous, or mixed his-
tology (fig. S13). AT2 lineage markers, including
prosurfactant protein C (pro-SPC) and MHC
Class II, were low or absent, and basal epithe-
lial markers such as Sox2 and keratin 18 were
variably expressed (fig. S13). Using scRNA-seq,
we also observed a notable increase in the highly
undifferentiated, basal stem-like state that fol-
lowed combined loss of Pten and expression of
Myc in AT2 cells—not seen with either genetic
perturbation alone (Fig. 4C). Expectedly, this
enrichment in the undifferentiated state was
also associated with greater Myc transcriptional
output (Fig. 4D).
Pulmonary basal cells efficiently generate SCLC
The basal stem-like program associated with
AT2 cells capable of adaptation to Myc sup-
ported the possibility that an intermediate state
during HT may be basal-like. More generally,
this raised the possibility that the basal cell
may serve as a cell of origin for SCLC, as
speculated by others (49). However, targeting
basal cells in mouse models is limited by their
anatomic location, noted to be refractory to viral
infections delivering Cre (50). Indeed, we ob-
served an absence of tdTom+ cells in the lungs
of mice 1 month after labeling when using a
Krt5CreERT2 allele to target basal cells (fig. S14A).
However, following regeneration in the proxi-
mal lung induced by naphthalene damage of
secretory cells (51, 52), tdTom+ cells were de-
tected within the lungs of mice, without a
significant change in the trachea or thymus (fig.
S14B). As basal cells are known to serve as
multipotent progenitors (fig. S14C) (51–53), we
then crossed mice to delete Rb1 and/or Trp53
from the basal lineage trace and found that Rb1
loss alone was sufficient to skew cells toward
a neuroendocrine fate (fig. S14D).
Consistent with this, we observed fully pene-
trant neuroendocrine tumorigenesis in mice that
have lost both Rb1 and Trp53 with (Krt5>RPMT)
or without (Krt5>RPT) expression of Myc from
the basal lineage (fig. S14E); however, tumors
arose with significantly shorter latency in mice
expressing Myc. Expression of the conditional
tdTomato reporter allele was easily observed
throughout the keratinized epithelium of mice
after recombination, but tumorigenesis was re-
stricted to the proximal airway (fig. S14, F
and H). Histologically, both models produced
SCLC-like tumors; however, we noted that most
tumors were adjacent to or within the thymus
and did not invade the lungs unless we damaged
lungs with naphthalene (fig. S14, G to I). These
data suggest that basal cells may serve not only
as cells of origin for SCLC but also that basal-cell
progeny and anatomic location can be influ-
enced by genotype and injury.
An efficient model of AT2-derived SCLC after
the loss of Pten
Deletion of Pten was sufficient to lower the
barrier to transformation by Myc in the AT2
lineage; however, the resulting tumors were
not neuroendocrine (fig. S13). We suspected
that Rb1 loss was required (10) —in addition to
adaptation to Myc—for neuroendocrine trans-
formation. Thus, to recapitulate bona fide trans-
formation from an AT2 cell to SCLC efficiently,
we generated a model in which we could delete
Rb1, Trp53, and a copy of Pten, and express Myc
and tdTom (RPPtenMT). If tumorigenesis was
initiated in PNECs, there was no difference in
latency between the RPMT and RPPtenMT
models, but a significant difference emerged
if the model was initiated in AT2 cells (Fig. 4E).
AT2-derived RPPtenMT tumor cells displayed
neuroendocrine expression profiles most sim-
ilar to de novo SCLC (Fig. 4F and fig. S15, A and
B). We also noted two subpopulations of cells
stratified by expression of Ascl1 (fig. S15B).
Analysis of differential gene expression in the
Ascl1low group was notable for expression of
neuronal genes, including Stnm2, Nfix, and
Mapt. The Ascl1low group also expressed high
levels of NeuroD1, which is consistent with prior
work showing Myc driving subtype plasticity
from an Ascl1high to Ascl1low/NeuroD1high tran-
scriptional profile in multiple models of SCLC
(54–58) (fig. S15C). Pathology revealed exten-
sive heterogeneity with mixing of classic SCLC
and large-cell neuroendocrine (LCNEC) tumor
features, marked by variable expression of neu-
ronal and neuroendocrine markers (fig. S15D).
Rb1 loss is necessary but insufficient for fully
penetrant LUAD to SCLC transformation
The ERPMT model provided an efficient system
to study HT, but we had not addressed a core
requirement for Myc in this process. We gen-
erated another model in the absence of Myc
overexpression (ERPT) and found that although
some animals had recurrent disease after the
removal of DOX (ERPT on > off), the pene-
trance of the phenotype was incomplete (Fig.
5A). Moreover, whereas the de novo ERPT
LUAD showed an absence of Ascl1 and Myc
protein, only some recurrent tumors appeared
to be histologically consistent with SCLC, fur-
ther limiting the utility of the ERPT model in
recapitulating HT (Fig. 5, B to D). Although
recurrence was incomplete, the ERPT tumors
recurring as SCLC did show an increase in
Myc protein (Fig. 5C). We analyzed the tran-
scriptomes of ERPT and ERPMT LUAD tumor
cells compared with those of normal AT2 cells
and found that both the ERPT and ERPMT
6 of 11
RES EARCH | R E S E A R C H A R T I C L E

A B

C D

E F

Fig. 4. Loss of Pten in the AT2 lineage removes the barrier to Myc
transformation. (A) Survival of mice where Myc (n = 12), PtenFl/WT (n = 5), or
the combination of these alleles (n = 23) are initiated in AT2 cells by using a
single copy of SpcCreERT2. Data were censored between 250 and 325 days in the
nonlethal arms. (B) Comparative histology of whole lungs from representative
Spc>Myc, Spc>PtenFl/+;Myc, Spc>PtenFl/Fl;Myc, or Spc>PI3KLSL-E545K;Myc mice
at ~3 months after labeling. Higher-magnification regions (boxed) are provided at
right; scalebars, 200 mm. (C) Bar plot showing fraction of each epithelial lineage
archetype detected per sample as in Fig. 2G. (D) Bar plot of hallmark gene sets
significantly enriched [false discovery rate (FDR) < 0.01] within the undifferentiated
cell state of the Spc>PtenFl/+;Myc sample pool. (E) Effect of Pten deletion combined
with deletion of p53 and Rb1 and expression of Myc and tdTomato (RPPtenMT or
RPMT) in neuroendocrine (blue) or AT2 (red) cells; n = 10 per arm with x axis split
for clarity, ****P < 0.0001. (F) Clustered heatmap of z-normalized imputed
expression of AT2 and PNEC signature genes and model oncogenic drivers (Myc
and Tg.EGFR) for all tumor-epithelial cells from the RPPtenMT model (green) and the
de novo LUAD (red) and SCLC (blue) models described in Fig. 1. Genes and cells
are clustered by using the average Euclidean distance method.

models transcriptionally maintained expres- model, further supporting the role Myc over- studies have compared unrelated de novo
sion of AT2 genes. However, Myc and em- expression has in HT after EGFR withdrawal LUAD with transformed SCLC (T-SCLC) in an
bryonic stem cell target genes were elevated (Fig. 5E). True paired cases of human LUAD attempt to understand this phenomenon as
in the ERPT model and increased in the ERPMT before and after HT are limited, but some it relates to de novo SCLC tumorigenesis

Gardner et al., Science 383, eadj1415 (2024) 9 February 2024 7 of 11

RES EARCH | R E S E A R C H A R T I C L E
A B
E F
G H
Fig. 5. Rb1 loss cooperates with Myc expression to facilitate HT.
(A) Survival of mice with the ERPT genotype (no MycT58A transgene) initiated
with Ad5.Spc-Cre on DOX (ERPT on; dark green, n = 23), or on and then off DOX
once mice developed advanced disease (ERPT on > off; light green, n = 20);
***P < 0.001. As before, DOX diet was removed when individual mice exhibited
signs of labored breathing and/or substantial weight loss with a hunched
appearance. For each model, cohorts of mice were followed until approximately
three times the median latency elapsed, at which point lungs were collected
and the study was ended. (B) Histologic appearance of “ERPT on” lungs
performed with H&E and IHC staining for Ascl1 and Myc (scalebar, 100 mm).
Gardner et al., Science 383, eadj1415 (2024) 9 February 2024
C D
I
(C) Similar to (B), now with mice on and then off DOX, representative of a SCLC-
like tumor. (D) As shown in (B), now with mice on and then off DOX,
representative of a LUAD-like tumor. Pathologic interpretation is provided below
each representative example. (E) (Left) Dot plot showing frequency of
expressing cells (node size) and log-transformed expression (node color) of
AT2 marker genes in normal AT2, ERPT LUAD, and ERPMT LUAD models. Genes
shown are expressed in at least 25% of cells within at least one condition. (Right)
KDE plots showing mean log-transformed expression by condition for select
gene signatures. (F) Volcano plot showing differentially expressed genes from
bulk RNAseq of human transformed SCLC versus LUAD. Genes from the
8 of 11
RES EARCH | R E S E A R C H A R T I C L E

Hallmark MYC Targets V1 signature are colored according to conditional
enrichment, top genes [abs(log2FC) > 1 and FDR < 1 × 10–5, where “abs” is
absolute and FC is “fold change”] are labeled, and the pathway normalized
enrichment score (NES) and FDR from the gene set enrichment analysis (GSEA)
are inset (bottom right). (G) Tumorigenesis initiated with an adenoviral (n =
10; Ad5.Spc-Cre) or AT2 lineage-trace allele (n = 19; SpcCreERT2) in the EPMT
model (Rb1WT/WT) produces LUAD that does not relapse following DOX removal
(n = 5 per group), ***P < 0.001. (H) Representative sagittal lung H&E sections
from each group in (G) on DOX at point of moribund disease or 1 month after the
removal of DOX from an otherwise moribund animal. (I) Clustered heatmap of gene
signatures differentially enriched between the tumor-epithelial cells of the EPMT
and ERPMT models before and after DOX removal. All pathways are significantly
enriched (NES > 0 and FDR < 1 × 10–5) in at least one condition. Less significant
signatures (FDR < 0.01) are transparent, and signatures not meeting this threshold
are blank. Rows and columns are clustered by using the complete Manhattan
distance method and metric. (J) Bar plot showing fraction of each epithelial lineage
archetype detected per sample (as shown in Fig. 2G) for the EPMT (left) and ERPMT
(right) models before and after removal of DOX.

(1, 3, 59, 60). We reanalyzed a publicly avail-
able cohort and could demonstrate that Myc
target genes were differentially increased in
T-SCLC as compared with LUAD, suggesting
that Myc may have a role in facilitating HT
in a clinical setting (Fig. 5F).
Rb1 loss cooperates with expression of Myc to
facilitate neuroendocrine transformation
To compare the relative dependence of Rb1 loss
and Myc overexpression for HT, we generated a
model in which Rb1 was WT (EPMT) and
initiated tumorigenesis in AT2 cells using Ad5.
Spc-Cre or a lineage trace (SpcCreERT2>EPMT).
If on DOX, then both models produced LUAD
with differences in latency likely reflecting the
broad initiation achievable when using a lineage-
trace allele as compared with sparse infection
using inhaled adenovirus. (Fig. 5, G and H).
LUAD did not recur in EPMT mice taken off
DOX. Through a comparison of de novo LUAD
and off-DOX residual cells from tumors of
varying Rb1 status, we observed marked up-
regulation of transcriptional programs associated
with HT after DOX removal that were dependent
on loss of Rb1 (Fig. 5I). Additionally, the
relative abundance of lineage archetypes shifts
within these models in response to DOX re-
moval suggests that, in the presence of Rb1 loss,
high Myc expression selects for a residual state
that is stem-like and capable of full neuro-
endocrine transformation (Fig. 5J). Thus,
these events cooperate, as neither Rb1 loss
nor Myc overexpression alone are sufficient
for fully penetrant HT.
We conclude that the AT2 cell is highly re-
fractory to transformation by oncogenic Myc,
as are many other cell types within the lung,
the noted exception being the PNEC. Intolerance
of AT2 cells to Myc can be relieved through
activation of the Akt signaling pathway, such as
through the deletion of the tumor suppressor
Pten. Although this generates a permissive stem-
like state, the full conversion to a Myc-driven,
high-grade neuroendocrine cancer requires the
additional loss of Rb1.
Discussion
Carcinogenesis and related processes, such as
tumor progression, therapy resistance, and HT,
remain incompletely understood. To understand
the mechanism by which HT occurs in lung
cancer, we have developed mouse models in
which lung tumorigenesis can be initiated in
different cell lineages to follow the transfor-
mation of EGFR-driven LUAD to Myc-driven
SCLC. In doing so, we demonstrate that HT
can be simplified conceptually to a change in
oncogenic drivers as the tumor cells transition
between states that resemble cells in the AT2
and the PNEC lineages. The driver oncogene
in the PNEC lineage is Myc, and the bottleneck
in HT can be relieved by mechanisms that allow
an AT2 cell to become stem-like and capable
of tolerating Myc as an oncogenic driver.
Understanding the events that facilitate HT
clinically has been limited by the lack of sam-
ples obtained from the same patient before,
during, and after HT. Comparisons of de novo
SCLC to transformed SCLC have highlighted
pathways activated after neuroendocrine dif-
ferentiation (59). A more recent study relied
on putative HT samples in which a LUAD
oncogenic driver was detected in a histologically
confirmed SCLC. When compared to de novo
SCLC, these cases were enriched for mutations
that activate the PI3K signaling pathway (61).
Consistent with this result, we found that loss
of Pten or activation of Pik3ca was sufficient
to break a barrier in the AT2 lineage to trans-
formation by Myc but insufficient to lead to
neuroendocrine transformation.
Previous work has established a combina-
tion of genetic events capable of reprogramming
human cell types to neuroendocrine cancers
(62); however, the intermediate steps in this
process remain unclear. We have built upon
this foundational work by developing intact
genetically engineered animals and sampling
tumor cells throughout the lifetime of HT.
Moreover, our findings show that although Rb1
loss is necessary for HT to a neuroendocrine
cancer, it is the extinction of the driver onco-
gene transcriptionally that is required before
the emergence of neuroendocrine features.
Beyond the genetic manipulations and changes
in gene expression described here, we speculate
that there are likely other mechanisms that
allow for such lineage conversion, cellular plas-
ticity, or transdifferentiation. Lineage tracing
has demonstrated that basal progenitors can
give rise to PNECs (63, 64). More recently,
tracheal basal cells have been shown to dif-
ferentiate toward a PNEC fate under hypoxic
conditions, but the intermediate transcriptional
steps in this complex process are uncharted (65).
In our attempts to describe the transition state
between LUAD and SCLC, we found that these
tumor cells appear “basal-like” on the basis of
their transcriptional profile but lack definitive
basal lineage markers. Instead, these cells are
more accurately described as “lineage negative,”
stem-like progenitors. Such cells have been
described to arise in the mouse lung following
major airway injury (66). Thus, the airway cell
most capable of such plasticity may be the pul-
monary basal cell or a stem-like cell that has yet
to be fully characterized.
Lastly, it is unclear whether therapeutic tar-
geting of pathways facilitating HT in patients
poised to undergo HT will be efficacious. Our
results suggest that noninvasive monitoring
for activation of the Akt signaling pathway
(such as the appearance of PI3K mutations
in circulating tumor DNA) may serve to alert
physicians to the likelihood of HT before the
emergence of an aggressive, transformed SCLC.
It is not yet known whether direct targeting
of Myc in such Myc-driven scenarios (including
HT or de novo SCLC tumorigenesis) will be
fruitful; however, strategies to inhibit the
transcriptional activity of Myc proteins have
advanced substantially over the last decade and
might ultimately have clinical utility in multi-
ple contexts (43, 67, 68).
Limitations of the study
There are several key limitations to this study.
First, we are attempting to understand HT by
modeling a human phenomenon in the mouse,
where the complexity of cell types and micro-
environments is not identical to that found in
humans. Recent descriptions of the human dis-
tal airway have found a diversity of cell types
in human lungs that are not observed in the
mouse, including specialized regenerative cells
with hybrid alveolar-secretory character and
multiple subtypes of basal cells (19, 69, 70). For
example, basal cells are not present in the mouse
lung unless the airway is damaged, which we
demonstrated using naphthalene. Furthermore,
most laboratory mice are housed under pathogen-
free conditions. We know that lifetime exposure
to various carcinogens and particulates funda-
mentally alters the microenvironment of the
lung, including the likelihood that a cancer may
develop (71, 72). Second, there are processes
in human cells not found in the mouse that
may be critical for HT, including the role for

Gardner et al., Science 383, eadj1415 (2024) 9 February 2024 9 of 11

RES EARCH | R E S E A R C H A R T I C L E
APOBEC-mediated hypermutation. Studies
of EGFR-mutated LUAD have found APOBEC
mutagenesis signatures after treatment with
EGFR-targeted therapies (11, 73, 74), and this
mutational signature is enriched after HT (1, 61).
It is unclear whether APOBEC mutations are
directly responsible for activating the PI3K
signaling pathway, (hypothetically) relieving
intolerance of the AT2 cell to Myc (1, 59, 61);
however, other studies would suggest that this
is possible (75, 76). Lastly, we were unable to
initiate tumorigenesis in the ERPMT model
with the SpcCreERT2 allele. This proved impossible
because in the mouse genome, the Rb1 locus is
~3 Mb from the Sftpc locus (GRCm38/mm10).
Such proximity precludes generation of the
desired genotype with both homozygous floxed
copies of Rb1 and the SpcCreERT2 lineage-trace
allele.
RE FE RENCES AND N OT ES
1. J. K. Lee et al., Clonal History and Genetic Predictors of
Transformation Into Small-Cell Carcinomas From Lung
Adenocarcinomas. J. Clin. Oncol. 35, 3065–3074 (2017).
doi: 10.1200/JCO.2016.71.9096; pmid: 28498782
2. M. F. Zakowski, M. Ladanyi, M. G. Kris; Memorial
Sloan-Kettering Cancer Center Lung Cancer OncoGenome
Group, EGFR mutations in small-cell lung cancers
in patients who have never smoked. N. Engl. J. Med. 355,
213–215 (2006). doi: 10.1056/NEJMc053610;
pmid: 16837691
3. L. V. Sequist et al., Genotypic and histological evolution of lung
cancers acquiring resistance to EGFR inhibitors. Sci. Transl.
Med. 3, 75ra26 (2011). doi: 10.1126/scitranslmed.3002003;
pmid: 21430269
4. H. Beltran et al., Divergent clonal evolution of castration-
resistant neuroendocrine prostate cancer. Nat. Med. 22,
298–305 (2016). doi: 10.1038/nm.4045; pmid: 26855148
5. D. Hirano, Y. Okada, S. Minei, Y. Takimoto, N. Nemoto,
Neuroendocrine differentiation in hormone refractory prostate
cancer following androgen deprivation therapy. Eur. Urol. 45,
586–592, discussion 592 (2004). doi: 10.1016/
j.eururo.2003.11.032; pmid: 15082200
6. Á. Quintanal-Villalonga et al., Lineage plasticity in cancer: A
shared pathway of therapeutic resistance. Nat. Rev. Clin. Oncol.
17, 360–371 (2020). doi: 10.1038/s41571-020-0340-z;
pmid: 32152485
7. J. George et al., Comprehensive genomic profiles of small cell
lung cancer. Nature 524, 47–53 (2015). doi: 10.1038/
nature14664; pmid: 26168399
8. C. M. Rudin et al., Molecular subtypes of small cell lung cancer:
A synthesis of human and mouse model data. Nat. Rev. Cancer
19, 289–297 (2019). doi: 10.1038/s41568-019-0133-9;
pmid: 30926931
9. M. G. Oser, M. J. Niederst, L. V. Sequist, J. A. Engelman,
Transformation from non-small-cell lung cancer to small-cell
lung cancer: Molecular drivers and cells of origin. Lancet Oncol.
16, e165–e172 (2015). doi: 10.1016/S1470-2045(14)71180-5;
pmid: 25846096
10. M. J. Niederst et al., RB loss in resistant EGFR mutant lung
adenocarcinomas that transform to small-cell lung cancer.
Nat. Commun. 6, 6377 (2015). doi: 10.1038/ncomms7377;
pmid: 25758528
11. M. Offin et al., Concurrent RB1 and TP53 Alterations Define
a Subset of EGFR-Mutant Lung Cancers at risk for Histologic
Transformation and Inferior Clinical Outcomes. J. Thorac.
Oncol. 14, 1784–1793 (2019). doi: 10.1016/j.jtho.2019.06.002;
pmid: 31228622
12. T. J. Desai, D. G. Brownfield, M. A. Krasnow, Alveolar progenitor
and stem cells in lung development, renewal and cancer.
Nature 507, 190–194 (2014). doi: 10.1038/nature12930;
pmid: 24499815
13. D. B. Frank et al., Early lineage specification defines alveolar
epithelial ontogeny in the murine lung. Proc. Natl. Acad. Sci.
U.S.A. 116, 4362–4371 (2019). doi: 10.1073/pnas.1813952116;
pmid: 30782824
Gardner et al., Science 383, eadj1415 (2024) 9 February 2024
14. A. N. Nabhan, D. G. Brownfield, P. B. Harbury, M. A. Krasnow,
T. J. Desai, Single-cell Wnt signaling niches maintain stemness
of alveolar type 2 cells. Science 359, 1118–1123 (2018).
doi: 10.1126/science.aam6603; pmid: 29420258
15. Cancer Genome Atlas Research Network, Comprehensive
molecular profiling of lung adenocarcinoma. Nature 511,
543–550 (2014). doi: 10.1038/nature13385; pmid: 25079552
16. Y. Ouadah et al., Rare Pulmonary Neuroendocrine Cells Are
Stem Cells Regulated by Rb, p53, and Notch. Cell 179,
403–416.e23 (2019). doi: 10.1016/j.cell.2019.09.010;
pmid: 31585080
17. C. S. Kuo, M. A. Krasnow, Formation of a Neurosensory Organ
by Epithelial Cell Slithering. Cell 163, 394–405 (2015).
doi: 10.1016/j.cell.2015.09.021; pmid: 26435104
18. P. Sui et al., Pulmonary neuroendocrine cells amplify allergic
asthma responses. Science 360, eaan8546 (2018).
doi: 10.1126/science.aan8546; pmid: 29599193
19. K. J. Travaglini et al., A molecular cell atlas of the human lung
from single-cell RNA sequencing. Nature 587, 619–625 (2020).
doi: 10.1038/s41586-020-2922-4; pmid: 33208946
20. K. Politi et al., Lung adenocarcinomas induced in mice by
mutant EGF receptors found in human lung cancers respond to
a tyrosine kinase inhibitor or to down-regulation of the
receptors. Genes Dev. 20, 1496–1510 (2006). doi: 10.1101/
gad.1417406; pmid: 16705038
21. K. D. Sutherland et al., Cell of origin of small cell lung cancer:
Inactivation of Trp53 and Rb1 in distinct cell types of adult
mouse lung. Cancer Cell 19, 754–764 (2011). doi: 10.1016/
j.ccr.2011.04.019; pmid: 21665149
22. C. E. Barkauskas et al., Type 2 alveolar cells are stem
cells in adult lung. J. Clin. Invest. 123, 3025–3036 (2013).
doi: 10.1172/JCI68782; pmid: 23921127
23. A. M. Laughney et al., Regenerative lineages and immune-
mediated pruning in lung cancer metastasis. Nat. Med. 26,
259–269 (2020). doi: 10.1038/s41591-019-0750-6;
pmid: 32042191
24. Y. He et al., MHC class II expression in lung cancer. Lung
Cancer 112, 75–80 (2017). doi: 10.1016/j.lungcan.2017.07.030;
pmid: 29191604
25. T. Yazawa et al., Lack of class II transactivator causes severe
deficiency of HLA-DR expression in small cell lung cancer.
J. Pathol. 187, 191–199 (1999). doi: 10.1002/(SICI)1096-9896
(199901)187:2<191::AID-PATH206>3.0.CO;2-3;
pmid: 10365094
26. L. M. Marsh et al., Surface expression of CD74 by type II
alveolar epithelial cells: A potential mechanism for macrophage
migration inhibitory factor-induced epithelial repair. Am. J.
Physiol. Lung Cell. Mol. Physiol. 296, L442–L452 (2009).
doi: 10.1152/ajplung.00525.2007; pmid: 19136583
27. S. Bae, Y. Bessho, M. Hojo, R. Kageyama, The bHLH gene
Hes6, an inhibitor of Hes1, promotes neuronal differentiation.
Development 127, 2933–2943 (2000). doi: 10.1242/
dev.127.13.2933; pmid: 10851137
28. J. D. Buenrostro, B. Wu, H. Y. Chang, W. J. Greenleaf,
ATAC-seq: A Method for Assaying Chromatin Accessibility
Genome-Wide. Curr. Protoc. Mol. Biol.109, 21.29.1–21.9.9
(2015). doi: 10.1002/0471142727.mb2129s109;
pmid: 25559105
29. P. A. Ewels et al., The nf-core framework for community-
curated bioinformatics pipelines. Nat. Biotechnol. 38, 276–278
(2020). doi: 10.1038/s41587-020-0439-x; pmid: 32055031
30. S. Heinz et al., Simple combinations of lineage-determining
transcription factors prime cis-regulatory elements required
for macrophage and B cell identities. Mol. Cell 38, 576–589
(2010). doi: 10.1016/j.molcel.2010.05.004; pmid: 20513432
31. Y. Oren et al., Cycling cancer persister cells arise from lineages
with distinct programs. Nature 596, 576–582 (2021).
doi: 10.1038/s41586-021-03796-6; pmid: 34381210
32. M. D. Borromeo et al., ASCL1 and NEUROD1 Reveal
Heterogeneity in Pulmonary Neuroendocrine Tumors and
Regulate Distinct Genetic Programs. Cell Rep. 16, 1259–1272
(2016). doi: 10.1016/j.celrep.2016.06.081; pmid: 27452466
33. A. Cutler, L. Breiman, Archetypal Analysis. Technometrics 36,
338–347 (1994). doi: 10.1080/00401706.1994.10485840
34. M. Lange et al., CellRank for directed single-cell fate mapping.
Nat. Methods 19, 159–170 (2022). doi: 10.1038/s41592-021-
01346-6; pmid: 35027767
35. M. C. Wagle et al., A transcriptional MAPK Pathway Activity
Score (MPAS) is a clinically relevant biomarker in multiple
cancer types. NPJ Precis. Oncol. 2, 7 (2018). doi: 10.1038/
s41698-018-0051-4; pmid: 29872725
36. G. Chen et al., Targeting TM4SF1 exhibits therapeutic potential
via inhibition of cancer stem cells. Signal Transduct. Target.
Ther. 7, 350 (2022). doi: 10.1038/s41392-022-01177-7;
pmid: 36229443
37. X. Ouyang et al., WDR72 Enhances the Stemness of Lung
Cancer Cells by Activating the AKT/HIF-1a Signaling Pathway.
J. Oncol. 2022, 5059588 (2022). doi: 10.1155/2022/5059588;
pmid: 36385964
38. M. S. Kowalczyk et al., Single-cell RNA-seq reveals changes in
cell cycle and differentiation programs upon aging of
hematopoietic stem cells. Genome Res. 25, 1860–1872 (2015).
doi: 10.1101/gr.192237.115; pmid: 26430063
39. N. Tsuboyama et al., Therapeutic targeting of BAP1/ASXL3
sub-complex in ASCL1-dependent small cell lung cancer.
Oncogene 41, 2152–2162 (2022). doi: 10.1038/s41388-022-
02240-x; pmid: 35194152
40. Z. Zhao et al., PAX9 Determines Epigenetic State Transition
and Cell Fate in Cancer. Cancer Res. 81, 4696–4708 (2021).
doi: 10.1158/0008-5472.CAN-21-1114; pmid: 34341073
41. Y. Xia et al., Targeting CREB Pathway Suppresses Small Cell
Lung Cancer. Mol. Cancer Res. 16, 825–832 (2018).
doi: 10.1158/1541-7786.MCR-17-0576; pmid: 29523765
42. M. A. Gregory, Y. Qi, S. R. Hann, Phosphorylation by glycogen
synthase kinase-3 controls c-myc proteolysis and subnuclear
localization. J. Biol. Chem. 278, 51606–51612 (2003).
doi: 10.1074/jbc.M310722200; pmid: 14563837
43. H. Han et al., Small-Molecule MYC Inhibitors Suppress Tumor
Growth and Enhance Immunotherapy. Cancer Cell 36,
483–497.e15 (2019). doi: 10.1016/j.ccell.2019.10.001;
pmid: 31679823
44. S. Naranjo et al., Modeling diverse genetic subtypes of lung
adenocarcinoma with a next-generation alveolar type 2
organoid platform. Genes Dev. 36, 936–949 (2022).
doi: 10.1101/gad.349659.122; pmid: 36175034
45. D. J. Murphy et al., Distinct thresholds govern Myc’s biological
output in vivo. Cancer Cell 14, 447–457 (2008). doi: 10.1016/
j.ccr.2008.10.018; pmid: 19061836
46. A. Kauffmann-Zeh et al., Suppression of c-Myc-induced
apoptosis by Ras signalling through PI(3)K and PKB. Nature
385, 544–548 (1997). doi: 10.1038/385544a0; pmid: 9020362
47. H. Taniguchi et al., A resource of Cre driver lines for genetic
targeting of GABAergic neurons in cerebral cortex. Neuron 71,
995–1013 (2011). doi: 10.1016/j.neuron.2011.07.026;
pmid: 21943598
48. E. Y. Chen et al., Enrichr: Interactive and collaborative HTML5
gene list enrichment analysis tool. BMC Bioinformatics 14, 128
(2013). doi: 10.1186/1471-2105-14-128; pmid: 23586463
49. S. Lázaro et al., Differential development of large-cell neuroendocrine
or small-cell lung carcinoma upon inactivation of 4 tumor suppressor
genes. Proc. Natl. Acad. Sci. U.S.A. 116, 22300–22306 (2019).
doi: 10.1073/pnas.1821745116; pmid: 31611390
50. G. Ferone et al., SOX2 Is the Determining Oncogenic Switch in
Promoting Lung Squamous Cell Carcinoma from Different Cells
of Origin. Cancer Cell 30, 519–532 (2016). doi: 10.1016/
j.ccell.2016.09.001; pmid: 27728803
51. J. R. Rock et al., Basal cells as stem cells of the mouse trachea
and human airway epithelium. Proc. Natl. Acad. Sci. U.S.A. 106,
12771–12775 (2009). doi: 10.1073/pnas.0906850106;
pmid: 19625615
52. K. U. Hong, S. D. Reynolds, S. Watkins, E. Fuchs, B. R. Stripp,
Basal cells are a multipotent progenitor capable of renewing
the bronchial epithelium. Am. J. Pathol. 164, 577–588 (2004).
doi: 10.1016/S0002-9440(10)63147-1; pmid: 14742263
53. B. Hogan, P. R. Tata, Cellular organization and biology of the
respiratory system. Nat. Cell Biol. (2019). doi: 10.1038/s41556-
019-0357-7; pmid: 31346286
54. A. S. Ireland et al., MYC Drives Temporal Evolution of Small Cell
Lung Cancer Subtypes by Reprogramming Neuroendocrine
Fate. Cancer Cell 38, 60–78.e12 (2020). doi: 10.1016/
j.ccell.2020.05.001; pmid: 32473656
55. R. R. Olsen et al., ASCL1 represses a SOX9+ neural crest stem-
like state in small cell lung cancer. Genes Dev. 35, 847–869
(2021). doi: 10.1101/gad.348295.121; pmid: 34016693
56. D. Lissa et al., Heterogeneity of neuroendocrine transcriptional
states in metastatic small cell lung cancers and patient-derived
models. Nat. Commun. 13, 2023 (2022). doi: 10.1038/
s41467-022-29517-9; pmid: 35440132
57. J. M. Chan et al., Signatures of plasticity, metastasis, and
immunosuppression in an atlas of human small cell lung
cancer. Cancer Cell 39, 1479–1496.e18 (2021). doi: 10.1016/
j.ccell.2021.09.008; pmid: 34653364
58. C. M. Gay et al., Patterns of transcription factor programs and
immune pathway activation define four major subtypes of
SCLC with distinct therapeutic vulnerabilities. Cancer Cell 39,
10 of 11
RES EARCH | R E S E A R C H A R T I C L E
346–360.e7 (2021). doi: 10.1016/j.ccell.2020.12.014;
pmid: 33482121
59. A. Quintanal-Villalonga et al., Multiomic analysis of lung tumors
defines pathways activated in neuroendocrine transformation.
Cancer Discov. 11, 3028–3047 (2021). doi: 10.1158/2159-8290.
CD-20-1863; pmid: 34155000
60. S. Sivakumar et al., Integrative Analysis of a Large Real-World
Cohort of Small Cell Lung Cancer Identifies Distinct Genetic
Subtypes and Insights into Histologic Transformation. Cancer
Discov. 13, 1572–1591 (2023). doi: 10.1158/2159-8290.CD-22-
0620; pmid: 37062002
61. S. Sivakumar et al., Integrative analysis of a large real-world
cohort of small cell lung cancer identifies distinct genetic
subtypes and insights into histological transformation. Cancer
Discov. 13, 1572–1591 (2023). doi: 10.1158/2159-8290.CD-22-
0620; pmid: 37062002
62. J. W. Park et al., Reprogramming normal human epithelial
tissues to a common, lethal neuroendocrine cancer lineage.
Science 362, 91–95 (2018). doi: 10.1126/science.aat5749;
pmid: 30287662
63. D. T. Montoro et al., A revised airway epithelial hierarchy
includes CFTR-expressing ionocytes. Nature 560, 319–324
(2018). doi: 10.1038/s41586-018-0393-7; pmid: 30069044
64. H. Mou et al., Airway basal stem cells generate distinct
subpopulations of PNECs. Cell Rep. 35, 109011 (2021).
doi: 10.1016/j.celrep.2021.109011; pmid: 33882306
65. M. Shivaraju et al., Airway stem cells sense hypoxia and differentiate
into protective solitary neuroendocrine cells. Science 371, 52–57
(2021). doi: 10.1126/science.aba0629; pmid: 33384370
66. A. E. Vaughan et al., Lineage-negative progenitors mobilize to
regenerate lung epithelium after major injury. Nature 517,
621–625 (2015). doi: 10.1038/nature14112; pmid: 25533958
67. M. E. Beaulieu et al., Intrinsic cell-penetrating activity propels
Omomyc from proof of concept to viable anti-MYC therapy.
Sci. Transl. Med. 11, eaar5012 (2019). doi: 10.1126/
scitranslmed.aar5012; pmid: 30894502
68. T. E. Speltz et al., Targeting MYC with modular synthetic
transcriptional repressors derived from bHLH DNA-binding
domains. Nat. Biotechnol. 41, 541–551 (2023). doi: 10.1038/
s41587-022-01504-x; pmid: 36302987
69. M. C. Basil et al., Human distal airways contain a multipotent
secretory cell that can regenerate alveoli. Nature 604, 120–126
(2022). doi: 10.1038/s41586-022-04552-0; pmid: 35355013
Gardner et al., Science 383, eadj1415 (2024) 9 February 2024
70. P. Kadur Lakshminarasimha Murthy et al., Human distal lung
maps and lineage hierarchies reveal a bipotent progenitor.
Nature 604, 111–119 (2022). doi: 10.1038/s41586-022-04541-3;
pmid: 35355018
71. W. Hill et al., Lung adenocarcinoma promotion by air
pollutants. Nature 616, 159–167 (2023). doi: 10.1038/s41586-
023-05874-3; pmid: 37020004
72. C. Mascaux et al., Immune evasion before tumour invasion in
early lung squamous carcinogenesis. Nature 571, 570–575
(2019). doi: 10.1038/s41586-019-1330-0; pmid: 31243362
73. C. Martínez-Ruiz et al., Genomic-transcriptomic evolution in
lung cancer and metastasis. Nature 616, 543–552 (2023).
doi: 10.1038/s41586-023-05706-4; pmid: 37046093
74. P. Selenica et al., APOBEC mutagenesis, kataegis,
chromothripsis in EGFR-mutant osimertinib-resistant lung
adenocarcinomas. Ann. Oncol. 33, 1284–1295 (2022).
doi: 10.1016/j.annonc.2022.09.151; pmid: 36089134
75. R. Buisson et al., Passenger hotspot mutations in cancer driven
by APOBEC3A and mesoscale genomic features. Science 364,
eaaw2872 (2019). doi: 10.1126/science.aaw2872;
pmid: 31249028
76. S. Henderson, A. Chakravarthy, X. Su, C. Boshoff, T. R. Fenton,
APOBEC-mediated cytosine deamination links PIK3CA helical
domain mutations to human papillomavirus-driven tumor
development. Cell Rep. 7, 1833–1841 (2014). doi: 10.1016/
j.celrep.2014.05.012; pmid: 24910434
ACKN OWLED GMEN TS
Funding: This work was supported by the National Cancer
Institute (NCI) NIH cooperative agreement U01CA224326 (to H.V.);
NCI NIH grants P01CA120964 and R35CA197588 (to L.C.C.); NCI
NIH grants R01CA256188, R01CA280414, R01CA280572, and
R21CA266660-01 (to A.M.L.); the Burroughs Wellcome Fund (to
A.M.L.); the Lung Cancer Research Foundation (to A.M.L.); the
Damon Runyon Cancer Research Foundation fellowship DRG-2343-18
(to E.E.G.); and NIH training grant T32 GM132083 (to E.M.E).
The Varmus lab is supported in part by the Lewis Thomas
University Professor endowment. Author contributions:
Conceptualization: E.E.G. and H.V. Methodology: E.E.G., E.M.E., M.J.H.,
K.L., J.T., B.D.S., L.C.C., and A.M.L. Experimental Design: E.E.G.,
E.M.E., A.M.L., and H.V. Visualization: E.E.G., E.M.E., M.J.H., and
A.M.L. Funding acquisition: E.E.G., L.C.C., A.M.L., and H.V. Supervision:
H.V. and A.M.L. Pathology review: C.Z. Writing – original draft: E.E.G.
Writing – review and editing: all authors. Competing interests: L.C.C.
is a cofounder and member of the Scientific Advisory Board (SAB)
and holds equity in Faeth Therapeutics, Volastra Therapeutics, and
Larkspur Therapeutics. He is also a cofounder, former member of the
SAB and holds equity in Agios Pharmaceuticals and Petra
Pharmaceuticals (now owned by Loxo@Lilly). These companies are
developing novel therapies for cancer. L.C.C.’s laboratory has
previously received some financial support from Petra
Pharmaceuticals. None of these companies are currently providing
support for the Cantley laboratory. H.V. is a member of the SABs of
Volastra, Dragonfly Therapeutics, and Surrozen. None of these
companies are currently providing support for the Varmus laboratory.
All other authors declare no competing interests. Data and materials
availability: Lead contact: Requests for resources should be directed
to and will be fulfilled by Eric E. Gardner (eeg2001@med.cornell.edu)
or Ashley M. Laughney (asl4003@med.cornell.edu). Materials: All
mouse models, organoids derived from mice, and plasmids described
in this work will be made available to investigators through an
institutional or third-party Material Transfer Agreement (MTA) upon
reasonable request. Select plasmids will be submitted to Addgene
upon manuscript acceptance. Not all mouse strains are currently
active in the Varmus Laboratory colony—please contact Eric E. Gardner
for specific information and availability. Data and code: The
processed single-cell data and relevant code, including Docker
environments with Jupyter notebooks demonstrating key analyses,
are available on either the Gene Expression Omnibus (GEO no.
GSE248207) or Laughney Lab GitHub (https://github.com/
LaughneyLab/Lung_Histological_Transformation). License
information: Copyright © 2024 the authors, some rights reserved;
exclusive licensee American Association for the Advancement of
Science. No claim to original US government works. https://www.
science.org/about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adj1415
Materials and Methods
Supplementary Text
Figs. S1 to S15
Tables S1 to S4
References (77–115)
MDAR Reproducibility Checklist
Submitted 23 June 2023; accepted 8 December 2023
10.1126/science.adj1415
11 of 11
RES EARCH

◥ a preference for ORS, a preference for anti-

RESEARCH ARTICLE SUMMARY biotics, or no preference. To measure the effect
of financial incentives, some of the stand-
HEALTH ECONOMICS ardized patients assigned to the no-preference
arm informed the provider that they would
What drives poor quality of care for child diarrhea? purchase medicines from a different location,
thereby eliminating the provider’s financial
Experimental evidence from India incentive to recommend more-lucrative treat-
ments. Finally, to estimate the effect of ORS
Zachary Wagner*, Manoj Mohanan, Rushil Zutshi, Arnab Mukherji, Neeraj Sood stock-outs, we randomly assigned all providers
in half of the 253 towns to receive a 6-week
supply of ORS.
INTRODUCTION: Diarrhea is a leading cause of patients prefer non-ORS treatments (e.g., anti-
death in children, with nearly 500,000 young biotics) or dislike ORS because of poor taste, RESULTS: We found that having standardized
lives lost to diarrhea each year. Almost all lack of observable symptom relief, and percep- patients express a preference for ORS increased
these lives could be saved with a low-cost and tions that ORS is not a real medicine. Second, ORS prescribing by 27 percentage points com-
widely available treatment: oral rehydration providers could be responding to financial pared with no preference; 28% prescribed or
salts (ORS). However, at the present time, incentives to sell more-profitable alternatives. dispensed ORS when standardized patients
nearly half of diarrhea cases around the world ORS is inexpensive, and antibiotics generate expressed no preference, and 55% prescribed
do not receive ORS. Millions of young lives nearly double the profit. Finally, providers or dispensed ORS when standardized patients
could be saved if we can find ways to increase often have ORS stock-outs (out-of-stock events) expressed an ORS preference (96% increase).
ORS use. and might prefer to dispense something they We show that this is mainly because providers
Even when children seek care from a health have in stock to not lose out on the sale. Each of think only 18% of their patients want ORS on
care provider for their diarrhea, as most do, these potential barriers to ORS prescribing average, when, in reality, ORS was the most
they often do not receive ORS. Surprisingly, suggests a different solution. preferred treatment reported by patients in
most health care providers in developing We used a randomized controlled trial to household surveys. Eliminating stock-outs
countries know that ORS is a lifesaving and simultaneously study the role of these three increased ORS provision by 7 percentage
inexpensive treatment for child diarrhea, yet leading explanations for the underprescrib- points on average and by 17 percentage points
few prescribe it. This know-do gap has puzzled ing of ORS. More than 2000 providers across among clinics that sell, rather than prescribe,
experts for decades and cost millions of lives. 253 medium-sized towns in the Indian states medicine. Eliminating financial incentives to
of Karnataka and Bihar participated in the sell more-lucrative medicines had no effect on
RATIONALE: To develop interventions that in- study. To measure the effect of perceived average but did increase ORS prescribing at
crease ORS prescribing, we must have a clear patient preferences, we had standardized pa- pharmacies by 9 percentage points. By com-
understanding of why providers do not pre- tients (actors trained to act as patients) make bining these results with the prevalence of
scribe ORS even though they know it is the unannounced visits where they presented a each barrier estimated through provider and
standard of care. There are several leading case of diarrhea for their 2-year-old child, and household surveys, we estimate that provider
explanations. First, providers might think that we randomly assigned whether they expressed misperceptions that patients do not want ORS
explain 42% of underprescribing, whereas
Gain in ORS prescribing or dispensing Percent contribution to the problem stock-outs and financial incentives explain
by removing each barrier only 6 and 5%, respectively.
100%
Share of cases prescribed or dispensed ORS

CONCLUSION: Provider misperceptions that pa-

tients do not want ORS play the biggest role
80% 42% 47% ? in the underprescribing of ORS and are 6 to
No financial incentives Provider Other 10 times more important than financial in-
No stock-outs perceptions barriers
of patient
centives or stock-outs. These results suggest
60% that interventions to change provider misper-
All patients preferences
show preference ceptions of patients’ ORS preferences should
for ORS be aggressively explored because they have
40%
ORS 6% 5% the potential to substantially increase ORS use.
These interventions could target patients or
Status quo: Providers’
Stock-outs caretakers and encourage them to express
ORS prescribed financial
20%
incentives an ORS preference when they seek care, or they
or dispensed
42% of the time could target providers directly and inform
0% them that ORS preferences are more common

Provider perceptions that patients do not want ORS are the most important barrier to ORS pre-
than they think.
▪
scribing. The figure shows that if all patients showed a preference for ORS, it would increase ORS
prescribing by 24.5 percentage points, from 42 to 66.5%. Thus, provider perceptions of patient preferences The list of author affiliations is available in the full article online.
*Corresponding author. Email: zwagner@rand.org
explain 42% of the problem, whereas stock-outs and financial incentives explain only 6.4 and 5%,
Cite this article as Z. Wagner et al., Science 383, eadj9986 (2024).
respectively. Estimates are based on the effects of removing each barrier combined with estimates on the DOI: 10.1126/science.adj9986
prevalence of each barrier. The parameters used for these calculations are listed in table S21. In the
status quo, 16% of patients showed a preference for ORS, 42% of providers had ORS in stock, and all READ THE FULL ARTICLE AT
providers had a financial incentive. https://doi.org/10.1126/science.adj9986

606 9 FEBRUARY 2024 • VOL 383 ISSUE 6683 science.org SCIENCE

RES EARCH

◥ inappropriate antibiotic prescribing in other

RESEARCH ARTICLE settings (15, 16). Finally, private providers often
have ORS stock-outs (out-of-stock events) and
HEALTH ECONOMICS might prefer to dispense something that they
have in stock to not lose out on the sale (5, 17).
What drives poor quality of care for child diarrhea? Each of these potential barriers to ORS pre-
scribing suggests a different solution. It is cri-
Experimental evidence from India tical to understand how much each barrier
contributes to inappropriate care in order to
Zachary Wagner1,2*, Manoj Mohanan3, Rushil Zutshi1,2, Arnab Mukherji4, Neeraj Sood5,6 focus resources on interventions that would
be most effective at increasing ORS use.
Most health care providers in developing countries know that oral rehydration salts (ORS) are a To the best of our knowledge, this is the first
lifesaving and inexpensive treatment for child diarrhea, yet few prescribe it. This know-do gap has study to simultaneously evaluate and quantify
puzzled experts for decades. Using randomized experiments in India, we estimated the extent to which the extent to which each of these key barriers
ORS underprescription is driven by perceptions that patients do not want ORS, provider’s financial contributes to underprescribing of ORS. In
incentives, and ORS stock-outs (out-of-stock events). Patients expressing a preference for ORS this study, we used several randomized con-
increased ORS prescribing by 27 percentage points. Eliminating stock-outs increased ORS provision by trolled trials with 2282 private providers
7 percentage points. Removing financial incentives did not affect ORS prescribing on average but did across 253 medium-sized towns in India to
increase ORS prescribing at pharmacies. We estimate that perceptions that patients do not want ORS answer the following research questions: (i)
explain 42% of underprescribing, whereas stock-outs and financial incentives explain only 6 and What is the effect of provider perceptions of
5%, respectively. patients’ treatment preferences on ORS and
antibiotic prescribing? (ii) What is the effect

D
iarrhea is the second leading cause of
death for children in low- and middle-
income countries (LMICs), with nearly
half a million children under 5 years old
dying each year (1). This is true even
though nearly all such deaths could be pre-
vented with a simple and inexpensive medi-
cine: oral rehydration salts (ORS). ORS has
been lauded as one of the most important
medical advances of the 20th century (2), yet it
has been underutilized for decades (3). At pres-
ent, nearly half of diarrhea cases around the
world do not receive ORS (4). Millions of
young lives could be saved if we can find ways
to increase ORS use.
Lack of access to health care in LMICs
likely explains some portion of low ORS use.
However, most children with diarrhea visit a
health care provider for treatment, and many
of these children still do not receive ORS (5, 6).
Providers often prescribe antibiotics instead
of ORS, which do not address dehydration
and are not appropriate for common viral
diarrhea pathogens. Inappropriate prescribing
for diarrhea is particularly egregious among
private providers, who treat most childhood
illness in LMICs (5–7). In India, where nearly
a quarter of all child deaths from diarrhea
occur, more than 75% of caretakers seek care
for a child’s diarrhea from private providers,
and 45% do not receive ORS (8).
1
Department of Economics, Sociology and Statistics, RAND
Corporation, Santa Monica, CA, USA. 2Pardee RAND
Graduate School, Santa Monica, CA, USA. 3Sanford School of
Public Policy, Duke University, Durham, NC, USA. 4Center for
Public Policy, Indian Institute of Management Bangalore,
Bangalore, Karnataka, India. 5Sol Price School of Public
Policy, University of Southern California, Los Angeles, CA,
USA. 6Schaeffer Center for Health Policy and Economics,
University of Southern California, Los Angeles, CA, USA.
*Corresponding author. Email: zwagner@rand.org
There is little evidence to date documenting
why so many children fail to receive ORS when
they visit a private health care provider. This is
a critical global health question that has been
researched for decades (5, 6, 9, 10) and that
many global funders have tried to address
(11, 12), yet we still know little about the root
causes of the problem. This makes it chal-
lenging to design effective interventions to
increase ORS prescribing in the private sector.
Prior work documents that even when private
providers know that ORS is the appropriate
treatment for diarrhea, which most do, they
still fail to prescribe it (9). This implies that
lack of knowledge is unlikely to be an im-
portant driver, and educating providers about
ORS is unlikely to increase ORS dispensing.
Several studies that focused on increasing
provider knowledge as a means of improving
ORS use have been ineffective (12–14). Why is it,
then, that so many children who seek care for
diarrhea in the private sector do not receive ORS?
There are several remaining explanations
for why private providers underprescribe ORS.
First, providers might think that patients prefer
something different than ORS (e.g., an anti-
biotic), and private providers could be particu-
larly responsive to patient preferences to ensure
that patients keep supporting their business.
Providers might think that patients prefer
non-ORS treatments or dislike ORS because of
poor taste, lack of observable symptom relief
(it treats and prevents dehydration rather than
diarrhea symptoms), and a perception that
ORS is not a real medicine (it is a powder
that is mixed with water, rather than a pill or
injection). Second, private providers could be
responding to financial incentives to sell more-
profitable alternatives. ORS is inexpensive, and
antibiotics generate nearly double the profit.
There is evidence that financial incentives drive
of removing the provider’s financial incentive
to sell more lucrative medicines on ORS and
antibiotic prescribing? (iii) What is the effect of
eliminating stock-outs on ORS and antibiotic
prescribing? and (iv) What portion of the
problem do each of these barriers contribute?
Method
Sample
Our study took place in two Indian states:
Karnataka and Bihar. We chose states that are
very different from one another in socio-
economic status and diarrhea care, seeking to
ensure that our results were representative of
a broad population. Bihar, which is in the east,
is one of the poorest states in India, with 46%
of the adult population having no schooling
and only 42% having completed at least sec-
ondary school (8). By contrast, Karnataka, which
is in the south, has above average per capita
income, with only 26% of the adult population
having no schooling and 62% having com-
pleted at least secondary school. Additionally,
Bihar has below average ORS use (57% of cases
are treated with ORS compared with the na-
tional average of 61%), and Karnataka has
above average ORS use (70% of cases treated
with ORS) (8). We sampled 253 medium-sized
towns [see fig. S1 for a map of our study lo-
cations and section 1.1 of the supplementary
materials (SM) for a full description of our
sampling process]. We attempted to enroll all
private providers who treated child diarrhea
based on a census we created. We were able to
contact 59% of those identified in the census,
and 69% of those identified consented to
participate in the study.
Of the 2451 providers we enrolled, we were
able to collect prescribing data at follow-up for
2282 (6.9% lost-to-follow-up rate). Table S1
shows the characteristics of the 2282 providers
in our sample, which were collected through a

Wagner et al., Science 383, eadj9986 (2024) 9 February 2024 1 of 8

RES EARCH | R E S E A R C H A R T I C L E
baseline provider survey. Ninety-two percent
of providers were male, with an average age of
44 years; providers had 18.5 years of experi-
ence on average, and 86% said that they would
give ORS to a child with diarrhea based on a
case vignette presented to the provider. Pro-
viders saw an average of 24.7 patients per day
and 6.3 diarrhea cases per week; 56% dis-
pensed medications directly to patients, and
52% had ORS available at baseline. Our sample
includes four different types of providers: 20%
were providers with an MBBS degree (similar
to an MD degree in the United States); 37%
were providers with a degree in traditional
medicine, including Ayurveda, Yoga and Natu-
ropathy, Unani, Siddha and Homeopathy
(AYUSH); 21% were rural medical practitioners
(RMPs), who typically lack formal training but
still practice medicine; and 22% were phar-
macies. MBBS, AYUSH, and RMP providers
generally own their own practice and run it
as a business. Most only have one provider
who sees patients. We refer to this grouping
of providers (i.e., excluding pharmacies) as
“private clinics.” Pharmacies in our sample are
stand-alone businesses that sell medicines.
Providers who work at pharmacies generally
have no medical training but often advise
patients on treatment options for child diarrhea
and are often the first providers to whom
caretakers go to seek care. Many clinics have an
attached pharmacy that is part of the same busi-
ness, and we interpreted these entities as clinics
that sell medicine and not pharmacies. Tables
S2 and S3 show that the providers who were lost
to follow-up had similar characteristics across
study arms (only 2 of 28 comparisons of provider
characteristics were statistically significant).
Table S4 shows characteristics of caretakers
in the areas served by the providers in our sam-
ple who recently sought care for their child’s
diarrhea; 57% of caretakers in our setting sought
treatment at a private clinic, 31% sought treat-
ment at a private pharmacy, and only 9% sought
treatment in the public sector. Of those who
sought care from a private provider, 42% were
prescribed ORS and 49% were prescribed anti-
biotics (caretakers are often not able to distin-
guish between antibiotics and other medicines).
These numbers are similar to estimates from
demographic and health surveys (8).
In each state, we trained actors to play the
role of a father seeking care for their 2-year-old
child who had been having diarrhea for 2 days.
We recruited 40 actors in each state to go
through an extensive 2-week standardized
patient (SP) training and selected the top
25 actors to make the SP visits. The extensive
2-week training included memorizing a script
as well as responses to common questions and
practice visits with real providers.
All SPs presented a case of a 2-year-old child
who had been having uncomplicated diarrhea
for 2 days with clinical presentation of a case
Wagner et al., Science 383, eadj9986 (2024) 9 February 2024
caused by rotavirus. Half the SPs presented a
moderate case (four to five loose stools the
previous night) and the other half a severe
case (10 to 12 loose stools the previous night
and symptoms of dehydration), but both cases
were severe enough to require ORS. The stan-
dard of care for a case of rotavirus as reported
in the Indian Academy of Pediatrics consensus
statement is ORS combined with zinc, and no
antibiotics (18). Although antibiotics are appro-
priate for some cases of child diarrhea, we
designed the case with clinical presentation
(no blood in stool, feces quality not sticky or
smelly, and short duration) such that it was
clear that antibiotics would not be necessary.
The only difference between the SP roles were
(i) the opening statement, in which they ex-
pressed a preference (four different statements,
which are described more below) and (ii) the
severity of the diarrhea episode (moderate or
severe). Each SP played all eight possible SP
roles to help control for individual SP effects.
We received institutional review board (IRB)
approval from RAND’s Human Subjects Pro-
tections Committee in the United States and
the Karesa Independent Ethics Committee
in India as well as approval from the Indian
Council of Medical Research. See section 1.3.1
of the SM for more details on the SP design
and overall data collection process.
Experimental design
Our SP experiment included four study arms.
To estimate the effect of provider perceptions
of patient preferences on ORS prescribing, we
randomized the treatment preferences that
SPs expressed to the provider during the open-
ing statement; some expressed a preference
for ORS, some expressed a preference for anti-
biotics, and some expressed no preference (see
table S20 for exact opening statements). To
estimate the effect of financial incentives to
sell more-lucrative medicines, we randomized
half of the SPs who expressed no preference
to also inform the provider that they would
purchase any prescribed medication from a
pharmacy in their hometown, thus eliminating
the financial incentive at the point of sale. To
estimate the effect of stock-outs, we random-
ized half of the providers to receive a 6-week
supply of ORS before the SP visit. This gen-
erated exogenous variation in stock-outs and
thus enabled us to isolate the causal effect of
stock-outs on ORS dispensing. Section 1.3 of
the SM describes the experimental design in
more detail. We used these three experiments
to test the following main hypotheses:
H1: Providers will be more likely to prescribe
ORS (antibiotics) when the patient shows a
preference for ORS (antibiotics).
H2: Providers will be more likely to pre-
scribe ORS and less likely to prescribe antibiotics
when there is no financial incentive to sell more-
lucrative medicines.
H3: Providers will be more likely to recom-
mend ORS when they have it available to dis-
pense at their clinic.
We then combined the causal effects esti-
mated in the randomized controlled trials (RCTs)
with population level estimates of the prevalence
of each barrier (e.g., the share of providers who
had an ORS stock-out) to estimate the effect of
eliminating each barrier on ORS prescribing
in the population. This allowed us to estimate
what ORS prescribing would look like in the
private sector if each barrier were removed.
Results
In this section, we report all analyses and re-
sults according to a preanalysis plan that we
registered at ClinicalTrials.gov (NCT04833790)
and at the American Economic Association’s
RCT registry (AEARCTR-0007276). Our empir-
ical approach to estimate results is described in
more detail in section 1.6 of the SM.
Preliminary analyses
Table 1 describes status quo prescribing or
dispensing for child diarrhea among providers
who were not assigned to receive ORS supply
and where the SP expressed no preference. With
no intervention, 28% of providers prescribed or
dispensed ORS to SPs, although 86% said they
would prescribe ORS in a provider survey (see
table S1), and 69% prescribed or dispensed anti-
biotics, although only 50% said they would. This
shows that the know-do gap in child diarrhea
care uncovered by Mohanan et al. (9) persists
more than a decade later. This also highlights
that a lack of knowledge about the standard of
care is likely not the root cause of under-
prescribing of ORS. On average, providers dis-
pensed 1.8 different medicines, and the visit cost
INR 115 (USD 1.44). Providers spent about 4 min
on average with our SPs. Providers in Bihar
were more likely to prescribe ORS and/or anti-
biotics, spend more time with the SPs, and
dispense medication directly to the patient com-
pared with providers in Karnataka. Figure S2
shows SP treatment outcomes with more gran-
ularity. Table S5 shows that pharmacies were least
likely to prescribe or dispense ORS (16%) and that
MBBS providers were most likely to prescribe or
dispense ORS (43%). Table S6 shows that pro-
viders in the four SP study arms were balanced
on key characteristics, and table S7 shows that
the two ORS supply arms were also balanced.
Primary analyses
Our primary analyses followed a preregistered
analysis plan, and all subgroups were pre-
specified (19). This section describes the re-
sults of our main hypotheses.
H1: Showing a preference for ORS nearly
doubled ORS prescribing or dispensing
Table 2 shows that when SPs expressed a pref-
erence for ORS, providers were 27 percentage
2 of 8
RES EARCH | R E S E A R C H A R T I C L E

to only a 3.5 percentage point increase in ORS

Table 1. Prescribing in no-preference control group. Data were recorded from SP visits from SPs dispensing (p = 0.049). Using a two-stage least
who showed no treatment preference and providers who did not receive ORS supply. squares instrumental variables regression,
we estimated that the effect of having ORS
in stock increases the probability of dispens-
Pooled Bihar Karnataka ing ORS by 6.8 percentage points. However,
Outcome
(n = 275) (n = 146) (n = 129) these improvements were entirely driven by
Prescribed or dispensed ORS 0.276 0.308 0.240
.....................................................................................................................................................................................................................
clinics that sell medicines (a prespecified
Prescribed or dispensed antibiotics 0.691 0.781 0.589
.....................................................................................................................................................................................................................
subgroup). In these clinics, the ORS supply
Prescribed or dispensed zinc 0.091 0.123 0.054
.....................................................................................................................................................................................................................
intervention increased ORS dispensing by
Prescribed or dispensed ORS and zinc 0.047 0.068 0.023
.....................................................................................................................................................................................................................
8.3 percentage points and having ORS in stock
Directly dispensed any medicines 0.476 0.521 0.426
.....................................................................................................................................................................................................................
increased ORS dispensing by 16.6 percent-
Number of medicines prescribed or dispensed 1.8 2.2 1.5
.....................................................................................................................................................................................................................
age points (middle panel of Table 5). By con-
Cost of visit (INR) 116 157 69
.....................................................................................................................................................................................................................
trast, having ORS in stock did not change
Time spent with provider (min) 4.2 4.8 3.5 ORS dispensing among providers who do not
.....................................................................................................................................................................................................................
sell medicines (bottom panel of Table 5), even
though the ORS supply intervention increased
points more likely to prescribe or dispense mainly because of inaccurate provider percep- the probability of having ORS in stock by a
ORS (96% increase); 28.0% prescribed or dis- tions of patient preferences. Figure S4 shows similar amount among these clinics. Our main
pensed ORS when SPs expressed no preference, that when providers perceive that more of analyses of the ORS supply intervention ex-
and 55.1% prescribed or dispensed ORS when their patients want ORS, they are more likely clude pharmacies because nearly all pharma-
SPs expressed an ORS preference (p < 0.01; top to prescribe or dispense it; providers who think cies had ORS in stock in the control group
panel of Table 2). Providers were also 10 per- >30% of their patients want ORS are about (table S14). This suggests that stock-outs are
centage points less likely to prescribe antibiotics twice as likely to prescribe or dispense ORS not driving ORS dispensing at pharmacies.
when SPs expressed a preference for ORS (14% compared with providers who think <10% of However, our conclusions were similar when
decrease). When SPs expressed a preference for their patients want ORS (p < 0.01). we included pharmacies (table S15).
(inappropriate) antibiotics, providers were 7.3 The improvements from the ORS supply
percentage points more likely to prescribe anti- H2: Eliminating financial incentives intervention were entirely driven by increased
biotics (10% increase); 70.4% prescribed anti- increased ORS prescribing or dispensing at ORS dispensing in Bihar. The intervention
biotics when SPs expressed no preference, and pharmacies but not at clinics increased ORS dispensing by 8.0 percentage
77.7% prescribed antibiotics when SPs expressed Figure S5 shows that the revenue and profits points (66%) in Bihar on average (p < 0.05)
a preference for antibiotics (p < 0.01). Expressing from antibiotics are substantially larger than and by 11.6 percentage points among clinics
a preference for antibiotics did not change ORS those from ORS. On average, ORS prescribing that sell medicines (table S16). By contrast,
prescribing. was not significantly different when SPs pur- the average effects and effects for clinics that
Table S8 shows that the effect of showing an chased medicines from the provider (financial sell medicines in Karnataka were close to
ORS preference was strongest among providers incentive) compared with when SPs informed zero and not significant (table S17).
with no formal training, with a 47 percentage the provider that they would purchase elsewhere Providers might switch from writing ORS
point increase for pharmacies and a 41 per- (no financial incentive; top panel of Table 4). prescriptions to dispensing ORS directly to
centage point increase for RMPs, and weakest However, pharmacies (a prespecified subgroup) patients after receiving ORS supply. Thus, one
among the providers with the most training, were 9 percentage points more likely to pre- potential spillover effect of the supply inter-
with an 8.9 percentage point increase for MBBS scribe ORS when the financial incentive was vention is a reduction in ORS prescriptions,
providers (p value of difference in effect was removed (50% increase), which suggests that which would mitigate the benefits of increased
<0.01 for both pharmacies and RMPs com- low ORS prescribing at pharmacies might be ORS dispensing. This did not seem to happen.
pared with MBBS). There was no difference partly explained by financial incentives. The ORS prescribing or dispensing increased by
in the preference effect by state (table S9) or effect for RMPs was also of an important mag- roughly the same amount as ORS dispensing
by case severity (table S10). We also found no nitude (13 percentage points; table S9) but alone (Table 5), which suggests that the ad-
difference in the prices paid across the pref- only statistically significant in the model that ditional patients who were dispensed ORS would
erence arms (fig. S3). included covariates. not have received an ORS prescription absent
Similar to ORS prescribing, removing the the intervention.
Providers (inaccurately) think that patients financial incentive to sell medicines did not
do not want ORS affect antibiotic prescribing on average. However, Providers’ perceptions that patients do not want
Using data from caretaker and provider surveys, removing financial incentives to sell medicines ORS accounts for 42% of ORS underprescribing
we found that patient preferences are a barrier may have reduced antibiotic prescribing among To inform which interventions are likely to be
to ORS prescribing not because caretakers do MBBS providers, a 15 percentage point reduction most effective at increasing ORS provision, it
not want ORS but rather because providers that was only significant in adjusted models. The was important to assess the extent to which each
inaccurately think that most patients do not effect of financial incentives was similar across barrier that was examined above contributes
want ORS. Table 3 shows that although 48% states (table S12) and by case severity (table S13). to underprescribing of ORS at the population
of caretakers reported ORS as the best treat- level. For example, if most patients already
ment for child diarrhea (far higher than any other H3: Reducing stock-outs led to only a show a preference for ORS, then patient pref-
treatment), only 16% expressed a preference small improvement in ORS dispensing erences are not contributing much to the prob-
for ORS when seeking care. As a result, pro- The ORS supply intervention increased the lem, even if showing an ORS preference increases
viders think that only 18% of their patients likelihood that clinics had ORS on site to dis- ORS prescribing or dispensing. We combined
want ORS. Thus, preferences appear to be a pense by 52.1 percentage points, from 28.2 to our regression estimates of the effect of each
barrier to ORS prescribing or dispensing 80.4% (top panel of Table 5). However, this led barrier on ORS prescribing or dispensing with

Wagner et al., Science 383, eadj9986 (2024) 9 February 2024 3 of 8

RES EARCH | R E S E A R C H A R T I C L E

visiting a private provider (42%) and the share

Table 2. Effect of patient preferences on prescribing for child diarrhea. Data on prescribing are that showed a preference for ORS (16%). If,
from surveys completed by SPs after a visit with the provider. Treatment effects were estimated using instead, all patients showed a preference for
linear regression, with standard errors clustered by town. The “controls” column includes the severity ORS (i.e., 84 percentage point increase, from
of the case, whether the provider received free ORS supply, the provider’s training, whether the 16 to 100%,) it would increase ORS prescribing
provider dispenses medicines, whether the provider knew that ORS was the correct treatment at by 24.5 percentage points, from 42% of care-
baseline, the provider’s age and gender, the number patients seen per day, the number of diarrhea takers receiving an ORS prescription to 66.5%
cases seen per week, the provider’s years of experience, and whether the provider had ORS in stock (Fig. 1). Fifty-eight percent of providers in the
at baseline. Standard errors, clustered by town, are in parentheses. ***p < 0.01; **p < 0.05; *p < 0.1. control group experienced ORS stock-outs at
endline; eliminating these stock outs would
increase ORS prescribing by an additional 3.7
Outcome Patient preference Treatment effect percentage points to 70.2%. Finally, eliminating
ORS preference versus no-preference
.....................................................................................................................................................................................................................
financial incentives to sell medicines would in-
No preference ORS preference crease prescribing by another 2.4 percentage
No controls Controls
(n = 560) (n = 572)
.....................................................................................................................................................................................................................
points to 73%. Other barriers not addressed by
0.270*** 0.270*** this study explain the remaining 27% of cases
Prescribed or dispensed ORS 0.280 0.551
(0.029) (0.029)
.....................................................................................................................................................................................................................
that would not receive ORS even after these
−0.100*** −0.098*** three barriers were eliminated. Thus, provider
Prescribed or dispensed antibiotics 0.704 0.603
(0.027) (0.026)
.....................................................................................................................................................................................................................
perception of patient preferences explains
Antibiotics preference versus no-preference
.....................................................................................................................................................................................................................
42% of the problem (24.5/58), stock-outs ex-
No preference Antibiotics preference plain 6.4% (3.7/58), financial incentives ex-
No controls Controls
(n = 560) (n = 569)
.....................................................................................................................................................................................................................
plain 5% (2.4/58), and other barriers explain
−0.006 −0.006 47% (27/58). See section 2 of the SM for a more
Prescribed or dispensed ORS 0.280 0.274
(0.025) (0.025)
.....................................................................................................................................................................................................................
detailed description of these calculations.
0.073*** 0.080***
Prescribed or dispensed antibiotics 0.704 0.777 Discussion
(0.023) (0.023)
..................................................................................................................................................................................................................... Summary of findings
Despite widespread awareness and availability
of ORS, uptake has remained low for decades
across the globe (4). As a result of this know-do
Table 3. Caretaker preferences. Data from caretakers are from household surveys of caretakers
gap, nearly half a million children continue to
who had a child with a recent case of diarrhea. Caretaker perception of best treatment was obtained
die from diarrhea each year (1). The reasons
by asking caretakers “what is the best treatment for child diarrhea?” (unprompted). Caretaker
for the lack of ORS prescribing despite high
preference expressed to provider was obtained by asking caretakers if they “asked at all about any
provider knowledge of ORS appropriateness
treatments” and “if so, which ones?” Provider perception of caretaker preferences was obtained
have remained elusive. To our knowledge, this
by asking providers what share of their patients have a preference for the different treatments.
is the first study to simultaneously evaluate the
Expressing a preference for antibiotics was rare, so we grouped this response with “other medicine
role of multiple factors and to use randomized
preference.” The numbers to the left of the percentages are the total number of participants who
experiments to rigorously estimate which fac-
reported each response.
tors are the main drivers of low ORS prescrib-
ing rates. We found that provider perceptions
Diarrhea treatment preference Response of patient preferences play the biggest role in
underprescribing of ORS; indeed, provider
Caretaker perception of best treatment
..................................................................................................................................................................................................................... perceptions of patient preferences are 6 to
ORS preference 584 (48%)
..................................................................................................................................................................................................................... 10 times more important in explaining ORS
Antibiotics preference 176 (14%)
..................................................................................................................................................................................................................... underprescribing compared with financial in-
Other medicines preference 472 (40%)
..................................................................................................................................................................................................................... centives or stock-outs. When a patient simply
No preference 54 (4%)
..................................................................................................................................................................................................................... signals a preference for ORS, the likelihood
Caretaker preference expressed to provider
..................................................................................................................................................................................................................... that the patient is prescribed or dispensed
ORS preference 191 (16%)
..................................................................................................................................................................................................................... ORS by the provider is doubled. Notably, the
Other medicine preference 37 (3%)
..................................................................................................................................................................................................................... problem is not driven by low patient demand
No preference 972 (81%)
..................................................................................................................................................................................................................... for ORS; 48% of caretakers reported ORS as
Provider perception of caretaker preferences
..................................................................................................................................................................................................................... the best diarrhea treatment, far higher than
ORS preference 18%
..................................................................................................................................................................................................................... any other treatment, but only 16% of caretakers
Antibiotics preference 12%
..................................................................................................................................................................................................................... surveyed expressed an ORS preference to the
Other medicine preference 12%
..................................................................................................................................................................................................................... provider when seeking care for their child’s
No preference 58%
.....................................................................................................................................................................................................................
diarrhea. As a result, providers perceived that
only 18% of caretakers have an ORS preference.
It was also rare for caretakers to express a
estimates of the prevalence of each barrier in type based on caretaker reports; 35% of diarrhea preference for alternative treatments. Taken
the Indian population to estimate the contri- cases go to pharmacies, 23% go to MBBS pro- together, these findings suggest that interven-
bution of each barrier to underprescribing. viders, 12% go to RMPs, and 29% go to AYUSH tions to change provider misperceptions of
For each effect size, we used the provider providers. We used surveys of caretakers whose patient preferences for ORS should be aggre-
type–specific effects from tables S8, S11, and S14 child had a recent diarrhea episode to estimate ssively explored because they have the poten-
weighted by the market share for each provider the share of caretakers who received ORS when tial to substantially increase ORS use.

Wagner et al., Science 383, eadj9986 (2024) 9 February 2024 4 of 8

RES EARCH | R E S E A R C H A R T I C L E

understand why MBBS providers fail to

Table 4. Effect of provider financial incentives on prescribing for child diarrhea. Data on prescribe ORS.
prescribing are from surveys completed by SPs after a visit with the provider. Treatment effects are
estimated using linear regression, with standard errors clustered by town. The “controls” column includes The role of ORS stock-outs and financial
the severity of the case, whether the provider received free ORS supply, the provider’s training, whether the incentives to sell medicines
provider dispenses medicines, whether the provider knew that ORS was the correct treatment at baseline, We found that ORS stock-outs pose an import-
the provider’s age and gender, the number patients seen per day, the number of diarrhea cases seen per ant barrier to ORS prescribing for some pro-
week, the provider’s years of experience, and whether the provider had ORS in stock at baseline. Standard viders. But the stock-out effect is nuanced—it
errors, clustered by town, are in parentheses. ***p < 0.01; **p < 0.05; *p < 0.1. was only important among clinics that sell
medicines and was not important at phar-
macies because nearly all had ORS available
Outcome Financial incentive assignment Treatment effect at baseline. Thus, appropriate targeting of
All providers who sell medicine
.....................................................................................................................................................................................................................
interventions to address stock-outs is extremely
Purchase from Purchase important. Moreover, we found that stock-outs
No
provider elsewhere Controls only affected ORS dispensing in Bihar, and
controls
(n = 332) (n = 321)
.....................................................................................................................................................................................................................
increasing ORS supply in Karnataka had no
0.046 0.042 effect on ORS dispensing. This could partly be
Prescribed or dispensed ORS 0.256 0.302
(0.030) (0.030)
.....................................................................................................................................................................................................................
explained by the fact that the distribution of
−0.013 −0.023 provider types is different between the two
Prescribed or dispensed antibiotics 0.702 0.689
(0.035) (0.033)
.....................................................................................................................................................................................................................
states; Bihar has more RMPs, who sell medi-
Pharmacies
.....................................................................................................................................................................................................................
cines, and Karnataka has more MBBS doctors,
Purchase from Purchase who generally do not sell medicines (table S1).
No
provider elsewhere Controls However, more work is needed to fully under-
controls
(n = 137) (n = 135)
.....................................................................................................................................................................................................................
stand why increases in ORS supply in Karnataka
0.090** 0.091** had no effect on ORS dispensing.
Prescribed or dispensed ORS 0.182 0.272
(0.045) (0.045)
.....................................................................................................................................................................................................................
The ORS stock-out effect does not appear to
−0.009 −0.015 be driven by general supply-chain issues because
Prescribed or dispensed antibiotics 0.781 0.772
(0.049) (0.050)
.....................................................................................................................................................................................................................
nearly all pharmacies had ORS available. Ins-
Clinics that sell medicine
.....................................................................................................................................................................................................................
tead, it appears to be driven by clinics that sell
Purchase from Purchase medicines failing to order sufficient ORS inven-
No
provider elsewhere Controls tory and avoiding recommending ORS if it was
controls
(n = 197) (n = 222)
.....................................................................................................................................................................................................................
not in stock. This could be because they want to
0.006 −0.002 sell products that they have in stock to not lose
Prescribed or dispensed ORS 0.310 0.315
(0.045) (0.045)
.....................................................................................................................................................................................................................
the sale. By contrast, providers who do not sell
−0.015 −0.039 medicines are accustomed to recommending
Prescribed or dispensed antibiotics 0.650 0.635
(0.047) (0.043) rather than dispensing medicines, so giving
.....................................................................................................................................................................................................................
them ORS did not increase their rate of dis-
pensing ORS.
We found that financial incentives to sell
What interventions might work to change We found substantial heterogeneity in the medicine are an important barrier to prescrib-
provider perceptions and how to target effect of signaling preferences for ORS by pro- ing at pharmacies; pharmacies were 50% more
interventions vider type. The providers with the most train- likely to prescribe ORS when the financial
At present, it is not clear what interventions ing (MBBS providers) prescribed or dispensed incentive was removed. About one-third of
can effectively change providers’ misperception ORS at the highest rates (although still far caretakers seek care for a child’s diarrhea from
that patients have a distaste for ORS. These lower than optimal) and were the least sensi- pharmacies in our study area (table S4), so this
interventions could target patients or caretakers tive to patient preferences. This could be be- represents a nontrivial barrier to ORS use.
and encourage them to express an ORS pref- cause the expertise of these providers is more Thus, interventions to restructure financial
erence when they seek care. This could be done trusted and their business model relies less on incentives, such as providing an extra financial
through community-level social marketing or satisfying patient requests. Pharmacies and incentive to encourage pharmacists to prescribe
mass-media campaigns, but similar campaigns rural medical providers were most responsive ORS, could be effective. It is also important to
have had only modest success in the past (10). to patient preferences. These providers, who note that our study likely underestimates the
This could also be done directly at health lack formal training, might be less confident effects of financial incentives on prescribing or
clinics through prominently displayed posters in their medical expertise, and their businesses dispensing ORS and antibiotics. For example,
that encourage caretakers to request ORS for might rely more on appeasing customer re- providers might have the habit of dispensing
child diarrhea. A similar intervention was not- quests for medicines and tests. This suggests antibiotics instead of ORS owing to the higher
ably effective at encouraging passengers to that interventions that address patient prefer- per-unit profits from selling antibiotics. Because
speak up about risky driving on public buses ences are likely to be most effective if they of this habit formation, these providers will
(20). Alternatively, interventions could target target non-MBBS providers, which account continue to prescribe antibiotics even when
providers directly and inform them that ORS for about three-quarters of the market share the patient informs them that they will pur-
preferences are more common than they think for child diarrhea cases in the private sector. chase the medicines elsewhere. In addition,
and that patients generally prefer ORS to other Moreover, this suggests that among MBBS our study only addressed point-of-sale profit
treatments. Future work should rigorously test providers, other factors that were not in- incentives. It is possible that providers have
these various approaches to assess which one vestigated in this study are likely driving the additional financial incentives to encourage
is most effective at increasing ORS prescribing. problem. More work is needed to better repeat business or through kickbacks from

Wagner et al., Science 383, eadj9986 (2024) 9 February 2024 5 of 8

RES EARCH | R E S E A R C H A R T I C L E

pharmaceutical companies. Thus, some finan-

Table 5. Effect of ORS supply on prescribing for child diarrhea. Data on prescribing and cial incentives likely remained even when the
dispensing are from surveys completed by SPs after a visit with the provider. Data on ORS stock are SP was purchasing elsewhere.
from follow-up provider surveys completed on the same day as the SP visit. The effect of the supply
intervention was estimated using linear regression, with standard errors clustered by town. The effect Study contributions and policy implications
of ORS in stock was estimated using two-stage least squares regression with random assignment Our study fills gaps and extends the existing
as an instrument for having ORS in stock. Standard errors, clustered by town, are in parentheses. literature in several ways. First, to our knowledge,
***p < 0.01; **p < 0.05; *p < 0.1. NA, not applicable. we provide the most comprehensive evidence to
date on why one of the most important health
technologies in history, ORS, is often not pre-
Outcome ORS supply assignment Treatment effect scribed. Previously, several papers documented
All clinics
.....................................................................................................................................................................................................................
the problem of suboptimal ORS prescribing
Control Given supply Effect of supply Effect of ORS (4, 6, 9) but lacked extensive evidence that is
(n = 885) (n = 891) intervention in stock
.....................................................................................................................................................................................................................
needed to understand why this occurs. Our
0.521*** study provides new evidence to help guide inter-
ORS in stock at follow-up 0.282 0.804 NA
(0.033)
.....................................................................................................................................................................................................................
ventions to address the problem.
0.035** 0.068** Second, we contribute to sparse, but grow-
Dispensed ORS 0.087 0.122
(0.018) (0.033)
.....................................................................................................................................................................................................................
ing, evidence on what drives poor quality of
0.026 0.051 care in LMICs. Recent reports on quality of
Prescribed or dispensed ORS 0.349 0.376
(0.029) (0.056)
.....................................................................................................................................................................................................................
care from the Lancet Commission and the
Clinics that sell medicines
.....................................................................................................................................................................................................................
National Academies, in addition to a large of
Control Given supply Effect of supply Effect of ORS body of research, highlight the urgent need
(n = 385) (n = 370) intervention in stock
.....................................................................................................................................................................................................................
for global quality improvements (21–23), yet
0.499*** we know little about what drives poor quality
ORS in stock at follow-up 0.382 0.881 NA
(0.042)
.....................................................................................................................................................................................................................
of care in LMICs and how to improve it.
0.083*** 0.166*** Third, in the existing literature, it was unclear
Dispensed ORS 0.111 0.195
(0.029) (0.057)
.....................................................................................................................................................................................................................
why inappropriate care persists despite high
0.081** 0.161** levels of provider knowledge of appropriate
Prescribed or dispensed ORS 0.319 0.400
(0.040) (0.078)
.....................................................................................................................................................................................................................
care. This know-do gap phenomenon exists
Clinics that do not sell medicines
.....................................................................................................................................................................................................................
across a wide range of lifesaving medical inter-
Control Given supply Effect of supply Effect of ORS ventions, including triaging for chest pain, di-
(n = 501) (n = 521) intervention in stock
.....................................................................................................................................................................................................................
agnostics for tuberculosis, and appropriate
0.541*** management of respiratory conditions (9, 24, 25).
ORS in stock at follow-up 0.208 0.749 NA
(0.037)
.....................................................................................................................................................................................................................
Banerjee et al. suggested that patients not trust-
0.002 0.006 ing providers partly explained poor clinical
Dispensed ORS 0.069 0.071
(0.017) (0.032)
.....................................................................................................................................................................................................................
performance despite high levels of knowledge
−0.014 −0.023 (26). Kovacs et al., who also used variation in
Prescribed or dispensed ORS 0.372 0.358
(0.035) (0.066) SP patient profiles, showed that patients who
.....................................................................................................................................................................................................................
volunteered more information about their
symptoms to the provider reduced the know-
do gap (27). Our study extends this literature
by showing that providers’ misperceptions of
patient preferences plays a key role in explaining
Gain in ORS prescribing or dispensing Percent contribution to the problem
by removing each barrier know-do gaps.
100% Fourth, our study carries policy implications
for other countries beyond India. We contrib-
Share of cases prescribed or dispensed ORS

ute to the mixed scientific literature on patient

80% 42% 47% ? preferences and supplier-induced demand for
No financial incentives Provider Other inappropriate treatments. Currie et al., who
No stock-outs perceptions barriers used a similar SP design as our study, showed
60% of patient that patient demand and financial incentives
All patients preferences
show preference both drive inappropriate antibiotic prescribing
for ORS in China (15, 16). Similarly, Lopez et al. found
40% that private providers in Mali are responsive
ORS 6% 5%
to patient preferences for antimalarials even
Status quo: Providers’ when they are inappropriate (28). However,
Stock-outs financial
20% ORS prescribed Lagarde and Blaauw, who also used variation
or dispensed incentives
in SP profiles, showed that financial incentives
42% of the time
0% are not important for inappropriate antibiotic
prescribing in South Africa and that providers
Fig. 1. Contribution of each barrier. Estimates are based on the effects of removing each barrier combined overprescribe antibiotics even when they bear
with estimates on the prevalence of each barrier. The parameters used for these calculations are listed in the cost of the drugs (29). Moreover, studies
table S21. In the status quo, 16% of patients showed a preference for ORS, 42% of providers had ORS from Kenya and Tanzania have shown that pa-
in stock, and all providers had a financial incentive. tient preferences for antibiotics play a limited

Wagner et al., Science 383, eadj9986 (2024) 9 February 2024 6 of 8

RES EARCH | R E S E A R C H A R T I C L E
role in inappropriate antibiotic prescribing in
those countries (30, 31). Our work extends this
literature, showing that inappropriate prescrib-
ing could also be driven by providers’ percep-
tions of patient preferences (e.g., that they do
not want ORS) rather than patients’ expressed
preferences.
Lastly, we also found that showing a pref-
erence for ORS reduced antibiotics prescrib-
ing, which suggests some degree of perceived
substitution between ORS and antibiotics.
This is further supported by table S23, which
shows that prescriptions of ORS alone in-
creased and antibiotics alone decreased when
patients showed an ORS preference. The fact
that these two treatments are sometime seen
as substitutes has important global health impli-
cations. Antibiotic resistance to diarrhea patho-
gens is a global crisis (32), and India is one of
the largest contributors to global antibiotic
resistance (33). Child diarrhea is one of the
most common illnesses for which antibiotics
are prescribed inappropriately, and India
accounts for the most cases of child diarrhea
of any country in the world (34). Thus, chang-
ing provider perceptions of ORS preferences
in India could help curb global antibiotic
resistance.
Limitations and future directions
This study has several limitations. First, owing
to ethical issues, we could not have a child
present during SP visits. Although the care
received by SPs is similar to the care reported
in caretaker surveys, it is possible that the
treatment effects that we estimated would be
different if a child were present. Second, our
estimation of the contribution of each barrier
requires many assumptions that are uncertain.
For example, measuring caretakers’ ORS and
treatment preferences is complex, and it is
possible that our measures, such as the frac-
tion of households that report ORS as best
treatment that was estimated using simple
questions on a household survey, might not
fully capture treatment preferences of pa-
tients. Third, although the effects that we
estimated for each barrier have high internal
validity, the effects in our experiment might
not generalize to scenarios in which inter-
ventions to address each barrier are scaled
up more broadly. For example, patients who
ask for ORS might increase ORS use, but this
in turn might increase the price of ORS, which
could reduce ORS use in future periods. It is
difficult to estimate such general equilibrium
effects in an experimental setting. Finally, al-
though we conducted this research in two
distinct states to improve generalizability,
our study setting might not generalize to the
rest of India because we excluded smaller
towns and larger cities from the sample.
Medium-sized towns could have different case
mixes and provider types and thus different
Wagner et al., Science 383, eadj9986 (2024) 9 February 2024
quality of care than more rural or more densely
populated areas (35). Although a study using a
similar SP case to this study documented sim-
ilarly poor quality in rural settings in Bihar (9),
it is possible that the effects that we estimated
would be different. Moreover, we were only
able to enroll 42% of eligible providers, and it
is unclear whether these providers are repre-
sentative of the population of providers in our
study areas.
Future studies could address questions that
remain unanswered from this investigation.
We do not have good data on the additional
barriers that are driving underprescribing of
ORS that were not addressed by this study
and that explain roughly 47% of the problem.
Why is it that some providers who know ORS
is appropriate, have it available to dispense, and
have a patient who shows an ORS preference
still fail to prescribe ORS? More research is
needed to identify these additional barriers.
Conclusions
A long-standing puzzle in global health has
been that providers do not prescribe ORS for
child diarrhea even though they know it is the
standard of care. We show that providers’
perceptions that patients do not prefer ORS
explain a large portion the problem. Interven-
tions to change providers’ perceptions of
patients’ ORS preferences have the potential
to increase ORS use and reduce child mortality
from diarrhea.
Methods summary
All methods are described in detail in the SM.
The following is a summary of key aspects of
our methodology.
Sampling
Our sampling strategy consisted of three steps:
(i) sampling districts, starting with those with
the highest diarrhea prevalence; (ii) sampling
all towns within a district with a population
larger than 10,000 and smaller than 150,000
according to the latest available census (2011);
and (iii) recruiting all private providers who
treat child diarrhea in the town. To recruit
providers, we first conducted a census of all
clinics and RMPs in the town. We used this
census to enroll and survey all providers who
were available for an interview and provided
consent to participate in the study. We also
identified all pharmacies in the town and then
randomly sampled an average of two phar-
macies per town that were independent estab-
lishments and not attached to a clinic or RMP.
Interventions
SP visits
The preferences and financial incentives inter-
ventions were both done through variation
in the way that SPs presented a case of child
diarrhea when visiting the provider. All SP
visits presented a case of a 2-year-old child who
had been having diarrhea for 2 days. Half of the
SPs presented a moderate case and the other
half a severe case, but both cases were severe
enough to require ORS. The case was designed
such that antibiotics would not be appropriate
(no blood in stool, feces quality not sticky or
smelly, and short duration). The moderate
case had four to five loose stools the previous
night, and the child was taking fluids. In the
severe case, the child had 10 to 12 loose stools
the previous night, was not taking fluids or
food, and showed symptoms of dehydration
(low energy and sunken eyes). The only dif-
ferences between the SP roles were (i) the
opening statement, in which they expressed
a preference or not, and (ii) the severity of the
diarrhea episode (moderate or severe). Each
provider was randomly assigned to receive one
of the eight SP roles. Each SP played all eight
possible SP roles to help control for individual
SP effects. We trained actors extensively on
how to present the case and how to play each
role. All SPs visited the providers without a
child, which is common practice in India.
Preferences intervention
During the SP’s opening statement when they
first encountered the provider, some SPs ex-
pressed a preference for ORS, some expressed
a preference for antibiotics, and some ex-
pressed no preference. All opening statements
are provided in table S20. SPs who expressed
a medicine preference showed the doctor a
picture of used packaging of the medicine on
their phone, indicated that they had used this
medicine the previous time their child had
diarrhea, and asked if they could use it again.
SPs who expressed no preference just told the
doctor that their child had diarrhea and asked
for a recommendation.
Financial incentives intervention
We randomly assigned half of the SPs who ex-
pressed no preference to inform the provider
that they only wanted a treatment recommen-
dation (i.e., they would not purchase any treat-
ment) because their sick child was back in their
hometown. They were only visiting the current
town for work, and they would have a relative
deliver the treatment to the wife who was home
with the child. This reduced the influence of
financial incentives in the provider’s decision to
prescribe one treatment over another.
ORS supply intervention
To investigate the effect of ORS stock-outs, we
layered on a cluster-randomized experiment
with the SP experiments described above. In
particular, we randomly assigned all providers
in half of the towns to receive 60 packets of
ORS (about a 6-week supply). We distributed
ORS to providers upon enrollment, which
took place 2 to 3 weeks before the SP visits.
7 of 8
RES EARCH | R E S E A R C H A R T I C L E
Data collection
We collected four different types of data in this
study. First, we conducted baseline provider 5.
surveys, which took place directly after pro-
vider enrollment. Second, we conducted care-
taker surveys, which were conducted at caretaker 6.
homes (n = 963 in Bihar, and n = 237 in
Karnataka) and recorded caretaker treatment
preferences, treatment-seeking behavior, and 7.
provider interactions for caretakers whose
child had a recent case of diarrhea. Third, we
conducted SP debrief surveys within an hour
8.
of the SPs visiting the provider, which re-
corded information on what happened during
the visit. Fourth, we conducted follow-up sur- 9.
veys with providers within a few hours of the
SP visit to measure ORS stock-outs at the time
of the SP visit. 10.
Outcomes
We prespecified two primary outcomes: (i)
11.
whether the provider prescribed or dispensed
ORS and (ii) whether the provider prescribed
or dispensed antibiotics. We coded these binary
12.
outcomes to 1 if the SP was given the treatment
directly, was given a written prescription, or
was verbally recommended the treatment. When
assessing the impact of the ORS supply inter-
vention, we also separately analyzed ORS dis- 13.
pensing (given directly to the SP) because ORS
supply is most likely to affect dispensing ORS
rather than prescribing ORS.
14.
Statistical analysis
We estimated all regressions using linear proba-
bility models with standard errors clustered by
15.
town in accordance with our preanalysis plan.
We estimated unadjusted models and adjusted
models that controlled for the severity of the
case, whether the provider received free ORS 16.
supply, the provider’s training, whether the
provider dispenses medicines, whether the pro-
vider knew ORS was the correct treatment at 17.
baseline, the provider’s age and gender, the
18.
number of patients seen per day, the number
of diarrhea cases seen per week, the provider’s
years of experience, and whether the provider
19.
had ORS in stock at baseline. We chose co-
variates based on a priori expectations about
which variables would affect ORS prescrib-
20.
ing or dispensing.
RE FE RENCES AND N OT ES
1. J. Perin et al., Global, regional, and national causes of under-5 21.
mortality in 2000-19: An updated systematic analysis with
implications for the Sustainable Development Goals. Lancet
Child Adolesc. Health 6, 106–115 (2022). doi: 10.1016/S2352-
22.
4642(21)00311-4; pmid: 34800370
2. Water with sugar and salt. Lancet 2, 300–301 (1978).
pmid: 79090
3. B. C. Forsberg, M. G. Petzold, G. Tomson, P. Allebeck, 23.
Diarrhoea case management in low- and middle-income
countries—An unfinished agenda. Bull. World Health Organ.
85, 42–48 (2007). doi: 10.2471/BLT.06.030866; 24.
pmid: 17242757
4. C. T. Sreeramareddy, Y.-P. Low, B. C. Forsberg, Slow progress
in diarrhea case management in low and middle income
Wagner et al., Science 383, eadj9986 (2024) 9 February 2024
countries: Evidence from cross-sectional national surveys,
1985-2012. BMC Pediatr. 17, 83 (2017). doi: 10.1186/s12887-
017-0836-6; pmid: 28320354
Z. Wagner, M. Shah, N. Sood, Barriers to use of oral
rehydration salts for child diarrhea in the private sector:
Evidence from India. J. Trop. Pediatr. 61, 37–43 (2015).
doi: 10.1093/tropej/fmu063; pmid: 25389183
N. Sood, Z. Wagner, Private sector provision of oral rehydration
therapy for child diarrhea in sub-Saharan Africa. Am. J. Trop.
Med. Hyg. 90, 939–944 (2014). doi: 10.4269/ajtmh.13-0279;
pmid: 24732456
K. A. Grépin, Private sector an important but not dominant
provider of key health services in low-and middle-income
countries. Health Aff. (Millwood) 35, 1214–1221 (2016).
doi: 10.1377/hlthaff.2015.0862; pmid: 27385236
Demographic and Health Surveys Program (DHS), “India
Demographic and Health Survey 2019-2020”
(ICF International, 2022).
M. Mohanan et al., The know-do gap in quality of health care
for childhood diarrhea and pneumonia in rural India. JAMA
Pediatr. 169, 349–357 (2015). doi: 10.1001/
jamapediatrics.2014.3445; pmid: 25686357
F. Lam et al., Effect of enhanced detailing and mass media
on community use of oral rehydration salts and zinc
during a scale-up program in Gujarat and Uttar Pradesh.
J. Glob. Health 9, 010501 (2019). doi: 10.7189/jogh.09.010501;
pmid: 30546870
Abt Associates, Strengthening Health Outcomes through the
Private Sector (SHOPS) (2018); https://www.abtassociates.
com/projects/strengthening-health-outcomes-through-the-
private-sector-shops.
M. Mohanan, K. S. Babiarz, J. D. Goldhaber-Fiebert,
G. Miller, M. Vera-Hernández, Effect of a large-scale social
franchising and telemedicine program on childhood
diarrhea and pneumonia outcomes in India. Health Aff. 35,
1800–1809 (2016). doi: 10.1377/hlthaff.2016.0481;
pmid: 27702952
W. Friedman, B. Woodman, M. Chatterji, Can mobile phone
messages to drug sellers improve treatment of childhood
diarrhoea?—A randomized controlled trial in Ghana.
Health Policy Plan. 30, i82–i92 (2015). doi: 10.1093/heapol/
czu122; pmid: 25759456
M. Mohanan et al., Evaluation of a social franchising and
telemedicine programme and the care provided for childhood
diarrhoea and pneumonia, Bihar, India. Bull. World Health
Organ. 95, 343–352E (2017). doi: 10.2471/BLT.16.179556;
pmid: 28479635
J. Currie, W. Lin, J. Meng, Addressing antibiotic abuse
in China: An experimental audit study. J. Dev. Econ. 110,
39–51 (2014). doi: 10.1016/j.jdeveco.2014.05.006;
pmid: 26949284
J. Currie, W. Lin, W. Zhang, Patient knowledge and antibiotic
abuse: Evidence from an audit study in China. J. Health Econ.
30, 933–949 (2011). doi: 10.1016/j.jhealeco.2011.05.009;
pmid: 21733587
S. Wilson, S. Morris, S. Gilbert, “ORS case study: India” (Bill &
Melinda Gates Foundation, 2012).
S. Bhatnagar, N. Bhandari, U. C. Mouli, M. K. Bhan; IAP National
Task Force, Consensus Statement of IAP National
Task Force: Status report on management of acute diarrhea.
Indian Pediatr. 41, 335–348 (2004). pmid: 15123862
Z. Wagner, Manoj Mohanan, Rushil Zutshi, Arnab Mukherji,
N. Sood, “What drives poor care for child diarrhea?: A
standardized patient experiment in India” (AEARCTR-0007276,
AEA RCT Registry, 2023). doi: 10.1257/rct.7276-2.2
J. Habyarimana, W. Jack, Heckle and Chide: Results
of a randomized road safety intervention in Kenya.
J. Public Econ. 95, 1438–1446 (2011). doi: 10.1016/
j.jpubeco.2011.06.008
National Academies of Sciences, Engineering, and Medicine,
“Crossing the global quality chasm: Improving health care
worldwide” (National Academies Press, 2018).
M. E. Kruk, M. Pate, The Lancet global health Commission on
high quality health systems 1 year on: Progress on a global
imperative. Lancet Glob. Health 8, e30–e32 (2020).
doi: 10.1016/S2214-109X(19)30485-1; pmid: 31839136
J. Das, J. Hammer, K. Leonard, The quality of medical advice in
low-income countries. J. Econ. Perspect. 22, 93–114 (2008).
doi: 10.1257/jep.22.2.93; pmid: 19768841
J. F. Donohue et al., Bridging the “know-do” gaps in five non-
communicable diseases using a common framework driven by
implementation science. J. Healthc. Leadersh. 15, 103–119
(2023). doi: 10.2147/JHL.S394088; pmid: 37416849
25. M. Mohanan, J. D. Goldhaber-Fiebert, S. Giardili,
M. Vera-Hernández, Providers’ knowledge of diagnosis and
treatment of tuberculosis using vignettes: Evidence from rural
Bihar, India. BMJ Glob. Health 1, e000155 (2016). doi: 10.1136/
bmjgh-2016-000155; pmid: 28588984
26. A. Banerjee, A. Chowdhury, J. Das, J. Hammer, R. Hussam,
A. Mohpal, “The market for healthcare in low income
countries,” Working paper, Harvard Business School, Boston,
MA, 23 July 2023.
27. R. J. Kovacs, M. Lagarde, J. Cairns, Can patients improve the
quality of care they receive? Experimental evidence from
Senegal. World Dev. 150, 105740 (2022). doi: 10.1016/
j.worlddev.2021.105740; pmid: 35115735
28. C. Lopez, A. Sautmann, S. Schaner, Does patient demand
contribute to the overuse of prescription drugs? Am. Econ.
J. Appl. Econ. 14, 225–260 (2022). doi: 10.1257/app.20190722
29. M. Lagarde, D. Blaauw, Overtreatment and benevolent provider
moral hazard: Evidence from South African doctors. J. Dev.
Econ. 158, 102917 (2022). doi: 10.1016/j.jdeveco.2022.102917
30. J. King, T. Powell-Jackson, J. Hargreaves, C. Makungu,
C. Goodman, Pushy patients or pushy providers? Effect of
patient knowledge on antibiotic prescribing in Tanzania.
Health Aff. 41, 911–920 (2022). doi: 10.1377/
hlthaff.2021.01782; pmid: 35666967
31. A. Kwan, C. E. Boone, G. Sulis, P. J. Gertler, Do private
providers give patients what they demand, even if it is
inappropriate? A randomised study using unannounced
standardised patients in Kenya. BMJ Open 12, e058746
(2022). doi: 10.1136/bmjopen-2021-058746; pmid: 35304401
32. R. Neupane et al., Antibiotic resistance trends for
common bacterial aetiologies of childhood diarrhoea in
low- and middle-income countries: A systematic review.
J. Glob. Health 13, 04060 (2023). doi: 10.7189/jogh.13.04060;
pmid: 37475599
33. Center For Disease Dynamics, Economics, and Policy (CDDEP),
“The state of the world’s antibiotics 2015” (CDDEP, 2015).
34. R. C. ReinerJr et al., Mapping geographical inequalities in
childhood diarrhoeal morbidity and mortality in low-income
and middle-income countries, 2000-17: Analysis for the
Global Burden of Disease Study 2017. Lancet 395, 1779–1801
(2020). doi: 10.1016/S0140-6736(20)30114-8;
pmid: 32513411
35. A. Kwan et al., Use of standardised patients for healthcare
quality research in low- and middle-income countries.
BMJ Glob. Health 4, e001669 (2019). doi: 10.1136/bmjgh-2019-
001669; pmid: 31565413
36. Z. Wagner, M. Mohanan, R. Zutshi, A. Mukherji, N. Sood, What
drives poor quality of care for child diarrhea? Experimental
evidence from India [Dataset]. Dryad (2023); https://doi.org/
10.5061/dryad.3tx95x6nk.
AC KNOWLED GME NTS
We thank S. Banerjee, S. Patil, S. Slyvia, D. Levine, R. Jain,
E. Okeke, L. Were, S. Linnemayr, F. Lam, J. Goldberg, and
attendees of the American Society of Health Economists
(ASHEcon), International Health Economics Association (IHEA),
and Academy Health conferences for helpful input on the paper.
We also thank the field team, particularly J. Krishnappa and
A. Chittedam. Funding: This research was funded by the National
Institute of Diabetes and Digestive and Kidney Diseases (principal
investigator Z.W.; grant 5R01DK126049). Author contributions:
Conceptualization: Z.W., M.M., A.M., N.S.; Methodology: Z.W.,
M.M., R.Z., A.M., N.S.; Investigation and execution: Z.W., M.M., R.Z.,
A.M., N.S.; Funding acquisition: Z.W., M.M., A.M., N.S.; Writing –
original draft: Z.W., R.Z.; Writing – review and editing: Z.W., M.M.,
R.Z., A.M., N.S. Competing interests: The authors declare
that they have no competing interests. Data and materials
availability: All study data and replication materials are available
on the Dryad repository (36). License information: Copyright © 2024
the authors, some rights reserved; exclusive licensee American
Association for the Advancement of Science. No claim to original
US government works. https://www.science.org/about/science-
licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adj9986
Materials and Methods
Figs. S1 to S8
Tables S1 to S23
References (37–42)
MDAR Reproducibility Checklist
Submitted 31 July 2023; accepted 8 November 2023
10.1126/science.adj9986
8 of 8
RES EARCH

◥ ments at field sites located in eastern Washington,

RESEARCH ARTICLE USA. We observed approximately 300 flowers
over 200 total hours (110 hours at night and
POLLINATION 90 hours in the day). This location provides an
ideal site to examine plant-pollinator interactions
Olfaction in the Anthropocene: NO3 negatively affects and the impacts of anthropogenic pollutants
(tables S1 and S2). During our observations,
floral scent and nocturnal pollination flowers were visited by diverse pollinators
(Fig. 1, A to C, and table S2), particularly nocturnal
J. K. Chan1,2†, S. Parasurama1, R. Atlas2‡, R. Xu2,3, U. A. Jongebloed2, B. Alexander2, hawkmoths and diurnal bees. Diurnal pollinators
J. M. Langenhan4, J. A. Thornton2*, J. A. Riffell1* included species of bees, flies, and butterflies.
Nocturnal and crepuscular pollinators include
There is growing concern about sensory pollutants affecting ecological communities. Anthropogenically moths [mainly hawkmoths, including Hyles lineata
enhanced oxidants [ozone (O3) and nitrate radicals (NO3)] rapidly degrade floral scents, potentially (hereafter, Hyles) and Manduca spp. (hereafter,
reducing pollinator attraction to flowers. However, the physiological and behavioral impacts on Manduca)] and Lasioglossum bees (Fig. 1D). To
pollinators and plant fitness are unknown. Using a nocturnal flower-moth system, we found that assess the contribution of the pollinator commu-
atmospherically relevant concentrations of NO3 eliminate flower visitation by moths, and the reaction of NO3 nity to O. pallida pollination, we conducted a series
with a subset of monoterpenes is what reduces the scent’s attractiveness. Global atmospheric models of pollinator-exclusion treatments, including
of floral scent oxidation reveal that pollinators in certain urban areas may have a reduced ability to perceive bagging (to prevent pollinator visits) and cross-
and navigate to flowers. These results illustrate the impact of anthropogenic pollutants on an animal’s pollinating individual flowers by hand, which
olfactory ability and indicate that such pollutants may be critical regulators of global pollination. were later assessed for fruit set. Plants in bagged
treatments had significantly fewer fruit sets than

H
uman activities have drastically changed
the environment, including introducing
stimuli detected and processed by animals’
sensory systems. Human introduction of
noise, artificial lights, or anthropogenic
chemicals—called sensory pollutants—can change
animal behavior and fitness by providing new
stimuli or modifying naturally occurring stim-
uli (1, 2). Noise pollution has been found to
negatively affect the fitness of birds, mammals,
and insects (1, 3–5), and light pollution in urban
areas has been implicated in the mortality of
migrating birds (4). By contrast, much less is
known about the effects of airborne pollutants
on animal olfactory systems and the corres-
ponding ecological effects (1, 6, 7). Recent
studies have shown that high concentrations
of diesel exhaust, or tropospheric ozone, can
affect insect odor recognition by potentially
degrading the compounds in the scent (7–13).
However, studies often do not reflect the natu-
ral spatial and temporal dynamics of atmo-
spheric processing of the odors, and there is
still a lack of understanding of how the deg-
radation of natural scents by air pollution
affects animal behavior and ecological inter-
actions [however, (14, 15)].
Plant-pollinator interactions are essential
for ecological communities and may be especially
susceptible to anthropogenic pollutants (7, 9, 10).
1
Department of Biology, University of Washington, Seattle,
WA 98195, USA. 2Department of Atmospheric Sciences,
University of Washington, Seattle, WA 98195, USA. 3Center
for Earth System Science, Tsinghua University, Beijing 100084,
China. 4Department of Chemistry, Seattle University, Seattle,
WA 98122, USA.
*Corresponding author. Email: jriffell@uw.edu (J.A.R.); joelt@uw.
edu (J.A.T.)
†Present address: National Biodiversity Future Center (NBFC),
Department of Biology, University of Naples Federico II, Naples,
Italy.
‡Present address: Laboratoire de Météorologie Dynamique/École
Polytechnique, Paris, France.
Many pollinators navigate long distances by
the scents released from flowers (16), and scent
compounds can be quickly degraded in the
atmosphere by reaction with hydroxyl radicals
(OH), nitrate radicals (NO3), and ozone (O3)
pollutants that are formed from natural sources
and anthropogenic emissions, such as vehicle
emissions (7, 11). During the daytime, O3 photo-
lysis by sunlight in the presence of water vapor
leads to OH, the primary oxidizing agent in the
atmosphere (17). However, NO3 is often the
dominant oxidant at night in polluted regions
(18). At night, NO3 is formed from the reaction
of O3 with NO2 and achieves high abundances
because of the lack of NO3 photolysis (18, 19).
Research has demonstrated that NO3 is a
dominant nighttime pollutant in the troposphere
that reacts much faster than O3 to volatiles,
including monoterpenes (20, 21). Despite the
nocturnal predominance of NO3 as a possible
oxidant of floral scent in polluted regions, we
know little about the relative effects of NO3
and O3 on pollinator olfactory behaviors and
how these oxidants could affect local and global
plant-pollinator interactions.
Oenothera pallida pollination
In the North American Deserts ecoregions,
Oenothera pallida (Fig. 1) release a strong floral
scent that attracts a rich diversity of pollinators,
including crepuscular hawkmoths, which navi-
gate many kilometers to locate patches of flowers
(22, 23). In these areas, hawkmoths and
O. pallida will experience varying levels of
anthropogenic and naturally produced O3 and
NO3, with elevated levels of nocturnal NO3
near or downwind from urban centers (24).
To understand the importance of various
nocturnal and diurnal pollinators to O. pallida,
in June and July of 2017 and 2018 we conducted
pollinator observation and exclusion experi-
those of unbagged plants (pairwise comparison
of proportions with Holm correction, P < 0.001)
(Fig. 1E and tables S3 and S4). The exclusion
of nocturnal pollinators also resulted in fewer
fruit sets than those of no-treatment controls
(P = 0.0082), which is consistent with findings
from Gregory (22) that night-blooming Oenothera
species are pollinated by hawkmoths. These
hawkmoths use their olfactory sense to navigate
over kilometer distances and locate patches of
Oenothera flowers (25). At our field site, Hyles
and Manduca were observed to visit O. pallida
flowers throughout the night (Fig. 1D and
table S2).
Atmospheric oxidation of O. pallida
floral scent
To understand how atmospheric pollution may
affect floral scents, we first characterized the
O. pallida floral scent and identified the prin-
cipal bioactive compounds for attracting polli-
nators, especially the hawkmoths (Hyles and
Manduca) (22). Floral scents were collected in
the field by using headspace traps that allowed
the collection of the floral scent compounds.
The scent samples were then analyzed by use
of gas chromatography (GC) with mass spectrom-
etry (fig. S1 and tables S5 and S6), allowing
identification of the compounds in the scent.
To identify volatile compounds that pollinators
might use to detect the flowers, we performed
gas chromatography coupled with electro-
antennographic detection (GC-EAD) using
Megachile bees and male Hyles and Manduca
moths (Fig. 2A and fig. S2). All pollinators were
sensitive to many of the same compounds in
the scent, including monoterpenes such as
cis-b-ocimene and b-pinene. In particular, the
hawkmoths had similar antennal responses and
were especially sensitive to monoterpenes (cis-
b-ocimene and b-pinene) (Fig. 2A). We created a
floral odor composed of moth antennal-active

Chan et al., Science 383, 607–611 (2024) 9 February 2024 1 of 5

RES EARCH | R E S E A R C H A R T I C L E

A B C

a
D E
0.50 Adrenidae 0.8 a,b
Crabronidae a,b
Apoidea (Lassioglossum)
Apidae
Other Hymenoptera
Megachilidae b
Proportion visitation

Diptera Proportion fruit set

Other Insecta

0.25 Papilionidae 0.4

Sphingidae (Manduca, Hyles)
Microlepidoptera
Other Lepidoptera

0 0
Diurnal Crepuscular/Nocturnal Unbagged Crossed Pollinator Nocturnal Diurnal
visitors visitors exclusion pollinator pollinator
exclusion exclusion

Fig. 1. O. pallida pollinator assemblage at near-pristine sites in eastern
Washington, USA. (A) Example of O. pallida habitat in sandy areas in sagebrush
steppe of the Columbia Plateau (Echo basin, Washington, USA). (B and C) Major
pollinators of O. pallida, including (B) Hyles moths and (C) Megachile bee species.
(D) Diurnal and nocturnal pollinators visiting and pollinating O. pallida. There
was a significant difference in pollinator assemblages (Z = 17.67, P < 0.0001).
(E) Pollinator exclusion treatments. Treatments included unbagged
(no-treatment control flowers), crossed (bagged flowers, total exclusion of
pollinators with manual cross-pollination between individual plants), pollinator
exclusion (bagged flowers, total pollinator exclusion treatment), nocturnal
pollinator exclusion (nocturnal and crepuscular pollinators excluded), and
diurnal pollinator exclusion (diurnal pollinators excluded). Bars are the mean
± SEM. Letters denote groups whose members are not statistically different
from each other.

compounds (table S6), and with an emission rate
similar to that of the O. pallida flower, for use in
subsequent laboratory experiments (tables S6 to S8).
To examine the relative atmospheric degra-
dation of floral volatile organic compounds
(VOCs) by NO3 and O3 (Fig. 2B), we used an
atmospheric pressure flow reactor coupled
to a time-of-flight mass spectrometer (LToF;
TOFWERK AG, Thun, Switzerland) with chem-
ical ionization by benzene cations (26) and
proton-transfer reactions (Vocus PTR; TOFWERK
AG, Thun, Switzerland) (27) to measure the
concentrations of floral volatiles in real time
(Fig. 2C) (28). This system allowed us to mea-
sure the degradation of the floral volatiles under
realistic atmospheric conditions and timescales
and to scale our measurements to a variety of
different conditions and environments (fig. S3).
Exposure of the floral scent to both O3 and
NO2 [120 parts per billion (ppb) and 60 ppb,
respectively]—corresponding to urban environ-
ments and downwind from urban areas (29, 30),
and leading to the presence of NO3 and N2O5
(fig. S4)—decreased concentrations of certain
constituents in the scent, particularly certain
monoterpenes. By contrast, other scent com-
pounds, such as 2-methyl butanal oxime, showed
little change in concentration (fig. S5). These
effects were dose-dependent: Increasing or
decreasing the NO3 exposure time or concen-
tration caused corresponding changes in the
processing of the monoterpenes. Reaction prod-
ucts of the monoterpenes with O3 and NO3 were
also identified in the scent (fig. S6). Experiments
that tested the relative oxidation by NO3 and O3
on the individual compounds alone showed
similar results: Monoterpenes (such as b-pinene
and cis-b-ocimene) were sensitive to the pollu-
tants. However, whereas the monoterpenes were
partially oxidized by O3 (decreased by ~30%),
these compounds were severely degraded in
the presence of NO3 (decreased by 84 and 67%,
respectively; P < 0.001, Welch’s t test/Mann-
Whitney U-test) (Fig. 2D and table S9), con-
firming previous work (20, 21) and emphasizing
the role of NO3 in the atmospheric oxidation of
floral scents.
NO3 suppresses plant-pollinator visitations
We next conducted laboratory and field experi-
ments to determine how atmospheric oxida-
tion affects hawkmoths’ ability to locate
O. pallida scent sources and flower visitation.
We performed laboratory wind tunnel experi-
ments, which enabled reproducible simulation
of the physicochemical conditions present in
the field and determination of the impacts of

Chan et al., Science 383, 607–611 (2024) 9 February 2024 2 of 5

RES EARCH | R E S E A R C H A R T I C L E

Fig. 2. Sensitivity of floral odor to degradation by A eucal. -ocimene

free radicals. (A) GC-EAD traces of male (top) Hyles Oenothera pallida -p
-pinene
lineata and (bottom) Manduca sexta antennal aldoxime GC-FID
benz.
response to O. pallida night scent sample from the
field. Responses to the monoterpenes b-pinene and
cis-b-ocimene, 2-methylbutanal oxime (aldoxime),
Manduca sexta EADs
benzaldehyde (benz.), and eucalyptol (euca.) are
highlighted in gray. (B) Chemical equations for
the formation of the nitrate radical (NO3) from Hyles lineata 0.5 mV
nitrogen dioxide (NO2) and ozone (O3). Dinitrogen 1 min
pentoxide (N2O5) forms reversibly from NO3 and NO2
and acts as a reservoir of NO3. During the day, B C Flow tube
Floral VOCs (fVOCs)
ultraviolet light causes the dissociation of NO3 to NO2 + O3 NO3 + O2 degraded PTR-ToF-MS /
NO3 +fVOC
NO2 and O, preventing the buildup of NO3. NO3 fVOCs CI-ToF-MS
reaction with floral volatile organic compounds
NO3 + NO 2 NO2
(fVOCs) rapidly yields reaction products that are NO3 + NO2 N2O5
not detected by the moths. (C) Schematic of the
NO3 + fVOC Products
setup for generating NO3 from NO2 and O3 and hv O3 generator
oxidizing the fVOCs in a flowtube. (D) Example traces NO3 NO2 + O N2/NO2
of O3 followed by NOx oxidation of 2-methylbutanal
oxime (orange line), b-pinene (blue line), and D
cis-b-ocimene (green line), measured with a Vocus- aldoxime
1
PTR-TOF mass spectrometer. (Inset) Table of
Ion count (normalized)

% change % change
measured volatiles and their degradation rates under fVOCs with O3 with NO3
0.8
aldoxime No change No change
NOx and O3 oxidation (all differences P < 0.001, benzaldehyde -9 -12
Welch’s t test/Mann-Whitney U test) (table S9). The 0.6
-ocimene -30 -67
eucalyptol +4 +2
O3 and NO2 concentrations in the flowtube were -pinene -33 -84
120 and 60 ppb, respectively, which correspond to the 0.4 -ocimene
upper range of highly polluted urban environments -pinene
(29, 30). The reaction time in the flowtube was 73 s, 0.2 O3 NO3
which simulates the impacts on odor transmission 0 100
within 50 to 100 m from the odor source. Time (sec.)

Fig. 3. The impacts of NO3 on moth visitation to A B

flowers. (A) Diagram of experimental setup and Wind 3D tracking
cameras
Hyles lineata a Manduca sexta a
hawkmoth flight path in the wind tunnel upwind
1 a
toward the odor source. (B) Wind tunnel behavioral
Proportion floral visits

results for Hyles and Manduca at 0.5 m/s laminar b

airflow. Individual male hawkmoths were released 2 m b b
Flo
w 0.5
directly downwind of the odor source, and the tub
e
1m

proportion of moths that attempted feeding from the

source was recorded. NO3 oxidation conditions are
equilibrium N2O5 from 120 ppb O3 and 60 ppb NO2 at 0 0 0
Floral NO3 No odor Floral O NO3 NO3 No odor NO2
room temperature with 73 s reaction time in a glass O3 NO3 odor oxid. ctl. odor oxid. oxid. mix. ctl. ctl.
3

flowtube. O3 and NO2 oxidation conditions include just

the O3 or NO2 component of the NO3 treatment. C D
Controls were done with dry, filtered air. Another odor 0.4 a
a
treatment of NO3-proxy mixture (NO3 mix.) was
Pollinator flower visits (/h)

a
performed with a synthetic floral odor containing 84%
less b-pinene and 67% less b-ocimene than the
original floral odor to simulate oxidation by NO3. 0.2
Bars are the mean ± SEM (n = 15 to 35 moths per b
b
treatment). (C) Photograph of the field site. (D) Field
behavioral results showing the hourly nocturnal visits
to a single scent source of each treatment. Five
0
treatments were performed simultaneously, including Floral NO3 oxid. No odor Live flower Live
odor floral odor ctl. scent flower
a real flower treatment, real floral scent treatment,
floral odor treatment, NO3 oxidized floral odor
treatment, and clean air control. All treatments except the real flower involved air from the scent treatment delivered through a humidified filter paper cone. We
performed 43 to 152.5 observation hours for each treatment (total of approximately 344 hours). Bars are the mean ± SEM.

Chan et al., Science 383, 607–611 (2024) 9 February 2024 3 of 5

RES EARCH | R E S E A R C H A R T I C L E

NO3 and O3 levels found in either the pol-
luted urban or near-pristine environments
(Fig. 3A). For Hyles, NO3 oxidation of the floral
odor eliminated their behavioral attraction
(Fig. 3B). For Manduca, which was more sen-
sitive to the floral odor (fig. S7), NO3 oxidation
resulted in a 50% decrease in Manduca visi-
tation rate (P = 0.047, comparison of popula-
tion proportions) and to a level that was not
significantly different from that of the solvent
(no flower odor) control. By contrast, O3 oxidation
of the floral odor—at O3 concentrations typical of
highly polluted environments (120 ppb)—had no
impact on hawkmoth visitation (Fig. 3B).
Our scent oxidation experiments showed that
a subset of monoterpenes in the floral odor were
degraded with NO3 exposure. However, the de-
cline of hawkmoth attraction could be due to the
decreased monoterpene concentrations or the
moth’s perception of the oxidation products
from the odor reacting with NO3. Examination
of the oxidation products from our flowtube
experiments showed the production of organic
nitrates associated with the odor compounds
(fig. S6). We created a synthetic mixture of
relationship between hawkmoth visitation and
the resulting fruit set. From our experiments,
an untreated O. pallida flower was visited by a
hawkmoth approximately twice per night (1.9 ±
0.9), whereas flower visitation in the oxidized
scent treatment fell to 0.57 (±0.28) visits per
night, or a 70% (±20%) drop in visitation. In our
experiments, hawkmoths and crepuscular bees
account for 40% (±10%) of all fruits (table S3);
thus, a 70% drop in visitation will cause a 28%
(±11%) reduction in the total fruit set. Although
these results do not consider the effects of
diurnal pollinators—such as bees, some of
which are affected by the degradation of flower
compounds by high-concentration pollutants
such as diesel exhaust or O3 (7, 9, 10, 14, 31)—
they do illustrate the potential impact of field-
relevant concentrations of NO3 and O3 on
plant-pollinator interactions.
A B
distance (km)
Global modeling of pollutant impacts
Our research and work with other pollinators
such as bees (7, 11) show that oxidation of
specific monoterpenes in the floral odors
causes declines in attraction. NO3 dominates the
nighttime oxidation of biogenic VOCs, particu-
larly near and downwind of urban areas where
nitrogen dioxide and O3 are elevated. To estimate
the potential impacts of anthropogenic enhance-
ments to O3 and NO3 concentrations on pollinator
olfactory navigation, we used global distributions
of O3 and NO3 concentrations simulated by the
GEOS-Chem global chemical transport model
(25, 32). The GEOS-Chem model—which couples
meteorology (for example, temperature, three-
dimensional wind fields, precipitation, and bound-
ary layer heights) (33) with chemical emissions
and mechanisms—allows simulation of atmo-
spheric composition at local to global scales

distance (km)
O. pallida scent compounds to simulate the
recognition

recognition
>5.0 >5.0
Pollinator

Pollinator
selective depletion of monoterpenes by NO3
(called the NO3-proxy mixture). The NO3-proxy
mixture (with 84% less b-pinene and 67% less 0 0

b-ocimene, but lacking the oxidation products) With O3 chemistry With NO3 chemistry
-150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150
elicited significantly fewer responses than did longitude longitude
the untreated flower odor and the same amount
of responses as those with the clean air (no odor)
control and the NO3-degraded scent that con- C
tained the oxidation products. Controls of the
floral odor exposed to nitrogen dioxide (NO2;
the nonreactive precursor to NO3) or O3 alone
(to control for any physiological effects on the
moth) were not significantly different from
responses to the untreated floral odor (Fig. 3B).
Taken together, our results show that the moth’s
inability to navigate and recognize the flower is 100
of pre-industrial

from NO3 selectively degrading a subset of com- 80

Current %

pounds in the scent and not because of the moth 60

perception of the oxidation products.
To determine how oxidation of floral scents
may affect plant-pollinator interactions in the
field, we conducted experiments to test various
scent oxidation treatments in our Grants Pass,
Washington, site (Fig. 3C). Treatments included
real flowers, artificial flowers emitting the floral
untreated or NO3-degraded scent, or artificial
flowers not emitting a scent (visual control) (28).
Visitation rates to the floral odor—both real
floral odor (one flower) and our synthetic floral
odor—and to real flowers were not signifi-
cantly different (Fig. 3D). By contrast, the
visitation rate to floral odor exposed to NO3
was significantly less than to the untreated
odor (P = 0.027, generalized linear model,
Poisson, logistic link) (tables S10 and S11) and
not significantly different from the clean air
control (Fig. 3D). To establish the effects of
scent oxidation on fruit set, we determined the
40
20
0
Fig. 4. Global impacts of O3 and NO3 on pollinator recognition. (A) Map of floral scent recognition distance
by using O3 degradation of the volatiles b-pinene and cis-b-ocimene, with degradation thresholds of 84 and 67%,
respectively, and horizontal wind speed from the bottom grid of the GEOS-Chem model. The NO3 and O3
distributions were generated by using GEOS-Chem standard 12.1.0 with the 2013 emissions inventory and 2013
meteorology with a 2°-by-2.5° grid and 72 vertical levels to 0.01 hPa for the monthly average of January 2013
to February 2014. The bottom vertical level and the average concentrations for July 2013 (northern latitudes)
and January 2013 (southern latitudes)—summer periods when the pollinators were present—were used. (B) Map
of floral scent–recognition distance by using NO3 degradation of the volatiles b-pinene and cis-b-ocimene with
the same conditions as in (A), and with data from the same GEOS-Chem model run. (C) Map of 2013 floral scent–
recognition distance from (B) divided by preindustrial floral scent–recognition distance as a percentage. The
preindustrial NO3 and O3 distributions were generated by using GEOS-Chem 13.2.1 classic with the 2013
nonanthropogenic emissions inventory and 2013 meteorology with a 4°-by-5° grid and 72 vertical levels to
0.01 hPa for January 2013 to December 2013. The bottom vertical level and the average concentrations for
July 2013 (northern latitudes) and January 2013 (southern latitudes) were used. The outputs for the plots in (A) to
(C) were masked by using a land-ocean mask to remove the values over bodies of water.

Chan et al., Science 383, 607–611 (2024) 9 February 2024 4 of 5

RES EARCH | R E S E A R C H A R T I C L E
and has been evaluated against observations in
several studies (28). Using known rate constants
for reactions of monoterpenes with O3 and NO3,
we calculated the distance for floral scents to be
oxidized to a level unrecognizable by hawkmoths,
given the simulated concentrations of O3 and
NO3 and wind velocities within the lowest grid
level in the GEOS-Chem model (34). A scent-
recognition distance was computed for each lo-
cation and for the respective summer months
(July for the Northern Hemisphere and Janu-
ary for the Southern Hemisphere) and plotted
as a series of global maps that illustrate the
differing impacts of NO3 and O3 on pollinator
perception of the floral scents (Fig. 4).
Results from the model show that NO3 deg-
radation of monoterpenes in the floral scent
has a more severe impact on recognition dis-
tance than that by O3 alone (Fig. 4, A and B)
and that scent-recognition distances are reduced
to below 400 m in many populated areas.
Regions with the most severe impacts from
NO3 include North America, Europe, Central
Asia, the Middle East, and southern Africa.
In addition, we performed a simulation of the
preindustrial atmosphere using GEOS-Chem
to assess the percent change in scent-recognition
distance that has occurred since the pre-
industrial era (Fig. 4C). The comparison map
shows that in most populated regions of the
world, there has been a 75% or more decrease in
scent-recognition distances since preindustrial
times (Fig. 4C). In certain sparsely populated
areas (such as Greenland), NO3-related changes
are relatively small and do not affect the scent-
recognition distances. In other areas (parts of
Southern Africa), scent-recognition distances
may be unchanged relative to preindustrial
times because of both natural O3 and NOx emis-
sions and chemical feedbacks that stem from
their couplings with arboreal VOC emissions
(fig. S9 and table S13) (19). Geographic areas
may thus differ widely in the impact of NO3
on pollinator recognition of floral scents. Over
the past 10 years, annual variation in NO3 may
occur, especially in regions with biomass burn-
ing or other meteorological effects. However,
when comparing across years (2013, 2019, and
2021), the model showed similar global trends in
O3 and NO3 concentrations and distributions,
with level impacts orders of magnitude greater
when compared with preindustrial conditions
(table S13) (30), which is consistent with anthro-
pogenic inputs of NO3 around urban areas.
Our results demonstrate that atmospheric O3
and NO3 oxidation affects nocturnal pollinator
visitations in the field by changing floral scent
chemical composition, reducing scent-recognition
distances. As a further example of the impacts
of NO3 on many common floral volatiles, we
compared the oxidation rates of diverse com-
pounds with NO3 and O3 under mean environ-
mental conditions of the northern latitudes
(fig. S8B and table S12). Building on past atmo-
Chan et al., Science 383, 607–611 (2024) 9 February 2024
spheric chemistry work on the effects of NO3
(20, 21), our results show that most floral com-
pounds had significantly greater reactivity
toward typical NO3 concentrations than O3,
except for the sesquiterpenes a-humulene and
b-caryophyllene (fig. S8B and table S12). The
monoterpenes were the most susceptible to O3
and NO3 degradation of the floral scent com-
pounds analyzed, with certain monoterpenes
being more reactive than others. Many studies
(7, 11, 35–37) have established that monoter-
penes are ubiquitous floral volatiles important
for scent recognition by pollinators; our results
demonstrate that certain subsets of compounds
in the scent are more sensitive to anthropogenic
pollutants and generalize to other systems that
use NO3-reactive monoterpenes as key volatiles
for scent recognition. Although other classes
of volatiles such as sesquiterpenes, green leaf
volatiles, and aromatics have different reac-
tivity profiles, our analytical framework can
be used to estimate the impacts of O3 and NO3
on scent-recognition distances in other systems
if the relevant volatiles and rate constants can be
determined.
Olfaction and chemical signaling mediate
diverse ecological and evolutionary processes,
including predator-prey interactions, host selec-
tion, and mate selection (38, 39). At the popula-
tion level, our results indicate that nitrate radicals,
stemming from nitrogen oxide pollution, nega-
tively affect both plants (by decreasing fitness)
and insect pollinators (by decreasing their abil-
ity to locate nectar resources) and in a region-
specific manner. Future work is needed to
determine the community-level response to
anthropogenic pollutants and to identify how
different ecological processes are affected.
Anthropogenic pollutants are temporally and
regionally variable, and it will be necessary to
characterize these impacts in different geo-
graphic locations to understand and ultimate-
ly mitigate these effects.
RE FERENCES AND NOTES
1. D. M. Dominoni et al., Nat. Ecol. Evol. 4, 502–511 (2020).
2. W. Halfwerk, H. Slabbekoorn, Biol. Lett. 11, 20141051 (2015).
3. B. M. Siemers, A. Schaub, Proc. Biol. Sci. 278, 1646–1652 (2011).
4. B. M. Van Doren et al., Proc. Natl. Acad. Sci. U.S.A. 114,
11175–11180 (2017).
5. A. C. Owens et al., Biol. Conserv. 241, 108259 (2020).
6. E. Agathokleous, Z. Feng, J. Peñuelas, Trends Ecol. Evol. 37, 939–941 (2022).
7. R. D. Girling, I. Lusebrink, E. Farthing, T. A. Newman, G. M. Poppy,
Sci. Rep. 3, 2779 (2013).
8. B. Cook et al., J. Chem. Ecol. 46, 987–996 (2020).
9. F. Démares, L. Gibert, P. Creusot, B. Lapeyre, M. Proffit,
Sci. Total Environ. 827, 154342 (2022).
10. S. Dötterl, M. Vater, T. Rupp, A. Held, J. Chem. Ecol. 42, 486–489 (2016).
11. G. Farré-Armengol et al., New Phytol. 209, 152–160 (2016).
12. J. D. Fuentes, M. Chamecki, T. Roulston, B. Chen, K. R. Pratt,
Atmos. Environ. 141, 361–374 (2016).
13. Q. S. McFrederick, J. C. Kathilankal, J. D. Fuentes, Atmos. Environ.
42, 2336–2348 (2008).
14. J. M. W. Ryalls et al., Environ. Pollut. 297, 118847 (2022).
15. J. M. W. Ryalls et al., Proc. Biol. Sci. 289, 20221692 (2022).
16. J. A. Riffell, L. Abrell, J. G. Hildebrand, J. Chem. Ecol. 34,
837–853 (2008).
17. D. H. Ehhalt, Phys. Chem. Chem. Phys. 1, 5401–5408 (1999).
18. N. L. Ng et al., Atmos. Chem. Phys. 17, 2103–2162 (2017).
19. R. A. Graham, H. S. Johnston, J. Phys. Chem. 82, 254–268 (1978).
20. R. Atkinson, S. M. Aschmann, A. M. Winer, J. N. Pitts,
Environ. Sci. Technol. 18, 370–375 (1984).
21. R. Atkinson, S. M. Aschmann, A. M. Winer, J. N. Pitts,
Environ. Sci. Technol. 19, 159–163 (1985).
22. D. P. Gregory, Aliso 5, 385–419 (1964).
23. R. E. Stockhouse, Am. Midl. Nat. 96, 241–245 (1976).
24. H. Simon, A. Reff, B. Wells, J. Xing, N. Frank, Environ. Sci. Technol.
49, 186–195 (2015).
25. W. A. Haber, G. W. Frankie, Biotropica 21, 155–172 (1989).
26. A. Lavi, M. P. Vermeuel, G. A. Novak, T. H. Bertram,
Atmos. Meas. Tech. 11, 3251–3262 (2018).
27. J. Krechmer et al., Anal. Chem. 90, 12011–12018 (2018).
28. Materials and methods are available as supplementary materials.
29. M. Huang et al., Atmos. Chem. Phys. 17, 5721–5750 (2017).
30. M. Khan et al., Atmos. Res. 164-165, 347–357 (2015).
31. M. Vanderplanck et al., Antioxidants 10, 636 (2021).
32. D. A. Satterfield, T. S. Sillett, J. W. Chapman, S. Altizer,
P. P. Marra, Front. Ecol. Environ. 18, 335–344 (2020).
33. R. Gelaro et al., J. Clim. 30, 5419–5454 (2017).
34. K. R. Mylne, Boundary-Layer Meteorol. 60, 15–48 (1992).
35. J. T. Knudsen, R. Eriksson, J. Gershenzon, B. Ståhl, Bot. Rev.
72, 1–120 (2006).
36. G. Farré-Armengol, M. Fernández-Martínez, I. Filella,
R. R. Junker, J. Peñuelas, Front. Plant Sci. 11, 1154 (2020).
37. J. A. Riffell, H. Lei, L. Abrell, J. G. Hildebrand, Science 339,
200–204 (2013).
38. T. Eisner, J. Meinwald, Proc. Natl. Acad. Sci. U.S.A. 92, 1 (1995).
39. B. S. Hansson, M. C. Stensmyr, Neuron 72, 698–711 (2011).
40. J. A. Riffell, Olfaction in the Anthropocene: NO3 negatively impacts
floral scent and nocturnal pollination [Dataset]. Dryad (2024);
https://doi.org/10.5061/dryad.vt4b8gtzk.
41. J. A. Riffell, riffelllab/Chan-et-al: Chan et al - Olfaction in
the Anthropocene (version v1). Zenodo (2024);
https://doi.org/10.5281/zenodo.8411927.
42. The International GEOS-Chem User Community, geoschem/
geos-chem: GEOS-Chem 12.1.0 (12.1.0). Zenodo (2018);
https://doi.org/10.5281/zenodo.1553349.
43. The International GEOS-Chem User Community, geoschem/
GCClassic: GEOS-Chem 13.2.1 (13.2.1). Zenodo (2021); https://
doi.org/10.5281/zenodo.5500717.
AC KNOWLED GME NTS
The authors thank T. Daniel and J. Hille Ris Lambers for technical advice
and assistance, B. Nguyen for help in the moth insectary, E. Sammeth
and R. Mettey for help with the insectary and GC-EAD, and
R. Wolf for permission to use his image in Fig. 1B. We thank
G. Davidowitz and C. Francois for providing the Hyles lineata and the
Washington Department of Fish and Wildlife for access to the field sites.
Funding: This work was supported by Air Force Office of Scientific
Research grants FA9550-21-1-0101, FA9550-20-1-0422, and AWD-
004055-G4 a.m.01 (J.A.R.); the National Science Foundation under
grants 2121935 (J.A.R.) and 2202287 (B.A.); the National Institutes of
Health under grant R01AI148300 (J.A.R.); the Human Frontier Science
Program under grant RGP0044/2021 (J.A.R.); the Danish National
Research Foundation under grant DNRF168 (J.K.C.); an Endowed
Professorship for Excellence in Biology (J.A.R.); and endowments
from L. Riddiford, J. Truman, R. T. Paine, J. S. Edwards, and B. Hall
(J.K.C.). Author contributions: Conceptualization: J.K.C.,
J.A.T., and J.A.R. Methodology: J.K.C., R.A., S.P., R.X., U.A.J., B.A.,
J.M.L., J.A.T., and J.A.R. Investigation: J.K.C., R.A., S.P., R.X., U.A.J., B.A.,
J.M.L., J.A.T., and J.A.R. Visualization: J.K.C., R.A., J.A.T., and J.A.R.
Funding acquisition: J.K.C., R.A., J.A.T., and J.A.R. Project administration:
J.A.T. and J.A.R. Supervision: J.K.C., J.A.T., and J.A.R. Writing – original
draft: J.K.C., J.A.T., and J.A.R. Writing – review and editing: J.K.C., R.A.,
S.P., R.X., U.A.J., B.A., J.M.L., J.A.T., and J.A.R. Competing interests:
The authors declare that they have no competing interests. Data and
materials availability: Data are available at (28) and (40), and custom
code is available at (41). GEOS-Chem code is available at (42) and
(43). License information: Copyright © 2024 the authors, some rights
reserved; exclusive licensee American Association for the Advancement
of Science. No claim to original US government works. https://www.
science.org/about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adi0858
Materials and Methods
Equation S1
Figs. S1 to S9
Tables S1 to S13
References (44–79)
MDAR Reproducibility Checklist
Submitted 4 April 2023; accepted 4 January 2024
10.1126/science.adi0858
5 of 5
RES EARCH

MASS SPECTROMETRY excitation for analysis of the enantiomeric ions.

The ions were trapped in the LIT by an RF field,
Differentiating enantiomers by directional rotation which set the E/N value above 1 MTd. Two ac
resonance excitations were applied in both
of ions in a mass spectrometer x and y directions, which induced the com-
plex motions of the trapped ions (Fig. 1B and
Xiaoyu Zhou1, Zhuofan Wang1, Shuai Li1, Xianle Rong2, Jiexun Bu3, Qiang Liu2, Zheng Ouyang1,4* fig. S1). An excited ion could move around the
center of the ion trap (macro motion) while
Conventional mass spectrometry does not distinguish between enantiomers, or mirror-image isomers. rotating around its center of mass. The ions of
Here we report a technique to break the chiral symmetry and to differentiate enantiomers by inducing both enantiomers had similar macro motions
directional rotation of chiral gas-phase ions. Dual alternating current excitations were applied to and rotational motions (Fig. 1B and fig. S2).
manipulate the motions of trapped ions, including the rotation around the center of mass and macro Simulations were carried out using a pair of
movement around the center of the trap. Differences in collision cross section were induced, which enantiomers of mass to charge ratio (m/z) 551
could be measured by ion cloud profiling at high resolutions above 10,000. High-field ion mobility and (see supplementary materials for detailed infor-
tandem mass spectrometry analyses of the enantiomers were combined and implemented by using mation for the setup of initial conditions and
a miniature ion trap mass spectrometer. The effectiveness of the developed method was demonstrated the induced differences in collisional cross sec-
with a variety of organic compounds including amino acids, sugars, and several drug molecules, as tion). The rotating electric field, ER-motion, drove
well as a proof-of-principle ligand optimization study for asymmetric hydrogenation. the rotations of the ions (Fig. 1C), and the
rotational directions of the ions could be tuned

B
iological systems normally favor one of
the two mirror-image forms, or enantio-
mers, of the amino acids, sugars, and
nucleotides, respectively, present in pro-
teins, carbohydrates, DNA, and RNA
(1–4). Specifically, chiral amino acids mostly
exist as L-enantiomers, whereas carbohydrates
are mainly D-enantiomers (5). Owing to the
intrinsic chiral environment of biological sys-
tems, enantiomeric drug molecules often show
different physiological behaviors and pharma-
cological activities (6, 7). Using the therapeutic
drug thalidomide as an example, the R-enantiomer
is sedative, whereas the S-enantiomer is terato-
genic (8, 9) and its misuse led to birth defects
in more than 10,000 newborns (10). Differen-
tiation of enantiomers is crucial for drug devel-
opment as well as for disease diagnosis (4, 11).
Analysis of enantiomers can be achieved
based on their differential interaction with
electromagnetic fields, as manifested in opti-
cal rotation of polarization, circular dichroism
(12, 13), or anomalous dispersion in x-ray crys-
tallography (14, 15); or through interactions
with chiral selectors, such as the stationary
phase in chromatography (16). Modern mass
spectrometry (MS) systems serve as a powerful
platform for high-throughput analysis of com-
plex mixtures with high structural specificity.
Although determination of enantiomeric excess
is often done by coupling MS with gas chro-
matography or high-performance liquid chro-
matography (HPLC) with chiral stationary
phases, direct differentiation of enantiomers
by MS has remained as a major challenge. Ion
mobility MS (IMMS) could be used to perform
1
State Key Laboratory of Precision Measurement Technology
and Instruments, Department of Precision Instrument,
Tsinghua University, Beijing 100084, China. 2Department of
Chemistry, Tsinghua University, Beijing 100084, China.
3
PURSPEC Technology (Beijing) Ltd., Beijing 100084, China.
4
Tsinghua Shenzhen International Graduate School, Tsinghua
University, Shenzhen 518055, China.
*Corresponding author. Email: ouyang@tsinghua.edu.cn
chiral analysis, whereas chiral references need
to be introduced either for chemical interac-
tions prior to IMMS analysis (17–20) or for gas-
phase collisions during tandem IMMS analysis
(17, 18, 21–25).
Recently, we developed an IMMS method
simply by using an ion trap for both high-field
ion mobility separation and mass analysis (26).
Ion cloud profiling is performed under high
E/N conditions [1 MTd; Td: Townsend number;
E, the field strength of the radiofrequency (RF)
field; N, the number density of the background
gas], which led to super high resolutions above
10,000 (see supplementary materials for defi-
nition) for identifying the differences in mo-
lecular structures using high-field ion mobility.
The motivation for this work was to explore
the possibility of directly distinguishing enan-
tiomers using this high-resolution IM method
with small differences in collision cross sec-
tions that could be induced by electromagnetic
fields inside the ion trap. It has been demon-
strated that differentiation effects could be
induced by electromagnetic fields, often based
on the polarity differences of the enantiomers
(27–29). Mirrored propeller configurations have
been used to illustrate the differences that
could be induced for enantiomers owing to
the directional rotations (29–31) (Fig. 1A).
Hypothetically, manipulating the rotation
directions using electromagnetic fields should
be more effective for ionic than neutral states
of the chiral compounds. In this work, we dis-
covered differences in collision cross sections
induced for enantiomeric ions in the ion cloud
profiling process and subsequently developed
a method to control the directional rotations
of the ions using dual alternating current (ac)
excitation, which enabled a direct separation
and analysis of the enantiomers.
A Mini b miniature mass spectrometer
[PURSPEC Technology (Beijing) Ltd., Beijing,
China] was modified with a dual-LIT (linear
ion trap) configuration (32) and a dual-ac
by varying the phase difference of the ac exci-
tations, Dϕ ¼ ϕy  ϕx , where ϕx and ϕy repre-
sent the phases of the ac applied in the x and
y directions of the trap, respectively. For in-
stance, for Dϕ ¼ T p=2, R- and S-enantiomer
ions had opposite rotational motions. The dif-
ferences in the directional rotations resulted
in differences in the drag force in the macro
motion of the ions owing to the collisions
with the background gas molecules. This led
to differences in the amplitude of the macro
motion (Fig. 1C) and the possible differenti-
ation of the trapped enantiomeric ions. Through
the scan of the ac voltage, the amplitudes of
the enantiomeric ions increased accordingly
and the ions were ejected sequentially from
the trap (Fig. 1D).
Scope and sensitivity of enantiomer
differentiation
To experimentally test the directional rota-
tion effect described above for differentiating
enantiomeric ions, we first selected a pair
of enantiomers, (R)-(-)-1,1′-Bi-2-naphthol bis
(trifluoromethanesulfonate) and (S)-(+)-1,1′-
Bi-2-naphthol bis(trifluoromethanesulfonate),
abbreviated as R- and S-binaphthyl-triflate
respectively, for verification of the hypothe-
sis and predictions by simulations (Fig. 2A).
Binaphthyl-triflates have been previously se-
lected as a model system in testing the propeller
concept for enantiomer separation (29–31). The
two naphthyl groups of relatively large cross
sections serve as two blades of the propeller
and potentially could help to enhance the dif-
ference in the drag force effect once direc-
tional rotations are established. Various types
of derivatizations do not necessarily compli-
cate the simulation described above, where
the experimental measured differences were
used to calibrate the damping coefficients for
the simulations (see supplementary materials
for detailed information). The simulated dual-ac
excitation and the macro-motion amplitudes
are shown in Fig. 2, B and C, respectively. The

Zhou et al., Science 383, 612–618 (2024) 9 February 2024 1 of 7

RES EARCH | R E S E A R C H A R T I C L E

A Distinguishable? C =- / = /
ER-motion ER-motion
15E4 15E4
R R
S S

Ey [V/m]

Ey [V/m]
No 0 0

Mirrored motions - 15E4 - 15E4

-15E4 0 15E4 -15E4 0 15E4
Ex [V/m] Ex [V/m]
Macro motion Macro motion
1.0 1.0
R R
S S
Yes

y / r0
y / r0
0.0 0.0

Symmetry-breaking motions
-1.0 - 1.0
-1.0 0.0 1.0 -1.0 0.0 1.0
x / r0 x / r0

B D
R-motion =- / Ejection
1.00
Ion Trapping Boundary
R
Macro motion S
r / r0

0.50
y

x
0.00
0.0 2.5 5.0 4.184 4.188
-300 300 Time [ ms]
Enantiomer [V]

Fig. 1. The principle, instrumental setup, and simulations for macro
motions and directional rotations of enantiomeric ions trapped in LIT.
(A) Schematics of mirrored (upper) and symmetry-breaking (lower) rotational
motions. (B) Schematic of the macro motions with rotational (R-) motions of a
pair of enantiomeric ions in a RF trapping field. Two additional ac excitations
were applied in the x and y directions to manipulate the directional rotations.
(C) Simulations of the rotating E field that drove the rotations (upper),
which resulted in a detectable difference in amplitude of macro motion of the
enantiomeric ions (lower). The phase differences, Dϕ ¼ p=2 (left) and p=2
(right), of the ac applied in the x and y directions enabled tunable directional
rotations; the rotating R- and S-enantiomers had different collision cross sections
and experienced different drag forces through their macro motions, resulting
in differences in amplitudes. (D) Macro-motion amplitudes of the simulated
ion trajectories during the ion cloud profiling process with ac amplitude scan.
Enantiomeric ions were ejected sequentially, depending on their damping
coefficients, when their oscillation amplitudes exceeded the trap boundary,
indicated by the dashed line. The simulated R- and S-enantiomers of m/z
551 were set with identical initial conditions of ion motion but different
reduced damping coefficient: b′ ¼ 3.230 × 10−4 for the R-form and b′ ¼ 4.560 ×
10−4 for the S-form. b′ ¼ 2b=Wm, where W is the angular frequency of the
RF field, m is ion mass, and b is the damping coefficient of the R- or
S-enantiomers.

motion patterns of the ions could be tuned by
varying the phase difference Dϕ of the dual-ac
signals. It was clear that for Dϕ < 0, the macro-
motion amplitudes of S-binaphthyl-triflate ions
were greater than those of the R-enantiomers,
and the reverse was true forDϕ > 0. The dual-ac
excitation with Dϕ ¼ T p=2 should produce the
strongest effective rotating E fields and generate
the largest directional rotation effect for ion
motion separation (Fig. 2D and fig. S3). Ex-
perimental measurements performed accordingly
by ion cloud profiling of protonated ions of
m/z 551, produced by nano-ESI (electrospray
ionization), confirmed the detectable differences
in macro-motion amplitude when Dϕ ≠ 0, as
well as the maximum differences at Dϕ ¼ þp=2
or p=2 with reversed ion-ejection orders (Fig.
2E). The enantiomer mixture solution was

Zhou et al., Science 383, 612–618 (2024) 9 February 2024 2 of 7

RES EARCH | R E S E A R C H A R T I C L E

Fig. 2. Characterization of directional rotation effect for enantiomer
separations. (A) Molecular structures of R- and S-binaphthyl-triflate enantiomers.
(B) Rotating dual-ac excitation fields. (C) Lissajous figures of ion macro motion.
(D) Measured VAC values of S-binaphthyl-triflate (red, upper) and R-binaphthyl-triflate
(blue, upper) as a function of Dϕ. (E) Profiling spectra of R- and S-binaphthyl-triflate
mixture with phase difference, Dϕ. The resolution is defined as VAC =DVAC, where
VAC and DVAC are the ac excitation voltage and the full width at half maximum
of the peak, respectively. Error bars represent the mean ± SD of five replicates.
The curves were fit by B-spline. R- and S-binaphthyl-triflate concentrations
were 75 and 25 mM, respectively.

prepared with concentrations of 75 and 25 mM
for R- and S-binaphthyl-triflate, respectively. Reso-
lutions >10,000 were obtained for analysis of the
ionic enantiomers by ion cloud profiling. Drag
forces of the ions could be characterized by the
excitation voltage at the ion ejection, Vac. The Vac
values of the R- and S-binaphthyl-triflate ions as
a function of Dϕ are shown in Fig. 2D, and the
corresponding spectra are shown in Fig. 2E and
fig. S4. We optimized the E/N conditions by
adjusting the gas pressure and the trapping
voltage VRF of the RF field for maximizing
the enantiomer separation and the analysis
performance using ion cloud profiling (fig. S5).
This method can be easily implemented using
an ion trap, which is also capable of MS/MS anal-
ysis for structural confirmation, and therefore
holds potential for a broad range of applica-
tions. We further demonstrated the applicability
of this method for analyzing enantiomers of
a variety of different chemical biological com-
pounds, including amino acids, carbohydrates,
small-molecule drugs, and metabolite com-
pounds. For drug compounds, enantiomers with
opposite handedness may have completely op-
posite effects in the human body (8, 9). For
instance, R-thalidomide is sedative while
S-thalidomide is teratogenic, and D-penicillamine
is antiarthritic whereas L-penicillamine is toxic.
Consequently, a fast and facile means for iden-
tifying the chirality of the organic compounds
is of considerable interest for drug discovery as
well as drug compound synthesis. In this study,
R- and S-penicillamine as well as R- and
S-thalidomide were analyzed, with the ion cloud
profiling spectra recorded as shown in Fig. 3, A
and B, respectively. Another example shown
in Fig. 3C is the analysis of enantiomeric mix-
tures of 2-hydroxyglutarate (2HG). The 2HG
serves as a metabolite biomarker, but the L/D
ratio of enantiomeric forms is as an indication
of the cause of disease. Whereas mutations in

Zhou et al., Science 383, 612–618 (2024) 9 February 2024 3 of 7

RES EARCH | R E S E A R C H A R T I C L E

A Penicillamine B Thalidomide C 2HG

D D S L
S L
100 100 100
L R D

Rel. Int.

Rel. Int.
Rel. Int.
0 0 0
D S L
100 100 100
L R D

L 0 0 D 0
34 38 42 R 40 44 48 36 40 44
VAC [mV] VAC [mV] VAC [mV]

D Glucose E Mannose F Palonosetron

S,R,S,S (L) S,S,R,R (D) R,R
L L S,S
100 100 100
D D R,R

Rel. Int.

Rel. Int.
Rel. Int.

0 0 0
L L S,S
100 100 100
D R,R
D

0 0 0
R,S,R,R (D) 33 37 41 R,R,S,S (L) 33 37 41 36 40 44
S,S
VAC [mV] VAC [mV] VAC [mV]

G H Leu I
Leu Phe Trp 100 D L 3
Trp
L
Peak Separation

0
Rel. Int.

Phe 2 Phe
100
VAC [mV]

D L Leu
0 1
100 Trp
D L
0
D 0
140 145 150 155
34 38 42
VAC [mV] [Å]

Fig. 3. High-resolution separation of chiral compounds. Structures of
molecule with single stereocenter and ion cloud profiling spectra of enantiomer
mixtures of L- and D-forms (or R- and S-forms) at Dϕ ¼ p=2 (upper) and p=2
(lower): (A) penicillamine, (B) thalidomide, (C) 2HG. Structures of molecule
with multiple stereocenters and ion cloud profiling spectra of enantiomer mixtures of
L- and D-forms (or R- and S-forms) at Dϕ ¼ p=2 (upper) and p=2 (lower):
(D) glucose, (E) mannose, and (F) palonosetron. (G) Molecular structures and
(H) Ion cloud profiling spectra of amino acid enantiomers: leucine (Leu),
phenylalanine (Phe), tryptophan (Trp). (I) Measured peak separation DVAC values
of amino acid enantiomers plotted as a function of collision cross section (CCS); CCS
values are from reference (47). The chiral center (red) and ionization site (blue)
are marked in the molecular structures.

the genes encoding isocitrate dehydrogenase
result in an elevation of D-2HG and acute leu-
kemias and brain tumors (11, 33), some other
metabolism errors would lead to a higher
concentration of L-2HG (34, 35). MS analysis
of 2HG for intro-surgery examination has
been developed using miniature MS systems
with direct sampling ionization (36, 37). With
the method reported here, differentiation of
the 2HG enantiomeric forms could be readily
implemented for point-of-care purposes.
This method has been tested for analyzing
enantiomers with multiple stereocenters. Mono-
saccharides with four chiral centers, glucose and
mannose, were analyzed with spectra shown
in Fig. 3, D and E, respectively. Palonosetron with
two stereocenters was also analyzed (Fig. 3F),
and its S enantiomeric form is used to treat
nausea and vomiting caused by chemotherapy
(38). High resolutions above 10,000 were ob-
tained for analysis of all of these chiral com-
pounds, and the order of the enantiomer ejection
could be switched by adjusting Dϕ. In another
experiment, we selected a group of three amino
acids with increasing sizes for the side chains
leucine, phenylalanine, and tryptophan (Fig. 3G).
As shown in Fig. 3H, the order of ejection for
L- and D-enantiomeric forms is the same at
the same Dϕ, which is expected because they
all have similar configurations with the amine
group charged. Further peak separations were
observed for amino acids with larger side chains
(Fig. 3H). This could also be explained with the
propeller analogy (Fig. 1A). The enantiomeric
configurations with larger collision cross sec-
tions are expected to show a larger difference
in terms of colliding with gas-phase molecules
with symmetry-breaking motions. The measured
peak separation DVAC values for each pair of
enantiomers are plotted as a function of colli-
sion cross section as shown in Fig. 3I, which
confirms this hypothesis.

Zhou et al., Science 383, 612–618 (2024) 9 February 2024 4 of 7

RES EARCH | R E S E A R C H A R T I C L E
Fig. 4. Determination of enantiomeric ratio.
(A) Ion cloud profiling spectra and (B) calibration
curve for S- and R-binaphthyl-triflate mixtures. For
quantitative analysis of the mixture, the concentra-
tion of S-binaphthyl-triflate was 10 mM and the actual
e.e. = ðcR  cS Þ=ðcR þ cS Þ varied from 0 to 0.82,
where cR and cS are the concentrations of R- and
S-binaphthyl-triflate, respectively. Measured e.e. =
ðIR  IS Þ=ðIR þ IS Þ, where IR and IS are the intensity of
R- and S-binaphthyl-triflate. Each value represents
the mean ± SD of 15 replicates. (C) Profiling spectra
and (D) calibration curve for S- and R-thalidomide
mixtures. For quantitative analysis of the mixture,
the concentration of S-thalidomide was 20 mM and
the actual e.e. varied from 0 to 0.98. Each value
represents the mean ± SD of 10 replicates.
We further explored the possibility of quan-
tifying enantiopurity on the basis of the direc-
tional rotation effect using ion cloud profiling.
The analysis was performed for a series of
mixtures of R- and S-binaphthyl-triflate, with
enantiomeric excess (e.e.) values from 0 to 0.82.
The ion cloud profiling spectra were recorded
as shown in Fig. 4A, and the measured e.e.
values were calculated and plotted against
the actual values as shown in Fig. 4B. Cali-
bration curves for pure samples are shown
in fig. S7. We characterized the quantitation of
enantiopurity for the drug thalidomide (Fig. 4,
C and D), with e.e. values for the mixture sam-
ples ranging from 0 to 0.98, which were con-
firmed by circular dichroism measurements
(fig. S8). Good linearity was obtained for the
calibration.
Application to asymmetric catalysis
optimization
This developed method could be used to facil-
itate the optimization of reaction conditions
for enantioselective synthesis. For demonstra-
tion, we selected asymmetric hydrogenation
(AH), an important type of reaction owing to
Zhou et al., Science 383, 612–618 (2024) 9 February 2024
its pronounced efficiency, atom economy, and
extensive substrate versatility (39, 40). An
essential facet of AH development involves the
systematic and high-throughput exploration
of reaction conditions to identify the optimal
catalysts for specific substrates that enable
high product enantioselectivity and yield (41–43).
Currently, chiral-phase HPLC serves as the main
analysis method for this purpose. The proce-
dures for using chiral-phase HPLC to analyze
newly synthesized compounds, however, typi-
cally require relatively large (greater than milli-
gram) sample amounts and can be laborious
and time-consuming, especially when compared
with the method introduced here (fig. S9) (42).
In this study, the ion cloud profiling with
direction ion rotation was used for screening
ligands in manganese-catalyzed asymmetric
hydrogenation of quinoline (Fig. 5A and fig. S10)
(44–46). A stainless-steel wire was inserted into
the crude reaction product solution, adsorbing
less than 10 ng on its tip (see supplementary
materials for the method of estimation), and
then inserted into a nano-ESI capillary con-
taining 10 ml of acetonitrile. A high voltage was
then applied to the metal wire to produce ions
through nano-ESI, which were subsequently
analyzed by the miniature dual-LIT mass spec-
trometer (Fig. 5, B and C, and fig. S9B). Tandem
mass spectrometry and ion cloud profiling with
directional ion rotation were performed for
reaction product compounds at m/z 298, with
both MS/MS and chiral-specific IMS spectra
obtained for structural analysis (Fig. 5, C to E).
A complete procedure for the entire analysis
took less than 1 min. The IMS spectra were
recorded for hydrogenation products from
reactions with four different catalyst ligands—
s
L1, sL2, sL3, and sL4 (Fig. 5A and fig. S10, sL:
ligands with S configuration)—as shown in
Fig. 5E. The e.e. values associated with use of
the four ligands were calculated and confirmed
in separate measurements with chiral-phase
HPLC (Figs. S12 and S13), indicating an opti-
mal reaction condition using ligand sL4 with
an e.e. obtained at 89.3% ± 0.6% for the en-
antioselective synthesis.
Our study shows that when the enantiomers
are trapped as ions, the directional rotational
effect can break the chiral symmetry sufficient-
ly to separate them based on differential colli-
sion behavior. This helped to physically separate
5 of 7
RES EARCH | R E S E A R C H A R T I C L E

Fig. 5. Rapid screening of catalysts for A

asymmetric hydrogenation. (A) Asymmetric Catalyst with Ligand S
sL1, sL2, sL3 or sL4
hydrogenation of 2-(3,4-dimethoxyphenethyl) quino-
+ H2
line catalyzed by manganese catalysts containing
various ligands: sL1, (SC,RFC)-N-2-(1H-imidazol-2-yl)-
+
1-(2-bis[3,5-di-phenylphenyl]phosphine)-ferrocenyle-
thylamine (pink); sL2, (SC,RFC)-N-2-(1H-4(5)- R
phenyl-imidazol-2-yl)-1-(2-bisphenylphosphine)-
ferrocenylethylamine (orange); sL3, (SC, RFC)-N-2-
(1H-imidazol-2-yl)-1-(2-diphenylphosphino)- BB HV
ferrocenylethylamine (blue); sL4, (SC, RFC)-N-2-
(1H-benzo[d]imidazol-2-yl)-1-(2-diphenylphosphino)- < 10 ng nanoESI Mini MS
ferrocenylethylamine (purple). Here, C and FC
stand for central chirality and planar chirality,
respectively. (B) Procedure for direct analysis of
crude reaction product using a miniature mass Crude products 10 µL acetonitrile
spectrometer. Crude product of less than 10 ng was
sampled by the stainless-steel wire, transferred
C Products D m/z 298.4
to the nanoESI tip, diluted by 10 ml of acetonitrile, 205.3
100 m/z 298.4 100
266.4
and subjected to ESI for MS/MS analysis and
205.3
e.e. analysis. (C) MS spectra of the crude reaction
Rel. Int.

MS/MS 298.4

Rel. Int.
product. (D) MS/MS spectra of the isolated 50 50
hydrogenation product ion (left) with structure
shown to indicate the fragmentation (right).
(E) Ion cloud profiling spectra (left) and measured 0 0 266.4
e.e. values of reaction products (right) using 250 300 350 150 250 350
m/z m/z
catalyst with different ligands conditions: sL1(pink);
E
s
L2(orange); sL3(blue); sL4(purple). Measured S R
e.e. = ðIS  IR Þ=ðIR þ IS Þ, where IR and IS are the 89% 100
intensity of R- and S-2-(3,4-dimethoxyphenethyl)- 76%
1,2,3,4-tetrahydroquinoline. Each value represents the sL4

mean ± SD of 10 replicates. Chiral analysis

e.e. [%]
sL3

34% 50
sL2

6%
sL1
0
34 38 42 sL1 sL2 sL3 sL4

VAC [mV] Ligand

the enantiomer ions in the trap through colli-
sions with background gas molecules. This work
demonstrates the selection of enantiomers
without requiring chiral references and simple
implementation using an ion trap, which can
be readily applied with the use of small ins-
truments and simple procedures for a broad
range of applications. Further investigation
and development are needed to establish a
comprehensive understanding of the corre-
lations between the excitation inducing the
directional rotation effect and elucidation of
the enantiomeric structures. Determination
of the absolute chiral configurations might be
challenging using this method alone; however,
in combination with structure analysis by MS/
MS, using a single ion trap analyzer might
provide viable solutions for fast chiral analysis
at high specificity. We have demonstrated a
quantitation at e.e. up to 98%, although higher
e.e. values for measurement might be desirable
for other applications. This, however, could 14. K. Harrison, A. S. Mackay, L. Kambanis, J. W. C. Maxwell,
be intrinsically limited by the dynamic range R. J. Payne, Nat. Rev. Chem. 7, 383–404 (2023).
15. Y. Ma, P. Oleynikov, O. Terasaki, Nat. Mater. 16, 755–759
of ion traps, mostly due to the space charge (2017).
problem. The use of ion traps of larger trapping 16. A. Furusho et al., Anal. Chem. 91, 11569–11575 (2019).
capacity, or careful control of the number of 17. C. Xie et al., Anal. Chem. 93, 859–867 (2021).
18. H. Wang, F. Wu, F. Xu, Y. Liu, C.-F. Ding, Anal. Chem. 93,
trapped ions, could help to overcome this 15096–15104 (2021).
limitation. 19. W. A. Tao, R. G. Cooks, Anal. Chem. 75, 25A–31A (2003).
20. E. Zlibut, J. C. May, J. A. McLean, J. Am. Soc. Mass Spectrom.
RE FERENCES AND NOTES 33, 996–1002 (2022).
21. J. R. Enders, J. A. McLean, Chirality 21 (suppl 1), E253–E264
1. Y. Xu, T. F. Zhu, Science 378, 405–412 (2022).
2. J. S. Richardson, Nature 268, 495–500 (1977). (2009).
3. H. Lam et al., Science 325, 1552–1555 (2009). 22. X. Yu, Z.-P. Yao, Anal. Chim. Acta 968, 1–20 (2017).
4. Y. Liu, Z. Wu, D. W. Armstrong, H. Wolosker, Y. Zheng, Nat. Rev. 23. W. A. Tao, D. Zhang, E. N. Nikolaev, R. G. Cooks, J. Am. Chem.
Chem. 7, 355–373 (2023). Soc. 122, 10598–10609 (2000).
5. A. Banreti et al., Nat. Commun. 13, 7059 (2022). 24. D.-Q. Han, Z.-P. Yao, Trends Analyt. Chem. 123, 115763 (2020).
6. S. Lechner et al., Nat. Commun. 14, 3548 (2023). 25. P. Dwivedi et al.., Anal. Chem. 78, 8200–8206 (2006).
7. J. F. Beary 3rd, C. R. Eaton, Science 257, 145–145 (1992). 26. X. Zhou, Z. Wang, J. Fan, Z. Ouyang, Nat. Commun. 14, 1535
8. D. V. Augusti, R. Augusti, F. Carazza, R. G. Cooks, Chem. (2023).
27. D. Patterson, M. Schnell, J. M. Doyle, Nature 497, 475–477
Commun. (19): 2242–2243 (2002).
9. E. Dorey, Nat. Biotechnol. 18, 1239–1240 (2000). (2013).
10. N. Vargesson, Birth Defects Res. C Embryo Today 105, 140–156 28. M. Avalos et al., Chem. Rev. 98, 2391–2404 (1998).
(2015). 29. J. B. Clemens, O. Kibar, M. Chachisvilis, Nat. Commun. 6, 7868
(2015).
11. G. Notarangelo et al., Science 377, 1519–1529 (2022).
12. S. Daly, F. Rosu, V. Gabelica, Science 368, 1465–1468 (2020). 30. D. Schamel et al., J. Am. Chem. Soc. 135, 12353–12359 (2013).
13. R. M. Kim et al., Nature 612, 470–476 (2022). 31. Y. Zhang et al., Nat. Commun. 10, 3742 (2019).

Zhou et al., Science 383, 612–618 (2024) 9 February 2024 6 of 7

RES EARCH | R E S E A R C H A R T I C L E
32. X. Liu, X. Wang, J. Bu, X. Zhou, Z. Ouyang, Anal. Chem. 91,
1391–1398 (2019).
33. J.-A. Losman et al., Science 339, 1621–1625 (2013).
34. J. Zhao et al., Nat. Metab. 3, 1372–1384 (2021).
35. D. Ye, K.-L. Guan, Y. Xiong, Trends Cancer 4, 151–165 (2018).
36. R. Zou et al., Anal. Chem. 91, 1157–1163 (2019).
37. F. Pu et al., Anal. Bioanal. Chem. 411, 1503–1508 (2019).
38. A. De Leon, Proc. Bayl. Univ. Med. Cent. 19, 413–416 (2006).
39. R. Noyori, Angew. Chem. Int. Ed. 41, 2008–2022 (2002).
40. L. Alig, M. Fritz, S. Schneider, Chem. Rev. 119, 2681–2751
(2019).
41. M. R. Friedfeld, H. Zhong, R. T. Ruck, M. Shevlin, P. J. Chirik,
Science 360, 888–893 (2018).
42. C. C. Wagen, S. E. McMinn, E. E. Kwan, E. N. Jacobsen, Nature
610, 680–686 (2022).
43. A. R. Rosales et al., Nat. Catal. 2, 41–45 (2019).
44. C. Liu et al., Angew. Chem. Int. Ed. 60, 5108–5113 (2021).
Zhou et al., Science 383, 612–618 (2024) 9 February 2024
45. C. Liu, M. Wang, Y. Xu, Y. Li, Q. Liu, Angew. Chem. Int. Ed. 61, License information: Copyright © 2024 the authors, some rights
e202202814 (2022). reserved; exclusive licensee American Association for the
46. C. Liu, X. Liu, Q. Liu, Chem 9, 2585–2600 (2023). Advancement of Science. No claim to original US government
47. X. Zheng et al., Chem. Sci. 8, 7724–7736 (2017). works. https://www.sciencemag.org/about/science-licenses-
journal-article-reuse
ACKN OWLED GMEN TS
We thank Y. Xia, X. Ma, and W. Zhang (Tsinghua University) for SUPPLEMENTARY MATERIALS
helpful discussions, and J. Yang (Tsinghua University) for help with
science.org/doi/10.1126/science.adj8342
the organic synthesis. Funding: This work was supported by
Supplementary Text
National Natural Science Foundation of China, grant nos.
Figs. S1 to S13
21627807, 21934003, and 22227807. Author contributions:
Tables S1 to S4
Conceptualization: X.Z., Z.O. Investigation: X.Z., Z.W., S.L., J.B.,
Equations S1 to S14
X.R., Q.L. Writing – original draft: X.Z., Z.O. Writing – review and
References (48–52)
editing: X.Z., Z.O. Competing interests: Z.O. is the founder and
J.P. is an employee of PURSPEC Technology, which is developing Submitted 18 July 2023; resubmitted 22 August 2023
miniature MS systems. Data and materials availability: All data Accepted 3 January 2024
are available in the main text or the supplementary materials. 10.1126/science.adj8342
7 of 7
RES EARCH

GEOLOGY the mine has an origin distinct from that of

the gas. The water percolates from the upper
A deep reservoir for hydrogen drives intense levels of the mine, mainly draining through
the shafts, whereas the lower levels of the mine
degassing in the Bulqizë ophiolite are apparently dry as miners reported no evi-
dence of water inflow during the excavation
Laurent Truche1*, Frédéric-Victor Donzé1, Edmond Goskolli2, Bardhyl Muceku3, Corinne Loisy4, of the deepest galleries (28). However, this
Christophe Monnin5, Hugo Dutoit1, Adrian Cerepi4 does not rule out the involvement of meteoric
distal water or water from other origins in ser-
Deep crustal production of hydrogen (H2) is a potential source of primary energy if recoverable pentinization elsewhere in the geological forma-
accumulations in geological formations are sufficiently large. We report direct measurements of an tion or below. We measured a gas flow rate of
elevated outgassing rate of 84% (by volume) of H2 from the deep underground Bulqizë chromite 5 ± 1 L/s (at 25°C and 1.031 × 105 Pa) from
mine in Albania. A minimum of 200 tons of H2 is vented annually from the mine's galleries, making several vigorous bubbling zones located in a
it one of the largest recorded H2 flow rates to date. We cannot attribute the flux solely to the release of small 30 m2 pool (28). This gas is composed of
paleo-fluids trapped within the rocks or to present-day active and pervasive serpentinization of H2 (84.0 vol%) and CH4 (13.2 vol%) with minor
ultramafic rocks; rather, our results demonstrate the presence of a faulted reservoir deeply rooted in concentrations of N2 (2.7 vol%). Therefore, the
the Jurassic ophiolite massif. This discovery suggests that certain ophiolites may host economically amount of H2 discharged in the gallery from
useful accumulations of H2 gas. this single pool is 11 t/yr (∼30 kg/day) to the
minimum, as we did not account for the

H
ydrogen (H2), like electricity, is a carbon-
free energy carrier that plays an impor-
tant role in modern industry and in the
energy transition; however, most H2 is
manufactured using natural gas through
a process that consumes energy and releases
large amounts of carbon dioxide into the atmo-
sphere. A previously overlooked geologic source
of H2 could contribute to diversifying our en-
ergy mix and reducing the carbon footprint of
our economy.
Although geologic H2 plays a central role in
the abiotic synthesis of simple organic com-
pounds in the Earth’s crust and in supporting
deep microbial communities (1–4), the high mobil-
ity of the molecule combined with its high
reactivity were thought to prevent accumula-
tion in the subsurface except in rare cases (5–8).
This paradigm is called into question by recent
findings of H2-rich fluids in surface seeps and
underground mines or boreholes in specific
geological settings, along with a reanalysis
of historical drilling data (9–14). In particular,
H2-rich fluids produced by subsurface reactions
accompanying the serpentinization of ultramafic
rocks such as peridotites are well-documented
in uplifted orogenic and ophiolitic bodies (15–21).
Despite these observations, our understanding
of the processes and settings most conducive
to the formation of substantial accumulations
of H2 are still evolving. Notwithstanding the
possible importance of geologic H2 as a clean
fuel or as an energy source for life, the current
knowledge of H2 occurrences within the Earth’s
lithosphere is limited, with recoverable and eco-
nomic resources of H2 poorly quantified. The
main reason for this lack of fundamental un-
1
Université Grenoble Alpes, CNRS, IRD, ISTERRE, UMR 5275,
Grenoble, France. 2National Agency of Natural Resources,
NANR-AKBN, Tirana, Albania. 3Polytechnic University of
Tirana, FGJM, Department of Earth Sciences, Tirana, Albania.
4
Université Bordeaux, CNRS, Bordeaux INP, EPOC, UMR
5805, Pessac, France. 5Université Toulouse 3, CNRS, IRD,
GET, UMR 5563, Toulouse, France.
*Corresponding author. Email: laurent.truche@univ-grenoble-alpes.fr
derstanding of the H2 system stems from the in-
herent challenges in sampling deep geological
fluids and the lack of deep infrastructure tar-
geting potentially fertile H2 settings. This leads
to the pivotal question of whether there is a H2
geological system comparable to that of petro-
leum. Our study unveils a high emission rate of
almost pure geologic H2, suggesting the poten-
tial for a new extractable primary energy source.
The Bulqizë mine: geological context and
intense H2 outgassing
We conducted an exploration campaign in the
deep underground chromite mine of Bulqizë
in Albania (Fig. 1), where the presence of
flammable gas was first reported in 1992 at a
depth of 620 m. After the discovery of the
gas, major explosions occurred in 2011, 2017,
and 2023. The Bulqizë chromite mine is one of
the largest chromium extraction sites in the
world, with total recovery of over 20 million
tons of high-grade ore averaging 35 wt% Cr2O3.
The mine is situated within the Bulqizë Jurassic
ultramafic massif (fig. S6), approximately 40 km
northeast of Tirana. This massif is a component
of the giant Eastern Mediterranean supra-
subduction-zone ophiolite belt that extends from
Turkey to Slovenia over more than 3000 km
and represents one of the largest and most
complete preserved segments of oceanic litho-
sphere on Earth (22–26). The massif spans an
area of 370 km2 to a depth of 6 km and hosts
numerous folded and faulted concordant chro-
mite ore bodies embedded in the mantle se-
quence (26–28).
We observed very intense outgassing in the
deeper levels of the mine, specifically at depths
ranging from 500 to 1000 m below the surface
(Fig. 1 and Table 1). The most intense gas
discharge in the mine galleries is located in a
tectonic zone, i.e., a highly faulted domain,
where focused and intense bubbling is visible
in drainage pools and runoff streams located
at level L19 (fig. S1 and movie S1). The water in
many minor bubbling points.
Since 2017, advancements in monitoring tech-
nology have enabled precise measurements of
the H2 flow rate throughout the mine (28).
This has been made possible by the installation
of H2 sensors and flow meters on both the
ventilation circuit of inner shaft N9 (at L19)
and a dense network of 38 boreholes, extending
from L17 to L21 (170 m deep) through the fault
zone (figs. S2 and S3). The stale air from L19 to
L21, containing 0.40 vol% H2, is discharged
through shaft N9 at a flow rate of 840 Nm3/min
(29). This leads to a H2 flow rate of 3.4 Nm3/min
or 158 t/yr. Notably, the small 30 m2 pool emits
by itself 7% of the H2 flow vented through shaft
N9, demonstrating the importance of the fault
zone as a major drain or reservoir for H2 (28).
The 38 interconnected boreholes were drilled
to manage the discharge of H2 from the tectonic
zone intersecting the mine galleries and the ore
body. The boreholes are constantly flushed
by 4530 Nm3/hr of air. The exhaust gases, con-
taining 1.20 vol% H2, are channeled outside the
mine through a dedicated pipe, thus providing
an additional H2 flow rate of 54 Nm3/hr
(42 t/yr). This flow rate has remained cons-
tant over a six-year observation period. Only
a fraction of the total amount of air vented
from the mine is monitored for H2. Indeed,
100 Nm3/sec of air is circulated through the
mine by the main ventilation system. Thus, a
minimum of 200 tons of H2 are released from
the mine every year (1.0 × 108 mol/yr). Such a
relatively large flow of H2 greatly exceeds the
few outgassing rates previously reported from
dry seeps and hyperalkaline springs hosted in
ophiolites (Table 1). The H2 outgassing rates
we report are minimum values that have been
accurately measured and are not extrapolated
from single point surface measurements and
diffusive flux models.
Revealing the presence of a deep reservoir
The elevated H2 output raises the question of
whether the outgassing results from active

Truche et al., Science 383, 618–621 (2024) 9 February 2024 1 of 4

RES EARCH | R E S E A R C H A R T I C L E

Fig. 1. Schematic 3D view of the Bulqizë underground chromite mine. (A) Map of the deeper levels with location of the H2-bearing fault zone indicated.
(B) Locations where H2 outgassing rates have been measured. The entrance of the mine is at an altitude of 840 m above mean sea level (amsl).

Table 1. Outgassing rates of H2 from different sites. Concentrations of H2 and CH4 in the free gas phase from the Bulqizë mine (N2 and O2 concentrations
are given in table S1) and other ophiolite-hosted seeps and bubbling pools.

Area H2 flow H2 CH4 H2/CH4

H2 outgassing site Ref.
(m2) (t/yr) (vol%) (vol%) (vol/vol)
Oman, Haylayn pool (bubbles + diffuse) 20 ∼200 0.158 86.4 6.7 12.9
............................................................................................................................................................................................................................................................................................................................................
Oman, Misfah pool (bubbles + diffuse) 20 ∼1000 0.056 66.9 7.2 9.3
............................................................................................................................................................................................................................................................................................................................................
Turkey, Chimaera (diffuse dry seeps) 21 2000 3.5 9.9 87.0 0.11
............................................................................................................................................................................................................................................................................................................................................
Albania, Bulqizë mine, L19 pool (focused bubbling) This study 30 11 84.0 13.2 6.4
............................................................................................................................................................................................................................................................................................................................................
Albania, Bulqizë mine, L17 tectonic zone (boreholes) This study 400 42 1.20 0.15 8.0
............................................................................................................................................................................................................................................................................................................................................
Albania, Bulqizë mine, level L19 (shaft N9) This study ∼20,000* 158 0.40 0.05 8.0
............................................................................................................................................................................................................................................................................................................................................
*The area of level L19 (~20,000 m2) corresponds to the horizontal projection of the mining works’ footprint, and not to the H2 outgassing area, which is unknown but mostly concerns the fault
zone. These estimates provide an indication of the scale of the sites, even if the areas of surface and underground outgassing sites are not directly comparable.

serpentinization or from the liberation of entire Bulqizë massif. Our aim is to differen- particularly relevant to the ophiolitic context
fossil H2 entrapped within the rock. Evaluating tiate between active (present-day) flows and (5, 8, 17–21, 28, 30).
the volume of rock involved in this H2 gas gen- stocks (i.e., accumulations), with the latter Our first scenario considers both the decrepi-
eration and outgassing is therefore crucial. We arising from a combination of ancient (possi- tation of fluid inclusions and the release of H2
explored three scenarios that could account for bly fossil) and ongoing H2 generation processes. occluded in microporous minerals and micro-
the observed H2 flow rate of 1.0 × 108 mol/yr, Although multiple sources of geologic H2 in fractures. In this case, H2 is considered a paleo
emphasizing the respective roles of the fault the subsurface are recognized (6, 8), our anal- byproduct of the early stages of a serpentiniza-
zone, the mine’s drainage volume, and the ysis concentrates on three scenarios that are tion process that started 165 to 160 Myr ago (26).

Truche et al., Science 383, 618–621 (2024) 9 February 2024 2 of 4

RES EARCH | R E S E A R C H A R T I C L E
Fig. 2. Three different scenarios for H2 production. Life span of a constant H2 flow rate of 1.0 × 108 mol/yr
as function of the drained volume of rock and porosity, according to three different scenarios. Scenario 1:
decrepitation of paleo H2 occluded in the rock (dark blue); Scenario 2: active serpentinization at low temperature
(yellow); Scenario 3: release of H2 trapped inside the fault zone which acts as a reservoir (green). The gray
area represents the estimated volume range of the fault zone acting as a reservoir.
In our second scenario, we assume that the ob-
served H2 flow within the mine results from
present-day low-temperature serpentinization.
In our third scenario, we assume that the re-
lease of H2 occurs when a previously sealed
fault zone is opened during mining opera-
tions. In this latter case, H2 is stored within
the fractures and connected porosity of the
fault zone at a filling flow rate substantially
lower than the outgassing rate measured today
within the mine. For all three scenarios, we
calculated the life span of the measured cur-
rent H2 flow rate in the mine as a function of
the drained volume of rock (Fig. 2). We assume
a rock density of 3000 kg/m3 and consider an
average temperature and hydrostatic pressure
of 100°C and 30 MPa at 3 km depth, respec-
tively. We also use these parameters to estimate
an equivalent porosity containing gaseous H2
(28, 31). We also consider a 3-km average depth
for the ophiolite layer due to its bowl-shaped
geometry, with a maximum depth of 6 km (26).
In scenario 1, which involves the release of
paleofluids occluded in the rocks (20, 28, 30,
32, 33), we measured a range of H2 content
from 7.5 ×10−6 to 39.0 ×10−6 mol/kgrock
(average = 15.4 × 10−6 mol/kgrock; n = 9 samples)
Truche et al., Science 383, 618–621 (2024) 9 February 2024
in the bulk harzburgite, dunite, and chromi-
tite rock samples collected in the deepest levels
of the mine (table S2 and fig. S4). Given that the
H2 flow rate in the mine has remained constant
during the last 6 years, this requires a drained
rock volume of 5 to 27 km3. If we assume that
the mine is the sole outlet for H2, this implies
that the total volume of ophiolite constituting
the entire Bulqizë massif (i.e., 370 km2 × 3 km)
would be depleted in 250 to 1300 years. Thus,
this scenario seems unlikely, as all H2 within the
massif would have vanished instantaneously on
a geological timescale.
In scenario 2, which involves the direct con-
sequences of an active serpentinization process
at depth in the presence of water, a volume
of 1.2 × 10−4 km3 must be altered every year
to account for the observed H2 flow rate of
1.0 × 108 mol/yr, assuming that 0.3 mol of H2
are produced per kg of serpentinized peridotite
at temperatures below 100°C (20). If we con-
sider a peridotite area of 1 km2, this would
result in a serpentinization front advancing
by 0.15 m per year. Notably, this rate is 250 times
faster than the regional uplift rate of 0.6 mm/yr
for the Bulqizë ophiolite (34). Thus, if the uplift
is negligible and the velocity of the serpentini-
zation front remained constant over time, the
reaction front would have advanced beyond the
maximum thickness of the ophiolite (6 km) in
40,000 years, implying that the H2 production
potential should be exhausted by now. A simi-
lar conclusion can be reached by considering a
maximum mine drainage volume of 135 km3,
corresponding to the effective rainwater catch-
ment area of the mine of 45 km2 (fig. S7) and a
rock thickness of 3 km. In this case, the whole
drainage volume should have been fully serpen-
tinized in only 13.5 Ma whereas final exhumation
of the massif is dated at 15 to 45 Ma (24). An-
other independent line of evidence to rule
out present-day active serpentinization as
the source of H2 monitored in the mine comes
from the comparison of the observed H2 flow
rate with the global estimates of H2 production
rate from both the oceanic lithosphere and the
Precambrian continental crust. The latter ranges
from 1.0 × 1010 to 1.2 × 1012 mol/yr (35–37), mean-
ing that the contribution from the Bulqizë mine
alone would be 0.01 to 1% of this global flux. This
amount is an unrealistic contribution, consider-
ing the small size of the Bulqizë ultramafic massif
compared with both the 80,000-km-long oceanic
ridge system and the 1.06 × 108 km2 Precambrian
shield surface area. Therefore, H2 produced by
serpentinization or any other processes must
have accumulated over a long period of time in
a reservoir rock, the most realistic one being the
fault zone as indicated by its intense degassing.
This conclusion does not mean that decrepitation
(scenario 1) or serpentinization (scenario 2) do
not happen, but they alone cannot explain the
actual resulting flow rate.
In scenario 3, we consider that the flow mea-
sured in the mine results from the degassing
of H2 gas trapped in the fault zone only after
being produced. The mine perforated the
top of the fault zone at several points when it
reached a depth ranging from 0.5 to 1 km,
releasing the gas stored in this sealed volume,
which acts as a porous reservoir. Based on in
situ observations (28), the fault zone is ~10 m
wide, with a length varying from 100 m to
1 km, and a maximum height of 5 km (with
the sealed top being around 500 m deep in
the mine). These dimensions yield volumes of
rock ranging from 5.0 × 10−3 to 5.0 × 10−2 km3
(represented by the gray area in Fig. 2). The
total pore space volume generated by the overall
amount of fractures present in this highly
damaged zone is unknown, so we assume
that the equivalent porosity has an average
value of 5% at depth as measured in fault
zones in Oman ophiolites (38). Thus, a fault
volume of 1.3 × 10−3 km3 with a porosity of 5%
would be sufficient to sustain the observed H2
flow rate for 6 years. Considering the estimated
range of fault volume and the same porosity,
the measured flow rate could be sustained for
25 to 238 years. In other words, the total amount
of H2 stored in the fault zone would range from
3 of 4
RES EARCH | R E S E A R C H A R T I C L E
5000 to ~50,000 tons. The temperature ranging
from 40° to 160°C (28), along with the dry con-
ditions in the deep horizons of the ophiolite
may have hampered consumption of H2 by both
microbial activity (too hot) and abiotic redox
(too cold) reactions, at least in the deepest part
of the reservoir. Hydrogen-rich fluids isolated
in the crust for billions of years have already
been reported in deep fracture networks from
underground mines and boreholes in Pre-
cambrian shields demonstrating that H2 can
accumulate over geological timescales (9, 39).
Implications for geologic H2 exploration
What sets our discovery apart is the large flux
of almost pure H2 gas we have observed. In the
context of energy transition, our findings could
substantially affect the ongoing search for new
energy resources. We reveal that ophiolites,
which are mantle rocks from the oceanic crust
obducted onto continents, not only constitute
effective source rocks, but also have the poten-
tial to host high-quality, H2-rich gas reservoirs.
These geological formations, prevalent across
all the continents, extend beyond being mere
geological anomalies. In fact, many ophiolites
worldwide have been found to contain springs
or seeps that release H2 gas. Within this geo-
logical context, H2 exhibits potential for com-
mercial extraction as it can be focused into
fracture zones and trapped.
In the past, the oil and gas industry has
largely ignored ophiolites, considering them
unsuitable for hydrocarbon resource exploita-
tion. However, these onshore formations might
offer potential for large-scale H2 accumulations
and therefore constitute a promising target for
H2 exploration. A key point is the presence of
drainage systems such as faults and tectonic
zones in these geological settings. As a result, a
holistic understanding of the tectonic and
petrophysical factors influencing the migra-
tion pathways and accumulation of H2 is
needed to guide exploration. The configuration
and properties of the seal remain uncertain, but
the ore body and the necking of the faults may
play crucial roles. Chromitite, commonly found
Truche et al., Science 383, 618–621 (2024) 9 February 2024
in ophiolites, is noteworthy as several known
H2-rich seeps are situated near chromite mines
(32, 40, 41). The possible connection between
chromitite and H2 emissions, though not yet
confirmed, must be ascertained. Finally, in the
emerging quest for geologic H2, it is crucial to
consider both the nature of these H2 resources—
whether they are ancient (fossil) or recent—and
the potential impact on deep-seated microbial
ecosystems that thrive on H2 (42, 43). A com-
prehensive understanding of these factors is
essential to mitigate risks and ensure a sus-
tainable development of H2 resources.
RE FERENCES AND NOTES
1. D. S. Kelley et al., Science 307, 1428–1434 (2005).
2. C. Greening et al., ISME J. 10, 761–777 (2016).
3. L. Truche, T. M. McCollom, I. Martinez, Elements 16, 13–18
(2020).
4. E. P. Reeves, J. Fiebig, Elements 16, 25–31 (2020).
5. N. J. Smith, T. J. Shepherd, M. T. Styles, G. M. Williams,
“Hydrogen exploration: A review of global hydrogen
accumulations and implications for prospective areas in
NW Europe” in Petroleum Geology Conference Series.
(The Geological Society, 2005); vol. 6, pp. 349–358.
6. V. Zgonnik, Earth Sci. Rev. 203, 103140 (2020).
7. C. J. Boreham et al., APPEA J. 61, 163 (2021).
8. A. V. Milkov, Earth Sci. Rev. 230, 104063 (2022).
9. B. Sherwood Lollar et al., Astrobiology 7, 971–986 (2007).
10. V. A. Nivin, Appl. Geochem. 74, 44–55 (2016).
11. L. Truche et al., Earth Planet. Sci. Lett. 493, 186–197 (2018).
12. J. Guélard et al., Geochem. Geophys. Geosyst. 18, 1841–1865
(2017).
13. A. Prinzhofer, C. S. Tahara Cissé, A. B. Diallo, Int. J. Hydrogen
Energy 43, 19315–19326 (2018).
14. N. Lefeuvre et al., Geochem. Geophys. Geosyst. 22,
e2021GC009917 (2021).
15. C. Neal, G. Stanger, Earth Planet. Sci. Lett. 66, 315–320
(1983).
16. T. A. Abrajano et al., Chem. Geol. 71, 211–222 (1988).
17. G. Etiope et al., Appl. Geochem. 84, 286–296 (2017).
18. E. T. Ellison et al., J. Geophys. Res. Solid Earth 126,
e2021JB021981 (2021).
19. C. Monnin et al., J. Geophys. Res. Biogeosci. 126,
e2021JG006243 (2021).
20. J. A. Leong et al., Geochim. Cosmochim. Acta 347, 1–15
(2023).
21. G. Etiope, Int. J. Hydrogen Energy 48, 9172–9184 (2023).
22. A. H. F. Robertson, Lithos 65, 1–67 (2002).
23. Y. Dilek, H. Furnes, M. Shallo, Lithos 100, 174–209
(2008).
24. B. Muceku et al., Terra Nova. 20, 180–187 (2008),
25. Y. Dilek, H. Furnes, Lithos 113, 1–20 (2009).
26. A. Meshi, F. Boudier, A. Nicolas, I. Milushi, Int. Geol. Rev. 52,
117–141 (2009).
27. L. Hoxha, Mitt. Ges. Geol. Bergbaustud. Osterr. 48, 52–58
(2007).
28. Supplementary Text, Materials and Methods are available as
supplementary materials.
29. Nm3/min is a SI unit for volumetric flow rate of gas
normalized at a temperature of 0°C and a pressure of
101.325 kPa.
30. N. G. Grozeva, F. Klein, J. S. Seewald, S. P. Sylva, Philos. Trans.
A Math. Phys. Eng. Sci. 378, 20180431 (2020).
31. F.-V. Donzé, L. Truche, Zenodo (2023): 10.5281/zenodo.8308027.
32. G. Etiope et al., Sci. Rep. 8, 8728 (2018).
33. F. Klein, N. G. Grozeva, J. S. Seewald, Proc. Natl. Acad. Sci. U.S.A.
116, 17666–17672 (2019).
34. J. Carcaillet, J. L. Mugnier, R. Koçi, F. Jouanne, Quat. Res. 71,
465–476 (2017).
35. S. L. Worman, L. F. Pratson, J. A. Karson, E. M. Klein, Geophys.
Res. Lett. 43, 6435–6443 (2016).
36. A. S. Merdith et al., Geochem. Geophys. Geosyst. 21,
e2019GC008869 (2020).
37. B. S. Lollar, T. C. Onstott, G. Lacrampe-Couloume, C. J. Ballentine,
Nature 516, 379–382 (2014).
38. I. Katayama et al., Tectonophysics 814, 228978 (2021).
39. G. Holland et al., Nature 497, 357–360 (2013).
40. C. Bohdanowicz, AAPG Bull. 18, 750–760 (1934).
41. G. Etiope, M. Schoell, H. Hosgörmez, Earth Planet. Sci. Lett.
310, 96–104 (2011).
42. W. J. Brazelton et al., PeerJ 5, e2945 (2017).
43. E. M. Fones et al., ISME J. 13, 1750–1762 (2019).
AC KNOWLED GME NTS
We are grateful to AlbChrome Ltd. mining company and its CEO,
C. Acar; the miners, engineers (B. Raman and V.O. Karatay); and the
director of the Bulqizë mine (M. Demçe) for their strong support in
the field. We thank P. Kelemen and two anonymous reviewers for their
insights and manuscript reviews. Funding: This project has received
fundings from the French National Research Agency (grant
agreement ANR-20-CE01-0020) and from the CNRS (MITI 2022.2).
L.T. acknowledges support from the Institut Universitaire de France.
Author contributions: Funding acquisition: A.C. and L.T. Project
administration: A.C., B.M., E.G., and L.T. Fieldwork: All authors.
Laboratory analysis: L.T., A.C., and C.L. Visualization: L.T., H.D., and
F.V.D. Writing – original draft: L.T. and F.V.D. Writing – review and
editing: All authors. Competing interests: Authors declare that they
have no competing interests. Data and materials availability: All
data are available in the main text or the supplementary materials.
The MATLAB script used in this study is available at Zenodo (31).
License information: Copyright © 2024 the authors, some rights
reserved; exclusive licensee American Association for the
Advancement of Science. No claim to original US government
works. https://www.sciencemag.org/about/science-licenses-journal-
article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adk9099
Materials and Methods
Supplementary Text
Figs. S1 to S10
Tables S1 to S3
References (44–50)
Movie S1
Submitted 18 September 2023; accepted 10 January 2024
10.1126/science.adk9099
4 of 4
RES EARCH

BIOSYNTHESIS epoxidated to yield a 4(20)b-epoxy-5a-acetoxy

intermediate (called the epoxide ring), and
Characterization and heterologous reconstitution of then ring expansion from the epoxide ring to
the oxetane ring occurs via a rearrangement
Taxus biosynthetic enzymes leading to baccatin III reaction (22) (fig. S2C).
Our previous study demonstrated that the
Bin Jiang1†, Lei Gao2†, Haijun Wang2†, Yaping Sun1,3†, Xiaolin Zhang1, Han Ke2, Shengchao Liu1, CYP725A subfamily, which contains all char-
Pengchen Ma4, Qinggang Liao1, Yue Wang1, Huan Wang1, Yugeng Liu1, Ran Du1, Torben Rogge4‡, acterized oxygenases involved in the skeleton
Wei Li1, Yi Shang5, K. N. Houk4, Xingyao Xiong1, Daoxin Xie6, Sanwen Huang1, tailoring of paclitaxel, occurs exclusively in the
Xiaoguang Lei2,7*, Jianbin Yan1* Taxus genus (14, 23–28). Given that the oxe-
tane ring is only known to be produced by
Paclitaxel is a well known anticancer compound. Its biosynthesis involves the formation of a highly Taxus plants (29, 30), the oxygenase respon-
functionalized diterpenoid core skeleton (baccatin III) and the subsequent assembly of a phenylisoserinoyl sible for forming the oxetane ring is likely
side chain. Despite intensive investigation for half a century, the complete biosynthetic pathway of derived from this subfamily. Therefore, we iso-
baccatin III remains unknown. In this work, we identified a bifunctional cytochrome P450 enzyme lated the reaction substrate from Taxus mairei,
[taxane oxetanase 1 (TOT1)] in Taxus mairei that catalyzes an oxidative rearrangement in paclitaxel oxetane taxa-4(20),11-diene-2a,5a,7b,9a,10b,13a-hexaol-
formation, which represents a previously unknown enzyme mechanism for oxetane ring formation. hexa-acetate (taxadiene hexa-acetate, 1) (31)
We created a screening strategy based on the taxusin biosynthesis pathway and uncovered the enzyme and developed a two-step screening strategy
responsible for the taxane oxidation of the C9 position (T9aH1). Finally, we artificially reconstituted a to determine if any member of the CYP725A
biosynthetic pathway for the production of baccatin III in tobacco. subfamily could catalyze the oxidation of the
double-bond moiety.

P
aclitaxel, which is derived from the sec-
ondary metabolism of Taxus genus plants
in the Taxaceae family, has been clin-
ically used to treat various cancers (1–3).
Paclitaxel contains a structurally com-
plex 6-8-6 tricyclic carbon skeleton that bears
nine stereocenters, one notable oxetane ring
motif, and one phenylisoserine chain (3, 4).
Paclitaxel can pass through nanopores in the
microtubule wall and interact with tubulin on
the lumen surface of microtubules to disrupt
microtubule dynamics, which triggers cyto-
toxic effects against cancer cells (5, 6). The
overall conformational rigidification of the tri-
cyclic carbon skeleton is generated by the
oxetane ring of paclitaxel (7).
One strategy presently used to produce
paclitaxel is chemical semisynthesis, in which
paclitaxel precursors, such as baccatin III, are
isolated from plant cell cultures or raw tree
biomass of Taxus and subsequently chemically
1
Shenzhen Branch, Guangdong Laboratory for Lingnan Modern
Agriculture, Key Laboratory of Synthetic Biology, Ministry of
Agriculture and Rural Affairs, Agricultural Genomics Institute at
Shenzhen, Chinese Academy of Agricultural Sciences, Shenzhen,
China. 2Beijing National Laboratory for Molecular Sciences,
Peking-Tsinghua Center for Life Sciences, Key Laboratory of
Bioorganic Chemistry and Molecular Engineering of Ministry
of Education, College of Chemistry and Molecular
Engineering, Peking University, Beijing, China. 3State Key
Laboratory for Conservation and Utilization of Subtropical
Agro-Bioresources, College of Life Sciences, South China
Agricultural University, Guangzhou, China. 4Department of
Chemistry and Biochemistry, University of California, Los
Angeles, Los Angeles, CA, USA. 5Yunnan Key Laboratory of
Potato Biology, The CAAS-YNNU-YINMORE Joint Academy of
Potato Sciences, Yunnan Normal University, Kunming, China.
6
Peking-Tsinghua Center for Life Sciences, School of Life
Sciences, Tsinghua University, Beijing, China. 7Institute for
Cancer Research, Shenzhen Bay Laboratory, Shenzhen,
China.
*Corresponding author. Email: jianbinlab@caas.cn (J.Y.);
xglei@pku.edu.cn (X.L.)
†These authors contributed equally to this work.
‡Present address: Institute of Chemistry, Technical University of
Berlin, Straße des 17. Juni 115, 10623 Berlin, Germany.
converted into paclitaxel (8). The semisynthesis
strategy relies heavily on natural resources
and artificial cultivation of Taxus plants, which
cannot meet the growing market demand
(9). With the development of synthetic biol-
ogy, biosynthetic strategies to produce pacli-
taxel in a green and sustainable manner have
become extremely attractive (10, 11).
Since the structure of paclitaxel was first
identified in 1971 (3), elucidating its biosyn-
thetic pathway has always been a challenging
task in natural product research. To date, ap-
proximately 20 possible enzymatic reactions
have been reported (12). The reaction steps
can be divided into three critical processes,
including the formation of a taxane skeleton
with a 4(20)-ene-5a-ol moiety from the sub-
strate geranylgeranyl pyrophosphate (GGPP),
the biosynthesis of baccatin III through the
epoxidation of the C4 and C20 double bonds,
and the attachment of a phenylisoserine chain
to the C13 position of baccatin III to generate
paclitaxel (fig. S1). C13 side chain formation
and its combination with baccatin III to pro-
duce paclitaxel have been well studied (13).
However, several essential steps in forming
baccatin III remain unknown, especially oxetane
ring formation and C9 oxygenation (fig. S1)
(14, 15). As a result, the upstream and down-
stream steps of paclitaxel biosynthesis cannot
be linked, and a complete paclitaxel biosynthet-
ic pathway has so far been lacking (16–18).
Results
Identification of TOT1 for oxetane ring formation
The oxetane ring of paclitaxel is a notable
chemical motif and contributes to the molecule’s
anticancer mechanism (19–21). The generally
accepted proposal for oxetane ring formation
in paclitaxel involves the following chem-
ical processes: first, the 4(20)-ene-5a-yl ace-
tate moiety (called the double-bond moiety) is
We first divided the genes in the CYP725A
subfamily into three groups (groups I, II, and
III) based on phylogenetic analysis (Fig. 1A)
and simultaneously expressed all genes in each
group in tobacco leaves by Agrobacterium-
mediated transformation. When the expres-
sion of exogenous genes reached a high level on
the fourth day after Agrobacterium infiltrat-
ion, we injected substrate (1) into the cor-
responding leaf region where the infiltration
was performed. After a reaction for 1 day, the
leaf metabolites were extracted with methanol
and analyzed by liquid chromatography mass
spectrometry (LC-MS) (Fig. 1B). We found that
the substrate peak area in the group II sample
was considerably decreased, producing a pro-
minent peak for the oxidation product (2)
with the same retention time and MS spectra
as 1-dehydroxybaccatin IV, which contains the
oxetane ring (fig. S3A); these results indicate
that the desired enzymes were present in
group II (Fig. 1C). By contrast, only substrate
peak (1) could be detected in the samples of
groups I and III, without any other peak of
substrate oxidation (Fig. 1C).
To determine which gene in group II oxi-
dized the substrate, we next expressed all
genes of this group in tobacco leaves individ-
ually and performed substrate-feeding exper-
iments separately. The reaction catalyzed by
Chr9_74725878 exhibited a strongly decreased
substrate peak (1) and increased oxetane pro-
duct peak (2), indicating that the enzyme en-
coded by Chr9_74725878 could catalyze oxetane
ring formation (Fig. 1D).
In addition to the oxetane product peak,
the enzyme catalyzes the formation of anoth-
er oxidative product (3) (Fig. 1, C and D). Its
retention time and MS spectra were consistent
with those of baccatin I (fig. S3B), which con-
tains a C4,C20-epoxide ring. To rule out the
possibility that other cytochrome P450 enzymes
in tobacco may be involved in the reaction, we

Jiang et al., Science 383, 622–629 (2024) 9 February 2024 1 of 8

RES EARCH | R E S E A R C H A R T I C L E
Fig. 1. TOT1 is responsible for oxetane formation in Taxus plants. (A) Phylogenetic
analysis of the CYP725A subfamily genes (protein sequence identity <90%) in
Taxus mairei (Taxus). The neighbor-joining tree was constructed by Interactive Tree
of Life (iTOL) software. Evolutionary distances were analyzed by the p-distance
method. Bootstrap values (based on 1000 replicates) are indicated at the tree nodes.
(B) Workflow for activity screening of candidate genes belonging to the CYP725A
subfamily. (C) Extracted ion chromatograms (EICs) for taxadiene hexa-acetate (1),
1-dehydroxybaccatin IV (2), and baccatin I (3) in Nicotiana benthamiana leaves
expressing candidate genes from different groups with the infiltration of taxadiene
Jiang et al., Science 383, 622–629 (2024) 9 February 2024
hexa-acetate (1). N. benthamiana leaves expressing the GFP gene with the infiltration
of compounds 1 to 3 were used as negative controls. Ac, acetyl. (D) EICs for taxadiene
hexa-acetate (1), 1-dehydroxybaccatin IV (2), and baccatin I (3) in N. benthamiana
leaves expressing the single gene in group II with the infiltration of compound 1.
The negative controls were the same as those used in (C). (E) In vivo assay to
characterize the function of TOT1 in insect cells. Insect cells coexpressing TOT1 and
TcCPR were resuspended in 1 ml of medium [containing 4% v/v dimethyl sulfoxide
(DMSO)] and then incubated with 1 (50 mM) at 28°C and 100 rpm for 24 hours. Insect
cells expressing the empty vector (EV) and T5aH were used as negative controls.
2 of 8
RES EARCH | R E S E A R C H A R T I C L E

Fig. 2. The 5a-acetyl 4, 20-ene moiety is indispensable for the TOT1-mediated
oxetane formation reaction. (A) EICs for substrates 4 or 5 and their desired
oxidative product molecular weights (m/z of 731 for 4, m/z of 601 for 5, m/z of 747
for the oxidative product of 4, and m/z of 617 for the oxidative product of 5) in insect
cells expressing TOT1. EV and T5aH were used as negative controls. Insect cells co-
expressing TOT1 and TcCPR were resuspended in 1 ml of medium (containing 4% v/v
DMSO) and then incubated with 4 or 5 (50 mM) at 28°C and 100 rpm for
24 hours. (B) EICs for substrate 1 and its oxidative products 2 and 3 in insect cells
expressing TOT1. Insect cells coexpressing TOT1 and TcCPR were resuspended in
1 ml of medium (containing 4% v/v DMSO) and then incubated with compound
1 (50 mM), 2 (50 mM), or 3 (4.5 mM) at 28°C and 100 rpm for 24 hours. EV and T5aH
were used as negative controls. (C) Summary of the catalytic functions of TOT1.

assayed the reaction with an enzyme heterol-
ogously expressed in insect cells. The substrate
1 could not be oxidized by the negative-control
cells that expressed T5aH (Fig. 1E). By con-
trast, insect cells expressing Chr9_74725878
produced two clear oxidative product peaks,
with the primary product being the oxetane
ring product (2) and the minor product being
the epoxide product (3) (Fig. 1E), a finding
that was supported by systematically char-
acterizing the chemical structures of 2 and
3, including 1D and 2D nuclear magnetic reso-
nance (NMR) spectra (figs. S4 and S5 and
table S1). We also performed kinetic char-
acterization of Chr9_74725878 using insect
microsomes. The Michaelis constant (Km) value
for substrate 1 was determined as 14.14 ± 3.15 mM
(fig. S6), which is comparable to the Km values
of the other reported cytochrome P450s in
the baccatin III biosynthesis (23–28). Knock-
down of Chr9_74725878 in Taxus cells was
able to decrease the content of both baccatin
III and paclitaxel, highlighting its essential
role in paclitaxel biosynthesis (fig. S7). Togeth-
er, these results demonstrated that Chr9_
74725878 mainly catalyzed an oxidative rear-
rangement of the double-bond moiety to form
an oxetane ring; consequently, we named the
enzyme taxane oxetanase 1 (TOT1).
Reaction mechanism of oxetane ring formation
in taxanes
Based on structural analysis and biomimetic
synthesis, many studies have proposed that
the epoxide ring formation is a prerequisite for
generating the oxetane ring (15, 32–35). The
identification of TOT1 enabled us to examine
the proposed mechanism. When the C5 acetyl
group was replaced with cinnamoyl (4), or
there was no substitution (5) (fig. S8), these
substrates did not form oxidative oxetane by
TOT1 (Fig. 2A), which supports that the acetyl
group is essential for the subsequent oxidative
rearrangement reaction. TOT1 did not pro-
duce any new products in the presence of
compound 2 (Fig. 2B), demonstrating the ir-
reversibility of the rearrangement reaction.
When we used compound 3 as the substrate to
test whether TOT1 could convert the epoxide
ring to the oxetane ring, no new peak was de-
tected when compound 3 was incubated with
TOT1 in the insect cell assay (Fig. 2B), which
differed from the positive control with com-
pound 1. TOT1 is therefore a bifunctional
oxygenase that directly converts the alkene
moiety into the epoxide and the oxetane ring.
However, TOT1 cannot function as an isomer-
ase to transform the epoxide ring into the
oxetane ring (Fig. 2C). The epoxide ring is thus
not an essential intermediate for oxetane ring
formation.
To investigate the reaction mechanism for
the formation of the epoxide and oxetane
rings by TOT1, we first applied AlphaFold2 to
obtain a high-confidence predicted structure
of TOT1 (fig. S9). By docking compound 1 into
the active pocket of modeled TOT1, we ob-
tained the most populated conformation of
1 after three 200-ns simulations (fig. S10A).
The dihedral ∠C4-C5-O1(Ac)-C1(Ac) is mainly
distributed at approximately −60° (fig. S10B).
With the most representative structure of
1 from molecular dynamics simulation, we
performed density functional theory (DFT)
calculations using an enzyme-free computa-
tional truncated model (36, 37). As shown in
Fig. 3, the attack on C20 by the O of Fe=O
involves an energy barrier of 25.1 kcal/mol,
which leads to radical intermediate Int0 (fig.
S11). A more stable oxonium ion intermediate
(Int1) is formed when the acetyl group is ro-
tated, with a very low barrier and CO bond
formation. Then, oxetane product 2 is formed
by a shift in O(Fe) to carbon and regeneration
of the acetate group, with a 6.8-kcal/mol en-
ergy barrier. Int1 can also produce epoxide
ring product 3 via TS2 with a higher energy
barrier (10.7 kcal/mol). Epoxide and oxetane
can both be produced from Int1 in the TOT1
enzyme. The epoxide does not need to serve as
a precursor for the oxetane. The generation
of oxetane ring product 2 is energetically and
kinetically favored over epoxide 3.
Functional screening and identification of T9aH1
for C9 hydroxylation
In addition to the oxetane ring structure, bac-
catin III contains seven mono-oxidation sites.
Because the mono-oxidation reaction at the C9
position likely occurs simultaneously with the
oxidation reactions at other sites, the mono-
oxidation intermediates at the C9 position

Jiang et al., Science 383, 622–629 (2024) 9 February 2024 3 of 8

RES EARCH | R E S E A R C H A R T I C L E
Fig. 3. Reaction energy profile of the TOT1-mediated oxidation reaction using a model system for the axial cysteine-ligated Fe-porphyrin catalyst.
Relative Gibbs energies (DG) are reported, which are shown in purple for the doublet electronic state and green for the quartet electronic state. Int, intermediate;
TS, transition state.
could not be isolated; as a result, the enzymes
that catalyze C9 oxidation have not been iden-
tified during the past 30 years (16, 38). To
uncover the unknown C9 hydroxylase, we es-
tablished a screening strategy in tobacco by
constructing the taxusin biosynthesis pathway
de novo (Fig. 4A). Taxusin (6) is an interme-
diate in an alternative route of paclitaxel bio-
synthesis (4, 39). It bears four acetyl groups
at positions C5, C9, C10, and C13 in its struc-
ture (Fig. 4A), which originates from cyto-
chrome P450-catalyzed hydroxylation and
acyltransferase-mediated acetylations. Except
for the unknown C9 hydroxylase, other en-
zymes that possess terpene synthase activity
(40), hydroxylase activities (23–25), and acetyl-
transferase activities have been reported
(41, 42), including TXS, T5aH, T10bH, T13aH,
and TAT1 to TAT4 (Fig. 4A). In the screening
strategy, we used the Agrobacterium-mediated
transient expression system to simultaneously
coexpress all known enzymes of taxusin bio-
synthesis with a single C9 hydroxylase candi-
date in tobacco leaves. Based on this strategy,
C9 hydroxylase could be identified by de-
tecting the presence of taxusin in transformed
leaf samples.
To obtain the C9 hydroxylase candidates,
we first determined the tissue-specific accu-
mulation of taxusin. Through LC-MS analysis,
we found that taxusin was highly accumu-
lated in Taxus roots, with little distribution
in needles and bark (fig. S12A). Furthermore,
we found that the known genes of taxusin
Jiang et al., Science 383, 622–629 (2024) 9 February 2024
biosynthesis were highly expressed in roots,
exhibiting a strong correlation with the con-
tent of taxusin (Fig. 4B). By further analyzing
the tissue-specific expression patterns between
CYP725As and known genes involved in tax-
usin biosynthesis, we screened 17 CYP725A
candidates, as shown in Fig. 4B. Next, we in-
dividually transformed the candidate genes
into the tobacco chassis containing known
genes of taxusin biosynthesis and found that
the coexpression of the Chr9_26460669 gene
produced a new peak 6 that shared the same
retention time and tandem MS (MS/MS) spec-
tra with taxusin in the metabolites of tobacco
(Fig. 4C and fig. S12B). The results demon-
strated that the protein encoded by the Chr9_
26460669 gene possesses the desired C9 hy-
droxylase activity, and we named it taxane 9a-
hydroxylase 1 (T9aH1).
Artificial reconstitution of the baccatin III
biosynthetic pathway in tobacco
After identifying the two essential genes (TOT
and T9aH) for baccatin III synthesis, we at-
tempted to coexpress the two new genes with
other known genes involved in baccatin III
biosynthesis (TXS, T5aH, T13aH, T2aH, T7bH,
TAT, and TBT) to determine whether we can
artificially reconstruct the biosynthetic pathway
of baccatin III in tobacco. The results showed
that a new product (peak 7) could be detected
when T9aH and TOT were coexpressed with the
seven known genes of the baccatin III biosynthe-
tic pathway, and the peak product exhibited the
same mass [mass/charge ratio (m/z) 609 (M +
Na)+] and retention time as baccatin III (7)
(Fig. 5A). MS/MS analysis further demonstrated
that the new product 7 was baccatin III (Fig. 5B).
We further investigated whether these nine
genes form the critical pathway that leads
to the synthesis of baccatin III in tobacco. As
shown in Fig. 5A, baccatin III could hardly be
detected when any of the nine genes were
absent in the coexpression system. The result
indicated that the nine genes constitute the
core pathway in baccatin III biosynthesis. Al-
though T10bH has long been regarded as an
essential enzyme for baccatin III biosynthesis
(25), the results demonstrated that enzymes
in the core combination could replace the oxi-
dation function of T10bH at the C10 position.
In addition, 10-deacetylbaccatin III-10-O-acetyl
transferase (DBAT) was reported as the critical
enzyme in the C10 acylation of baccatin III
(12, 43). Our results showed that baccatin III
could still be synthesized in the absence of
DBAT (Fig. 5A), which is consistent with our
observation that TAT might also catalyze the
C10 acylation (fig. S13).
Functional cooperativity of the core genes in
baccatin III biosynthesis
To clarify the actions of baccatin III biosyn-
thetic genes in Taxus plants, we performed
transcriptional analysis to determine the ex-
pression profile of these genes in different tis-
sues. TOT and T9aH were mainly expressed in
roots rather than bark and needles regardless
4 of 8
RES EARCH | R E S E A R C H A R T I C L E

Fig. 4. Functional screening and identification of T9aH1 for C9 hydroxylation. (A) The screening scheme of T9aH and chemical structure of taxusin.
(B) Heatmap showing the gene expression patterns of T9aH1 candidate genes and related known genes involved in taxusin biosynthesis in different Taxus tissues.
The color scale bar represents relative expression levels. (C) EICs for taxusin (6) in the different tobacco leaf samples that coexpressed T9aH1 candidates or the GFP
gene with the known genes of the taxusin biosynthesis pathway.

of sex differences, and similar expression pat-
terns to those of T2aH, T7bH, TAT, and TBT
were observed (Fig. 5C). Their expression pat-
terns are more similar to those of genes such
as PAM, BAPT, and DBTNBT, which are rela-
ted to the downstream steps of paclitaxel syn-
thesis; the patterns are different from those of
upstream genes for paclitaxel synthesis, in-
cluding TXS, T5aH, and T13aH, which were
highly expressed in roots and needles (Fig. 5C).
The results suggested that the expression of
the nine genes was differentially regulated
at the tissue level. In addition, genes such as
TOT and T9aH exhibited a stronger expres-
sion correlation with the downstream genes of
paclitaxel synthesis.
Previous studies have shown that jasmo-
nate is an essential regulatory hormone for
paclitaxel biosynthesis (14, 44). We therefore
treated Taxus needles with jasmonate at dif-
ferent times to evaluate the coordination of
these baccatin III synthesis core genes. The
results showed that these genes could be in-
duced after 4 hours of treatment and that ex-
pression peaked at approximately 24 hours
and then gradually decreased (Fig. 5D and
fig. S14), which is an expression pattern simi-
lar to those of downstream genes BAPT and
DBTNBT (Fig. 5D and fig. S14). By contrast,
the expression pattern of T14bH differed from
those of the other nine genes. The highest ex-
pression peak of T14bH appeared at a 4-hour
point and then decreased rapidly under jas-
monate treatment (Fig. 5D). Given that T14bH
is in an oxidation branch of paclitaxel precur-
sors that does not lead to paclitaxel synthesis
(26), the different expression pattern of T14bH
implied that another regulatory mechanism
occurred between the baccatin III pathway
and the branch pathway that did not produce
paclitaxel. Moreover, the coexpression network
showed a close correlation between the expres-
sion of each of the nine genes, but expression
of T14bH did not exhibit significant correlations
with these genes, further suggesting that a coor-
dinated regulation occurs among the nine core
genes (Fig. 5E). In addition, although T14bH is
located in the group 9.2 in chromosome 9 (14),

Jiang et al., Science 383, 622–629 (2024) 9 February 2024 5 of 8

RES EARCH | R E S E A R C H A R T I C L E

Fig. 5. Pathway constitution and functional cooperativity of baccatin III
biosynthetic enzymes. (A) EICs for baccatin III (7) in tobacco leaves expressing
eight genes with different gene combinations (shown as groups 2 to 10,
respectively). Tobacco leaves that expressed the 9 necessary genes were used
as a positive control (shown as group 1). (B) MS/MS spectrum of baccatin III
produced in tobacco (black) and its standard (red). (C) Heatmap showing the
expression patterns of the paclitaxel biosynthetic genes in different tissues
(needle, bark, and root) of female (F) and male (M) Taxus plants. The color scale
bar represents relative expression levels. (D) The expression pattern of
biosynthetic genes of baccatin III at different times after jasmonate treatment
(100 mM MeJA). The significance analysis was conducted with a two-sided
Student’s t test. CK, the housekeeping gene (Taxus actin 1, ctg19083.1), was
used as a control. Data are means ± SE (n = 3 biological replicates). *P < 0.05;
**P < 0.01; ***P < 0.001. (E) Coexpression network of the core genes and
T14bH. The Pearson correlation coefficients of the genes are displayed on the
net. The orange solid and the gray dashed lines represent P < 0.05 and P > 0.05,
respectively. The thickness of the lines indicates the r value of the correlation
coefficient. (F) The genomic location of TOT and T9aH with respect to other
genes. The known cytochrome P450s, acyltransferases, and TXS involved in
the paclitaxel biosynthesis and functionally uncharacterized CYP725As are
indicated with blue and red lines, respectively. (G) Subcellular location of TOT,
T9aH, and T5aH. T5aH, an ER-localized cytochrome P450, was used as a
positive control. Colocalization analysis between different channels was
performed by analyzing the gray value distributions of the white line in the region
of interest. Scale bars are 5 mm. (H) Illustration of the biosynthetic process of
baccatin III in plant cells.

standing between T5aH and TOT (Fig. 5F), the The spatial organization of enzymes was enzymes, we fused the green fluorescent pro-
different expression patterns also supported assumed to benefit metabolite synthesis and tein (GFP) gene with the genes individually
the regulation complexity of taxane metabolic regulation (45, 46). To investigate the sub- at their C termini and performed a thorough
pathways. cellular location of the six cytochrome P450 analysis. As shown in Fig. 5G, similar to T5aH

Jiang et al., Science 383, 622–629 (2024) 9 February 2024 6 of 8

RES EARCH | R E S E A R C H A R T I C L E
(18), TOT and T9aH displayed a comparable
distribution pattern to that of the endoplasmic
reticulum (ER) marker, demonstrating that
TOT and T9aH were located at the ER. Fur-
thermore, we found that other core compo-
nents responsible for baccatin III synthesis,
including T13aH, T7bH, and T2aH, were also
localized to the ER (fig. S15A). Combined with
the chloroplast localization of taxadiene syn-
thase (fig. S15B) (47) and cytoplasmic local-
ization of acyltransferases (12, 48), the ER
localization of the cytochrome P450 enzymes
indicated that highly synergistic regulation
occurred between the chloroplast, ER, and cyto-
plasm (Fig. 5H). First, isopentenyl diphosphate
(IPP) and dimethylallyl diphosphate (DMAPP)
are generated through the 2-C-methyl-D-erythritol
4-phosphate (MEP) pathway (18) and then con-
densed by geranylgeranyl diphosphate syntha-
ses (GGPPS) to form GGPP (49), the starting
substrate for taxane diterpenoids. Next, GGPP
is cyclized and transformed into taxadiene by
terpene synthase (TXS) in chloroplasts (40, 47).
Subsequently, taxadiene is transferred to the
cytoplasm probably by contact sites of plastid-
ER (50); synergistically modified by six ER-
associated P450s, including T2aH, T5aH, T7bH,
T9aH, T13aH, and TOT, as well as two cyto-
plasmic acyltransferases (TAT and TBT); and
eventually converted into baccatin III (Fig. 5H).
Discussion
Natural oxetane-containing compounds (OCCs)
are structurally characterized by enormous
diversity and complexity, thereby conferring
particular advantages and opportunities for
modern drug discovery and development. More
than 600 different OCCs have been found in
animals, plants, and microorganisms, although
most are produced by plants (30). The fol-
lowing reaction mechanisms have been dis-
covered for the biosynthesis of natural oxetane
skeletons: [2+2] cycloaddition (51), ring con-
traction (52, 53), and ring expansion (54) (fig.
S2). [2+2] Cycloaddition and ring contraction
mechanisms have been found in animals and
microorganisms. By contrast, natural OCCs,
which may originate from ring-expansion re-
actions, are only found in plants. In this study,
we identified the cytochrome P450 enzyme
TOT1 in Taxus mairei that catalyzes an un-
usual oxidative rearrangement reaction to
produce the oxetane ring (Figs. 1 to 3); this
reaction represents a distinctive chemical logic
for oxetane ring formation and reshapes our
knowledge on the origin of oxetane in nature.
The catalytic flexibility or promiscuity of
cytochrome P450s contributes to metabolic
bifurcation, thereby fostering structural diver-
sity among diterpenoids (55, 56). Our findings
show that only nine enzymes are needed to
synthesize baccatin III from GGPP, which is
far less than the at least 13 or 14 enzymes that
were thought to be required (11, 22, 54, 57, 58).
Jiang et al., Science 383, 622–629 (2024) 9 February 2024
The discovery also highlights that catalytic
promiscuity is somewhat common in the cyto-
chrome P450s and acyltransferases of baccatin
III biosynthesis. Previous studies have dem-
onstrated that T5aH can catalyze the forma-
tion of many unknown oxygenated products
other than taxadien-5a-ol (59–61), and recent-
ly it was revealed that this enzyme even has
weak cyclase activity leading to the oxetane
structure (62), which indicates its functional
promiscuity. T13aH could catalyze C13 hydro-
xylation with taxadien-5a-ol and taxadien-5a-yl
acetate as substrates, thereby exhibiting a
substrate flexibility similar to that of cyto-
chrome P450s in paclitaxel biosynthesis (24).
We found that T13aH could oxidize taxadiene
to produce many hydroxylated products,
one of which was confirmed to be taxadiene-
5a,10b,13a-triol (fig. S16), which suggests that
the function promiscuity of T13aH is far be-
yond present understanding. Therefore, it is
possible that the C1 and C10 hydroxylation
activity could be fulfilled by one or more en-
zymes among T5aH, T13aH, T9aH, T2aH, TOT,
and T7bH, presenting promising avenues for
further investigation.
Owing to the functional promiscuity of
these cytochrome P450s, the metabolic path-
way constituted by the nine core genes is
believed to be a complex network rather than
a linear one, which makes it quite challenging
to fully decipher the role of each gene in bac-
catin III biosynthesis. Meanwhile, owing to
the possible bifurcation pathway, the overall
yield of baccatin III in tobacco is low (~50 ng
per g dry weight). Future research will focus
on elucidating the specific catalytic orders and
substrate promiscuities of these nine enzymes
and find the rate-determining step for meta-
bolic engineering optimization. With more ef-
fort from a broad field of scientists, including
natural product chemists, plant physiologists,
and synthetic biologists, green and efficient
manufacturing of paclitaxel may be achieved
through synthetic biology in the near future.
RE FERENCES AND NOTES
1. A. Ahmed Khalil et al., Curr. Pharm. Des. 28, 3363–3373
(2022).
2. S. M. Swain, S. F. Honig, M. C. Tefft, L. Walton, Invest. New Drugs
13, 217–222 (1995).
3. M. C. Wani, H. L. Taylor, M. E. Wall, P. Coggon, A. T. McPhail,
J. Am. Chem. Soc. 93, 2325–2327 (1971).
4. E. Baloglu, D. G. Kingston, J. Nat. Prod. 62, 1448–1472
(1999).
5. F. Naaz, M. R. Haider, S. Shafi, M. S. Yar, Eur. J. Med. Chem.
171, 310–331 (2019).
6. H. Freedman, J. T. Huzil, T. Luchko, R. F. Ludueña,
J. A. Tuszynski, J. Chem. Inf. Model. 49, 424–436
(2009).
7. Y. Fu et al., Curr. Med. Chem. 16, 3966–3985 (2009).
8. J. C. Borah, J. Boruwa, N. C. Barua, Curr. Org. Synth. 4,
175–199 (2007).
9. A. Sharma et al., Biotechnol. Bioprocess Eng. 27, 706–728
(2022).
10. H. Wang et al., Acta Pharm. Sin. B 11, 3322–3334
(2021).
11. M. Sabzehzari, M. Zeinali, M. R. Naghavi, Biotechnol. Adv. 43,
107569 (2020).
12. T. Wang et al., Molecules 26, 2855 (2021).
13. R. Sanchez-Muñoz et al., Front. Bioeng. Biotechnol. 8, 410
(2020).
14. X. Xiong et al., Nat. Plants 7, 1026–1036 (2021).
15. R. Croteau, R. E. Ketchum, R. M. Long, R. Kaspera,
M. R. Wildung, Phytochem. Rev. 5, 75–97 (2006).
16. Y. Tong, Y. Luo, W. Gao, Phytochem. Rev. 21, 863–877
(2022).
17. I. Mutanda, J. Li, F. Xu, Y. Wang, Front. Bioeng. Biotechnol. 9,
632269 (2021).
18. J. Li et al., Nat. Commun. 10, 4850 (2019).
19. F. Guéritte-Voegelein, D. Guénard, P. Potier, J. Nat. Prod. 50,
9–18 (1987).
20. C. S. Swindell, S. F. Britcher, J. Org. Chem. 51, 793–797
(1986).
21. D. Willenbring, D. J. Tantillo, Russ. J. Gen. Chem. 78, 723–731
(2008).
22. S. Jennewein, M. R. Wildung, M. Chau, K. Walker,
R. Croteau, Proc. Natl. Acad. Sci. U.S.A. 101, 9149–9154
(2004).
23. J. Hefner et al., Chem. Biol. 3, 479–489 (1996).
24. S. Jennewein, C. D. Rithner, R. M. Williams,
R. B. Croteau, Proc. Natl. Acad. Sci. U.S.A. 98,
13595–13600 (2001).
25. A. Schoendorf, C. D. Rithner, R. M. Williams,
R. B. Croteau, Proc. Natl. Acad. Sci. U.S.A. 98,
1501–1506 (2001).
26. S. Jennewein, C. D. Rithner, R. M. Williams, R. Croteau,
Arch. Biochem. Biophys. 413, 262–270 (2003).
27. M. Chau, R. Croteau, Arch. Biochem. Biophys. 427, 48–57
(2004).
28. M. Chau, S. Jennewein, K. Walker, R. Croteau, Chem. Biol. 11,
663–672 (2004).
29. S. B. Horwitz, Cell 177, 502–505 (2019).
30. V. Vil et al., Appl. Microbiol. Biotechnol. 103, 2449–2467
(2019).
31. D. D. C. de Marcano, T. Halsall, J. Chem. Soc. D 1969,
1282–1283 (1969).
32. D. D. C. de Marcano, T. Halsall, J. Chem. Soc. D 1970,
1381–1382 (1970).
33. M. M. Francl, G. Hansell, B. P. Patel, C. S. Swindell, J. Am.
Chem. Soc. 112, 3535–3539 (1990).
34. J. L. Giner, J. A. Faraldos, Helv. Chim. Acta 86, 3613–3622
(2003).
35. R. Kaspera, J. L. Cape, J. A. Faraldos, R. E. Ketchum,
R. B. Croteau, Tetrahedron Lett. 51, 2017–2019
(2010).
36. S. P. de Visser, F. Ogliaro, N. Harris, S. Shaik, J. Am.
Chem. Soc. 123, 3037–3047 (2001).
37. J. Soler, S. Gergel, C. Klaus, S. C. Hammer, M. Garcia-Borràs,
J. Am. Chem. Soc. 144, 15954–15968 (2022).
38. H. G. Floss, U. Mocek, in TAXOL: Science and Applications,
M. Suffness, Ed. (CRC Press, 1995), pp. 191–208.
39. M. Miyazaki, K. Shimizu, H. Mishima, M. Kurabayashi,
Chem. Pharm. Bull. 16, 546–548 (1968).
40. M. R. Wildung, R. Croteau, J. Biol. Chem. 271, 9201–9204
(1996).
41. D. Hampel, C. J. D. Mau, R. B. Croteau, Arch. Biochem. Biophys.
487, 91–97 (2009).
42. M. Chau, K. Walker, R. Long, R. Croteau, Arch. Biochem. Biophys.
430, 237–246 (2004).
43. K. Walker, R. Croteau, Proc. Natl. Acad. Sci. U.S.A. 97, 583–587
(2000).
44. Y. Yukimune, H. Tabata, Y. Higashi, Y. Hara, Nat. Biotechnol.
14, 1129–1132 (1996).
45. L. J. Sweetlove, A. R. Fernie, Annu. Rev. Plant Biol. 64, 723–746
(2013).
46. C. H. Abrahamson, B. J. Palmero, N. W. Kennedy,
D. Tullman-Ercek, Annu. Rev. Biophys. 52, 553–572
(2023).
47. D. C. Williams et al., Arch. Biochem. Biophys. 379, 137–146
(2000).
48. K. Walker, A. Schoendorf, R. Croteau, Arch. Biochem. Biophys.
374, 371–380 (2000).
49. J. Hefner, R. E. B. Ketchum, R. Croteau, Arch. Biochem. Biophys.
360, 62–74 (1998).
50. P. Mehrshahi, C. Johnny, D. DellaPenna, Trends Plant Sci. 19,
501–507 (2014).
51. Q. Cheng, D. C. Lamb, S. L. Kelly, L. Lei, F. P. Guengerich,
J. Am. Chem. Soc. 132, 15173–15175 (2010).
52. J. Bridwell-Rabb, A. Zhong, H. G. Sun, C. L. Drennan, H.-W. Liu,
Nature 544, 322–326 (2017).
53. M. Hecker, V. Ullrich, J. Biol. Chem. 264, 141–150
(1989).
7 of 8
RES EARCH | R E S E A R C H A R T I C L E
54. M. Hezari, R. Croteau, Planta Med. 63, 291–295
(1997).
55. U. Bathe, A. Tissier, Phytochemistry 161, 149–162
(2019).
56. W. Gao et al., Org. Lett. 11, 5170–5173 (2009).
57. R. W. Alexander, Trends Biochem. Sci. 26, 152
(2001).
58. Y. Zhang et al., Mol. Plant 16, 1951–1961 (2023).
59. V. G. Yadav, J. Mol. Catal. B Enzym. 110, 154–164
(2014).
60. Q. Wu, Z. Huang, J. Wang, H. Yu, J. Xu, Bioresour. Bioprocess.
9, 82 (2022).
61. J. C.-T. Liu, R. De La Pena, C. Tocol, E. S. Sattely,
BioRxiv 2023.09.27.559859 [Preprint] (2023); https://doi.org/
10.1101/2023.09.27.559859.
62. Y. Zhao et al., J. Am. Chem. Soc. 146, 801–810
(2024).
63. Q. Liao, Contigs of taxus genome. Figshare (2021); https://doi.
org/10.6084/m9.figshare.15000672.v1.
ACKN OW LEDG MEN TS
We thank the staff (P. Lu, T. He, B. Niu, Y. Peng, M. Su,
F. Wang, and X. Liao) at the Agricultural Genomics Institute at
Shenzhen and J. Zhou at Peking University for help with activity
screening and MS analysis. We thank W. Wang (Engineer Center
of Pharmaceutical Technology, Tsinghua University) for assistance
with target compound detection in cell lines. We acknowledge
computational calculation support from the High-performance
Jiang et al., Science 383, 622–629 (2024) 9 February 2024
Computing Platform of Peking University. Funding: This work
was funded by the National Key R&D Program of China (grants
2023YFA0915800, 2018YFA0903200, 2020YFA0907900,
2021YFC2102900, 2022YFC3401500, 2022YFC2502500, and
2023ZD04076); the National Natural Science Foundation of China
(grants 22193073, 22322701, 21661140001, 92253305, and
22101009); the Science Technology and Innovation Commission of
Shenzhen Municipality of China (grant ZDSYS20200811142605017);
the Innovation Program of Chinese Academy of Agricultural Sciences
(CAAS) and the Elite Young Scientists Program of CAAS; the National
Institutes of Health National Institute of Allergy and Infectious
Diseases (grant R01 AI 141481 to K.N.H.); and the Beijing National
Laboratory for Molecular Sciences (grant BNLMS-CXX-202106).
Author contributions: J.Y. and X.L. conceived the study.
B.J. and Ya.S. cloned the genes and heterogeneously expressed them
in tobacco with help from Y.W. and W.L. S.H., Yi.S., and X.X. helped
connect Taxus materials. L.G. performed the purification and
synthesis of the substrates or standards from Taxus extracts. B.J. and
Ya.S. conducted the gene function confirmation. B.J. and X.Z.
performed subcellular localization and colocalization analyses. X.Z.,
S.L., Q.L., and Hu.W. conducted quantitative real-time polymerase
chain reaction (PCR) and the transcriptional and bioinformatic
analyses, respectively. Ha.W., L.G., and R.D. analyzed the samples
using LC-MS. H.K., P.M., Y.L., T.R., and K.N.H. performed the DFT
calculations. J.Y., X.L., D.X., S.H., B.J., L.G., Ha.W., Ya.S., and X.Z.
wrote the paper with input from all the authors. Competing
interests: J.Y., S.H., B.J., and Ya.S. are inventors on the
patent applications related to this work (CN202310496624.6,
CN202310961179.6, and CN202311137424.8) that were submitted by
the Agricultural Genomics Institute at Shenzhen. The other authors
declare no competing interests. Data and materials availability: The
cds and protein sequence files of CYP725As, including TOT
(Chr9_74725878, ctg447_gene.1; CYP725A37, named by D. Nelson)
and T9 (Chr9_26460669, ctg2120_gene.5; CYP725A55, named by
D. Nelson), are available at Figshare (63). Cartesian coordinates of
computed structures are provided in data S1. All other data are
presented in the text or supplementary materials. Requests for
materials should be addressed to corresponding author J.Y. License
information: Copyright © 2024 the authors, some rights reserved;
exclusive licensee American Association for the Advancement of
Science. No claim to original US government works. https://www.
science.org/about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adj3484
Materials and Methods
Figs. S1 to S16
Tables S1 and S2
Data S1
References (64–85)
MDAR Reproducibility Checklist
Submitted 22 June 2023; accepted 10 January 2024
Published online 25 January 2024
10.1126/science.adj3484
8 of 8
RES EARCH

QUANTUM GASES light sheets, we observed running and standing

waves of second sound, demonstrating multi-
Thermography of the superfluid transition ple reflections of entropy waves from the walls
of the box. Our thermography works across the
in a strongly interacting Fermi gas superfluid transition, allowing the observation
of a pronounced peak in thermal diffusion atTc,
Zhenjie Yan1†, Parth B. Patel1, Biswaroop Mukherjee1, Chris J. Vale2, characteristic of critical behavior expected near
Richard J. Fletcher1, Martin W. Zwierlein1* second-order phase transitions.

Spectral thermometry
Heat transport can serve as a fingerprint identifying different states of matter. In a normal liquid, a
hotspot diffuses, whereas in a superfluid, heat propagates as a wave called “second sound.” Direct The working principle of our method is sketched
imaging of heat transport is challenging, and one usually resorts to detecting secondary effects. In this in Fig. 1, A to D. In rf spectroscopy, interacting
study, we establish thermography of a strongly interacting atomic Fermi gas, whose radio-frequency atoms are ejected from the many-body system
spectrum provides spatially resolved thermometry with subnanokelvin resolution. The superfluid phase into an initially unoccupied internal spin state
transition was directly observed as the sudden change from thermal diffusion to second-sound (28). For interacting gases, the resulting spectra
propagation and is accompanied by a peak in the second-sound diffusivity. This method yields the full depend on temperature. At high temperatures,
heat and density response of the strongly interacting Fermi gas and therefore all defining properties of when the thermal de Broglie wavelength is
Landau’s two-fluid hydrodynamics. shorter than both scattering length and inter-
particle distance, the spectra approach the bare,

H
eat transport is a ubiquitous phenome-
non at work in everything from steam
engines to the formation of stars, and it
dictates how energy, information, and
entropy flow in the system. In conven-
tional materials, heat, mass, and charge are all
transported by the motion of (quasi)particles,
such as electrons in metals. This common
origin of transport results, for example, in the
Wiedemann-Franz law, relating thermal and
electrical conductivity. However, in strongly
correlated systems, such as high-temperature
superconductors (1), neutron stars (2), and the
quark-gluon plasma of the early universe (3),
the notion of a quasiparticle is poorly defined.
It is unknown whether there is a common
relaxation rate for heat, density, and spin trans-
port (4) or if strong correlations separate these
phenomena. Understanding the flow of entro-
py is at the forefront of current research, with
powerful theoretical models connecting ther-
mal flow in quantum systems to gravitational
duals (3, 5). Directly measuring thermal trans-
port, as distinct from mass or charge transport,
is thus of great relevance for elucidating the origin
of heat dissipation in strongly correlated matter.
Strongly interacting atomic Fermi gases near
a Feshbach resonance provide an ideal platform
for quantitative studies of fermion transport
(6–10). As a result of scale invariance in
resonant Fermi gases (11), measurements
performed in one system constrain the equa-
tion of state and transport properties of other
strongly interacting Fermi systems, including
neutron matter at densities 25 orders of mag-
1
MIT-Harvard Center for Ultracold Atoms, Research
Laboratory of Electronics, and Department of Physics,
Massachusetts Institute of Technology, Cambridge, MA
02139, USA. 2Optical Science Centre and ARC Centre of
Excellence in Future Low-Energy Electronics Technologies,
Swinburne University of Technology, Melbourne 3122, Australia.
*Corresponding author. Email: zwierlei@mit.edu.
†Present address: Department of Physics, University of California,
Berkeley, CA 94720.
nitude higher. The system features the largest
superfluid transition temperature Tc, relative to
its density, of all known fermionic systems (12).
In this study, we introduce a thermography
method to image heat in interacting quantum
gases. The method requires only a temperature-
dependent spectral response that can be locally
resolved. In the case of the Fermi gas that we
studied here, the radio-frequency (rf) spectrum
is temperature dependent (13, 14). We spatially
resolved this spectral response and directly mea-
sured heat transport in the strongly interacting
Fermi gas.
The nature of heat transport can help dis-
tinguish states of matter. In ordinary liquids,
heat transport is purely diffusive and governed
by thermal conductivity. By contrast, in super-
fluids, heat propagates as a wave called “second
sound.” The two-fluid model of superfluidity
introduces normal and superfluid components
that can move in and out of phase (15, 16). This
gives rise to two distinct sound modes, first and
second sound, corresponding to a density and
an entropy wave (17). The speed of second
sound c2 is a direct measure of the superfluid
fraction rS =rN, the ratio of the superfluid den-
sity rS to the normal component density rN
(18). Its attenuation yields the second-sound
diffusivity D2, which involves the thermal con-
ductivity, bulk, and shear viscosities (17, 19).
Consequently, we observe a dramatic change
in thermal transport as the Fermi gas is cooled
belowTc. Simultaneously recording the complete
density and heat response of the system to a
known external perturbation allows us to com-
pletely characterize the two-fluid hydrodynamics
of the strongly interacting Fermi gas (19, 20).
Previous studies of thermal transport in quan-
tum gases relied on the weak coupling between
the density and temperature of the gas (21–24).
This allowed the observation of second sound
in Bose (25, 26) and Fermi gases (21, 24, 27) but
without directly measuring heat propagation.
By using a homogeneous box potential formed by
unshifted response for an isolated atom. Con-
versely, at low temperatures, the spectra display
interaction-induced shifts known as “clock
shifts.” In the particular case of attractive two-
component Fermi gases, at zero temperature
the spectral peak is shifted by approximately
the pairing energy EB of fermion pairs (13),
and at nonzero temperature, broken pairs con-
tribute to the response at lower frequencies
(Fig. 1A). For a fixed detuning w0 on the flank
of the spectrum, the rf response is sensitive
to changes in temperature (Fig. 1B). As the rf
response can be spatially resolved, this allows
for a direct measurement of the local tem-
perature from a single image of rf-transferred
atoms.
As an exemplary application of this method,
we may detect second sound in the fermionic
superfluid, which is a wave in the gas of excita-
tions that, close to Tc , consists predominantly
of broken pairs (Fig. 1C). A suitably detuned rf
drive can transfer atoms from the gas of exci-
tations, yielding a direct, local measure of heat
(Fig. 1D). We stress that the method does not
depend on this simplified picture of broken
pairs and only relies on the temperature dep-
endence of the rf spectrum. It therefore applies
in a wide range of temperatures set by the mag-
nitude of clock shifts, which for the unitary
Fermi gas are on the scale of the Fermi tem-
perature (13).
Our experiment began with a uniform fermi-
onic superfluid trapped in a cylindrical box
potential whose axial direction is defined as
the z axis, formed by an equal mixture of re-
sonantly interacting fermions in the first (1)
and third (3) hyperfine state of 6Li at a Feshbach
resonance (magnetic field, 690 G) (29). The
density of n0 ¼ 0:75 mm3 per spin state corre-
sponds to a Fermi energy of EF ¼ h  10:5 kHz
and a Fermi temperature of TF ¼ EF =kB ≃
500 nK, where h is Planck’s constant and kB
is the Boltzmann constant. To create temper-
ature gradients in the superfluid gas, we res-
onantly excited a standing wave of second

Yan et al., Science 383, 629–633 (2024) 9 February 2024 1 of 5

RES EARCH | R E S E A R C H A R T I C L E
A B
C D
NF
SF
z two-dimensional (2D) temperature profile along
E F
x hydrodynamics for superfluids was provided
z Observation of heat propagation
Fig. 1. Direct local thermography using rf spec-
troscopy. (A) A sketch of rf spectra at various
temperatures for the unitary Fermi gas (13). Blue,
gray, and red lines correspond to the rf response
IðwÞ at successively higher temperatures. (B) At
fixed frequency w0 on the flank of a spectrum [black
dotted line in (A)], the rf response is sensitive to
temperature and serves as a local thermometer.
(C) In a simplified picture, the superfluid component
(SF) consists of fermion pairs, whereas the normal
fluid (NF) is composed of broken pairs. (D) The
unpaired atoms are transferred to a weakly inter-
acting state by an rf pulse and subsequently imaged
to determine the spatial distribution of the normal
component density. (E and F) In situ observation of
a second-sound wave after resonant gradient
excitation. Shown are the column density and local
temperature, respectively, from simultaneous in situ
absorption images of unperturbed (3) and rf-
transferred (2) atoms, with density n and
temperature variation DT, averaged along the x axis,
shown below. The vertical dotted line marks the
edge of the box potential (half maximum of
potential). The black dashed line in (F) is a fit to the
fundamental eigenmode in the box [eq. S1 in (18)].
Second sound has a significant effect on the
temperature, but not the density.
sound using an oscillating potential gradient
along the z axis (Fig. 2A). Our thermography
uses rf transfer of atoms from state 1 into the
initially unoccupied state f ≡ 2. Simultaneous
in situ absorption images of atoms in states 2
and 3 along one of the radial direction (y axis)
yield the original gas density nðx; z Þ (Fig. 1E),
as well as the density nf ðx; z Þ of rf-transferred
atoms, carrying the information on the local
temperature (Fig. 1F). The rf thermometer is
calibrated on gases in thermal equilibrium by
Yan et al., Science 383, 629–633 (2024) 9 February 2024
recording the dependence of nf on temper-
@T jn, and density, @n jT (18). This method
ature, @n @nf
f
of calibrating spectral responses versus each
thermodynamic variable while holding other
parameters constant can be applied universally.
More generally, all that is required for the ob-
servation of thermal transport is access to any
local observable that is sensitive to tempera-
ture, meaning that it can be achieved even with-
out a calibrated thermometer. Integrating the
the uniform x axis yields a 1D temperature pro-
file, DT ðz Þ, the deviation in temperature from
the equilibrium state, with a precision of 500 pK
from a single image, as shown in Fig. 1F. The data
reveal an essentially flat density in the presence of
a ~8-nK temperature difference across the box.
Armed with the ability to spatially resolve tem-
perature in the strongly interacting Fermi gas,
we directly observed second sound as the free
back-and-forth sloshing of heat after resonant
gradient excitation (Fig. 2, B to D). Figure 2B
shows the measured temperature variation
DT ðx; z; t Þ obtained at various times after second-
sound generation. Figure 2C presents the time
evolution of the 1D temperature profiles
DT ðz; t Þ, and Fig. 2D shows the correspond-
ing evolution of the amplitude DT ðk1 ; t Þ of
the first spatial Fourier mode supported by
the axial box length L ¼ 91 mm (kj ¼ jp=L), all
clearly demonstrating the wave-like propaga-
tion of heat. Here, the absolute temperature of
the gas in equilibrium, obtained from expan-
sion (14), was T ¼ 63ð2Þ nK ¼ 0:125ð5ÞTF ,
or T ¼ 0:75ð3Þ Tc when compared with the su-
perfluid transition temperature Tc ¼ 0:167TF
reported in (12). A damped sinusoidal fit
to DT ðk1 ; t Þ yielded a speed of second sound
of c2 ¼ w=k ¼ 3:57 ð2Þmm=s , corresponding
to about a tenth of the Fermi velocity c2 ¼
0:092ð2ÞvF . From the measured damping
rate G, we obtained a diffusivity of second sound
D2 ¼ G=k2 ¼ 2:44ð11Þℏ=m. As was found for
the diffusivities of spin (30, 31), momentum (32),
and first sound (33), a natural scale for the dif-
fusivity of second sound is reduced Planck’s
constant ℏ = h/2p divided by the particle mass
m (27, 34). This scale directly emerges in a
strongly interacting quantum fluid from a
mean-free path of carriers of approximately
one interparticle spacing d, and characteristic
speeds of ℏ=md given by Heisenberg’s uncer-
tainty (30). A similar scale of diffusivity is also
measured for second sound in the strongly
interacting bosonic superfluid 4He (35), where-
as the more weakly interacting fermionic 3He
in its superfluid A1 and B phases displays much
larger values that are many hundreds to thou-
sands of times ℏ=m (36).
Thermography provides an unprecedented
view of the superfluid transition in the strongly
interacting Fermi gas. Figures 2, F and G, show
the transition from heat diffusion in the normal
state to wave-like propagation of heat, second
sound, in the superfluid. For these data, we
created a local hotspot on one side of the box
by locally applying an intensity-modulated
optical grating (Fig. 2E). Modulation at ∼2 kHz
efficiently creates high-frequency phonons that
rapidly decay into heat (33, 37), creating a
temperature profile with good overlap with
the j ¼ 1 mode. The subsequent evolution of
the temperature amplitude DT ðk1 ; T Þ displays
a striking change in character from exponen-
tial decay above Tc to the damped sinusoid of
second sound below Tc .
Entropy and density response functions
The full linear response theory of two-fluid
over half a century ago by Hohenberg and
Martin (19). Under an external potential that
acts on the density n with wave vector k and
frequencyw, systems respond through changes
in their density n as well as their temperature
or equivalent entropy densitys. Thermography
enables us to obtain the corresponding response
functions, not only cn;n ðk; wÞ but alsocs;n ðk; wÞ.
These encode all the thermodynamic and two-
fluid hydrodynamic information of the unitary
Fermi gas (18–20).
To determine the linear response functions,
we apply a potential gradient, oscillating at fre-
quency w. The steady-state temperature change
DT ðk1 ; wÞ and density change Dnðk1 ; wÞ, mea-
sured after an integer number of oscillation
cycles, yield the respective out-of-phase response
functions (19, 20). The change in entropy per
particle, Ds, is linked to the temperature and
density variation by the equation of state.
For our scale invariant, unitary Fermi gas, this
connection is provided by the specific heat per
particle cV at constant density (11, 12)
 
DT 2 Dn
Ds ¼ cV  ð1Þ
T 3 n0
Measurements of fractional temperature and
density variations thus directly yield the entro-
py variation in units of cV . Figures 3, A and B,
display the entropy and density response of the
superfluid in a frequency range that solely
excites the lowest spatial mode (j ¼ 1), the
sloshing mode. The density reveals a dominant
peak attributed to first sound near 90 Hz
(33) and a faint signature of second sound at
20 Hz, expected in a gas of nonzero expan-
sivity, where density and temperature are
coupled. However, in the entropy channel,
whose signal derives predominantly from
the rf transfer (18), the strong second-sound
peak indicates a large response. This directly
demonstrates that second sound in the unitary
Fermi gas is predominantly an entropy wave,
whereas first sound is essentially isentropic.
This is similar to the case in superfluid 4He
2 of 5
RES EARCH | R E S E A R C H A R T I C L E

A Oscillating E Oscillating
7 µm
Gradient Optical Grating
NF
SF x

z z
B F G

Fig. 2. Direct observation of the superfluid transition from heat propaga-
tion in a strongly interacting Fermi gas. (A) Generating second sound with an
oscillating potential gradient for data shown in (B) to (D) at a temperature of
T ¼ 63 nK or 0:75Tc. (B) In situ thermographs at times t ¼ 0; 26, and 54 ms
after second-sound excitation. (C) Time evolution of the axial temperature
profiles, revealing the wave-like propagation of heat. (D) Amplitude of the first
spatial Fourier mode of the temperature profiles DT(k1, t) versus time (gray
circles). A fit to a damped sinusoid (dashed line) gives the speed and attenuation
rate of second sound. (E) Local heating with an intensity-modulated optical
grating for data shown in (F) to (G). (F) Time evolution of temperature
amplitudes DT(k1, t) (solid circles) and fits (dashed lines) at various gas
temperatures. The dotted lines show the DT ¼ 0 line for each temperature. The
fitting method used in (D) and (F) is indicated by eq. S24 in (18). (G) Two-
dimensional interpolation with Gaussian smoothing of temperature amplitudes
versus time across the superfluid transition. In (D) and (E), the initial
temperature variation for each time trace is normalized to be 1.

(17) but drastically different from the case in
2D and 3D Bose gases, in which density and
entropy are strongly coupled (25, 26, 38). In
Figs. 3, C and D, we show the thermal evo-
lution of the entropy and density responses
in the first spatial Fourier mode, which serve
as a direct measurement of the out-of-phase
entropy-density [Imcs;n ðk1 ; wÞ] and density-
density [Imcn;n ðk1 ; wÞ] response functions (18).
The measured response functions completely
encode all information about the two-fluid hy-
drodynamics in a unitary Fermi gas (18–20).
The peak positions and widths give the speeds
and diffusivities of first and second sound.
The speed of first sound is a direct measure
of the energy of the gas (33), and the speed of
second sound yields the superfluid density.
The height of the second-sound peak in the
entropy-density response is given by the ex-
pansivity ap of the gas, and the weight of the
second-sound versus the first-sound response
in the density-density response directly equals
g  1 , where g ¼ cp =cV is the ratio of heat
capacities at constant pressure and density.
The thermodynamic quantities ap and g are
related by the isothermal compressibility kT ,
the heat capacity, and temperature by g  1 ¼
T a2p =ðnkT cV Þ, and in particular for the uni-
tary gas simply by g  1 ¼ 23 ap T .
Heat transport across superfluid transition
Figure 4A shows the speed of second sound,
measured consistently with our three inde-
pendent methods: free evolution after resonant
excitation of the second-sound mode (yellow
squares), local heating (red diamonds), and
steady-state response functions (blue circles).
The superfluid fraction is obtained fromc2 and
the previously measured equation of state
(12, 18) and is shown in Fig. 4B. The measure-
ments show a qualitative agreement with
Nozières and Schmitt-Rink theory (39, 40)
(dotted-dashed line), although their absolute
value of Tc differs from experiment. Our super-
fluid fraction agrees well with the result re-
constructed for the homogeneous case from
the second-sound measurement in a quasi-1D
trapped gas in (21), which relied on the same
equation of state from (12). With the local
heating method (red diamonds), we are able
to observe the continuous evolution of c2 and
rS from a finite value in the superfluid phase
to zero in the normal phase. The phase-
transition temperature Tc obtained from this
measurement is consistent with the equilib-
rium thermodynamic measurement (12) (the
vertical gray area) and the onset of pair conden-
sation (7, 13), which we have measured here as
well (Fig. 4C). As is expected, there is a clear
quantitative difference between the superfluid
fraction, which saturates to unity at tempera-
tures T ≲ 0:1TF , and the pair condensate
fraction, which remains ≲0:75 . The super-
fluid density quantifies the portion of the
fluid that flows without friction. Formally, it
measures the rigidity against phase twists,

Yan et al., Science 383, 629–633 (2024) 9 February 2024 3 of 5

RES EARCH | R E S E A R C H A R T I C L E

A B A

C D B

Fig. 3. Steady-state entropy and density response of the unitary Fermi superfluid. Shown are the
(A) change in entropy per particle (Ds) and (B) density (Dn) after excitation by an integer number of
cycles of an oscillating axial potential gradient (Fig. 2E). For frequencies below 50 Hz, the drive duration is
5 cycles at an amplitude of g ¼ h  2:12 Hz=mm; for frequencies above 50 Hz, we drive for 20 cycles at an
oscillation amplitude of g ¼ h  0:85 Hz=mm. The gas temperature is T=Tc ¼ 0:75. Amplitudes of the Fig. 4. The speed and diffusivity of second
first spatial Fourier mode are shown in (C) and (D) for various temperatures in the superfluid phase. The sound. (A) Speed of second sound, normalized by
solid lines are fits using the full entropy- and density-response function from two-fluid hydrodynamics the Fermi velocity, as a function of temperature,
[eqs. S9 and S10 in (18)]. determined by fitting the steady-state response
functions (blue circles), and the free evolution of
whereas the condensate fraction is a measure ity is not precisely known but is estimated to second sound after resonant gradient excitation
for the number of fermion pairs at zero–center be on the order of Tc (43, 44). A quantitative (yellow squares) or after local heating (red diamonds).
of mass momentum. In the zero-temperature analysis of critical behavior, such as the mea- The first-sound speed measured from the response
limit, the entire system is superfluid, but only surement of critical exponents, is prevented by functions (gray circles) is also shown. The dotted-
a fraction of fermion pairs are condensed, the residual inhomogeneity of the gas den- dashed line indicates Nozières-Schmitt-Rink theory
owing to quantum depletion and Pauli block- sity, giving a variation of DðT =Tc Þ ∼ 5  103, (39). (B) The superfluid fraction of the unitary Fermi
ing (6, 7, 9). and by the finite size of our system. Indeed, gas obtained from the speed of second sound
A further dramatic signature of the super- even for the lowest spatial mode j ¼ 1, second [symbols as shown in (A); also see eq. S13 in (18)].
fluid transition is seen in the temperature sound becomes overdamped (G ≳ 2w) within The blue shaded area indicates the uncertainty from
dependence of the second-sound diffusivity D2 3% of Tc. At low temperatures T =Tc < 0:6, D2 the equation of state. Solid green circles indicate the
in the superfluid state, and thermal diffusion is again seen to rise significantly, which we superfluid fraction obtained from quasi-1D experiments
in the normal state, shown in Fig. 4D. We attribute to the diverging mean-free path of (21) that also utilized the MIT equation of state of
observe a striking peak in this transport co- phonons, the only remaining contribution at the unitary Fermi gas (12). (C) Pair condensate fraction
efficient within a range DT ≈ 0:1Tc around the low temperatures once pair-breaking exci- measured with the rapid-ramp technique to detect
critical temperature of superfluidity, rising above tations are frozen out. fermion pair condensates (13). (D) Second-sound
a background minimum value of about 2ℏ=m Above the transition temperature, the second- diffusivity obtained from various methods [symbols
up to nearly three times this value. This be- sound mode evolves into a thermal-diffusion as shown in (A)]. The vertical gray area shown in all
havior echoes that found in liquid 4He (35, 41) mode whose diffusivity is directly given by ther- panels indicates the uncertainty of critical temper-
near its superfluid transition, associated with mal conductivity k: D2 ¼ k=ncP (19, 20, 45, 46). ature from (12).
classical criticality. Indeed, the order param- We therefore find quantum-limited thermal
eters of both the Fermi superfluid and liquid diffusion ∼2ℏ=m (47), similar to prior results for sumably because Bragg scattering as a density
helium belong to the same 3D XY static uni- spin (30, 31), momentum (32), and first-sound probe becomes insensitive to heat propagation
versality class, and also the same [model F diffusion (33) in the unitary gas. However, the above Tc . Away from Tc , the values for D2 re-
in (42)] dynamic universality class, dictating nonmonotonous behavior of second-sound ported in (27) were about half of what we
a behavior D2 ºjTc  T jn=2 near the transi- diffusivity, with steep rise at low temperatures observed. Given that the experiment in (27)
tion, with critical exponent n ≈ 0:672, as ob- and around Tc, has not been observed in other used a much higher wave vector and corre-
served in 4He (41). Related critical behavior for transport coefficients. spondingly more elevated frequencies, the gas
the speed of second sound c2 ºðTc  T Þn=2 and The second-sound diffusivity D2 was inde- may no longer have been hydrodynamic but
rS ºðTc  T Þn is qualitatively consistent with pendently measured with Bragg scattering (27), instead entered the collisionless regime, which is
the steep slopes we observe close to Tc in and a small rise in the second-sound damping similar to the behavior for high-momentum first
these quantities. For the unitary Fermi gas, rate approaching Tc was observed. However, a sound in (33, 37). Assuming the hydrodynamic
the width of the region governed by critical- peak in D2 near Tc could not be resolved, pre- relation G ¼ D2 k2 for such modes will yield

Yan et al., Science 383, 629–633 (2024) 9 February 2024 4 of 5

RES EARCH | R E S E A R C H A R T I C L E
too small a value for D2 . By contrast, in the
present work using thermography, we veri-
fied hydrodynamic scaling by exciting also
the second ( j ¼ 2) spatial mode supported by
the box, finding within error bars identical
values of D2 [fig. S4 in (18)].
In the superfluid regime of the unitary Fermi
gas, there are three contributions to second-
sound diffusion: thermal conductivity k, shear
viscosity h, and bulk viscosity z3 from normal-
superfluid counterflow (36, 48). Although it is
known that z3 ¼ 0 for a pure phonon gas with
linear dispersion (49), in the range T =Tc ≳ 0:5,
the normal fluid is dominated by pair-breaking
excitations. In this case, all three contributions
are of similar importance (36, 48). Assuming
z3 ¼ 0 in this regime, as was done in (27), is
not warranted, and obtaining viscosity and
thermal conductivity from first- and second-
sound diffusion alone is not possible.
Outlook
Direct measurement of heat transport has been
a long-standing goal in quantum gas experi-
ments. Thermography now opens the door to
study a host of intriguing nonequilibrium phe-
nomena, from nonlinear heat waves to quench
dynamics (50, 51) and even far-from-equilibrium
phenomena such as prethermal states (52, 53).
Using tomographic imaging techniques (54), the
complete 3D spectral response can be measured,
enabling the investigation of transverse entropy
transport in anisotropic or inhomogeneous sys-
tems. For thermodynamic systems with addi-
tional degrees of freedom beyond density and
temperature, for example, spin-imbalanced
systems, additional independent measurements
such as probes of the local spin polarization can
be supplemented to fully determine thermody-
namic response functions. The spectral response
continues to serve as a channel highly sensitive
to temperature. Therefore, our spectroscopic
thermometry method may be applicable to
other quantum gas platforms, including Bose
Yan et al., Science 383, 629–633 (2024) 9 February 2024
gases, Bose-Fermi mixtures, impurity systems,
and Hubbard quantum simulators (55).
RE FERENCES AND NOTES
1. P. A. Lee, N. Nagaosa, X.-G. Wen, Rev. Mod. Phys. 78, 17–85
(2006).
2. M. G. Alford, L. Bovard, M. Hanauske, L. Rezzolla, K. Schwenzer,
Phys. Rev. Lett. 120, 041101 (2018).
3. A. Adams, L. D. Carr, T. Schäfer, P. Steinberg, J. E. Thomas,
New J. Phys. 14, 115009 (2012).
4. S. A. Hartnoll, Nat. Phys. 11, 54–61 (2015).
5. M. Blake, R. A. Davison, S. Sachdev, Phys. Rev. D 96, 106008
(2017).
6. S. Giorgini, L. P. Pitaevskii, S. Stringari, Rev. Mod. Phys. 80,
1215–1274 (2008).
7. W. Ketterle, M. W. Zwierlein, Riv. Nuovo Cim. 31, 247 (2008).
8. M. W. Zwierlein, Novel Superfluids (Oxford Univ. Press, 2015),
pp. 269–422.
9. W. Zwerger, in Quantum Matter at Ultralow Temperatures,
M. Inguscio, W. Ketterle, S. Stringari, G. Roati, Eds.
(IOS Press, 2016), pp. 63–142.
10. S. Krinner, T. Esslinger, J.-P. Brantut, J. Phys. Condens. Matter
29, 343003 (2017).
11. T.-L. Ho, Phys. Rev. Lett. 92, 090402 (2004).
12. M. J. H. Ku, A. T. Sommer, L. W. Cheuk, M. W. Zwierlein,
Science 335, 563–567 (2012).
13. B. Mukherjee et al., Phys. Rev. Lett. 122, 203402 (2019).
14. Z. Yan et al., Phys. Rev. Lett. 122, 093401 (2019).
15. L. Tisza, Nature 141, 913 (1938).
16. L. Landau, Phys. Rev. 60, 356–358 (1941).
17. I. M. Khalatnikov, An Introduction to the Theory of Superfluidity
(CRC Press, 1965).
18. See supplementary materials.
19. P. Hohenberg, P. Martin, Ann. Phys. 34, 291–359 (1965).
20. L. P. Kadanoff, P. C. Martin, Ann. Phys. 24, 419–469 (1963).
21. L. A. Sidorenkov et al., Nature 498, 78–81 (2013).
22. J.-P. Brantut et al., Science 342, 713–715 (2013).
23. D. Husmann et al., Proc. Natl. Acad. Sci. U.S.A. 115, 8563–8568 (2018).
24. D. K. Hoffmann et al., Nat. Commun. 12, 7074 (2021).
25. P. Christodoulou et al., Nature 594, 191–194 (2021).
26. T. A. Hilker et al., Phys. Rev. Lett. 128, 223601 (2022).
27. X. Li et al., Science 375, 528–533 (2022).
28. S. Gupta et al., Science 300, 1723–1726 (2003).
29. B. Mukherjee et al., Phys. Rev. Lett. 118, 123401 (2017).
30. A. Sommer, M. Ku, G. Roati, M. W. Zwierlein, Nature 472,
201–204 (2011).
31. C. Luciuk et al., Phys. Rev. Lett. 118, 130405 (2017).
32. C. Cao et al., Science 331, 58–61 (2011).
33. P. B. Patel et al., Science 370, 1222–1226 (2020).
34. Z. Yan et al., Bull. Am. Phys. Soc. 64, Abstract D07.00006 (2019).
35. W. B. Hanson, J. R. Pellam, Phys. Rev. 95, 321–327 (1954).
36. D. Vollhardt, P. Wölfle, The Superfluid Phases of Helium 3
(Taylor & Francis, 1990).
37. C. C. N. Kuhn et al., Phys. Rev. Lett. 124, 150401 (2020).
38. L. Verney, L. Pitaevskii, S. Stringari, Europhys. Lett. 111, 40005 (2015).
39. E. Taylor et al., Phys. Rev. A 80, 053601 (2009).
40. P. Nozières, S. Schmitt-Rink, J. Low Temp. Phys. 59, 195–211 (1985).
41. L. S. Goldner, N. Mulders, G. Ahlers, J. Low Temp. Phys. 93,
131–182 (1993).
42. P. C. Hohenberg, B. I. Halperin, Rev. Mod. Phys. 49, 435–479 (1977).
43. E. Taylor, Phys. Rev. A 80, 023612 (2009).
44. T. Debelhoir, N. Dupuis, Phys. Rev. A 93, 023642 (2016).
45. B. Frank, W. Zwerger, T. Enss, Phys. Rev. Res. 2, 023301
(2020).
46. P. Patel, “Quantum transport in strongly interacting, ultracold
Fermi gases in box potentials,” thesis, Massachusetts Institute of
Technology (2022).
47. X. Wang, X. Li, I. Arakelyan, J. E. Thomas, Phys. Rev. Lett. 128,
090402 (2022).
48. H. Smith, H. H. Jensen, Transport Phenomena (Clarendon Press, 1989).
49. M. A. Escobedo, M. Mannarelli, C. Manuel, Phys. Rev. A 79,
063623 (2009).
50. M. W. Zwierlein, C. H. Schunck, C. A. Stan, S. M. F. Raupach,
W. Ketterle, Phys. Rev. Lett. 94, 180401 (2005).
51. P. Dyke et al., Phys. Rev. Lett. 127, 100405 (2021).
52. M. Gring et al., Science 337, 1318–1322 (2012).
53. C. Eigen et al., Nature 563, 221–224 (2018).
54. M. J. H. Ku, B. Mukherjee, T. Yefsah, M. W. Zwierlein, Phys. Rev.
Lett. 116, 045304 (2016).
55. C. Gross, I. Bloch, Science 357, 995–1001 (2017).
56. Z. Yan et al., Replication Data for: Thermography of the superfluid
transition in a strongly interacting Fermi gas, version 1, Harvard
Dataverse (2023); https://doi.org/10.7910/DVN/KKLTXW.
AC KNOWLED GME NTS
We thank B. Svistunov, N. Prokof’ev, W. Zwerger, and in particular
the late L. Pitaevskii for illuminating discussions. Funding: This
work was supported by the NSF (Center for Ultracold Atoms Award
Nos. PHY-1734011 and PHY-2012110), the Air Force Office of
Scientific Research (FA9550-16-1-0324 and MURI Quantum Phases
of Matter FA9550-14-1-0035), the Office of Naval Research
(N00014-17-1-2257), and the Vannevar Bush Faculty Fellowship
(ONR No. N00014-19-1-2631). Author contributions: Z.Y., P.B.P.,
B.M., and R.J.F. contributed to building the experimental setup.
Z.Y., P.B.P., and R.J.F. performed the measurements, and analyzed
the data. All authors contributed to the interpretation of the data
and the preparation of the manuscript. Competing interests: The
authors declare that they have no competing interests. Data and
materials availability: All data shown in this work can be found at
Harvard Dataverse (56). License information: Copyright © 2024 the
authors, some rights reserved; exclusive licensee American
Association for the Advancement of Science. No claim to original US
government works. https://www.science.org/about/science-
licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adg3430
Materials and Methods
Supplementary Text
Figs. S1 to S10
References (57, 58)
Submitted 28 December 2022; accepted 10 January 2024
10.1126/science.adg3430
5 of 5
RES EARCH

METALLURGY istence of coarse columnar b grains and a

heterogeneous distribution of phases (Fig. 1,
Ultrauniform, strong, and ductile 3D-printed titanium A to C). This results in highly nonuniform,
position-dependent tensile properties from
alloy through bifunctional alloy design L-PBF, as we demonstrate (Fig. 1, D and E) and
as has been demonstrated in other studies across
Jingqi Zhang1†, Michael J. Bermingham1†, Joseph Otte2, Yingang Liu1, Ziyong Hou3,4,5, Nan Yang1, multiple 3D printing technologies (27–29).
Yu Yin1, Mohamad Bayat6, Weikang Lin1, Xiaoxu Huang3,4, David H. StJohn1, Matthew S. Dargusch1* We show that the single addition (up to 5.0 wt %)
of elements from the b-isomorphous group
Coarse columnar grains and heterogeneously distributed phases commonly form in metallic alloys [in this case, we selected Mo; see section on
produced by three-dimensional (3D) printing and are often considered undesirable because they can alloy design in the materials and methods (30)]
impart nonuniform and inferior mechanical properties. We demonstrate a design strategy to unlock into Ti-5553 powder to form a composite blend
consistent and enhanced properties directly from 3D printing. Using Ti−5Al−5Mo−5V−3Cr as a model achieves bifunctionality: (i) During 3D printing,
alloy, we show that adding molybdenum (Mo) nanoparticles promotes grain refinement during some of the Mo particles partially melt, but the
solidification and suppresses the formation of phase heterogeneities during solid-state thermal cycling. core survives to nucleate fine grains during solid-
The microstructural change because of the bifunctional additive results in uniform mechanical properties ification and prevent coarse columnar grains
and simultaneous enhancement of both strength and ductility. We demonstrate how this alloy can be from forming. (ii) The dissolved Mo solute
modified by a single component to address unfavorable microstructures, providing a pathway to achieve stabilizes the b phase and suppresses the forma-
desirable mechanical characteristics directly from 3D printing. tion of isothermal w phases and a phases under
solid-state thermal cycling. As a result, the micro-

T
hree-dimensional (3D) printing or additive
manufacturing (AM) of metals and alloys
typically involves multiple physical and
metallurgical phenomena that impart
complex microstructures and varied mech-
anical properties in the fabricated products (1–7).
For metallic alloys, which solidify with a cubic
crystal structure, columnar grains often prevail
in the 3D-printed part (8, 9) because grains
with the easy growth <100> directions tend
to align closely to the maximum temperature
gradient of the melt pool and grow epitaxially
from the partially melted layers (10). Although
this highly textured columnar grain structure
can be beneficial for certain applications, in
most cases it is undesirable because it de-
grades the mechanical performance and results
in mechanical property anisotropy (11, 12). Ac-
cordingly, extensive effort has been devoted to
transforming the coarse columnar grain struc-
ture into fine equiaxed grains to achieve su-
perior and isotropic mechanical properties (13–17).
Generally, the columnar-to-equiaxed transition
(CET) and grain refinement can be promoted
through process control and by adding grain-
refining inoculants (12, 18). The former typically
includes manipulating the 3D printing process-
ing parameters or introducing external inter-
ferences (16, 17). However, the effectiveness and
practicalities of this approach are limited to
specific alloys or 3D printing technologies.
1
School of Mechanical and Mining Engineering, The
University of Queensland, St. Lucia, Brisbane, QLD, Australia.
2
Centre for Microscopy and Microanalysis, The University of
Queensland, St. Lucia, Brisbane, QLD, Australia.
3
International Joint Laboratory for Light Alloys (Ministry of
Education), College of Materials Science and Engineering,
Chongqing University, Chongqing, China. 4Shenyang National
Laboratory for Materials Science, Chongqing University,
Chongqing, China. 5Department of Materials Science and
Engineering, KTH Royal Institute of Technology, Stockholm,
Sweden. 6Department of Mechanical Engineering, Technical
University of Denmark, Lyngby, Denmark.
*Corresponding author. Email: m.dargusch@uq.edu.au
†These authors contributed equally to this work.
Alternatively, the metallurgical approach through
additives has proven highly effective but often
results in an undesirable loss in ductility owing
to the formation of brittle second phases (9).
Therefore, simultaneously addressing the coarse
columnar grains and eliminating property anis-
otropy without adversely affecting the ductility
is highly desirable.
A further complication is that many allotropic
alloy systems, including titanium alloys, are also
susceptible to the heterogeneous distribution
of phases associated with the solid-state thermal
cycling experienced during the 3D printing pro-
cess (19–22). This poses an additional challenge
to achieve uniform mechanical properties of
3D-printed parts made from these alloys (23, 24).
The localized heating, cooling, and reheating
nature of 3D printing effectively creates dy-
namic in situ heat treatments that encourage
the decomposition of initially formed phases
and/or the precipitation of new phases through
solid-state phase transformations (20, 25). Be-
cause thermal cycles are spatially variable, the
associated heat treatments can produce an
inhomogeneous distribution of phase along
the building direction of the part, thereby
resulting in the spatial variation of mecha-
nical properties (21, 22). Postprinting heat
treatments can be effective in mitigating these
phase heterogeneities but introduce delays
and additional costs and are not effective
in refining textured columnar grains (26).
The confluence of these issues has made it
extremely challenging to achieve uniform
and superior mechanical properties in the
as-fabricated state.
We demonstrate a design strategy to address
this challenge by simultaneously controlling
the grain structure and constituent phases in
products manufactured by laser powder bed
fusion (L-PBF). We selected the Ti−5Al−5Mo−
5V−3Cr (Ti-5553) metastable b titanium alloy
as a model alloy because it shows the coex-
structural change because of the bifunctional Mo
addition not only results in an improved uniform-
ity in mechanical properties, but simultaneously
enhances the strength and ductility. In general,
the design strategy provides a pathway for achiev-
ing uniform and enhanced mechanical properties
in parts produced by 3D printing.
Mechanical properties
We added 2.5 wt % and 5.0 wt % Mo to Ti-5553
(which we refer to as Ti-5553+2.5Mo and Ti-
5553+5Mo, respectively) by mechanical mixing
and produced the Ti-5553 parts with and with-
out Mo additions using refined L-PBF process-
ing parameters for titanium alloys (30). Given
the fact that the part dimension and size can
affect the thermal history and consequently the
mechanical properties of the part (31), we adopted
two types of part geometries (dog-bone– and
cuboid-shaped parts; fig. S2) to evaluate the
effect of Mo additions on mechanical proper-
ties. For simplicity, we discuss the representa-
tive data of Ti-5553+5Mo unless otherwise
stated. We compare the tensile engineering
stress-strain curves of Ti-5553 and Ti-5553+5Mo
specimens (Fig. 2A), which were machined from
the cuboid parts. In marked contrast to Ti-5553,
which exhibits inferior tensile properties and
substantial property variation throughout the
parts, Ti-5553+5Mo shows enhanced and more
uniform mechanical properties (see fig. S3
for fracture analysis). We also found the high
consistency in mechanical properties of Ti-
5553+5Mo in the dog-bone–shaped parts (fig.
S4), regardless of the change in part geom-
etry. To evaluate the degree of anisotropy of
the tensile ductility, we compared the tensile
ductility data of Ti-5553+5Mo with those of
Ti-5553 and similar alloys that have a small
deviation from the chemical composition of Ti-
5553—that is, Ti−5Al−5Mo−5V−3Cr−1Zr (Ti-55531)
and Ti−5Al−5Mo−5V−1Cr−1Fe (Ti-55511) (Fig. 2B)
(27, 28, 32–34). Generally, a substantial deviation
of the ductility data point from the blue dashed

Zhang et al., Science 383, 639–645 (2024) 9 February 2024 1 of 7

RES EARCH | R E S E A R C H A R T I C L E

Fig. 1. Microstructures and mechanical properties of Ti-5553 produced by
L-PBF. (A) The coexistence of coarse columnar b grains and spatially dependent
phases in Ti-5553 produced by L-PBF. (1) Schematic illustration of the L-PBF
process. (2) EBSD IPF map showing coarse columnar b grains along the building
direction (BD). (3) SEM-BSE micrographs showing the phase distribution along
the BD. The yellow arrows point out a phases with a darker contrast in the
b-Ti matrix. (B) Schematic illustration of the microstructure heterogeneity in
terms of columnar b grains and heterogeneously distributed phases on the
cross-section S−S (the yz-plane), as indicated in (A). (C) TEM micrographs of
Ti-5553. (1) Dark-field TEM image showing a phases. (2) TEM SAED pattern from
the [111]b zone axis showing the presence of a phases. The key diagram of
the diffraction is shown in (3). Note that there are three a variants with the
 
same zone axis of 2110 , which grow in different directions. (4) TEM SAED from
the [113]b zone axis showing the existence of isothermal w phases. The key
diagram of the diffraction is shown in (5). Note that there are two w variants.
(D) Tensile engineering stress-strain curves of Ti-5553 horizontal tensile
specimens (1, 3, 5, 7, 9, and 11). The inset shows how the horizontal tensile
specimens were machined from the Ti-5553 part. (E) Tensile engineering stress-
strain curves of Ti-5553 vertical tensile specimens (1, 3, 5, 7, 9, and 11). The inset
shows how the vertical tensile specimens were machined from the Ti-5553 part.

line indicates a high degree of anisotropy in the
tensile ductility. Ti-5553+5Mo clearly shows
higher and more isotropic ductility compared
with Ti-5553 and similar alloys.
We also compared the yield strength and the
elongation to failure of Ti-5553+5Mo with those
of Ti-5553 (as well as those of Ti-55531 and
Ti55511) produced in the as-fabricated L-PBF
state and under postprinting heat treatment
conditions (Fig. 2C) (29, 32, 34–38). Compared
with Ti-5553 and its similar alloys in the as-
fabricated state, Ti-5553+5Mo shows compara-
ble yield strength but notably higher ductility.
Postprinting heat treatment is commonly used
to balance the mechanical properties of Ti-
5553 produced by L-PBF. Although a high yield
strength (>1100 MPa) can be achieved under
certain heat treatment conditions, the ductil-
ity often deteriorates substantially, with an
elongation to failure of <10%, which limits the
use in safety-critical applications. For exam-
ple, a minimum elongation to failure of 10% is
recommended for the use of Ti−6Al−4V, which
is the so-called workhorse in the titanium in-
dustry (20). By contrast, without the need for
downstream heat treatments, Ti-5553+5Mo shows
an excellent balance of strength and ductility
directly from L-PBF, which makes it stand out
from Ti-5553 and the associated similar alloys
in the strength-ductility map.
Grain structures
The exceptionally uniform and enhanced mech-
anical properties of Ti-5553+5Mo relative to
Ti-5553 require a careful examination of mi-
crostructures and the associated underlying
mechanisms. We first performed microfocus
computed tomography (micro-CT) of Ti-5553
and Ti-5553+5Mo to evaluate the part quality.
Overall, both Ti-5553 and Ti-5553+5Mo show a
very high density, with total pore volume frac-
tions of 0.004024% and 0.001589%, respectively
(fig. S5 and table S1). Such a high density indi-
cates that porosity is unlikely to be responsible
for the highly scattered tensile properties of
Ti-5333 (Fig. 1, D and E) and is also in line with
the high consistency in mechanical perform-
ance of Ti-5553+5Mo (Fig. 2A).
To reveal the effect of Mo additions on
the grain structure, we carried out electron
backscatter diffraction (EBSD) characteriza-
tion of Ti-5553 and Mo-doped Ti-5553 (Fig. 3;
the EBSD maps of Ti-5553+2.5Mo are provided
in fig. S6). The microstructures of Ti-5553 con-
sist of relatively large grains along the scanning
direction (Fig. 3A, panel 1) and coarse columnar
b grains along the building direction (Fig. 3A,
panel 2), which exhibit strong crystallographic
textures (fig. S7A). The addition of 5.0 wt % Mo
to Ti-5553 results in a notable change in the
grain structure and associated crystallographic
texture. Numerous fine equiaxed grains (~20 mm
in diameter) are evident, forming along the edges

Zhang et al., Science 383, 639–645 (2024) 9 February 2024 2 of 7

RES EARCH | R E S E A R C H A R T I C L E

Fig. 2. Mechanical properties of Ti-5553 and Ti-5553+5Mo produced by
L-PBF. (A) Tensile engineering stress-strain curves of Ti-5553 and Ti-5553+5Mo
specimens, which were machined from the cuboid parts (as shown in the inset).
Note that the tensile curves of Ti-5553+5Mo are from the selected tensile specimens
(1, 3, 5, 7, 9, and 11), which are the same as Ti-5553 as shown in Fig. 1, D and E.
(B) Comparison of the vertical and horizontal elongation with failure of Ti-5553,
Ti-55531, Ti-55511, and Ti-5553+5Mo produced by multiple 3D printing
technologies with and without postprinting heat treatment (HT). Error bars represent
the standard deviation of the elongation to failure, but some are smaller than the size
of the data symbols. LMD, laser metal deposition; WAAM, wire arc additive
manufacturing. (C) Comparison of the tensile properties of Ti-5553+5Mo with those
of Ti-5553, Ti-55531, and Ti-55511 in both as-built L-PBF and he43at-treated states.

of scanning tracks of Ti-5553+5Mo (Fig. 3B,
panel 1). By contrast, the microstructure of Ti-
5553+5Mo is characterized by fine equiaxed
grains and narrow columnar grains along the
building direction (Fig. 3B, panel 2). Close
examination of the microstructure reveals
a periodic distribution of fine and columnar
grains. Unlike in Ti-5553, where the highly
textured columnar grains span across multiple
layers, the length scale of columnar grains in Ti-
5553+5Mo is determined by the melt pool
size, and the crystallographic texture becomes
random and weak (fig. S7B).
During solidification, the segregating alloy
solute can restrict the growth of the solid phase
by generating constitutional supercooling (DTCS),
which plays a key role in promoting the CET and
grain refinement (39, 40). Two parameters are
often used to provide an indication of how
likely a solute is to generate constitutional
supercooling and promote the CET. The first is
the growth restriction factor Q, which repre-
sents the initial rate that DTCS develops. The
second is the freezing range (DTFR), also known
as the supercooling parameter, which repre-
sents the maximum possible DTCS that can
develop (41). Despite being heavily alloyed,
Ti-5553 contains alloying elements Al, Mo, V,
and Cr, which show rather limited constitu-
tional supercooling (in terms of Q and DTFR)
in titanium (18). The total Q and DTFR values
based on the real chemical composition of
Ti-5553 are calculated to be about 10.4 K and
10.9 K, respectively (30) (table S3). These values
are far below those reported in other studies
to achieve substantial grain refinement by
solute—for example, Q = 62 K and DTFR = 603 K
in Ti−8.5Cu (9, 42). Therefore, it is unsurprising
that the L-PBF–produced Ti-5553 shows the
typical columnar grain structure (Fig. 3A). The
addition of Mo (2.5 wt % and 5.0 wt %) to Ti-5553
results in a pronounced change in the grain
structure (Fig. 3B and fig. S6). Such a notable
microstructural change could not be primarily
attributed to the generation of constitutional
supercooling because the Mo addition up to
5.0 wt % results in maximum total alloy Q and
DTFR values of 12.5 K and 12.8 K, respectively
(table S3), which assumes that all of the added
Mo particles have fully dissolved in the titanium
matrix. However, a dense population of residual
Mo particles exists in the microstructure (fig. S5,
D and E), which indicates incomplete dissolu-
tion of the added Mo particles. Despite the
apparent lack of Mo solute toward DTCS, the
transient nature of Mo particle dissolution
and potential higher local concentration around
the particles (>5.0 wt % Mo) may generate fur-
ther constitutional supercooling and help contrib-
ute toward the grain refinement in concert with
other mechanisms.
Mo-rich particles
To gain a deeper insight into the role of Mo
addition in solidification, we then focused on
the topmost layers of Ti-5553 and Ti-5553+5Mo.
Unlike the lower region of the fabricated part,
where the microstructure can be substantially
altered owing to the deposition of subsequent
layers, the topmost layer only undergoes one
or several thermal cycles because of the depo-
sition of neighboring tracks. In principle, such

Zhang et al., Science 383, 639–645 (2024) 9 February 2024 3 of 7

RES EARCH | R E S E A R C H A R T I C L E

Fig. 3. Microstructure characterization of Ti-5553 and Ti-5553+5Mo.
(A) EBSD IPF maps (overlaid with the image quality map and high-angle grain
boundaries with the misorientation angle ≥10°) of the xy-plane (along the
scanning direction) (1) and the xz-plane (along the building direction) (2) of the
Ti-5553 sample. (B) EBSD IPF maps (overlaid with image quality map and
high-angle grain boundaries with the misorientation angle ≥10°) of the xy-plane
(1) and xz-plane (2) of the Ti-5553+5Mo sample. (C) EBSD maps of the top
surface layers of Ti-5553. (1) IPF map overlaid with high-angle grain boundaries.
(2) KAM map. (D) EBSD maps of the top surface layers of Ti-5553+5Mo. (1)
IPF map overlaid with high-angle grain boundaries. (2) KAM map. (E) SEM-BSE
micrograph of the top surface of Ti-5553+5Mo. (F) SEM-EDX maps showing
the elemental distribution in the region in (E).

thermal cycling is much weaker compared with
that experienced by the lower region, and hence
the topmost layer can provide a window for
understanding the role of Mo in solidification.
We show the EBSD inverse pole figure (IPF)
and kernel average misorientation (KAM) maps
of the microstructure close to the top surfaces
of Ti-5553 (Fig. 3C) and Ti-5553+5Mo (Fig. 3D),
respectively. Ti-5553 exhibits coarse columnar
grains in the upper layer (Fig. 3C), whereas Ti-
5553+5Mo is dominated by fine equiaxed grains
(Fig. 3D). Further characterization of Ti-5553+
5Mo by scanning electron microscopy (SEM)–
backscattered electrons (BSE) shows that there are
some particles located at the center of the equiaxed
dendrite grains (Fig. 3E and fig. S8), which is a
typical characteristic of a nucleating particle
seeding a grain. Such particles are enriched
in Mo, as evidenced by SEM–energy-dispersive
x-ray spectroscopy (EDX) mapping (Fig. 3F).
To further explore these Mo-rich particles,
we performed SEM-EDX and transmission elec-
tron microscopy (TEM)–EDX line-scanning
across the interface between the Mo-rich par-
ticles and the titanium matrix (fig. S9 and
Fig. 4A). The concentration profile obtained
using SEM-EDX shows that there is a gradual
change from the Mo particle to the titanium
matrix (fig. S9, B to D), which is also confirmed by
TEM-EDX line-scanning (Fig. 4B). Additionally,
the high-resolution TEM imaging shows that the
Mo-rich particle matches perfectly with the tita-
nium matrix with a coherent interface (Fig. 4C).
Our SEM and TEM observations suggest
that the transitional region across the Mo par-
ticle and titanium matrix plays an important
role in grain refinement. Because these charac-
terizations were performed on the samples that
have undergone both solidification and solid-
state transformations during L-PBF, it remains
unclear whether the transitional region at the
interface develops in the melt pool or results
from solid-state diffusion after solidification.
Unfortunately, this process cannot readily be
characterized by current experimental tech-
niques. To help understand the dissolution
of Mo particles during solidification, we per-
formed diffusion simulation using DICTRA
(diffusion-controlled transformation) (Fig. 4D).
The simulation is based on the assumption that
an individual Mo particle with a radius of 2.5 mm
is presented in the Ti-5553 melt pool (with a
simulation cell of 100 mm) and then evolves
during the heating and the subsequent cooling
processes before solidification. It is evident that
during the heating process (from time = 0 s to
time = 7.68 × 10−4 s) and the cooling process
(from time = 7.68 × 10−4 s to time = 1.54 × 10−3 s),
significant diffusion of Mo solute from the Mo
particle to the Ti-5553 melt leads to a Mo-rich
region surrounding the particle. Careful exami-
nation of the composition profiles of Mo in the
DICTRA simulation and the SEM-EDX charac-
terization reveals that the dissolution region
from simulation is slightly larger than that from
experimental characterization. This observation
can be explained by the fact that, under realistic
conditions, the fluid flow within the melt pool
encourages the Mo-rich solute to rapidly diffuse
and mix with the bulk liquid, thereby leading
to a narrower Mo-rich region. Nevertheless,
the DICTRA simulation supports our hypoth-
esis that the transitional region across the Mo
particle and titanium matrix can occur during
the moments before and during solidification,
although it would be further facilitated by sub-
sequent solid-state diffusion during cooling
after solidification.
On the basis of the experimental and sim-
ulation results, it appears clear that the con-
tribution of Mo additions to grain refinement

Zhang et al., Science 383, 639–645 (2024) 9 February 2024 4 of 7

RES EARCH | R E S E A R C H A R T I C L E

Fig. 4. TEM characterization and DICTRA simulation of the interface
between Mo particles and the titanium matrix. (A) Bright-field TEM
micrograph showing the interface of a Mo particle and the titanium matrix. Insets
show the SAEDs from the [011] zone axis of Mo and b-Ti. (B) TEM-EDX line-
scanning showing the compositional profile across the interface of the Mo
particle and titanium matrix. (C) High-resolution TEM (HRTEM) image showing
the coherent interface between the Mo particle and titanium matrix (1), along
with the SAEDs of Mo (2) and Ti matrix (3) taken from the ½113 zone axis.
(D) DICTRA simulation showing the diffusion across the Mo particle and titanium
melt during the heating and cooling processes of L-PBF. The inset shows a
higher magnification of the composition profile of Mo and the boundary
movement resulting from the dissolution process.

works in two ways: (i) The solid cores that
survive in partially dissolved Mo particles act
as heterogeneous nucleation sites for the new
grains. (ii) The Mo solute associated with the
dissolution of Mo particles surrounding the
solid core potentially generates a supercooled
region that enhances heterogeneous nucleation
efficiency of the partially melted Mo particles.
We schematically illustrate the mechanism of
grain refinement in fig. S10. Upon being heated
to above the melting point of Ti-5553, some Mo
particles undergo partial dissolution during
L-PBF (fig. S10, A and B, stages a and b). Mo
has a very high melting point of 2783 K—much
higher than the 1877 to 1933 K of the b-Ti phase
in Ti-5553 (43)—and is thermodynamically quasi-
stable in the titanium melt. Given the high
reactivity of liquid titanium, such character-
istics are essential to ensure that some solid
Mo particles survive in the melt long enough
for grain nucleation. During the laser melting
process, the partial dissolution of Mo particles
creates an enveloping liquid enriched in Mo
solute (fig. S10, A and B, stage b). Although the
Mo-rich solute in this enveloping liquid is ex-
pected to rapidly diffuse and mix with the bulk
liquid, the continued dissolution (or melt-back)
of the particle provides a steady flux of Mo
solute. The local Mo composition in this region
may reach very high levels well beyond the no-
minal bulk alloy composition. The DICTRA simu-
lation supports concentrations of ~20 wt % Mo
extending a few micrometers into the melt;
however, a very steep concentration gradient
closer to the particle-liquid interface exists, so
the local concentration may exceed this. The
local Q and DTFR values increase to ~18.5 K at
20 wt % Mo but would increase further with
the concentration of Mo solute. This Mo-rich
liquid surrounding the surviving solid Mo core
may provide a supercooled zone if the tempera-
ture of the melt pool drops rapidly (fig. S10B,
stages c and d). Although this supercooled zone
arises as a result of localized composition varia-
tions, its generation is conceptually different
from the traditional notion of constitutional
supercooling, which develops as a result of
the solute rejection into the melt ahead of a
growing solid-liquid interface, thereby restrict-
ing the growth of the solid phase (39). In the
current case, the surrounding enriched Mo
solute has developed from Mo particle dis-
solution, creating a rich transient Ti-Mo liquid
(richer in Mo compared with the nominal alloy
composition) that facilitates the development
of a supercooled zone ahead of the particle as
the melt temperature falls (fig. S10B, stages
c and f). Consequently, the confluence of these
factors and the presence of multiple Mo par-
ticles spatially distributed throughout the melt
results in the fine equiaxed grain structure of
Ti-55553+5Mo.
Constituent phases
In addition to refining the grain structure, the
Mo addition also alters the constituent phases
in the microstructure in the solid state. Al-
though some of the larger Mo powders evi-
dently survive to nucleate equiaxed grains,
many of the smaller particles added are likely
fully dissolved in the melt pool and therefore
increase the overall Mo solute concentration
in the Ti-5553 alloy [see (30) for particle size
distribution]. This inevitably affects the phase
stability and microstructural evolution. We show
the phase distribution along the building direc-
tion of the Ti-5553 part (Fig. 5, A and C), and, as
expected, there is a clear variation in the dis-
tribution of a phase from the bottom to the
top of the part. The presence of a phase is fur-
ther confirmed by x-ray diffraction (XRD) (Fig.

Zhang et al., Science 383, 639–645 (2024) 9 February 2024 5 of 7

RES EARCH | R E S E A R C H A R T I C L E

5B). The TEM selected-area electron diffraction
(SAED) pattern in Fig. 1C has some discrete
diffraction spots, which are the signature for
the isothermal w phase. This observation is con-
sistent with the coexistence of the isothermal
w phase and the a phase in Ti-5553 upon being
heat treated in the temperature range of 523
to 773 K (44, 45). Our differential scanning calo-
rimeter (DSC) measurement further supports
the formation of these phases in this temper-
ature range (fig. S11). The addition of Mo tends
to stabilize the b phase and suppress the pre-
cipitation of the a phase. We observed that the
addition of Mo up to 5.0 wt % results in an
intensity reduction of the a phase in the XRD
spectra (Fig. 5B). Although other phases possib-
ly may be present in Ti-5553+5Mo that we
could not detect within the sensitivity of our
XRD, if other phases do exist, they have no
notable effect on the mechanical properties or
uniformity. In contrast to Ti-5553, Ti-5553+5Mo
shows solidification cellular structures along
the building direction (Fig. 5, D and E) without
any evidence of the needle-like a phase. Further
SEM-EDX of the cellular structures reveals that
the boundary (dark region) of the cellular struc-
tures is enriched in Ti solute (Fig. 5E and fig. S12).
This observation is expected when Ti is alloyed
with b-isomorphous elements such as Ta, W,
Nb, and Mo because these elements have
partition coefficients greater than unity (18).
The formation of cellular structures has also
been reported in high-solute–containing alloys
produced by L-PBF—for example, 316L stain-
less steel (3), Ni-based superalloy (46), and
Ti−42Nb alloy (47)—and is commonly asso-
ciated with the microsegregation of solute
in solidification.
By promoting grain refinement during solid-
ification and stabilizing the b phase under
solid-state thermal cycling, the single, small Mo
addition results in two notable changes in the
mechanical property of Ti-5553: (i) a simulta-
neously improved yield strength and ductility
and (ii) an enhanced uniformity in mechanical
properties (Fig. 2A and table S2). The strength
increase is attributed primarily to the combined
effect of solid solution strengthening of Mo
solute and grain refinement, based on the esti-
mation of strengthening contributions (30) (fig.
S16). Unlike the addition of b-eutectoid elem-
ents, which increase the strength at the expense
of ductility (48), the Mo addition simulta-
neously enhances the strength and ductility.
The ductility improvement is mainly attributed
to the suppression of precipitation of metastable
phases by Mo solute as opposed to other po-
tential effects, such as the transformation-
induced plasticity effect (30). The coherent
interface between the Mo particle and Ti alloy
may not adversely affect the fatigue perform-
ance (30). Furthermore, the nanoparticles con-
tribute to the high uniformity of mechanical
properties of Ti-5553+5Mo because of their
bifunctional role in the elimination of both
textured columnar grain and phase hetero-
geneities. The primary aim of our design strat-
egy is to demonstrate that we can eliminate the
coexistence of columnar grains and phase het-
erogeneities through a single additive. Because
we have not fully explored the compositional
space, we do not know whether 5.0 wt % Mo
is the optimal addition level, so further fine-
tuning may lead to greater improvement in
mechanical properties.
Conclusion
We explored how to simultaneously address
the formation of columnar grains and hetero-
geneously distributed phases in the Ti-5553 alloy
produced by L-PBF. We have shown that the

Fig. 5. Phase analysis of Ti-5553 and Mo-doped Ti-5553. (A) Schematic
illustration showing microstructural examinations of different locations (1, 2, and
3) in a Ti-5553 cubic part. (B) XRD spectra showing the effect of Mo additions on
the phase change. The inset shows the b(110) peak shifts toward lower angles.
arb. units, arbitrary units. (C) SEM-BSE images showing the heterogeneous
phase distribution along the building direction of the Ti-5553 part. The numbers
in the upper right corners indicate the locations where the SEM-BSE was
performed, as schematically illustrated in (A). These SEM-BSE images were
taken using a concentric backscattered detector at a low accelerating voltage
(3 kV) and a small working distance (6 mm). The phase contrast stems partially
from electron channeling (25). (D) SEM-BSE images showing solidification
cellular structures in Ti-5553+5Mo. The numbers indicate the locations where
the SEM-BSE was performed, as schematically illustrated in (A). (E) SEM-EDX
maps showing elemental segregation in cellular structures.

Zhang et al., Science 383, 639–645 (2024) 9 February 2024 6 of 7

RES EARCH | R E S E A R C H A R T I C L E
addition of Mo up to 5.0 wt % to Ti-5553 re-
sults in substantial CET and grain refinement,
which we attributed to heterogeneous nuclea-
tion on partially unmelted Mo particles, with
the dissolved Mo solute developing a super-
cooling zone that enhances the grain refine-
ment efficiency. Additionally, the dissolved Mo
solute enables the elimination of phase hetero-
geneities in Ti-5553 by stabilizing the b phase.
Compared with Ti-5553, Ti-5553+5Mo exhibits
higher strength, higher ductility, and more uni-
form tensile properties. More broadly, we anti-
cipate that the design strategy is likely to be
applicable beyond the titanium alloy consid-
ered here and could guide the design of other
alloys because both columnar grains and het-
erogeneously distributed phases are frequently
reported in a wide range of metallic alloys
produced by 3D printing (10, 23, 24, 49–52).
RE FE RENCES AND N OT ES
1. D. Herzog, V. Seyda, E. Wycisk, C. Emmelmann, Acta Mater.
117, 371–392 (2016).
2. T. DebRoy et al., Prog. Mater. Sci. 92, 112–224 (2018).
3. Y. M. Wang et al., Nat. Mater. 17, 63–71 (2018).
4. R. Cunningham et al., Science 363, 849–852 (2019).
5. P. Kürnsteiner et al., Nature 582, 515–519 (2020).
6. C. Panwisawas, Y. T. Tang, R. C. Reed, Nat. Commun. 11, 2327 (2020).
7. T. Zhang et al., Science 374, 478–482 (2021).
8. J. H. Martin et al., Nature 549, 365–369 (2017).
9. D. Zhang et al., Nature 576, 91–95 (2019).
10. T. M. Pollock, A. J. Clarke, S. S. Babu, Metall. Mater. Trans. A
51, 6000–6019 (2020).
11. B. E. Carroll, T. A. Palmer, A. M. Beese, Acta Mater. 87,
309–320 (2015).
12. T. DebRoy, T. Mukherjee, H. L. Wei, J. W. Elmer, J. O. Milewski,
Nat. Rev. Mater. 6, 48–68 (2021).
13. M. J. Bermingham, D. H. StJohn, J. Krynen, S. Tedman-Jones,
M. S. Dargusch, Acta Mater. 168, 261–274 (2019).
14. Q. Tan et al., Acta Mater. 196, 1–16 (2020).
15. Y. Liu et al., Acta Mater. 220, 117311 (2021).
16. P. Kontis et al., Acta Mater. 177, 209–221 (2019).
17. C. J. Todaro et al., Nat. Commun. 11, 142 (2020).
18. D. Zhang et al., Metall. Mater. Trans. A 51, 4341–4359
(2020).
Zhang et al., Science 383, 639–645 (2024) 9 February 2024
19. M. M. Kirka, P. Nandwana, Y. Lee, R. Dehoff, Scr. Mater. 135,
130–134 (2017).
20. W. Xu et al., Acta Mater. 85, 74–84 (2015).
21. T. Mukherjee, T. DebRoy, T. J. Lienert, S. A. Maloy,
P. Hosemann, Acta Mater. 209, 116775 (2021).
22. J. Zhang et al., Nat. Commun. 13, 4660 (2022).
23. P. C. Collins, D. A. Brice, P. Samimi, I. Ghamarian, H. L. Fraser,
Annu. Rev. Mater. Res. 46, 63–91 (2016).
24. J. J. Lewandowski, M. Seifi, Annu. Rev. Mater. Res. 46, 151–186
(2016).
25. J. Haubrich et al., Acta Mater. 167, 136–148 (2019).
26. M. Laleh et al., Prog. Mater. Sci. 133, 101051
(2023).
27. C. Hicks, G. Sivaswamy, T. Konkova, P. Blackwell, Mater. Sci.
Eng. A 825, 141928 (2021).
28. A. Caballero et al., Philos. Mag. 102, 2256–2281 (2022).
29. J. Zhang, M. Bermingham, J. Otte, Y. Liu, M. Dargusch,
Scr. Mater. 223, 115066 (2023).
30. Materials and methods are available as supplementary
materials.
31. S. Huang, P. Kumar, W. Y. Yeong, R. L. Narayan, U. Ramamurty,
Acta Mater. 225, 117593 (2022).
32. H. Deng et al., J. Alloys Compd. 810, 151792 (2019).
33. Z. Liu et al., Mater. Sci. Eng. A 798, 140093 (2020).
34. A. Zafari, E. W. C. Lui, M. Li, K. Xia, J. Mater. Sci. Technol. 105,
131–141 (2022).
35. H. Schwab, F. Palm, U. Kühn, J. Eckert, Mater. Des. 105, 75–80
(2016).
36. H. D. Carlton, K. D. Klein, J. W. Elmer, Sci. Technol. Weld. Join.
24, 465–473 (2019).
37. T. Zhang et al., Mater. Sci. Eng. A 835, 142624 (2022).
38. B. He, J. Li, X. Cheng, H.-M. Wang, Mater. Sci. Eng. A 699,
229–238 (2017).
39. D. H. StJohn, M. Qian, M. A. Easton, P. Cao, Acta Mater. 59,
4907–4921 (2011).
40. M. J. Bermingham, D. Kent, H. Zhan, D. H. Stjohn,
M. S. Dargusch, Acta Mater. 91, 289–303 (2015).
41. M. A. Easton, D. H. StJohn, Acta Mater. 49, 1867–1878
(2001).
42. M. S. K. K. Y. Nartu et al., Nat. Commun. 14, 3288
(2023).
43. C. Leyens, M. Peters, Eds., Titanium and Titanium Alloys:
Fundamentals and Applications (Wiley, 2005).
44. N. G. Jones, R. J. Dashwood, M. Jackson, D. Dye, Acta Mater.
57, 3830–3839 (2009).
45. Y. Zheng et al., Acta Mater. 103, 850–858 (2016).
46. T. Keller et al., Acta Mater. 139, 244–253 (2017).
47. S. Tekumalla, M. Seita, S. Zaefferer, Acta Mater. 262, 119413
(2024).
48. Z. Xiong, X. Pang, S. Liu, Z. Li, R. D. K. Misra, Scr. Mater. 195,
113727 (2021).
49. S. A. Mantri, R. Banerjee, Addit. Manuf. 23, 86–98
(2018).
50. M. Duport et al., Addit. Manuf. 61, 103340 (2023).
51. Y. Tian et al., Metall. Mater. Trans. A 45, 4470–4483
(2014).
52. S. H. Sun et al., Acta Mater. 64, 154–168 (2014).
53. J. Zhang et al., Data from: Ultra-uniform, strong and ductile 3D
printed titanium alloy through bifunctional alloy design,
dataset, Dryad (2024); https://datadryad.org/stash/dataset/
doi:10.5061/dryad.6t1g1jx50.
AC KNOWLED GME NTS
We thank M. Y. Lu for providing the nanoindentation equipment.
We acknowledge the Australian Microscopy and Microanalysis
Research Facility at the Centre for Microscopy and Microanalysis
(CMM), The University of Queensland, for the facilities and
technical assistance. Funding: This study received support from
Australian Research Council Research Hub for Advanced
Manufacturing of Medical Devices grant IH150100024 (M.S.D.);
Australian Research Council Discovery Project grant DP220102748
(M.J.B.); National Key Research and Development Program of
China grant 2021YFB3702101 (X.H. and Z.H.); and the “111” Project
from the Ministry of Education and the State Administration of
Foreign Experts Affairs of China, grant B16007 (X.H. and Z.H.).
Author contributions: Conceptualization: J.Z., M.J.B., X.H., and
M.S.D. Methodology: J.Z., J.O., Y.L., N.Y., Y.Y., and W.L. Investigation:
J.Z., M.J.B., J.O., Z.H., and M.B. Funding acquisition: M.S.D., D.H.S.,
and M.J.B. Project administration: M.S.D. Supervision: M.J.B., X.H.,
D.H.S., and M.S.D. Writing – original draft: J.Z., M.J.B., and M.S.D.
Writing – review & editing: J.Z., M.J.B., J.O., Y.L., Z.H., N.Y., Y.Y., M.B.,
W.L., X.H., D.H.S., and M.S.D. Competing interests: The authors
declare that they have no competing interests. Data and materials
availability: All data are available in the main text, in the
supplementary materials, or on Dryad (53). License information:
Copyright © 2024 the authors, some rights reserved; exclusive
licensee American Association for the Advancement of Science. No
claim to original US government works. https://www.science.org/
about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adj0141
Materials and Methods
Supplementary Text
Figs. S1 to S18
Tables S1 to S5
References (54–73)
Submitted 1 June 2023; resubmitted 11 October 2023
Accepted 18 December 2023
10.1126/science.adj0141
7 of 7
RES EARCH

PLANT REPRODUCTION would be expected if they were at the same

cell cycle stage (fig. S1, E and F). From these two
A paternal signal induces endosperm proliferation independent analyses, we conclude that before
fertilization, the EC is in G2 phase, whereas the
upon fertilization in Arabidopsis CC has entered but not completed S phase.
The fact that the CC did enter but not com-
Sara Simonini*, Stefano Bencivenga, Ueli Grossniklaus* plete S phase suggested that DNA replication
would resume upon fertilization. To test this
In multicellular organisms, sexual reproduction relies on the formation of highly differentiated hypothesis, we incubated inflorescences of
cells, the gametes, which await fertilization in a quiescent state. Upon fertilization, the cell cycle Arabidopsis with the nucleotide analog 5-
resumes. Successful development requires that male and female gametes are in the same phase of ethynyl-2′-deoxyuridine (EdU) to visualize DNA
the cell cycle. The molecular mechanisms that reinstate cell division in a fertilization-dependent synthesis in situ before fertilization and at 4, 6,
manner are poorly understood in both animals and plants. Using Arabidopsis, we show that a sperm- and 8 HAP (Fig. 1, G to I, and fig. S1G). The
derived signal induces the proliferation of a female gamete, the central cell, precisely upon unfertilized CC incorporated no EdU, indicat-
fertilization. The central cell is arrested in S phase by the activity of the RETINOBLASTOMA ing the absence of DNA synthesis (Fig. 1I). This
RELATED1 (RBR1) protein. Upon fertilization, delivery of the core cell cycle component CYCD7;1 observation was supported by the presence, in
causes RBR1 degradation and thus S phase progression, ensuring the formation of functional the EC and CC, of transcripts encoding the
endosperm and, consequently, viable seeds. replication licensing factor CTD1a, correlating
with the absence of DNA synthesis (35, 36)

S
exual reproduction entails specification
of the germline and the formation of male
and female gametes, which fuse during
fertilization (1). In flowering plants, fer-
tilization involves two pairs of gametes
that are produced by multicellular gameto-
phytes. The male gametophyte (pollen) germi-
nates to form a pollen tube that transports two
sperm cells to the female gametophyte. The
latter develops within the ovule and contains
two female gametes: the egg cell (EC) and the
central cell (CC). During double fertilization,
one sperm fuses with the EC and the other
with the CC (2). The fertilized EC develops into
the embryo, whereas the fertilized CC forms
the endosperm, a placenta-like tissue sustain-
ing embryonic growth (3).
Male and female gametes are in a quiescent
state, which ensures that they do not divide in
the absence of fertilization and mediates syn-
chronization of their cell cycles when their
nuclei fuse upon fertilization. In plants, de-
pending on the species, sperm are arrested
in the G1 (4–10) or G2 phase (11, 12), whereas it
is unclear at which cell cycle stage the female
gametes are arrested (10–14).
More than 100 factors govern cell cycle arrest
and progression (15, 16), and their deregulation
often leads to reduced fertility because of ab-
errant gametogenesis and/or embryogenesis
(17–21). In the model plant Arabidopsis, mu-
tations in some cell cycle genes affect only the
CC: some cause proliferation in the absence of
fertilization, whereas others cause a lack of
division upon nuclear fusion (22–26). These
observations have fueled the hypothesis that a
mechanism preventing cell division operates
in the CC, which is lifted by a fertilization-
dependent signal (13, 27, 28).
Institute of Plant and Microbial Biology and Zurich-Basel
Plant Science Center, University of Zurich, CH-8008 Zurich,
Switzerland.
*Corresponding author. Email: grossnik@botinst.uzh.ch (U.G.);
sara.simonini@botinst.uzh.ch (S.S.)
Quiescent ECs and CCs arrest at different
cell cycle stages
We assessed the stage at which mature female
gametes arrest by analyzing the expression of
the cell cycle machinery using available tran-
scriptome datasets (29–31). We found that the
CC expressed high levels of virtually all transcripts
encoding factors involved in and necessary for
DNA replication, whereas this was not the case
for the egg and pollen (Fig. 1A and table S1).
This suggested that the CC might be arrested
in S phase. To test this, we measured the DNA
content of the EC and CC by staining ovules
with propidium iodide (PI) (32) at 0 to 10 hours
after pollination (HAP) (Fig. 1, B to E). We
quantified fluorescence intensity in the nuclei as
a proxy for the DNA amount in a given nucleus
(C-value) (Fig. 1, B to F). Sporophytic cells of
the ovule are diploid (2n) and may be at any
stage of the cell cycle. When they are in G1 or G2
phase, their ploidy/C-value is 2n/2C or 2n/4C,
respectively, whereas cells in S phase have C-
values between 2C and 4C. Indeed, sporophytic
cells showed a range of ploidy/C-values from
2n/2C to 2n/4C (Fig. 1F). The CC is homo-
diploid at maturity, so we expected a ploidy/
C-value of 2n/2C if it was in G1 phase and 2n/4C
if it was in G2 phase. However, unfertilized CCs
(0 HAP) had an average ploidy/C-value of 2n/3C
(Fig. 1F and fig. S1A), indicating that they had
entered but not completed S phase. By contrast,
unfertilized ECs, which are haploid (1n), had an
average ploidy/C-value of 1n/2C (Fig. 1F, fig. S1A).
Thus, their DNA had been replicated (2C) and
they were in G2 phase.
We also quantified the DNA content of ECs
and CCs by scoring the fluorescence intensity
of histone H2B fused to the fluorescent protein
tdTomato expressed under the control of two
different promoters, pGRP1::H2B-tdTomato (33)
and pRPS5a::H2B-tdTomato (34) (fig. S1, B to F).
We found that the chromatin content of the CC
was ~1.5 times (mean = 1.73, n = 58) greater than
that of the EC, but not two times greater, as
(table S1). We also detected a green fluores-
cent protein (GFP)–tagged version of CTD1a
in immature (Fig. 1J) and unfertilized CCs
(Fig. 1, K to N). These data suggest that the
unfertilized CC is arrested in S phase at matu-
rity and, unlike the EC, does not reach the G2
phase (Fig. 1O).
The CC completes S phase after fertilization
After fertilization, we observed changes in the
CC. First, the DNA level increased from 2n/3C
to about 3n/5C, indicating successful fertil-
ization (Fig. 1F and fig. S1A) and supporting
previous observations that sperm has a DNA
content of 1n/2C (11). Second, at ~8 HAP, the
ploidy/C-value of the fertilized CC had increased
to 3n/6C, indicating that it was now in G2 phase
(Fig. 1F and fig. S1A) and thus that DNA syn-
thesis occurred after fertilization. Third, in some
ovules, the EC had a ploidy/C-value of 2n/4C,
whereas that of the CC remained 2n/3C (fig. S1A)
with a sperm nucleus visible in its proximity (fig.
S1A). This was consistent with nonsynchronous
nuclear fusion (karyogamy) in ECs and CCs
(28, 37–40) and indicates that sperm delivery
to the female gametophyte is insufficient to
trigger cell cycle reactivation in the CC. Rather,
karyogamy is required, because an increase in
the ploidy/C-value occurred only in CCs with a
ploidy/C-value of 3n/5C. Consistent with this
increase in the ploidy/C-value, we also ob-
served EdU staining, indicative of active DNA
synthesis, specifically in the CCs of fertilized
ovules (Fig. 1, H and I), and depletion of
CTD1a-GFP, typically occurring when the S
phase is initiated, exclusively in the CC (Fig. 1,
L to N). The EC behaved very differently upon
fertilization: its ploidy/C-value increased from
1n/2C to 2n/4C (Fig. 1F and fig. S1A) without
detectable EdU incorporation (Fig. 1I) or per-
sistent CTD1a-GFP fluorescence (Fig. 1, K to
N). These observations confirm that mature
ECs and sperm cells are arrested in G2 phase (11)
and reveal that replication of the CC genome is
completed only after karyogamy (Fig. 1O).

Simonini et al., Science 383, 646–653 (2024) 9 February 2024 1 of 8

RES EARCH | R E S E A R C H A R T I C L E
Fig. 1. Quiescent ECs are in the G2 phase of the cell cycle and CCs arrest
in S phase, which is completed only upon fertilization. (A) Circular heatmap
of the relative expression levels of transcripts encoding cell cycle components in
the EC, CC, and pollen (PO). Genes are organized according to their expression
at the different phases of the cell cycle from G1 to M, as indicated. A single gene
may be represented in more than one phase according to its function. (B to E) Repre-
sentative images of PI-stained ovules showing the female gametophyte at various times
(in HAP), as indicated, and after the first endosperm division. Endo, endosperm; Zyg,
zygote; SC, sperm cell. (F) Box plots showing quantification of PI staining, as in (B) to
(E), indicating the ploidy/C-values of sporophytic (Sp) cells, CC (dark green), EC (light
green), endosperm (EN), and zygote (ZY) at the indicated times (in HAP). (G and
H) Representative images of EdU staining of ovules at the indicated time points after
pollination. (I) Box plots showing quantification of EdU signals as in (G) and (H) in Sp
Simonini et al., Science 383, 646–653 (2024) 9 February 2024
cells, CCs (orange), and ECs (red) at the indicated time points (in HAP). (J to M) Confocal
fluorescence microscopy analyses of CTD1a-GFP (green) in the female gametophyte before
(J) and after (K) fusion of the polar nuclei (Pn) in the CC before fertilization and in the
primary endosperm nucleus after fertilization [(L) and (M)]. Cell walls were stained with PI
(purple). Fertilized ovules are identified by accumulation of PI in the degenerated synergid
cell (Syn). (N) Quantification of CTD1a-GFP signal intensity in the nuclei of Sp cells, ECs
(purple), and CCs (pink) in unfertilized and fertilized ovules. Values for each category are
normalized to the averaged signal intensity value of 10 Sp cells surrounding the female
gametophyte. (O) Schematic representation summarizing the cell cycle phases
of the EC and CC relative to fertilization. Unfertilized mature ECs are in G2 phase,
unfertilized CCs are arrested in S phase, and fertilized CCs have completed S phase.
Scale bars, 20 mm. n.s., not significant; *P < 0.01; **P < 0.001; and ***P < 0.0001
according to a t test.
2 of 8
RES EARCH | R E S E A R C H A R T I C L E

Degradation of RBR1 in the CC is required for
cell cycle progression
To investigate the mechanism underlying
karyogamy-dependent DNA synthesis in the CC,
we focused on RETINOBLASTOMA RELATED1
(RBR1) (41–44), an evolutionarily conserved,
potent inhibitor of entry into and progression
through the S phase (45–47). In Arabidopsis, rbr1
mutant CCs undergo uncontrolled divisions to
produce an endosperm-like structure (23) (Fig.
2, A and B), suggesting that RBR1 may be in-
volved in preventing S phase completion in
the CC. We investigated RBR1 dynamics in the
CC around fertilization by analyzing expression
of a yellow fluorescent protein (YFP)–tagged
version of the protein, RBR1-YFP (48). Indeed,
RBR1 accumulated in unfertilized CCs, as
previously described (49) (Fig. 2, C and D).
At 7 to 8 HAP, however, CCs exhibited a weaker
or undetectable fluorescent signal, before RBR1-
YFP reappeared in both endosperm nuclei after
the first division (Fig. 2, C to G).
To determine whether RBR1 degradation in
the CC is mediated by the 26S proteasome, as
it is in animals and other plant tissues (48, 50),
we treated pistils with the 26S proteasome in-
hibitors epoxomicin, syringolin-A (SylA), and
MG-132 and imaged RBR1-YFP around fertiliza-
tion. When treated with epoxomicin or SylA,
the RBR1-YFP signal persisted in the fertilized

Fig. 2. Degradation of RBR1 in the CC is required for cell

cycle progression. (A and B) Cleared, unfertilized ovules of
RBR1/rbr1-3 plants showing a wild-type (WT) female gametophyte
(A) and a mutant female gametophyte (B), in which the CC
has undergone divisions in absence of fertilization. The CC
nucleolus (A) and the endosperm-like nucleoli (B) are artificially
colored cyan. Scale bars, 20 mm. (C) Quantification of RBR1-YFP
signal intensity in Sp cells and CCs of unfertilized and fertilized
ovules. Values for each category are normalized to the averaged
signal intensity of 10 Sp cells surrounding the female gametophyte.
Scale bars, 20 mm. (D to G) Representative images of ovules
expressing RBR1-YFP (yellow) at various time points (in HAP),
as indicated. The contrast of the image in (F) was increased
to reveal the faint RBR1-YFP signal in the nucleus of the
synergid cell (white arrowhead). Cell walls were stained
with PI (purple). Scale bars, 20 mm. (H) Quantification of
RBR1-YFP signal in Sp cells, unfertilized (Unfe) CCs (light green),
and fertilized (Fert) CCs (dark green) in inflorescences treated
with a mock solution or with the proteasome inhibitors epoxomicin,
syringolin-A (Syl-A), and MG-132. Values for each category are
normalized to the averaged signal intensity value of 10 Sp
cells surrounding the female gametophyte. (I to K) Cleared
seeds of inflorescences treated with a mock solution (I),
epoxomicin (J), and Syl-A (K). At the top right corner is the
percentage of seeds showing the corresponding phenotype.
Embryo and endosperm nucleoli are artificially highlighted in
yellow and cyan, respectively. Scale bars, 50 mm. n.s.,
not significant and ***P < 0.0001 according to a t test.

Simonini et al., Science 383, 646–653 (2024) 9 February 2024 3 of 8

RES EARCH | R E S E A R C H A R T I C L E
CC at ~8 HAP, whereas it disappeared from
mock-treated pistils (Fig. 2H), indicating that
these inhibitors prevented or slowed down RBR1-
YFP degradation in the CC nucleus. Although
MG-132 is reported to inhibit RBR1 degrada-
tion (48), it had no statistically significant effect
on RBR1-YFP stability in our experiments (Fig.
2H). This might be explained by the chemical
properties of the inhibitors: MG-132 is a re-
versible inhibitor that is highly unstable in water,
whereas epoxomicin and Syl-A are high-affinity,
water-stable, irreversible inhibitors.
We repeated the epoxomicin and Syl-A treat-
ments and left pollinated pistils to develop for
another 4 days without inhibitors. A single
treatment of ~2 hours led to the formation of
seeds with apparently normal, globular-stage
embryos that were surrounded by a few, mas-
sively enlarged endosperm nuclei (Fig. 2, I to K).
Similar endosperm phenotypes were previously
observed in plants lacking components of the
DNA replication machinery (25–27) or CULLIN4
(51), an E3 ubiquitin-ligase involved in protein
degradation by the 26S proteasome. The sim-
ilarities between the phenotype of these mutants
and our inhibitor treatments support the con-
clusion that protein degradation and DNA syn-
thesis are required for normal endosperm
development.
CCs express several CYCLIN-DEPENDENT
KINASES (CDKs) (19) that, by dimerizing with
cyclins, phosphorylate RBR1 and provoke its
degradation (fig. S2, A to E). To prevent RBR1
degradation before fertilization, the CC is
equipped with KIP-RELATED PROTEINS
(KRPs) that bind to CDKs and inhibit their
function. In ick1,2,6,7 mutant plants, which
lack four KRPs (52), we observed that 13.1%
of unfertilized ovules contained two to four
endosperm-like nuclei, indicating division of
the CC in the absence of fertilization (fig. S2,
A to D). Together with the data described above
(fig. S2E), this indicates that RBR1 persistence
in the CC results in cell cycle arrest, prevent-
ing progression to the G2 phase. Moreover,
RBR1 degradation is a prerequisite for S phase
completion, ultimately ensuring that synchro-
nized maternal and paternal genomes initiate
proper endosperm development.
Sperm cell cycle components are delivered to
the female gametes upon fertilization
To understand how RBR1 degradation is in-
duced upon fertilization in the CC, we isolated
CCs by laser-assisted microdissection at various
time points after pollination and analyzed their
transcriptomes (Fig. 3A; table S2; and fig. S3, A
to D). RBR1 protein degradation in the fertilized
CC occurs at ~8 HAP (Fig. 2, C to G). We there-
fore searched for transcripts that were the
most abundant at this time. Such a peak in
abundance might be caused by fertilization-
dependent de novo transcription or by delivery
of paternal transcripts from the sperm to the
Simonini et al., Science 383, 646–653 (2024) 9 February 2024
CC upon gamete fusion. Because RBR1 degra-
dation occurs through a cell cycle–regulated
mechanism (45, 53–55), we focused on transcripts
encoding cell cycle components (Fig. 3B). Tran-
scripts of the CYCD7;1 gene, encoding a cyclin
D-type family member, were absent from un-
fertilized CCs but highly abundant at 8 HAP.
D-type cyclins mediate entry into the S phase
(56, 57). CYCD7;1 interacts with RBR1 and pro-
motes its degradation (58, 59) and is the only
D-type cyclin that is highly expressed in pollen
(31, 58). Moreover, ectopic expression of CYCD7;1
induces CC proliferation in the absence of fer-
tilization (60). We thus hypothesized that
CYCD7;1 is a paternal signal that is stored in
the sperm and delivered to the CC upon fer-
tilization, thus provoking RBR1 degradation
and S phase completion.
To test whether CYCD7;1 is a paternal signal,
we looked for CYCD7;1 mRNA in sperm
and unfertilized ovules by in situ hybrid-
ization on WT ovules fertilized with pollen
expressing a pCYCD7;1::CYCD7;1-YFP transgene
(58). A YFP-specific antisense probe (Fig. 3, C
to J) allowed us to distinguish maternal and
paternal transcripts in fertilized CCs. In
pCYCD7;1::CYCD7;1-YFP plants, signal was
detected during pollen development (Fig. 3D)
and in sperm nuclei of mature pollen grains
and elongating pollen tubes (Fig. 3, E and F).
At ~4 HAP, a punctate signal was observed
in WT ovules, indicative of CYCD7;1-YFP tran-
scripts in delivered sperm nuclei (Fig. 3H). In
some fertilized ovules at 4 HAP, we detected
signals confined to the nuclei of CCs after
karyogamy (Fig. 3I), whereas at 8 HAP, they
were in the cytoplasm of ECs and CCs (Fig. 3J).
The presence of paternal CYCD7;1 transcripts
in fertilized female gametes was confirmed in
reciprocal crosses between WT and cycd7;1-
1 mutant plants using a probe specific to the
WT CYCD7;1 allele (fig. S4, E to J).
We then analyzed the expression of a nuclear-
localized YFP protein under control of the
pCYCD7;1 promoter (58), allowing us to mo-
nitor its spatiotemporal activity. In ovules, no
YFP signal was detected in unfertilized or fer-
tilized ECs or in CCs (Fig. 3L). These data indi-
cate that CYCD7;1 is transcribed in pollen, where
its mRNA is stored in the sperm nuclei and
delivered to the female gametes upon fertilization.
The CYCD7;1 protein is present in mature
pollen (58, 61), so we wondered whether it is
also stored in sperm nuclei and delivered to
female gametes. We fertilized WT plants with
pollen from pCYCD7;1::CYCD7;1-YFP transgenic
plants and analyzed CYCD7;1-YFP fluorescence.
CYCD7;1-YFP was detected in sperm nuclei
of mature pollen (Fig. 3M), in growing pollen
tubes (Fig. 3N), and in fertilized EC and CC
nuclei (Fig. 3, O and P). These data demon-
strate that CYCD7;1 is a paternally derived fac-
tor, with both transcripts and proteins being
delivered to the female gametes. The restricted
nuclear localization of mRNA and protein
likely ensures that CYCD7;1 is active only upon
successful karyogamy and not simply after
sperm delivery.
Paternal CYCD7;1 promotes cell cycle
progression of the CC
To investigate whether CYCD7;1 promotes RBR1
degradation in the CC, we studied the dynam-
ics of RBR1 degradation upon pollination by
crossing pRBR1::mCherry-RBR1 plants (62) with
pCYCD7;1::CYCD7;1-YFP pollen. We found that
RBR1 depletion in the CC occurred after
CYCD7;1 delivery (fig. S4, A to C), consistent
with the notion that CYCD7;1 promotes RBR1
degradation. Moreover, CC-specific (Fig. 4, A
and B) or ubiquitous (Fig. 4C) ectopic expres-
sion of CYCD7;1 induced the development of
endosperm-like structures in unfertilized ovules,
consistent with a previous report (60). The in-
teraction between RBR1 and CYCD7;1, which is
required for phosphorylation and thus degra-
dation of RBR1, is mediated by a Leu-x-Cys-x-
Glu (LxCxE) motif (59, 63). To determine
whether this interaction is necessary for RBR1
degradation, we ubiquitously expressed a
CYCD7;1 LxCxE mutant variant (pRPL18::
CYCD7;1mut) and found that it did not induce
CC proliferation in unfertilized ovules (Fig. 4,
A to D), indicating that the CYCD7;1-RBR1 inter-
action is required for the S phase to proceed in
the CC upon fertilization.
In plants expressing RBR1-YFP and CYCD7;1
in unfertilized CCs (pMEA::CYCD7;1), the RBR1-
YFP signal was very weak or undetectable, and
endosperm-like structures formed (Fig. 4, E and
F). These findings indicate that CYCD7;1 alone
is sufficient to induce RBR1 degradation in the
CC and, consequently, to stimulate endosperm
proliferation.
To investigate whether CYCD7;1 exerts pa-
ternal control over cell division in fertilized CCs,
we used the T-DNA insertion mutants cycd7;1-
1 (58), cycd7;1-2 (58), and cycd7;1-3 and a
new mutant allele created by CRISPR-Cas9,
cycd7;1CRISPR (fig. S4D), in reciprocal crosses with
WT plants. If paternal delivery of CYCD7;1 were
required for S phase completion in the CC, then
its absence would delay division only when
cycd7;1 mutants were used as a male parent,
but not as a female parent, in these crosses.
We scored the percentage of fertilized ovules
with undivided CCs (Fig. 4G) or two, four, or
eight endosperm nuclei (Fig. 4, H to J). In
crosses of the WT with cycd7;1 mutant pollen,
we observed a significant delay in the first CC
division (Fig. 4L and fig. S4E). At 10 HAP, <5%
of fertilized ovules in crosses between the
WT and cycd7;1 mutant pollen contained
two endosperm nuclei, compared with 30% in
crosses between two WT plants (Fig. 4L and
fig. S4E). WT and cycd7;1 mutant pollen tubes
grew at similar rates (fig. S5A), indicating that
delayed endosperm initiation was not caused
4 of 8
RES EARCH | R E S E A R C H A R T I C L E

by delayed fertilization. The endosperm of seeds
that had received a paternal cycd7;1 mutant
allele started to proliferate at ~16 HAP (Fig. 4L),
suggesting that other D-type cyclins might com-
pensate for the absence of CYCD7;1. To investi-
gate this possibility, we created a quadruple cycd
mutant: cycd3;1 cycd3;2 cycd3;3 cycd7;1 (cyc4D;
fig. S5, B and C). When crossing the WT with
cyc4D pollen, the first endosperm division was
even more delayed than in crosses with cycd7;1
single mutants (99% of ovules at 10 HAP had
an undivided CC; Fig. 4L), consistent with other
D-type cyclins compensating for the absence
of CYCD7;1. Embryo development was unaf-
fected, but we observed several fertilized
ovules with undivided CC nuclei next to elon-
gating zygotes (Fig. 4K). Moreover, the delay in
initiating endosperm division was exacer-
bated when cyc4D females were pollinated
with cyc4D pollen (Fig. 4M). This suggests
that the CC becomes transcriptionally active
soon after fertilization, producing cell cycle
components including D-type cyclins. Early
activation of maternal CYCD7;1 was also ob-
served in zygotes (64). Consistent with this
conclusion, we detected a YFP signal in plants
carrying a maternal pCYCD7:1::YFP-YFPnls
allele in endosperm containing four nuclei
(Fig. 4N), indicating activation of CYCD7;1
transcription at ~20 HAP.
To demonstrate a causal effect between
CYCD7;1 delivery by the sperm and RBR1
degradation, we crossed plants expressing
RBR1-YFP with WT, cycd7;1, or cyc4D pollen
and analyzed fluorescence after fertilization.
RBR1-YFP fluorescence was depleted from
fertilized CCs at ~8 HAP only when WT pollen
was used, whereas the RBR1-YFP signal per-
sisted in CCs fertilized with cycd7;1 or cycd4D
sperm (Fig. 4O). Stability of RBR1-YFP fluores-
cence was associated with a lack of DNA syn-
thesis upon fertilization. CCs fertilized by cycd7;1
or cyc4D sperm failed to reach the G2 phase after
fertilization; their ploidy/C-values remained at
~3n/5C at 8 HAP, when all CCs fertilized by
WT sperm had reached the 3n/6C value (Fig. 4P).
We conclude that RBR1 depletion in the CC fails
to occur at fertilization in the absence of paternal
CYCD7;1 delivery, thereby delaying completion of
S phase and progression to G2 phase.
Discussion
We describe a mechanism ensuring that re-
activation of the cell cycle in the CC occurs

Fig. 3. Paternal CYCD7;1 mRNA and protein are delivered by the sperm cell
to the female gametes upon fertilization. (A) Schematic of unfertilized and
fertilized female gametophytes used for laser-assisted microdissection of the CC
or endosperm. Laser-isolated cell sections were used for transcriptome analysis
of the CC at various time points (in HAP), as indicated. Key developmental events
and cells (synergid cells: yellow; degenerated synergid cell: brown; unfertilized EC:
red; fertilized EC: purple; unfertilized CC: cyan; fertilized CC: blue; endosperm
nuclei: pink) are indicated. (B) Heatmap of cell cycle–related genes (right) with
transcript levels that vary across the four time points indicated. (C to F) In situ
hybridization with a YFP antisense probe on tissues of pCYCD7;1::CYCD7;1-YFP
(abbreviated as pCYCD7;1::D7;1-YFP in the figure) plants showing no signal in
unfertilized ovules (C), but a positive signal in anthers (D), and the nuclei of
sperm cells (SC, arrowheads) in mature pollen grains (E) and growing pollen tubes
(F). (G and H) In situ hybridization with a YFP antisense probe on WT pistils
pollinated with pCYCD7:1::CYCD7;1-YFP pollen at the indicated time points (in
HAP), showing a signal in the discharged SCs [(H), arrowhead], the CC nucleus
[(I), arrowhead], and the nuclei and cytosol of ECs and CCs (J). (K and L) Confocal
fluorescence microscopy of YFP-YFPnls driven by the pCYCD7;1 promoter in
unfertilized (K) and fertilized (L) ovules shows no YFP signal in ECs and CCs. Cell walls
were stained with PI (purple). (M and N) Confocal fluorescence microscopy of
mature pollen grains (M) and pollen tubes (PT) (N) of a pCYCD7;1::CYCD7;1-YFP
plant showing CYCD7;1-YFP in SC nuclei. Pollen grain cell walls were stained with PI
(purple). (O and P) Confocal fluorescence microscopy of unfertilized (O) and fertilized
(P) WT ovules pollinated with pCYCD7;1::CYCD7;1-YFP pollen showing CYCD7;1-YFP
in the fertilized CC and EC. Cell walls were stained with PI (purple). Scale bars in (E),
(L), and (N), 10 mm; scale bars in (C), (D), (F) to (K), (M), (O), and (P), 20 mm.

Simonini et al., Science 383, 646–653 (2024) 9 February 2024 5 of 8

RES EARCH | R E S E A R C H A R T I C L E
Fig. 4. Paternally derived CYCD7;1 promotes cell cycle progression and
RBR1 degradation in the CC. (A and B) Cleared ovules of WT and hemizygous
pMEA::CYCD7;1/- plants with CC-specific expression of CYCD7;1 showing a WT-
looking ovule with an undivided CC (A) and an ovule in which the CC has
proliferated (n = 28 of 30 independent transformants). The percentage of ovules
with the corresponding phenotype is indicated in the top right corner. The CC
nucleolus (A) and endosperm-like nucleoli (B) are artificially colored cyan. (C and
D) Cleared ovules of hemizygous pRPL18::CYCD7;1/- and pRPL18::CYCD7;1mut/-
plants with ubiquitous expression of CYCD7;1 and CYD7;1mut, respectively,
showing an ovule in which the CC has proliferated [n = 18 of 26 independent
Simonini et al., Science 383, 646–653 (2024) 9 February 2024
transformants (C)] and a WT-looking ovule (D). The percentage of ovules with
the corresponding phenotype is indicated in the top right corner. Endosperm-like
nucleoli (C) and the CC nucleolus (D) are artificially colored cyan. (E) Quantification
of the RBR1-YFP signal in undivided CCs and endosperm-like nuclei of
hemizygous pMEA::CYCD7;1/- plants 5 days after emasculation (DAE). For each
ovule showing CC proliferation, four representative nuclei were selected for
quantification. (F) Example of the RBR1-YFP signal (bottom) in an ovule of a
hemizygous pMEA::CYCD7;1/- plant showing CC proliferation in the absence of
fertilization (top). Endosperm-like nucleoli are artificially colored cyan. Syn,
synergid cells. (G to J). Cleared fertilized ovules showing four stages of
6 of 8
RES EARCH | R E S E A R C H A R T I C L E

endosperm development: fertilized CC (G) and two (H), four (I), and eight (J)
endosperm nuclei. The CC nucleolus and endosperm-like nucleoli are artificially
colored cyan. ZY, zygote. (K) Cleared WT ovule fertilized with cycd4D pollen at
12 HAP showing an undivided CC nucleus (artificially colored cyan) next to
the elongating ZY artificially colored yellow. (L) Classification of fertilized ovules
(%) based on the number of endosperm nuclei at the indicated time points (in HAP)
in crosses between WT × WT, WT × cycd7;1-1, and WT × cyc4D plants. Sample
size is indicated on the right side of each bar. The color key is indicated at the bottom
of panel (M). The P value was calculated using a chi-square test with Egon
Pearson correction where appropriate. (M) Classification of fertilized ovules (%)
based on the number of endosperm nuclei at 24 HAP in reciprocal crosses
between WT, cycd7;1-1, and cyc4D individuals. Sample size is indicated on the
right side of each bar. The color key is indicated at the bottom. The P value
was calculated using a chi-square test with Egon Pearson correction where
appropriate. (N) Confocal fluorescence microscopy of ovules derived from a
cross of a pCYCD7;1::YFPYFPnls female with WT pollen at 16 HAP showing
activation of the maternal pCYCD7;1::YFPYFPnls transgene. Asterisks mark
endosperm nuclei with a YFPYFPnls signal. (O) Box plots showing quantification
of RBR1-YFP signal in unfertilized CCs and in CCs of pistils pollinated with WT,
cycd7;1, or cyc4D pollen. n.s., not significant; **P < 0.001; and ***P < 0.0001
according to a t test. (P) Quantification of ploidy/C-values by PI staining of
CCs and ECs of WT pistils pollinated with WT, cycd7;1, or cyc4D pollen at 8 HAP.
n.s., not significant; **P < 0.001; and ***P < 0.0001 according to a t test.
Scale bars, 20 mm.

precisely upon fertilization. We propose that,
upon delivery of CYCD7;1 by the sperm, an
active CDK-CYCD7;1 complex forms and initiates
RBR1 degradation (fig. S6A). Several CDKs are
highly expressed in the CC (19) and are known
to be important for endosperm development
(19, 65), but RBR1 is not degraded before fer-
tilization. This is likely because there are no
D-type cyclins to activate CDKs before their
delivery by the sperm, but also because several
KRPs are present that inhibit CDK function.
The observation that the quiescent CC is
arrested in S phase is unusual and opens new
questions. S phase is typically arrested when
the DNA integrity checkpoint has not been
satisfied, e.g., because of DNA damage, but also
upon oxidative stress or depletion of nucleo-
tides. Upon DNA damage, pRB, the animal
homolog of RBR1, localizes to replication origins,
arresting DNA synthesis and inhibiting S phase
progression by attenuating cyclin and Cdk activ-
ity (47, 66–68). The replication licensing factor
CDT1, which usually gets degraded upon S phase
entry, also accumulates on damaged chromatin
(69). Thus, in addition to their canonical activity
at the G1-to-S transition, animal pRB and CTD1
function during S phase arrest, consistent with
our observations in the Arabidopsis CC.
We can only speculate about the cause of
S phase arrest in the CC. Consistent with the
findings in animals mentioned above, the CC
nucleus may experience either DNA damage
caused by extensive DNA demethylation
occurring before fertilization (70) or oxidative
stress (71). Consistent with the latter, oxidative
stress–related genes are highly enriched among
genes changing expression in the CC upon fer-
tilization (fig. S6B).
Paternal CYCD7;1-dependent RBR1 degrada-
tion seems specific to the CC because the
fertilized EC does not rely on paternal CYCD7;1
to initiate the first division. This is as expected
because a cell in G2 phase does not require DNA
synthesis regulated by RBR1 degradation
for cell cycle progression. Our findings suggests
that the EC and CC, although genetically iden-
tical, use different pathways to integrate cell cycle
progression with developmental programs that
may rely on regulating a component’s activity
rather than its expression, as we showed for RBR1.
Previously, only two factors were known to
be required specifically in one female gamete.
The SHORT SUSPENSOR transcript accumu-
lates in sperm and is delivered to the female
gametes, where it controls development of
the embryonic suspensor (72), and peptides
encoded by the EMBRYO SURROUNDING
FACTOR1 gene family accumulate in the CC
and nonautonomously control suspensor
development (73). By contrast, CYCD7;1 is a
paternal, sperm-derived signal that specifically
controls CC proliferation and thus endosperm
formation.
Our findings provide insights into how a cell
determines when it is time to divide, a ques-
tion that is of particular relevance to female
gametes because embryogenesis or endosperm
development fail if division occurs before fer-
tilization. Given the conserved roles of the
factors involved, understanding the fertilization-
dependent molecular mechanisms that control
endosperm initiation might help in the design
of strategies to manipulate seed development
in crops.
RE FERENCES AND NOTES
1. H. Sawada, N. Inoue, M. Iwano, in Sexual Reproduction in
Animals and Plants, (Springer, 2014); https://link.springer.
com/book/10.1007/978-4-431-54589-7.
2. G. N. Drews, A. M. G. Koltunow, in The Arabidopsis Book
(ASPB, 2011), vol. 2011; https://doi.org/10.1199/tab.0155.
3. F. Berger, Curr. Opin. Plant Biol. 6, 42–50 (2003).
4. Y. Matsuda, T. Saito, M. Koseki, T. Z. Shimada, Plant Physiol. 9,
1–6 (1990).
5. L. J. Goff, A. W. Coleman, Dev. Biol. 102, 173–194 (1984).
6. Z. Pónya et al., Protoplasma 208, 163–172 (1999).
7. Y. Sukawa, T. Okamoto, Plant Reprod. 31, 107–116 (2018).
8. H. L. Mogensen, N. Leduc, E. Matthys-Rochon, C. Dumas,
Planta 197, 641–645 (1995).
9. H. L. Mogensen, P. B. Holm, Plant Cell 7, 487–494
(1995).
10. X.-Q. Liu et al., Plant Physiol. 185, 137–145 (2021).
11. W. E. Friedman, Development 126, 1065–1075 (1999).
12. H. Q. Tian, T. Yuan, S. D. Russell, Sex. Plant Reprod. 17,
243–252 (2005).
13. F. Berger, Y. Hamamura, M. Ingouff, T. Higashiyama, Trends
Plant Sci. 13, 437–443 (2008).
14. J. Chen et al., Plant Cell 29, 2106–2125 (2017).
15. C. Gutierrez, in The Arabidopsis Book (ASPB, 2009), vol. 2009;
https://doi.org/10.1199/tab.0120.
16. M. Malumbres, M. Barbacid, Nat. Rev. Cancer 9, 153–166 (2009).
17. S. Simonini et al., Dev. Cell 56, 1945–1960.e7 (2021).
18. G. J. Campbell, E. L. Hands, M. Van de Pette, Int. J. Mol. Sci. 21,
5343 (2020).
19. M. K. Nowack et al., Dev. Cell 22, 1030–1040 (2012).
20. C. Ebel, L. Mariconti, W. Gruissem, Nature 429, 776–780
(2004).
21. M. Romeiro Motta et al., EMBO Rep. 23, e53995 (2022).
22. L. Borghi et al., Plant Cell 22, 1792–1811 (2010).
23. A. J. Johnston, E. Matveeva, O. Kirioukhova, U. Grossniklaus,
W. Gruissem, Curr. Biol. 18, 1680–1686 (2008).
24. D. A. Ni et al., New Phytol. 184, 311–322 (2009).
25. P. S. Springer, D. R. Holding, A. Groover, C. Yordan,
R. A. Martienssen, Development 127, 1815–1822 (2000).
26. M. A. Collinge, C. Spillane, C. Köhler, J. Gheyselinck,
U. Grossniklaus, Plant Cell 16, 1035–1046 (2004).
27. M. K. Nowack et al., Nat. Genet. 38, 63–67 (2006).
28. S. J. Aw, Y. Hamamura, Z. Chen, A. Schnittger, F. Berger,
Development 137, 2683–2690 (2010).
29. M. W. Schmid et al., PLOS ONE 7, e29685 (2012).
30. A. Schmidt et al., PLOS Biol. 9, e1001155 (2011).
31. A. E. Loraine, S. McCormick, A. Estrada, K. Patel, P. Qin,
Plant Physiol. 162, 1092–1109 (2013).
32. W. She, D. Grimanelli, C. Baroux, J. Vis. Exp. 88, e51530
(2014).
33. D. Susaki et al., PLOS Biol. 19, e3001123 (2021).
34. S. Adachi et al., Proc. Natl. Acad. Sci. U.S.A. 108,
10004–10009 (2011).
35. B. Desvoyes, A. Arana-Echarri, M. D. Barea, C. Gutierrez,
Nat. Plants 6, 1330–1334 (2020).
36. H. Nishitani, S. Taraviras, Z. Lygerou, T. Nishimoto, J. Biol. Chem.
276, 44905–44911 (2001).
37. Y. Hamamura et al., Nat. Commun. 5, 4722 (2014).
38. Y. Hamamura et al., Curr. Biol. 21, 497–502 (2011).
39. M. Ingouff, Y. Hamamura, M. Gourgues, T. Higashiyama,
F. Berger, Curr. Biol. 17, 1032–1037 (2007).
40. G. C. Pagnussat et al., Development 132, 603–614 (2005).
41. Q. Xie, A. P. Sanz-Burgos, G. J. Hannon, C. Gutiérrez, EMBO J.
15, 4900–4908 (1996).
42. G. Grafi et al., Proc. Natl. Acad. Sci. U.S.A. 93, 8962–8967
(1996).
43. L. J. Kong et al., EMBO J. 19, 3485–3495 (2000).
44. B. Desvoyes, C. Gutierrez, EMBO J. 39, e105802 (2020).
45. R. A. Weinberg, Cell 81, 323–330 (1995).
46. K. E. Knudsen et al., Mol. Cell. Biol. 20, 7751–7763 (2000).
47. D. Avni et al., Mol. Cell 12, 735–746 (2003).
48. A. Cruz-Ramírez et al., Cell 150, 1002–1015 (2012).
49. P. E. Jullien et al., PLOS Biol. 6, e194 (2008).
50. P. Sdek et al., Mol. Cell 20, 699–708 (2005).
51. E. Dumbliauskas et al., EMBO J. 30, 731–743 (2011).
52. L. Cao et al., PLOS Genet. 14, e1007230 (2018).
53. M. B. Boniotti, C. Gutierrez, Plant J. 28, 341–350 (2001).
54. I. Takahashi, S. Kojima, N. Sakaguchi, C. Umeda-Hara,
M. Umeda, Plant Cell Rep. 29, 307–315 (2010).
55. H. Nakagami, K. Kawamura, K. Sugisaka, M. Sekine,
A. Shinmyo, Plant Cell 14, 1847–1857 (2002).
56. C. E. Cockcroft, B. G. den Boer, J. M. Healy, J. A. Murray,
Nature 405, 575–579 (2000).
57. M. Meijer, J. A. H. Murray, Plant Mol. Biol. 43, 621–633 (2000).
58. A. K. Weimer et al., Development 145, dev160671 (2018).
59. J. L. Matos et al., eLife 3, e03271 (2014).
60. E. Sornay, C. Forzani, M. Forero-Vargas, W. Dewitte,
J. A. H. Murray, Plant J. 84, 41–55 (2015).
61. J. Adrian et al., Dev. Cell 33, 107–118 (2015).
62. S. Biedermann et al., EMBO J. 36, 1279–1297 (2017).
63. J. O. Lee, A. A. Russo, N. P. Pavletich, Nature 391, 859–865
(1998).
64. P. Zhao et al., Dev. Cell 49, 882–893.e5 (2019).
65. H. Takatsuka, C. Umeda-Hara, M. Umeda, Plant J. 82,
1004–1017 (2015).
66. E. S. Knudsen, C. Buckmaster, T. T. Chen, J. R. Feramisco,
J. Y. Wang, Genes Dev. 12, 2278–2292 (1998).

Simonini et al., Science 383, 646–653 (2024) 9 February 2024 7 of 8

RES EARCH | R E S E A R C H A R T I C L E
67. A. Yen, R. Sturgill, Exp. Cell Res. 241, 324–331 (1998).
68. S. P. Angus et al., Mol. Cell. Biol. 24, 5404–5420 (2004).
69. A. Kanellou, N. N. Giakoumakis, A. Panagopoulos, S. C. Tsaniras,
Z. Lygerou, Anticancer Res. 40, 2449–2456 (2020).
70. T.-F. Hsieh et al., Science 324, 1451–1454 (2009).
71. M. V. Martin, D. F. Fiol, V. Sundaresan, E. J. Zabaleta,
G. C. Pagnussat, Plant Cell 25, 1573–1591 (2013).
72. M. Bayer et al., Science 323, 1485–1488 (2009).
73. L. M. Costa et al., Science 344, 168–172 (2014).
ACKN OW LEDG MEN TS
We thank B. Scheres (Wageningen University and Research),
D. Bergmann (Stanford University), C. Gutierrez (CSIC Centro de
Biología Molecular ‘Severo Ochoa’), A. Schnittger (University of
Hamburg), D. Kurihara (Nagoya University), H. Wang (University of
Saskatchewan), the Nottingham Arabidopsis Stock Center, and the
Arabidopsis Stock Center IJPB-INRAE for providing seeds;
Z. Hasenkamp and R. Dudler (University of Zurich) for the gift of
Syl-A; M. W. Schmid (MWSchmid GmbH) for analysis, figure
preparation, and deposition of the LAM-RNA-Seq data; M. Nowack
(VIB-Ghent), C. Gutierrez, B. Desvoyes (CSIC Centro de Biología
Molecular ‘Severo Ochoa’), and M. Affolter (Biozentrum, University
Simonini et al., Science 383, 646–653 (2024) 9 February 2024
of Basel) for helpful discussions or comments on the manuscript;
C. Featherstone (Life Science Editors) for professional editorial
assistance during manuscript preparation; C. Baroux (University of
Zurich) for suggestions on image acquisition; C. Eichenberger,
F. Pasquer, A. Bolaños, D. Guthörl, and P. Kopf (University of
Zurich) for general laboratory support; and members of the
CHERI-T meeting for helpful discussions. Funding: This project
was supported by the University of Zurich, the European Union’s
Horizon 2020 research and innovation program (Marie
Sklodowska-Curie grant 798953 to S.B.), and the Swiss National
Science Foundation (grant 31003A_179553 to U.G.). Author
contributions: U.G. conceived, supervised, and raised funding for
the project. S.S. conceived, designed, and performed the
experiments. S.S. and U.G. analyzed and interpreted the data and
wrote the manuscript. S.B. generated the CRISPR-Cas9 mutants
and overexpression lines, provided technical support for the LAM
RNA-Seq, and commented on the manuscript. Competing
interests: S.S., S.B., and U.G. are inventors on patent application
EP22186785.6 submitted by the University of Zurich covering a
method for the induction of autonomous endosperm development.
Data and materials availability: All data are available in the
manuscript or the supplementary materials; the RNA-Seq raw data
have been deposited at the National Center for Biotechnology
Inormation Sequence Read Archive under accession number
PRJNA989505. License information: Copyright © 2024 the
authors, some rights reserved; exclusive licensee American
Association for the Advancement of Science. No claim to original
US government works. https://www.science.org/about/science-
licenses-journal-article-reuse. This research was funded in whole or in
part by the Swiss National Science Foundation (grant 31003A_179553),
a cOAlition S organization. The author will make the Author
Accepted Manuscript (AAM) version available under a CC BY public
copyright license.
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adj4996
Materials and Methods
Figs. S1 to S6
Tables S1 to S3
References (74–85)
MDAR Reproducibility Checklist
Submitted 18 July 2023; accepted 4 January 2024
10.1126/science.adj4996
8 of 8
RES EARCH

BIOGEOGRAPHY Landward retreat of the escarpment repre-

sents a major change to the landscape because
Escarpment evolution drives the diversification migration of the main water divide of the
island produces a cascade of downstream dy-
of the Madagascar flora namic changes to the topography, river net-
work topology, and erosion rates through time
Yi Liu1,2*†, Yanyan Wang3*†, Sean D. Willett3, Niklaus E. Zimmermann1,2‡, Loïc Pellissier1,2‡ (24). It is our hypothesis that these transient
and long-lived landscape perturbations result
Madagascar exhibits high endemic biodiversity that has evolved with sustained and stable rates of in continuously changing distributions of hab-
speciation over the past several tens of millions of years. The topography of Madagascar is itat in eastern Madagascar, providing a mech-
dominated by a mountainous continental rift escarpment, with the highest plant diversity and rarity anism for allopatric or parapatric speciation.
found along the steep, eastern side of this geographic feature. Using a process-explicit model, We tested this hypothesis by comparing space
we show that precipitation-driven erosion and landward retreat of this high-relief topography and time patterns of geographic and habitat
creates transient habitat organization through multiple mechanisms, including catchment change predicted by a landscape evolution
expansion, isolation of highland remnants, and formation of topographic barriers. Habitat isolation model with plant phylogenies, distribution
and reconnection on a million-year timescale serves as an allopatric speciation pump creating patterns, and the resulting biogeographic struc-
the observed biodiversity. turing of the flora.

Timing and spatial pattern of

M
adagascar exhibits disproportionately
high biodiversity (1, 2), which has ari-
sen in large part from its isolation from
other landmasses (3). Most species in
Madagascar evolved since its separa-
tion from Africa 120 million years ago (Ma)
and from the Seychelles and India about
90 Ma (3, 4), resulting in extensive endemism.
Although there is evidence of colonization of
Madagascar by specific groups, including ver-
tebrates (5), in the early Cenozoic, most of the
subsequent speciation has been vicariant, with
>94% of mammals and reptiles and 82% of
vascular plants being endemic to the island
(1). However, isolation alone does not explain
the high rate of in situ speciation of so many
taxa, and within-island geological processes
and their impacts on habitat distributions in
space and time could help explain how this
diversity has evolved (1, 2, 6).
The primary drivers of diversification in
Madagascar have been suggested to be the
diverse climate and changes in physical geo-
graphy (6). The wet, resource-rich rainforest in
eastern Madagascar is starkly different from
the temperate forests and subarid and arid
bioclimatic zones of the west, and these bio-
climatic zones have existed since at least the
Oligocene (7), providing opportunities for adapt-
ive speciation to the variety of habitats along
this environmental gradient (8, 9). The high-
relief topography of Madagascar has aided in
the formation of refugia during Quaternary
climate cycles (6) and provided mountain and
river barriers to the dispersion of vertebrates,
driving microendemism (10). Although these
and other mechanisms for diversification have
been proposed, no single process has been iden-
1
Swiss Federal Research Institute (WSL), 8903 Birmensdorf,
Switzerland. 2Department of Environmental Systems
Science, ETH Zürich, 8092 Zürich, Switzerland. 3Department of
Earth Sciences, ETH Zürich, 8092 Zürich, Switzerland.
*Corresponding author. Email: yi.liu@wsl.ch (Y.L.);
yanyan.wang@erdw.ethz.ch (Y.W.)
†These authors contributed equally to this work.
‡These authors contributed equally to this work.
tified that can explain why Madagascar has
such high biodiversity relative to other tropical
forest regions around the world (2).
Biodiversity developed through vicariant spe-
ciation is often associated with tectonically
active regions, where the processes of moun-
tain building lead to complex topography,
creating new habitats, generating environ-
mental gradients, and fragmenting existing
habitats (11–15). Madagascar does not fit this
hypothesis well because tectonic activity since
the rifting of Gondwanaland has been mini-
mal and localized (3, 4). However, Madagascar
is surrounded by passive continental margins
formed during the rifting from Africa, Australia,
and, most recently, India (16), and these mar-
gins include physiographic escarpments, par-
ticularly on the east coast, which represents
the youngest margin (17). The main water divide
of the island marks the westward limit of the
eastern escarpment, which is characterized by
an increase in elevation of up to 2 km over a
distance of <100 km (Fig. 1). The steep topo-
graphic gradient of the escarpment and the
lack of a corresponding topographic gradient
to the west implies a contrast in erosion rate,
which drives inland retreat of the escarpment
(Fig. 1 and fig. S1). The escarpment appears to
have retreated at a nearly constant rate over
the past 90 million years (Myr) and has been
estimated to be from 0.5 to nearly 2 km/Myr
(18) (fig. S2). The escarpment retreat is super-
imposed on a regional geodynamic uplift that
has occurred over the past 30 to 55 Myr in
response to deep mantle upwelling that has
affected the entire island (19–21). The up-
welling has maximum values in western and
northern Madagascar (Fig. 1B), and it is con-
sistent with Paleogene-Neogene marine sedi-
ments that are currently at an elevation of a
few hundred meters (20). The island also has
rift-related seismicity and volcanism as well as
surface faulting in the Alaotra-Ankay graben
system, which has been active since the Mio-
cene (22, 23).
plant diversification
We used published phylogenies to estimate the
rate of accumulation of lineages and the age
distribution of extant species. We generated
100 phylogenies of 8884 Madagascar seed
plants (77% of the island flora) based on a
dated megaphylogeny (25) with polytomies
randomly resolved. We additionally consid-
ered six time-calibrated maximum credibility
trees for specific clades with species endemic
to Madagascar as examples (fig. S3). The line-
age through time plots—indicating the num-
ber of species that diversified from birth-death
processes per clade through time—show a
near-steady rate of accumulation over at least
the past 45 Myr, which is supported by diver-
sification models (Fig. 1C and tables S1 and
S2). Although there is structure to the rate of
accumulation, including an increase toward
the present day for some clades, the timescale
of accumulation is tens of millions of years,
and the smooth species age distribution of
extant Madagascar plants reflects long, steady
processes of species accumulation. In addition,
there is a lack of information regarding extinc-
tion rates, owing to the paucity of fossil data in
Madagascar, which pulls the accumulation
curve toward the present (26).
Plant species richness is highly variable
across Madagascar, with the highest species
richness occurring in eastern Madagascar, as-
sociated with both higher precipitation and
the high relief of the eastern escarpment (Fig.
2, A to C). Among the 8884 seed plants mapped
on the island, ~72% of the species occur
along or below the escarpment of eastern
Madagascar (Fig. 2A and fig. S4). Orchidaceae
(86%), Rubiaceae (80%), and Asteraceae (84%)
are the richest families of species predomi-
nantly occurring in the escarpment area (Fig. 2D).
Transient geographic and habitat complexity
We reconstructed surface elevation change
over the past 45 Myr, including mechanisms of
escarpment retreat, subsidence in and around

Liu et al., Science 383, 653–658 (2024) 9 February 2024 1 of 6

RES EARCH | R E S E A R C H A R T I C L E

Fig. 1. Surface uplift, erosion, and plant lineage accumulation in Madagascar.
(A) Escarpment position estimates over the past 45 Myr based on modern retreat
rates (arrows). Escarpment retreat has resulted in nearly 2 km of erosion and
surface lowering. (B) Spatial distribution of additional surface uplift processes,
including dynamic uplift from mantle flow (contours of cumulative uplift), active
surface faulting (black lines), and volcano formation (red polygons). Seismicity is
indicated by yellow circles. AAG, Alaotra-Ankay graben. (C) The upper panel
illustrates the estimated time frame of landscape change [solid bars indicating age
constrained by geological evidence (4, 18–23, 44–48) and the dashed line indicating
poorly constrained age]. The orange line indicates average change of area below
the escarpment. The lower panel displays the accumulation of plant lineages
over time. The mean ages of 8884 extant seed plant species, estimated from
100 reconstructed phylogenetic trees, are indicated by the dark gray curve. The
horizontal light gray lines represent the standard deviation around the mean of
individual species ages. The colored lines represent the lineage through time plots of
six example clades from dated phylogenies (49–54).

the Alaotra-Ankay graben system, dynamic
uplift from mantle flow, and formation of
the volcanic edifices in northern and central
Madagascar (Fig. 2B and figs. S5 and S6). The
reconstruction of elevation change shows that
the north and west of Madagascar are domi-
nated by dynamic uplift, punctuated by local,
topographic change associated with volcanism.
By contrast, eastern Madagascar is dominated
by surface lowering by erosion associated with
the passage of the migrating escarpment (Fig.
2, B and E), with additional subsidence caused
by extensional faulting.
Our results indicate a close correspondence
between positive or negative changes in eleva-
tion, precipitation, and species richness (Fig. 2,
A to C). There are strong correlations between
richness and escarpment retreat (fig. S7; Spear-
man’s correlation coefficient r = 0.52, P <
0.001) and between richness and precipita-
tion (fig. S7; Spearman’s r = 0.66, P < 0.001).
Other metrics of elevation change have a weaker
correlation with species richness (fig. S7). There
are clear regional patterns, with the highest
richness corresponding to large increases or
decreases in elevation in the north and east of
the island, respectively (fig. S8). The orographic
forcing of precipitation by the escarpment to-
pography results in a strong correlation be-
tween escarpment retreat and precipitation
rate (fig. S7) so that we cannot statistically
separate the relative importance of climate
and landscape change on species richness.
Notably, statistical analyses do not capture
the nonlocal and complex relationships where
speciation is driven by time-dependent land-
scape change. In this work, we take a different
approach and attempt to characterize the land
surface and habitat changes predicted by
escarpment retreat to make more detailed
predictions of the processes contributing to
speciation.
To approach a causal understanding of the
processes at play, we constructed a landscape
evolution model simulating fluvial erosion of
an escarpment on the edge of a preexisting,
topographically elevated, but weathered high-
land (Fig. 3, fig. S9, table S3, and movie S1)
(24). The model enables documentation of the
topography and habitat heterogeneity in space
and time to assess potential mechanisms for
vicariant speciation. We use the landscape
evolution model Divide and Capture (DAC) (27),
modified to predict the dynamic plant habitat
as a function of surface elevation, slope, and
aspect. The model shows that, consistent with
theory (28, 29), the escarpment retreats from
the coast at a constant rate over time (24). As
the escarpment retreats, drainage basins be-
come longer and wider, but growth rates are
heterogeneous, and the main water divide de-
velops sinuosity as geographically isolated drain-
age basins grow at different rates (Fig. 3).
We provide a metric of landscape transience
by defining habitat patches in the model as
contiguous regions larger than 1 km2, with
elevation, slope, and aspect that lie within a
specified range (see materials and methods).
Through the course of the model, habitat
patches are dynamic, altering their shape, size,
and distance from the coastline (Fig. 3, C and
D). Within the escarpment region, patches fre-
quently appear, disappear, fragment, or merge
(Fig. 3D, fig. S10, and movies S2 and S3), pro-
moting population isolation, allopatric spe-
ciation, and microendemism. Splitting and
merging events in the model occur at a rate
of several hundred per million years (Fig. 3D),
which corresponds to an average of one event
(either a split or a merge) every 2 to 5 Myr for
each habitat patch (Fig. 3, D and E, and figs.
S11 to S13), a rate consistent with speciation
times in angiosperms (30). These habitat ge-
ometry dynamics are a direct consequence of

Liu et al., Science 383, 653–658 (2024) 9 February 2024 2 of 6

RES EARCH | R E S E A R C H A R T I C L E

Fig. 2. Biodiversity of seed plants, topographic change, and mean annual
precipitation in Madagascar. (A) Species richness map of 8884 seed plants on
the contemporary Madagascar landscape. (B) Estimate of topographic change
over the past 45 Myr based on combined effects of the four processes listed
n Fig. 1C. (C) Map of contemporary annual precipitation in Madagascar [available
at https://chelsa-climate.org (55)]. (D) Total number of species occurring on
Madagascar (turquoise) and distributed along the escarpment (dark turquoise)
in the top 20 species-rich families: Ac, Acanthaceae; Ap, Apocynaceae; Ar,
Arecaceae; Asp, Asphodelaceae; Ast, Asteraceae; B, Balsaminaceae; C,
Cyperaceae; E, Euphorbiaceae; F, Fabaceae; Lam, Lamiaceae; Lau, Lauraceae;
Ma, Malvaceae; Me, Melastomataceae; Ol, Oleaceae; Or, Orchidaceae; Ph,
Phyllanthaceae; Po, Poaceae; R, Rubiaceae; Sal, Salicaceae; Sap, Sapindaceae.
(E) Topographic reconstruction of escarpment retreat of the selected region
(a-a’, window size: 50 × 145 km) over the past 45 Myr.

the escarpment migration process; by contrast,
models of mountain uplift with no escarpment
or water divide migration exhibit different be-
havior, with an initial, short-duration tran-
sient stage followed by almost no change in
habitat pattern (figs. S11 and S14 and movie S4).
The model predicts that escarpment physio-
graphic features (Fig. 3D) disrupt north-south
connectivity, fostering species turnover and
local endemism. We test this prediction by
coupling species range maps and phylogenies
and identifying biogeographic realms charac-
terized by marked phylogenetic turnover (Fig.
4A) along the escarpment. We clustered 10
realms by similarity and found that they are
nested within three main groups differenti-
ated by distance to the water divide and lati-
tude with a north-south segregation (Fig. 4, B
and C). Geographic parameters, specifically the
combination of elevation and latitude, are ef-
fective predictors (13.9% of classification error
rate) of biogeographic realms, whereas biocli-
matic parameters provide weaker associations
(table S4). Latitudinal segregation of biogeo-
graphic realms also occurs between families
(Fig. 4D and fig. S15) and genera (figs. S16 and
S17). Similar patterns have been observed in
other taxonomic groups (6, 31). In addition, we
computed the Margalef index as a metric of
microendemism (32), documenting a high oc-
currence of rare species in the upper escarpment
(Fig. 4D and fig. S18). Many species occurring
in the escarpment are known only from a few
localities, where 10% display extremely narrow
ranges (distances between occurrences of <25 km;
fig. S19). Biogeographic segregation along the

Liu et al., Science 383, 653–658 (2024) 9 February 2024 3 of 6

RES EARCH | R E S E A R C H A R T I C L E

Fig. 3. Landscape evolution model simulating retreat of a segment of the
Madagascar escarpment. (A) Perspective view of the escarpment with three
drainage basins highlighted. Main water divide migrates inland at a rate of 2 km/Myr,
expanding the area of the coastal plain. (B) Topographic isolines during escarpment
retreat. (C) Nominal plant habitat distributions defined from elevation, slope, and
aspect of the landscape (table S5). (D) Examples of the area evolution of habitat patch
groups within the model; the bold lines indicate a focus patch, and the fine lines
show neighbor patches that merge or split with a focus example. Birth, death, and
area changes are also indicated. (E) Number of habitat patches (black lines), as
defined in (C), and the rate of birth, death, split, and merge (events per million years)
grouped into three elevation ranges. Feature (a) indicates isolated highland remnants.
Basin (b) indicates a catchment that initially drained from highland elevations but
is subsequently confined to lowland regions. Basin (c) is an expanding basin,
increasing in drainage area.

escarpment, a high rate of occurrence of rare
species, and a high frequency of restricted dis-
tribution (fig. S20) are characteristics that sup-
port the thesis that habitat fragmentation along
the escarpment controls speciation.
Changes in topography associated with es-
carpment retreat may affect speciation rates
through a number of potential mechanisms.
First, the westward migration of the escarpment
and its orographic influence on precipitation
(Fig. 2 and fig. S1C) implies an expansion of
the tropical forest habitats with their greater
productivity and opportunities for ecological
specialization (12, 14). Second, escarpment mi-
gration implies movement of the main water
divide through drainage area capture—a dis-
crete, episodic process whereby individual
river basins advance their headwater reaches
into the highlands (Fig. 3A). The episodic
nature of this process implies downstream
variations in erosion rate and surface mor-
phology in space and time, leading to tempo-
rally variable habitat and the strength of river
barriers. Third, differential retreat rates imply
that some catchments become isolated from
the main divide [e.g., basin (b) in Fig. 3, A and
B]. Many of these geographic features have

Liu et al., Science 383, 653–658 (2024) 9 February 2024 4 of 6

RES EARCH | R E S E A R C H A R T I C L E

Fig. 4. The biogeographic realms of seed plants in eastern Madagascar.
(A) Realms are defined by phylogenetic turnover (p-bsim) in range maps.
(B) Elevational range of each biogeographic realm. (C) Dendrogram representing
the similarities in the composition of plant lineages between the 10 distinct
biogeographic realms. (D) Density of breaks separating different biogeographic
realms along the escarpment based on the biogeographic realms of the top
20 species-rich families in low-elevation (coastal) and high-elevation areas. To
compare the Margalef rarity along the escarpment and the western region at
the same latitude, we estimated the rarity on a 50-km-resolution grid and
subsequently averaged within each interval.

been demonstrated to be important in driving
speciation (2). For example, isolated catchments
have served as centers for endemism (6), and
large rivers have acted as barriers to lemur
dispersion (10). Although most past studies
have focused on animals, and it is difficult to
associate specific speciation events with spe-
cific processes, the close correlation of species
richness and rarity with the active escarpment
provides circumstantial evidence that some
combination of these processes has been re-
sponsible for an increased speciation rate
across multiple taxa.
Differentiating climate from geomorphic process
The escarpment and topographic highlands
have a strong orographic effect on the precip-
itation pattern across Madagascar such that
the bioclimatic zone and the escarpment geo-
morphic domain correspond almost perfectly,
and both correlate with the region of high spe-
cies richness, which makes it difficult to dis-
tinguish between high speciation rate and a
high, climate-modulated species carrying ca-
pacity. Tropical moist forests, such as those in
eastern Madagascar, are associated with a
higher biomass, capable of sustaining higher
levels of diversity through larger populations
and specific niche formation (33), and the
tropical forest area increases in size as the es-
carpment moves west. Adaptation along the
west-east climate gradient offers opportunity
for dry-climate species to adapt to the wetter
environments, or vice versa. Climate change
associated with moist-dry cycles during the
Quaternary may also have affected speciation
rate through habitat fragmentation and recon-
nection (34). However, there are three lines of
evidence that support geomorphic-driven vi-
cariance. First, the phylogenetic analysis (Fig.
1C and table S2) shows a near-constant rate of
diversification over at least the late Neogene,
with no acceleration during the Quaternary,
when climate cycles became more pronounced
(35). Second, the latitudinal turnover and spe-
cies rarity distribution (Fig. 4D and fig. S18)
have no direct climatic association, given the
lack of north-south climate variability. By con-
trast, the landscape evolution model predicts
transient changes in habitat connectivity in
the north-south direction, associated with
changes in the large rivers and interfluvial
ridges, consistent with the species turnover
pattern. Third, the timescale of habitat frag-
mentation by climate change is much faster
than what we predict for geomorphic tran-
sience. Quaternary climate cycles have periods
from 22,000 to 100,000 years (35), whereas
geomorphic transience in Madagascar creates
habitat patches with a typical isolation time
of several million years (Fig. 3, D and E), more
consistent with the empirical observations for
speciation time in plants (30).
Implications for the importance of geomorphic
vicariant speciation
Madagascar can serve as an example of how
speciation is enhanced by comparably small-
scale geomorphic landscape dynamics driving
microendemism. Although the importance of
catchment geography for Madagascar’s micro-
endemism has been recognized (2), our analy-
ses suggest that catchment geometry and other
geomorphic characteristics are highly transient
in response to escarpment migration. Lateral
migration of an escarpment or a water di-
vide is more disruptive to a landscape com-
pared with vertical tectonic uplift because
disruption of the planform geometry of a
river network triggers a downstream chain of
perturbation, creating yet more heterogeneity
in physiography and longer response times
(compare Fig. 3 with fig. S11) (36). These per-
turbations result in lower habitat connectivity
(37), fragmentation, and reconnection (fig. S11).
This process is evident in the segregation of
lineages and the spatial pattern of diversity
and endemism. Although we focused on plant
habitat parameters in our model, the limited
range sizes and latitudinal segregation are
representative of other taxa in Madagascar,
including lemurs (6), amphibians, and rep-
tiles (31), which show distributions closely
matching features of the escarpments.
There are other settings around the world
where geomorphic disequilibrium and high
biodiversity coexist. The Western Ghats of
India on the conjugate margin to Madagascar
also display a high level of plant endemism
(38). The rift margins of Brazil, South Africa,
and parts of Australia similarly stand out as
regions with transient escarpments and high
endemism (30, 39). Likewise, southeastern
North America retains a high degree of geo-
morphic disequilibrium and aquatic species
diversity 180 Myr after its formation as a con-
tinental margin (40, 41). High biodiversity
observed in other tectonic settings character-
ized by geomorphic transience and habitat
fragmentation includes the East African rift (42)
and the Andean foreland (43). Our results

Liu et al., Science 383, 653–658 (2024) 9 February 2024 5 of 6

RES EARCH | R E S E A R C H A R T I C L E
suggest that the complex dynamic response of
the landscape to tectonic forcing can persist
for tens or even hundreds of millions of years
and that the consequent disequilibrium leads
to shifting patterns of habitat connectivity in
both space and time, acting as a speciation
pump to increase biodiversity long after the
cessation of tectonic activity.
RE FE RENCES AND N OT ES
1. A. Antonelli et al., Science 378, eabf0869 (2022).
2. M. Vences, K. C. Wollenberg, D. R. Vieites, D. C. Lees,
Trends Ecol. Evol. 24, 456–465 (2009).
3. M. J. de Wit, Annu. Rev. Earth Planet. Sci. 31, 213–248
(2003).
4. A. D. Gibbons, J. M. Whittaker, R. D. Müller, J. Geophys.
Res. Solid Earth 118, 808–822 (2013).
5. A. Crottini et al., Proc. Natl. Acad. Sci. U.S.A. 109, 5358–5363
(2012).
6. L. Wilmé, S. M. Goodman, J. U. Ganzhorn, Science 312,
1063–1065 (2006).
7. N. A. Wells, in The Natural History of Madagascar,
S. M. Goodman, J. P. Benstead, Eds. (Univ. of Chicago Press,
2003), pp. 16–34.
8. A. D. Yoder et al., Proc. Natl. Acad. Sci. U.S.A. 113, 8049–8056
(2016).
9. J. L. Younger et al., Mol. Phylogenet. Evol. 139, 106535
(2019).
10. S. M. Goodman, J. U. Ganzhorn, J. Biogeogr. 31, 47–55
(2004).
11. C. Badgley et al., Trends Ecol. Evol. 32, 211–226 (2017).
12. P. Descombes et al., Glob. Ecol. Biogeogr. 27, 233–244
(2018).
13. C. Hoorn, V. Mosbrugger, A. Mulch, A. Antonelli, Nat. Geosci. 6,
154 (2013).
14. A. Antonelli et al., Nat. Geosci. 11, 718–725 (2018).
15. C. Rahbek et al., Science 365, 1114–1119 (2019).
16. J. O. Thompson, M. Moulin, D. Aslanian, P. De Clarens,
F. Guillocheau, Earth Sci. Rev. 191, 26–56 (2019).
17. A. Gilchrist, M. Summerfield, Nature 346, 739–742
(1990).
18. Y. Wang, S. D. Willett, D. Wu, N. Haghipour, M. Christl,
Geochem. Geophys. Geosyst. 22, e2021GC009979 (2021).
19. G. G. Roberts, J. D. Paul, N. White, J. Winterbourne,
Geochem. Geophys. Geosyst. 13, Q04004 (2012).
20. S. N. Stephenson, N. J. White, T. Li, L. F. Robinson,
Earth Planet. Sci. Lett. 519, 61–69 (2019).
Liu et al., Science 383, 653–658 (2024) 9 February 2024
21. S. N. Stephenson et al., Geochem. Geophys. Geosyst. 22,
e2020GC009624 (2021).
22. T. M. Kusky, E. Toraman, T. Raharimahefa,
C. Rasoazanamparany, Gondwana Res. 18, 274–294
(2010).
23. M. Mietton et al., Catena 165, 22–41 (2018).
24. Y. Wang, S. D. Willett, D. Wu, J. Geophys. Res. Earth Surf. 128,
e2022JF007001 (2023).
25. S. A. Smith, J. W. Brown, Am. J. Bot. 105, 302–314 (2018).
26. Y. Xing, R. H. Ree, Proc. Natl. Acad. Sci. U.S.A. 114,
E3444–E3451 (2017).
27. L. Goren, S. D. Willett, F. Herman, J. Braun, Earth Surf. Process.
Landf. 39, 522–545 (2014).
28. J. Braun, Gondwana Res. 53, 209–224 (2018).
29. S. D. Willett, S. W. McCoy, H. W. Beeson, Nature 561, 528–532
(2018).
30. J. Igea, A. J. Tanentzap, Ecol. Lett. 23, 692–700 (2020).
31. J. L. Brown, A. Cameron, A. D. Yoder, M. Vences, Nat. Commun.
5, 5046 (2014).
32. B. J. Enquist et al., Sci. Adv. 5, eaaz0414 (2019).
33. P. W. Richards, Biol. J. Linn. Soc. 1, 149–153 (1969).
34. J.-L. Mercier, L. Wilmé, Madag. Conserv. Dev. 8, 63–68 (2013).
35. L. E. Lisiecki, M. E. Raymo, Quat. Sci. Rev. 26, 56–69
(2007).
36. J. S. Scheingross, A. B. Limaye, S. W. McCoy, A. C. Whittaker,
Nat. Rev. Earth Environ. 1, 661–676 (2020).
37. T. Salles, P. Rey, E. Bertuzzo, Earth Surf. Dyn. 7, 895–910
(2019).
38. N. R. Gunawardene et al., Curr. Sci. 93, 1567–1572 (2007).
39. G. Murali, R. Gumbs, S. Meiri, U. Roll, Sci. Adv. 7, eabe5582
(2021).
40. S. F. Gallen, Earth Planet. Sci. Lett. 493, 150–160 (2018).
41. M. F. Stokes et al., Science 380, 855–859 (2023).
42. A. J. Plumptre et al., Biol. Conserv. 134, 178–194 (2007).
43. F. A. S. Cassemiro et al., Proc. Natl. Acad. Sci. U.S.A. 120,
e2211974120 (2023).
44. G. Eagles, H. H. Hoang, Geophys. J. Int. 196, 1–14 (2014).
45. L. Melluso et al., J. Petrol. 46, 1963–1996 (2005).
46. J. M. Bardintzeff, J. P. Liégeois, B. Bonin, H. Bellon,
G. Rasamimanana, Gondwana Res. 18, 295–314 (2010).
47. K. E. Samonds et al., J. Vertebr. Paleontol. 29, 1233–1243
(2009).
48. E. Laville et al., Bull. Soc. Géol. France 169, 775–788
(1998).
49. J. Hackel et al., J. Biogeogr. 45, 750–761 (2018).
50. A. Gamisch, K. Winter, G. A. Fischer, H. P. Comes, New Phytol.
231, 1236–1248 (2021).
51. S. Federman et al., BMC Evol. Biol. 15, 216 (2015).
52. C. G. Boluda, C. Christe, Y. Naciri, L. Gautier, Taxon 71,
360–395 (2022).
53. D. J. Hearn, M. Evans, B. Wolf, M. McGinty, J. Wen, J. Syst. Evol.
56, 340–359 (2018).
54. S. Faurby, W. L. Eiserhardt, W. J. Baker, J.-C. Svenning,
Mol. Phylogenet. Evol. 100, 57–69 (2016).
55. D. N. Karger et al., Sci. Data 4, 170122 (2017).
56. Y. Wang, yanyanwangesd/DAC-weaksurfacelayer: DAC-
Escarpment-weaksurfacelayer, version 1.0.0, Zenodo (2023);
https://doi.org/10.5281/zenodo.7686096.
57. Y. Wang, yanyanwangesd/landscape_physiographic_
patchness: Initial release, version 1.0.0, Zenodo (2023);
https://doi.org/10.5281/zenodo.8302036.
58. Y. Liu, Y. Wang, S. Willett, N. E. Zimmermann, L. Pellissier,
Escarpment evolution drives the diversification of the
Madagascar flora, EnviDat (2023); https://www.doi.org/
10.16904/envidat.387.
AC KNOWLED GME NTS
We thank the reviewers who provided constructive comments that
improved the manuscript. Funding: Y.L. thanks the Chinese
Scholarship Council (CSC) (201904910589) and WSL for its
financial support. Support was received from the ETH Research
Commission through the BECCY project. Author contributions:
N.E.Z. and L.P. jointly supervised Y.L. Conceptualization:
Y.L., Y.W., S.D.W., N.E.Z., and L.P.; Methodology: Y.L., Y.W., S.D.W.,
N.E.Z., and L.P.; Investigation: Y.L., Y.W., S.D.W., N.E.Z., and L.P.;
Visualization: Y.L. and Y.W.; Funding acquisition: Y.L., S.D.W.,
N.E.Z., and L.P.; Project administration: Y.L., N.E.Z., and L.P.;
Supervision: S.D.W., N.E.Z., and L.P.; Writing: Y.L., Y.W., S.D.W., N.E.Z.,
and L.P. Competing interests: The authors declare that they have
no competing interests. Data and materials availability: The
landscape evolution model used in this study, DAC (56), and codes for
the habitat patch change analysis of the landscape evolution models
(57) are available at Zenodo repositories. The data generated,
analyzed, and presented in this study are available from the EnviDat
repository (58). License information: Copyright © 2024 the authors,
some rights reserved; exclusive licensee American Association for the
Advancement of Science. No claim to original US government works.
https://www.science.org/about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adi0833
Materials and Methods
Figs. S1 to S20
Tables S1 to S7
References (59–101)
MDAR Reproducibility Checklist
Movies S1 to S4
Submitted 14 April 2023; accepted 4 January 2024
10.1126/science.adi0833
6 of 6
RES EARCH

PLANT SCIENCE sociated with oil glands. Genetic mapping of

the oil glandless trait in HP has been proven
Molecular regulation of oil gland development and technically challenging as a result of poly-
embryony (a kind of apomixis) and extended
biosynthesis of essential oils in Citrus spp. juvenile phase in Citrus. To address these
technical hurdles, a solution was found through
Hongxing Wang1†, Jie Ren1†, Shiyun Zhou1, Yaoyuan Duan1, Chenqiao Zhu1, Chuanwu Chen2, the discovery of a monoembryonic Hong Kong
Ziyan Liu1, Qingyou Zheng1,3, Shu Xiang1, Zongzhou Xie1, Xia Wang1,3, Lijun Chai1,3, Junli Ye1, kumquat (HK) (Fortunella hindsii, Mini-Citrus)
Qiang Xu1,3, Wenwu Guo1,3, Xiuxin Deng1,3*, Fei Zhang1,3* that possesses a shortened juvenile phase simi-
lar to annual crops (18).
Secretory structures in terrestrial plants serve as reservoirs for a variety of secondary metabolites. We crossed the glandless HP with the oil
Among these, the secretory cavity of the Rutaceae family is notable for containing essential oils with a gland–bearing HK (fig. S3). The F1 population
wide range of applications. However, the molecular basis underlying secretory cavity development is displayed the presence of oil glands, aligning
unknown. Here, we reveal a molecular framework for Citrus oil gland formation. Using genetic mapping with the oil gland presence in HK. The F2 gen-
and genome editing, we demonstrated that this process requires LATE MERISTEM IDENTITY1 (LMI1), eration arising from each monoembryonic F1
a key regulator of leaf serration. A conserved GCC box element of the LMI1 promoter recruits plant displayed the segregation ratio of 3:1
DORNROSCHEN-like (DRNL) for transcriptional activation. This DRNL-LMI1 cascade triggers MYC5 among with or without oil glands in their
activation, facilitating the development of oil glands and the biosynthesis of essential oils. Our findings leaves (fig. S3). These results suggested a mono-
spotlight cis-regulatory divergence within leaf shape genes, propelling novel functional tissue formation. genic recessive gene that exerts influence over
gland development. In addition, HP exhibits

T
o defend against herbivores and patho-
gens, vascular plants have evolved vari-
ous types of defense mechanisms, which
involve physical and chemical defenses
such as thorns and various compounds
in secretory structures, respectively. Secretory
structures include epidermal glandular tri-
chomes, subepidermal secretory cavities, in-
ternal resin ducts, and laticifers (1). Secretory
cavities are prevalent among terrestrial plants,
particularly in plant families such as Rutaceae,
Hypericaceae, Myrtaceae, and Leguminosae
(2–5). Within the Rutaceae family, secretory
cavities are a notable trait and are commonly
referred to as oil glands in Citrus plants. These
oil glands are present in various above-ground
parts of Citrus plants, with a noteworthy con-
centration in the pigmented pericarp region of
the fruits (fig. S1). Within Citrus oil glands, a
wide range of secondary metabolites are stored
as essential oils (EOs), including monoterpenes,
fatty acids, and flavonoids (6–8). Citrus EOs
have shown considerable practical value in
flavor, cleaning, fragrance, and agrochemical
industries, and constituted more than 30% of
all EOs sold in 2022.
Molecular pathways specifying plant secre-
tory structures have been studied mainly in
glandular secretory trichomes, such as those
in Artemisia and tomato (9). These studies have
revealed that the initiation of epidermal cell
reprogramming for the formation of multicel-
lular secretory structures is orchestrated by
1
National Key Laboratory for Germplasm Innovation &
Utilization of Horticultural Crops, Huazhong Agricultural
University, Wuhan 430070, China. 2Guangxi Key Laboratory
of Germplasm Innovation and Utilization of Specialty
Commercial Crops in North Guangxi, Guangxi Academy of
Specialty Crops, Guilin 541004, China. 3Hubei Hongshan
Laboratory, Wuhan 430070, China.
*Corresponding author. Email: feizhang@mail.hzau.edu.cn
(F.Z.); xxdeng@mail.hzau.edu.cn (X.X.D)
†These authors contributed equally to this work.
class IV homeodomain leucine-zipper (HD-
ZIP IV) transcription factors (9–13). Despite
some insights into the underlying morpho-
genetic and cellular mechanisms driving the
formation of secretory cavities (2, 14), the mo-
lecular basis of the early developmental pro-
cesses behind subepidermal secretory structures,
such as secretory cavities, remains unknown.
This lack of understanding primarily stems
from the absence of good model systems and
genetic tools for comprehensive investigations.
We have recently established the Citrus thorn
as a model system for stem cell activity regu-
lation using an efficient CRISPR-Cas9 system
in the Citrus hybrid Carrizo citrange (Citrus
sinensis 'Washington' sweet orange × Poncirus
trifoliata) (15, 16). This system can also be used
in identifying mechanisms for the develop-
ment of functional tissues such as oil glands
as a result of its simplifying nature: radial
symmetry, terminal differentiation, and clear
tip-base boundary. We report a molecular
framework that governs the development of
Citrus oil glands, which was uncovered through
the combined utilization of the Citrus thorn
system and a naturally occurring glandless
kumquat mutant.
Hua Pi glandless mutant encodes for LMI1
The Hua Pi kumquat (HP) (Fortunella crassifolia)
stands out as the Citrus cultivar that com-
pletely lacks oil glands in its leaves and a con-
siderably diminished number of oil glands in
its fruits (17) (Fig. 1, A and B). This distinctive
trait is observed in contrast to the Rongan
kumquat (RA), from which it originated as a
spontaneous seedling mutant (fig. S2). Notably,
HP exhibits heightened sensitivity to tobacco
cutworms (Spodoptera litura) in both field and
laboratory conditions compared with RA (Fig.
1, C and D, fig. S2). This phenomenon could
potentially result from the absence of “bad
taste” defense chemicals that are typically as-
characteristic smooth leaf margins. This trait
is found to be closely associated with the oil-
glandless phenotype across the F2 population.
Our BSA-seq results showed a single broad
peak within a 4.5-Mb region on chromosome 6
that represented the HP mapping interval
(Fig. 1E and fig. S4). Using fine mapping, we
reduced the mapping interval to a 413-kb re-
gion containing 54 protein-coding genes (Fig.
1E and fig. S4). Among these genes, only two
showed reduced expression in HP compared
with RA (fig. S5A and data S1). We reasoned
that this gene should be expressed in the tis-
sue containing oil glands such as the thorn
base but might be less expressed in glandless
tissue such as the thorn tip (fig. S5B and data
S2). With this scenario in mind, we further
narrow down the candidate to one possible
gene (fig. S5C). Resequencing of the HP ge-
nome shows that this gene is in an 8.9-kb frag-
ment deleted in HP (Fig. 1E and fig. S5D). This
gene encodes a nucleus-localized class I homeo-
domain leucine-zipper (HD-ZIP I) family tran-
scription factor (fig. S6), the only Citrus ortholog
of Arabidopsis thaliana LATE MERISTEM
IDENTITY1 (AtLMI1) and Cardamine hirsuta
REDUCED COMPLEXITY (RCO), key regulators
of leaf serration and dissected-leaf formation,
respectively (19, 20).
We generated loss-of-function mutants and
performed complementation assays to deter-
mine the role of CsLMI1. Because Carrizo cit-
range has normal oil glands and a well-established
plant transformation system, it was used to
trace the relationship between RA and HP.
We used the CRISPR-Cas9 system to knock out
CsLMI1 by three different gRNAs individually
or two in combination (Fig. 1F and fig. S7). All
Carrizo citrange cr-cslmi1 mutants showed a
smooth leaf margin phenotype, akin to the
Arabidopsis atlmi1 mutant (20), suggesting
the conserved role of CsLMI1 in promoting leaf
serration. In addition, those mutants displayed
an oil-glandless phenotype in all aerial organs

Wang et al., Science 383, 659–666 (2024) 9 February 2024 1 of 8

RES EARCH | R E S E A R C H A R T I C L E
Fig. 1. Leaf shape regulator LMI1 is required for oil gland development in
Citrus. (A) and (B) Oil gland phenotype of fruits and leaves of Rong An Kumquat
(RA) and glandless Hua Pi kumquat (HP). HP is a spontaneous seedling
mutant of RA. (C) Representative of leaf disc feeding assay at 0 hours (top)
and 12 hours (bottom) after placing tobacco cutworm (Spodoptera litura) stage-
3 caterpillars. The same results were observed from five biological repeats.
(D) Leaf feeding rate of S. litura to HP and RA. ****P < 0.0001 (Student’s t test).
(E) Cloning of the HP locus, wherein the glandless HP was crossed to the oil
gland–bearing monoembryonic Mini-Citrus, and the F2 populations resulting from
Wang et al., Science 383, 659–666 (2024) 9 February 2024
selfing the monoembryonic F1 were used for mapping. Bulked-segregant analysis
coupled to whole genome sequencing (BSA-Seq) mapped the HP locus to a
4.5-Mb region whereas fine mapping by markers defined the candidate region of
413-kb, containing 54 genes. The candidate gene LMI1 (pink) is in an 8.9-kb
deleted region in HP. (F) CRISPR-Cas9–induced homozygous mutations in
CsLMI1 of Carrizo citrange. Arrowheads indicate the mutation sites. Blue color
indicates the homeodomain and orange indicates the associated leucine zipper
domain. aa, amino acids; +1 bp, 1 base pair insertion that caused frameshift
and premature translation stops. (G) Complementation of glandless mutants by
2 of 8
RES EARCH | R E S E A R C H A R T I C L E
transgenic expression of LMI1. A genomic LMI1 DNA fragment from Mini-Citrus
containing a 5048-bp upstream promoter, a 1865-bp gene region, and a 1193-bp
downstream regulatory region was introduced into the glandless F2 plants containing
the homozygous HP mutation. +C, includes the complementation construct. (H) The
CsLMI1 promoter transcriptional reporter. A 5048-bp upstream promoter was used to
such as leaves, demonstrating the indispensable
role of an LMI1-like gene in oil gland formation
(fig. S7). The overall phenotypes of the cr-cslmi1
mutants are reminiscent of the recessive HP
mutant, indicating that CsLMI1 is the causative
gene for the HP mutant phenotype. To further
validate this conclusion, we introduced a CsLMI1
genomic clone (CsLMI1 genomic sequence with
a 5-kb upstream promoter and 1-kb down-
stream regulatory sequence) into the glandless
F2 plants or HP mutants and found that this
transgene fully complemented F2 homozygous
mutant plants (four independent transgenic
plants) and HP mutants (seven independent
transgenic plants) (Fig. 1G and fig. S8). Collec-
tively, these genetic results demonstrated that
the CsLMI1 is responsible for the glandless and
smooth leaf margin phenotype of HP.
LMI1 expression expands in Rutaceae species
We next examined the expression pattern of
CsLMI1 in young shoots and leaves of Carrizo
citrange through reporter gene assays and
RNA in situ hybridization. Using a 5-kb up-
stream promoter of CsLMI1 to drive the eGFP
reporter, we observed a clear and specific sig-
nal in oil glands along with signals at leaf mar-
gins (Fig. 1H). CsLMI1 mRNA was identified at
the leaf or leaflet margins of young leaves
while being absent in the shoot apical meri-
stem (Fig. 1I). This pattern mirrors the ex-
pression behavior of AtLMI1 in Arabidopsis,
indicating a conserved function of LMI1 in leaf
shape regulation. During leaf development,
CsLMI1 displayed scattered expression, consti-
tuting epidermal cells and the immediate layers
beneath, which eventually form the oil gland
(Fig. 1, I and J). This distinct and consistent ex-
pression pattern of CsLMI1 is observed through-
out the stages of oil gland development, leading
up to the beginning of lumen formation.
To test whether LMI1 is conserved for oil
gland development in the Rutaceae family, we
selected Chinese box orange (Atalantia buxifolia,
a primitive Citrus species), Wampee (Clausena
lansium, an exotic fruit species), and orange
jasmine (Murraya paniculata) for LMI1 ex-
pression pattern analysis using RNA in situ
hybridization. These species were selected as
they all have oil glands in their leaves and
their genomes are readily available (fig. S9).
LMI1 genes in A. buxifolia, C. lansium, and
M. paniculata all exhibited a similar expres-
sion pattern to LMI1 in the Australia finger
lime (Citrus australasica) during the devel-
opment of leaf oil glands (fig. S9). These re-
Wang et al., Science 383, 659–666 (2024) 9 February 2024
drive the eGFP reporter in Carrizo citrange. (I and J) CsLMI1 in situ expression of
the shoot apex (I) and developing leaves (J) of Carrizo citrange. Longitudinal sections
were used. A star marks the shoot apical meristem, empty arrowheads point to the
signal at leaf margin tissues, and blue arrowheads point to developing oil glands. Scale
bars, 0.5 cm in (A), (B), and (F); 1 cm in (C); 1 mm in (G); 200 mm in (H) to (J).
sults indicate a conserved role of LMI1 genes
for gland development in the Rutaceae family.
A GCC box is necessary for oil gland initiation
We next explored the specific cis-regulatory ele-
ment of CsLMI1 that plays a role in oil gland
initiation. We began by confirming the discern-
ible expression pattern discrepancy of CsLMI1
between the thorn base and tip through quan-
titative real-time PCR (qRT-PCR) and RNA
in situ hybridization (Fig. 2, A and B). This
observation prompted us to conduct genome-
wide measurement of chromatin accessibility
through assay for transposase-accessible chro-
matin using sequencing (ATAC-seq) in both
thorn base and tip tissues (Fig. 2C). In the
promoter region of CsLMI1, we identified two
ATAC-seq peaks that exhibited significant dif-
ferences between the base and tip tissues (peak
around -3895bp, distal; peak around -280bp,
proximal). A closer examination of the distal
segment of the LMI1 promoters revealed a con-
served GCC box across all Citrus species as well
as in several species within the Rutaceae family,
including A. buxifolia, C. lansium, M. paniculata,
and Zanthoxylum bungeanum (Fig. 2D), sug-
gesting that this cis-regulatory element could
be linked with control of transcriptional activation
of the LMI1 gene and development of oil glands.
To uncover the role of this cis-regulatory
element, we designed two gRNAs aimed at
targeting the GCC box independently and in
combination, using CRISPR-Cas9. Among the
18 distinct transgenic plants harboring con-
structs that targeted the GCC box within the
CsLMI1 promoter, five lines exhibited a notice-
able reduction in the size and number of oil
glands, although no distinction was observed
in leaf margins as compared with the wild type
(WT) (Fig. 2, E and F). Notably, all five lines
bearing oil gland phenotypes had deletions of
the GCC box, with 40 to 70% determined by a
next generation sequencing (NGS) approach.
These plants also contained the mutation type
of single base pair insertions or deletions
(indels) within the GCC box. Conversely, the
other 13 lines, which exhibited no noticeable
phenotypic alteration, displayed only single
base pair indels or no mutations. Consistent
with the oil gland phenotype, the lines man-
ifesting the oil gland phenotype displayed a
notable reduction in CsLMI1 expression. By
contrast, the lines harboring single base pair
indels did not exhibit any substantial altera-
tion in CsLMI1 expression (Fig. 2G). This find-
ing strongly suggests that the evolutionarily
conserved GCC element box within the CsLMI1
promoter is pivotal for its transcriptional activa-
tion, eventually influencing the development of
oil glands.
DRNL binds to the GCC box to activate LMI1
To identify regulatory factors that contribute
to the activation of CsLMI1 by interacting with
the GCC box element, we leveraged the Citrus
thorn system. Given that GCC box elements
are recognized by APETALA2/Ethylene re-
sponsive factor (AP2/ERF) transcription fac-
tors (21), we focused on AP2/ERF members
that exhibited significantly higher expression
at the thorn base with glands in comparison to
the glandless thorn tip (fig. S10 and data S2).
Through this strategy we identified four AP2/
ERF members, one of which is the nucleus-
localized single Citrus ortholog of the Arabi-
dopsis DRN and DRNL genes (fig. S11) (22, 23).
DRN and DRNL genes are key regulators of
embryonic patterning, floral meristem identity,
and lateral organ initiation (22–24). Knocking
out CsDRNL—but not the other three genes—
resulted in a lack of oil glands in leaves in ad-
dition to the smooth leaf margin phenotype,
similar to the cr-cslmi1 mutant (Fig. 3B and
fig. S12), suggesting a potential collaborative
role of CsDRNL and CsLMI1 in regulating both
leaf shape and oil gland development in Citrus.
We next determine the relationship between
CsDRNL and CsLMI1 (Fig. 3C). A qRT-PCR
analysis showed that the expression of CsLMI1
in the developing leaves of cr-csdrnl mutants
was reduced to 40% as compared to the WT.
Conversely, in Carrizo citrange plants with
constitutive CsDRNL overexpression (fig. S13),
CsLMI1 expression was increased to ∼60 times
that of the WT level and the oil gland density of
those plants was substantially increased. We
also generated inducible lines by fusing CsDRNL
with the glucocorticoid receptor (p35S:CsDRNL-
GR) (fig. S13) (25). After induction of CsDRNL
activity, we found a considerable increase in
CsLMI1 expression. Collectively, these findings
substantiate that CsDRNL functions to activate
CsLMI1 expression in vivo, thereby establish-
ing a molecular and functional link between
these two key transcription factors in the
regulation of oil gland development and leaf
serration.
We determined the expression pattern of
CsDRNL by GFP reporter gene assays and RNA
in situ hybridization (Fig. 3D and fig. S14).
When a 5.8-kb regulatory sequence of CsDRNL
was used to drive the eGFP reporter gene,
3 of 8
RES EARCH | R E S E A R C H A R T I C L E

Fig. 2. Divergence of the LMI1 promoter contributes to oil gland development.
(A) qRT-PCR analysis of CsLMI1 at thorn base and tip. (Left) tip and base regions
collected when the young thorn was ~1 cm long. Blue dashed line indicates the
site cut for sampling. Error bars are one standard deviation (SD) from three
biological replicates. ***P < 0.001 (Student’s t test). (B) CsLMI1 in situ expression at
the young thorn base. Signals can be observed from epidermis to inner oil gland
tissues. (C) ATAC-seq analysis of the open chromatin at the CsLMI1 locus on
chromosome 6 (chr6). Peak indicates more open chromatin in that region. Blue
boxes indicate exons of CsLMI1 and the gray box represents the 3′ untranslated
region. The data were representative of three biological repeats, which have similar
patterns. (D) Alignment of the distal enhancer containing the GCC box of Rutaceae
family species. (E) CRISPR-Cas9 editing of the GCC box at the CsLMI1 promoter
of Carrizo citrange. The target sequence and PAM sequence were indicated by blue
and orange, respectively. The 1-bp insertion for the #3 plant is in bold. The
percentage indicates the rate of the mutation type in an average of 2500 reads
per sample determined by deep sequencing. (F) Number of oil glands per 25 mm2
counted under a microscope. Bars represent means ± SD (n = 6). ***P < 0.001
(Student’s t test). (G) qRT-PCR analysis of CsLMI1 in the promoter mutants with
GCC box editing. Error bars are one SD from three biological replicates. ***P < 0.001
(Student’s t test). Scale bars, 50 mm in (B) and 0.5 cm in (E).

distinct signals were observed in developing
oil glands within young leaves, mainly restricted
within 1 to 2 cells at the epidermis (176 out of 222
developing oil glands) (Fig. 3D and fig. S14A).
RNA in situ hybridization also revealed that
CsDRNL expression could be mainly de-
tected at the epidermis and occasionally at
subepidermal cells as oil gland initials (fig.
S14B). The expression pattern of CsDRNL in oil
gland initials closely aligns with the pheno-
types observed in the cr-csdrnl mutants.
Notably, CsLMI1 mRNA was also detected
at the oil gland initials, implying a potential
overlap with CsDRNL expression. To validate
this, we employed RNA in situ hybridization to
examine consecutive sections using different
probes. The results unequivocally revealed mul-
tiple shared expression sites between CsDRNL
and CsLMI1. CsDRNL's expression was more
prominent toward the epidermis, whereas
CsLMI1 exhibited the ability to extend the ex-
pression signals deeper into the tissues (Fig. 3,
E and F). These findings provide evidence of
the joint involvement of CsDRNL and CsLMI1
in oil gland development.
We next investigated the direct regulation of
the CsLMI1 promoter by CsDRNL. Using elec-
trophoretic mobility shift assay (EMSA), we
assessed the binding affinity of CsDRNL to
the CsLMI1 promoter. Notably, band shifts were
observed only when the WT probe from the
distal segment of the CsLMI1 promoter was
incubated with the CsDRNL protein. This in-
teraction was absent when mutated probes
with GCC box deletions were used (Fig. 3G).
Notably, the presence of a 1-bp insertion with-
in the GCC box only slightly affected the bind-
ing of CsDRNL. This observation indicates
that the CsDRNL protein can tolerate slight
variations, which aligns with our previous find-
ings that such variations within the GCC box
do not affect CsLMI1 expression or oil gland
development. ChIP-qPCR analysis further con-
firmed the binding of CsDRNL to the CsLMI1
promoter in vivo (Fig. 3H).
The CsDRNL-CsLMI1 regulatory module
activates CsMYC5
To uncover the mechanism underlying how
CsLMI1 orchestrates oil gland formation we
performed a transcriptome analysis of cr-cslmi1
mutant leaves to identify genes under the con-
trol of CsLMI1. This analysis revealed that
compared with the WT, 463 genes were sig-
nificantly down-regulated in the cr-cslmi1
mutants whereas only 30 genes were up-
regulated (fig. S15A and data S3). To identify
genes specifically related to oil gland devel-
opment, we included genes that were more
highly expressed in thorn bases compared to
thorn tips (data S2). This approach led us to
identify 185 genes as downstream candidates
of CsLMI1 (fig. S15B). Out of these genes, 13
encoded transcription factors, including CsLMI1
itself, suggesting that CsLMI1 has a positive
feedback for its own expression. Subsequent
CRISPR-Cas9–mediated knockout of other 12
genes individually pinpointed Carrizo citrange
CsMYC5—the closest homolog to Arabidopsis
MYC5 (26, 27)—as a pivotal gene regulating
cavity formation (Fig. 4A). In the cr-csmyc5
mutants, the number of secretory cavities ap-
peared normal except that they lacked ar-
ranged sheath cells and epithelial cells (Fig.
4B and fig. S16A). As such, these oil glands

Wang et al., Science 383, 659–666 (2024) 9 February 2024 4 of 8

RES EARCH | R E S E A R C H A R T I C L E
Fig. 3. DRNL directly activates the LMI1 promoter to promote oil gland initia-
tion. (A) qRT-PCR analysis of CsDRNL in thorn base and tip. Error bars are one SD
from three biological replicates. ***P < 0.001 (Student’s t test). (B) Leaf margin and
oil gland phenotypes of wild-type, cr-csdrnl mutants. Two independent cr-csdrnl
mutants were shown. (C) qRT-PCR analysis of CsLMI1 in cr-csdrnl mutants, p35S::
CsDRNL transgenic plants, and CsDRNL overexpression-inducible activity plants.
Young leaves (~1 cm long) were used. OX, overexpression; DNRL-GR OX, over-
expression of DRNL-GR fusion, the fusion protein activity can be induced by 20 mM
dexamethasone (DEX), 0.1% DMSO was used as MOCK. Error bars are one SD from
three biological replicates. *P < 0.05, ***P < 0.001 (Student’s t test). (D) The CsDRNL
promoter transcriptional reporter. A 5.8-kb upstream of the translational start codon
Wang et al., Science 383, 659–666 (2024) 9 February 2024
was used to drive the eGFP reporter. (E and F) CsDRNL and CsLMI1 in situ expression
in consecutive cross sections. Common sites and CsLMI1 unique signals are indicated
by orange and blue arrowheads, respectively. (G) EMSA assays, in which purifed His-
NusA-CsDRNL fusion protein and FAM labeled probes were used. dP1 and iP1 were
mutated probes, including deletion of the GCC box (orange color) or 1-bp insertion in
the GCC box, respectively. 2X and 5X indicates 2 and 5 times unlabeled competitor
probe, respectively. N, no protein control. (H) ChIP-qPCR assay in young Carrizo
citrange shoots with p35S::3xFLAG-CsDRNL-GR. DEX-treated young shoots and anti-
FLAG antibody were used for the ChIP assay. **P < 0.01, (Student’s t test). In (A),
(C) and (H), bars represent mean ± SD (n = 3). Scale bars, 0.5 cm in (B), 100 mm in (D)
and 50 mm in (E) and (F).
5 of 8
RES EARCH | R E S E A R C H A R T I C L E
Fig. 4. The DRNL-LMI1 pathway regulates oil gland formation through MYC5.
(A) CRISPR-Cas9–induced homozygous mutations in CsMYC5 (cr-csmyc5) of
Carrizo citrange. Arrowheads indicate the mutation site. aa, amino acids; +1 bp,
1 base pair insertion which caused frameshift and premature translation stop.
(B) Semi-thin cross section of the fully expanded leaves of WT and cr-csmyc5.
L, lumen; E, epithelial cell; S, sheath cell. (C) qRT-PCR analysis of CsMYC5
expression in young leaves of cr-csdrnl and cr-cslmi1 mutants. Error bars are one
SD from three biological replicates. ***P < 0.001 (Student’s t-test). (D) CsMYC5
in situ expression of cross section of a leaf axial. T, thorn; B, bract; S, stem.
(E and F) In situ expression of CsMYC5 and CsLMI1 in paired consecutive sections.
Wang et al., Science 383, 659–666 (2024) 9 February 2024
Orange arrows in (E) and (F) indicate common sites for CsLMI1 and CsMYC5
expression and blue arrowheads represent CsLMI1 unique expression sites. (G) A
regulatory pathway for Citrus oil gland development. Cell enlargement and cell
division for oil gland initials begin at the epidermis, with the expression of
CsDRNL as a marker. CsDRNL activates CsLMI1 by binding to the GCC box at its
promoter. CsLMI1 maintains its own expression during subsequent gland
development and triggers the expression of CsMYC5 at a later stage before cavity
formation. Eventually, CsMYC5 promotes the formation of cavities and the
expression of genes involved in essential oil biosynthesis. Stars mark the oil
gland cells. Scale bars, 0.1 cm in (A), 20 mm in (B) and 50 mm in (D) to (F).
6 of 8
RES EARCH | R E S E A R C H A R T I C L E
lacked EOs which indicates that CsMYC5 might
function at a later stage for specifying func-
tional oil glands. This observation was consis-
tent with qRT-PCR results that demonstrated
a substantial decrease in CsMYC5 expression
in cr-csdrnl and cr-cslmi1 mutants (Fig. 4C).
Further insights into CsMYC5 expression
were investigated through promoter-reporter
and RNA in situ hybridization assays, which
revealed its presence exclusively in the inner
cell layers around the lumen formation stage
in leaves, stems, and thorn bases (Fig. 4D and
fig. S16B). Notably, the expression of CsMYC5
overlapped with that of CsLMI1 at later stages
of oil gland development (Fig. 4, E and F).
These results suggest CsMYC5 as a downstream
gene within the CsDRNL-CsLMI1 pathway gov-
erning oil gland development.
Subsequent comparative transcriptomic analy-
sis identified 303 genes that were down-regulated
in cr-csmyc5 mutants (fig. S17A and data S4),
with 261 of these genes (86%) also being down-
regulated in cr-cslmi1 mutants. Among these
genes, half (131) exhibited higher expression in
thorn bases than in thorn tips, allowing them
to function as prime candidates for later-
stage oil gland development. This subset in-
cluded eight genes encoding transcription
factors from various families such as AP2/ERF
(CsADAP, CsERF114, CsERF13.2, CsERF13.3),
WRKY (CsWRKY75.1 and CsWRKY75.2), NAC
(ANAC042), and MADS (AGL8). We also un-
covered many genes encoding key enzymes
involved in potential roles within EOs bio-
synthesis in this group (fig. S17A). For in-
stance, nine genes encoding terpene synthases,
which catalyze the initial steps of terpene
biosynthesis, were identified. Additionally, six
O-methyltransferase family genes, with poten-
tial involvement in flavonoid O-methylation,
were present. Moreover, genes encoding en-
zymes related to lipid metabolism and lipid-
transfer proteins were identified. This prompted
us to validate the differential expression of these
structural genes involved in terpene and lipid
metabolism through qRT-PCR (fig. S17, B to G).
The results showed that the expression of these
genes is dependent on CsDRNL, CsLMI1, and
CsMYC5, thus uncovering regulatory genes and
indicating that oil gland formation is a prereq-
uisite for triggering the biosynthesis of EOs.
Discussion
In this study, we have uncovered a regulatory
framework for secretory cavity development
from initiation to cavity formation (Fig. 4G).
This marks a major step toward understanding
the molecular foundation behind the formation
of secretory cavities and the accumulation of
related secondary metabolites in terrestrial
plants. At the center of this pathway lies the
HD-ZIP I family transcription factor CsLMI1,
which has a role in both the facilitation of ser-
rated leaf formation and the promotion of oil
Wang et al., Science 383, 659–666 (2024) 9 February 2024
gland development in Citrus. Notably, MpC1HDZ,
the solitary HD-ZIP I family member in the
liverwort Marchantia polymorpha, also plays
a role in driving oil cell differentiation and the
subsequent accumulation of terpenoid com-
pounds within oil bodies (28). This function
aids in safeguarding plants against arthropod
herbivores. Moreover, a HD-ZIP I family gene
in cucumbers regulates glandular and non-
glandular trichome development but not ini-
tiation of such structures (29). These examples
suggest that the co-option of members of this
same family of transcription factors for differ-
ent aspects of secretory structure development
in widely divergent species may be surprising-
ly common (30).
The evolution of upstream enhancers has
emerged as a critical factor influencing the
expression and function of LMI1-like genes,
contributing substantially to leaf morphol-
ogy (19, 31–33). Our findings reveal that the
GCC box cis-regulatory element within LMI1
genes has expanded its role to drive expression
specifically for gland development in Rutaceae
species. Transcription activator CsDRNL binds
this GCC box element and activates CsLMI1 tran-
scription. This mechanism likely serves to rewire
an initial signal originating from the epi-
dermis and sub-epidermis of oil gland primor-
dia toward the internal structural patterning.
Notably, similar to CsLMI1, CsDRNL is also
implicated in controlling the development of
serrated leaves in Citrus. This suggests the
existence of similar regulatory mechanisms
involving cell proliferation or expansion in
both oil gland formation and leaf serration
production, and the latter process is known to
be regulated by the interplay of auxin and cyto-
kinin (34). Furthermore, DRNL's homolog is
indispensable in leaf initiation and promotes
the development of peltate trichomes in toma-
toes, which are multicellular secretory struc-
tures on the surface of the epidermis (10, 35).
This suggests that DRNL proteins possess the
capability to stimulate both inward and out-
ward gland formation in Rutaceae and Sola-
naceae, respectively. This versatility indicates
their pivotal role in shaping the diverse
array of secretory structures in these plant
families.
The CsDRNL-CsLMI1 pathway operates to
activate CsMYC5, promoting oil gland differ-
entiation and orchestrating the expression of
structural genes that participate in the bio-
synthesis of EOs, encompassing terpenoids,
flavonoids, and lipids, which serve as a robust
chemical defense against herbivores, such as
caterpillars. The regulation of diverse struc-
ture genes by CsMYC5 could be achieved through
the action of other transcription factors such
as CsERF13s. Notably, the MpERF13 gene is
required for oil body formation and terpenoid
biosynthesis in M. polymorpha (36). Overall,
CsMYC5 plays a role in promoting a secondary
metabolism similar to that of its closest cotton
homolog GoPGF, but its involvement in gland
development was restricted to later stages un-
like GoPGF, which controls the initiation of
glandular trichomes (37, 38). This observation
suggests that distinct developmental programs
are involved in governing secretory cavity for-
mation and glandular trichome initiation.
The CsDRNL-CsLMI1-CsMYC5 pathway dem-
onstrates its capacity to regulate oil gland
development not only in leaves but also in
fruits. This is supported by the detection of
their expression within developing oil glands
present in fruit peels (fig. S18). Thus, our in-
vestigations provide key insights on the regu-
latory pathway and mechanisms that underlie
secretory cavity development and the synthe-
sis of valuable EOs. This understanding lays
the foundation for potential applications in
synthetic biology, enabling the protection of
plants and the generation of germplasms for
industries involved in scent and flavoring.
REFERENCES AND NOTES
1. B. M. Lange, Annu. Rev. Plant Biol. 66, 139–159 (2015).
2. T. G. Knight, A. Klieber, M. Sedgley, Ann. Bot. 88, 1039–1047
(2001).
3. J. F. Richit, S. V. N. Díaz, L. F. P. Dick, J. E. A. Mariath,
Am. J. Bot. 110, e16248 (2023).
4. D. Ciccarelli, A. C. Andreucci, A. M. Pagni, Ann. Bot. 88,
637–644 (2001).
5. A. P. Fortuna-Perez et al., Perspect. Plant Ecol. Evol. Syst. 48,
125588 (2021).
6. N. Mahato et al., Crit. Rev. Food Sci. Nutr. 59, 611–625
(2017).
7. K. Fisher, C. Phillips, Trends Food Sci. Technol. 19, 156–164
(2008).
8. S. S. Voo, H. D. Grimes, B. M. Lange, Plant Physiol. 159, 81–94
(2012).
9. C. Chalvin, S. Drevensek, M. Dron, A. Bendahmane, A. Boualem,
Trends Plant Sci. 25, 477–487 (2020).
10. M. Wu et al., Dev. Cell 58, 278–288.e5 (2023).
11. T. Yan et al., New Phytol. 218, 567–578 (2018).
12. T. Yan et al., New Phytol. 213, 1145–1155 (2017).
13. C. Yang et al., Proc. Natl. Acad. Sci. U.S.A. 108, 11836–11841
(2011).
14. M. Bai et al., J. Exp. Bot. 71, 4812–4827 (2020).
15. F. Zhang et al., Curr. Biol. 30, 2951–2961.e5 (2020).
16. F. Zhang, C. LeBlanc, V. F. Irish, Y. Jacob, Plant Cell Rep. 36,
1883–1887 (2017).
17. Q. Ma et al., Front. Plant Sci. 12, 759968 (2021).
18. C. Zhu et al., Plant Biotechnol. J. 17, 2199–2210 (2019).
19. D. Vlad et al., Science 343, 780–783 (2014).
20. L. A. Saddic et al., Development 133, 1673–1682 (2006).
21. D. M. Eklund et al., Plant Physiol. 157, 2069–2080
(2011).
22. J. W. Chandler, W. Werr, J. Exp. Bot. 68, 3457–3472
(2017).
23. T. Kirch, R. Simon, M. Grünewald, W. Werr, Plant Cell 15,
694–705 (2003).
24. J. W. Chandler, M. Cole, A. Flier, B. Grewe, W. Werr,
Development 134, 1653–1662 (2007).
25. T. Aoyama, N. H. Chua, Plant J. 11, 605–612 (1997).
26. S. Song et al., Plant Cell Physiol. 58, 1752–1763 (2017).
27. P. Figueroa, J. Browse, Plant J. 81, 849–860 (2015).
28. F. Romani et al., Curr. Biol. 30, 2815–2828.e8 (2020).
29. X. Liu, E. Bartholomew, Y. Cai, H. Ren, Front. Plant Sci. 7, 1187
(2016).
30. F. Romani et al., J. Exp. Bot. 73, 4427–4439 (2022).
31. R. J. Andres et al., Proc. Natl. Acad. Sci. U.S.A. 114, E57–E66
(2017).
32. F. Vuolo et al., Genes Dev. 30, 2370–2375 (2016).
33. A. Sicard et al., Curr. Biol. 24, 1880–1886 (2014).
34. S. Navarro-Cartagena, J. L. Micol, Trends Plant Sci. 28, 54–73
(2023).
7 of 8
RES EARCH | R E S E A R C H A R T I C L E
35. Y. Capua, Y. Eshed, Proc. Natl. Acad. Sci. U.S.A. 114,
3246–3251 (2017).
36. T. Kanazawa et al., Nat. Commun. 11, 6152 (2020).
37. Z. N. Zhang et al., New Phytol. 241, 314–328 (2024).
38. D. Ma et al., Nat. Commun. 7, 10456 (2016).
39. H. Wang et al., F. Zhang, Molecular regulation of oil gland
development and biosynthesis of essential oils in Citrus spp., Dryad
(2024); https://doi.org/10.5061/dryad.4xgxd25h6.
ACKN OW LEDG MEN TS
We thank W. Liu and A. Saffer for critical reading of the manuscript. We
thank F. Yang, S.P. Yan, D. Jackson, and J. Murray for helpful
discussions. We thank members of the Irish lab at Yale University for
comments on the manuscript. We thank P.W. Wang and X.L. Qu for
the assistance for imaging. Funding: This work was funded by the
following: the National Natural Science Foundation of China (NSFC)
key program grant (32230095) to X.X.D.; the National Excellent
Young Scientists (overseas) Fund of NSFC to F.Z.; the Foundation of
Hubei Hongshan Laboratory to F.Z. (2021hszd016). Author
Wang et al., Science 383, 659–666 (2024) 9 February 2024
contributions: H.X.W., J.R., X.X.D., and F.Z. designed the research; described biological material can be obtained from F.Z. License
H.X.W. mapped the HP locus and conducted most of the research on information: Copyright © 2024 the authors, some rights reserved;
CsLMI1 and CsMYC5; J.R. established the Citrus thorn system for exclusive licensee American Association for the Advancement of
screening oil gland developmental regulators and conducted all Science. No claim to original US government works. https://www.
research on CsDRNL and the CsLMI1 promoter. S.Y.Z. performed sciencemag.org/about/science-licenses-journal-article-reuse
paired in situs and LMI1 in Citrus relatives. S.Y.Z., Z.Y.L., Q.Y.Z., and
S.X. generated other CRISPR mutants. Y.Y.D. and Q.Y.Z. contributed
to data analysis. C.W.C. coordinated the investigation of HP orchards SUPPLEMENTARY MATERIALS
in Rong An county. C.Q.Z., Z.Z.X., L.J.C., and J.L.Y. contributed new
science.org/doi/10.1126/science.adl2953
reagents. X.W. and Q.X. contributed to evolutionary analysis. X.X.D.,
Materials and Methods
W.W.G., and F.Z. supervised the project. H.X.W., J.R., X.X.D., and F.Z.
Figs. S1 to S18
wrote the first draft of the manuscript and all authors contributed
Table S1
revisions. Competing interests: The authors declare that they
References (40–57)
have no competing interests. Data and materials availability: The
MDAR Reproducibility Checklist
NGS data (RNA-seq, whole genome sequencing, and ATAC-seq)
Data S1 to S4
have been deposited to the National Genomics Data Center (NGDC,
https://ngdc.cncb.ac.cn) under Bioproject numbers: PRJCA020044,
PRJCA019979 and PRJCA019977 and also at Dryad (39). All other Submitted 10 October 2023; accepted 29 December 2023
data are available in the main text or supplementary materials. The 10.1126/science.adl2953
8 of 8
RES EARCH

PALEOCLIMATE mineral approach has been limited to non-iron-

oxide–bearing combinations [e.g., (15)] with rel-
Oxygen isotope ensemble reveals Earth’s seawater, atively poor sensitivities (<3‰ shift in D18Omin-min
between 0° and 100°C) that do not provide
temperature, and carbon cycle history satisfactory precision for historical application
(fig. S6).
Terry Isson* and Sofia Rauzi To reconstruct the history of seawater d18O
and temperature using the paired mineral ap-
Earth’s persistent habitability since the Archean remains poorly understood. Using an oxygen isotope proach, we assembled a comprehensive compila-
ensemble approach—comprising shale, iron oxide, carbonate, silica, and phosphate records—we tion of oxygen isotopes in marine sedimentary
reconcile a multibillion-year history of seawater d18O, temperature, and marine and terrestrial clay archives spanning the Archean to present
abundance. Our results reveal a rise in seawater d18O and a temperate Proterozoic climate distinct to (Fig. 1), along with corresponding temper-
interpretations of a hot early Earth, indicating a strongly buffered climate system. Precambrian ature-dependent mineral-water oxygen isotopic
sediments are enriched in marine authigenic clay, with prominent reductions occurring in concert with offsets (figs. S2 to S5) (25). Each record is
Paleozoic and Cenozoic cooling, the expansion of siliceous life, and the radiation of land plants. These sampled from the 70th to 97th percentile to
findings support the notion that shifts in the locus and extent of clay formation contributed to minimize any potential diagenetic influence
seawater 18O enrichment, clement early Earth conditions, major climate transitions, and climate classically viewed to shift sedimentary d18O
stability through the reverse weathering feedback. values more negative [e.g., (13, 26, 27)] and
eliminate positive outliers. Our results indicate
that stable Proterozoic seawater d18O is termi-

E
vidence of life and liquid water extends
across Earth’s multibillion-year history (1).
Yet the reasons for Earth’s persistent
habitability and conditions that ensued
after the widespread formation of oceans
and subaerial continents remain poorly con-
strained. A gradual decrease in the partial pres-
sure of carbon dioxide in the atmosphere (pCO2)
in step with solar brightening is viewed to have
been necessary to sustain above-freezing condi-
tions (2–5). While variation in solid Earth de-
gassing (6) and crustal reactivity (7, 8), the
silicate weathering feedback (9, 10), and the
reverse weathering feedback (11) have been
suggested to contribute to this long-term pCO2
trend, the efficacy of these factors and timing of
proposed transitions remain debated (8, 12).
Furthermore, there is a lack of consensus on
the history of surface temperatures [e.g.,
(13, 14)]. Together this has limited our under-
standing of the conditions under which life
originated and evolved. We make the case
that pairing the sedimentary oxygen isotope
(d18O) records of carbonate, silica, or phosphate
with that of iron oxide provides a powerful
means of simultaneously reconstructing the
evolution of temperature and the oxygen isotope
composition of seawater. By examining the shale
record against these results, a coeval history of
marine authigenic and terrestrial clay abun-
dance can be determined, providing insight
into global carbon cycle evolution.
Sedimentary records of oxygen isotopes in
silica, carbonate, and phosphate exhibit a secular
rise since the Archean (Fig. 1). This trend is
widely believed to reflect the combined history
of seawater d18O and temperature. However,
isolating the effects of each factor through the
coupling of these records has not been viable
Te Aka Mātuatua, University of Waikato (Tauranga), Bay of
Plenty, Tauranga, New Zealand.
*Corresponding author. Email: tisson@waikato.ac.nz
[e.g., (15)], resulting in a broad range of plau-
sible seawater d18O and temperature histories
often cast as extremes (fig. S1). One end-member
calls upon increasing seawater d18O, imply-
ing relatively invariant surface temperatures
throughout Earth history (14, 16–18). The other
end-member dictates that early oceans were
significantly hotter, implicitly suggesting that
seawater d18O has been well buffered at ~0 per
mil (‰) (13, 19, 20). A more nuanced approach
that allows for the coevolution of seawater d18O
and temperature is adopted by some investi-
gations (21–23). To distinguish between these
hypotheses, we demonstrate that the advent of
an iron oxide d18O record (24) enables the use of
independent mineral pairs to decipher Earth’s
climatic evolution.
Mineral pairs reconcile Earth’s temperature
and seawater d18O evolution
The history of seawater d18O and temperature
can be simultaneously acquired by coupling
expressions that describe the temperature
dependence of mineral-water oxygen isotopic
offsets (D18Omin-H2O) for two distinct archives
(25). In this paired mineral approach, coupling
phases with relatively more distinct D18Omin-H2O
temperature relationships generates a greater
sensitivity to changes in temperature and seawater
d18O. Whereas carbonate, silica, and phosphate
have similarly strong temperature sensitivi-
ties (25), iron oxides exhibit a distinctively weak
D18Omin-H2O temperature dependence under en-
vironmentally relevant conditions (Fig. 2A and
figs. S2 to S5). The inclusion of iron oxide in any
mineral pair therefore gives rise to a more pre-
cise system for reconstructing temperature and
parent fluid d18O. This is reflected by the strong
temperature-dependent mineral-mineral oxygen
isotopic offset (D18Omin-min) of the iron oxide–
carbonate, iron oxide–silica, and iron oxide–
phosphate pairs—the focus of this study—which
possess a >11‰ range between 0° and 100°C
(Fig. 2, B and C, and fig. S3). Thus far, the paired
nated by a ~7‰ rise to near modern values
(Fig. 3). Modest temperatures of ~2° to 18°C
(1s) characterize the Proterozoic, and maximum
early Paleozoic values of ~26° to 46°C (1s) are
succeeded by an overall decline toward the
modern, with notable cooling observed through
the Paleozoic and Cenozoic (Fig. 3). All three
mineral pairs generate similar evolutionary
trends, reconciling a unified history of seawater
oxygen isotope ratios and temperature from
four superficially distinct sedimentary d18O
records. This suggests that the unique trajec-
tories of each record may simply reflect dif-
ferences in the temperature sensitivities of
individual mineral-water isotopic offsets. A
greater net Phanerozoic d18O rise is expressed
in the more temperature–sensitive phases of
carbonate and silica as compared with the
less sensitive phosphate (Fig. 1 and figs. S7 and
S8). The smallest increase in d18O is observed
in iron oxides with a temperature-insensitive
D18Omin-H2O, predominantly reflecting seawater
d18O evolution.
Our paired mineral results diverge from sug-
gestions that seawater d18O has been well buf-
fered at near modern levels by reactions with
ocean crust [e.g., (28)] and interpretations of
a hot early ocean based chiefly on ancestral
protein reconstructions (29), altered ocean
crust d18O signatures (30), assessments of sili-
con and oxygen isotopes in marine cherts that
assume an invariant seawater d18O (13), the
paired use of phosphate-silica d18O (15), and bulk
carbonate clumped isotope thermometry [e.g.,
(25, 31)]. Instead, our results substantiate the
seawater d18O history proposed on the basis
of the iron oxide d18O record (24), clumped
isotope analysis of well-preserved fossil shells
(32), and key geologic water cycle models (14, 18)
which demonstrate that an invariant ophiolite
d18O record does not negate the existence of
more 18O-depleted early oceans (fig. S1). Further
refinement of Archean seawater d18O and tem-
peratures may improve our understanding of

Isson et al., Science 383, 666–670 (2024) 9 February 2024 1 of 5

RES EARCH | R E S E A R C H A R T I C L E

Fig. 1. Oxygen isotope records of marine sediments. (A to D) Silica,
carbonate, phosphate, and iron oxide d18O records, respectively (25). The
shaded region indicates the sampling window (70th to 97th percentile). Dashed
lines indicate a linear regression through the 97th percentile. (E) Marine
shale d18O (25). Shaded region is the sampling envelope (16th to 84th
percentile), and dashed lines are the linear regressions to the full dataset.
Solid line is the loess smoothed curve. In (A) to (E), the slopes of the linear
regressions are given in the top right corner. (F) Offsets between the d18O of the
shale record with that of carbonate, silica (chert), phosphate, and iron oxide
relative to the modern.

the timing of continental emergence (14, 23)
and the environment and conditions that sup-
ported the origin of life.
Several key features arise from the paired
mineral temperature and seawater d18O recon-
struction that require explanation (Fig. 3).
Earth’s surface temperature has been excep-
tionally well buffered within a narrow range
(between ~4° and 37°C) since the Proterozoic,
suggesting the persistence of robust climate-
stabilizing feedbacks throughout Earth history.
The Proterozoic displays a more stable climate
relative to the Phanerozoic, which is charac-
terized by a greater spread in temperature and
frequent icehouse-greenhouse transitions. The
early Paleozoic is the warmest interval of the
past 2 billion years, and Paleozoic and Ce-
nozoic cooling occurs in step with estimates of
CO2 decline (4, 33, 34). Finally, a considerable
increase in seawater d18O is observed between
the late Proterozoic and Paleozoic. Notably,
these key transitions in climate stability, tem-
perature, and seawater d18O take place in con-
cert with prominent deviations between the
marine shale d18O record and that of other
sedimentary archives—features that we dem-
onstrate are intimately linked.
Decoupling of the shale oxygen isotope record
from other sedimentary archives
The evolution of marine shale d18O stands out
from all other sedimentary archives (Fig. 1). A
modest Precambrian rise in shale d18O of
~1.9‰ per billion years (Gyr−1) is followed by
a net decrease through the Phanerozoic. A de-
cline in marine shale d18O during the Paleozoic
is succeeded by an increase in the Mesozoic
and then a decrease through the Cenozoic. This
deviates from the overall rising d18O trend ex-
hibited by silica, carbonate, and phosphate since
the Archean. The decoupling of the shale record
from that of the other sedimentary phases is
highlighted by a prominent ~7 to 16‰ increase
in the offset between shale d18O and that of
other archives through the Phanerozoic, with
notable transitions taking place during the
Paleozoic (~6 to 7‰) and Cenozoic (2 to 7‰)
(Fig. 1F). Any framework used to explain the
observed trends in marine shale d18O must
account for both a broad secular rise over time
and its decoupling from other sedimentary
archives through the Phanerozoic. Diagenetic
alteration, which is viewed to drive older and
increasingly altered archives toward more-
negative values, cannot account for the progres-
sively more 18O-depleted shale oxygen isotope
signatures through the Phanerozoic. This is
consistent with the view that a high water-to-
rock ratio is uncommon in mudrocks, which
are notoriously impermeable and therefore
resistant to postdepositional alteration (35).
Here, we demonstrate that variations in the
detrital-to-authigenic clay content of marine
shales provides a basis for its decoupling from
the other chiefly authigenic archives—carbonate,
silica, phosphate, and iron oxide.
A record of marine and terrestrial clay
abundance through time
We propose that the history of clay production
at Earth’s surface can be investigated using
the oxygen isotopic signature of marine shales.
Sources of clay to marine shales include terres-
trially derived weathering products (36) and
marine authigenic clay (37, 38), which possess
unique oxygen isotopic signatures [~16 ±
5.1‰ and ~31 ± 4.9‰ (2s), respectively, in
the modern system] owing to the distinct com-
position and temperature of terrestrial weather-
ing fluids and seawater (figs. S9 and S10). On this
basis, a simple mixing model can be used to quan-
tify the relative abundance of marine authigenic

Isson et al., Science 383, 666–670 (2024) 9 February 2024 2 of 5

RES EARCH | R E S E A R C H A R T I C L E

Fig. 2. Temperature-dependent mineral-water and mineral-mineral oxygen isotope offsets. Temperature-dependent mineral-water (A) and mineral-mineral
(B) oxygen isotope offsets, relative to corresponding values at 0°C. (C) Net change in mineral-mineral oxygen isotope offset between 0° and 100°C. Separate hematite
and goethite expressions were used for all calculations in this study; an average iron oxide curve is rendered here for simplicity.

and terrestrial clay in marine shales as follows
d18 Oclay  d18 Otc
fmac ¼
d18 Omac  d18 Otc
where fmac is the fraction of oxygen bound in
marine authigenic clay relative to the total clay
pool, d18Oclay represents the oxygen isotopic
signature of the total clay fraction, and d18Otc
and d18Omac are the oxygen isotopic signatures
of terrestrial clay and marine authigenic clay,
respectively. Initial terrestrial and marine authi-
genic clay d18O are sampled from the modern
distribution (36–38). The evolution of d18Oclay
is determined from the shale record, whereas
d18Otc and d18Omac are determined according
to the oxygen isotope and temperature histories
of seawater and meteoric water (25) (fig. S11).
This view expands on initial interpretations of
marine shale d18O, which suggest that terres-
trial weathering clay products are the only clay
source [e.g., (39, 40)].
To determine the evolution of terrestrial and
marine clay d18O, we used the temperature and
seawater d18O histories generated from the iron
oxide–bearing mineral pairs (Fig. 3, A and B).
Meteoric water d18O was determined on the
basis of both the temperature and the oxygen
isotopic composition of seawater (25). We found
that while modern to late Paleozoic marine
shales display d18O values that overlap signif-
icantly with that of terrestrial clay, shale d18O
gravitates toward that of marine authigenic
clay between the Carboniferous and Cambrian
(Fig. 3C). These results indicate that changes
in temperature and seawater d18O can only
partly account for marine shale d18O evolution
and that variations in shale d18O are difficult
to explain if terrestrial weathering products are
the only clay source.
Extracting the history of marine authigenic
and terrestrial clay abundance from the shale
d18O record (25) reveals drastic shifts in fmac
ð1Þ
(Fig. 3). An elevated Proterozoic marine authi-
genic clay fraction of ~0.35 to 0.6 is followed
by a prominent decline through the Paleozoic
toward modest values of <0.025. A decrease in
fmac is also observed through the Cenozoic
from between ~0.08 and 0.25 to <0.04. To
ensure a balanced perspective not limited to
our preferred temperature and seawater d18O
history based on mineral pairs, we additionally
reconstructed fmac using the end-member sce-
narios of constant seawater d18O and constant
temperature (25) and found that Paleozoic and
Cenozoic shifts in fmac are similarly reproduced
(fig. S12).
Notably, the timing of the transitions in fmac
are coincident with the divergence of the shale
d18O record from all other sedimentary archives,
indicating that this decoupling is driven by a
shift in the source of clay to marine shales
(Fig. 1F). Overall, these results indicate a ~1 to
2 order magnitude and ~2 to 50 times change
in the marine authigenic clay content of shales
during the Paleozoic and Cenozoic, respective-
ly, and that Precambrian sediments are ~10 to
120 times richer in marine authigenic clay than
are their modern counterparts. The possibility
of an Archean waterworld earlier in Earth history
would imply an additional shift in clay produc-
tion from a system chiefly dominated by marine
synthesis.
Biological innovations, clay production,
and climate on Earth
Our fmac reconstruction provides direct geo-
chemical evidence for a prominent shift in the
locus of clay mineral production at Earth’s
surface coincident with key climatic, seawater
d18O, evolutionary, and ecological transitions.
We attribute these shifts to variations in the
intensity of marine and terrestrial clay synthesis.
The Paleozoic fmac transition, coincident with evi-
dence for increased clay retention on land (41),
supports rising terrestrial clay production likely
driven by the expansion of rooted flora. All else
being equal, this would imply a decrease in
delivery of clay-forming ingredients to the
marine realm as dissolved constituents and thus
marine clay authigenesis and fmac. Evolutionary
shifts in fmac also correspond to proposed
intervals of declining marine clay authigenesis
associated with the Paleozoic rise of silicifying
sponges and radiolarians, and Cenozoic pro-
liferation of diatoms (11). In this framework,
the expansion of siliceous life (42, 43) reduced
marine clay authigenesis by lowering dissolved
silica levels (11) and transitioning silica deposi-
tion toward deeper environments less rich in
other key clay-forming ingredients such as
aluminum (44), cation release from marine
weathering (45), and alkalinity generated
from anaerobic respiration (11). Consequently,
changes in marine authigenic clay formation
likely contributed to both the Paleozoic and
Cenozoic transitions in fmac. If correct, these
shifts in marine clay authigenesis imply a fun-
damental restructuring of the globally coupled
carbon-silica cycle that regulates long-term
climate on Earth.
Carbon introduced to the ocean-atmosphere
reservoir by solid Earth degassing is ultimately
sequestered through the weathering of silicate
minerals when coupled to the formation of
carbonate and silica (8, 9, 46). The efficiency of
this carbon sequestration process is governed
by the formation of marine authigenic clay,
which consumes alkalinity and reliberates CO2
captured through the initial weathering process

Isson et al., Science 383, 666–670 (2024) 9 February 2024 3 of 5

RES EARCH | R E S E A R C H A R T I C L E

back into the atmosphere (8, 47, 48). Simply,

marine clay authigenesis, commonly referred
to as reverse weathering, leads to the retention
of CO2 within the ocean-atmosphere system
(fig. S13). The balance of weathering and reverse
weathering therefore plays a direct role in
regulating atmospheric CO2. As marine clay
authigenesis is sensitive to pH, this process is
proposed to stabilize Earth’s climate through
the reverse weathering feedback (11, 49)—
elevating atmospheric pCO2 lowers marine pH
and reduces reverse weathering CO2 release,
whereas a decrease in pCO2 increases marine
pH and fosters more-vigorous authigenic clay–
driven carbon recycling. Changes in the inten-
sity of reverse weathering therefore provide a
mechanistic link between the deviations in
marine shale d18O from other sedimentary
archives coincident with transitions in temper-
ature and seawater d18O.
To quantify the effect of variation in reverse
weathering on atmospheric pCO2 evolution,
we adopted a standard global carbon-silica
cycle model (50). The modern climate system
was used as a baseline, and changes in silica
and alkalinity consumption by means of marine
clay authigenesis were prescribed (25). An
alkalinity-to-silica (Alk:Si) uptake ratio was
sampled between 0.3 and 4, on the basis of
clay mineral species commonly expressed in
sedimentary archives (25). We find that atmos-
pheric pCO2 levels proposed for the Eocene (51)
and Proterozoic (3, 5) can be sustained by
modest changes in reverse weathering well
within the range supported by the fmac record
(Fig. 3E).
These results substantiate the view that
enhanced marine authigenic clay formation
contributed to the elevated atmospheric pCO2
levels necessary to maintain a habitable cli-
mate, providing a viable solution to the faint
young Sun paradox (11). Further, this framework
intimately links the evolution of clay mineral
production on Earth to key climatic and
biotic transitions, supporting the notion that
the rise of siliceous organisms contributed to a
Precambrian-Phanerozoic shift in climate stab-
ility and to Paleozoic and Cenozoic cooling.
This explanation for intervals of Phanerozoic
cooling is distinct from postulations of changes
in Earth’s surface reactivity (7, 52), tectonic de-
gassing (53, 54), continental configuration, uplift, Fig. 3. A history of seawater d18O, temperature, and shale marine authigenic clay content based on
and crustal composition (55, 56). the paired mineral approach. (A and B) Seawater d18O and temperature reconstructions based on 104 model
Finally, a Precambrian-to-Phanerozoic decline simulations using the iron oxide–phosphate, iron oxide–silica, and iron oxide–carbonate pairings. Error bars (2s)
in CO2 release from clay mineral production reflect sampling from the 70th to 97th percentiles of the carbonate, silica, and phosphate records. Dot-dash
likely also contributed to rising seawater d18O. lines are loess smoothed curves. (C and D) Results from 104 model simulations. (C) Marine authigenic clay,
All else being equal, a reduction in clay syn- terrestrial clay, and shale (clay fraction) d18O. Dashed lines indicate the mean, while the shaded regions
thesis would necessitate less silicate weathering represent the middle 68% of the data. (D) fmac results with the frequency in color contours. (E) Results of
to balance solid Earth CO2 degassing. This im- 104 successful carbon-silica cycle simulations. Histograms represent the relative frequency of simulations
plies a diminished intensity of low-temperature that reproduce atmospheric pCO2 and seawater pH estimates for the Eocene (blue bars) based on boron
crustal alteration that preferentially depletes isotopes (51) and the Proterozoic (red bars) required to sustain above-freezing conditions (3, 5). Atmospheric
seawater of 18O. pCO2 is represented as times preindustrial (×PIAL). The change in reverse weathering required to reproduce
When interpreted collectively, the sedimen- the atmospheric pCO2 and seawater pH estimates is normalized to the modern value (×modern). Dashed
tary oxygen isotope records suggest that the vertical lines represent preindustrial values.

Isson et al., Science 383, 666–670 (2024) 9 February 2024 4 of 5

RES EARCH | R E S E A R C H A R T I C L E
evolution and rise of complex life over Earth’s
multibillion-year history took place under a
stable and clement climate and that early
oceans were relatively more deplete in 18O. A
late-Precambrian shift toward a more variable
Phanerozoic climate supports the idea that
silica and marine authigenic clay–rich early
oceans primed the Earth system to dampen
climate forcings through a strengthened reverse
weathering feedback (11). Our analysis also
reveals a contemporaneous decline in tem-
perature and the marine authigenic clay content
of shales coincident with the rise of silicifying
organisms and land plants. This provides the
first direct geochemical evidence for an inti-
mate link between clay formation, seawater
d18O, and global climate. Overall, these results
provide an impetus to deepen our understand-
ing of how life both responds to and shapes cli-
mate on Earth.
RE FE RENCES AND N OT ES
1. S. J. Mojzsis, T. M. Harrison, R. T. Pidgeon, Nature 409,
178–181 (2001).
2. D. O. Gough, Sol. Phys. 74, 21–34 (1981).
3. J. F. Kasting, Precambrian Res. 34, 205–229 (1987).
4. J. F. Kasting, Science 259, 920–926 (1993).
5. I. Halevy, A. Bachan, Science 355, 1069–1071 (2017).
6. E. Tajika, T. Matsui, Earth Planet. Sci. Lett. 113, 251–266
(1992).
7. R. A. Berner, Philos. Trans. R. Soc. London Ser. B 353, 75–82
(1998).
8. L. R. Kump, S. L. Brantley, M. A. Arthur, Annu. Rev. Earth
Planet. Sci. 28, 611–667 (2000).
9. J. C. Walker, P. Hays, J. F. Kasting, J. Geophys. Res. 86,
9776–9782 (1981).
10. R. A. Berner, A. C. Lasaga, R. M. Garrels, Am. J. Sci. 283,
641–683 (1983).
11. T. T. Isson, N. J. Planavsky, Nature 560, 471–475 (2018).
12. C. Keller, B. Wood, Nature 364, 223–225 (1993).
13. F. Robert, M. Chaussidon, Nature 443, 969–972 (2006).
Isson et al., Science 383, 666–670 (2024) 9 February 2024
14. J. B. Jaffrés, G. A. Shields, K. Wallmann, Earth Sci. Rev. 83,
83–122 (2007).
15. J. Karhu, S. Epstein, Geochim. Cosmochim. Acta 50, 1745–1756
(1986).
16. J. Veizer et al., Chem. Geol. 161, 59–88 (1999).
17. E. C. Perry Jr., Earth Planet. Sci. Lett. 3, 62–66 (1967).
18. J. F. Kasting et al., Earth Planet. Sci. Lett. 252, 82–93 (2006).
19. L. P. Knauth, S. Epstein, Geochim. Cosmochim. Acta 40,
1095–1108 (1976).
20. L. P. Knauth, D. R. Lowe, Geol. Soc. Am. Bull. 115, 566–580
(2003).
21. D. Herwartz, A. Pack, T. J. Nagel, Proc. Natl. Acad. Sci. U.S.A.
118, e2023617118 (2021).
22. H. C. Olson, N. Drabon, D. T. Johnston, Earth Planet. Sci. Lett.
591, 117603 (2022).
23. B. W. Johnson, B. A. Wing, Nat. Geosci. 13, 243–248 (2020).
24. N. Galili et al., Science 365, 469–473 (2019).
25. See supplementary materials.
26. L. S. Land, Geochim. Cosmochim. Acta 59, 2843–2844 (1995).
27. E. T. Degens, S. Epstein, Am. Assoc. Pet. Geol. Bull. 46,
534–542 (1962).
28. K. Muehlenbachs, R. Clayton, J. Geophys. Res. 81, 4365–4369
(1976).
29. E. A. Gaucher, S. Govindarajan, O. K. Ganesh, Nature 451,
704–707 (2008).
30. C. Holmden, K. Muehlenbachs, Science 259, 1733–1736 (1993).
31. K. D. Bergmann, S. A. Al Balushi, T. J. Mackey, J. P. Grotzinger,
J. M. Eiler, J. Sediment. Res. 88, 960–979 (2018).
32. G. A. Henkes et al., Earth Planet. Sci. Lett. 490, 40–50 (2018).
33. D. L. Royer, Geochim. Cosmochim. Acta 70, 5665–5675 (2006).
34. G. L. Foster, D. L. Royer, D. J. Lunt, Nat. Commun. 8, 14845
(2017).
35. L. S. Land, F. L. Lynch, Geochim. Cosmochim. Acta 60,
3347–3352 (1996).
36. I. N. Bindeman, G. Bayon, J. Palandri, Earth Planet. Sci. Lett.
528, 115851 (2019).
37. S. M. Savin, S. Epstein, Geochim. Cosmochim. Acta 34, 43–63
(1970).
38. T. Cole, Geochim. Cosmochim. Acta 49, 221–235 (1985).
39. I. Bindeman, A. Bekker, D. Zakharov, Earth Planet. Sci. Lett.
437, 101–113 (2016).
40. I. N. Bindeman et al., Nature 557, 545–548 (2018).
41. W. J. McMahon, N. S. Davies, Science 359, 1022–1024 (2018).
42. R. Siever, Geochim. Cosmochim. Acta 56, 3265–3272 (1992).
43. P. Cermeño, P. G. Falkowski, O. E. Romero, M. F. Schaller,
S. M. Vallina, Proc. Natl. Acad. Sci. U.S.A. 112, 4239–4244
(2015).
44. J. E. Mackin, R. C. Aller, Cont. Shelf Res. 6, 245–262 (1986).
45. S. Geilert et al., Nat. Commun. 14, 1676 (2023).
46. H. C. Urey, Proc. Natl. Acad. Sci. U.S.A. 38, 351–363
(1952).
47. T. T. Isson et al., Global Biogeochem. Cycles 34,
e2018GB006061 (2020).
48. F. T. Mackenzie, R. M. Garrels, J. Sediment. Petrol. 36,
1075–1084 (1966).
49. L. G. Sillén, Science 156, 1189–1197 (1967).
50. T. T. Isson et al., Nat. Commun. 13, 3509
(2022).
51. J. W. Rae et al., Annu. Rev. Earth Planet. Sci. 49, 609–641
(2021).
52. J. K. Caves Rugenstein, D. E. Ibarra, F. von Blanckenburg,
Nature 571, 99–102 (2019).
53. T. D. Herbert et al., Science 377, 116–119 (2022).
54. R. D. Müller et al., Nature 605, 629–639 (2022).
55. G. Li, H. Elderfield, Geochim. Cosmochim. Acta 103, 11–25
(2013).
56. Y. Goddéris et al., Nat. Geosci. 10, 382–386 (2017).
AC KNOWLED GME NTS
Funding: This work was supported by the Marsden Fund (Royal
Society of New Zealand) under contract MFP-UOW2010 (T.I.) and
by a Rutherford Discovery Fellowship (Royal Society of New
Zealand) (T.I.). Author contributions: T.I. conceived of the
ideas, designed the research, and carried out the modeling work.
T.I. and S.R. were responsible for development, visualization,
revision, and literature compilation. T.I. and S.R. wrote the
manuscript. Competing interests: The authors declare that they
have no competing interests. Data and materials availability: All
data are available in the manuscript or the supplementary
materials. License information: Copyright © 2024 the authors,
some rights reserved; exclusive licensee American Association for
the Advancement of Science. No claim to original US government
works. https://www.science.org/about/science-licenses-journal-
article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adg1366
Supplementary Text
Figs. S1 to S23
Tables S1 to S13
References (57–326)
Submitted 4 December 2022; accepted 5 January 2024
10.1126/science.adg1366
5 of 5
 WORKING LIFE
By Yaowu Zhang

The gift of feedback

T
he comments from reviewers on my first submission to an international journal were brutally
honest. Grammar, punctuation, word choice—all fell short of academic standards. Many phrases
were awkward or unclear. The writing was riddled with redundancies. “This paper … doesn’t meet
the criteria of academic writing,” one reviewer wrote. I knew my writing wasn’t perfect. I grew up
in rural China, and English didn’t come naturally to me. But the words still stung, causing me to
question whether my aspiration to become a clinician-scientist was out of step with my abilities.
Only later did I learn a lesson that helped me on my writing journey: Academic English is a second lan-
guage for everyone, even native speakers—and getting feedback is part of the learning process.

I began to study English when I was
10 years old. I learned what I needed
to pass the tests, but the subject felt
remote. Amid the serene landscapes
of rural China, the rhythm of life
harmonized more with the fields
than with the flicker of computer
screens.
Only later, during medical school
in Beijing, did I realize that English
would play a pivotal role in my future.
I entered the program because of my
desire to help patients as a clinician.
But in time a passion for research
emerged, inspiring me to embark on
a research project.
It soon became clear that many
landmark studies in my field were
published in English. If I wanted
“We are all perpetual learners,
to succeed as a researcher, I would forever refining our craft.”
need to publish in international
journals as well, which felt like a formidable challenge.
When the time came to write up my findings, I spent days
sifting through articles, checking English-Chinese dictionar-
ies, and organizing my thoughts into a coherent structure.
Sometimes I would stare at the screen, feeling overwhelmed
by the complexity of English writing. After countless hours of
writing and revising, I finally had a draft.
Some of my colleagues were co-authors, but I didn’t seek
feedback from them before submitting to a journal. After
pouring so much time into my writing, I felt the words were
my children; I didn’t want others to disrupt them. The reality
was also that my colleagues weren’t strong English writers
either, and we weren’t accustomed to commenting on one
another’s writing.
So, the first real feedback I received was from the anony-
mous peer reviewers. I took their critiques personally and felt
disheartened that, despite my best efforts, I clearly needed to
do more work on my writing.
I tried to solve the problem mostly
by myself, meticulously dissecting
journal articles to learn how they
were put together. That helped, but it
wasn’t until after I had finished medi-
cal school and moved on to a post-
doc in the United States that my real
breakthrough occurred.
During our first lab meeting, the
director of the lab, a native English
speaker with a prolific publication
record, candidly revealed his early-
career struggles with publishing. Jour-
nal reviewers had even criticized his
writing. His openness helped me see
that academic English is a challenge
for many scientists, not just me, and
it freed me to share my own struggles.
The director reassured me that I
wasn’t alone in this journey, saying,
“Yaowu, remember, we are a team; we
struggle and succeed together.” I quickly learned that the re-
search group had a culture of giving and receiving feedback.
And in the months that followed, my perspective shifted from
seeing feedback as a form of criticism to seeing it as a valu-
able source of professional growth.
Since then, I have actively sought out comments on manu-
scripts by sending them to co-authors, as well as to peers in
my department and elsewhere. I have also let go of feeling
that all of the words have to be solely my own. Science is col-
laborative, and academic writing can be a group effort, too.
So, to my fellow second-language English writers, I offer
this: Do not let the fear of language barriers hold you back.
Seek diverse perspectives, embrace feedback, and remember
that, in the realm of academic writing, we are all perpetual
learners, forever refining our craft. j
Yaowu Zhang is a neurosurgeon at Beijing Tiantan Hospital, Capital
Medical University. Send your career story to SciCareerEditor@aaas.org.

674 9 FEBRUARY 2024 • VOL 383 ISSUE 6683 science.org SCIENCE