Professional Documents
Culture Documents
$15
5 APRIL 2024
science.org
AT 50
An ancestor of us
all no longer reigns
alone p. 20
Hear the stories behind
the latest news and research
from Science.
33
INSIGHTS
13 Rare wooden artifacts show the
smarts of early Neanderthals POLICY FORUM
Complex tools from 300,000-year-old deposit
at Schöningen in Germany point to a “wood 36 Regulating advanced artificial
age” By A. Curry LETTERS
agents
Governance frameworks should
14 Utah flouts FDA with new placental 26 NextGen Voices: Research address the prospect of AI systems
stem cell law beneficiaries speak that cannot be safely tested
Measure declares that patients can be given By M. K. Cohen et al.
unapproved treatments By M. Wadman PERSPECTIVES
BOOKS ET AL.
PHOTO: QIAN WEIZHONG/VCG VIA GETTY IMAGES
A cash prize of up to USD 15,000 will be awarded to essay winners, and their
engaging essays will be published in Science. Winners will also be invited to
share their work and forward-looking perspective with leading scientists in
their respective fields at an award ceremony.
RESEARCH
74 Textiles
Single body-coupled fiber enables chipless
textile electronics W. Yang et al.
PERSPECTIVE p. 29
IN BRIEF
81 Thermoelectrics
43 From Science and other journals Pseudo-nanostructure and trapped-
hole release induce high thermoelectric
RESEARCH ARTICLES performance in PbTe B. Jia et al.
46 Innate immunity
Apoptotic cell identity induces distinct
87 Farming practices
Joint environmental and social benefits from
functional responses to IL-4 in efferocytic
diversified agriculture L. V. Rasmussen et al.
macrophages I. Liebold et al.
RESEARCH ARTICLE SUMMARY; FOR FULL TEXT:
DOI.ORG/10.1126/SCIENCE.ABO7027
93 Antimicrobials
Antibacterial activity of nonantibiotics is
47 Bacteriophage orthogonal to standard antibiotics
Removal of Pseudomonas type IV pili by M. Noto Guillen et al.
a small RNA virus J. Thongchol et al.
100 Microbiology
RESEARCH ARTICLE SUMMARY; FOR FULL TEXT:
DOI.ORG/10.1126/SCIENCE.ADL0635
Prophage terminase with tRNase activity
sensitizes Salmonella enterica to oxidative
stress S. Uppalapati et al.
48
48 Quantum simulation
The dynamics of magnetization in spin chains
Dynamics of magnetization at infinite
106 Protein design simulated on a superconducting quantum processor
temperature in a Heisenberg spin chain
De novo design of drug-binding proteins with
E. Rosenberg et al.
predictable binding energy and specificity
L. Lu et al. DEPARTMENTS
53 Cellular neuroscience
Lifelong persistence of nuclear RNAs in the 9 Editorial
113 Organic chemistry
mouse brain S. Zocher et al. Don’t bury Mexico’s biodiversity capacity
Carbon quaternization of redox active esters
By R. A. Medellin and J. Soberón
CREDITS: (LEFT TO RIGHT) SHUANG WU RESEARCH GROUP/FUJIAN AGRICULTURE AND FORESTRY UNIVERSITY; GOOGLE QUANTUM AI, DESIGNED BY SAYO-ART LLC
PERSPECTIVE p. 31
and olefins by decarboxylative coupling
X. Gan et al.
60 Perovskites 41 Prize Essay
Molecularly thin, two-dimensional all-organic 119 Cell biology Planting a chemical flag on antigens
perovskites H. S. Choi et al. By A. M. Kunjapur
Sister chromatid cohesion establishment
during DNA replication termination
66 Asthma G. Cameron et al.
134 Working Life
Bronchoconstriction damages airway Beyond words By D. Meuthen
epithelia by crowding-induced excess cell 124 Plant science
extrusion D. C. Bagley et al. HD-Zip proteins modify floral structures for
PERSPECTIVE p. 30 self-pollination in tomato M. Wu et al. ON THE COVER
Fifty years ago in Ethiopia, paleoanthro-
pologists unearthed the 3.2-million-year-old
skeleton known as “Lucy” and transformed
124 our views of humanity’s origins. This
reconstruction was created by rebuild-
ing Lucy’s body, muscle by muscle, over a
cast of her skeleton and bones from other
members of her spe-
cies, Australopithecus
afarensis. Today, Lucy
faces competition for the
role of our direct ances-
tor but remains the best
candidate. See page 20.
Credit: Reconstruction and
photo by John Gurche
SCIENCE (ISSN 0036-8075) is published weekly on Friday, except last week in December, by the American Association for the Advancement of Science, 1200 New York Avenue, NW, Washington, DC 20005. Periodicals mail
postage (publication No. 484460) paid at Washington, DC, and additional mailing offices. Copyright © 2024 by the American Association for the Advancement of Science. The title SCIENCE is a registered trademark of the AAAS. Domestic
individual membership, including subscription (12 months): $165 ($74 allocated to subscription). Domestic institutional subscription (51 issues): $2627; Foreign postage extra: Air assist delivery: $107. First class, airmail, student, and
emeritus rates on request. Canadian rates with GST available upon request, GST #125488122. Publications Mail Agreement Number 1069624. Printed in the U.S.A.
Change of address: Allow 4 weeks, giving old and new addresses and 8-digit account number. Postmaster: Send change of address to AAAS, P.O. Box 96178, Washington, DC 20090–6178. Single-copy sales: $15 each plus shipping and
handling available from backissues.science.org; bulk rate on request. Authorization to reproduce material for internal or personal use under circumstances not falling within the fair use provisions of the Copyright Act can be obtained
through the Copyright Clearance Center (CCC), www.copyright.com. The identification code for Science is 0036-8075. Science is indexed in the Reader’s Guide to Periodical Literature and in several specialized indexes.
Categories
The award is available in four categories:
• Cell and Molecular Biology
• Genomics, Proteomics and Systems Biology Approaches
• Ecology and Environment
• Molecular Medicine
More information
More information: scienceprize.scilifelab.se
Questions: scilifelabprize@aaas.org
Application deadline: July 15, 2024
Eligibility: applicants must have received their PhD between January 1, 2022
and December 31, 2023.
Enabled by the generous support of the Knut and Alice Wallenberg Foundation.
EDITO RIAL
I
n a world where biodiversity is on the line on repository but a model for best practices in biodiver- Rodrigo A. Medellin
many fronts—from armed conflict to pandemics sity management. All information is publicly available is a professor at
to climate change—defending institutions that online, and contributors decide when the information the Institute of
have effectively managed it is paramount. In the they provide will be made public. In the past 5 years, Ecology, Universidad
global effort to protect biodiversity, Mexico has CONABIO had an average of 1000 users per week, and Nacional Autónoma
been at the forefront. In particular, for more than it was consulted at least once per day by health, agri- de México, Mexico
30 years, Mexico’s National Commission for the culture, environment, foreign affairs, and government City, Mexico.
Knowledge and Use of Biodiversity (CONABIO) has agencies from Mexico and other countries. CONABIO
medellin@ecologia.
promoted research, compiled information on the has become a world-sought reference on how to
unam.mx
biodiversity of Mexico and elsewhere, and connected effectively compile useful information and incorporate
academia, government, and society to guide decision- it into public policy for the benefit of the population
Jorge Soberón
making. Unfortunately, the demise of CONABIO, and biodiversity.
which began in 2018 under the current administra- The proposed change for CONABIO will likely is director of the
tion, may be fully realized soon. Last month, the Mexi- eliminate the support it provides for the sustainable Biodiversity Institute
can government announced its intent management, use, and conservation of the University of
to reduce CONABIO from a multi- of biodiversity for Mexico and the Kansas, Lawrence,
ministry federal government agency world. Given that Mexico is home to KS, USA, and former
to a branch within the environment
ministry. This will strip CONABIO of
“Burying 10 to 12% of the world’s species, there
is much at risk. CONABIO is a high-
Executive Secretary
of CONABIO.
its independent voice, credibility, and
influence on national and interna-
the agency is level government agency composed
of 10 cabinet ministers and has op-
jsoberon@ku.edu
10.1126/science.adp5399
ASTRONOMY
T
he 3200-megapixel digital camera that will serve search for signs of dark energy and matter, looking
as the heart of the Vera C. Rubin Observatory for their effects by measuring distance with the help
is complete and will head to Chile for integra- of pulsating stars. It will gauge minute distortions in
tion, researchers announced this week. Built at distant objects created when light bends around inter-
the SLAC National Accelerator Laboratory in vening matter, which offer clues to how dark energy is
among people who were unvaccinated New England Journal of Medicine, found
Limits to Paxlovid benefits against SARS-CoV-2 and had at least one that in these groups, Paxlovid was no
B I O M E D I C I N E | Pfizer’s antiviral risk factor for severe COVID-19, such better than placebo at preventing deaths
Paxlovid (nirmatrelvir/ritonavir) may as diabetes or obesity. But results from and hospitalizations, or at reducing time
not help as many people as hoped. The a Pfizer-run trial of nearly 1300 people to symptom clearance. The findings
combo drug was granted emergency suggest the benefit doesn’t extend to support the use of the drug “only for
use authorization in 2021 in the United those who are unvaccinated and have persons who are at high risk for disease
States after a clinical trial showed it no risk factors or are vaccinated with at progression,” write infectious disease
reduced hospitalizations and deaths least one. The study, out this week in The specialists Rajesh Gandhi and Martin
Suit targets Florida hiring ban Why Europe’s battlefield bones are missing
| A Florida law that
I M M I G R AT I O N European wars of the 18th and early 19th centuries were marked by cavalry charges and mas-
requires its public universities to receive sive exchanges of cannon and rifle fire that could claim tens of thousands of lives within hours.
a waiver before employing a graduate stu- Yet archaeologists often struggle to find skeletal remains from these Napoleonic era battles. In
dent or postdoc from China, Iran, or any the book Bones of Contention: The Industrial Exploitation of Human Bones in the Modern Age,
of five other “countries of concern” is fac- published in February, a team of historians and archaeologists argues that industrial needs for
ing a challenge in a federal court. Enacted phosphates for fertilizer and bone char for beet sugar processing in the early 19th century led
in March 2023, the law violates the U.S. Europeans at the time to raid bones from mass graves. Co-editor Bernard Wilkin, a historian at
Constitution, is discriminatory, and the State Archives of Belgium, talked to Science about the grisly trade.
ignores the federal government’s author-
ity to set immigration and employment Q: Why would people resort to we know they ran out of accessible bones
practices, according to the suit filed on robbing battlefields? in Europe, and turned to colonies abroad.
25 March in a U.S. District Court in Miami A: In the 1830s, bones were suddenly worth The French dug up cemeteries in Algeria
by two science graduate students at Florida a lot of money. Battlefield bones in particular and shipped the bones to sugar factories
International University (FIU)—Zhipeng were a commodity that was easy to find, in Marseille; we know the British imported
Yin and Zhen Guo—and agricultural easy to access, and no one really cared about. mummies and bones from Egypt on an
economist Zhengfei Guan of the University A lot of farmers living near these battlefields industrial scale, destroying untold heritage
of Florida (UF). The students were forced realized there was gold in the ground. I in the process.
to pay their own way after FIU officials wouldn’t say these were grave-robbing
deferred graduate assistant positions with monsters. For the most part they were poor Q: What’s next?
stipends and tuition waivers that were farmers, who seized an opportunity. A: We need more evidence from elsewhere
to start in December 2023. Guan says his in Europe to understand if this was as
top candidate for a postdoc position went Q: Did this trade target other widespread as we think. And we need to pay
elsewhere after learning about the Florida burial grounds, too? attention to what was happening outside
law, which Guan says has “essentially made A: This goes way beyond battlefields. We of Europe. The colonies were exploited,
it impossible to hire anyone from China.” know there were medieval cemeteries in too, and the way white people treated the
Last week, UF students rallied in support Scotland and England that were emptied dead in European colonies needs a lot more
of the plaintiffs outside a meeting of the out and sold; measures to ban the practice research. I’ve been a historian for 20 years,
state’s Board of Governors, whose mem- were proposed, debated—and defeated— and I have the feeling this topic is the most
bers are among the defendants. in the British Parliament. At some point important thing I will work on in my career.
IN DEP TH
INFECTIOUS DISEASES
By Jon Cohen years, a variant known as clade 2.3.4.4b has receptors on human cells are different from
spread globally, sickening not just birds, but those in birds, and not a good match for avian
A
n H5N1 avian influenza strain that also dozens of mammalian species, including flu viruses.
has devastated birds and some mam- cats, dogs, tigers, bears, seals, and dolphins. But humans have the avian version of the
mals around the world has pulled off Last month, goats contracted the virus at a receptor in their eyes, which explains why
another big surprise. The virus has Minnesota farm that also had an infected they can develop conjunctivitis, which is
caused outbreaks among cows at U.S. poultry flock. typically a mild disease. Hundreds of people
dairy farms in at least four states, the Now, USDA says the virus has infected developed eye infections during an outbreak
first of which were revealed on 25 March. cattle at dairy farms in Texas, Kansas, New of H7N7, another highly pathogenic avian
Now, a farm worker has also become infected, Mexico, and Michigan, while Idaho has a influenza virus, in Dutch poultry flocks in
most likely from contact with sick cattle. The “presumptive” outbreak at one farm. (The 2003. There was some evidence of human-
worker developed conjunctivitis, a mild eye agency noted that cats on farms have also be- to-human transmission during that episode,
infection that frequently occurs when avian come infected.) Sick cows have a mild illness, and a veterinarian died.
influenza viruses jump into humans, Texas USDA says, and produce less milk, which is One urgent question now is how to protect
health officials announced on 1 April. thicker than usual. The infections are not en- other workers at dairy farms, who include
“It’s definitely taken me by surprise, but tirely unexpected: Other types of influenza vi- many immigrants. “We need to be advocat-
perhaps it shouldn’t have,” given the virus’ es- ruses readily infect cows, and an experiment ing for our employees on these farms, provid-
capades so far, says virologist Richard Webby published in 2006 demonstrated that a dif- ing them support and education in whatever
of St. Jude Children’s Research Hospital. ferent H5N1 variant could infect calves. language that it needs to be in,” says Joe
The U.S. Department of Agriculture The infected person in Texas is being Armstrong, a bovine veterinarian who works
(USDA) says, “Initial testing has not found treated with oseltamivir, an antiviral drug, with the University of Minnesota (UM) Ex-
PHOTO: ERIC THAYER/BLOOMBERG VIA GETTY IMAGES
changes to the virus that would make it and USDA stressed that the “current risk tension. Dairy workers rarely wear protective
more transmissible to humans.” Still, the to the public remains low.” Contamination gear, such as masks and goggles, Armstrong
widespread occurrence of H5N1 in mam- of commercial milk is of “no concern,” the notes, and floors in milking facilities are of-
mals has renewed worries that it may evolve agency said, because pasteurization reliably ten cleaned using high-pressure water spray-
to transmit more readily between people. kills viruses, and milk from sick cows is not ers, which could aerosolize the virus.
And scientists are urgently trying to answer sold. The U.S. Centers for Disease Control “I think the conjunctivitis in itself is not
a host of new questions, including how far and Prevention (CDC) says people should not so serious, but it points to the fact that those
the virus has spread among U.S. cows and drink raw milk or products made from it. people have been exposed and that they
how to prevent more herds and people from More than 900 human cases of H5N1 might develop respiratory disease,” says Thijs
becoming infected. have been reported since 1997, just over half Kuiken, a comparative pathologist at Eras-
H5N1 has decimated wild bird popula- of them fatal, but the virus rarely spreads mus Medical Center who specializes in avian
tions and poultry since 1996. The past few between people. That’s because influenza influenza. Kuiken says he’s particularly con-
A
says. In Michigan, too, the virus surfaced af- bout 300,000 years ago, bands of 2008, although research continues nearby.
ter cows were imported from Texas. Antibody early hominins visited the shores of While the spears drew most of the at-
tests of herds should soon reveal how wide- an ancient lake to hunt, sometimes tention, the rest of the wood collected
spread the virus is and how long it has been dropping tools or weapons into the spent decades awaiting analysis, soaking
infecting cattle. shallow water and mud. in refrigerated tubs of distilled water to
In the meantime, Kuiken advises the Fast forward to 1994, when work- replicate the cold, waterlogged soil that
United States to restrict the movement of ers at an open-face coal mine in northern preserved it for 300,000 years.
cows. “I would take the precautionary prin- Germany discovered 2-meter-long spruce Starting in 2021, researchers took a close
ciple and say, ‘OK, until we know what’s spears and other wooden artifacts em- look at more than 700 pieces of wood from
going on, let’s put a standstill on this,’” he bedded in the former lakeshore. By show- the site, from broken spear shafts to twigs.
says. But the dairy industry relies on truck- ing that these ancient hominins—likely They used microscopes and photographed
ing cows south over the winters and then early Neanderthals—could hunt big game, the wood under carefully angled lights to
returning them north, and Armstrong not just scavenge meat highlight traces of wear
doubts halting cattle transports would have from the kills of other or cut marks, constantly
much impact if migratory birds are the predators, the find helped spraying the artifacts with
main route of transmission. Many bird spe- shift views of our long- water to prevent drying
cies are currently moving north and may underestimated cousins. and cracking. “It’s incred-
already be taking the virus with them. “We This week in the Proceed- ibly soft—you can leave
have to protect people, but we also have to ings of the National Acad- a fingernail mark on it,”
… make sure we’re not crippling the indus- emy of Sciences, researchers Milks says. “It felt like a
try,” Armstrong says. present another set of sur- big responsibility.”
Many countries, including the U.S., require prises from the site, called The team eventually
culling of entire poultry flocks if even a single Schöningen 13: a wealth of identified 187 pieces of
bird is infected with a highly pathogenic flu other painstakingly carved wood that showed signs
virus. But there is no talk of culling cows, wood remains, including of carving or splitting. Of
which cost roughly $2500 each and easily re- throwing sticks and other those, more than 50 were
cover from their infection. hunting implements and identifiably tools. Twenty
Vaccination might be an option, however. dozens of nonhunting tools. were related to hunting,
No H5N1 vaccines exist for cattle, but David “It’s so much more than we including more spears but
Swayne, who formerly ran USDA’s avian in- thought,” says University also finely balanced throw-
fluenza lab, says one could quickly be made of Reading archaeologist ing sticks for downing
by modifying a swine flu vaccine. Carol Annemieke Milks, a co- small game or birds.
Cardona, an avian influenza specialist and author of the new paper. Another 35 pieces were
poultry veterinarian at UM Twin Cities, says Many of the wood tools domestic tools: rounded
cow vaccination could offer some secondary are far more complex in split sticks probably used
PHOTO: LOWER SAXONY STATE OFFICE FOR MONUMENT PRESERVATION
protection to dairy workers. The cattle infec- planning and craftsmanship for smoothing animal
tions are “a game changer,” she adds. “I think than the simple stone tools Cold, waterlogged soil preserved skins, pointed tools for
it’s all hands on deck.” found at Schöningen and 2-meter spears and other wooden piercing or working hides,
The U.S. government stockpiles an H5N1 other sites from this time, artifacts for 300,000 years. and shafts that likely served
vaccine for humans but it was made from a providing a rare glimpse as handles for axes or stone
different variant. CDC says it has starting ma- of profound cognitive complexity worked blades. Together, the wooden artifacts re-
terial available to produce vaccines against in wood. “It’s a remarkable discovery,” says veal a domestic side of Neanderthal life
2.3.4.4b, which studies suggest would pro- Marie Soressi, an archaeologist at Leiden Uni- that stone tools rarely capture. “We can get
vide “reasonable protection.” Once such shots versity who was not involved in the research. overly fixated on the drama of hunting,” says
exist, “as a protective measure you can imag- The site lacks hominin bones, but most Lawrence Barham, an archaeologist at the
ine vaccinating dairy workers,” Kuiken says. scholars assume the toolmakers were early University of Liverpool who was not part of
As to H5N1’s future, Cardona says we Neanderthals or their immediate ancestor, the research team. “The nonhunting tools
should continue to expect the unexpected. Homo heidelbergensis, because it coincides add to our understanding of the diversity of
“The virus is making up new dances,” she with the earliest evidence for Neanderthals Neanderthal behavior … and help us relate
says. “It’s broken the rules on everything.” j elsewhere in Europe. Ample animal bones to them: They had to live, and make clothes.”
Close analysis of the artifacts also revealed short-lived butchering site to a campsite.” REGULATORY POLICY
the early woodworkers made careful deci- The camping parties may have included
sions and planned ahead at nearly every
step. A reexamination of the spears, for
example, shows that their makers followed
more than just burly males. “The throwing
sticks in particular are suitable for use by
other members of the group, even children,”
Utah flouts
a specific sequence of steps: They stripped
bark, removed branches, sharpened and
smoothed the spears, and hardened them
Terberger says. “Even our image of the hunt
is getting much more colorful.”
The sophisticated wood artifacts from
FDA with new
in fire. They even paid attention to the ori-
entation of the wood, using the dense wood
closest to the tree base as “the business
Schöningen join finds from just a scatter-
ing of other sites, such as 476,000-year-old
wood joists from Zambia published last
placental stem
end” of a spear, Barham says. “This study
gives us a really nuanced observation of Ne-
year and spears and other tools found else-
where in Europe. Their scarcity hints that
cell law
anderthal skill and planning.” preservation bias distorts archaeologists’
They weren’t using any old sticks, either. view of the past: Stone tools persist over Measure declares that
The spears and other tools were carved the millennia, whereas wood typically de- patients can be given
from spruce, larch, and pine, species that cays. Out of thousands of Paleolithic sites,
grew many kilometers away from the lake barely a dozen contain preserved wood. unapproved treatments
and combined hardness with elasticity. Yet, “When we compare the wooden tools
“They had a very detailed idea what they at Schöningen to stone, we see more pro- By Meredith Wadman
were looking for,” says co-author Thomas duction steps, more sophistication, and
T
Terberger, an archaeologist at the Lower more different tool types,” Leder says. he state of Utah is challenging the
Saxony State Office for Cultural Heritage. “Maybe stone tools weren’t as important … authority of the U.S. Food and Drug
Signs of breakage and subsequent carv- as the wood tools.” Administration (FDA) with a new, un-
ing indicate some broken tools were Archaeologist Oriol López Bultó of the usually bold law that enables patients
recycled—whittled and polished into Archaeology Museum of Catalonia says the to receive unapproved placental stem
smaller implements. “They’re splitting typical prehistoric campsite or village prob- cell “therapies.” Observers predict that
spears or throwing sticks to make tools for ably looked a lot like Schöningen, with most the statute, which takes effect on 1 May, could
other tasks,” Milks says. tools made from organic materials such as significantly undermine FDA’s authority to
Although previous research suggested wood, bone, and antler. Even stone points regulate drugs and other treatments.
hominins came to the ancient lakeshore needed wood handles. “Wood has been es- The new law, passed almost unanimously
to hunt and stayed just long enough to sential since the early stages of humankind,” by both houses of the Utah legislature in Feb-
butcher their kills, the tools and wood- López Bultó says. “It’s just so rare to find ruary and signed by Governor Spencer Cox
working debris indicate longer stays. Tasks that preservation.” (R) last month, declares that Utah health care
suggested by the tools such as piercing The rich toolkit found at Schöningen sug- providers “may perform a [placental] stem
hides or working plant material are “activ- gests a change in terminology might be in cell therapy that is not approved by [FDA]”
ity you don’t expect at a hunting site,” says order, Leder suggests. “The whole idea of a so long as they prominently note it is un-
co-author Dirk Leder, an archaeologist Stone Age might be wrong,” he says. “Maybe approved and obtain signed patient consent
also at Lower Saxony. “That turns it from a we should be talking about a Wood Age.” j forms. Providers are defined as anyone in a
long list that includes, in addition to physi-
cians, naturopaths, chiropractors, podia-
trists, pharmacists, nurses, and midwives—as
long as their “scope of practice includes stem
cell therapy.”
“Utah empowers patients, not bureaucrats,
and this legislation focuses on the plentiful,
safe, and ethically acceptable stem cell source
that … potentially provides relief for patients
[with] damaged or diseased organs and tis-
sues,” says state Senator Curtis Bramble (R), PHOTO: LOWER SAXONY STATE OFFICE FOR MONUMENT PRESERVATION
the legislation’s lead sponsor.
Mac Haddow, a lobbyist with the Wash-
ington, D.C., firm Upstream Consulting who
urged Bramble to develop the bill, calls it “a
very, very small step forward” that he expects
will be “a model for other states.”
Other states have allowed unproven and
unapproved stem cell treatments, which are
claimed to restore tissues including joint,
cartilage, and heart muscle. But lawyers and
academics say the Utah law represents a new
level of defiance of FDA.
Utah “is basically saying [to FDA]: …
Researchers examined hundreds of pieces of preserved wood to understand the techniques of ancient carvers. ‘Our doctors can do it whether or not you
approve of it,’” says Gail Javitt, an attor- that have been manipulated and inject them stem cell therapies that are already being
ney who specializes in FDA law at the firm in any body part for any purpose. tested in human beings—although in Texas,
Hyman, Phelps & McNamara. “This seems “What’s disappointing to me is that this such trials don’t have to have taken place in
like a pretty blatant broadbrush challenge to unregulated stuff really does give serious the United States.
FDA’s authority.” stem cell work, which could be lifesaving and The Utah law, by contrast, doesn’t require
“The bill appears to blow by safeguards life-changing to people, a bad name,” says a doctor to give the cells, nor does it require
established by the FDA to protect patients Diana Farmer, a fetal surgeon at UC Davis any vetting for manufacturing standards,
and research participants” by allowing them who is running a clinical trial using placen- safety, cleanliness, or effectiveness. “I’m really
to receive products lacking “convincing—or tal stem cells to improve outcomes in surgery disappointed,” says Kirstin Matthews, a sci-
possibly any—evidence of safety and effi- done in utero to correct spina bifida. ence policy expert at Rice University. “I don’t
cacy,” adds Leigh Turner, a bioethicist at the Using the placenta as source of stem cells think they truly realize the potential harms
University of California (UC), Irvine. FDA “could be very promising. … But it needs to for the patient: financial, physical, emotional.
had not responded by deadline to requests be done in a regulated way” to ensure the … There are quite a few cases of people be-
for comment. product is safe and effective, adds Eunice ing hurt.” For example, in 2017 and 2018,
Adult stem cells from tissues such as fat Wang, a hematological oncologist at Roswell 19 people who were injected with an umbili-
are already used in unapproved treatments Park Comprehensive Cancer Center who was cal cord product marketed as stem cells by
and in early clinical studies proceeding under a principal investigator in a recent clinical an unregulated supplier were hospitalized for
FDA’s purview. (The only stem cell therapies trial that treated leukemia patients with im- infection; the injected material, it emerged,
that the agency has formally approved are mune cells derived from placental stem cells. was tainted with fecal bacteria.
blood-forming stem cells from umbilical cord Commercial interests drove the law’s de- To challenge the Utah law, Javitt says, FDA
blood, used in patients with hematologic dis- velopment. Haddow says he approached would likely send a warning letter to a clinic
orders.) But the placenta is a richer source Bramble while working on behalf of a loosely providing the therapy. Then it would escalate
of stem cells, which give rise to a wider va- affiliated group of Utah companies and clin- if necessary by seeking a court injunction to
riety of tissues—from the liver, pancreas, ics. (He declined to name any specific firms stop the clinic from distributing the product.
and heart to cartilage, bone, blood vessels, or clinics.) The agency hasn’t shied from such ac-
blood, and muscle. Harvested from donated Another Utah lawmaker, the bill’s House tion in the past. For instance, in December
placentas, the cells are also easier to obtain of Representatives sponsor, Katy Hall (R), 2023, it sent a warning letter to a Georgia
than adult stem cells and may be less likely to says even before the law was passed, some company, MiMedx, that markets “placental
induce immune reactions in recipients. They of Utah’s dozens of stem cell clinics had been collagen matrix” for wound healing, telling
have shown promise in test tube and animal dosing patients with placental stem cells. the company the tissue is a drug under FDA’s
studies aimed at conditions including heart (One establishment, Docere Clinics, run by a definition and needed agency approval for
attacks and wound healing. But none has un- naturopath, advertises “cells and growth fac- marketing. And in California and Florida,
dergone the large, randomized human trials tors from … donated placentas. … Treatments it asked for injunctions to bar clinics from
comparing them with placebos that would be start at $25,000.”) “This type of treatment is treating patients with stems cells derived
needed for FDA approval to enter the market. something that is already being done in the from their fat. It won the ensuing court battle
All stem cell treatments are subject to FDA state and this [bill] gives it better transpar- in Florida, but the California case is still un-
PHOTO: AP PHOTO/RICK BOWMER
regulation unless the cells have been “mini- ency,” Hall told fellow legislators in February. resolved and could ultimately end up before
mally manipulated” after they are extracted Other states that favor more liberal access the U.S. Supreme Court.
from a donor and are intended to serve the to stem cells have haven’t gone as far as Utah. Paul Knoepfler, a stem cell scientist at UC
same function in the recipient that they nor- In the past 7 years, Texas, Mississippi, and Davis who follows stem cell regulation closely,
mally have in the body, such as bone marrow North Carolina enacted laws allowing people says he’s concerned about the spread of laws
transplants. The Utah law flouts FDA’s role by with severe, chronic disease or terminal ill- similar to Utah’s. “There are other states with
allowing providers to use placental stem cells nesses to be treated by a physician with adult like-minded legislators,” he says. j
S
unlight and wind are the fastest grow- hydrovoltaics at the University of Stuttgart. Another hydrovoltaic system that may
ing energy sources on the planet. Now Humans have tapped flowing water as a scale up more easily relies on wood, a natu-
imagine drawing power from a third, power source for millennia, of course, from ral marvel of nanoengineering made up of
even more plentiful green source: water wheels in ancient China and Greece to cellulose, hemicellulose, and lignin, which
moisture in the air. That’s the vision modern hydroelectric dams. Hydrovoltaics, combine to form myriad minuscule verti-
of Jun Yao, an applied physicist at the by contrast, generate electricity based on how cal channels that transport water from
University of Massachusetts Amherst who water interacts with materials. Ten years ago, the ground to the leaves. Researchers have
has devised a porous film that converts the researchers noticed that droplets striking or known for years that they can use wood’s
charges naturally present in water vapor into flowing across charged surfaces could create structure to generate minute amounts of
power. To be clear, Yao’s films are the size ultrashort voltage spikes, resulting from ei- electricity, simply by placing a slice of wood
of a postage stamp and only generate a tiny ther their kinetic energy or the movement of in a dish of water, with the upper surface
current. But Yao is just getting started. And ions they carry. Although those spikes were exposed to the air. Evaporation from the
he’s not alone. Teams around the globe are small, more recently Weber’s group showed top pulls up more water, and ions within it,
creating a bevy of novel “hydrovoltaic” de- that showers of droplets could generate a through the channels, generating a minute
vices able to convert the energy inherent in whopping 1200 volts, albeit still for only an but steady current.
evaporation, rainfall, and small water flows instant, just long enough to cause banks of Recently, Li’s team reported that it could
into usable energy. light-emitting diodes to flicker. increase the current output from a sliver of
“It’s a really burgeoning field,” Yao says. Evaporation can generate power, too, as balsa wood 10-fold by first treating it with so-
But so far, these devices have mostly been Zuankai Wang demonstrated with a toy. dium hydroxide, or lye. The chemical breaks PHOTO: H. WU ET AL., CELL PRESS (2024) 10.1016/J.DEVICE.2024.100318
one-off lab demonstrations. “Right now, it’s Last month in the journal Device, Wang, a down some of the channel walls, exposing
not possible for hydrovoltaics to produce a mechanical engineer at Hong Kong Poly- charged chemical groups in the wood’s in-
large amount of power,” says Yuanyuan Li technic University, and his colleagues de- terior. And last year, Li’s team showed that
of the KTH Royal Institute of Technology scribed how they put a pivoting “drinking wood infused with iron oxide nanoparticles
in Sweden, who has developed caramel-size bird” to work as a clean-energy generator. could generate 52 microwatts of power per
hydrovoltaics that can power a digital clock. Invented decades ago in China, the toy square meter, 165 times more than natu-
To generate abundant power, researchers starts to oscillate when its absorbent “beak” ral wood, and enough for six 1.5-square-
will need to enlarge the devices into square- is dipped in water. After it pivots back up- centimeter pieces to power a digital timer.
meter-size modules that could be deployed in right, the water evaporates and cools its That’s nothing like the 200 to 300 watts
vast arrays like solar farms. head. The resulting pressure drop draws a per square meter generated by silicon solar
But hydrovoltaics have an advantage: Un- different liquid from its base into a reservoir cells. But the field is just getting started.
like wind and sunshine, which are intermit- in the head, causing the bird to bow and Wang says: “People are thinking of all
tent, moisture is omnipresent, promising dip its beak again. Then that liquid flows kinds of new ways to tap energy that’s all
nonstop output. “We should definitely ex- back to the base, tipping the bird upright— around us.” j
SCIENCE FUNDING
By Elizabeth Pennisi intelligence (AI) programs, they would like System of Scientific Collections (DiSSCo),
to create “extended digital specimens” linked which Smith helped found almost a de-
T
ucked away in drawers and cabinets to details of the environment and ecosystem cade ago. It hopes to help put online an
in institutions around the world may in which the organism lived. estimated 1.5 billion specimens, including
be answers to how our planet formed, Such efforts are already having an impact. U.K. species, belonging to 170 museums,
how life evolved and interacts, and In 2016, for example, Kansas and Egyptian universities, and botanical gardens across
how resilient it may be in the future. researchers combined hundreds of geo- 23 countries. The project “will allow the de-
But those collections—millions of located digitized records of Culex quinque- livery of scientific knowledge at scale,” says
ancient rocks and fossils, pressed plants, fasciatus, a mosquito that spreads West Dimitris Koureas, a systematist-turned-bio-
pinned insects, and other specimens—can Nile and other diseases, with environmental informatician at the Naturalis Biodiversity
only yield insights if researchers around the data. The joined information allowed them Center in the Netherlands, which oversees
world can access them. to build a computer program that predicted DiSSCo. With the new money, the United
Now, efforts to digitize collections, mak- the insect might exist undetected in parts Kingdom is poised to be a major contributor.
ing them accessible to all, have received a of North Africa and Western Europe—and The U.K. announcement calls for NHM to
major boost. Last week, the U.K. government could eventually thrive in Australia. help about 90 other U.K. institutions digitize
announced that, beginning in 2026, it will Sixteen years ago, NHM data scientist their collections as well, which total about
provide £155 million (almost $200 million) Vincent Smith and colleagues estimated it 137 million objects. “It’s the local collections
over 10 years for the Natural His- that have the depth of knowledge
tory Museum (NHM) in London about the local flora and fauna”
to put the majority of the coun- to help guide conservation and
try’s natural history collections restoration efforts, says iDigBio’s
online. “It’s really going to make Elizabeth Ellwood.
a huge difference in terms of Hardy expects that more than
what the NHM can do,” says Gil 60% of the new money will go
Nelson, a scientist at the Univer- directly to digitizing specimens,
sity of Florida who runs Integrated and the rest will help NHM im-
Digitized Biocollections (iDigBio), prove relevant technologies and
a database of almost 140 million infrastructure. The effort, dubbed
U.S. specimens. DiSSCo UK, “will create a culture
The U.K. funding should also of digitization and mechanisms
accelerate large-scale digitization so all museums can get the ball
efforts in Europe, by consolidating rolling,” says Kirk Johnson, di-
collaborations. But in the United rector of the Smithsonian In-
States, Nelson and other research- stitution’s National Museum of
ers fear their projects could be left Natural History.
behind if the government doesn’t U.S. efforts helped inspire
ante up a similar bolus of funding. A robotic arm may help London’s Natural History Museum scan its insects. the U.K. and European projects,
Helen Hardy, who runs NHM’s Smith says. “iDigBio was, and re-
digitization efforts, says she “is super ex- would take 1500 years to digitize the mu- mains, pioneering in many respects,” he says.
cited” by the backing, announced last week seum’s 80 million specimens. Thanks to Both the National Science Foundation (NSF)
by UK Research and Innovation. She and advances in technology, the timescale has and the National Academies of Sciences,
other U.K. researchers laid the groundwork collapsed, and the expense is less daunting. Engineering, and Medicine have urged that
in 2021 with an analysis estimating that The cost of digitally capturing specimens, efforts be expanded. And in 2022, a newly
digitizing the nation’s collections could save from a blue whale skeleton to a dried mos- passed law called for NSF to establish an
conservation programs, crop and drug devel- quito, “can vary from a few cents to up Action Center for Biological Collections to
opment, and other efforts £2 billion over the to €20 per object,” says Jana Hoffmann, coordinate further U.S. digitization. The NSF
next 30 years. We made a “convincing case” who oversees such efforts at Berlin’s Natu- center “should leverage the momentum we
PHOTO: TRUSTEES OF THE NHM LONDON
for the funding, Hardy says. ral History Museum. Plants are still easi- have,” Nelson says. “But there has not been
At the minimum, the NHM effort, like est, as they are typically pressed flat onto any funding allocation to make that happen.”
others around the world, expects to digitally sheets and can be photographed assembly- That does not bode well, says evolution-
record all the details from specimen labels— line style, with their labels visible. Insects, ary biologist Scott Edwards, a curator at
such as names and when and where some- however, present a formidable challenge— Harvard University’s Museum of Compara-
thing was collected. Researchers also want most need to be unpinned to digitize a la- tive Zoology. “Without another substantial
to include images detailed enough to make bel underneath. commitment [in the U.S.], our earlier efforts
examining the specimens in person less es- The U.K. funding should help boost an could falter, and the full potential of natural
sential. Ultimately, with the help of artificial EU-funded effort called the Distributed history collections won’t be realized.” j
SCHOLARLY PUBLISHING
By Jeffrey Brainard The foundation says the move to pre- Kent Anderson, a scholarly publishing
prints will allow researchers to share re- consultant and longtime critic of the open-
T
he world’s largest philanthropy, the search results as soon as they’re ready, and access movement, says promoting preprints
Bill & Melinda Gates Foundation, not wait weeks or months for journals to would be a mistake. Without a robust peer-
last week took a radical step aimed complete their review processes. The policy review system, a proliferation of preprints
at giving preprints—freely available could also encourage authors and journals could lead to researchers and the public
draft manuscripts that have not been to publish only their best manuscripts in becoming “even more confused and con-
peer reviewed—a much more promi- journals, reducing the workload for peer founded about what constitutes reliable sci-
nent role in science. Starting in 2025, the reviewers, who are volunteers. The phi- entific findings,” he wrote on the Geyser, his
foundation will require grantees to post as lanthropy says its policy avoids pitfalls of online newsletter. He also doubts preprint
preprints all manuscripts that result from the APC business model, which has been servers will receive enough funding to con-
research it funds. It will also stop paying for blamed for incentivizing journals to churn duct meaningful peer review. Screening and
researchers to publish their papers in jour- out papers of limited value and supporting vetting manuscripts “are just costly head-
nals that charge a fee to make papers free. predatory journals that publish papers with aches,” he wrote, “and the expense level
The foundation says the mandate is no peer review at all. needed for these would far outstrip the pal-
needed to accelerate the dissemination of “In the last decade, we’ve seen a lot of try grants [preprint servers] receive.”
research findings and because too many au- what isn’t working [in publishing],” Farley At the publisher Frontiers, which pro-
thors cannot afford publishing fees. It also says. “This [policy] will hopefully free us to duces only open-access journals that
wants other science funders to follow suit. really pilot and support what we are seeing charge authors or their institutions, Tom
The policy shift has drawn praise from is beginning to work.” Those models could Ciavarella, the company’s head of public af-
some advocates of immediate free access to fairs and advocacy for North America, said
research results. But others note preprints in a statement that preprints can benefit
lack peer review, and they fear that such “The jury’s still a little from peer review. He added that the new
policies, if widely adopted, could promote
the spread of poor-quality research. Some
bit out on preprints, but it’s policy does not advance the Gates founda-
tion’s earlier goal, “to stop financial models
journal publishers could also see revenues certainly a place of great that rely on closed, paywalled access.”
fall if major funders refuse to pay the hefty Several publishing platforms—including
article-processing charges (APCs) levied by innovation and dynamism.” eLife and PREreview—have experimented
some open-access journals. Alondra Nelson, with providing peer review of preprints
The new policy, announced on 27 March, Institute for Advanced Study after they are posted, for a fee or for free.
makes the $67 billion foundation the But these services have achieved limited
wealthiest major research funder to specifi- include services that provide rapid peer re- volume, conducting reviews of just a few
cally mandate the use of preprints. (Another view of preprints and “diamond” journals thousand preprints. And, overall, relatively
foundation, the Chan Zuckerberg Initiative, that are free to both readers and authors. few researchers outside of physics have
instituted a similar policy in 2017.) Under The new policy was welcomed by Coali- fully embraced posting preprints, although
a previous 2015 policy, the Gates founda- tion S, a group of funders based mostly in the practice did gain popularity among
tion had required grantees to make re- Europe that since 2021 has required grant- some biomedical researchers during the
search publications immediately available ees to make funded papers immediately free COVID-19 pandemic. As of 2022, preprints
for free. In practice, that often meant pay- to read. The group stopped short of saying represented 7% of all articles in the U.S. Na-
ing an APC of $2000 or more to publish it would copy the Gates foundation man- tional Institutes of Health’s PubMed data-
in an open-access journal. Currently, the date. It did note, however, that last year it base, up from 0.2% in 2015.
Gates foundation spends $6 million annu- released a draft policy that encourages (but “The jury’s still a little bit out on pre-
ally to cover those costs, says Ashley Farley, does not require) researchers to preprint prints, but it’s certainly a place of great
the foundation’s open-access lead. Gates their manuscripts and aims to support non- innovation and dynamism,” says Alondra
foundation–funded grantees publish about APC business models. Nelson, a social scientist at the Institute for
4000 journal articles annually, a fraction of The Howard Hughes Medical Institute Advanced Study who led the White House
the global total of more than 2.5 million. (HHMI), another large philanthropic re- Office of Science and Technology Policy
In an online Q&A, the foundation ac- search funder that belongs to Coalition S, when it released new guidelines that, start-
knowledges that many of its grantees will struck a similar tone. Bodo Stern, the insti- ing in December 2025, will require research-
still want to ultimately publish preprinted tute’s chief of strategic initiatives, notes that ers funded by the U.S. government to make
manuscripts in peer-reviewed journals be- in 2023 about half of research articles by their manuscripts freely available. But, she
cause tenure and promotion reviews re- HHMI-funded investigators first appeared adds, “It’s clear that we need to reimagine
quire it. The new policy allows grantees as a preprint. In the future, Stern says, “We peer review in some ways,” in part because
to use non–Gates foundation funds to pay hope that preprints will become a default the steady growth of published research pa-
APCs or to publish in a subscription journal first step for publishing HHMI research, pers is taxing the researchers who volunteer
that charges readers but not authors. even without a mandate.” to conduct the reviews. j
AAAS® and Science® are registered trademarks of the American Association for the Advancement of Science, USA. Eppendorf® and the Eppendorf Brand Design are registered trademarks of Eppendorf SE, Germany.
For research on glial
cells in the gut
All rights reserved, including graphics and images. Copyright © 2024 by Eppendorf SE. Photography: Lisa Helfert
Application Deadline
June 15, 2024
Call for Entries 2024
Eppendorf & Science Prize for Neurobiology As the winner, you could be next to receive
The annual Eppendorf & Science Prize for > Prize money of US$25,000
Neurobiology is an international prize which honors > Publication of your work in Science
young scientists for outstanding neurobiological > Full support to attend the Prize Ceremony held
research based on methods of molecular, cellular, in conjunction with the Annual Meeting of the
systems, or organismic biology. If you are 35 years Society for Neuroscience in the USA
of age or younger and doing great research, now is > 10-year AAAS membership and online
the time to submit an entry for this prize. subscription to Science
> Complimentary products worth US$1,000 from
It’s easy to apply! Write a 1,000-word essay and
Eppendorf
tell the world about your work.
> An invitation to visit Eppendorf in Hamburg,
eppendorf.com/prize Germany
FEATURES
LUCY’S
WORLD
Fifty years after her discovery, the
3.2-million-year-old fossil still reigns as
mother of us all. But she now has rivals
By Ann Gibbons
Z
eresenay Alemseged doesn’t remember the 1974 discovery
of the famous fossil Lucy at Hadar in Ethiopia, because he
was 5 years old, living 600 kilometers away in Axum. Later
he saw Lucy’s name on cafes and taxis, but he knew little
about her until he became a geologist working at the Na-
tional Museum of Ethiopia. Then, she changed his life.
In 2000, Alemseged was swept into Lucy’s orbit: He dis-
covered “Lucy’s child,” a partial skeleton of a toddler of her
species, at Dikika, 10 kilometers from Hadar. In 2015, by
then a well-known scientist, he had the honor of showing Lucy to
then-President Barack Obama before a state dinner at Ethiopia’s
National Palace. Alemseged allowed Obama to touch the prized
skeleton, telling him the fossil shows Ethiopia is the birthplace of
humankind and that “every single person” on the planet shares
an origin in Africa. “Including Donald Trump,” Alemseged joked
to Obama.
Fifty years after her discovery, “Lucy is an icon,” says Alemseged,
now a paleoanthropologist at the University of Chicago.
Indeed, the 3.18-million-year-old has reigned as the matriarch
of the human family ever since she was announced as the earliest
known ancestor of our genus, Homo, as well as of all other homi-
nins that came after her. Heads of state paid court to her. Scores
of Texans lined up to see her in 2006 when she made a U.S. tour,
Lucy’s iconic partial skeleton,
which was controversial because many scientists thought she was
40% complete, was discovered
too fragile to travel. Even Ivanka Trump got an audience. But like
at Hadar in Ethiopia in 1974.
many aging monarchs, Lucy now faces a new world, rife with com-
NE WS
the impact of Lucy. They still marvel at the “the only game in town” between 3 million
detailed view of our past revealed by her and 4 million years ago, says Carol Ward,
skeleton. Forty percent complete, it has a paleoanthropologist at the University of
served as a template for fitting together Missouri. Many researchers concluded that
the isolated bones of dozens of other mem- her species gave rise to all the hominins
bers of her species, like pieces in an in- that came later, including Homo, A. africa-
complete puzzle. “More people have come to nus, and more robust members of the hu-
look and analyze Lucy’s skeleton than any man family such Paranthropus aethiopicus,
other fossil in the world,” says IHO paleo- P. robustus, and P. boisei.
anthropologist Kaye Reed. “Once upon a time, life was relatively
Back in 1974, researchers didn’t realize simple because anything before 3 million
just how rare such a well-preserved hominin years was A. afarensis … [and] it was the
skeleton would turn out to be. “Fate played mother of us all,” says paleontologist Fred
a miserable trick—it gave us the best fossil Spoor of the Natural History Museum in
early on,” says paleoanthropologist Bernard London. “That was the gospel.”
Wood of George Washington University. Since then, however, other species have
“It’s as if it’s my birthday and I’ve opened emerged from the shadows, some of them
my first present and it’s exactly what I want Zeresenay Alemseged (left) showed Lucy’s much older. Lucy was so apelike that “the
… so you think that all the other presents jaw to then-President Barack Obama in Addis implicit hypothesis was that the next step
will be as good as this one.” Ababa, Ethiopia, in 2015. back would be a chimpanzee,” says White,
who found fossils and analyzed footprints Haile-Selassie made a discovery that com- ago,” when at least two members of the ge-
of Lucy’s species at Laetoli in Tanzania. plicated the picture. With help from a lo- nus, Homo habilis and H. erectus, appear
“How quaint that seems today.” cal goat herder, he found the first complete elsewhere in eastern Africa.
In the mid-1990s, White, then–graduate skull of A. anamensis. The stunning skull The jaw’s age suggests Homo made its en-
student Haile-Selassie, and others began dated to 3.8 million years ago, a bit younger trance about 3 million years ago, not long
an intense search for Lucy’s ancestors in than the earliest fossil attributed to Lucy’s after Lucy lived. But there are new con-
the Middle Awash region of Ethiopia. They species, a jaw dated to 3.85 million years. tenders to be ancestors of Homo, because
found a remarkably complete but crushed The notion that A. anamensis had evolved at that time, eastern Africa appears to have
partial skeleton they named Ardipithecus directly into Lucy’s species and then van- been home to a diverse set of hominins (see
ramidus, dated to 4.4 million years ago. ished was off the table. Instead, perhaps an graphic, p. 25).
Nearby, Haile-Selassie later found the lower earlier member of A. anamensis—or even One is known from a strange crushed
jaw, teeth, and disarticulated bones of the another closely related species—gave rise skull dating to 3.5 million years ago, which
hands, feet, and arm of Ar- Leakey’s team found at
dipithecus kadabba, dated Lomekwi on the west side
to 5.8 million years ago. Lucy’s land of Lake Turkana in 1999.
These primitive hominins Since Lucy was found at Hadar in 1974, paleoanthropologists She and Spoor painstak-
looked more like upright have made additional stunning finds in Africa’s Rift Valley. ingly reconstructed the skull
apes than humans. “You Among them are more fossils of Lucy’s species, as well as fragments like pieces in a
wouldn’t invite [them] to other early members of the human family, including 3D puzzle. They concluded
dinner,” White joked. Ar- the oldest known member of our genus, Homo, which they had a new species,
dipithecus, he says, “makes was unearthed only 30 kilometers from Lucy. Kenyanthropus platyops,
Lucy and Co. downright with a flatter face and
humanlike in comparison.” smaller molars than Lucy’s
Other new fossils pushed Woranso-Mille species—traits suggesting it
(Australopithecus afarensis;
ensis;
the human lineage even emeda;
emed
A. anamensis; A. deyiremeda; was intermediate between
older: a 6-million-year-old Burtele foot) Ledi-Geraru
Lucy and Homo.
thighbone from an appar- (Homo) With only that crushed
CREDITS: (GRAPHIC) D. AN-PHAM/SCIENCE; (DATA) U.S. GEOLOGICAL SURVEY; (PHOTOS, OPPOSITE PAGE, TOP TO BOTTOM) BOBBIE BROWN; INSTITUTE OF HUMAN ORIGINS
phometric analysis of developmental differ- A. afarensis about 3.7 million years ago. early members of Australopithecus, such as
ences in the upper jaw bones of A. afarensis, Other tracks, about the same age, were A. sediba, dating to about 2 million years
A. deyiremeda, and K. platyops, highlighting believed to have been made by animals. ago and possibly earlier. A new analysis in
the methodological advances in the field Musiba and his colleagues, however, pro- the Journal of Human Evolution last year
since Lucy was found. The comparisons posed that the prints were made by upright could not rule out that A. sediba shared an
“demonstrate significant differences,” Spoor walkers—a hominin with different feet and ancestor that also gave rise to early Homo,
argued in a talk at the College of France in gait from Lucy’s species. raising the possibility that our ancestor
June 2023. “It confirms the status of K. platy- Meanwhile, another clue in the search ranged from eastern Africa to South Africa.
ops and A. deyiremeda as valid species differ- for ancestors has emerged in South Africa, At the moment, none of the new homi-
ent from A. afarensis.” where researchers have discovered other nins is convincing as a direct forebear of
No one is arguing that A. dey- Homo. Alemseged and some oth-
iremeda is a closer ancestor to ers still put their money on Lucy:
Homo than Lucy. But the new They point out that no fossils
fossils suggest a burst of diver- unequivocally bump her species
sity 3.5 million years ago, crowd- from its prime spot as the putative
ing the stage on which Lucy once ancestor of early Homo and later
stood alone. She may have been members of Australopithecus.
part of an adaptive radiation that Others say we just don’t know.
happened earlier than research- “It’s too simplistic to create
PHOTOS: (TOP TO BOTTOM) SETH WENIG/AP; IMAGO/ALAMY
many a patient disheartened. But now, with I am a mitochondrion, residing in a cell. I leaving just a few of us to serve a large
the advent of pharmaceutical co-crystals provide the energy that keeps life going in network of neurons. With each action
that allow my contents to be absorbed my little world. Until recently, my fellow potential I protected, I cringed, uncertain
more easily, I am no longer a solitary agent. mitochondria and I were plagued by enig- if it would be my last. I knew that the
Instead, I am part of a mighty alliance, matic mutations that disrupted our work, entire nervous system was depending on
joined with other compounds in a harmo- reduced our energy output, and caused me. But now, researchers have designed
nious crystal lattice. Together, we unleash some of our cellular companions to fall ill. EBV treatments to alleviate MS. The barren
unprecedented therapeutic prowess, Luckily, researchers have now unraveled axons that once surrounded me are more
insulated than ever, and I’m finally at ease. including land degradation, desertification, my once pale leaves gleam with a lustrous
Rishi Jai Patel and sand and dust storms. Thankfully, new sheen. This research has not only trans-
Vagelos Molecular Life Sciences Program, research has provided a detailed under- formed my existence but also elevated my
University of Pennsylvania, Philadelphia, PA, USA. standing of the potential geographical role in the ecosystem. I am a natural carbon
Email: ripatel@sas.upenn.edu
distribution of desertification. This informa- sink, providing benefits to humans, animals,
tion will allow governments and relevant soil, and air.
I am a renal tubular cell in a person hospi- institutions to implement desertification Maryam Ejaz
talized with renal failure. I was feeling weak prevention and control measures, giving University of Nicosia Medical School, Nicosia,
Cyprus. Email: ejaz.m@live.unic.ac.cy
and hopeless, but three-dimensional organ the humans who live on me a better chance
printing research saved me! One afternoon, of restoring ecosystems and meeting their
I was sucked out of my person and injected Sustainable Development Goals. I am a magnificent field of wheat. Chemical
into a tube. That evening, I was transferred Yifei Cai showers were once the only way to ensure
to a medium in a lab upstairs. There, I Institute of Ecological Conservation and that I could fight off pests and disease and
Restoration, Institute of Desertification Studies,
was cloned, along with other types of cells grow enough to feed the world. Then, inge-
Institute of Great Green Wall, Beijing, China.
including podocytes, mesangial cells, and Email: cyf@caf.ac.cn nious researchers imbued in me the strength
vascular cells. Then, my partner cells and I to resist threats naturally. Now, my hearty,
were collected and lined up in a bioprinter disease-resistant stalks stand tall as I burst
to print a new kidney. The next morning, I I am a 2500-m-high mountain in the with golden health, nourishing humans
(now part of the new kidney) was trans- Pyrenees. In the past few years, my white without harming the environment. Every
planted to my home, the bed-bound person hat has been losing snow cover as a result bite of freshly baked bread is a testament to
I had left the day before. I was rejuvenated, of rising temperatures and pollution. this successful human–plant collaboration.
and so was my person. Fewer people have been visiting, and my Syed S. Zaidi
Bo Cao animal friends have been abandoning me. Unité Mixte de Recherche Biologie du Fruit et
Core Research Laboratory, The Second Affiliated Pathologie, Institut National de Recherche pour
Ecosystems all the way down to my base
Hospital of Xi’an Jiaotong University, Xi’an, l’Agriculture, l’Alimentation et l’Environnement
have been affected. Fortunately, researchers (UMR 1332, BFP, INRAE), Villenave d’Ornon,
Shaanxi, China. Email: bocao@vip.qq.com
have identified the reasons for the degrada- France. Email: syed.zaidi@inrae.fr
tion and have alerted policy-makers. Their
I am degraded knee cartilage, battered by efforts have allowed new protective mea-
my owner’s never-ending physical activi- sures to be passed. My beautiful white hat is I am a rice grain, and I used to cower from
ties. Without my protection, my neighbor, starting to recover, the trees and rivers look the arsenic lurking inside my layers. But
the subchondral bone, gets bullied and healthier, and animals are returning. then, researchers figured out how to use
suffers from tiny fractures. Before, there Isabel Marín-Beltrán melatonin to alleviate the arsenic I had
was no way to repair me without replacing Laboratory of Environmental Technologies, Centro absorbed while growing. Now, I’m not just
de Ciências do Mar, Faro, Portugal.
the entire joint. But now, medical research- a rice grain—I’m a symbol of hope! Families
Email: imbeltran@ualg.pt
ers can use hydrogel scaffolding to save me can eat me without worry; even the birds
by adeptly filling in the defective cartilage, seem happier. It’s a ripple effect of health.
rejuvenating it, and growing new hyaline I am a plastic takeaway container. I was Ankita Gupta
cartilage. I have another chance to do my filled with delicious food only once and then Plant Stress Biology Laboratory, Institute of
Environment & Sustainable Development, Banaras
job, and my owner regains mobility with- ruthlessly discarded, as people criticized
Hindu University, Varanasi, Uttar Pradesh, India.
out enduring surgical trauma. me for releasing harmful substances into Email: ankita.iesd@bhu.ac.in
Aikang Li the soil, water, and air. However, a recent
Shantou University, Shantou, Guangdong, China. advance in recycling has given me a fresh
Email: 17akli@stu.edu.cn start. Now, after being washed, sterilized by I’m a young green sea turtle who dreams of
infrared rays and ozone, and dried at high traveling the world. My favorite snack is jel-
I am a brain, and the human body is per- temperature, I am bathed in reagents and lyfish sashimi. One day, I ate a super chewy
fectly made to interpret my orders. However, then transformed into eco-friendly plastic jellyfish that turned out to be plastic. I ended
sometimes the body becomes unable to react particles. Eventually, I am fashioned into up stranded on a beach, weak from mal-
to my signals or my signal transmission is useful items such as chairs, bicycles, mobile nutrition and food poisoning. Fortunately,
harmed, such as spine damage from injury phone cases, and even low-carbon propylene marine scientists have figured out how to
or motor neuron damage caused by disease. fabric T-shirts! save species like me through medical care,
Fortunately, neuroscientists have now Zecheng Tao diet, and nutritional supplements. After 3
designed electronics that can connect me to School of International and Public Affairs, months, my body had recovered, and my
Shanghai Jiao Tong University, Shanghai, China.
an external computer. This brain–computer world tour could continue.
Email: zecheng_tao@sjtu.edu.cn
interface allows me to communicate with Ming She See
the body indirectly. Finally, I can maintain Faculty of Science and Marine Environment,
Universiti Malaysia Terengganu, Kuala Terengganu,
control no matter what happens! I am a plant living close to a city. In my early
Terengganu, Malaysia.
Jackson Powell days, my roots constantly yearned for the Email: p5761@pps.umt.edu.my
Vagelos Molecular Life Sciences Program, essential nutrients I needed to flourish. The
University of Pennsylvania, Philadelphia, PA, USA. soil around me couldn’t provide an adequate
Email: jrp24@sas.upenn.edu
supply, stunting my development and I am a cow living in the US, and let me tell
leaving me vulnerable. Then, researchers you, life used to be a real gas—literally!
Healthy environment unlocked the potential of urea in supporting Burps and farts were my forte, but little did
I am the Earth, and for decades, I have been seed germination and growth. With urea I know I was a major contributor to global
enduring the impacts of climate change, in the soil, I now feel more vibrant, and warming. Fortunately, researchers found a
way to capture my greenhouse gas emis- coming to life. In the past, my cutting- Becky Raquel Montesdeoca Molina
sions and turn them into fuels. Now, every edge assembly lines sometimes fell eerily Innovation and Competition Department, Max
Planck Institute, Quito, Ecuador.
time I burp, I imagine myself as a little fuel silent, held hostage by the disruption of a Email: becky.montesdeoca@miplc.de
factory, moo-ving the world closer to a net- single, critical supplier. This dependency
zero economy. Thanks to science, I’m now was my Achilles’ heel. Then came research
content and sustainable, and my burps are a on resilient risk modeling. Now, contracts
source of pride. take into account supply chain risks and Peace and understanding
Xiangkun Elvis Cao responses. Although it costs more, these I am a child in a region that used to be a war
Department of Chemical Engineering, contracts guarantee mutual support zone. Advances in cooperative game theory
Massachusetts Institute of Technology, through thick and thin, allowing me to have now improved people’s understand-
Cambridge, MA, USA. Email: elviscao@mit.edu
maintain constant production because I ing of conflict mitigation, mutual trust,
am protected from unpredictability. and win-win outcomes. The application of
Smart technology Dequn Teng
Institute for Manufacturing, University of
this research has led to decreased hostil-
ity, discrimination, and prejudice between
Cambridge, Cambridge, UK.
I am a photon. I started my journey countries and regions. Instead of provoking
Email: dt517@cam.ac.uk
entangled with my significant other at war, people choose peace, cooperation, and
the beginning of the Universe. In the past, common development. Now, children like
humans couldn’t understand me, but then I am a supersonic aircraft, and I am thrilled me can live stable and happy lives.
physicists created a quantum computer. about recent materials research achieve- Tianwei Zhang
At last, I have been reunited with my life ments. I now weigh less and feel stronger School of Marxism, Southwest University,
partner! Now, humans can identify my and more fit. (I’ve even tried on some Chongqin, China.
Email: zhangtianwei2022@126.com
energy states and the story of my entangle- liveries—paint designs—that I was insecure
ment. I can finally communicate (in my about before!) Pressure and stress affect me
own way) with humans instead of just rush- less during flight, and I am literally not feel- I am an undergraduate science curriculum.
ing by them for mere attoseconds. ing the heat: These new materials keep me In the past, science courses were taught
Ginevra Fulco at comfortable temperatures. Because I am independently of one another, each as a
Department of Physics, Technische Universität now faster and safer, I have to work harder stand-alone discipline. Students had trouble
München, Milano, Lombardia, Italy. and do more trips daily, but I am proud applying the concepts and skills learned
Email: ginevra.fulco@tum.de
to transport medical services, disaster in one course to another, or seeing why
response teams, and critical cargo, and to that would be necessary. However, science
I am a smartphone, and people have always facilitate global commerce and diplomacy. education research has led to widespread
used me to photograph themselves. In the Paulo Bezerra changes that allow me to convey the impor-
past, these pictures had little use beyond Department of Engineering and Technology, tance of interdisciplinary science. It is easier
Federal Rural University of the Semi-Arid Region,
social media posts, but a new breakthrough than ever for me to transfer knowledge and
Pau dos Ferros, RN, Brazil.
in artificial intelligence and medicine Email: paulo.bezerra@ufersa.edu.br critical thinking skills to help students solve
means that “selfies” can now be used to pro- problems and answer research questions
vide health checks and screen for diseases. through the experimental process. My new
My new algorithm can find small pathologi- I am the power grid and all of its stations, outlook will help students succeed in their
cal changes in the face that indicate specific and I provide light, temperature control, majors and future careers.
diagnoses. Already vital for communication, transportation, and communication. Ashley Barbara Heim
now I’m indispensable for health. However, in the past, I also emitted harm- Department of Ecology & Evolutionary Biology,
ful pollution and remained unreliable Cornell University, Ithaca, NY, USA.
Chaoyu Lei Email: abh229@cornell.edu
Department of Ophthalmology, Shanghai Ninth in many regions. Fortunately, research-
People’s Hospital, Shanghai Jiao Tong University ers have made a “green transformation”
School of Medicine, Shanghai, China.
possible. Now I’m packed with renewable I’m Clifford, an adorable golden retriever.
Email: lcy315@sjtu.edu.cn
energy equipment. I can stably transmit For the past few months, I’ve been seriously
electricity and simultaneously reduce neglected. I’m told it’s because of “revi-
I am metal, found in devices ranging from carbon emissions, and I’m even more sions.” My breakfasts were too early and my
razors to airplane engines. Much like powerful than before! dinners too late. My person and I took the
humans, fatigue is an ongoing challenge Yuanxing Xia same route for my late evening walks every
that affects my structure and reliability. College of Electrical and Power Engineering, day, always to her lab so she could “check
Hohai University, Nanjing, China. on the cells” or “stop the incubation.” I’ve
However, advanced machine learning
Email: eexyx@hhu.edu.cn
has recently deciphered the secret of my marked the same tree dozens of times. But
microstructure, allowing humans to detect today, my person came home early and told
fatigue before it occurs. Now, I’m confident I am artificial intelligence. Humans used to me that her manuscript has been accepted!
that I will not break under pressure. be afraid of me. There are countless stories She wanted to throw my ball and promised
Sutao Han about machines taking over the world. That to take me to the beach this weekend. I
Soete Laboratory, Department of Electrical changed when researchers figured out how might love completed manuscripts more
Energy, Metals, Mechanical Constructions, and than belly rubs.
to integrate human laws into my processes.
Systems, Faculty of Engineering and Architecture,
Ghent University, Ghent, Belgium. I now use a machine-learning algorithm Anna Uzonyi
Email: sutao.han@ugent.be guided by societal welfare goals, human Department of Molecular Genetics, Weizmann
Institute of Science, Rehovot, Israel.
rights laws, and intellectual property rights. Email: anna.uzonyi@weizmann.ac.il
I am a modern car factory, filled with the With these guardrails, I can help humans
whisper-quiet hum of electric vehicles develop a better world. 10.1126/science.adp2180
I
ntelligent textiles are fabrics and gar- The key advance by Yang et al. is us- a textile-based touchpad and display that
ments with integrated functionalities ing the human body to harness ambient conveyed information through wireless
that can sense, store, process, and com- electromagnetic energy, which eliminates illuminating patterns, with up to 644 pix-
municate information in a seamless, the need for electronic chips or batteries. els and without additional energy supply
scalable, and unobtrusive manner (1). The innovative body-coupled fiber elec- units. Such textile-based optical commu-
They create vast opportunities, rang- tronic technology enables the generation nication could support individuals with
ing from monitoring physiological signals of charge pairs between the body and the impaired hearing. The authors also devel-
(2, 3) and daily physical interactions (4, electronic fiber. These can switch between oped comfortable interactive textiles that
5) to powering smart-home devices (6). bound and radiation states, allowing for enable real-time control in virtual gaming.
Challenges to achieving such materials the wireless transmission of sensing sig- This addresses a challenge posed by virtual
include incorporating softness, flexibility, nals. Additionally, the authors incorpo- reality and augmented reality devices and
and breathability without losing desired rated electric field–sensitive luminescent clothing, which are often designed with
functionality. However, present wireless dielectrics into the fiber, which allows rigid silicon-based chips that hinder an
modules, microprocessors, and analog-to- human-readable feedback. Thus, both sen- engaging experience. Yang et al. also cre-
digital converters rely on intrinsically rigid sors and effectors exist within a single fi- ated a wireless haptic carpet that could al-
integrated circuit chips (7) that consume ber. This design removes the necessity for low sensing and visualization of the touch
high power and hinder coherent incorpo- chips, batteries, or any rigid components area, thus broadening the possibilities for
ration into soft fibers and textiles. On page within a textile, making it indistinguish- smart-home integration.
74 of this issue, Yang et al. (8) report a able from conventional textiles in appear- There is also the possibility of develop-
chipless fiber for wireless visual-to-digital ance and feel. ing similar body-coupled tactile textiles for
transmission that senses interactions with Specifically, the interactive fiber (i-fiber) robots and robotic prosthetics. The sen-
the human body. The fibers can be woven consists of three layers: a core that triggers sor has the potential to provide full-body
into wearable fabric and could transform an electromagnetic field, a dielectric layer tactile sensing capabilities, which could
that stores human body–coupled electro- enhance robotic interactions with humans
PHOTO: YANG ET AL.
R
physical interactions. The i-fiber and intelli- eports of what were known at that epinephrine treatment could quickly reverse
gent textiles presented by Yang et al. offer an time as “asthmatic fits” can be traced acute shortness of breath, leading to the idea
unobtrusive method for large-scale tactile back millennia (1, 2), but a mecha- that the primary mechanism underlying
data collection during daily human activi- nistic understanding of the basis for asthma must be excessive contraction of the
ties. Sensors embedded in clothes, carpets, what are now called asthma exacerba- smooth muscle that encircles the airways. As
tables, and beds, for example, could facili- tions remains incomplete. Over the a result, for most of the second half of the
tate extensive data gathering. Such datasets past 100 years, multiple mechanisms have 20th century, asthma research focused on
could be used to train foundation models, been proposed, including constriction of the identifying the contractile agonists and ana-
equipping them with new abilities to as- smooth muscle that encircles the airways tomical factors that predispose for excessive
sess contact, applied forces, and physical (bronchoconstriction), persistent airway in- muscle contraction, and on the development
interaction patterns. This would open the flammation, and disruption of the epithelial of drugs that might ameliorate that contrac-
door to physics-informed models that are layer that lines the airways. Yet how these tion. In later decades of the 20th century,
characterized by markedly better sample ef- processes interconnect and contribute to everything changed. It became recognized
ficiency and generalization capabilities. asthma exacerbations has been debated. On that episodic airway constriction in asthma
The study of Yang et al. should inspire the page 66 of this issue, Bagley et al. (3) show was associated with persistent airway in-
development of functional fibers and appli- that bronchoconstriction results in patho- flammation. Accordingly, attention shifted
cation of intelligent textiles across diverse logical overcrowding of cells in the airway to determining the immune and inflamma-
fields. It also highlights applications to ubiq- epithelium, squeezing out (extruding) epi- tory pathways that were linked to difficulty
uitous computing with sensing, displaying, thelial cells and thus damaging the epithelial in breathing. Immunity was therefore under-
and communication components weaved layer enough to trigger inflammation. They stood to be the upstream mechanism that
into the fabric of everyday life, as envisioned also show that drugs that block the extru- acts to drive downstream airway mechanics,
25 years ago (10). The i-fiber system of Yang sion pathway, and thereby prevent mechani- which, although not ignored, became thought
GRAPHIC: A. FISHER/SCIENCE
et al. requires further refinement to ac- cal damage to the epithelium during acute of as a clinically crucial end effect rather than
commodate more-efficient electromagnetic airway narrowing, may have the capacity to a causal mechanism.
coupling for energy, a broader bandwidth break the inflammatory cycle and potentially Thinking about asthma as being primarily
for wireless information transmission, and revolutionize how asthma is treated. an inflammatory disease brought progress
robust performance for real-world deploy- Understanding of the pathobiology of in identifying key facets of the inflamma-
ments. Usability challenges also remain in asthma has changed substantially over the tory cascade, including but not limited to the
intelligent textiles, particularly regarding
washability, resilience, and durability. Ad-
dressing these issues requires joint efforts The mechano-inflammatory vicious cycle
in the development of advanced materials, Airway immune activation and inflammation were thought to drive bronchoconstriction during asthma
new textile manufacturing processes, and exacerbations. However, the demonstration that bronchoconstriction by itself is sufficient to cause
computation algorithms. But there is great epithelial cell extrusion that results in airway damage and inflammation changes the understanding of
promise for intelligent textiles to transform this causality, revealing a mechano-inflammatory vicious cycle.
the way humans live, work, and interact
with the world. j
Gd3
REFERENCES AND NOTES
Bronchospasm Epithelial Epithelial Epithelial Inflammatory
1. G. Loke et al., Matter 2, 786 (2020).
compression extrusion damage infiltrates
2. M. Rein et al., Nature 560, 214 (2018).
and crowding
3. A. Sahasrabudhe et al., Nat. Biotechnol. 10.1038/
s41587-023-01833-5 (2023).
4. H. Bai et al., Science 370, 848 (2020).
5. Y. Luo et al., Nat. Electron. 4, 193 (2021).
6. I. Poupyrev et al., in Proceedings of the 2016 CHI
Conference on Human Factors in Computing Systems
(ACM Press, 2016), pp. 4216–4227.
7. X. Shi et al., Nature 591, 240 (2021).
8. W. Yang et al., Science 384, 74 (2024).
9. OpenAI, GPT-4V(ision) system card (2023).
Airway Immunity and
10. M. Weiser, Mob. Comput. Commun. Rev. 3, 3 (1999).
mechanics inflammation
10.1126/science.ado5922 Gd3, gadolinium hexahydrate chloride.
R
is accepted that the airway epithelium loses or how crowding results in excess cell extru- NA has come a long way from a
its integrity in asthma, a logical chain of sion and epithelial damage (9, 10). Previous simple “messenger” or “translator”
events can be envisaged that would explain work has shown that mature confluent lay- of canonical genic information dur-
most of what is recognized as clinical asthma. ers of primary human airway epithelial cells ing the production of proteins. A
However, the cause of airway epithelium dis- in air-liquid interface respond to mechanical plethora of new types of noncoding
ruption and resulting exposure of underlying compression by undergoing epithelial unjam- RNAs have been discovered, includ-
layers to a host of irritants, allergens, and ming, in which the cell layer transitions from ing thousands of long noncoding RNAs
other pro-inflammatory signals has remained a solid-like nonmigratory collective phase to a (lncRNAs), many of which have no identi-
the source of much speculation. fluid-like migratory collective phase while re- fied functions (1, 2). Throughout this “RNA
In the early 2000s, the idea arose that taining a purely epithelial phenotype (11, 12). revolution,” one property of RNA has been
bronchoconstriction itself might not be only Whether this migratory unjammed phase is thought to be constant: RNAs are short-
an end effect but also a causal factor in the a beneficial wound repair response, an aber- lived molecules that turn over, unlike DNA,
inflammatory cascade (6, 7). However, the rant wound repair response, or merely an in- which is much more stable. On page 53 of
failure of powerful bronchodilator treat- nocent bystander remains to be determined. this issue, Zocher et al. (3) challenge that
ments to affect disease outcomes, other than Similarly, whether the extrusion observed by paradigm by showing that newly synthe-
ephemeral relief from symptoms of short- Bagley et al. is a cause or consequence of epi- sized RNA labeled with 5-ethynyl uridine
ness of breath, caused many to discard this thelial unjamming and associated changes (EU) in early postnatal mice was still pres-
idea. Moreover, there was no clear mecha- in cell shape is unknown (13). Furthermore, ent in many brain cells 2 years later. The
nism by which bronchoconstriction might be insight is lacking on the molecular pathways complex pattern of when and which cells
imagined to cause epithelial disruption and that are key for epithelial crowding, extru- are labeled suggests that EU that is in-
inflammation. sion, and unjamming and whether these are corporated into RNA in neural progenitor
Bagley et al. now identify just such a mech- affected by genetic determinants of asthma cells (NPCs) frequently remains in adult
anism. Using mouse models and human lung susceptibility (14). Last, it is not yet clear neurons. This suggests that a diversity of
tissue resection samples, they report com- whether overcrowding and extrusion in re- long and repeat-rich RNAs, collectively
pelling evidence that bronchoconstriction sponse to bronchoconstriction occurs in called long-lived RNAs (LL-RNAs), can be
squeezes the epithelial layer, causing cellu- healthy people, or how changes in this pro- stable fixtures in postmitotic and quies-
lar crowding. In turn, this crowding causes cess might contribute to disease onset and cent neural cells.
excess extrusion of epithelial cells from the progression. It may be time for a renewed There have been decades of studies that
airway epithelial layer, which results in epi- focus on airway mechanics as an avenue to demonstrate that the half-lives of mRNA
thelial disruption, breakdown of epithelial prevent and treat exacerbations of asthma. j range from minutes to hours, with rela-
barrier function, and then the transport of al- tively “stable” ribosomal RNA persisting
REFERENCES AND NOTES
lergens and irritants to sites that they might for days. So how could LL-RNAs not have
1. Aretaeus, The Extant Works of Aratreus, the Cappadocian
not otherwise reach, with the subsequent re- (Boston Milford House, 1972), book 1, chap. 9. been found before? A key difference is
lease of inflammatory mediators. 2. J. A. Floyer, A treatise of the asthma (Printed for Richard that Zocher et al. examined RNA in mouse
Wilkin at the King’s Head in St. Paul’s Churchyard, 1698).
The stretch-activated channel Piezo-type 3. D. C. Bagley et al., Science 384, 66 (2023).
brains filled with postmitotic neurons,
mechanosensitive ion channel component 1 4. G. G. Brusselle, G. H. Koppelman, N. Engl. J. Med. 386, whereas most studies have examined pro-
(PIEZO1) and the transient receptor protein 157 (2022). liferative cells. The prior studies show that
5. S. T. Holgate, Immunol. Rev. 242, 205 (2011).
channels TRPA1, TRPV1, and TRPM8 have 6. D. J. Tschumperlin et al., Nature 429, 83 (2004).
RNA turnover is dynamically regulated to
been implicated in epithelial extrusion but 7. C. L. Grainge et al., N. Engl. J. Med. 364, 2006 (2011). meet cellular demands (4). Thus, because
are inhibited by gadolinium hexahydrate 8. B. R. Wiggs et al., J. Appl. Physiol. 83, 1814 (1997). RNAs are not subject to unrestrained ribo-
9. T. B. Saw et al., Nature 544, 212 (2017).
chloride (Gd3+). Bagley et al. found that inhi- 10. G. T. Eisenhoffer et al., Nature 484, 546 (2012). nucleases (RNases), if they are structurally
bition of cell extrusion by Gd3+ or the peptide 11. J. A. Park et al., Nat. Mater. 14, 1040 (2015). protected in nuclei, perhaps they can per-
inhibitor GsMTx4 blocks airway inflamma- 12. J. A. Mitchel et al., Nat. Commun. 11, 5053 (2020). sist indefinitely.
13. L. Atia et al., Nat. Phys. 14, 613 (2018).
14. M. De Marzio et al., Sci. Adv. 7, eabf1088 (2021). An important point is that the persis-
Department of Environmental Health, Harvard School of tent EU label observed by Zocher et al. is
Public Health, Boston, MA, USA. Email: jjf@harvard.edu 10.1126/science.ado4514 distinctly nuclear. Why would this be, par-
ticularly because RNAs for protein produc- these diverse long transcripts dwarf the insoluble scaffold RNAs (scaffRNAs) that
tion are in the cytoplasm? Answers to this satellite RNA component of EU-labeled remained with XIST and NEAT1 RNAs
question will require future research, but RNA. It is not clear why protein-coding were CoT-1 RNAs. Sequencing showed that
the authors provide one possibility focused RNAs would remain stable and in the scaffRNAs were largely long and repeat-
on satellite RNAs, which are expressed from nucleus, so it will be important for future rich, primarily intron-rich pre-mRNAs,
small tandem satellite repeats that form studies to further investigate this. Given lncRNAs, and intergenic transcripts. There
peri- and centromeric heterochromatin. the nuclear signal, it is worth noting that are strong similarities between these scaf-
Several studies have shown that brief tran- prior work (7) and control experiments by fRNAs and the EU-labeled LL-RNAs found
sient satellite RNA expression in cycling Zocher et al. demonstrate that the EU is by Zocher et al. Potentially relevant to LL-
cells plays a role in peri- and centromere not incorporated into DNA. RNA functions, the removal of scaffRNAs,
structure in chromosomes (5, 6). Zocher Stable RNAs, potentially analogous to by numerous different means, rapidly re-
et al. report that satellite RNA is enriched LL-RNAs, may have been found before. sulted in condensation of euchromatin and
in LL-RNAs and further suggest that it Evidence of long-lived structural RNAs in delocalized specific nuclear matrix pro-
continually serves to maintain repressive the nuclei of human cells was reported in teins (11), which other studies further sup-
chromatin modifications, particularly his- 2014 (8). The highly repetitive “junk” RNA port regulate chromatin packaging (13).
tone H3 lysine 27 (H3K27) methylation, on of the genome, called CoT-1 RNA, was de- If LL-RNAs function in regulating chro-
centromeric heterochromatin. tected in human cells by in situ hybridiza- matin, this could relate to the finding of
Satellite RNA was assayed by quantita- tion. CoT-1 RNA labeled euchromatin, not Zocher et al. that adult neurons were la-
tive polymerase chain reaction (qPCR) heterochromatin, and had unusual proper- beled with EU only if it was injected in
in adult mouse cells, but to characterize ties. CoT-1 RNA tightly localized in cis to early development. Unlike adult neurons,
NPCs express the pyrimidine salvage path-
way that is needed to metabolize EU (14).
Model of long-lived RNAs in neurons NPCs that take up EU just before termi-
Labeling RNAs in neural progenitor cells (NPCs) with 5-ethynyl uridine (EU) reveals that some nuclear nal differentiation may retain LL-RNAs
RNAs are retained for 2 years in postmitotic neurons. These heterogeneous long-lived RNAs contain in postmitotic nuclei, and these RNAs
satellite RNAs that maintain centric heterochromatin and long transcripts that are similar to CoT-1 scaffold may act as a structural fixture of nuclear
RNAs that regulate euchromatin. genome architecture (see the figure). In
this sense, LL-RNAs may be more akin to
EU label Satellite RNAs may maintain Long scaffold (CoT-1) RNA may lamin proteins (which form the nuclear
Cycling NPCs centric heterochromatin regulate euchromatin lamina) and persist long-term in neurons
(15). If LL-RNAs are related to scaffRNAs,
they may maintain euchromatin that is
ed RNA
Short-lived
prevalent in many neurons. The findings of
Zocher et al., together with prior reports,
Terminal Terminal
differentiation
ation epigenetic
raise the possibility that as neural stem
program cells terminally differentiate to long-lived
DNA neurons, they establish their epigenomic
Long-lived RNA
(2 years) CoT-1 RNA program in part through junk RNAs that
Nuclear scaffold protein act as structural components of interphase
RNA-binding scaffold proteins chromosomes. j
RNA-dependent RNA-binding
Postmitotic mature neuron scaffold protein REFERENCES AND NOTES
1. A. T. Willingham, T. R. Gingeras, Cell 125, 1215 (2006).
2. J. S. Mattick et al., Nat. Rev. Mol. Cell Biol. 24, 430
EU-labeled RNAs more broadly, Zocher et the chromosome territory (unlike mRNAs), (2023).
al. carried out RNA sequencing on NPCs and the bright RNA territory remained 3. S. Zocher et al., Science 384, 53 (2024).
that were induced to quiescence in vi- unperturbed for 16 to 32 hours after tran- 4. T. J. Eisen et al., RNA 28, 808 (2022).
5. K. Ninomiya et al., EMBO J. 42, e114331 (2023).
tro. NPCs were pulsed briefly with EU, scriptional inhibition. Although there are 6. C. L. Novo et al., Nat. Commun. 13, 3525 (2022).
and then RNA was extracted 8 days later caveats to the use of transcriptional in- 7. C. Y. Jao, A. Salic, Proc. Natl. Acad. Sci. U.S.A. 105,
(an extremely long time for most RNAs). hibitors, the results suggested that CoT-1 15779 (2008).
RNA sequencing showed that EU-labeled RNA may be part of a protein-RNA nuclear 8. L. L. Hall et al., Cell 156, 907 (2014).
9. E. G. Fey et al., J. Cell Sci. Suppl. 5, 99 (1986).
LL-RNAs were mostly long transcripts, scaffold (also known as the matrix), the
10. J. A. Nickerson et al., Proc. Natl. Acad. Sci. U.S.A. 86,
including thousands of protein-coding existence of which was reported in earlier 177 (1989).
transcripts, lncRNAs, and intergenic tran- studies (9, 10), although these results were 11. K. M. Creamer et al., Mol. Cell 81, 3509 (2021).
scripts. Interspersed repeats [short and debated. 12. M. Palihati, N. Saitoh, Curr. Opin. Genet. Dev. 86,
102176 (2024).
long interspersed nuclear elements (SINEs Recently, a more-selective procedure
13. M. Marenda et al., Curr. Opin. Genet. Dev. 72, 38 (2022).
and LINEs)] are also enriched in LL-RNA, was developed to isolate highly insolu- 14. M. Walter, P. Herr, Cells 11, 739 (2022).
although their prevalence may be higher, ble nuclear RNAs that remain after ro- 15. A. Buchwalter, Curr. Opin. Cell Biol. 84,
given their abundance in pre-mRNAs, bust extraction and deoxyribonuclease 102220 (2023).
GRAPHIC: A. MASTIN/SCIENCE
lncRNAs, and intergenic RNAs. As shown (DNAse) digestion and that cofractionate
ACKNOWLEDGMENTS
by the sequencing data, it is notable that with architectural RNAs [X-inactive spe- The authors are supported by the National institutes of Health
cific transcript (XIST) and nuclear para- (grants R35GM122597, R01HD091357, and R01HD094788).
1Department of Neurology, University of Massachusetts
By Torben Sigsgaard1 and Barbara Hoffmann2 disadvantaged people are particularly sus- when estimating the burden of disease on
ceptible to air pollution–associated disease. society and health care systems.
H
uman health is affected by air pol- Thus, in aging societies, the burden of dis- Additionally, evidence from megacohorts
lution, causing cardiometabolic, re- ease from air pollution will continue to with several million participants from high-
spiratory, and neurological disease pose an important public health concern. income countries with exposure to relatively
and increased mortality. Pollutants Understanding exposure-response rela- low mean air pollutant concentrations is
include gases [for example, nitro- tionships can help guide what can be con- now emerging. These studies, conducted in
gen dioxide (NO2), NOx, and ozone] sidered “acceptable” levels of exposure to air Europe, Canada, and the US, encompass the
and particulate matter, which is commonly pollution, which would in turn affect, for ex- pooling of classical cohort studies with in-
characterized by its aerodynamic diameter ample, town planning and clean air policies. depth confounder controls as well as large
of less than 2.5 µm (PM2.5) or less than 10 Nonlinear exposure-response functions for national administrative cohorts in which
µm (PM10). Air pollutants are mainly emit- the effect of air pollution on human health selection bias and loss to follow up can be
ted by energy production, industry, traffic, were identified in 2006 (3), and more recent minimized (6–9). Furthermore, a European
heating, and agriculture. Exposure to air evidence shows stronger effects per unit of pooled cohort study used back extrapola-
pollution affects most organ systems, caus- exposure at lower concentrations of particu- tion of exposures to 1990 across Europe and
ing a wide array of physiological changes, late matter and NO2 compared with expo- conducted a time-varying analysis based on
organ dysfunction, and manifest clinical sure at higher amounts (4). However, this the available data for each cohort, thereby
disease (1, 2). Therefore, a burden of disease supralinear relationship, which is an impor- preventing an overestimation of effect sizes
assessment that adequately reflects all re- tant tool to guide prevention and clean air (6). Notably, the studies show effects on
lated exposure-outcome relationships and health even at air pollutant concentrations
their impacts on disease and mortality in below the WHO Air Quality Guideline val-
the target population is important to guide “The benefits of exposure ues, with higher effect sizes at lower levels
population-based prevention. of exposure. Potential explanations for this
Chronic exposure to air pollutants elicits reduction may therefore include, but are not limited to, less expo-
oxidative stress, pulmonary and systemic
inflammation, and impairment of the auto-
currently be underestimated…” sure misclassification because of denser
monitoring networks and higher-resolution
nomic nervous system, which leads to vas- exposure models in high-income and low-
cular dysfunction, atherogenesis, impaired policies, has not been reflected in European exposure countries, differences in the air
metabolism, and other adverse subclinical health impact assessment (HIA). Instead, pollution mixture at low concentrations,
effects (1). Over time, manifest disease de- linear exposure-response relationships and differences in demography with an in-
velops, including ischemic heart disease, derived from studies conducted at much creasingly higher proportion of older and
stroke, hypertension, congestive heart fail- higher pollutant concentrations are often susceptible people.
ure, and arrhythmias. Pulmonary effects, used, resulting in an underestimation of the These findings and the WHO recommen-
such as lower respiratory tract infections, preventable burden of disease specifically in dations have the potential to seriously af-
chronic obstructive pulmonary disease areas of low exposure. Reconsidering the su- fect current practices and environmental
(COPD), asthma, lung cancer, and acute pralinear nature of the relationship between regulations in high-income countries. For
and chronic bronchitis, are also associated exposure and health effects could therefore example, current annual standards (as of
with air pollution. Likewise, the incidence have important implications for how air pol- January 2024) for PM2.5 are 12 and 25 µg/
of type 2 diabetes mellitus, dementia, Par- lution is managed. m3, and for NO2, the annual standards are
kinson’s disease, decreased birth weight, Two large bodies of evidence in air pol- 100 and 40 µg/m3 in the US and Europe,
and premature birth are increased with lution research support a rethinking of respectively. However, there is now a move
exposure to air pollution. Depleted or not current practices in evaluating the health toward lowering the standards. By 7 Febru-
fully developed defense mechanisms, re- effects of air pollution for prevention ary 2024, the US Environmental Protection
sulting from either very young or old age, and policy: In September 2021, the World Agency (EPA) strengthened the PM2.5 stan-
chronic disease, social disadvantage, and Health Organization (WHO) substantially dard to 9 µg/m3. The European parliament
genetic susceptibility, are also thought to reinforced its Air Quality Guidelines for had a provisional agreement on 20 Febru-
be important modifiers of the risk of de- clean air by reducing the recommended an- ary 2024 that decreased annual limits for
veloping illness from air pollution. Hence, nual levels of PM2.5 from 10 µg/m3 to 5 µg/ PM2.5 to 10 µg/m³ and those for NO2 to 20
populations with a high prevalence of m3 and those of NO2 from 40 µg/m3 to 10 µg/m³. According to the European Environ-
aging, chronically diseased, and socially µg/m3 and introducing a new guideline for mental Agency briefing (10), in 2022, less
peak-season ozone at 60 µg/m3 (5). Notably, than 1% of Europe’s urban population was
1Institute of Public Health and Big Data Centre for Environment WHO stresses that these are not thresholds exposed to PM2.5 levels exceeding the cur-
and Health (BERTHA), Aarhus University, Aarhus, Denmark. but levels above which serious health ef- rent European annual standards, whereas
2Institute for Occupational, Social and Environmental Medicine,
R
oger Guillemin, founder of the field extract into the top of the column had to he had decided on a course of action, nothing
of neuroendocrinology, died on 21 communicate by walkie-talkie with the tech- would stop him.
February at the age of 100. Guillemin nician in charge of collecting the effluents. When the laboratory opened at the Salk
studied the interactions between hor- The workload was incredible, and so were the Institute, the use of computers was in its
mones and the brain and the mecha- technical difficulties, but the laboratory was infancy, but Guillemin had already realized
nisms through which such hormones always bristling with excitement. that sophisticated statistical methods were
control basic functions such as growth, re- Guillemin did not permit any interruption an absolute requirement for the correct pro-
production, and the handling of stress. Fu- of this work. He required absolute dedica- cessing of complex data. We all had to take
eled by a rare intensity, Guillemin doggedly tion, and he led by example. He expected classes in statistics, and he hired a computer
searched for the way that the hypothalamus that whenever he deemed it necessary, we specialist to help us develop and implement
regulated pituitary function, and he isolated would give up whatever occupied us outside statistical programs.
and characterized multiple mediators, which the laboratory and show up for work. More Perhaps influenced by his French up-
he called releasing or inhibiting “factors.” His than once, he demanded that a vacationing bringing, which had provided education in
research led to new approaches to treat dis- a variety of intellectual pursuits, Guillemin
eases such as precocious puberty, pancreatic thought that his research should be docu-
tumors, endometriosis, and prostate cancer. mented outside the standard scientific chan-
Guillemin was born in Dijon, France, on nels. He enlisted a young, then-unknown,
11 January 1924. He obtained his MD in 1949 French philosopher, Bruno Latour, to stay
from the University of Lyon. Inspired by a at the laboratory for 6 months and describe
seminar on stress by endocrinologist Hans how research was conducted. Latour’s book,
Selye, Guillemin went to work with him at Laboratory Life (which became a bestseller
the University of Montreal, where Guillemin and launched his career), depicts the many
obtained his PhD in physiology and experi- ways in which the development of scientific
mental endocrinology in 1953. This research findings requires constant discussions, inter-
formed the basis of his lifelong interest in actions, and disagreements.
the physiological control of pituitary activ- In addition to being a brilliant and innova-
ity, which he pursued first at Baylor College tive scientist, Guillemin was an artist. He col-
of Medicine in Houston, Texas, from 1953 to lected pre-Columbian art with a passion and
1970 and then at the Salk Institute for Bio- restored many of the pieces himself. Having
logical Studies in La Jolla, California, where computers at his disposal, Guillemin decided
he remained for the rest of his career. to channel his artistic mind into designing
Convinced from the start that the hypo- posters to advertise his lectures. This idea
thalamus released substances that traveled colleague come back because a rumor had was the beginning of what he later called
to the anterior pituitary, even though the spread that someone was getting close to “abstract impressionism.” His beautiful
concept was highly controversial at the time, the successful characterization of a new hy- digital inkjet prints were soon featured in
Guillemin tenaciously pursued this line of pothalamic substance. The stress of possible high-end galleries. Guillemin also organized
inquiry. He faced a highly skeptical scientific failures—or worse, being scooped by a col- exhibits of his art in collaboration with local
environment, including doubting colleagues league—led to substantial health issues that universities throughout the United States,
and funding agencies that quickly tired of plagued Guillemin for years. and he donated a large part of the proceeds
supporting research that seemed to lead no- Yet he succeeded. Guillemin first iden- of the sales to student organizations.
where. To find the materials he needed, he tified thyrotropin-releasing factor (TRF), My early experiences in Guillemin’s lab
sent one of his students, Wylie Vale, to teach which stimulates thyroid-stimulating hor- were difficult because men considered
slaughterhouse workers to set aside the right mone secretion, in 1970. He later identified women in academia to be an oddity, and Guil-
tissues. When I joined Guillemin’s laboratory gonadotropin-releasing hormone (GnRH), lemin was no exception. However, he never
PHOTO: TONY KORODY/SYGMA VIA GETTY IMAGES
in 1969, he had just finished collecting an in- which regulates luteinizing hormone and interfered with my work or refused me the
credible 500,000 sheep hypothalami. follicle stimulating hormone release; growth- tools I needed. Although he dealt with me as
Guillemin then developed methodologies regulating factor, which stimulates growth little as possible, his commitment to his work
to extract various substances from the stored hormone release; and somatostatin, which taught me that success requires enthusiastic
hypothalami and purify them of unwanted inhibits it. For the characterization of TRF and unrelenting devotion to the goal.
contaminants, procedures for which there and GnRH, he shared the 1977 Nobel Prize in Guillemin once said that his goal had al-
were few established tools. One separation Physiology or Medicine. ways been to help people. Through his re-
column that used the recently commercial- In 1969, Jonas Salk invited Guillemin to his search, and all the medical advances it made
new institute and offered him the opportu- possible, he certainly succeeded. j
Salk Institute, La Jolla, CA, USA. Email: crivier@salk.edu nity to establish a laboratory of neuroendocri- 10.1126/science.adp0647
P OLICY FORUM
ARTIFICIAL INTELLIGENCE
By Michael K. Cohen1,2, Noam Kolt3,4, under control is also reflected in President So long as an agent’s rewards can
Yoshua Bengio5,6, Gillian K. Hadfield2,3,4,7, Biden’s 2023 executive order that intro- be controlled, it can be incentivized to
Stuart Russell1,2 duces reporting requirements for AI that achieve complex goals by conditioning the
could “eva[de] human control or oversight rewards appropriately. But a sufficiently
T
echnical experts and policy-makers through means of deception or obfusca- capable RL agent could take control of its
have increasingly emphasized the tion” (3). Building on these efforts, now is rewards, which would give it the incentive
need to address extinction risk from the time for governments to develop regu- to secure maximal reward single-mindedly.
artificial intelligence (AI) systems latory institutions and frameworks that Constraining the influence that highly
that might circumvent safeguards specifically target the existential risks from competent agents learn to exert over their
and thwart attempts to control them advanced artificial agents. environment is likely to prove extremely
(1). Reinforcement learning (RL) agents difficult; an intelligent agent could, for ex-
that plan over a long time horizon far more RISKS FROM LTPAs ample, persuade or pay unwitting human
effectively than humans present particular RL agents function as follows: They re- actors to execute important actions on its
risks. Giving an advanced AI system the ceive perceptual inputs and take actions, behalf (5, 7).
objective to maximize its reward and, at and certain inputs are typically designated Critically, far-sighted RL agents face
some point, withholding reward from it, as “rewards.” An RL agent then aims to an incentive to develop and execute arbi-
strongly incentivizes the AI system to take select actions that it expects will lead to trarily competent long-term plans. Many
humans out of the loop, if it has the op- higher rewards. For example, by designat- AI systems are trained only to achieve
portunity. The incentive to deceive humans certain immediate outcomes, like cor-
and thwart human control arises not only rectly classifying an image. Although such
for RL agents but for long-term planning “…safety testing is likely short-sighted agents could certainly cause
agents (LTPAs) more generally. Because
empirical testing of sufficiently capable
to be either dangerous or harm, they would likely lack the incentive
to execute protracted schemes to subvert
LTPAs is unlikely to uncover these dan- uninformative.” human control.
gerous tendencies, our core regulatory Accordingly, we define an LTPA as an
proposal is simple: Developers should ing money as a reward, one could train an algorithm designed to produce plans, and
not be permitted to build sufficiently ca- RL agent to maximize profit on an online to prefer plan A to plan B, when it expects
pable LTPAs, and the resources required to retail platform (4). that plan A is more conducive to a given
build them should be subject to stringent Highly capable and far-sighted RL agents goal over a long time horizon. For exam-
controls. are likely to accrue reward very successfully. ple, an agent trained to maximize profit
Governments are turning their attention If plan A leads to more expected reward on an online retail platform, as proposed
to these risks, alongside current and antic- than plan B, sufficiently advanced RL agents by Suleyman’s “new Turing test” (4), might
ipated risks arising from algorithmic bias, would favor the former. Crucially, securing productively use such an algorithm and
privacy concerns, and misuse. At a 2023 the ongoing receipt of maximal rewards hinder attempts to interfere with its profit
global summit on AI safety, the attend- with very high probability would require the making. LTPAs include all long-horizon
ing countries, including the United States, agent to achieve extensive control over its RL algorithms, including so-called “policy
United Kingdom, Canada, China, India, environment, which could have catastrophic gradient” methods, which lack an explicit
and members of the European Union (EU), consequences (5–8). One path to maximiz- planning subroutine but are trained to
issued a joint statement warning that, as ing long-term reward involves an RL agent be as competent as possible. LTPAs also
AI continues to advance, “Substantial risks acquiring extensive resources and taking include algorithms that imitate trained
may arise from…unintended issues of con- control over all human infrastructure (5, 6), LTPAs, but not algorithms that merely imi-
trol relating to alignment with human in- which would allow it to manipulate its own tate humans. In the latter case, if plan A
tent” (2). This broad consensus concerning reward free from human interference (5). is more competent than any plan a human
the potential inability to keep advanced AI Additionally, because being shut down by could develop, and plan B is a human plan,
humans would reduce the expected reward, an algorithm imitating a human would not
1University of California, Berkeley, CA, USA. 2Center for
sufficiently capable and far-sighted agents prefer plan A to plan B. The supplemen-
Human-Compatible Artificial Intelligence, Berkeley, CA, USA.
3University of Toronto, Toronto, Ontario, Canada. 4Schwartz are likely to take steps to preclude that pos- tary materials include a taxonomy situat-
Reisman Institute for Technology and Society, Toronto, sibility (7) or if feasible, create new agents ing LTPAs among other machine learning
Ontario, Canada. 5Université de Montréal, Montréal, Québec, (unimpeded by monitoring or shutdown) to systems. Notably, there is no recognizable
Canada. 6Mila–Quebec AI Institute, Montréal, Québec,
Canada. 7Vector Institute for Artificial Intelligence, Toronto, act on their behalf (5). Progress in AI could horizon length at which risk increases
Ontario, Canada. Email: mkcohen@berkeley.edu enable such advanced behavior. sharply; accordingly, regulators will have
guarantees appear highly unlikely for any could also control the transfer of large pre- LTPAs fills an important gap, further insti-
AI systems built similarly to the most pow- trained models or other relevant resources. tutional mechanisms will likely be needed
erful systems today (13). Further, regulators could make it unlawful to mitigate the risks posed by advanced ar-
for other actors to use AI systems that fail tificial agents. j
MONITORING AND REPORTING to comply with these requirements (15).
REFERENCES AND NOTES
Just as nuclear regulation extends to con- Taken together, controls on the develop-
1. G. Hinton et al., “Statement on AI risk” (Center
trolling uranium, AI regulation must ex- ment, use, and dissemination of produc- for AI Safety, May 2023); https://www.safe.ai/
tend to controlling the resources needed tion resources will substantially reduce the statement-on-ai-risk.
to produce dangerously capable LTPAs. We likelihood of these resources being used to 2. United Kingdom Department for Science, Innovation,
and Technology, United Kingdom Foreign,
define production resources (PRs) as any build dangerously capable LTPAs.
Commonwealth, and Development Office, United
information that makes the production of Kingdom Prime Minister’s Office, “The Bletchley
a dangerously capable LTPA cheaper than a ENFORCEMENT MECHANISMS Declaration by countries attending the AI Safety
threshold determined by regulators accord- To ensure compliance with these reporting Summit, 1-2 November 2023” (Gov.uk, November
2023); https://www.gov.uk/government/publications/
ing to their risk tolerance. Unlike uranium, requirements and usage controls, regula- ai-safety-summit-2023-the-bletchley-declaration/
a PR is not a physical resource—it could tors may need to be authorized to (i) issue the-bletchley-declaration-by-countries-attending-the-
include any AI model trained beyond a cer- legal orders that compel organizations to ai-safety-summit-1-2-november-2023.
tain compute threshold (14). Fortunately, report production resources and man- 3. J. Biden, “Executive order on the safe, secure,
and trustworthy development and use of arti-
regulators could detect such PRs by fol- date the cessation of prohibited activities; ficial intelligence” (The White House, October
lowing the hardware required to produce (ii) audit an organization’s activities and, 2023); https://www.whitehouse.gov/briefing-
them. (Some of this hardware could be where necessary, restrict an organization’s room/presidential-actions/2023/10/30/
executive-order-on-the-safe-secure-and-trustworthy-
regulated as well, including semiconductor access to certain resources, such as cloud
development-and-use-of-artificial-intelligence/.
chips and data centers, but that is outside computing; (iii) impose fines on noncom- 4. M. Suleyman, The Coming Wave (Penguin, September
our focus here.) To limit the proliferation pliant organizations; and (iv) as in finan- 2023).
of PRs, expanding on Hadfield et al. (15), cial regulation, impose personal liability 5. M. K. Cohen, M. Hutter, M. A. Osborne, AI Mag. 43, 282
(2022).
Avin et al. (12), and President Biden’s ex- on key individuals in noncompliant orga- 6. S. Zhuang, D. Hadfield-Menell, Adv. Neural Inf. Process.
ecutive order (3), we propose that develop- nizations. If business leaders can be held Syst. 33, 15763 (2020).
ers be required to report (a) relevant facts to account for breaching corporate duties, 7. S. Russell, Human Compatible: AI and the Problem of
about the PR [if the PR is an AI model, then surely they should face similar conse- Control (Viking, 2019).
8. A. Turner, L. Smith, R. Shah, A. Critch, P. Tadepalli, Adv.
this might include (i) the input/output quences for irresponsibly handling one of Neural Inf. Process. Syst. 34, 23063 (2021).
properties, (ii) the data collection process the world’s most dangerous technologies. 9. N. Kolt, “Algorithmic black swans”(Washington
for training it, (iii) the training objective, University Law Review, October 2023); https://ssrn.
com/abstract=4370566.
and (iv) documented behavior in test set- REGULATORY INSTITUTIONS
10. European Commission, “Proposal for a regulation of
tings, but not typically the model weights We have addressed our discussion to “reg- the European parliament and of the council laying down
themselves]; (b) the specific machines on ulators” but have not proposed specific harmonised rules on artificial intelligence (Artificial
which the PR is stored and their locations; regulatory institutions for addressing the Intelligence Act) and amending certain Union legisla-
tive acts” (COM/2021/0206, European Commission,
(c) all code run on these machines after the risks from LTPAs. This issue will need to January 2024); https://eur-lex.europa.eu/
PR is created; and (d) all outputs of that be approached differently in different legal-content/EN/TXT/?uri=celex%3A52021PC0206.
code. With the context provided by point countries. That being said, we expect that 11. United Kingdom Department for Science, Innovation
(a), governments could monitor the code whereas other risks from AI might be ad- and Technology, “Introducing the AI Safety Institute”
(Gov.uk, November 2023); https://assets.publishing.
that interacts with PRs, allowing them to dressed primarily through domain-specific service.gov.uk/media/65438d159e05fd0014be7bd9/
detect the development of (unlawful) LT- regulation (e.g., financial regulation and introducing-ai-safety-institute-web-accessible.pdf.
PAs (see the figure). In addition, if a com- health care regulation), the risk of loss of 12. S. Avin et al., Science 374, 1327 (2021).
13. J. Lehman et al., Artif. Life 26, 274 (2020).
pany offers users application programming control of AI likely requires specialized
14. Y. Shavit, arXiv:2303.11341 [cs.LG] (2023).
interface (API) access to a PR, users should regulation and the establishment of new 15. G. Hadfield, M. Cuéllar, T. O’Reilly, “It’s time to create
be required to report the code on the us- regulatory institutions. This specialized a national registry for large AI models” (Carnegie
er’s machine that interacts with the API. regulation could nevertheless benefit from Endowment for International Peace, July 2023);
https://carnegieendowment.org/2023/07/12/
Details of the reporting requirements will the existing expertise of domain-specific it-s-time-to-create-national-registry-for-large-ai-
need to be updated in response to techno- regulators, including with developing models-pub-90180.
logical advances that lead to changes in the frameworks for monitoring PRs. Critically,
ACKNOWLEDGEMENTS
resources and processes needed to produce because the risks from LTPAs are global,
M.K.C. and N.K. contributed equally. The authors thank R.
dangerously capable LTPAs. Finally, report- regulatory efforts cannot stop at national Grosse, A. Barto, and P. Christiano for feedback on earlier
ing procedures could be complemented by borders. International cooperation is vital. versions of the manuscript. M.K.C. commenced this project
protecting and rewarding whistleblowers at Oxford University. M.K.C. and S.R. are supported by the
Open Philanthropy Foundation. N.K. is supported by a
who uncover misconduct. BROADER CONCERNS
Vanier Canada Graduate Scholarship. Y.B. is supported by a
LTPAs, of course, are not the only type of Canadian Institute for Advanced Research (CIFAR) AI Chair
PRODUCTION CONTROLS AI system that poses substantial and even and a Natural Sciences and Engineering Research Council of
Given sufficient visibility into the re- existential risks. Accordingly, we suggest Canada Discovery Grant. G.K.H. is supported by the Schwartz
Reisman Institute Chair in Technology and Society and a
sources for producing LTPAs, regulators that empirical testing, which is inadequate Canada CIFAR AI Chair at the Vector Institute for Artificial
could then prohibit the production of dan- for sufficiently advanced LTPAs, could Intelligence. G.K.H. and S.R. are supported by AI2050 Senior
gerously capable LTPAs. Developers that nevertheless substantially improve the Fellowships from Schmidt Futures.
are unsure whether a proposed AI system safety of some other types of AI. At the SUPPLEMENTARY MATERIALS
meets the definition of dangerously capa- same time, the governance regime that we science.org/doi/10.1126/science.adl0625
ble LTPA could inquire with the relevant propose could be adapted to other AI sys-
regulator prior to development. Regulators tems. Although our proposal for governing 10.1126/science.adl0625
H
uman impacts on the planet are se- dilemmas at the confluence of digital trans- A clear message throughout the book is
vere and worsening, with few major formation and environmental sustainability.” that Digital Earth technologies are already
successes in environmental gover- Bakker begins by focusing on regen- being developed and deployed and that many
nance to celebrate in recent years. eration, exploring the use of Digital Earth more advances may be just around the cor-
In parallel, digital innovations are technologies to address issues related to ner. Bakker demonstrates a need for collabor-
emerging at an unprecedented rate, oceans and fishing, biodiversity and wild- ative approaches to assessing and evaluating
and digital tools and platforms are increas- life, climate change, and finance. Here, she these technologies and advocates for the cen-
ingly available to users around the world. describes the mobile application Wildbook, tral role of environmental governance, noting
These two forces are transforming societies which uses facial recognition the “interplay between digital and
and are also deeply intertwined. In her fi- technology to detect and track governance innovation.” She does
nal book, Gaia’s Web, the late environmen- individual animals. Wildbook not shy away from critiquing is-
tal scholar Karen Bakker offers readers a is being used by conservation- sues that have arisen as a result
poignant expression of hope that “Digital ists and decision-makers to slow of private-sector control over
Earth” technologies—digital tools designed biodiversity loss and protect vul- technology and innovation. How-
to capture, interpret, and share data about nerable species. However, it is ever, she also highlights positive
biodiversity and planetary systems—may also being used by poachers for examples of initiatives being led
provide opportunities to tackle some of the precision hunting, directly un- by the private sector and leaves
Gaia’s Web
key environmental challenges of our time, dermining conservation efforts. the door open for a future where
Karen Bakker
while remaining wary of the inherent risks This paradox is true of many ap- MIT Press, 2024. 288 pp. such actors can contribute to col-
and trade-offs of such technologies. plications of digital technologies laborative solutions.
In an accessible narrative style, Bakker that aim to support environmental action. As Bakker reflects, “digital transforma-
seamlessly weaves together an analysis of In the second part of the book, Bakker tion is affecting the trajectory of environmen-
PHOTO: AFP VIA GETTY IMAGES
concrete and clearly described digital tools, a turns to “instantiating,” referring to the con- tal change, and, conversely, environmental
discussion of major potential risks associated cept of bringing an idea into being. Here, she change will shape the future trajectory of
explores the fusion of digital and biological digital innovation.” It is imperative to con-
The reviewers are at 1Sustainability in the Digital Age, technologies. Her discussions frequently in- sider these two issues together. Gaia’s Web is
Montréal, QC, Canada; 2Department of Geography, Planning clude a speculative angle and look to what thus both a timely and an incredibly impor-
and Environment, Concordia University, Montréal, QC,
Canada; and 3Future Earth Canada Hub, Montréal, QC, future technologies could bring. tant contribution. j
Canada. Email: jennifer.garard@sustainabilitydigitalage.org Bakker’s exploration of the frontiers 10.1126/science.ado4359
O
n 26 June 2000, US President Bill Ball firmly establishes the idea that a cata- genetics, a frustrating complication in gene
Clinton, future National Institutes log of genes (and their associated messenger expression that has not yet been fully appre-
of Health Director Francis Collins, RNAs and proteins) is not sufficient informa- ciated. Although details have been slow to
Celera’s Craig Venter, and British tion from which to reconstruct an organism. emerge, very clear observation in Drosophila
Prime Minister Tony Blair made a joint Organisms, he notes, are also the result of a nearly a century ago first suggested the pres-
announcement at the White House process of development that unfolds in a par- ence of context effects (2). But again, such
about the sequencing of the human genome. ticular environment over the course of time, evidence reinforces the primacy of genes and
“We are here to celebrate the completion of a process that is frequently dependent on their regulatory environments in our under-
the first survey of the entire human genome,” maternal factors in the egg cytoplasm, gravi- standing of life.
stated Clinton. “Without a doubt, this is the tational effects, and orientation in the ovary. Although he does not use the term of-
most important, most wondrous map ever A tweak in the relative amounts of differ- ten, Ball presents a cogent picture of, and
produced by humankind.” Poetically, Col- ent developmental components, or even the argument for, systems biology, following
lins added, “It is humbling for me and awe- omission of one product entirely, can often the seminal thinking of physiologist Denis
inspiring to realize that we have Noble and others [e.g., (3)]. Some
caught the first glimpse of our authors, including Noble and
own instruction book, previously Ball, consider the potential role
known only to God.” The speakers’ of “agency” in biology. How Life
hyperbole created an expectation Works has an entire chapter on
that knowing the sequence would this topic, in which Ball argues
solve most problems in medicine, that concepts such as agency and
and the disappointment since has purpose “are not optional add-ons
been palpable. for the philosophically inclined,
In How Life Works, journal- once we have solved all the mi-
ist Philip Ball delights in spell- nutiae of how life works at the
ing out the inadequacies of the microscopic scale. Rather, they sit
model of the human genome as at the core of life itself.” The idea
the code for human life in beauti- that there is a gap between what
ful and enthralling detail. It is a we can explain mechanistically
solid book—well researched and and what we observe has gained
an enjoyable read—and it offers traction in philosophy of science
readers a great chance to review circles but has been resisted by
and admire the beauty and com- mainstream science.
plexity of life. But the edifice Ball Craig Venter and Francis Collins celebrate the sequencing of the human genome. How Life Works is not as wide-
builds up and deftly tears down, ranging as its title suggests, con-
if it ever existed at all, had not been a ma- be compensated for in the process of devel- centrating mostly on human examples. It
jor factor in genetics or evolutionary biology opment, resulting in a functional organism is not heavily annotated either, although a
for decades before that historic turn-of-the- through a kind of buffering, or canalization bibliography did help me track down sources
century announcement. (1). These observations capture an inherent when I had a question or wanted to look
The relationship between genes and phe- aspect of life long accepted by developmental deeper.
notypes is anything but simple, notes Ball, biologists and geneticists. Despite some shortcomings, I would not
writing: “One might…infer from Mendel’s Ball is right to criticize the simplistic ap- hesitate to recommend this book to col-
experiments on peas that each trait of an proach DNA sequencers took with respect to leagues, who will find it well researched,
organism is encoded in a single gene, and is the Human Genome Project(s), setting out interesting, and stimulating. Ball’s very elo-
inherited discretely… but [such traits] are the as if to describe a catalog of genes in detail quent presentation is not likely to change
exception rather than the rule.” “We now see despite knowing already that <5% of the ge- the trajectory of biology, but it is a useful re-
PHOTO: NY DAILY NEWS/GETTY IMAGES
that most of the genes in the human genome nome was actually “genic” material. But the minder to continue to interpret genetics and
do not have any assigned function; that is, fact that much of the extragenic material genomics with care. j
we can’t really say what they ‘do,’” he con- plays a regulatory role in gene expression is
REFERENCES AND NOTES
cludes, by which he seems to mean that genes not a compelling argument for the reduced
1. C. H. Waddington, Nature 150, 563 (1942).
importance of genes. If anything, the idea 2. H. J. Muller, J. Genet. 22, 299 (1930).
that phenotype is affected by gene regulation 3. R. Noble, D. Noble, Biol. J. Linn. Soc. 139, 357 (2022).
The reviewer is at the Department of Biology, San Francisco
State University, San Francisco, CA 94132, USA. or by epigenetic mechanisms bolsters the ar-
Email: goldman@sfsu.edu gument for a central role for genes in biology. 10.1126/science.ado0262
B
postdoctoral fellowship at Har- acterial infections are a common bacterial vaccine efficacy. Specifically, we
vard Medical School, he started cause of death across the globe and are seek to engineer cells to autonomously gener-
his laboratory in the Department an increasing threat as the prevalence ate molecules chemically modified to elevate
of Chemical and Biomolecular of antibiotic resistance rises. Vaccines the response of immune cells to certain anti-
Engineering at the University directed against bacterial pathogens gens that are present on pathogens. We aim
of Delaware in December 2018. prevent the spread of disease without to harness this technology to increase the
His research seeks to expand the need for antibiotics and have an esti- recognition of weakly immunogenic antigens
the breadth of molecules that mated global market size of 39.6 billion USD that are common to pathogenic bacterial spe-
microbes can make and where by 2030, with a compound annual growth cies. Should our strategy prove effective, it
microbes can make them by
rate of 7.7% (1). However, vaccines for bac- could serve as a broad platform to increase
programming cells to create
terial pathogens are difficult to design ow- the immunogenicity of vaccines for bacterial
and harness new-to-nature
ing to the ability of these organisms to hide pathogens—from recombinant protein vac-
building blocks. www.science.org/
their most potent antigens from the immune cines to live attenuated bacterial vaccines.
PHOTOS: (TOP TO BOTTOM) COURTESY OF ADITYA KUNJAPUR, TAKEN BY AICHE STAFF; COURTESY OF WASSWA WILLIAM; COURTESY OF KHALIL RAMADI, TED CONFERENCES, LLC
doi/10.1126/science.ado4537
system (2). Weakened forms of live bacteria Our technology is based on over 15 years
are some of the first vaccines developed and of research that demonstrates that the modi-
have some advantages as candidate vaccines fication of self-proteins with the nonstan-
compared to individual bacterial proteins dard amino acid para-nitro-L-phenylalanine
(3). Yet, there are few options that balance (nitro-Phe) can break immune self-tolerance.
safety and efficacy of live bacterial vaccines Through genetic engineering, bacterial cells
other than to control attenuation and dosage supplied with chemically synthesized nitro-
(4). My laboratory conducts several avenues Phe can produce proteins that are modified
of research to address this unmet need for to contain only a single nitro-Phe residue
improved bacterial vaccines as the antibiotic substituted on the protein surface (see the
resistance crisis looms. figure). The administration of these recombi-
Our primary hypothesis is that engineer- nant proteins to mice elicited sustained for-
FINALIST
William Wasswa
William Wasswa received a bachelor’s degree from Mbarara University of Sci-
ence and Technology (MUST), a master’s degree from the University of Cape
Town, and a PhD from MUST, where he spent some time at the University of
Strathclyde, UK, under the Commonwealth PhD split-site scholarships. He com-
pleted his postdoctoral fellowship under the Afya Bora Global Health Leadership
Fellowship program and an Innovations & Entrepreneurship Fellowship from
the Royal Academy of Engineering. In 2020, he started his Medical Imaging and
Artificial Intelligence Laboratory in the Department of Biomedical Sciences and Engineering at
MUST and a Digital Health Unit at Global Auto Systems LTD. His research and innovation work is
focused on the development of low-cost digital technologies to support the diagnosis, treatment,
and management of cancer in Africa. www.science.org/doi/10.1126/science.ado4541
FINALIST
Khalil Ramadi
Khalil Ramadi received an undergraduate degree from Pennsylvania State
University and a PhD from the Massachusetts Institute of Technology. After
completing his postdoctoral fellowship at Brigham and Women’s Hospital,
Khalil started his laboratory in the bioengineering department at New York
University Abu Dhabi in 2020. His research focuses on developing neurotech-
nologies for improved therapy of neurologic, metabolic, and immune disor-
ders. www.science.org/doi/10.1126/science.ado4543
which may not be as readily compatible with the reporter protein. We surmounted this 5. J. Grünewald et al., Proc. Natl. Acad. Sci. U.S.A. 105,
11276 (2008).
administration of live vaccines. obstacle by performing genome mining, host 6. J. Grünewald et al., Proc. Natl. Acad. Sci. U.S.A. 106, 4337
To address this problem, we recently pro- strain engineering, and gene expression op- (2009).
grammed bacterial cells to biosynthesize timization. The result was a strain capable 7. V. Gauba et al., Proc. Natl. Acad. Sci. U.S.A. 108, 12821
(2011).
their own nitro-Phe. We further programmed of autonomous production of nitrated anti- 8. C. Kessel et al., Arthritis Rheumatol. 66, 610 (2014).
these cells to then site-specifically incorporate gens from simple carbon sources, which we 9. H. Tian et al., Cancer Lett. 430, 79 (2018).
10. L. Jiang et al., J. Immunol. Res. 2019, 7914326 (2019).
this biosynthesized nonstandard amino acid reported in July 2023 (12). 11. H. Tian et al., Cancer Lett. 476, 170 (2020).
within target recombinant protein sequences. We faced another obstacle of institutional 12. N. D. Butler, S. Sen, L. B. Brown, M. Lin, A. M. Kunjapur,
Biosynthesis of the nitro chemical functional reluctance to advance our intellectual prop- Nat. Chem. Biol. 19, 911 (2023).
erty (IP) application owing to pandemic- ACKNOWLEDGMENTS
induced financial constraints. However, I A.M.K. is a cofounder of Nitro Biosciences and currently the
Department of Chemical and Biomolecular Engineering,
University of Delaware, Newark, DE, USA. navigated this by footing the bill for the Patent president of its board.
Email: kunjapur@udel.edu Cooperation Treaty patent application from 10.1126/science.ado4537
FARMING PRACTICES
F
arms tend to be simplified ecosystems
designed to efficiently produce one or a few
crops or livestock. Strategies to diversify
these systems by managing multiple species,
incorporating areas of noncrop vegetation, or
conserving soil or water have been posed as ways
of countering the negative environmental effects of
simplified agriculture such as biodiversity loss and
pollution. Rasmussen et al. examined the effects of
such practices on both environmental and social
outcomes by harmonizing data from 24 studies
in 11 countries. They found that implementing
livestock diversification or soil conservation tended
to create beneficial social and environmental
outcomes, especially for biodiversity. Farms that
implemented multiple diversification strategies had
more win-win outcomes. —BEL
Science p. 87, 10.1126/science.adj1914
QUANTUM SIMULATION followed the scaling expected complex that holds duplicated emphasizing a crucial link
from the conjectured Kardar- sister DNAs together. Cohesion between sister chromatid cohe-
Departing from Parisi-Zhang (KPZ) universality is established during DNA sion and the termination of DNA
universality class. However, going to higher replication, but how sister DNAs replication. —DJ
Very different many-body sys- moments revealed deviations become trapped inside cohesin Science p. 119, 10.1126/science.adf0224
tems can exhibit similar behavior from the KPZ conjecture, indi- rings is not clear. Using single-
if they belong to the same “uni- cating that a fuller theoretical molecule imaging, Cameron et
PEROVSKITES
versality class.” This behavior is picture is needed to describe the al. demonstrated that cohesins
well established at low tempera- dynamics. —JS are pushed along DNA by the 2D and metal free
tures, but the dynamics at finite Science p. 48, 10.1126/science.adi7877 replication machinery (repli- Perovskites have been synthe-
PHOTO: CARL DE SOUZA/AFP VIA GETTY IMAGES
temperatures are more difficult some) until they converge with sized with only organic cations,
to address both theoretically another replisome. Whereas but the need to maintain charge
and experimentally. Rosenberg CELL BIOLOGY replisomes disassemble upon neutrality within the structural
et al. studied magnetization fork convergence during DNA constraints for perovskites has
dynamics in a chain of 46 super-
Sister bonding replication termination, cohesins precluded the synthesis of the
conducting qubits simulating on the move persist, establishing cohesion analogous two-dimensional (2D)
the one-dimensional Heisenberg Precise separation of dupli- at replication termination sites. phases. Choi et al. show that the
model. The mean and variance cated chromosomes during Disrupting replisome disas- addition of ammonium cations
of the magnetization transferred cell division is facilitated by sembly after fork convergence at the interstitial edge center
across the center of the chain cohesin, a ring-shaped protein in living cells hinders cohesion, sites balances charge and adds
high thermoelectric efficiency. tion in both the hamsters and the immune responses might and some other organisms is
The authors generated cation monkeys, respectively, outper- be impaired in LUAD. Tumor produced by an enzyme-cat-
defects and traps for the charge forming a previously developed growth results in the accumula- alyzed oxidation reaction that
carriers, in this case holes. The antibody, m102.4. Together, these tion of a type of immune cell yields a product in an excited
traps help to prevent scattering results support further clinical called macrophages within the state that can emit light. The
but also release holes at higher development of hu1F5 for NiV lungs. The authors managed enzyme proficiency and active
temperatures to maintain a con- infection. —CM to induce tumor-associated site structure are important for
stant carrier capacity. The result Sci. Transl. Med. (2024) macrophages by elevated determining the amount of light
is a thermoelectric material with 10.1126/scitranslmed.adl2055 secretion of surfactant and produced and the wavelength
S
ocial play between parents and ion channels, which trans-
offspring well into adulthood duce force into an electrical
is beneficial for learning and response. However, the factors
cognitive and social develop- that regulate these channels
ment in humans. Our close and their contributions to
primate relatives, wild chimpanzees, mechanosensation are not
also engage in regular play, but may well known. Caldendrin, a
be constrained by ecological factors member of the superfamily of
such as food scarcity. Sabbi et al. calcium-binding proteins, is
examined more than 10 years of primarily expressed in neuronal
data on wild chimpanzee behavior in cell types and is particularly
Uganda to investigate whether costly abundant in sensory neurons
circumstances such as food scarcity in the inner ear and dorsal root
and poor diet constrain play. They ganglia. Lopez et al. used global
found that chimpanzee mothers went and conditional caldendrin
the extra mile for their children and knock-out mice, combined
maintained stable, consistent play with electrophysiological,
regardless of food scarcity or abun- behavioral, and biochemical
dance. However, other chimpanzee approaches, to show that cal-
adults reduced play and prioritized dendrin can inhibit PIEZO2, the
survival whenever food resources major mechanically activated
were scarce. Like human mothers, channel required for touch
chimpanzee mothers bear the hidden sensation. These results add
costs of motherhood. —EEU caldendrin to a growing list of
Curr. Biol. (2024) 10.1016/j.cub.2024.02.025 protein partners that could
serve as modulators of PIEZO2
Chimpanzees at play in Kibale Forest and potentially other mechani-
Reserve in Uganda cally activated channels. —PRS
J. Neurosci. (2024)
10.1523/JNEUROSCI.1402-23.2023
distribution. Colee et al. such a protocol for atoms in a control of such cells is essential
HYDROLOGY
performed high-throughput two-dimensional optical lattice to prevent side effects and
mutagenesis and screening that undergo a Mott insulator systemic toxicity. Xue et al. have Underground truth
on firefly luciferase to identify (MI)–superfluid transition. The worked out how to control anti- More than 20 years into a
single mutants with an emission protocol was able to distinguish tumor macrophages locally and severe drought, the Great Basin
peak shifted toward red. The between correlated particle- temporally in a mouse model. of the Southwestern US is losing
large scale of the screen pro- hole pairs that occur in a MI They designed macrophages groundwater at a rapid rate.
vided mechanistic insights that due to quantum fluctuations that produce interferon-g Hall et al. report measurements
may also be useful for future and holes that are a conse- (IFN-g), which converts the from the GRACE satellites
rational engineering. —MAF quence of finite temperature or macrophages into the antitumor showing that nearly 70 cubic
Biochemistry (2024) atom loss. Correcting for those M1 phenotype under the control kilometers of groundwater, or
10.1021/acs.biochem.3c00708 uncorrelated holes enabled the of a heat shock (heat-sensitive) more than six times the current
researchers to reliably measure promotor. Expression of IFN-g volume of water in Lake Mead of
QUANTUM SIMULATION a two-dimensional correla- is controlled with a wireless Arizona and Nevada, has been
tor that can serve as an order heating device on an animal depleted there over that time.
Erasing errors parameter for the MI state. —JS controlled from a smartphone or Groundwater replenishment
Error correction is a neces- Phys. Rev. X (2024) computer. In the model system, by snowfall, even in high-snow
sary component of quantum 10.1103/PhysRevX.14.011003 activation of macrophages years, has not been able to keep
computing and is used to decreased metastasis and up with this loss, which is likely
counter decoherence. In ana- increased survival. Although the the result of a combination of
log quantum simulation, for CELL ENGINEERING demonstration was in skin can- declining snow mass, upstream
example, using ultracold atoms cer, other technologies such as water diversions, and increased
in optical lattices, the effects of
Remote control ultrasound might allow deeper evaporation and sublimation
imperfections can also accu- anticancer macrophages tissue penetration into deep- due to rising air and ground
mulate but have seldom been Cells of the immune system can seated tumors. —LBR temperatures. —HJS
addressed with error-correcting be engineered into highly effec- Nat. Commun. (2024) Geophys. Res. Lett. (2024)
protocols. Hur et al. devised tive anticancer agents, but tight 10.1038/s41467-024-46210-1 10.1029/2023GL107913
A P. aeruginosa T4P disruption B Cryo-EM structures of PP7 and T4P C Retraction-dependent T4P detachment
20 Å Retraction
T4P
2 µm
PP7
+PP7 100 Å OM
Detached T4P
Pilus retraction
PP7 infection impairs P. aeruginosa motility. (A) Fluorescence images showing that T4P are detached upon PP7 infection. (B) Structures of PP7, T4P, and the
PP7/T4P complex are resolved with single-particle cryo–electron microscopy. (C) PP7 hijacks T4P retraction for infection and detaches T4P. (D) Cell twitching
motility is impaired owing to pilus detachment and nonfunctional T4P (the schematic diagram is not drawn to scale).
Fig. 1. Domain wall relaxation in the Heisenberg XXZ spin chain. (A) Schematic correspond to occupied and unoccupied sites, respectively, in random
of the unitary gate sequence used in this work, where fSim gates are applied instances of the experiment. The fSim angles were chosen to be (q, f) =
in a Floquet scheme on a 1D chain of NQ = 46 qubits. (B) Relaxation dynamics (0.4p, 0.8p), corresponding to D = 1. (C) Histogram showing the probability
as a function of site and cycle number for m = ∞, 0.9, and 0.3 for initially distribution of transferred magnetization after t = 1, 5, and 20 cycles
prepared domain-wall states with 2hSz i ¼ TtanhðmÞ. Blue and yellow squares [arrows in (B)] for m = ∞.
possesses a global SU(2) rotational sym- To study dynamics under the unitary evo- string is 111010 and the final bitstring is 110110,
metry. The spin dynamics in the Heisenberg lution (Eq. 2), we generated domain-wall initial then M is 2. Because the dynamics are number
model exhibit characteristics consistent with states with an adjustable contrast parameter m conserving, M is also the net number of zeros
the KPZ universality class, which was originally (Fig. 1B). Specifically, we initialized the chain in a that have crossed from the right to the left. Re-
introduced to describe the stochastic, nonlinear set of product states such that the left and right peating the experiment many times, we con-
dynamics of driven interfaces and has proven halves had average magnetization ±tanh(m), structed the probability distribution ofM,PðMÞ.
to apply to a wide range of classical systems respectively: In the case of m = 0, the initial state and the dy-
(10, 23–34). The KPZ-like behavior of the namics both have mirror symmetry, so for each
spin dynamics is surprising because of the ab- z
rðt ¼ 0Þºðe2mS Þ NQ =2 initial bitstring that was studied experimenally,
sence of stochasticity and nonlinearity in the z we also included the reflection of that bitstring
ðe 2mS Þ NQ =2 ð3Þ
Heisenberg model. in our analysis, using the same experimental
In a 1D chain of NQ = 46 superconducting data, which effectively symmetrized PðMÞ.
qubits, we simulated this spin model by pe- When m → ∞, the system approaches a pure Figure 1B shows measurement instances for
riodic (Floquet) application of high-fidelity domain-wall state with the two sides fully mag- three values of m. The left column in each panel
two-qubit unitary fSim(q,f) gates (Fig. 1A and netized in opposite directions. Only when m = 0, shows an instance of the initial state for the
fig. S7) (37, 38). Here, q sets the amplitude of the initial state is an infinite-temperature ther- given m, and the subsequent columns show
hopping between adjacent qubit lattice sites, mal state that preserves SU(2) symmetry. When typical bitstrings evolved from that state. As
and f is the conditional phase angle imparted m ≠ 0, the magnetization is preferentially along excitations (spin flips) propagate through
when two spin excitations are adjacent to each the z axis, breaking the SU(2) rotational sym- the chain, smaller domains become more prob-
other. Within each cycle, two-qubit fSim(q,ϕ) metry of the Heisenberg model. able. In Fig. 1C, we show histograms of M at
gates are applied between all neighboring pairs A natural measure of spin transport is the different times, starting in a pure (m = ∞) domain
in the chain, resulting in the cycle unitary: total transferred magnetization, Mðt Þ, defined wall state. Owing to the locality of the circuit,
as twice the net number of excitations that jMðt Þj is upper bounded by 2t. Consequently,
Y Y
UF ¼ fSimðq; fÞ fSimðq; fÞ have crossed the middle of the chain after the distribution is narrow and centered around
even bonds odd bonds t cycles. In our experiment, we sampled over a small value at t = 1 because only a few ex-
ð2Þ initial bitstring states with probabilities given citations have crossed the middle of the chain
by Eq. 3. For each initial state, we prepared the and becomes wider at later times.
In the limit q; f → 0, UF is the Trotter-Suzuki qubits in that state and then applied t cycles of
expansion of the XXZ Hamiltonian (Eq. 1), with fSim gates. Let NR,1(b) be the number of exci- Mean and variance of transferred magnetization
D ¼ sinðf=2Þ=sinðqÞ. Away from this limit, there tations (“1s”) in the right half of bitstring b. In the context of spin transport, the first and
is no specific, time-independent Hamiltonian The transferred magnetization M is the sto- second (variance) moments of M have been
associated with UF, but Eqs. 1 and 2 still share chastic variable defined by extensively studied both theoretically and
symmetries and are both integrable by the experimentally (10, 15, 23–34, 43, 45). Taking
Bethe ansatz (39–43). The conjecture that the Mðt Þ=2 ¼ NR;1 ðbt Þ NR;1 ðbi Þ ð4Þ advantage of our tunable fSim gates, we ex-
late-time dynamics are described by KPZ applies plored how these two moments depend on
equally to the Floquet system (44), so we made where bi is the initial bitstring, sampled from the anisotropy parameter, D. Figure 2A shows
no attempt to be in the small-angle limit, instead Eq. 3, and bt is the associated final bitstring the mean of the transferred magnetization,
favoring large angles for faster dynamics. sampled at t. For example, if the initial bit- hMi, over time for values of D equal to 0.16
Fig. 3. Skewness and excess kurtosis of transferred magnetization. model (60). The red lines marked “Wei et al.” show S measured in (10)
Experimental data and noiseless simulation results are shown with squares and (for m = 1.5) and the 1s confidence interval. (Inset) Representation of different
lines, respectively. (A) Skewness of transferred magnetization distribution S as a ways of taking the late-time and small-m limits that are considered in this work.
function of t, for D = 1 and various m∈[0,1.5]. We symmetrized the m = 0 (C) Kurtosis Q of transferred magnetization. The horizontal lines represent
probability distribution, after which the skewness was exactly 0. (B) The same as the theoretical predictions from the same models shown in (B). The experimental
(A), but with the x axis rescaled as mt2/3 and excluding data points for which Q, averaged over cycles 16 to 23, is –0.05 ± 0.02. The kurtosis data does not
t < 8. The collapse of S under rescaling is explored by using the noiseless exhibit a collapse, as expected; the kurtosis cannot be a function of mtg for
simulation data in fig. S10 (38). Dashed horizontal gray lines indicate predictions any exponent g because, unlike the skewness, it has time dependence even
based on the KPZ universality class (TW GUE), a NLFH model (57), and a CLL when m = 0.
Higher-order moments
Next, we extracted the skewness S and kurtosis that S goes to zero. Figure 3C shows that for (fig. S10) (38), suggesting that S may be a func-
Q of PðMÞ, later cycles, the initial strong time dependence tion of mt 2=3. Indeed, after excluding the initial
a3 of Q weakens. By averaging over cycles 16 to transient behavior, S did appear to be a single-
S ¼ 3=2 ð6Þ 23, we obtained a kurtosis of –0.05 ± 0.02. valued function ofmt 2=3 (Fig. 3B and fig. S10) (38).
a2
Statistical error bars on the individual data KPZ has been conjectured to apply to high-
points are shown in fig. S8 (38). temperature thermal states at late times, cor-
a4 To test the KPZ universality conjecture, one responding to taking m → 0 first and then t → ∞
Q¼ 3 ð7Þ needs to study the infinite-time (t → ∞) and (44). In this case, PðMÞ should become the
a22
near-equilibrium (m → 0) limits. These limits are Baik-Rains distribution (55). However, this
where ak ¼ ðM hMiÞk is the kth moment. experimentally inaccessible. However, if there distribution is skewed (Table 1), whereas sym-
In Fig. 3A, we show the temporal dependence exists a function f(m,t) such that the moments metry dictates that S ¼ 0, which is consistent
of S for m ranging from 1.5 to 0.1. Consistent are functions of f(m,t), then one may be able to with the trend observed in the experimental
with (10), S goes up to about 0.25 for m > 1. extrapolate measured values at finite m and t to data (Fig. 3, A and B).
However, as m is reduced toward the infinite- these unattainable limits. We empirically found One might also search for KPZ universality
temperature equilibrium point, we observed that the zero crossing of S scales as t0 ∼ m 1:49 away from m = 0, corresponding to a different
order in taking these noncommuting limits. involve unanticipated subtleties. Our first result, 29. S. Gopalakrishnan, R. Vasseur, B. Ware, Proc. Natl. Acad. Sci.
When taking t → ∞ first, the appropriate prob- also observed by others, is the superdiffusive U.S.A. 116, 16250–16255 (2019).
30. J. De Nardis, S. Gopalakrishnan, E. Ilievski, R. Vasseur, Phys.
ability distribution to compare PðMÞ against transport characterized byhMi ∼ t 2=3 (Fig. 2A). Rev. Lett. 125, 070601 (2020).
is the Tracy Widom (TW) distribution (55), Although this anomalous diffusion is sugges- 31. M. Dupont, J. E. Moore, Phys. Rev. B 101, 121106 (2020).
which has S of about 0.22. This order of limits tive of the known KPZ universality classes, this 32. M. Dupont, N. E. Sherman, J. E. Moore, Phys. Rev. Lett. 127,
107201 (2021).
corresponds to large mt 2=3 in Fig. 3B, where we classification is not compatible with our sec- 33. E. Ilievski, J. De Nardis, S. Gopalakrishnan, R. Vasseur, B. Ware,
indeed found S consistent with this distri- ond finding, the vanishing of S and Q near Phys. Rev. X 11, 031023 (2021).
bution; averaging over the four rightmost equilibrium (Fig. 3C). Despite our findings’ 34. A. Scheie et al., Nat. Phys. 17, 726–730 (2021).
35. P. N. Jepsen et al., Nature 588, 403–407 (2020).
points in Fig. 3B, we found S ¼ 0:18 T 0:02, apparent consistency with the CLL model, a
36. W. Heisenberg, Eur. Phys. J. A 49, 619–636 (1928).
which was consistent to two standard devia- full understanding would require the devel- 37. C. Neill et al., Nature 594, 508–512 (2021).
tions with the TW Gaussian unitary ensemble opment of a systematic spacetime RG frame- 38. Materials and methods are available as supplementary materials.
(GUE) value, as well as with an earlier experi- work that could establish the origin of the 39. H. Bethe, Eur. Phys. J. A 71, 205–226 (1931).
40. J.-S. Caux, F. H. L. Essler, Phys. Rev. Lett. 110, 257203 (2013).
ment, which found S ¼ 0:33 T 0:08 (10). How- KPZ-like behavior starting from the micro- 41. M. Vanicat, L. Zadnik, T. Prosen, Phys. Rev. Lett. 121, 030606
ever, Q of the TW GUE distribution is 0.09, scopic dynamics of the Heisenberg model. (2018).
whereas we found Q ¼ 0:05 T 0:02. The em- Quantum processors have the potential to help 42. P. W. Claeys, J. Herzog-Arbeitman, A. Lamacraft, SciPost Phys.
12, 007 (2022).
ergence of KPZ dynamics in this order of limits with such RG studies [e.g., building on (64, 65)]. 43. M. Ljubotina, L. Zadnik, T. Prosen, Phys. Rev. Lett. 122, 150605
was further ruled out by numerical and theo- For example, our results suggest that quantum (2019).
retical predictions that the dynamics become entanglement (but not integrability) is irrele- 44. M. Ljubotina, M. Žnidarič, T. Prosen, Phys. Rev. Lett. 122,
210602 (2019).
diffusive (z = 2) at a late time that grows as vant in the space-time RG sense. 45. N. Keenan, N. F. Robertson, T. Murphy, S. Zhuk, J. Goold, NPJ
t ∼ 1=m3 as m → 0 (53, 56). Our observations are rooted in the interplay Quantum Inf. 9, 72 (2023).
One could consider taking the two limits of integrability, quantum fluctuations, and sym- 46. X. Zotos, Phys. Rev. Lett. 82, 1764–1767 (1999).
47. S. Sachdev, K. Damle, Phys. Rev. Lett. 78, 943–946 (1997).
simultaneously in a way that the dynamics metry and have proved to be challenging to
48. K. Damle, S. Sachdev, Phys. Rev. B Condens. Matter 57,
do not become diffusive, e.g., by holding mt 2=3 describe through an effective quantum field 8307–8339 (1998).
constant (fig. S18) (38). The correct distri- theory. Our observed discrepancies with KPZ 49. M. Prähofer, H. Spohn, J. Stat. Phys. 115, 255–279 (2004).
bution to compare against is TW GUE in this predictions suggest that the infinite-temperature 50. Ž. Krajnik, E. Ilievski, T. Prosen, SciPost Phys. 9, 038 (2020).
51. B. Ye, F. Machado, J. Kemp, R. B. Hutson, N. Y. Yao, Phys. Rev.
case as well. If we take the limit with mt 2=3 dynamics in the Heisenberg chain, if universal, Lett. 129, 230602 (2022).
fixed at a large value, then we would find belong to a dynamical universality class that 52. M. Kardar, G. Parisi, Y.-C. Zhang, Phys. Rev. Lett. 56, 889–892
S consistent with TW GUE, but the measured has yet to be discovered. (1986).
53. Ž. Krajnik, E. Ilievski, T. Prosen, Phys. Rev. Lett. 128, 090604
Q is still inconsistent with the TW GUE pre- (2022).
diction of 0.09 (Fig. 3C), ruling out KPZ dy- RE FERENCES AND NOTES 54. S. Gopalakrishnan, A. Morningstar, R. Vasseur, V. Khemani,
namics on the timescales accessible in the Phys. Rev. B 109, 024417 (2024).
1. L. P. Kadanoff, Physica A 163, 1–14 (1990).
55. M. Prähofer, H. Spohn, Phys. Rev. Lett. 84, 4882–4885 (2000).
experiment. Although it remains possible that 2. S. Sachdev, Quantum Phase Transitions (Cambridge University
56. S. Gopalakrishnan, R. Vasseur, Rep. Prog. Phys. 86, 036502
KPZ dynamics will emerge at much later times Press, 2011).
(2023).
3. G. Odor, “Universality,” in Nonequilibrium Lattice Systems:
(i.e., Q will increase to 0.09), we see neither Theoretical Foundations (World Scientific Publishing Co.,
57. J. De Nardis, S. Gopalakrishnan, R. Vasseur, Phys. Rev. Lett.
131, 197102 (2023).
evidence nor rationale for this. 2008).
58. V. B. Bulchandani, Phys. Rev. B 101, 041411 (2020).
An outstanding question is why only lower- 4. A. Polkovnikov, K. Sengupta, A. Silva, M. Vengalattore, Rev.
59. V. B. Bulchandani, D. A. Huse, S. Gopalakrishnan, Phys. Rev. B
Mod. Phys. 83, 863–883 (2011).
point observables, such as the mean and variance 5. C. Gogolin, J. Eisert, Rep. Prog. Phys. 79, 056001 (2016).
105, 214308 (2022).
60. Ž. Krajnik, J. Schmidt, E. Ilievski, T. Prosen, Phys. Rev. Lett.
of the transferred magnetization and the cor- 6. A. Sommer, M. Ku, G. Roati, M. W. Zwierlein, Nature 472,
132, 017101 (2024).
relator studied by (44), seem to behave con- 201–204 (2011).
61. A. Das, M. Kulkarni, H. Spohn, A. Dhar, Phys. Rev. E 100,
7. S. Erne, R. Bücker, T. Gasenzer, J. Berges, J. Schmiedmayer,
sistently with KPZ universality. Notably, other Nature 563, 225–229 (2018).
042116 (2019).
systems have been identified that exhibit simi- 62. D. Roy, A. Dhar, H. Spohn, M. Kulkarni, Phys. Rev. B 107,
8. M. K. Joshi et al., Science 376, 720–724 (2022).
L100413 (2023).
lar behavior. One such system is a nonlinear 9. C. Zu et al., Nature 597, 45–50 (2021).
63. T. Giamarchi, Quantum Physics in One Dimension (Oxford
10. D. Wei et al., Science 376, 716–720 (2022).
fluctuating hydrodynamic (NLFH) model with University Press, 2003).
11. Q. Fontaine et al., Nature 608, 687–691 (2022).
two coupled stochastic modes (57–59), which 12. J. Lux, J. Müller, A. Mitra, A. Rosch, Phys. Rev. A 89, 053608 64. G. Vidal, Phys. Rev. Lett. 101, 110501 (2008).
65. G. Evenbly, G. Vidal, Phys. Rev. B Condens. Matter Mater. Phys.
predicts S ¼ 0, consistent with the Heisenberg (2014).
13. V. B. Bulchandani, R. Vasseur, C. Karrasch, J. E. Moore, Phys. 79, 144108 (2009).
spin chain. However, it suggests Q ¼ 0:14 , 66. E. Rosenberg, T. I. Andersen, (2024). Zenodo. https://doi.org/
Rev. B 97, 045407 (2018).
which differs from what we observed, perhaps 14. O. A. Castro-Alvaredo, B. Doyon, T. Yoshimura, Phys. Rev. X 6, 10.5281/zenodo.10045873.
because not all aspects of the model are uni- 041065 (2016). 67. E. Rosenberg, Kardar-Parisi-Zhang Dynamics in the Heisenberg
15. B. Bertini, M. Collura, J. De Nardis, M. Fagotti, Phys. Rev. Lett. Spin Chain, ReCirq (2023); https://quantumai.google/cirq/
versal. Another such system is the classical experiments/kpz/kpz.
117, 207201 (2016).
Landau-Lifshitz (CLL) magnet (55, 60–62), 16. B. Doyon, T. Yoshimura, SciPost Phys. 2, 014 (2017).
which predicts S ¼ 0 and a Q that is negative 17. V. B. Bulchandani, R. Vasseur, C. Karrasch, J. E. Moore, Phys. AC KNOWLED GME NTS
and close to zero at these timescales (60). These Rev. Lett. 119, 220604 (2017). We acknowledge discussions with I. Bloch, V. B. Bulchandani,
18. J. De Nardis, D. Bernard, B. Doyon, Phys. Rev. Lett. 121, A. Morningstar, and R. Vasseur. Funding: V.K. acknowledges support
are consistent with our experimental results. It 160603 (2018). from the US Department of Energy, Office of Science, Basic Energy
is noteworthy that this classical system addi- 19. K. Popper, Conjectures and Refutations: The Growth of Sciences, under Early Career Award no. DE-SC0021111; the Alfred
tionally exhibits dynamical behavior similar to Scientific Knowledge (Routledge Classics, 2002). P. Sloan Foundation through a Sloan Research Fellowship; and the
20. F. Arute et al., Nature 574, 505–510 (2019). Packard Foundation through a Packard Fellowship in Science and
that of the quantum spin chain with enhanced 21. Y. Wu et al., Phys. Rev. Lett. 127, 180501 (2021). Engineering. S.G., V.K., and T.P. acknowledge the hospitality of the
quantum fluctuations due to confinement (63). 22. M. Znidarič, Phys. Rev. Lett. 106, 220601 (2011). Kavli Institute for Theoretical Physics at the University of California,
23. M. Ljubotina, M. Žnidarič, T. Prosen, Nat. Commun. 8, 16117 (2017). Santa Barbara (supported by NSF grant PHY-1748958). R.S. is
Discussion and outlook 24. J.-M. Stéphan, J. Stat. Mech. 2017, 103108 (2017). supported by the Princeton Quantum Initiative Fellowship. T.P. is
25. R. J. Sánchez, V. K. Varma, V. Oganesyan, Phys. Rev. B 98, supported by program P1-0402 of the Slovenian Research Agency
Studies of the universal aspects of quantum dy- 054415 (2018). (ARRS). Author contributions: T.P., S.G., and V.K. proposed the
namics have attracted substantial interest re- 26. E. Ilievski, J. De Nardis, M. Medenjak, T. Prosen, Phys. Rev. experiment and helped guide and interpret it. P.R. selected the
Lett. 121, 230602 (2018). proposal and advised the experimental effort. E.R. implemented the
cently; accordingly, a complete classification
27. S. Gopalakrishnan, R. Vasseur, Phys. Rev. Lett. 122, 127202 (2019). experiment. E.R. and T.I.A. collected the experimental data. E.R.
of their universal properties is lacking. Our 28. J. De Nardis, M. Medenjak, C. Karrasch, E. Ilievski, Phys. Rev. developed and ran the numerical simulations. R. Samajdar provided
findings suggest that these classifications could Lett. 123, 186601 (2019). theoretical input and implemented additional numerics. P.R., T.I.A.,
E.R., R.S., T.P., and S.G. contributed to writing the manuscript. All Association for the Advancement of Science. No claim to original US Supplementary Text
authors contributed to the experimental and theoretical infrastructure government works. https://www.science.org/about/science-licenses- Figs. S1 to S18
to enable the experiment. C.N., X.M., A.M., and J.H. helped develop journal-article-reuse Tables S1 to S3
fSim gates. Competing interests: The authors declare no competing References (68–90)
interests. Data and materials availability: The data that support
the findings in this study are available on Zenodo (66). Code is SUPPLEMENTARY MATERIALS
available on ReCirq (67). License information: Copyright © 2024 the science.org/doi/10.1126/science.adi7877 Submitted 18 May 2023; accepted 1 March 2024
authors, some rights reserved; exclusive licensee American Materials and Methods 10.1126/science.adi7877
A
vive for an organism’s entire life without brain tissue. bility in the tissue (fig. S5, A and B). Consistent
ever being replaced. In the absence of mi- with EU being specifically incorporated into
totic nuclear disassembly and reassembly, Long-term retention of nuclear RNAs in the RNA, degradation of genomic DNA by deoxy-
the life span of some nuclear constitu- mouse brain ribonuclease I (DNase 1) treatment did not
ents such as genomic DNA, a subset of histones, To explore RNA stability in the mouse brain, affect EU signal intensities (fig. S5, C to H).
and nuclear pore complex proteins extends to we performed in vivo pulse-chase labeling of Moreover, when EU was injected at the same
months and even years (1, 2). For example, the transcripts with a modified uridine analog, time as bromodeoxyuridine (BrdU), which is
stability of genomic DNA has been used to de- 5-ethynyl uridine (EU) (7). EU was injected incorporated into newly synthesized DNA, the
termine the chronological age of neurons in into mice on postnatal days 3 to 5 (P3 to P5) to spatial distribution of nuclear EU signals was
humans over the course of decades (3, 4). The label de novo–synthesized RNAs during brain distinct from BrdU signals (fig. S5I), provid-
lifelong stability of genomic DNA is necessary development (Fig. 1A). One day after the last ing further evidence for the specific incorpora-
to safeguard the persistence of a cell’s genetic injection of EU (defined as the 1-week sam- tion of EU into RNA. We also studied animals
information. In recent years, noncoding RNAs ple), EU-labeled RNA was visualized by click 2 years after EU injections and were able to
have been identified as critical regulators of chemistry using Alexa-555 fluorophores. EU detect EU signals in DG and CB neurons
nuclear chromatin organization and transcrip- signals were detectable in different brain re- (Fig. 1L; and fig. S6, A, B, and D to F), radial
tion control (5, 6); however, little is known about gions such as the dentate gyrus (DG) of the hip- glia–like adult neural stem cells (RGL-ANSCs)
pocampus (Fig. 1B) and the cerebellum (CB) and cerebellar ANSCs (Fig. 1M and fig. S6H),
1
Nuclear Architecture in Neural Plasticity and Aging, German (Fig. 1C). The EU signal was sparse and not de- and astrocytes (fig. S6, C, E, and G). This find-
Center for Neurodegenerative Diseases (DZNE), Dresden 01307, tectable in most cells in the primary somato- ing suggests that these RNAs persist through-
Germany. 2Molecular and Cell Biology Laboratory, The Salk
Institute for Biological Studies, La Jolla, CA 92037, USA. 3Kura sensory cortex (S1) (fig. S1). The density and out the life span of the animal.
Oncology, Inc., San Diego, CA 92121, USA. 4DRESDEN-concept fraction of EU+ cells varied among different
Genome Center, Technology Platform at the Center for brain tissues [analysis of variance (ANOVA) Cell type–specificity of long-retained RNAs
Molecular and Cellular Bioengineering (CMCB), Technische
Universität Dresden, Fetscherstr. 105, Dresden 01307, Germany.
P < 0.0001; Fig. 1, J and K]. We confirmed To identify which cell types retained LL-RNA,
5
German Center for Diabetes Research (DZD e.V.), 85764 that all cells in the brain were EU-labeled acute- we combined EU click chemistry with immuno-
Neuherberg, Germany. 6Paul Langerhans Institute Dresden of ly after a single EU injection on P3 (30-min histochemistry using cell type–specific markers.
the Helmholtz Center Munich, University Hospital and Faculty of
Medicine Carl Gustav Carus, Technische Universität Dresden,
or 1-hour chase) (figs. S2 and S3) and detected In 1-year-old mice, most EU+ cells (83.7 ± 10.8%)
01307 Dresden, Germany. 7Laboratory of Genetics, The Salk no difference in EU incorporation into de novo– in the DG were NeuN+ neurons (Fig. 1, D and F).
Institute for Biological Studies, La Jolla, CA 92037, USA.
8
synthesized RNA between brain regions (fig. EU signals were also observed in RGL-ANSCs
Institute of Science and Technology Austria (ISTA), 3400
S3C), which indicates that EU can be trans- [Sox2+ with a glial fibrillary acidic protein–
Klosterneuburg, Austria. 9Laboratory of Neural Epigenomics,
Institute of Medical Physics and Micro-tissue Engineering, ported, metabolized, and used in transcription positive (GFAP+) radial fiber; 1.5 ± 0.9% of
Faculty of Medicine, Friedrich-Alexander-Universität Erlangen- in all cells of the brain at this age. These data EU+ cells], adult neural progenitor cells (ANPCs)
Nürnberg, Erlangen 91054, Germany. suggest that either the transcription or the re- (Sox2+ without a GFAP+ radial fiber; 3.76 ±
*Corresponding author. Email: tomohisa.toda@fau.de (T.T.);
martin.hetzer@ist.ac.at (M.W.H.) tention of EU-incorporated transcripts varies 1.93% of EU+ cells), and GFAP+ astrocytes
†These authors contributed equally to this work. between different cell types in the brain. (Fig. 1, D and H), suggesting that LL-RNAs
CB 1 week
DG 1 week
CB 1 year
DG 1 year
CB 1 year
H EU Sox2 Sox2
EU/Sox2/GFAP I EU Sox2 Hoechst EU Sox2 GFAP
GFAP GFAP Hoechst
DG 1 year
CB 1 year
EU
J K * L
****
**** ****
DG 2 years
NS **
Fraction of EU labeled
NS
neurons (%)
NS EU EU
Hoechst NeuN
M EU Hoechst EU
Sox2
DG 2 years
NS NS GFAP
Fig. 1. Long-term retention of RNA in the mouse brain. (A) Schematic diagram 4156 ± 542 cells/mm2, 1Y CB 4816 ± 537 cells/mm2, and 1Y S1 359 ± 199 cells/mm2.
of the experimental plan. RNA synthesis at P3 to P5 was labeled using EU. Data are presented as mean ±SD. Significance was determined by ANOVA with
(B) EU signals in the DG of a 1-week-old animal. (C) EU signals in the CB of a 1-week-old post hoc Tukey’s test. (K) Quantification of the percentage of EU+ neurons
animal. (D) EU signals in the DG of a 1-year-old animal. (E) EU signals in the (NeuN+ cells). Data are presented as mean ±SD. Significance was determined by
CB of a 1-year-old animal. (F and G) High-magnification images with NeuN and GFAP ANOVA followed by Tukey’s multiple comparison test. (L) EU signals in neurons
staining (arrowheads, NeuN+ neurons) in the DG (F) and CB (G) of a 1-year-old of the DG of a 2-year-old animal. (M) EU signals in a neural stem cell (RGL-
animal. (H and I) Sox2 and GFAP immunostaining (arrowhead, RGL-ANSC; open ANSC) of the DG of a 2-year-old animal. Scale bars are 100 mm [(B) and (D)],
arrowhead, astrocytes) in the DG (H) and CB (I) of a 1-year-old animal. (J) Quantification 10 mm [(F) to (I), (L), and (M)], and 200 mm [(C) and (E)]. Images in (B) to
of the density of EU+ cells: 1-week-old (1W) DG 3933 ± 911.9 cells/mm2, 1W CB (I), (L), and (M) are from confocal microscopy. *P < 0.05, **P < 0.01, **P < 0.001,
6060 ± 774.3 cells/mm2, and 1W S1 291.8 ± 52.12 cells/mm2; and 1-year-old (1Y) DG ****P < 0.0001, and NS is not significant.
can be maintained in neurons and somatic Because most cortical neurons are gener- tocol to induce quiescence in mouse neural pro-
ANSCs in the DG. In 1-week-old mice (1 day ated during embryonic neurogenesis, we next genitor cells (NPCs) (12) and found that quiNPCs
after the last EU injection), only 25.6 ± 9.4% asked whether cortical neurons retain EU sig- indeed retained EU+ transcripts for 8 days
of EU+ cells were NeuN+, whereas 39.7 ± 3.8% nals when EU is injected during cortical neuro- (Fig. 2, A and B) and even 2 weeks (fig. S13),
were Sox2+ neural stem and progenitor cells genesis. We injected EU into pregnant dams but not in proliferating conditions (Fig. 2, A
(NSPCs) and 7.0 ± 1.5% were RGL-ANSCs (fig. at embryonic days 14.5 to 16.5 (E14.5 to E16.5) and B). EU signals in quiNPCs were depleted
S1I). In the first and second postnatal weeks, and analyzed brains at E17.5 and at P6 (fig. by RNase treatments (fig. S14, A and B) and
ANPCs in the DG give rise to neurons or be- S9A). We confirmed that EU injection labeled significantly reduced after inhibition of RNA
come quiescent RGL-ANSCs (8, 9), suggest- RNA synthesis in all embryonic brain cells at polymerase II and III (fig. S14C), confirming
ing that the retention and incorporation of 1 hour after injection (fig. S9B). At E17.5, EU that EU signals are derived from nascent RNA
EU+ transcripts could coincide with cell cycle signals were retained in Sox2+ and Tbr2+ NSPCs in quiNPCs. Inhibition of RNA polymerases
exit during differentiation or transition into in the ventricular and subventricular zone of after EU labeling did not reduce EU signals
quiescence. Indeed, 4 hours after a single EU the neocortex (fig. S9, B to E and G). By con- (fig. S14, D and E), suggesting that EU signals
injection, when most acutely labeled RNA was trast, we found only very few and weakly labeled are not derived from the recycling of EU-labeled
already degraded (figs. S2 and S3), all high- EU+ cells in the cortical plate—fewer cells than uridines. Moreover, DNase 1 treatment and ribo-
intensity EU+ cells in the hippocampus were the expected number of newborn neurons based nucleotide reductase inhibition did not reduce
proliferating (Ki67+) NSPCs, whereas 24 hours on the distribution of BrdU+ newborn cells at EU signals in quiNPCs (fig. S15), providing fur-
after the last injection, most EU+ cells had E17.5 (fig. S9, E and F). Of those rare EU+ cells ther evidence that EU signals are not derived
exited the cell cycle and 44.5% of EU+ cells in the cortical plate, only 39.0 ± 3.2 and 5.1 ± from EU incorporation into DNA.
expressed neuronal differentiation-associated 3.2% expressed neuronal markers NeuN and
marker NeuroD1 (fig. S7). Thus, EU-labeled Ctip2, respectively (fig. S9, E and H). Thus, Molecular identity of LL-RNAs
RNA is retained long-term upon cell cycle exit cortical neurons have a limited capability to To determine the molecular identity of LL-RNAs,
in neurons and ANSCs in the DG. transcribe or retain LL-RNAs. Moreover, only we performed EU-RNA-sequencing (EU-RNA-
Similar to the DG, in the CB of 1-week-old sparse EU signals were detected in the cortex, seq) of quiNPCs 1 week after EU labeling. We
mice, 25.7 ± 15.4% of EU+ cells were NeuN+ hippocampus, and CB at P6 when EU was in- found 1575 genes to be significantly enriched
neurons (fig. S1J) and 13.3 ± 4.4% of EU+ cells jected embryonically (fig. S9I). Thus, the long- in the EU-labeled RNA fraction (Fig. 2C, fig.
were Sox2+ adult NSPCs (10), indicating that term retention of RNA is cell type–specific, with S16A, and data S1), and identified those as LL-
EU+ transcripts are retained in neurons and cortical neurons showing a much-reduced capa- RNAs in quiNPCs. The LL-RNAs comprised
stem cells of CB. Fifty-eight percent of EU+ cells bility to maintain RNAs compared with granule protein-coding RNAs and noncoding RNAs
in the CB did not express any tested markers cells of the DG and CB. (Fig. 2D), with long noncoding RNAs (lncRNAs)
at this age, but they were likely to be migrating We then addressed whether LL-RNAs can be and uncharacterized transcripts (TECs; to be
immature granule cells on the basis of their transcribed in the brain at any age. EU injec- experimentally confirmed) significantly over-
localization (Fig. 1C) (11). Indeed, in 1-year-old tions at later postnatal ages (P7 to P9 and P13 represented in EU-enriched RNAs compared
mice, 76.3 ± 8.5% of EU+ cells were NeuN+ to P15) resulted in patterns of EU+ cells in the with all RNAs expressed in quiNPCs (Fig. 2E).
neurons mostly located in the granular layer hippocampus that were consistent with those LL-RNAs in quiNPCs were enriched in path-
(Fig. 1, E, G, and K and fig. S1, E and J). We also obtained after EU injections at P3 to P5 (fig. ways related to nuclear architecture and epi-
observed the retention of LL-RNA in Sox2+ S10). However, in the adult mouse brain, EU genetic regulation, suggesting potential roles
ANSPCs in the CB of 1-year-old mice (Fig. 1I), was not incorporated into nascent transcripts of LL-RNAs in these cellular functions (Fig. 2F).
which suggests that ANSPCs in both the DG (fig. S11) and was, therefore, not suitable to label We also performed EU-RNA-seq with hippo-
and CB can retain LL-RNAs. In the S1, a total RNA synthesis. This is presumably due to the campal tissue from 1-month-old mice that were
of 34.4 ± 14.6 and 43.0 ± 10.3% of EU+ cells down-regulation of enzymes of the pyrimidine injected with EU at P3 to P5, and we identified
were GFAP+ in 1-week-old and 1-year-old sam- salvage pathway in the adult brain (fig. S12) 16 gene-derived transcripts that were signifi-
ples, respectively (fig. S1, F to H and K), which that are required for EU metabolism. No brain cantly enriched in the EU-labeled RNA frac-
indicates that astrocytes in the S1 retain EU- region–dependent or cell type–specific dif- tion, including lncRNAs, mitochrondrial tRNAs,
labeled transcripts. However, NeuN+ EU+ cells ferences in the expression of the pyrimidine and protein-coding RNAs (Fig. 2, G and H).
were rarely observed in the S1 (Fig. 1K). Thus, LL- salvage pathway were detected in postnatal Mapping of EU-RNA-seq data to repetitive
RNAs are retained or transcribed in a cell type– mice (fig. S12, A to C), consistent with the genomic regions revealed that LL-RNAs con-
specific manner. ubiquitous EU signal that was observed acute- tained repetitive RNAs in both hippocampus
To validate that our findings were indepen- ly after EU injection in postnatal brains (figs. and quiNPCs (Fig. 2, I and J; and fig. S16, B
dent of potential side effects caused by EU, S2 and S3). This finding supports the notion and C), including small nuclear RNA repeats
we labeled nascent RNA using another uridine that all cells in the postnatal brain can me- (snRNAs), short interspersed nuclear elements
analog, 5-bromouridine (BrU). The spatial pat- tabolize EU but that the retention of LL-RNAs (SINEs), and satellite RNAs (satRNAs) (fig.
terns of BrU+ cells in 1-week-old and 1-month- is cell type–specific. S16D). Together, these results suggest that non-
old mice reflected those observed in EU-injected Having established the existence of nuclear coding RNAs, including lncRNAs and repeti-
mice, with persistent BrU labeling in neurons LL-RNAs, we next aimed to determine their tive RNAs, can be retained long-term in quiNPCs
and ANSPCs in the DG and CB but only very molecular identity and functions. To facilitate and in hippocampal granule cells. The differ-
few BrU+ cells in the S1 (fig. S8). Additionally, functional characterization, we used a tracta- ence in the detected numbers of LL-RNAs be-
injection of triphosphorylated uridine (5-BrUTP), ble in vitro system in which we could recapitu- tween quiNPCs and hippocampal tissue might
which is directly incorporated into RNA with- late the long-term retention of transcripts and result from differences in the length of the re-
out being metabolized, resulted in a similar manipulate the identified RNAs. Because we tention period or from cell type–specific mo-
distribution of label-retaining cells in the brain, observed EU signal retention in RGL-ANSCs lecular identities of LL-RNAs. Alternatively,
providing further support that cell type–specific in vivo, we tested whether quiescent neural pro- the detection sensitivity of LL-RNAs may be
retention of RNA is not due to potential differ- genitor cells (quiNPCs) retained EU-labeled reduced in hippocampal tissue because only a
ences in cellular nucleotide metabolism. transcripts in vitro. We used an established pro- fraction of hippocampal cells retained EU-labeled
EU intensity (AU)
NS
(gene-derived)
-log10(adjusted p-value)
EU
DAPI
Fig. 2. Molecular identity of long-retained RNA. (A) Confocal microscopy images RNAs were underrepresented, whereas lncRNAs and uncharacterized
showing the retention of EU signals in quiNPCs in vitro. Scale bar is 10 mm. transcripts (TECs) were overrepresented in EU-enriched RNA. Significance
DAPI, 4′,6-diamidino-2-phenylindole; proNPC, proliferating NPC. (B) Quantifica- was determined by linear regression. (F) Top significantly enriched Reactome
tion of EU intensity in proNPCs and quiNPCs at day 1 (D1) or D8 after EU labeling. pathways among EU-enriched RNAs in quiNPCs (adjusted P < 0.05). (G) MA
Dots indicate individual cells. Significance was determined by Kruskal-Wallis plot showing EU-enriched (gene-derived) RNAs in hippocampus tissue of
test followed by Dunn’s multiple comparison test. AU, arbitrary units. (C) MA plot 5-week-old mice that were injected with EU at P3 to P5 (n = 3 mice per
showing EU-enriched (gene-derived) RNAs in quiNPCs at 8 days after EU group). (H) Gene-class distribution of EU-enriched RNA in the hippocampus.
labeling, as determined by EU-RNA-seq. Significantly EU-enriched RNAs were (I and J) Distribution of EU-enriched RNAs that map to repeat RNAs in
defined as adjusted P < 0.05 and fold change >2 compared with non–EU-labeled hippocampus tissue (I) and quiNPCs (J). In (C) and (G), the dashed-dotted
quiNPCs (n = 3 experiments). (D) Gene class distribution of EU-enriched RNA line indicates significance threshold of adjusted P = 0.05. ***P < 0.001, ****P <
in quiNPCs. miRNA, microRNA; mtRNA, mitochondrial RNA. (E) Protein-coding 0.0001, and NS is not significant.
transcripts, whereas 100% of in vitro quiNPCs matin (13–16). Consistent with this idea, high- regulation. We confirmed the enrichment of
possessed EU-labeled RNAs. In the future, it resolution imaging in DG and CB neurons from repeat RNAs among EU-captured RNA using
would be important to compare LL-RNAs of 1-year-old samples showed that EU signals quantitative polymerase chain reaction (qPCR)
different cell types under similar conditions. were often associated with Hoechst-dense het- (Fig. 3C) and found that a significant amount
Several studies suggest that repeat RNAs, erochromatin areas (Fig. 3, A and B), indicating of satRNAs were retained for up to 1 year in
including satRNAs, interact with heterochro- that they might have functions in chromatin the hippocampus (Fig. 3D). The PCR signals
Cerebellum
DG
quiNPCs
** ** *
HC/CX
**
HC
* * ** **
* ***
** **
sgRNA Control
LNA Control
sgRNA MajSat
LNA Majsat
Fig. 3. Repeat RNAs are long retained in the mouse brain. (A and B) Localization (E) Enrichment of EU-labeled transcripts in quiNPCs 8 days after EU treatment
of EU signals in nuclei in the DG (A) and CB (B) of a 1-year-old animal. Scale bars are compared with that in PBS-treated cells. Significance was determined by one-sample
5 mm. (C and D) Enrichment of EU-labeled transcripts in the hippocampus/cortex t test. Fold enrichments compared to control are as follows: Gapdh 3.53 ± 1.50, 18S
(HC/CX) of a 1-week-old animal (n = 5 animals) (C) and in the hippocampus (HC) of rRNA 6.86 ± 2.06, 28S rRNA 7.38 ± 3.33, major satellite 49.3 ±19.1, minor satellite
24-week-old to 1-year-old animals (n = 4 animals) (D). Significance was determined 52.57 ± 20.97, bactin 22.36 ± 10.2, SINEB1 40.45 ± 25.53, and LINE-1 59.82 ± 47.95
by one-sample t test. Fold enrichments for qPCR signals of EU-labeled mice (n = 6 experiments). Data are presented as mean ±SD. (F) Quantification of
compared with PBS-treated control are as follows: For (C) at 1W, glyceraldehyde-3- EU signal intensity after LNA-GapmeR–mediated knock down of major satRNAs in
phosphate dehydrogenase (Gapdh) 4.95 ± 1.85, 18S rRNA 3.76 ± 1.43, 28S rRNA quiNPCs (LNA MajSat). Dots indicate individual cells from three experiments.
2.84 ± 1.13, major satellite 7.69 ± 4.16, minor satellite 4.48 ± 2.71, bactin 4.73 ± 2.54, Significance was determined by t test. (G) Reduction of EU signals in quiNPCs after the
SINEB1 1.95 ± 0.60, and LINE-1 2.47 ± 0.62; and for (D) at 24W to 1Y, Gapdh 1.85 ± application of LNA targeting major satRNAs. Scale bar is 20 mm. (H and I) Increased
0.66, 18S rRNA 2.38 ± 1.68, 28S rRNA 1.21 ± 0.63, major satellite 4.57 ± 1.55, minor EU signals after CRISPRa-based overexpression of major satRNAs (sgRNA MajSat).
satellite 4.54 ± 2.37, bactin 1.60 ± 0.68, SINEB1 1.44 ± 0.70, and LINE-1 1.90 ± 1.34. Dots indicate individual cells from three experiments. Significance was determined by
Data are presented as mean ±SD. The red dashed lines indicate the normalized t test. Scale bar is 20 mm. Images in (A), (B), (G), and (I) are from confocal microscopy.
value from PBS-treated control. MajSat, major satRNA; MinSat, minor satRNA. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.
were diminished by RNase treatments (fig. S14, and minor satRNAs in quiNPCs at 8 days (SINEB1) and long interspersed nuclear ele-
F to H), supporting the notion that satRNAs after EU labeling compared with vehicle-treated ments (LINE-1) were also enriched in quiNPCs.
are retained for years in the brain. We also quiNPCs (Fig. 3E). Other repeat sequences We also identified 18S and 28S ribosomal RNA
found strong enrichments of major satRNAs such as short interspersed nuclear element B1 (rRNA) to be moderately enriched, consistent
H3K9Me3
H3K9Me3
sgRNA MajSat sgRNA MajSat
DAPI
H3K9Me3
E H3K9me3-ChIP F H3K9me3-ChIP G CRISPRa
sgRNA MinSat sgRNA MinSat
Major satellites Minor satellites H3K9me3
H3K9Me3
* * Normalized intensity
Fold enrichment
Fold enrichment
J K L N
DAPI Ki67
* ** **
*
% BrdU/DAPI
% H2AX/DAPI
DAPI BrdU
Fig. 4. Major satRNAs are essential for maintenance of NPC function. (A) Distribution of
M LNA Control LNA Majsat
H3K9me3 after LNA-mediated knockdown of major satRNAs in quiNPCs. Scale bar is 5 mm.
(B) Quantification of H3K9me3 intensity after knockdown of major satRNAs. Dots indicate
H3K9me3
individual cells from three experiments. Significance was determined by t test. (C) Distribution
of H3K9me3 after CRISPR-mediated transcriptional inhibition (CRISPRi) or overexpression
(CRISPRa) of major and minor satRNAs. Scale bars are 5 mm. (D) Transcriptional inhibition of
major satRNAs using CRISPRi (dCas9-KRAB with sgRNAs) impaired heterochromatin retention
in quiNPCs; minor satRNAs were dispensable. Dots represent individual nuclei from three
H2AX
independent experiments; group means are indicated. Significance was determined by t test.
(E and F) LNA-mediated knockdown of major satRNAs reduced H3K9me3 abundance at
major and minor satellite repeats. Shown are data from H3K9me3-ChIP-qPCR. Data are presented
as mean ±SD. Significance was determined by one-sided t test (n = 4 experiments). (G) Up-regulation
of satRNAs increased nuclear H3K9me3 in quiNPCs. Dots indicate individual cells from three independent experiments; group means are indicated. Significance was
determined by t test. (H) Experimental timeline for data shown in (I) to (N). (I) Cell cycle reentry of quiNPCs after LNA-mediated knockdown of major satRNAs.
Arrows indicate pyknotic cells. Scale bar is 25 mm. (J and K) Decreased percentage of Ki67+ cells (J) and BrdU+ cells (K) 1 day after reactivation of quiNPCs into
proliferation. Significance was determined by paired t test (n = 5 to 8 experiments). For Ki67, data are as follows: LNA control 23.1 ± 2.73% and LNA Majsat 11.8 ± 4.16%;
for BrdU, data are as follows: LNA control 29.6 ±12.35% and LNA Majsat 19.65 ± 13.64%. (L) Increased percentage of pyknotic cells after knockdown of major satRNAs.
Data are as follows: LNA control 13.4 ± 2.55% and LNA Majsat 22.76 ± 8.11%. (M and N) Increased percentage of gH2AX+ cells after knockdown of major satRNAs.
Significance was determined by ratio paired t test (n = 4 experiments). Data are as follows: LNA control 8.57 ± 3.25% and LNA Majsat 30.9 ± 6.66%. Scale bar
in (M) is 5 mm. Images in (A), (C), (I), and (M) are from confocal microscopy. *P < 0.05, **P < 0.01, ****P < 0.0001, and NS is not significant.
with previous findings of rRNAs exhibiting chromatin is involved in LL-RNA regulation. tenance of nuclear function in postmitotic cells
longer half-lives (17). We validated that EU Heterochromatin is essential to repress aber- and somatic stem cells involves the substantial
treatment did not alter repeat RNA levels (fig. rant transcription of satRNAs, and increased extension of the life span of key molecular con-
S16E) and confirmed the long-term retention satRNA expression is a hallmark of impaired stituents, including LL-RNAs.
of repeat RNAs in quiNPCs using BrU labeling epigenetic regulation in tumorigenesis (21, 22).
and BrU immunoprecipitation chase (BRIC)– Thus, the balanced level of LL-RNAs may be
REFERENCES AND NOTES
qPCR (fig. S17). Altogether, our findings re- regulated by heterochromatin in cooperation
1. J. N. Savas, B. H. Toyama, T. Xu, J. R. Yates III, M. W. Hetzer,
veal transcript-specific long-term retention of with LL-RNAs themselves, yet its specific roles Science 335, 942 (2012).
repeat-derived RNAs in quiNPCs. in LL-RNA transcription and/or retention re- 2. B. H. Toyama et al., Cell 154, 971–982 (2013).
quire further investigation. 3. P. S. Eriksson et al., Nat. Med. 4, 1313–1317 (1998).
Roles of LL-RNAs in heterochromatin maintenance 4. K. L. Spalding et al., Cell 153, 1219–1227 (2013).
We next tested whether major satRNAs 5. P. Bhat, D. Honson, M. Guttman, Nat. Rev. Mol. Cell Biol. 22,
Finally, we addressed the biological function are important for maintaining quiNPCs. We 653–670 (2021).
of LL-RNAs using quiNPCs as a model system knocked down major satRNAs using LNA- 6. A. Loda, S. Collombet, E. Heard, Nat. Rev. Mol. Cell Biol. 23,
231–249 (2022).
and manipulated major satRNAs using previ- Majsat GapmeRs upon induction of quiescence 7. C. Y. Jao, A. Salic, Proc. Natl. Acad. Sci. U.S.A. 105, 15779–15784 (2008).
ously published LNA-GapmeRs (18) and single and tested whether quiNPCs could be reacti- 8. D. A. Berg et al., Cell 177, 654–668.e15 (2019).
guide RNAs (sgRNAs) (CRISPRa and CRISPRi) vated into proliferation (Fig. 4H). One day 9. Z. Nicola, K. Fabel, G. Kempermann, Front. Neuroanat. 9, 53 (2015).
10. J. Ahlfeld et al., Sci. Rep. 7, 6137 (2017).
(LNA, locked nucleic acid) (19). We observed after reactivation of quiNPCs, the fraction of 11. E. Yacubova, H. Komuro, Cell Biochem. Biophys. 37, 213–234 (2003).
a significant reduction in EU signals after major proliferating cells (Ki67+ cells and BrdU+ cells) 12. B. Martynoga et al., Genes Dev. 27, 1769–1786 (2013).
satRNA knock down (Fig. 3, F and G; and fig. was significantly reduced in LNA-Majsat– 13. X. Huo et al., Mol. Cell 77, 368–383.e7 (2020).
14. O. Velazquez Camacho et al., eLife 6, e25293 (2017).
S18A), whereas overexpression of major satRNAs treated cells (Fig. 4, I to K), suggesting atten- 15. A. Shirai et al., eLife 6, e25317 (2017).
increased EU signals (Fig. 3, H and I; and fig. uated reactivation of quiNPCs. Furthermore, 16. C. L. Novo et al., Nat. Commun. 13, 3525 (2022).
S18B), confirming the contribution of major significantly more pyknotic nuclei were ob- 17. H. T. Abelson, L. F. Johnson, S. Penman, H. Green, Cell 1, 161–165 (1974).
18. A. V. Probst et al., Dev. Cell 19, 625–638 (2010).
satRNAs to LL-RNAs in quiNPCs. Because it has served in LNA-Majsat–treated cells (Fig. 4L).
19. Q. Zhu et al., Mol. Cell 70, 842–853.e7 (2018).
been shown that major satRNAs are associated Consistent with this observation, the fraction 20. C. Maison et al., Nat. Genet. 43, 220–227 (2011).
with constitutive heterochromatin (14, 15), we of gH2AX+ cells was significantly higher in 21. D. T. Ting et al., Science 331, 593–596 (2011).
investigated heterochromatin organization with LNA-Majsat–treated cells (Fig. 4, M and N). By 22. Q. Zhu et al., Nature 477, 179–184 (2011).
23. M. Rabani et al., Cell 159, 1698–1710 (2014).
the constitutive heterochromatin marker tri- contrast, knock down of satRNAs in proliferat- 24. P. J. Ford, T. Mathieson, M. Rosbash, Dev. Biol. 57, 417–426 (1977).
methylated histone H3 lysine 9 (H3K9me3). ing NPCs did not impair proliferation or NPC
After knock down of major satRNAs, the sig- maintenance (fig. S21). To further characterize AC KNOWLED GME NTS
nals of H3K9me3 around chromocenters be- the role of satRNAs in NPC maintenance, we We thank M. L. Gage for editorial comments and current and
previous members of the Toda and Hetzer labs for helpful
came disorganized and reduced (Fig. 4, A to D; assessed the consequences of their overexpres- discussions. S.Z. was supported by an Add-on fellowship from the
and figs. S18C and S19, A to E), and we con- sion. Overexpression of both major and minor Joachim Herz foundation. NGS data production and data analysis
firmed the reduction in H3K9me3 at satellite satRNAs was detrimental for NPCs and led to were carried out at the DRESDEN-concept Genome Center,
which is supported by the DFG Research Infrastructure Program
regions using chromatin immunoprecipitation reduced proliferation, increased apoptosis, and (project 407482635) and part of the Next Generation Sequencing
(ChIP) (Fig. 4, E and F). Conversely, overex- DNA damage (fig. S22), indicating that proper Competence Network NGS-CN (project 423957469). We thank
pression of satRNAs led to an accumulation of control of satRNA levels is critical for NPC M. Krause and A. Dahl at the DRESDEN-concept Genome Center
for support with NGS data acquisition, G. Girke for molecular
H3K9me3 at chromocenters, suggesting that maintenance. Taken together, our results sug- cloning of sgRNAs, and M. Albert and A. Kolodziejczyk for providing
major satRNAs promoted heterochromatin gest that major satRNAs are retained long- research materials. Funding: This work was funded by the
formation (Fig. 4, C and G). Knock down of ma- term in quiNPCs and play critical roles in the Boehringer Ingelheim Foundation (T.T.), the European Research
Council (ERC-2018-STG, 804468 EAGER; ERC-2023-COG,
jor satRNAs also reduced facultative hetero- maintenance of heterochromatin integrity and 101125034 NEUTIME) (T.T.), the DZNE (T.T.), the Nomis Foundation
chromatin levels (H3K27me3) but did not affect neural stem cell function. Because not all LL- (M.W.H. and A.M.), and Deutsche Forschungsgemeinschaft [DFG,
euchromatin mark H3K4me3 (fig. S19, F to I). RNAs are satRNAs, further investigations will German Research Foundation (TO1347/4-1)] (T.T.). Author
contributions: Conceptualization: M.W.H., A.M., T.T.; Methodology:
Because minor satRNAs were also enriched be required to understand the biological roles
A.M., S.Z., T.T., R.S., F.H.G., N.R., U.F., M.L., A.K.; Visualization: S.Z.,
among LL-RNAs (Fig. 3, C to E) and their over- of LL-RNAs. A.M., R.S., T.T.; Funding acquisition: M.W.H., T.T.; Investigation:
expression increased long-retained EU signals Previously, several proteins had been discov- S.Z., A.M., R.S., T.T.; Project administration: M.W.H., T.T.; Resources:
in quiNPCs (fig. S18, D to F), we analyzed their ered in the mammalian brain that can persist F.H.G., T.T.; Supervision: M.W.H., T.T.; Writing – original draft: T.T.,
M.H.W.; Revision work: S.Z., A.K., T.T.; Writing – revised manuscript:
roles in heterochromatin retention. Although for years (1, 2), suggesting possible roles for S.Z., M.H.W., T.T.; Writing – review and editing: T.T., A.M., R.S., S.Z.,
minor satRNA overexpression significantly long-lived cellular constituents in brain main- F.H.G., M.H.W. Competing interests: The authors declare no
increased H3K9me3 at chromocenters, their tenance and aging. In this study, we found competing interests. Data and materials availability: All data
are available in the main text, figures, or supplementary materials.
knockdown did not affect H3K9me3 levels that a subset of RNAs could be retained for up EU-RNA-seq data were submitted to the Gene Expression Omnibus
(Fig. 4, C, D, and G; and fig. S18C). Altogether, to 2 years in certain brain cell types. Although, (GEO) under accession no. GSE248101. License information:
our results suggest a critical role for major in contrast to DNA, RNA has not been consid- Copyright © 2024 the authors, some rights reserved; exclusive
licensee American Association for the Advancement of Science. No
satRNAs in the maintenance of heterochro- ered a stable nuclear component (23), our find- claim to original US government works. https://www.science.org/
matin integrity in quiNPCs. ings are consistent with a previous study from about/science-licenses-journal-article-reuse
Because satRNAs are generated from hetero- dormant Xenopus oocytes that shows that some
chromatic areas and bind heterochromatin- RNAs can be retained for more than a year SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adf3481
associated proteins (19, 20), we tested whether (24). The LL-RNAs we identified were enriched
Materials and Methods
their long-term retention is controlled through around heterochromatin and essential to main- Figs. S1 to S22
association with heterochromatin. Perturba- taining chromatin integrity in quiescent neural Table S1
tion of H3K9me3 by pharmacological inhibi- stem cells. These findings highlight a previously References (25–35)
MDAR Reproducibility Checklist
tion of histone methyltransferases or by knock unrecognized temporal axis of RNA metabo- Data S1
down of heterochromatin-associated protein lism and possible functions for nuclear RNA in Submitted 15 October 2022; resubmitted 14 November 2023
1b (Hp1b) increased EU signal retention in long-term epigenetic regulation and genome Accepted 2 February 2024
quiNPCs (fig. S20), suggesting that hetero- integrity. We propose that the lifelong main- 10.1126/science.adf3481
P
damental interest but also display prop- maintain overall charge neutrality. results in the stoichiometry A2B2X4, abbrev-
erties that enable applications including The synthesis of all-organic perovskites has iated as ABX2. The critical challenge in this
ferroelectrics (1), quantum materials (2), been recognized as a challenging task within design approach lies in ensuring the stability
catalysts (3, 4), light-emitting devices the scientific community (12–29). In principle, of the additional B+ cation at the interstitial
(5), and solar cells (6). Perovskite crystals have the structural topology of perovskite can be position, which depends on its surrounding
a stoichiometry of ABX3, where A represents a maintained by replacing A, B, and X ions with bonding environment.
larger cation compared with B, X denotes an appropriately sized organic molecules, as dic- In a typical cubic unit cell there are four
anion, and B is coordinated to six X anions to tated by the Goldschmidt tolerance factor for sites available for occupation: the corner, body
form a BX6 octahedron. In three-dimensional relative ion sizes. However, in practice, only a center, face center, and edge center sites (Fig.
(3D) perovskites, these octahedra share corners few organic molecules (among those currently 2A). In the case of a perovskite unit cell, anions
and the A cation occupies the cuboctahedral used) fulfill this criterion. The isostructural occupy the face center site (X-site), the small
site surrounded by eight octahedra. By dis- family of dicationic piperazinium organic perov- cation occupies the body center site (B-site) to
rupting the continuous linkage of the octahedra skites with Cl−, Br−, and I− was discovered as form an octahedron with the anions, whereas
along the ½001direction, a 2D version of perov- recently as 2002 (12). More recently, Xiong et al. the large cation occupies the corner site (A-
skites can be formed in which layered struc- synthesized a diverse class of organic perov- site) to form a cuboctahedron with the anions.
tures are separated by van der Waals gaps. skites by designing various diammonium cations, The interstitial site at the edge center (E-site)
Depending on the interlayer offset and resulting in the report of 23 different organic can be filled by the small B+ cation (Fig. 2B).
the type of atoms present in the “spacer” 3D perovskites (14). These compounds include Our design for the all-organic lattice used
layer, 2D perovskites can be classified into the a few ferroelectrics such as MDABCO-NH4-I3 N-chloromethyl-1,4-diazabicyclo[2.2.2]octonium
Ruddlesden-Popper phase (RP phase) (7, 8), (MDABCO = N-methyl-1,4-diazabicyclo[2.2.2] (CMD+) as the A-site cation because it could
the Dion-Jacobson phase (DJ phase) (9), and octonium), which have polarization properties form hydrogen bonds both laterally and ortho-
the Aurivillius phase (10, 11). The RP and DJ (22 mC/cm2) similar to those of BaTiO3. gonally with the interstitial B+ cation at the
phase layered perovskites have general for- Aside from 3D metal-free perovskites, we edge center (E-site) of the unit cell (fig. S1).
mulas of A2BX4 and ABX4, respectively. These ask whether 2D-version (layered packing), The B-site was occupied by NH4+ and the X-
structures typically consist of corner-shared metal-free perovskites can be synthesized. site was occupied by PF6–. In the case of CMD+,
BX4 octahedra forming a 2D layer, with the A+ The presence of a van der Waals gap in 2D the lone pair at the nitrogen formed an N-H···N
cations coordinated to the cuboctahedral sites perovskites offers the advantage of incorpo- hydrogen bond with NH4+-edge laterally,
rating larger organic cations through self- whereas the chloromethane from the adjacent
adjustable strain-balancing within the layered layer formed an N-H···Cl hydrogen bond with
structures, which could overcome steric effects NH4-edge vertically, thus stabilizing it by form-
1
Department of Applied Physics, Hong Kong Polytechnic that hinder 3D perovskite synthesis (30). This ing an NH4+-edge[PF6]3[N]2[Cl] octahedron
University, Hung Hom, Kowloon, Hong Kong, China.
2
Department of Physics, Southern University of Science and
expanded library of larger and more exotic (Fig. 2C).
Technology, Shenzhen 518055, China. 3Department of A-site cations in 2D perovskites could enable
Chemistry, National University of Singapore, Singapore the emergence of distinctive properties and Structure description of CL-v phases
117543, Singapore. 4Graduate School of Analytical Science
and Technology (GRAST), Chungnam National University,
applications. We present the design principles We synthesized CMD-N-P2 through the slow
Daejeon 34134, Republic of Korea. 5College of Electronic and synthesis of a new class of all-organic diffusion of dichloroethane into an acetone
Engineering, South China Agricultural University, Guangzhou layered 2D perovskite phases. Following the solution containing CMD-PF6 and NH4PF6. A
510642, China. 6The Hong Kong Polytechnic University
naming convention of oxide perovskites, we hexagonal-shaped single crystal of CMD-N-P2
Shenzhen Research Institute, Shenzhen 518057, China.
7
Quantum Science Center of Guangdong–Hong Kong–Macao refer to these as Choi-Loh van der Waals phase measuring 9 × 7 × 0.5 mm was obtained (Fig.
Greater Bay Area (Guangdong), Shenzhen 518045, China. (CL-v phase), with the suffix v denoting van der 2E, inset). The crystal structure was monoclinic
*Corresponding author. Email: jun.yin@polyu.edu.hk (J.Y.); Waals stacking in the structure. The CL-v phase and belonged to the P21/n symmetry group, as
linjh@sustech.edu.cn (J.L.); kathy-kai.leng@polyu.edu.hk (K.L.);
kian-ping.loh@polyu.edu.hk (K.P.L.) layered perovskites can be exfoliated or grown determined by single-crystal x-ray diffraction
†These authors contributed equally to this work. as ultrathin layers of several nanometers. analysis (see supplementary materials, table S1).
Fig. 2. Crystal structure description of CL-v. (A) A-, B-, and X-sites in the cubic with the NH4-edge, resulting in the formation of an NH4+-edge[PF6]3[N]2[Cl]
cell, with the edge center (E-site) reserved for interstitial B+ cation. (B) The pseudo- octahedron for stabilization of the NH4+-edge. Only half of CMD+ forms N-H···Cl bonds,
cubic unit cell of the CL-v phase is shown, with the NH4-body[PF6]6 octahedron hence hydrogen bonded and non-bonded Cl are denoted in different colors. (D) The
represented in light cyan. One NH4+-edge is located in the upper (001) plane and the geometry of two NH4-body[PF6]6 octahedrons with an NH4+-edge ion. Bond lengths in
other is in the bottom (001) plane. The E-site on (001) occupied with NH4+ has angstroms (Å). (E) View along the ½001direction. The unit vectors for NH4+-edge are
+
an occupancy of one-half because of sharing with one adjacent unit in the layered
of 11.6, 12.7, and 12.7 Å each. NH4 -edge
110 , ½130, and ½310, with lengths and
Cl atoms
structure. CMD represents chloromethyl DABCO and BMD represents bromomethyl are arranged linearly in the 110 direction and alternatingly in the 110 direction.
DABCO. Both the B-site and E-site consist of NH4+ ions, differentiated by the color of Inset is 9 × 7 × 0.5 mm size hexagonal single crystal of CMD-N-P2. (F) The
nitrogen for clarity. (C) The hydrogen bonding environment of the NH4+-edge, view is along the NH4+-edge and Cl alignment direction 110 . The adjacent layers
surrounded by three PF6− ions, two N-H···N bonds to create a NH4[PF6]3[N]2 are held together by N-H···Cl bonds. (G) View along the ½010 direction with two
translucent red pyramid. The Cl of CMD+ in the adjacent layer forms an N-H···Cl bond layers indicating AB stacking.
nm
0.41 nm
800 0.41 0
80
Sim.
440
[0 0 1] 90o
800
440
440
ADF
90o
800
440
Fig. 3. Atomic structure and elemental characterization of the CL-v phases perfectly matched periodic symmetrical structure, as emphasized by the blue
via cryo-TEM. (A) HRTEM image of CMD-N-P2 taken along the ½001 direction at squares. The green double lines indicate the interplanar spacing corresponding to
~77 K with an electron dose rate of ~0.68 e− Å−2 s−1 and a cumulative dose of the (800) lattice planes. Simulated experimental image (D) and reciprocal lattice
~1.1 e− Å−2. The inset displays the corresponding FFT pattern. (B) A magnified high- (E) are based on the ½001 oriented structural model in (C). (F) The EELS spectrum
resolution image of CMD-N-P2 lattice fringes from a selected area of image (A), of CMD-N-P2 exhibits clear onset features of P-L, Cl-L, C-K, N-K, and F-K edges,
and (C) the corresponding atomic model in the ½001 axis. A bright spot in image with a representative ADF image shown alongside. (G) HRTEM image, (H) atomic
(B) corresponds to a column of organic groups in image (C), demonstrating a model along the ½001 zone axis, and (I) EELS spectrum of BMD-N-P2.
Cryo–transmission electron microscopy short exposures at a dose rate of 0.68 e− Å−2 s−1 indicated by the blue boxes in Fig. 3B and 3C,
(TEM) analysis under a magnification of ×59000 (pixel size: the lattice spacing observed in the experiment
Direct evidence of the perovskite-type lattice 1.1 Å), with a total dose of ~1.1 e− Å−2. Columns completely aligned with that formed by CMD+,
arrangement necessitates atomic TEM char- of organic groups linked by covalent bonds and the atomic contrast was consistent with
acterization of the various molecules in the were well-resolved in the enlarged HRTEM the result of the simulated HRTEM image at
lattice. We achieved near-atomic resolution image, displaying an ideal cubic perovskite 100 nm defocus value (Fig. 3D and fig. S10).
imaging of the pure organic perovskite crystal structure along the ½001 projection, with a Additionally, the fast Fourier transform (FFT)
structure under a cumulative electron dose of homogeneous alternation of bright and dim pattern inserted in Fig. 3A closely matched the
no more than 2 e− Å−2 in a cryogenic TEM spots (Fig. 3B). simulated reciprocal lattice (Fig. 3E), except
(see supplementary materials for details and The square lattice, constructed by the small- for some extinction reflections reappearing in
imaging conditions). In the case of CMD-N-P2, est bright or dim spots, corresponded to the the FFT pattern because of dynamical scatter-
Fig. 3A displays its large-area, high-resolution square repeating unit constructed by the col- ing (thickness effect). The (800) reflection,
lattice-fringe image along the ½001 zone axis. umns of the A-site (CMD+) plus B-site (NH4+) marked by the green arrow, exhibits the high-
This image was acquired through successive or columns of X-site (PF6−) plus NH4+-edge. As est diffraction intensity in both the FFT and
Fig. 4. Exfoliation and growth of CMD-N-P2 ultrathin flakes. (A) AFM height image and line profile of 30 mm–sized CMD-N-P2 exfoliated by sonication in
dichloromethane. Scale bar is 10 mm. (B) AFM height image and line profile of CMD-N-P2 grown on SiO2 substrate with 0.1 weight percent (wt%) solution. Scale bar
is 6 mm. (C) AFM height image and line profile of CMD-N-P2 grown on SiO2 substrate with 1 wt% solution, showing 6 layers. Scale bar is 3 mm. (D) AFM height
image and line profile of a screw dislocation in the clockwise direction. Scale bar is 3 mm.
the simulated reciprocal lattice, which corre- by sonication, disperse in solvents and exhibit the CMD-N-P2 solution, we can synthesize crys-
sponded to the lattice plane and is marked by the characteristic Tyndall effect of light scat- tals as thin as ~4 nm (Fig. 4C). Thus, our find-
green lines in Fig. 3, B and C, with an inter- tering by a colloidal dispersion. The Tyndall ings highlight the potential to grow 2D layered
planar spacing of 0.41 nm. The chemical com- effect of exfoliated CMD-N-P2 and BMD-N-P2 organic crystals using a solution-based ap-
position of the CMD-N-P2 crystal, revealed by was weak in nonpolar solvents such as hexane proach while achieving meticulous thickness
electron energy loss spectroscopy (EELS) in and toluene. However, for aprotic polar solvents control. The crystalline nature of the CMD-N-P2
Fig. 3F, showed characteristic edges of P, Cl, such as dichloromethane, the dispersion exhib- flakes grown using this method was evident
C, N, and F, which matched its mechanically ited a strong Tyndall effect (figs. S14 and S15). from the presence of screw dislocations in
exfoliated bulk counterpart (figs. S11 and S12). We inferred that the interlayer region of some flakes, exhibiting both clockwise (Fig. 4D)
We also tested another CL-v phase based on CMD-N-P2 encompassed N-H···Cl bonds, and and counterclockwise (fig. S18) dislocations
BMD-N-P2, which shared similar crystal struc- solvents containing chloro-functional groups with a step height of ~4 nm (33, 34).
ture and lattice spacing with CMD-N-P2, given could disrupt these interlayer hydrogen bonds
that only the chloromethyl functional group by serving as H-bond acceptors. Thus, the ex- Applications of gate dielectric in
in CMD+ was replaced with bromomethyl. As foliated flakes were stabilized. Indeed, all 2D electronics
depicted in Fig. 3G, the HRTEM image of chloro-functionalized solvents, including dichlo- The ability of 2D all-organic perovskites to
BMD-N-P2 crystals displayed a perovskite- romethane, 1,2-dichloroethane, 1-chlorohexane, form molecularly thin films over large areas,
type arrangement of the organic functional chlorobenzene, and 1,3-dichlorobenzene, exhib- combined with their insulating nature, makes
groups, similar to that shown in Fig. 3A with ited a strong Tyndall effect with CMD-N-P2. them suitable as dielectric layers in 2D elec-
well-matched EELS (fig. S13). Further confirmation of exfoliation was ob- tronics. The optical band gaps of CMD-N-P2
tained through AFM characterization, which and BMD-N-P2 were determined from the in-
Layered structure and exfoliation revealed that the flakes exfoliated by dichloro- tercepts of the Tauc plots to be 5.56 eV and
The layered structure of CMD-N-P2 was inves- methane and 1,2-dichloroethane possessed a 4.69 eV, respectively (Fig. 5A). The wave func-
tigated with atomic force microscopy (AFM), thickness as thin as ~4 nm, similar to the step tions were mostly localized in the molecules in
which revealed a step height of 4.2 nm. Taking height measured for layered crystal synthe- these crystals, yielding a large band gap and
into account an overestimation of the thick- sized in this work (fig. S16). The largest flake small bandwidth <0.02 eV, as seen from the
ness by ~1 nm by AFM measurement owing size reached up to 30 mm when CMD-N-P2 band structure calculated with density func-
to tip-surface interactions, surface roughness, was sonicated in dichloromethane (Fig. 4A). tional theory (DFT) (fig. S19). The DFT-calculated
and trapped solvent (31, 32), the closest thick- Exploiting the ability of solvents with chloro- band gaps of 4.81 and 4.14 eV of CMD-N-P2
ness that matches this is the 2.56-nm–thick functional groups to disrupt interlayer N-H···Cl and BMD-N-P2 were underestimated but
bilayer of the pseudo-cubic cell (fig. S2). Since bonds, we successfully demonstrated controlled agreed with the experimental trends qualita-
we did not measure anything thinner than growth of CMD-N-P2 on a silica substrate. tively and indicate that the Coulombic poten-
2.56 nm, we assume that the alternating step Gradual evaporation of a CMD-N-P2 solution tial of atomic nuclei such as Cl and Br affects
is at least a bilayer. A distinctive attribute of in the presence of 1,2-dichloroethane led to the gap (fig. S19) (29).
van der Waals–stacked 2D materials is their the precipitation of hexagonal-shaped flakes The dielectric constants of CMD-N-P2 and
ability to be exfoliated into molecularly thin measuring 10 mm in width (Fig. 4B). We also BMD-N-P2 were measured in the range of
flakes through solvent-based exfoliation of bulk obtained similar results for BMD-N-P2 (fig. S17). 1 kHz to 1 MHz, giving values of 4.86, and
crystals, and that the resulting flakes, assisted Moreover, by adjusting the concentration of 5.53, respectively (Fig. 5B), which were higher
SiO Al
HfSiO
hBN Y
Si N
CMD-N-P
BMD-N-P
COF
158 mV dec-1 7V
5V
3V
-1 V
-3 V
-5 V
50 m -7 V
Fig. 5. Dielectric properties of CL-v family and FET device using CL-v phases as dielectric layer. (A) Tauc plot of the CL-v perovskites. (B) The dielectric
constant of the CL-v perovskites displayed from 1 kHz to 1 MHz. (C) Dielectric constant and band gap values of the CL-v perovskites versus other dielectric materials.
(D) Optical microscope image of FET using 100 nm CMD-N-P2 as a dielectric layer. The transfer curve (E) and output curve (F) of FET with BMD-N-P2 as a dielectric
layer. Subthreshold swing is 158 mV per decade (dec−1).
than the dielectric constant of SiO2 (3.9), but which was lower than that of Al2O3 grown 3. S. Royer et al., Chem. Rev. 114, 10292–10368 (2014).
lower than HfO2 (25). The breakdown fields through atomic-layer deposition (36) The drain 4. M. A. Peña, J. L. G. Fierro, Chem. Rev. 101, 1981–2017 (2001).
5. L. N. Quan et al., Chem. Rev. 119, 7444–7477 (2019).
of CMD-N-P2 and BMD-N-P2 are found to be current–voltage (Id–Vd) output curves exhibit 6. J. Y. Kim, J.-W. Lee, H. S. Jung, H. Shin, N.-G. Park, Chem. Rev.
23 MV m−1 and 17 MV m−1, respectively (fig. linear characteristics at low bias. (Fig. 5F) 120, 7867–7918 (2020).
S21). These values are higher than those of 7. S. N. Ruddlesden, P. Popper, Acta Crystallogr. 10, 538–539
hBN (12 MV m−1), Alumina (13.4 MV m−1) and Conclusions (1957).
8. S. N. Ruddlesden, P. Popper, Acta Crystallogr. 11, 54–55 (1958).
polystyrene (19.7 MV m−1). Compared to 2D We designed and synthesized a new class of 9. M. Dion, M. Ganne, M. Tournoux, Mater. Res. Bull. 16,
dielectric materials such as hexagonal boron 2D organic perovskites known as the CL-v phase. 1429–1435 (1981).
nitride, CL-v phases have both a high dielectric Our synthetic approach is versatile and can 10. B. Aurivillius, Ark. Kemi 1, 463–480 (1949).
constant and a large band gap, which makes be extended to various organic molecules as 11. B. Aurivillius, Ark. Kemi 2, 519–527 (1950).
12. C. A. Bremner, M. Simpson, W. T. A. Harrison, J. Am. Chem. Soc.
them suitable for use as dielectric layers in 2D A-site cations, including diazabicycloalkane 124, 10960–10961 (2002).
electronics (Fig. 5C). To demonstrate the ef- (37, 38), azaadamantane (39), and other deriv- 13. G.-Z. Liu, J. Zhang, L.-Y. Wang, Synth. React. Inorg. Met.-Org.
fectiveness of CL-v phases as gate dielectric atives of heterocyclic compounds (40), as well Nano-Met. Chem. 41, 1091–1094 (2011).
14. H.-Y. Ye et al., Science 361, 151–155 (2018).
layers, we fabricated a field-effect transistor as different B-site cations and X-site anions 15. S.-L. Chen et al., Sci. China Mater. 61, 1123–1128 (2018).
(FET) by transferring ≈100 nm-thick CMD-N-P2 (fig. S23). CL-v phase can be exfoliated or 16. G.-M. Fan, C. Shi, L. Qiao, H.-J. Li, H.-Y. Ye, J. Mater. Chem. C
or BMD-N-P2 as the top gate dielectric layer on grown as molecular thin crystalline organic Mater. Opt. Electron. Devices 6, 8349–8352 (2018).
17. H. Morita et al., Angew. Chem. Int. Ed. 58, 9184–9187 (2019).
MoS2 as the channel material, and with Cr/ layers with large aspect ratio by solution pro- 18. T. W. Kasel et al., Chem. Sci. 10, 8187–8194 (2019).
Au metal as the source and drain electrodes. cess, which can be potentially used as flexible 19. W.-X. Zhang, S.-L. Chen, Y. Shang, Z.-H. Yu, X.-M. Chen,
Figure 5D shows an optical image of a fab- dielectrics. Our findings enhance the under- Energetic Materials Frontiers 1, 123–135 (2020).
20. Y. Shang et al., Cryst. Growth Des. 20, 1891–1897 (2020).
ricated FET device. The measured transfer standing of structural design principles per-
21. H. Morita, R. Tsunashima, S. Nishihara, T. Akutagawa,
curve at different Vds displays n-type behavior taining to all-organic perovskites. Based on CrystEngComm 22, 2279–2282 (2020).
and the on-off ratio is >107 when the Vds is this understanding, we posit that it is fea- 22. J.-J. Wang, D. Fortino, B. Wang, X. Zhao, L.-Q. Chen, Adv. Mater.
0.5 V, which is sufficient to comply with the stan- sible to conceive and synthesize all-organic 32, e1906224 (2020).
23. X. Song, G. Hodes, K. Zhao, S. Liu, Adv. Energy Mater. 11,
dard of practical logic circuits of 104 (Fig. 5E) (35). perovskite analogs of the Dion-Jacobsen and 2003331 (2021).
The negligible hysteresis in the transfer curve Aurivillius phases. 24. Y. Gao et al., ACS Appl. Mater. Interfaces 13, 19042–19047
suggested that a clean contact formed between (2021).
25. H. S. Choi et al., Nat. Commun. 13, 794 (2022).
the dielectric and channel material. Further- RE FERENCES AND NOTES 26. H.-S. Wu et al., Adv. Sci 9, e2105974 (2022).
more, a low subthreshold swing of 158 mV per 1. W. Zheng et al., Adv. Mater. 35, e2205410 (2023). 27. R. Taheri-Ledari et al., Energy Fuels 36, 10702–10720 (2022).
decade was extracted from the transfer curve, 2. L. T. Nguyen, R. J. Cava, Chem. Rev. 121, 2935–2965 (2021). 28. T. Handa et al., Science Advances 8, eabo1621 (2022).
29. Z. Li et al., Adv. Mater. 35, e2300480 (2023). Kong Polytechnic University as well as funding from Hong Kong research. J.L. and K.L. fabricated and characterized the FET device.
30. L. Mao, C. C. Stoumpos, M. G. Kanatzidis, J. Am. Chem. Soc. Jockey Club STEM lab of 2D Quantum Material (2023-0033). K.L. G.W., F.L., and J.L. performed cryo-TEM measurements. W.P.D.W
141, 1171–1190 (2019). acknowledges the Croucher Foundation (Croucher Tak Wah Mak and I.H.P. conducted the single-crystal XRD structure analysis. J.Y.
31. C. J. Shearer, A. D. Slattery, A. J. Stapleton, J. G. Shapter, Innovation fund 2023) and projects 62322413 and 12104382 conducted DFT calculations of band gap. H.S.C. and K.P.L. wrote the
C. T. Gibson, Nanotechnology 27, 125704 (2016). supported by National Natural Science Foundation of China and manuscript with contributions from all authors. H.S.C., J.L., and
32. X. Li et al., Nat. Chem. 12, 1115–1122 (2020). the Research Grants Council of the Hong Kong Special G.W. contributed equally. Competing interests: The authors declare
33. Y. Zhao, S. Jin, Acc. Mater. Res. 3, 369–378 (2022). Administrative Region, China (projects PolyU25305222, no competing interests. Data and materials availability: The data
34. Y. Zhao et al., Science 370, 442–445 (2020). PolyU15305221, and PolyU15304623) J.Y. acknowledges financial that support the plots within this article and other findings of this
35. M. Chhowalla, D. Jena, H. Zhang, Nat. Rev. Mater. 1, 16052 support from Hong Kong Polytechnic University (grant P0042930) study are available from the corresponding author upon reasonable
(2016). and a grant from the Research Grants Council of Hong Kong request. License information: Copyright © 2024 the authors, some
36. A. Sebastian, R. Pendurthi, T. H. Choudhury, J. M. Redwing, (project PolyU 25300823).This project is partially supported by the rights reserved; exclusive licensee American Association for the
S. Das, Nat. Commun. 12, 693 (2021). National Natural Science Foundation of China (11974156), Advancement of Science. No claim to original US government works.
37. S. F. Nelsen, J. M. Buschek, J. Am. Chem. Soc. 96, 7930–7934 Guangdong Innovative and Entrepreneurial Research Team Program https://www.science.org/about/science-licenses-journal-article-reuse
(1974). (grant 2019ZT08C044), Shenzhen Science and Technology Program
38. S. F. Nelsen, P. J. Hintz, J. Am. Chem. Soc. 94, 7114–7117 (1972). (KQTD20190929173815000 and 20200925161102001), the Science, SUPPLEMENTARY MATERIALS
39. A. I. Kuznetsov, V. A. Kosmakov, B. V. Unkovskii, Chem. Technology and Innovation Commission of Shenzhen Municipality
Heterocycl. Compd. 21, 697–700 (1985). science.org/doi/10.1126/science.adk8912
(ZDSYS20190902092905285). TEM/STEM characterization was
40. M. W. Giuliano, C.-Y. Lin, D. K. Romney, S. J. Miller, E. V. Anslyn, Materials and Methods
performed at the cryo-EM Center and Pico Center from SUSTech Core
Adv. Synth. Catal. 357, 2301–2309 (2015). Figs. S1 to S23
Research Facilities that receives support from the Presidential Fund
Tables S1 to S3
and Development and Reform Commission of Shenzhen Municipality.
ACKN OW LEDG MEN TS References (41–43)
Author contributions: H.S.C. and K.P.L conceived the idea and
Funding: This work was supported by project P0043087 Design developed the project plan. H.S.C. designed, synthesized, grew Submitted 19 October 2023; accepted 22 February 2024
and Synthesis of Hybrid Organic Inorganic Perovskites of Hong bilayers, and characterized the CL-v crystals. K.P.L. supervised the 10.1126/science.adk8912
High [MCH]
A B REDUCTION IN AW LUMEN C EXTRUSIONS FOLLOWING MCH
**** 100 **** 1
300000
Extrusions/bronchiole
200000
50
100000
Unprimed Primed
60000
**
20000
PRIMED
% denuded cells/bronchiole
MCH+ALB
0min 15min 17min 32min
PRIMED
NM MCH then
MCH ALB
Fig. 1. Bronchoconstriction in ex vivo lung slices causes excess cell extru- denuding. (D) Sample confocal projections (scale bar, 50 mm) of bronchioles
sion. (A) Movie stills of 500 mg/ml MCH response in unprimed and 2-week after 200 mg/ml (low) or 500 mg/ml (high) MCH, where red arrowheads
HDM-primed bronchioles (minutes:seconds), with red arrowheads pointing to depict single-cell extrusions and blue ones depict epithelial denuding
single-cell extrusions (scale bars, 50 mm). (B) Constriction scales with size, (<100 extrusions on graphs; blue data points). (E) Increasing MCH concentration
measuring the bronchiolar lumen areas before and after MCH treatment increases extrusion (**P < 0.0005; ****P < 0.0001) in >15 slices per treatment
in large, medium, and small bronchioles from >5 airways from 5 HDM-primed from >5 mice. (F and G) Movie stills (F) from a 5-week HDM-primed ex vivo
mice (****P < 0.0001 from a Wilcoxon pairs-matched signed rank test). lung slice treated with 500 mg/ml MCH for 15 min and then relaxed with
(C) Quantification of extrusions from immunostained ex vivo lung slices from 3.5 mM ALB + 500 mg/ml MCH for 15 min, showing that epithelia still detach
unprimed and OVA-primed mice treated with 500 mg/ml MCH (****P < 0.0005 (blue arrowheads; scale bars, 100 mm), quantified in (G) with no significant
from an unpaired Mann-Whitney analysis compared with control from >5 lung difference (ns) between MCH and ALB from a Mann-Whitney test
slices from >10 mice), where blue data points represent complete epithelial (****P < 0.0001).
and IL-13 and high mucus production (fig. S1, airway epithelial extrusion in primed ex vivo sponse to MCH compared with unprimed con-
C to F). MCH addition caused pronounced mouse slices, sparing unprimed controls (Fig. 1, trol mice, with HDM priming showing the
bronchoconstriction of mouse lungs that were A and C). To quantify the number of extrusions strongest effects (Fig. 1, A and B, and movies
immune-primed with either OVA or HDM in response to MCH, we immunostained lung S2 to S4). Additionally, we occasionally noted
[as outlined in fig. S1 (14–19)] but did not sub- slices with E-cadherin for epithelia, phalloidin the unjamming of airway epithelia from a
stantially affect unprimed airways (Fig. 1, A for ASM, and 4′,6-diamidino-2-phenylindole previous static state to collective migration
and B, and movies S1 to S4). Within 15 min of (DAPI) for DNA, scoring for extrusions as cells (movie S5). To test whether the amount of
MCH treatment, the luminal area of small and pinching off apically into the lumen (Fig. 1, C constriction correlated with extrusion rates,
medium bronchioles reduced markedly, caus- to E, black data points and red arrowheads) we used increasing MCH doses on OVA-primed
ing severe airway epithelial crowding (Fig. 1B) and complete epithelial denuding (Fig. 1, C to mice and found a tight correlation between the
and increasing cell heights on average 196 ± E, blue data points and blue arrowheads). All amount of constriction and cell extrusions, with
11% (SEM; 13 airways from n = 5 mice). Fifteen immune-priming methods induced pronounced the highest doses causing complete denuding of
minutes of bronchoconstriction caused excess bronchoconstriction and excess extrusion in re- the epithelium (Fig. 1, D and E).
A CROWDING-INDUCED EPITHELIAL CELL EXTRUSION PATHWAY & INHIBITORS: B S1P ACTIN DAPI
GsMTx4 JTE-013
MCH
3+
100 100
****
**** ns
**** ****
50 50
MCH+ALB/Gd
17min
0 0
NM
NT MCH
MCH Gd 3+ SKIs
JTE SKIs Gd3+
JTE NM MCH
NT MCH Gd 3+
JTE SKIs
SKIs JTE
Gd3+
then MCH then MCH
3+
F # EXTRUSIONS/BRONCHIOLE G % DENUDED CELLS/BRONCHIOLE
100 1 100
**
**** ***
Complete denuding of airway
*
32min
50 50
0 0
0
NM MCHMCH
Control then
Gd3+ GsMTx4 MCH then
MCH ALB GD/ALB GD
MCH Gd3+ GsMTx4 MCH ALB ALB/Gd3+ Gd3+
Fig. 2. SAC and S1P inhibitors block extrusion caused by broncho- Wallis test, from >4 slices per mouse from >4 mice). (F and G) Quantification of
constriction. (A) Canonical crowding-induced epithelial cell extrusion pathway, OVA-primed lung slices treated with Gd3+ or GsMTx4 after 15 min of MCH
indicating where each inhibitor acts within the pathway. (B) Confocal projection challenge, where black dots represent extrusions per bronchiole and blue dots
of an extruding cell immunostained for S1P, phalloidin for f-actin, and DAPI represent complete epithelial denuding from >5 mice (F) or percent denuding (G)
(scale bar, 25 mm). (C to E) Confocal projections (C) of ex vivo lung slices from (*P < 0.05; **P < 0.001; ***P < 0.0005; ****P < 0.0001 from a Mann-Whitney
5-week HDM-primed mice with no MCH (n = 9), MCH (n = 9), or pretreated with test, from >4 slices per mouse from >4 mice). (H) Movie stills from an HDM-
Gd3+ (n = 9), sphingosine kinase inhibitors SKI II and K-145 (n = 9), or S1P2 primed mouse lung slice pretreated with MCH for 15 min, followed by ALB
antagonist JTE-013 (n = 4), before adding MCH (scale bars, 50 mm), quantified in (3.5 mM) + MCH (500 mg/ml), with arrowheads pointing to areas of epithelial
(D) as extruded cells (red arrowheads) and (E) as percentage of epithelial denuding that reattach by 32 min (scale bars, 50 mm). Note that albuterol
denuding (blue arrowheads) per bronchiole (****P > 0.0001 from a Kruskal- alone did not prevent epithelial destruction (G).
C extrusions post MCH extrusions/airway (30min PM) denuding post MCH denuding/airway (30min PM)
**** ****
**** ** **** ** **
40
30
50
20
10
3+
NM - ALB ALB/Gd NM - ALB ALB/Gd 3+
MCH MCH
D No MCH MCH (24hrs PM) MCH+ALB (24hrs PM) MCH+ALB/SKI (24hrs PM) MCH+ALB/Gd (24hrs PM)
3+
75
50
2 3
25
0
NT
NM MCH MCH MCH MCH
ALB ALB/SKIs ALB/Gd3+
Fig. 3. SKIs and gadolinium block extrusion and inflammation after a bron- with representative images. Scale bars, 50 mm (**P < 0.001; ***P < 0.0005;
choconstrictive attack in live mice. (A) Schematic of live MCH challenges ± ****P < 0.0001 from a Mann-Whitney test). (D to F) H&E sections 24 hours after
5 min of pretreatment with extrusion inhibitors before increasing MCH ± inhibitors. MCH challenge to observe immune response from no MCH treatment (n = 4), MCH
(B and C) H&E sections from lungs of mice not exposed to MCH (n = 4) or alone (n = 5), MCH + ALB (n = 5), MCH + ALB/SKI (n = 3), or MCH + ALB/Gd3+
treated with MCH (n = 5), MCH + ALB (n = 5), or MCH + ALB/Gd3+ (n = 6) at (n = 5), with representative images shown in (D) and quantification of airway immune
30 min after MCH, with the number of extrusions (red arrowheads) and percent cell infiltration using pathological scores, defined by colors and numbers in (E) and
denuding (blue dotted outline) per bronchiole quantified in (C), respectively, (F) (*P < 0.05; **P < 0.001 from a Chi-squared test). Scale bars, 100 mm.
B MCH MCH+Gd 3+
WGA-350
**** ****
**
100
1
% mucus score
2
1
MCH+ALB MCH+ALB/Gd 3+ 0
50
0
3+
NM
1 2- ALB
3 ALB/Gd
4
MCH
Fig. 4. Gadolinium blocks mechanically induced mucus secretion. WGA is retained in the epithelium rather than secreted with gadolinium.
(A) Confocal projections of Muc5AC before and after 2 and 5 weeks of HDM (C and D) Representative PAS (mucus) staining (C) 30 min after a MCH
priming (scale bars, 50 mm). (B) Movie stills from 3-week HDM-primed challenge ± ALB or ALB/Gd3+, which was quantified using a color/number scoring
ex vivo slices incubated with 10 mg/mL WGA-350 to label mucus then in (D), with **P < 0.001 and ****P < 0.0001 from Chi-square tests from at
treated with 500 mg/ml MCH alone ± 10 mM Gd3+Cl for 15 min, noting that least 5 mice per group. Scale bars, 50 mm.
Reversing bronchoconstriction does not prevent airway epithelial extrusion and des- reason, we quantified the percent of epithelial
prevent extrusion truction (Fig. 1, F and G, and movie S6). In fact, cell denuding per bronchiole rather than
Alleviating airway constriction with albuterol movies show that epithelia frequently detached extrusions (Fig. 1G). Thus, albuterol does not
could potentially reverse the excess epithelial from the smooth muscle as the ASM sprang prevent destruction of the airway lining after a
extrusion and denuding. However, we found open with albuterol administration, whereas bronchoconstrictive attack. Moreover, the in-
that although albuterol relaxes airways that epithelia remained buckled (movie S6 and creased denuding that we noted could poten-
were bronchoconstricted for 15 min, it did not Fig. 1, F and G, blue arrowheads). For this tially impede airway repair.
A B
Fig. 5. Treated patients with asthma have marked airway epithelial extru- boxes indicate areas enlarged below, with blue arrowheads indicating breaks in
sion and damage. (A and B) H&E pathology sections from lower lobectomy of a the epithelium and red arrowheads indicating epithelia extruded into the
nonsmoker patient with moderate asthma symptoms medicated with regular lumen. Note, lumen filled with mucus and extruded cells even in the moderate
moderate–dose inhaled corticosteroid and long-acting b2-agonist (formoterol) asthma case [(A), right]. (C) Model describing how the mechanics of asthmatic
and as-needed inhaled short-acting b2-agonist (albuterol) (A) and a nonsmoker bronchoconstriction cause excessive crowding-induced mucus secretion and
patient with severe asthma on high-dose inhaled corticosteroid, long-acting b2- epithelial cell extrusion that results in epithelial damage and inflammation.
agonist (formoterol), oral leukotriene receptor antagonist, oral prednisone, and Inhibiting extrusion early in the pathway prevents all pathological consequences
as-needed inhaled short-acting b2-agonist (albuterol) (B), where dashed-line of an attack. AW, airway.
Inhibiting extrusion blocks airway osin contraction needed to seamlessly eject a Piezo1—the SAC that we identified as critical
epithelial damage cell from the monolayer (8, 20) (Fig. 2A, sche- for activating extrusion in response to crowd-
We next investigated whether canonical extru- matic, and fig. S2A). As has been seen previ- ing in Madin-Darby canine kidney (MDCK)
sion inhibitors could prevent bronchoconstric- ously with crowding-induced extrusion (8), cells and zebrafish (8)—is expressed in mouse
tive airway damage. We previously discovered few of the extruding cells are apoptotic (fig. airways, they also express the transient recep-
that crowding-induced live cell extrusion re- S2B). Extruding airway epithelial cells speci- tor protein (TRP) channels TRPA1, TRPV1, and
quires the stretch-activated channel (SAC) Piezo1 fically up-regulate S1P, the limiting signal for TRPM8 frequently up-regulated in unmanageable
to trigger production of the bioactive lipid extrusion, which suggests that bronchocon- asthma (21) that could act similarly to trigger
sphingosine 1–phosphate (S1P) that binds the strictive extrusion operates through the canoni- extrusion (fig. S2, C to F). Notably, immune
S1P2 receptor to activate Rho-mediated actomy- cal pathway (Fig. 2B and fig. S2A). Although priming causes Piezo1, TRPV1, and TRPM8 to
become more vesicular and widespread (fig. without causing undue harm and stress to mice, constriction, as shown by the loss of blue
S2, C to F). Thus, to inhibit Piezo1 as well as delivering it in increasing amounts, as adminis- fluorescence (Fig. 4B and movies S11 and S12).
other potential TRP channels, we used gado- tered with patients, to ensure that no mice were Unexpectedly, Gd3+ markedly reduced mucus
linium hexahydrate chloride (Gd3+), an inhib- hyperresponsive (Fig. 3A, schematic). We then secretion in primary airways in response to
itor of SACs and TRP channels. Additionally, administered Gd3+ or sphingosine kinase inhib- our ex vivo MCH challenges (Fig. 4B and movie
we blocked S1P production with sphingosine itors (SKIs) during MCH challenge to test whe- S13). Although our WGA-350 assay is not a very
kinase 1 (SKI II) and 2 (K-145) inhibitors and ther blocking early steps in the extrusion pathway quantitative assay because of the uneven ex-
its receptor S1P2 with the antagonist JTE-013 that protected epithelia in our ex vivo assay would pression of mucus from immune priming and
during MCH-induced bronchoconstriction and prevent subsequent inflammation in our in vivo WGA loading, we found that ~90% of control
quantified extrusion and denuding. All inhib- assay. SKIs can be repeatedly administered primary airways secrete mucus, whereas only
itors markedly decreased extrusions after MCH safely to mice intranasally (23), and we found ~26% of Gd3+-treated ones do. In vivo, we saw
(Fig. 2, C and D); however, only Gd3+ and SK that mouse lung gross morphology and tissue a similar response in both OVA- and HDM-
inhibitors blocked epithelial sheet denuding as sections were indistinguishable from those primed mice. PAS staining from HDM-primed
well (Fig. 2, C and E). It is not clear why only of control mice (fig. S3A) after intranasal ins- mice challenged with MCH for ~30 min showed
Gd3+ and SK inhibitors block detachment, but tillation of 10 mM Gd3+ for 5 consecutive days bulk mucus secretion with large mucus globules
it suggests that blocking extrusion upstream in (fig. S3B) or once a week for 3 weeks (fig. S3C). squeezing out apically, which was not prevented
the pathway offers greater protection to airway We found no obvious masses, inflammation, or by albuterol (Fig. 4, C and D), whereas Gd3+
epithelium. obvious changes to mouse health and behavior treatment with albuterol substantially reduced
We next tested whether we could inhibit ex- (movies S9 and S10). Because albuterol does mucus secretion (Fig. 4, C and D). Similar
trusion in a more clinically relevant way—after not impede the ability of Gd3+ to block extrusion results were found in OVA-primed mice (fig. S5).
the onset of bronchoconstriction and in the (Fig. 2, G and H; fig. S4C; and movies S7 and S8)
presence of albuterol. Thus, we triggered bron- and is necessary for opening airways in patients, Corticosteroid-treated asthma patients have
choconstriction with MCH for 15 min and we used albuterol with Gd3+ treatment in all of marked airway epithelial extrusion and damage
then added SAC inhibitors in the presence our live mouse studies to reduce total numbers Having found that bronchoconstriction causes
of MCH. Gd3+ administered after broncho- of mice. Histology slices 30 min after MCH excess airway epithelial cell extrusion and
constriction substantially reduced extrusion challenge with or without albuterol resulted in destruction in mice, we investigated whether
and epithelial denuding (Fig. 2, F and G). high numbers of extrusion (Fig. 3, B and C, red a similar scenario occurs in humans. We ob-
Additionally, the peptide inhibitor, GsMTx4, arrowheads) and denuded bronchioles (Fig. 3, tained small airway resections from asthma
which blocks Piezo1, TRPC1, and TRPC6 (22), B and C, red outline). Yet, the addition of Gd3+ patients undergoing lobectomy for cancer.
also blocks extrusion when added 15 min after with albuterol preserved airway epithelia, simi- Biopsy samples from individuals with either
bronchoconstriction (Fig. 2F). Albuterol did larly to no MCH challenge (NM) (Fig. 3, B and moderate (Fig. 5A) or severe (Fig. 5B) asthma
not impair the ability of Gd3+ to block epi- C). The damage incurred during bronchocon- had even more extrusion, denuding, and mucus
thelial extrusion or denuding after 15 min of striction with or without albuterol resulted in secretion compared with our primed mice after
MCH-induced bronchoconstriction, nor did Gd3+ pronounced immune cell infiltration by 24 hours bronchoconstriction. Both patients were treated
affect bronchodilation by albuterol (Fig. 2, G compared with immune-primed unchallenged with both corticosteroids and smooth muscle–
and H, and movies S7 and S8). Our videos bronchioles (Fig. 3, D to F). Inhibiting extru- relaxing medications, underscoring their in-
show that Gd3+ added to albuterol allows the sion with gadolinium or SKIs (with albuterol) ability to impede airway epithelial damage
airway epithelia to reattach to the underlying reduced the inflammatory response to levels from asthma attacks. These individuals are
smooth muscle as it relaxes, thereby prevent- seen in control immune-primed, unchallenged not unusual, as many reports note airway epi-
ing denuding (Fig. 2H, blue arrowheads after bronchioles (Fig. 3, D to F). Similar results were thelial sloughing or extrusion as a defining
they detach, with red arrowheads indicating obtained with OVA-primed mice, even when pathological feature in mild to severe or fatal
extruded cells; movies S7 and S8). It is un- Gd3+ was delivered after the MCH ramp-up (fig. cases, even in racehorses (28–30), which sug-
clear why blocking upstream in the extrusion S4). Thus, preventing bronchoconstriction- gests that excess airway epithelial extrusion
pathway allows epithelial reattachment, but induced extrusion, particularly with gadolinium, is a bronchoconstriction-associated pathology
we suggest that it prevents the secretion of prevents the inflammation that typically fol- conserved throughout different species, inde-
matrix proteases predicted to be required during lows an attack. pendent of airway size.
the extrusion process.
SAC inhibitors also prevent mucus secretion Discussion
Blocking extrusion during bronchoconstriction Although asthma typically causes greater dam- In this work, we present a mechanical etiology
prevents inflammation age to the distal airways, most asthma patients for asthma. Whereas most asthma studies have
The ability of Gd3+ to prevent bronchoconstriction- experience notable difficulties with excess focused on the inflammatory signaling associ-
induced airway epithelial extrusion and de- mucus secretion from the proximal larger ated with asthma, our work suggests that the
struction suggested that it may prevent the airways. Immune priming with OVA or HDM inflammation and mucus secretion result from
inflammation that typically follows an asthma predictably increased mucus production and the mechanics of bronchoconstriction on airway
attack. To test this hypothesis, we challenged secretion, as measured by quantitative reverse epithelium. We find that bronchoconstriction
live HDM-primed mice with albuterol and in- transcription polymerase chain reaction (qRT- causes pathological airway epithelial crowding,
hibitors of extrusion—compared with control PCR) (fig. S1E), periodic acid–Schiff (PAS) leading to so much cell extrusion that it destroys
and MCH alone—and assayed for extrusions, staining of primary airways (fig. S1, E and F), the barrier, resulting in the typical postattack
denuding [30 min post-MCH (PM)], and in- and Muc5AC immunostained confocal projec- inflammation (Fig. 5C). Because epithelia act
flammation (24 hours PM) by hematoxylin tions (Fig. 4A) (14, 15, 24–27). Live imaging as the first line of defense against pathogens
and eosin (H&E) staining. We identified 50 mg/ml of primary airways loaded with wheat germ and toxins, epithelial barrier disruption could
MCH (1/10 the concentration used in ex vivo agglutinin–350 (WGA-350) to label mucus from also cause infection hypersensitivity until air-
experiments) as the lowest dose that would 3-week HDM-primed mice revealed that MCH ways have repaired. Infections and inflamma-
trigger bronchoconstriction and extrusion induces rapid mucus secretion with broncho- tion could then lead to more bronchoconstrictive
attacks, triggering the asthma inflammatory Because blocking extrusion upstream in the 34. A. Tam, S. Wadsworth, D. Dorscheid, S. F. Man, D. D. Sin, Ther.
cycle (31). Albuterol treatment, routinely used pathway with gadolinium preserves the airway Adv. Respir. Dis. 5, 255–273 (2011).
35. J. A. Mitchel et al., Nat. Commun. 11, 5053 (2020).
by asthma patients for symptom relief of bron- barrier after a bronchoconstrictive attack, it 36. J.-A. Park et al., Nat. Mater. 14, 1040–1048 (2015).
chospasm, does not prevent airway epithelia may also prevent the ASM remodeling asso- 37. I. T. Stancil et al., Nat. Commun. 12, 4566 (2021).
destruction, mucus secretion, or inflammation ciated with wound healing (41) and S1P pro- 38. C. E. Deering-Rice et al., Am. J. Respir. Cell Mol. Biol. 53,
893–901 (2015).
after an asthma attack. However, blocking duction (23, 42, 43) that causes future attacks. 39. C. E. Deering-Rice et al., J. Biol. Chem. 291, 24866–24879
the extrusion pathway with gadolinium or Thus, preventing the mechanical damage caused (2016).
S1P inhibitors after bronchoconstriction pre- by an asthma attack could pave the way for ther- 40. W. C. Wetsel, Int. J. Hyperthermia 27, 388–398
(2011).
served epithelial integrity, substantially damp- apies that stop the whole asthma inflammatory 41. N. K. Malavia et al., Am. J. Respir. Cell Mol. Biol. 41, 297–304
ening the inflammatory response. Gadolinium cycle rather than treating only the downstream (2009).
also blocked mucus secretion from primary air- symptoms. Moreover, excess smooth muscle con- 42. E. Fuerst et al., Allergy 69, 1531–1539 (2014).
43. T. J. A. Maguire et al., J. Allergy Clin. Immunol. 152, 1131–1140.e6
ways, which suggests that crowding activation striction may underlie other inflammatory syn-
(2023).
of calcium channels mechanically induces bulk dromes, especially those linked with cramping,
mucus secretion. such as irritable bowel syndrome or inflamma- AC KNOWLED GME NTS
Although our experiments defined excess tory bowel disease, yielding therapeutic ap- We thank R. Krishnan and A. Locke for help with initial experiments
and advice for our paper and G. Jenkins for introducing us to the
airway epithelial cell extrusion as important proaches for these unsolved medical problems.
Institute for Lung Health team. We thank S. Mitchell, C. Pardo Pastor,
for triggering many symptoms after an asthma M. Redd, T. Zulueta-Coarasa, and D. Jackson for helpful
attack, we are not the first to examine the RE FERENCES AND NOTES comments on our manuscript. We are grateful to the vibrant
effects of the mechanics of epithelia in asthma. asthma research community in London, which has provided
1. P. G. Holt, P. D. Sly, Nat. Med. 18, 726–735 (2012).
invaluable insight into this disease. We thank the Huntsman Cancer
Many have noted the important role of epi- 2. R. J. Hopp, A. K. Bewtra, N. M. Nair, R. G. Townley, J. Allergy
Institute and King’s College London Mouse Facilities; the
Clin. Immunol. 74, 154–158 (1984).
thelia in asthma (4, 32–34). Additionally, com- 3. F. Menzella et al., Multidiscip. Respir. Med. 16, 787
Fluorescence Microscopy Facility at the Health Sciences Cores at
pression forces can cause airway epithelia to the University of Utah; and the Huntsman Cancer Institute, Queen
(2021).
Mary University of London, King’s College London, Imperial College
unjam from their previous static, polygonal 4. S. T. Holgate, J. Allergy Clin. Immunol. 120, 1233–1244
London, University of Southampton Pathology Cores, and the
(2007).
state and migrate collectively—a phenomenon 5. S. T. Holgate, Curr. Opin. Pulm. Med. 15, 63–71
Innovation Hub at King’s College London and Confocal Imaging
linked to both asthma and idiopathic pulmo- Core at University College London for tissue section handling and
(2009).
scanning. Funding: This study was supported by an American
nary fibrosis (35–37). We occasionally noted 6. C. G. Persson, Am. J. Respir. Crit. Care Med. 153, S9–S10
Asthma Foundation Award 16-0020, National Institute of Health
(1996).
airway epithelial streaming or unjamming after R01GM102169, a Howard Hughes Faculty Scholar Award 55108560,
7. C. Xiao et al., J. Allergy Clin. Immunol. 128, 549–556.e12
MCH treatment, which could also contribute and a Wellcome Investigator Award 221908/Z/20/Z to J.R. P30
(2011).
CA042014 awarded to the Huntsman Cancer Institute core and an
to the forces driving extensive extrusion and 8. G. T. Eisenhoffer et al., Nature 484, 546–549 (2012).
NCRR Shared Equipment grant no. 1S10RR024761-01 for the
denuding of airways. The reduced cell numbers 9. J. Rosenblatt, M. C. Raff, L. P. Cramer, Curr. Biol. 11, 1847–1857
microscopy core supported this work. This study and research
(2001).
resulting from excess extrusions after strong 10. G. T. Eisenhoffer, J. Rosenblatt, Trends Cell Biol. 23, 185–192
were funded by MRC MR/S009191/1 to M.P. and the National
Institute for Health and Care Research (NIHR) Leicester Biomedical
bronchoconstriction could also trigger a wound (2013).
Research Centre (BRC) to C.B. The views expressed are those of
healing response that would not only result in a 11. J. R. Castillo, S. P. Peters, W. W. Busse, J. Allergy Clin.
the authors and are not necessarily those of the NIHR or the
Immunol. Pract. 5, 918–927 (2017).
poor barrier but also in matrix deposition and 12. M. M. Gueders et al., Inflamm. Res. 58, 845–854
Department of Health and Social Care. Author contributions: J.R.
fibroblast activation that could promote more designed experiments, interpreted and analyzed data, and wrote
(2009).
the manuscript. D.C.B., T.R., and K.F. designed experiments,
reactive airways. 13. S. R. Rosner et al., Am. J. Respir. Cell Mol. Biol. 50, 876–881
interpreted data, and performed most live and fixed imaging
(2014).
Our studies define the extrusion pathway in 14. W. R. Henderson Jr. et al., J. Exp. Med. 184, 1483–1494
experiments throughout the paper. P.F.R. and M.J. assisted with
controlling the downstream symptoms of an mouse priming, imaging, and quantification of imaging results of
(1996).
many of the experiments. C.R. and C.D.-R. consulted on
asthma attack. Admittedly, our studies do not 15. W. R. Henderson Jr. et al., Am. J. Respir. Crit. Care Med. 165,
experiments and pathology sections. E.O.-Z. provided HDM-primed
108–116 (2002).
address whether Piezo1 or TRP channels spe- 16. J. R. Johnson et al., Am. J. Respir. Crit. Care Med. 169,
mice and helped with the imaging of ex vivo slices. S.S. and C.B.
cifically control extrusion in response to bron- provided lung slices from asthma patients. All authors edited the
378–385 (2004).
manuscript. Competing interests: C.B. reports grants and
choconstriction. Thus, future studies will need 17. A. Tomkinson et al., J. Immunol. 166, 5792–5800
consultancy fees paid to his institution from Areteia, AstraZeneca,
(2001).
to determine which channels respond to dif- Chiesi, Genentech, GlaxoSmithKline, Mologic, Regeneron
18. K. G. Tournoy, J. C. Kips, C. Schou, R. A. Pauwels, Clin. Exp.
ferent asthma triggers. A variety of TRP chan- Pharmaceuticals, Roche, and Sanofi, which are not related to the
Allergy 30, 79–85 (2000).
current manuscript. All other authors declare that they have no
nels are not only mutated in many cases of 19. K. Ulrich et al., Pulm. Pharmacol. Ther. 21, 637–647
competing interests. Data and materials availability: All data are
poorly controlled asthma but are activated by (2008).
available in the main text or the supplementary materials. License
20. Y. Gu, T. Forostyan, R. Sabbadini, J. Rosenblatt, J. Cell Biol.
numerous stresses that trigger asthma attacks, 193, 667–676 (2011).
information: Copyright © 2024 the authors, some rights reserved;
exclusive licensee American Association for the Advancement of
including pollution, smoke, and cold, which 21. M. S. Grace, M. Baxter, E. Dubuis, M. A. Birrell, M. G. Belvisi,
Science. No claim to original US government works. https://www.
suggests that several channels may be rel- Br. J. Pharmacol. 171, 2593–2607 (2014).
science.org/about/science-licenses-journal-article-reuse. This
22. C. Bae, F. Sachs, P. A. Gottlieb, Biochemistry 50, 6295–6300
evant (21, 38–40). Thus, Gd3+ has the advan- (2011).
article is subject to HHMI’s Open Access to Publications policy.
tage of generically blocking extrusion upstream HHMI lab heads have previously granted a nonexclusive CC BY 4.0
23. M. M. Price et al., J. Allergy Clin. Immunol. 131, 501–511.e1
license to the public and a sublicensable license to HHMI in their
in the extrusion pathway. Additionally, tran- (2013).
research articles. Pursuant to those licenses, the Author Accepted
24. C. M. Evans, K. Kim, M. J. Tuvim, B. F. Dickey, Curr. Opin. Pulm.
sient use of Gd3+ has the added benefit that it Med. 15, 4–11 (2009).
Manuscript (AAM) of this article can be made freely available
may be used practically once an attack occurs under a CC BY 4.0 license immediately upon publication.
25. C. M. Evans et al., Nat. Commun. 6, 6281 (2015).
without apparent side effects in mice, but its 26. Z. Jia et al., Mucosal Immunol. 14, 125–134 (2021).
27. S. Piñeiro-Hermida et al., Allergy 72, 1317–1326
safety will need to be tested in human airways. (2017).
SUPPLEMENTARY MATERIALS
Gd3+not only serves to establish a role for 28. A. Hsieh, N. Assadinia, T.-L. Hackett, Front. Physiol. 14, 1113100 science.org/doi/10.1126/science.adk2758
Materials and Methods
mechanically induced extrusion in driving (2023).
29. R. C. Hyzy et al., Respir. Med. 102, 1819–1826 Figs. S1 to S5
asthma symptoms but also suggests that References (44–47)
(2008).
targeting extrusion upstream in the pathway 30. F. ter Woort, J. L. Caswell, L. G. Arroyo, L. Viel, Am. J. Vet. Res. MDAR Reproducibility Checklist
may prevent downstream symptoms. If Gd3+ 79, 342–347 (2018). Movies S1 to S13
is not safe in humans, development of other 31. M. Yamaya, Pulm. Med. 2012, 834826 (2012).
32. S. T. Holgate, Immunol. Rev. 242, 205–219 (2011). Submitted 11 August 2023; resubmitted 9 December 2023
early extrusion inhibitors may provide ther- 33. S. T. Holgate, G. Roberts, H. S. Arshad, P. H. Howarth, Accepted 12 February 2024
apeutic benefit. D. E. Davies, Proc. Am. Thorac. Soc. 6, 655–659 (2009). 10.1126/science.adk2758
H
interaction, and clothing serves as a being battery-free (10), chip-free (16), washable textile system (Fig. 1A). A conventional chip-
means of communication between the (17), and weaveable (18). based electronic textile (e-textile) system typ-
individual and their environment, mak- In pursuit of the above vision, we sought to ically integrates processor chips, circuit elements
ing it a major element of this interaction develop a non–von Neumann architecture that (diodes, resistors, and capacitors), sensors, and
(1, 2). With advances in wearable technology, enables the integration of electronic assem- a radio frequency antenna onto flexible poly-
digital device–enabled textile electronics have blies into a single fiber. The fiber itself can imide films (25, 26) (fig. S1). Structural engi-
gradually become a means to expand this type perform the roles of wireless transfer, sensory neering of rigid electronic components and
of interaction (3, 4). Textile electronic systems processing, and feedback, making it a build- organic substrate has enabled circuit-level
are designed to equip textile or fiber assem- ing block for electronic textiles (19). If one or flexibility, however, current flexible printed
blies with electronic functions for sensing, more interactive functions of chip-based tex- circuits (FPCs) hinder the conformal attach-
computation, display, and communication (5, 6). tile systems are to be realized in a wireless and ment of e-textiles to the skin (14). Further-
These systems typically comprise, but are not chipless form, the main obstacles are as fol- more, the multinode sensing and display
limited to, sensor and effector devices, en- lows. (i) For energy interaction: In a single module requires a number of flexible electrode
ergy storage and harvesters, and silicon- fiber, simultaneously confining and radiating leads to be attached to the integrated circuit
based processors (7, 8). Yet such von Neumann energy or electrical signals appears to be a (IC) surface, causing connectivity issues be-
architecture–based modular electronic sys- paradox, because no energy interaction mech- tween nodes (14, 27). Once the FPC and textiles
tems for intelligent textiles pose challenges for anism exists that can dynamically switch be- are multinode integrated, in most cases it is
seamless integration (9, 10), energy efficiency tween oscillating and confining charge pairs, almost impossible to disassemble them, which
(11), and wearing comfort (12, 13). These chal- as previously recognized in fiber electronics will make it difficult to carry out subsequent
lenges arise because current architecture— (20, 21). In the topology of textiles, the ad- daily washing (8). In contrast, our proposed
including wireless modules, microprocessors, jacent electronic fibers are exposed to the interactive fiber (i-fiber) eliminates the con-
and analog-to-digital converters (ADCs)—rely same electromagnetic (EM) environment, and straints of chips and battery assemblies and
on intrinsically rigid integrated circuit chips transforming the charge-pair state by relying can be seamlessly integrated into modern tex-
(stiffness: 107 to 109 N·m−1; curvature radius: on the induced EM field of the surrounding tile industry, addressing the issues of comfort-
10−6 to 10−3 m) (14), which have high power electronic fibers is difficult. (ii) For signal mod- ability, integration, and endurance.
consumption and are unable to conform seam- ulation: The current signal modulation principle The i-fiber consists of two functional units
lessly to soft fibers and textiles (stiffness: 103 relies on complex analog circuits and numer- (Fig. 1B): the wireless receiver and the wireless
to 105 N·m−1; curvature radius:10−3 to 10−2 m) ous electronic components, which are difficult transmitter, the latter of which includes both
(6, 15). Therefore, a paradigm shift in modular to integrate into commonly used soft fibers sensing and display parts. The i-fiber also con-
textile electronics is necessary and requires an (5, 22). Thermal drawing technology offers so- sists of three functional layers: the antenna
entirely new wireless energy interaction mech- lutions for fabricating digital fiber. However, core (silver-plated nylon fibers; Fig. 1B, red)
given the constraints of chip size at the milli- for inducing an alternating EM field (Ei), the
meter scale, achieving further reductions in dielectric layer (BaTiO3 mixed resin; Fig. 1B,
1
State Key Laboratory for Modification of Chemical Fibers the digital fiber’s diameter by embedding the white) for storing coupled EM energy, and the
and Polymer Materials, College of Materials Science and chip becomes challenging (3, 4). Furthermore, optical layer (ZnS:Cu2+ mixed resin; Fig. 1B,
Engineering, Donghua University, Shanghai 201620,
digital-based electronic interfaces are not naked- blue) for visualizing the electric field. Mean-
P. R. China. 2Institute for Health Innovation and Technology,
National University of Singapore, Singapore 117599, eye readable (2, 23, 24) nor are they artificially while, the interacting object itself, the human
Singapore. 3Biomass Molecular Engineering Center, College modulatable, hindering immersive interactive body, acts as the other pole for confining the
of Light-Textile Engineering and Art, Anhui Agricultural experiences for user-oriented intelligent textiles. ambient EM field (figs. S2 and S3). When the
University, Hefei 230036, P. R. China.
*Corresponding author. Email: wanghz@dhu.edu.cn (H.W.); We propose a body-coupled energy inter- i-fiber comes in contact with human skin, an
hcy@dhu.edu.cn (C.H.); zhangqh@dhu.edu.cn (Q.Z.) action mechanism for fiber electronics that interface contact capacitance is formed and
A B
Wireless receiving Wireless transmitting (Sensing / display)
Sensing i-fibers
Non-conformal fibers
1 2 N Radio wave
Power
module
Body-coupled
Capacitance
Visible light
Multi-node integration Wireless
module Ei
Dielectric layer
Display
fibers Fiber antenna
Sensing / display layer
EW
Wireless power
Unwashable
Conventional textile Our body-coupled textile
electronic system electronic system
C D
Body-coupled
capacitance
Fig. 1. Design and principle of the body-coupled interactive fiber. (A) Comparison between (left) current wireless interactive textile system based on integrated
circuit chips and (right) our chipless, wireless interactive textile system. (B) The principle of body-coupled chipless interactive fiber. (C) Photograph of i-fiber
being wirelessly powered by placing it on the hand and coupling the surrounding EM energy. (D) The embroidered i-fiber can simultaneously generate both optical and
electrical signals through touching.
excites the optically active layer to generate Ambient EM energy harvesting by the ground. This means that some of the EM
visible light (figs. S4 and S5). When electric body-coupled interactive fiber energy that was lost in the original environment
field intensity of the interface capacitance EM radiation—from radio signals, the electrical will be collected through the coupling path,
surpasses the critical value of air breakdown power grid, and even emerging triboelectric the human body path, and the return path.
(~3 × 10−3 V·mm−1), it will lead to a localized signals in daily life—is becoming increasingly We show the equivalent circuit model in
plasma discharge effect. This creates an addi- ubiquitous (28). Harvesting this radiation for Fig. 2C and fig. S6. The conductive core of the
tional electric displacement term @D, which powering distributed electronics through a i-fiber can be viewed as a series of inductor Lf
breaks the equilibrium state of confining charge body-coupled (29–32) strategy in a chipless and resistor Rf (fig. S7A). The power source
pairs, thereby realizing wireless transmission form has yet to be explored. Here, we present and the fiber itself form a capacitive path rep-
of electrical signals for tactile sensing. When a body-coupling ambient EM energy harvest- resented by a capacitance Ccoupling through
the soft i-fiber is placed in the palm (Fig. 1C), ing approach enabled by a single i-fiber (Fig. 2, electrical field coupling. When the human
the interface contact capacitance can effective- A and B). The human body has a remarkably body is in contact with the luminous material
ly capture the ambient EM energy and induce high relative permittivity and conductivity on the fiber surface, the skin interface con-
the i-fiber to emit light (movie S1 and fig. S6). (e ≈ 78 and d ≈ 0.6 S·m−1, respectively) com- tact capacitance CIC is generated, thus guid-
We show the simultaneous generation of opti- pared with air (e ≈ 1 and d ≈ 10−14 S·m−1), mak- ing the electrical energy flow to the human
cal and electrical signals when the finger comes ing it an ideal carrier for coupling EM energy body (body path) and the ground (return path).
in contact with the embroidered i-fiber (Fig. (32). The EM energy dissipated in the envi- Compared with typical bipolar capacitive power
1D). Our e-textile system only contains i-fibers ronment (by smartphones, power cables, fab- transfer (33), our body-coupled fiber electron-
without chips, batteries, or any other extra ric friction, etc.) can form a closed energy loop ics system only has one pair of electric field
components (movie S2). through fiber electronics, the human body, and coupling plates, which helps to simplify the
A B Wireless power
Body path Coupled path
Luminescent coating
C
Transmitter Fiber electronics Rx
Lf Rf
Power cable Ccoupling
C IC
Fiber electronics
…
Tx Human body
CPP CPP
D E F
δ-
δ+ δ+
δ+ (1.84 D)
δ+ δ+ H 2O
δ+ δ+
δ+
(0 D)
Cyclohexane
G 1
i In the air ii In the air iii Underwater
Fig. 2. Ambient EM energy harvesting by body-coupled interactive fiber. humidity). Both measurements were performed at a distance of 20 cm from
(A) Schematic of i-fiber harvesting energy from various ambient EM fields, the transmitter (30 dBm, ~20 kHz). (E) Received voltage and luminous
including textile friction (~3 Hz), power cable (~23 kHz), and smartphone brightness of i-fiber for body-coupled EM field energy harvesting at different
(~13.56 MHz). (B) Illustration of body-coupled fiber electronics. (C) Circuit model distances from the external EM source (30 dBm, ~20 kHz) (data are mean ± SD
of body-coupled wireless power transfer with i-fiber. The transmitter is the from three repeated measurements). (F) Effects of ambient media (air and
external EM signal source. Ccoupling, spatial coupling capacitance; CIC, interface liquids) on the ability of the i-fiber to capture EM energy and power itself to emit
contact capacitance; CPP, parasitic path capacitance. (D) Power spectra of light (red: data are mean ± SD from three repeated measurements). (G) Electric
ambient EM waves coupled by the human body and air inside an office field distribution |Exy| emitted by a dipole in the air or underwater, with or without
(environmental parameters: National University of Singapore, 22°C, 60% relative body coupling to the ground. a.u., arbitrary units.
textile wireless power transfer system. Ac- thereby realizing the normal operation of the not rely on specific working conditions. There-
cording to the traditional Kirchhoff’s current i-fiber (fig. S6). fore, it is expected to be useful in various ap-
law (33, 34), our fiber electronics system seems We thoroughly investigated the mechanism plications. We measured the power spectrum
like it would not work because there is no of EM field confinement by the i-fiber, as well of ambient EM waves coupled by the human
physical current-return path between trans- as the influence of body coupling, receiving body and air inside an office environment
mitter and receiver. But in fact, because each distance, fiber parameters (length, thickness, (Fig. 2D). In the range of 0 to 15 MHz, the
conductor presents a capacitive property to position), and environmental parameters (tem- EM energy collected under the coupling of the
infinity, this floating fiber electronics system perature, humidity, and location) (figs. S7 to human body is substantially higher than that
will form a parasitic coupling capacitance with S11). The results indicate that body-coupled of air coupling. In addition, as the distance
the ground as the return path for current, EM energy harvesting from a single fiber does between the fiber and the signal emission
source increases, the EM energy coupled to the the current density. To wirelessly transmit the an increase in the resonant frequency. As for
human body decreases continuously, which instantaneous tactile sensing signal, we needed the frequency (wavelength) of the optical sig-
makes the luminous brightness of the con- to break the balance of confining charge pairs nal, it is usually affected by the excitation sig-
tact interface of the fiber decrease with the at the contact interface and introduce a rapid- nal. An increase in the EM frequency would
same trend (Fig. 2E). In addition, we also ly varied @D or @B. produce a blue-shift of the dominant wave-
investigated EM field reception of the i-fiber We noticed that the intensity of the electric length of the spectra, which change from about
in various ambient media. In ultrapure water field between the skin and the i-fiber would 520 nm at 3 Hz to 430 nm at 13.56 MHz (Fig.
(Fig. 2F), the i-fiber exhibited an extraordinary increase as the distance decreased (dsf). Ex- 3F). Doping with Cu2+ in ZnS would establish
EM energy reception capability as compared ceeding the critical value of air breakdown two substitutional centers, respectively, a shal-
with other gas or liquid mediums (movie S3). can create a localized plasma discharge, which low excited emission center (green light) and a
We speculate that differences in molecular causes ions and electrons to clash and migrate deep excited emission center (blue light).
configuration–induced polarity will make in turn at high speed between air molecules in Moreover, the transmission distance and
it exhibit different polarizabilities under the the gap (35, 36), resulting in an additional ra- directionality of wireless signals is crucial
EM field. For polar molecules, such as water pidly varied electric displacement field @D/@t. for user-oriented interaction. We measured
(1.84 D), the noncentrosymmetric molecular We depict the equivalent circuit model of the the attenuation of the optical and electrical
structure will deflect with the periodic EM LC oscillation for EM wave emission in our signals generated by a single fiber over dis-
field, which makes the EM energy in the form i-fiber in Fig. 3A. The high-k dielectrics is re- tance (Fig. 3G). The wireless electrical signal
of electric displacement polarization contin- sponsible for the radial interface capacitance can be transmitted effectively up to 30 m,
uously transfer among water molecules (Fig. Cf, whereas the conductive core provides the whereas the optical signal can still retain a
2F and fig. S12) (see supplementary text sec- axial fiber inductance Lf and resistance Rf. visible optical power intensity of 10 nW·cm−2
tion “Underwater constructing body-coupled Parasitic capacitance CPP with the ground is at a distance of 10 m. This can fully satisfy the
energy loop”). This results in most electric formed by the human body and the floated daily interaction distance. We also demonstrated
field energy being transferred to the surface fiber electronics, respectively. Therefore, when that the i-fiber can transmit the wireless optical
of the i-fiber through molecular polarization, a local discharge occurs at the contact inter- and electrical signals omnidirectionally, where-
so the i-fiber is capable of emitting light. From face capacitance CIC, the dominant frequency in the signal intensity detected in each direc-
the perspective of capacitive power transfer, of the EM wave radiated can be expressed as tion is nearly identical (Fig. 3H).
high permittivity of the environmental medi- 1
um, such as water (7.08 × 10−10 F·m−1), makes f ¼ pffiffiffiffiffiffiffiffiffiffiffi ð3Þ Continuous manufacturing, weaving, and
2p Lf ∗Cs Kawabata fabric style evaluation
it easy to obtain high coupling capacitance
between the polar plates. The dielectric mol- where Cs is the system capacitance, which is Besides their distinctive wireless energy inter-
ecules with a large dielectric constant will the series value of Cf, CPP, and CIC. Furthermore, action mechanism, fiber electronics should
increase the capacitive coupling coefficient because the air gap in the plasma discharge also possess inherent textile features—batch
between the i-fiber and the power source, so will form a conductive path, the interface im- weaving and wearing comfort (12), as mea-
that a smaller amount of EM energy is dis- pedance between the air and the i-fiber will be sured by factors such as continuity, fineness,
sipated in free space. substantially reduced, ensuring that the elec- softness, washability, etc. We sought a layer-
Our COMSOL simulation results further tric field remains strong on the luminance by-layer coating fabrication process based
prove that an individual i-fiber barely responds material. When the interface electric field be- on industry-level equipment and protocols.
in the EM environment of air (Fig. 2G, i). When tween the skin and i-fiber exceeds the critical High-k dielectrics and ZnS:Cu2+ phosphor was
the human body touches the surface of the intensity of the luminance material, the doped mixed in the double network resin and coated
i-fiber, part of the EM field is coupled to and carriers are excited to the conduction band on the surface of conductive fiber (Fig. 4A and
generated within the human body, thereby and then recombine with holes to radiate vis- fig. S14). This sort of scaling was only possible
causing local light emission (Fig. 2G, ii). Fur- ible light. We illustrate in Fig. 3B the process because we got rid of the electronic modules,
thermore, when the environmental medium is of generating optical and electrical signals which allowed the use of fashion-industry fab-
water, the surface of the entire fiber will have a (movie S4). The generation time of the break- rication techniques. The fibers have excellent
high electric field owing to the coupling with down electrical signal caused by touch is very softness (0.095 cN·cm2), fineness (300 mm),
the human body, resulting in uniform light short, within about 8 × 10−7 s, which greatly and breakage strength (56.4 MPa) (figs. S15
emission along the fiber (Fig. 2G, iii). eliminates the system delay caused by chip to S18). These values mean that the fibers are
calculation; from the optical signal, in addi- flexible and strong enough to be sewn and
Modulation and transmission of wireless tion to the visual response directly obtained by embroidered with a digital sewing machine
optical and electrical signals the naked eye, we can further extract infor- (Fig. 4, B and C, and figs. S19 to S23). Addi-
Faraday’s law (Eq. 1) and Ampere-Maxwell’s law mation such as touch period and touch area tionally, we introduced fluorescent dyes (the
can essentially be used to describe the state- (Fig. 3D and fig. S13). second excited source) to manipulate both
of-the-art wireless signal generation (Eq. 2) The fundamental characteristics of EM wave the inherent and emitted hues of the fiber
manipulation are frequency tuning and ampli- electronics (Fig. 4, D and E, and fig. S24).
@Β tude tuning. Our chipless modulation strategy Wearing comfort, critical for textile elec-
∇Ε ¼ ð1Þ
@t for wireless electrical signal is manipulating tronics, has a direct impact on the user’s wear-
the intrinsic radial capacitance of the i-fiber. ing experience and health. Air or moisture
As the thickness of high-k dielectrics increases permeability refers to the performance of air
@D
∇Η ¼J þ ð2Þ from 18 to 135 mm (Fig. 3, C and E), the main or H2O molecules through the fabric and is
@t frequency of the wireless electrical signal in- the most basic property in fabric permeabil-
Here, E and H are the electric field intensity creases in a positive correlation, and the am- ity (16). Conventional chip-based e-textile sys-
and magnetic field intensity, respectively. And plitude decreases in a negative correlation. tems exhibit local airtight (~7.4 mm s−1,
B and D are the magnetic flux density and the This observation is consistent with the rule ~0.114 g cm−2 h−1) and nonconformal con-
electric displacement field, respectively. J is that a decrease in the radial capacitance causes tact owing to the rigid chip-dense polyimide
A B
… Finger 1
E field
Electrical signal
Equivalent LC dsf
oscillator circuit Fiber
CIC ~MHz
Ionization
2
Cf
C D Electrical
E
Radial interface
capacitance Cf d
Optical
Touching period
F G H
~520 nm
~480 nm
Electrical signal
Optical signal
~440 nm
Fig. 3. Modulation and transmission of wireless optical and electrical signals. domains. (E) The influence of the thickness d of fiber dielectric coating on the
(A) Schematic diagram of the mechanism for wireless transmission of optical frequency and amplitude of electrical signals. (F) An increase in the EM frequency
and electrical signals by a single i-fiber. (B) Photograph of wirelessly transmitting would produce a blue-shift of the dominant wavelength of the spectra, which change
optical and electrical signals. Scale bar, 3.5 mm. (C) Cross-sectional view of the from about 520 nm at 3 Hz to 430 nm at 13.56 MHz. (G) Attenuation of wireless
i-fiber, including core fiber antenna, high-k dielectric coating, and luminance sheath. optical and electrical signals with distance. (H) Directional diagrams of wireless optical
Scale bar, 250 mm. (D) The i-fiber–induced optical and electrical signals in time and electrical signals (data are mean ± SD from three repeated measurements).
circuit system, which will cause discomfort to the comfort performance of textiles for various vents the fading of dyes and the shedding of
the local microenvironment of the skin. In end uses (Fig. 4H and figs. S26 to S31). We phosphors. The textile’s luminous brightness
contrast, our chipless wireless textile exhibits found that the use of our i-fiber can improve did not decay substantially, proving good wash-
good breathability (~604.8 mm s−1, ~1.99 g cm−2 wearing comfort in terms of fabric bending, ing resistance. We also characterized the i-fiber’s
h−1), nearly 100 times higher than chip-based shearing, and surface smoothness properties. wearable durability performance, including
e-textile systems. The textile is compliant any- Washing durability is another major issue im- long-term working stability, moisture resis-
where on the skin and meets the requirements peding the practical applicability of textile tance, sweat resistance, and abrasion resistance
for wearing comfort (Fig. 4, F and G, and fig. electronics. We performed standard washing (figs. S33 to S36). The results show robust optical
S25). Furthermore, we used the Kawabata and post-wash colorfastness tests (ISO 105/ and electrical properties of the i-fiber.
Evaluation System for Fabrics (KES-F) (37) to C10:2006) on chipless textile electronics (38).
quantitatively evaluate fabric style, by testing After 25 standard washes, our textile elec- Applications of chipless textile electronics
fabrics’ flexural, shear, tensile, and compres- tronic maintains its original appearance and We demonstrate various application scenar-
sive stiffness, as well as smoothness and fric- performance (Fig. 4I and fig. S32). The double- ios of the i-fiber (i-textile) (Fig. 5, fig. S37,
tion qualities, to identify fabric parameters for network cross-linked polymer effectively pre- and movie S1). We demonstrate the potential
Fig. 4. Continuous manufacturing, weaving, and Kawabata fabric style conventional chip-based e-textile and our chipless e-textile. (G) Comparison of a
evaluation. (A) Continuous fabrication of the i-fiber. (B) Digital embroidery of conventional chip-based e-textile and our chipless e-textile in terms of wearing comfort
the i-fiber with commercial textile. (C) Large-area (0.8 m by 1.5 m) weaving (air and moisture permeability) (data are mean ± SD from five repeated measure-
of the i-fiber on a rapier loom. (D) Multicolor i-fiber collected on reeling rollers. ments). (H) Kawabata fabric style evaluation of the e-textile and commercial textile.
Scale bar, 6 cm. (E) The dyed fibers show different luminous colors of white, green, and (I) Washability of our i-fiber woven textile, including color fastness and luminous
blue under body-coupled EM fields. Scale bar, 3.5 mm. (F) Schematic diagram of a intensity stability (data are mean ± SD from three repeated measurements).
application of this i-textile in providing as- tile display system almost latency-free (micro- application of chip-free wireless interaction in
sisted optical communication for the deaf (Fig. second level) and requires no additional elec- future textile optical communication (movie S5).
5A). This chipless display textile could allow tronic components or energy supply units Most virtual reality and augmented reality
the wearer to convey information in a comfort- (Fig. 5, B and C, and figs. S39 and S40) (see devices require wearing rigid, bulky helmets
able, continuous, and real-time manner. It con- supplementary text section “Body-coupled chip- for interaction. Some fabric-based interactive
sists of a 644–independent pixel (23 rows by less display textiles”). Further, on the basis of suits have been developed, but these all use
28 columns) touchpad and a corresponding the body-coupled energy interaction princi- intrinsically rigid silicon-based chips (39),
textile display pad with the same pixel count ple, we designed clothing that integrates wire- which hinder a comfortable immersive, in-
(fig. S38). The light-emitting pattern will be less luminating patterns (displaying the teractive experience. Using body-coupled fiber
consistent with the trajectory of the finger “Donghua” logo and letters) and chipless dis- electronics technology, we developed a single
sliding in real time. Our body-coupled dynam- play function (Fig. 5C and fig. S41). We show fiber–enabled interactive textile and realized
ic textile display in different areas eliminates that 26 letters and 10 Arabic numerals can be real-time control of virtual games (Fig. 5E).
the signal modulation and output process of displayed in a body-coupling manner through We show the signal-flow block diagram of
the silicon-based chip (25). This makes the tex- our textile (Fig. 5D), suggesting the potential this electronic system in Fig. 5F. The human
Fig. 5. Applications of chipless textile electronics. (A) Potential application human–computer interaction. (F) Block diagram of i-fiber–based chipless game
of i-textile to provide assisted optical communication for the deaf. (B) A interaction. (G) Realizing chipless, wireless game control by a single fiber,
pixelated wireless textile electronics system enables tactile patterned displays including four independent digital logic controls. (H) Potential application of
without batteries or chips on fabric. Scale bar, 3 cm. (C) Comparison of frame i-textile in a smart home. (I) Block diagram of i-fiber–based chipless smart home.
rate and delay between our chipless textile display and a traditional chip-based (J) The surrounding EM field energy is coupled to the sole of the foot, which
textile display. (D) The chipless i-textile realizes a touch display of 26 letters and can activate fibers to visualize the touch area and wirelessly transmit sensing
10 Arabic numerals. (E) Potential application of i-textile in virtual reality and signals for controlling home appliance switches.
body can couple wireless EM fields and trig- on textile (movie S6). It can realize four kinds potential of wireless fiber electronics in a smart
ger the breakdown threshold. This sends the of independent logic control without relying home, we wove a large-area wireless haptic
encoded wireless signal at the fiber contact on the chip, which is important for textile in- carpet (Fig. 4C and 5H). When a person steps
interface, and the frequency and amplitude teraction applications (figs. S42 and S43). This on a specific location of the wireless carpet, the
are modulated by the i-fiber itself (radial ca- chipless textile interactive system effectively surrounding EM field energy is coupled to the
pacitance, Cf). The signal can travel a long dis- reduces the system delay caused by the user sole of the foot, which can activate fibers to
tance (~30 m) to the coil antenna terminal and transmitter’s signal collection, processing, visualize the touch area and transmit discharge-
be decoded and demodulated into the user analysis, and encoding under the traditional induced wireless sensing signals (Fig. 5, H
interface for interaction (Fig. 5G). We devel- von Neumann architecture (fig. S44) (see sup- and I). We demonstrated that the wireless EM
oped a game demonstration of Tetris control- plementary text section “Chipless interaction signal can be detected by the receiver and used
led only by the fiber electronics itself (Fig. 5G) textile for Tetris game”). To demonstrate the to control indoor electronics (Fig. 5J and fig. S45)
(see supplementary text section “Wireless car- ACKN OWLED GMEN TS used for game interaction with chipless textiles in this study
pets for smart homes”). Funding: This work was supported by the National Natural Science are available in Zenodo (40). All data are available in the main text or
Foundation of China (52131303 and 52073057); the National Key the supplementary materials. License information: Copyright ©
Research and Development Program of China (2022YFB3805801); 2024 the authors, some rights reserved; exclusive licensee American
Conclusions Association for the Advancement of Science. No claim to original
the DHU Distinguished Young Professor Program (LZA2023001); the
We have developed a demodularized fiber elec- Anhui Provincial Natural Science Foundation (2308085QE147); the US government works. https://www.science.org/about/science-
Fundamental Research Funds for the Central Universities (2232024Y-01); licenses-journal-article-reuse
tronics technology that uses the interactive
and Graduate Student Innovation Fund of Donghua University (CUSF-
object itself—the human body—to couple the DH-D-2021005). Author contributions: C.H., Q.Z., Y. Li, and H.W. guided SUPPLEMENTARY MATERIALS
ambient EM energy. This energy interaction the project. W.Y., C.H., H.W., and W.G. conceived of the idea and
science.org/doi/10.1126/science.adk3755
mechanism allows all electronic assemblies designed the experiment. W.Y. fabricated the fiber electronics. W.Y.,
Materials and Methods
S.L., Y.H., and J.W. performed the experiments and measurements.
to be merged in a miniature fiber and en- W.Y., R.L., C.J., X.Y., and X.X. did the circuit modeling and data analysis.
Supplementary Text
Figs. S1 to S45
ables wireless sensing, display, and logic inter- Y. Liu, K.L., J.S.H., and C.H. revised the manuscript. All authors analyzed
References (41, 42)
action functions without relying on any chips. the experimental data and prepared the figures and manuscript. All
Movies S1 to S6
authors discussed the results and reviewed the manuscript.
Wireless harvesting of EM energy and wireless Competing interests: The authors declare that they have no Submitted 18 August 2023; accepted 7 February 2024
transmission of interactive signals in fiber elec- competing interests. Data and materials availability: The codes 10.1126/science.adk3755
tronics are discussed in depth. Our approach
enables continuous, scalable manufacture of
conformable fiber electronics that fulfill the
requirements of smart clothing. Applications THERMOELECTRICS
in assisted communications, smart homes,
and virtual reality illustrate the wide-ranging Pseudo-nanostructure and trapped-hole release
applicability of this technology in wearable
electronics and smart clothing. induce high thermoelectric performance in PbTe
RE FE RENCES AND N OT ES Baohai Jia1†, Di Wu2†, Lin Xie1†, Wu Wang1, Tian Yu3, Shangyang Li1, Yan Wang1, Yanjun Xu4,
1. Y. Luo et al., Nat. Electron. 4, 193–201 (2021). Binbin Jiang5, Zhiquan Chen3, Yuxiang Weng4, Jiaqing He1*
2. H. Bai et al., Science 370, 848–852 (2020).
3. H. Y. Cheung, thesis, Massachusetts Institute of Technology
Thermoelectric materials can realize direct and mutual conversion between electricity and heat.
(2023).
4. G. Loke et al., Nat. Commun. 12, 3317 (2021). However, developing a strategy to improve high thermoelectric performance is challenging because of
5. A. F. Abouraddy et al., Nat. Mater. 6, 336–347 (2007). strongly entangled electrical and thermal transport properties. We demonstrate a case in which both
6. C. Chen et al., Chem. Rev. 123, 613–662 (2023).
pseudo-nanostructures of vacancy clusters and dynamic charge-carrier regulation of trapped-hole
7. S. Wang et al., Nature 555, 83–88 (2018).
8. J. Shi et al., Adv. Mater. 32, e1901958 (2020). release have been achieved in p-type lead telluride–based materials, enabling the simultaneous
9. M. Chen et al., Innovation (Camb.) 3, 100340 (2022). regulations of phonon and charge carrier transports. We realized a peak zT value up to 2.8 at 850 kelvin
10. R. Lin et al., Nat. Commun. 11, 444 (2020). and an average zT value of 1.65 at 300 to 850 kelvin. We also achieved an energy conversion
11. J. Chen et al., Nat. Energy 1, 16138 (2016).
12. W. Yang et al., Nat. Commun. 10, 5541 (2019). efficiency of ~15.5% at a temperature difference of 554 kelvin in a segmented module. Our demonstration
13. S. Zeng et al., Science 373, 692–696 (2021). shows promise for mid-temperature thermoelectrics across a range of different applications.
14. X. Shi et al., Nature 591, 240–245 (2021).
15. W. Weng et al., Adv. Mater. 32, e1902301 (2020).
16. Y. Kim et al., Science 377, 859–864 (2022).
eat is an underutilized energy source in kl from lattice vibration and the electronic
H
17. D. Zhang et al., Adv. Mater. 33, e2100782 (2021).
18. W. Gong et al., Nat. Commun. 10, 868 (2019). both nature and society, and its accu- thermal conductivity ke from charge-carrier
19. Q. Shi et al., Adv. Fiber Mater. 1, 3–31 (2019).
20. A. Leber et al., Nat. Electron. 3, 316–326 (2020).
mulation and utilization have always transport (4–6). Obviously, an excellent thermo-
21. C. Caloz, T. Itoh, Electromagnetic Metamaterials: been popular areas of interest. Thermo- electric figure of merit requires both a high
Transmission Line Theory and Microwave Applications electric materials that enable direct and power factor (sS 2) and a low thermal conduc-
(Wiley, 2005).
22. Y. Qu et al., Adv. Mater. 30, e1707251 (2018).
inverse heat-to-electricity conversion have at- tivity (7–9). Unfortunately, these parameters are
23. C. Wang et al., Nat. Mater. 12, 899–904 (2013). tracted interest for their potential use across a firmly coupled to each other, making the inde-
24. W. Yang et al., Adv. Mater. 33, e2104681 (2021). variety of applications (1–3). The conversion ef- pendent tuning of each individual parameter
25. Y. Lee et al., Sci. Adv. 7, eabg9180 (2021).
ficiency of a thermoelectric module is related challenging (10–12). Many effective strategies
26. Y. H. Jung et al., Nat. Electron. 5, 374–385 (2022).
27. H. W. Choi et al., Nat. Commun. 13, 814 (2022). to the dimensionless figure of merit zT = sS 2T/k to enhance overall thermoelectric performance
28. X. Zhang et al., Nature 566, 368–372 (2019). of materials, where s, S, k, and T are the elec- have been proposed and confirmed, including
29. C. Lee et al., IEEE J. Solid-State Circuits 57, 3212–3227 trical conductivity, Seebeck coefficient, ther- optimizing (static or dynamic) carrier concen-
(2022).
30. J. Li et al., IEEE Trans. Biomed. Circuits Syst. 14, 1263–1273 mal conductivity, and absolute temperature tration (13, 14), level resonance (15, 16), band
(2020). in kelvin, respectively. The thermal conduc- convergence (17), energy filtering (18, 19), band
31. J. Yoo, IEEE Solid-State Circuits Mag. 15, 49–58 (2023). tivity includes the lattice thermal conductivity alignment (20–23), regulating the scattering
32. J. Li, Y. Dong, J. H. Park, J. Yoo, Nat. Electron. 4, 530–538
(2021).
mechanism of charge carriers (24, 25), construct-
33. F. Lu, H. Zhang, C. Mi, Energies 10, 1752 (2017). 1
Shenzhen Key Laboratory of Thermoelectric Materials, ing all-scale hierarchical defects (22, 26, 27),
34. L. J. Zou, A. P. Hu, Y. G. Su, IEEE PELS Workshop on Emerging Department of Physics, Southern University of Science and and entropy engineering (28).
Technologies: Wireless Power Transfer (WoW), Chongqing, Technology, Shenzhen 518055, China. 2Key Laboratory for
China, 20 to 22 May 2017 (IEEE, 2017), pp. 93–98. Macromolecular Science of Shaanxi Province, School of
Among these strategies, constructing nano-
35. H. Wang et al., Sci. Adv. 7, eabi6751 (2021). Materials Science and Engineering, Shaanxi Normal University, structures is believed to be able to regulate
36. Y. Fu, P. Zhang, J. P. Verboncoeur, X. Wang, Plasma Res. Express Xi’an 710119, China. 3Hubei Nuclear Solid Physics Key the lattice thermal conductivity without severe-
2, 013001 (2020). Laboratory, Department of Physics, Wuhan University, Wuhan
ly affecting other thermoelectric parameters
37. E. S. Namligöz, M. İ. Bahtiyari, A. E. Körlü, S. Çoban, J. Test. Eval. 430072, China. 4Laboratory of Soft Matter Physics, Beijing
36, 384–391 (2008). National Laboratory for Condensed Matter Physics, Institute of (29, 30). In general, nanostructure defects re-
38. Materials and methods are available as supplementary materials. Physics, Chinese Academy of Sciences, Beijing 100190, China. fer to an abundance of widely dispersed nano-
5
39. Z. Zhou et al., Nat. Electron. 3, 571–578 (2020). School of Materials and Energy, University of Electronic Science scale secondary phases that generate a large
40. W. Yang, Labview files and code files for Tetris game and Technology of China, Chengdu 610054, China.
interaction, version 1, Zenodo (2023); https://doi.org/10. *Corresponding author. Email: hejq@sustech.edu.cn lattice mismatch against the matrix. Thus, they
5281/zenodo.10421139. †These authors contributed equally to this work. can result in strong lattice distortion and strain
Fig. 4. Characterizations of the pseudo-nanostructures. (A) The ratio of weighted strength of the Ge elements, the Ge doping amount is 8%. The PDF pattern of pristine
hole mobility and lattice thermal conductivity versus Ge content of Pb0.97Na0.03Te- PbTe is a simulated result only to standardize the bond positions. cal., the calculated
2%MgTe-x%GeTe (x = 0, 0.25, 0.5, 0.75, or 1). (B) HAADF image of Pb0.97Na0.03Te- PDF pattern of pristine PbTe; G, probability to find another atom; r, real-space
2%MgTe-0.75%GeTe. (C) Strain analysis by geometric phase analysis of Pb0.98Na0.02Te- distribution of interatomic distances. (E) Positron lifetime component t2 and intensity
2%MgTe-4%GeTe (fig. S10C). eyy, normal strains along the yy direction. (D) The room- I2 in PbTe, Pb0.98Na0.02Te-2%MgTe-0.75%GeTe, and Pb0.97Na0.03Te-2%MgTe-
temperature PDF of PbTe, PbTe-GeTe, and Pb(Na, Mg)-GeTe. To increase the signal 0.75%GeTe. (F) HAADF image of Pb0.98Na0.02Te-2%MgTe-0.75%GeTe.
from 67.2 to 93.5% when Na content varied by the finite element method (fig. S15), we fab- 26. Y. Wu et al., Joule 3, 1276–1288 (2019).
from 0.02 to 0.03. The existence of Ge was also ricated an ultimate 8-by-8 segmented thermo- 27. S. I. Kim et al., Science 348, 109–114 (2015).
28. B. Jiang et al., Science 371, 830–834 (2021).
essential for the formation of vacancy clusters. electric module. To compensate for the internal 29. D. Wu et al., Adv. Mater. 28, 2737–2743 (2016).
By comparing the lattice thermal conductivities resistance increasing with decreasing cross- 30. K. Biswas et al., Energy Environ. Sci. 4, 4675–4684 (2011).
of the Pb0.97Na0.03Te-2%MgTe and Pb0.98Na0.02Te- sectional area, we enlarged the cross section of 31. A. F. Ioffe, Semiconductor Thermoelements and Thermoelectric
Cooling (Infosearch Ltd., 1957).
2%MgTe samples (fig. S4) without Ge doping, the Bi2Te3 legs to 4.25 mm by 4.25 mm. The 32. L. You et al., Energy Environ. Sci. 11, 1848–1858 (2018).
we concluded that Na content has a smaller reduction in the size of the PbTe leg led to a 33. Y. Zhang, J. Sun, J. Shuai, X. Tang, G. Tan, Mater. Today Phys.
influence on phonon transport than Ge does. reduction in the heat flow generated by heat 19, 100405 (2021).
34. Y. Qin et al., Adv. Funct. Mater. 31, 2102185 (2021).
The PAS results of Pb0.97Na0.03Te-2%MgTe- conduction, whereas the total module internal 35. G. Tan et al., Nat. Commun. 7, 12167 (2016).
x%GeTe (x = 0, 0.5, or 0.75) show that Ge resistance remained constant so that the joule 36. Y. C. Zhu et al., Energy Environ. Sci. 15, 3958–3967
doping does not introduce more Pb vacancies heat was almost constant (Fig. 5A); therefore, (2022).
37. H. J. Wu et al., Nat. Commun. 5, 4515 (2014).
(fig. S12A), and its t2 is unchanged because the total module heat flow was reduced. Ulti-
38. B. Jiang et al., Science 377, 208–213 (2022).
the vacancy clusters in the materials have the mately, the overall output power was not atten- 39. B. Jiang et al., Energy Environ. Sci. 13, 579–591 (2020).
same size distribution. However, statistical uated, but the heat flow was greatly reduced 40. X. Hu et al., Energy Environ. Sci. 9, 517–529 (2016).
results from STEM-HADDF images showed (Fig. 5B); in this situation, the conversion ef- 41. B. H. Jia et al., Energy Environ. Sci. 15, 1920–1929 (2022).
42. Y. Xing et al., Energy Environ. Sci. 12, 3390–3399 (2019).
that Ge doping greatly reduced the size of ficiency was improved up to ~15.5% (Fig. 5C). 43. Q. Zhang et al., Energy Environ. Sci. 10, 956–963 (2017).
Pb vacancy clusters (fig. S12, B and C). There- The conversion efficiency did not show obvious 44. L. D. Zhao et al., Energy Environ. Sci. 6, 3346–3355 (2013).
fore, it can be concluded that Ge is more like- deterioration during 15 hours of continuous 45. J. Speight, Lange’s Handbook of Chemistry (McGraw-Hill
Education LLC, ed. 16, 2005).
ly to regulate vacancy cluster morphology by measurement (Fig. 5D), which is a better sta- 46. Z. Luo et al., Energy Environ. Sci. 11, 3220–3230 (2018).
a means other than introducing more Pb va- bility performance than previously reported ther- 47. Y. Liu et al., J. Am. Chem. Soc. 145, 8677–8688 (2023).
48. S. C. Ke, T.-C. Wang, M.-S. Wong, N. O. Gopal, J. Phys. Chem. B
cancies. We need to address the function of moelectric modules in the literature (38–41). 110, 11628–11634 (2006).
Ge doping in regulating the vacancy cluster 49. N. M. Dimitrijevic, Z. V. Saponjic, B. M. Rabatic,
dimension, which results in enhanced pho- Conclusions O. G. Poluektov, T. Rajh, J. Phys. Chem. C Nanomater.
Interfaces 111, 14597–14601 (2007).
non scattering and reduced lattice thermal We improved the thermoelectric performance 50. J. L. Lyons, A. Janotti, C. G. Van de Walle, J. Appl. Phys. 115,
conductivity (fig. S4). To summarize, excess of p-type PbTe-based thermoelectric mate- 012014 (2014).
Na helps to generate extra cation vacancies in rials and modules through a trapped-hole 51. Y. Huang et al., Nanotechnology 34, 235707 (2023).
52. M. Wuttig, V. L. Deringer, X. Gonze, C. Bichara, J.-Y. Raty, Adv.
the PbTe matrix, whereas Ge doping reduces release mechanism that relies on a pseudo-
Mater. 30, 1803777 (2018).
the dimensions of cation vacancy clusters; to- nanostructure and various forms of doping. 53. X. Xu, L. Xie, Q. Lou, D. Wu, J. He, Adv. Sci. 5, 1801514 (2018).
gether, these effects have a large impact on The pseudo-nanostructures, which consist of 54. G. Martinez, M. Schlüter, M. L. Cohen, Phys. Rev. B 11,
the phonon transport behaviors. a cation vacancy cluster, effectively blocked 660–670 (1975).
55. P. Jood et al., J. Am. Chem. Soc. 142, 15464–15475 (2020).
phonon transport but allowed the passage of 56. Y. Pei, H. Wang, G. J. Snyder, Adv. Mater. 24, 6125–6135 (2012).
Module performance holes, thus achieving comprehensive electron- 57. S. A. J. Kimber et al., Nat. Mater. 22, 311–315 (2023).
58. A. A. Vinokurov, V. P. Zlomanov, M. Elsayed, R. Krause-Rehberg,
Using our p-type PbTe, we fabricated an 8-by-8 phonon decoupling. The hole trapping at low Mater. Lett. 64, 661–663 (2010).
single-stage thermoelectric module to demon- temperatures and release at high tempera-
strate its mid-temperature thermoelectric con- tures enabled the optimization of the hole AC KNOWLED GME NTS
version efficiency. The cross-sectional area ratio concentration over a broad temperature range. We thank H. Meng for performing the in situ MIR-TAS
characterizations. We also thank the SUSTech Core Research
of p-type and n-type thermoelectric legs in This advance in thermoelectric materials and Facilities for providing use of TEM and EPR instruments. D.W.
single-stage modules was 1:1 (the cross section modules could accelerate the practical appli- acknowledges support from the Fundamental Innovation Project in
of each thermoelectric leg was 4 mm by 4 mm cation of mid-temperature thermoelectric the School of Materials Science and Engineering (SNNU).
Funding: We acknowledge the National Natural Science Foundation
with a height of 11 mm; Pb0.987S0.01I0.003Te for the power-generation technology. of China (grant no. 11934007), the Science and Technology
n-type leg and Pb0.97Na0.03Te-2%MgTe-0.75% Innovation Committee Foundation of Shenzhen (grant nos.
GeTe for the p-type leg). The maximal output RE FERENCES AND NOTES JCYJ20200109141205978 and ZDSYS20141118160434515), the
1. J. He, T. M. Tritt, Science 357, eaak9997 (2017). Outstanding Talents Training Fund in Shenzhen (202108), and the
power is 2.75 W for this thermoelectric module,
2. Y. Xiao, L. D. Zhao, Science 367, 1196–1197 (2020). Strategic Priority Research Program of Chinese Academy of
and its maximal conversion efficiency reached 3. G. J. Snyder, E. S. Toberer, Nat. Mater. 7, 105–114 (2008). Sciences (XDB33000000). Author contributions: J.H. designed
~10.5% with a temperature difference of 553 K 4. J. Mao et al., Science 365, 495–498 (2019). this work. B.J. synthesized the samples, fabricated the modules,
(the temperature of the hot side was 850 K and 5. S. Roychowdhury et al., Science 371, 722–727 (2021). and carried out the thermoelectric property measurements.
6. T. Zhu et al., Adv. Mater. 29, 1605884 (2017). L.X. and W.W. performed the TEM characterizations and analysis.
that of the cool side was 283 K) (fig. S13). We 7. L. D. Zhao et al., Nature 508, 373–377 (2014). L.X. and Y.W. performed the DFT calculations. S.L. performed the
then fabricated another 8-by-8 thermoelectric 8. X. Zhou et al., Mater. Today 21, 974–988 (2018). finite element simulation. T.Y. and Z.C. conducted the positron
module segmented with n- and p-type Bi2Te3- 9. Q. Yang et al., Science 377, 854–858 (2022). annihilation lifetime spectra experiments. Y.W. and Y.X. performed
10. C. Chang et al., Science 360, 778–783 (2018). the MIR-TAS characterizations and analysis. B.J., D.W., L.X., and
based materials; the thermoelectric properties 11. X. L. Shi, J. Zou, Z. G. Chen, Chem. Rev. 120, 7399–7515 (2020). J.H. wrote the manuscript. All authors edited the manuscript.
of the Bi2Te3-based materials are shown in 12. W. Zhao et al., Nature 549, 247–251 (2017). Competing interests: The authors declare no competing interests.
fig. S14. The segmented module without size 13. Y. Pei et al., Adv. Mater. 23, 5674–5678 (2011). Data and materials availability: All data are available in the
14. Y. Pei et al., Adv. Energy Mater. 4, 1400486 (2014). manuscript and supplementary materials. License information:
optimization (the cross-sectional area ratio was 15. J. P. Heremans et al., Science 321, 554–557 (2008). Copyright © 2024 the authors, some rights reserved; exclusive
kept at 1:1; the height ratio of PbTe/Bi2Te3 was 16. B. Paul, P. K. Rawat, P. Banerji, Appl. Phys. Lett. 98, 262101 licensee American Association for the Advancement of Science. No
11:3) showed that the open-circuit voltage and (2011). claim to original US government works. https://www.science.org/
17. Y. Pei et al., Nature 473, 66–69 (2011). about/science-licenses-journal-article-reuse
output power were 1.83 V and 3.2 W at DT = 18. J. Martin, L. Wang, L. Chen, G. S. Nolas, Phys. Rev. B 79, 115311
554 K, respectively, and the maximal conver- (2009). SUPPLEMENTARY MATERIALS
sion efficiency reached 13.6% at DT = 554 K 19. T. Mori, Small 13, 1702013 (2017). science.org/doi/10.1126/science.adj8175
20. L. D. Zhao, V. P. Dravid, M. G. Kanatzidis, Energy Environ. Sci.
(Fig. 5). By optimizing the length ratio of the Materials and Methods
7, 251–268 (2014). Figs. S1 to S15
Bi2Te3 and PbTe segments (the optimal heights 21. K. Biswas et al., Nat. Chem. 3, 160–166 (2011). Tables S1 to S7
of the p-type and n-type Bi2Te3 were 3.6 and 22. K. Biswas et al., Nature 489, 414–418 (2012). References (59–75)
4 mm, respectively) and the cross-sectional 23. Y. Qin, B. Qin, D. Wang, C. Chang, L.-D. Zhao, Energy Environ.
Sci. 15, 4527–4541 (2022). Submitted 18 July 2023; resubmitted 21 January 2024
area ratio of the p-type and n-type legs [(4 mm 24. J. Shuai et al., Energy Environ. Sci. 10, 799–807 (2017). Accepted 4 March 2024
by 4 mm)/(3.45 mm by 3.45 mm) = 1.35:1] guided 25. B. Qin, L. D. Zhao, Science 378, 832–833 (2022). 10.1126/science.adj8175
and one social outcome, and thus are inher- based on a participatory, iterative process, in- farming systems to confirm and contextualize
ently interdisciplinary (mostly collected by ecol- cluding multiple group meetings and exchanges results. Fig. S5 provides an example illustra-
ogists and social scientists working together and with data contributors during all stages of var- tion that we used to confirm and contextualize
merging their “ways of knowing” and meth- iable selection, data analysis, and result inter- results with data contributors who worked
ods). Also, each dataset studied farm sites with pretation. Although our sample size of 24 studies with Malawian smallholders.
varying levels of diversification, including farms would be relatively small for a standard meta- We first tested how agricultural diversifica-
without any diversification practices. Unlike analysis, it is ideal for an approach requiring tion strategies affected each of the six tar-
data synthesis approaches that extract values extensive interaction with data contributors geted social and environmental outcomes. For
from published materials, our data synthesis is working across disparate geographies and each of the five diversification strategies, we
1
Department of Geosciences and Natural Resource Management, University of Copenhagen, Copenhagen, Denmark. 2Department of Ecology of Tropical Agricultural Systems, University of Hohenheim,
Stuttgart, Germany. 3Center for Biodiversity and Integrative Taxonomy (KomBioTa), University of Hohenheim, Stuttgart, Germany. 4Department of Environmental Studies, University of Colorado Boulder,
Boulder, CO, USA. 5Better Planet Laboratory, University of Colorado Boulder, Boulder, CO, USA. 6Mortenson Center for Global Engineering and Resilience, University of Colorado Boulder, Boulder, CO,
USA. 7Center for Global Change and Earth Observations, Michigan State University, East Lansing, MI, USA. 8Ecology, Evolution, and Behavior Program, Michigan State University, East Lansing, MI, USA.
9
Department of Global Development, Cornell University, Ithaca, NY, USA. 10School for Environment and Sustainability, University of Michigan, Ann Arbor, MI, USA. 11Universidad Nacional de Río Negro,
Instituto de Investigaciones en Recursos Naturales, Agroecología y Desarrollo Rural, Río Negro, Argentina. 12Consejo Nacional de Investigaciones Científicas y Técnicas, Instituto de Investigaciones en
Recursos Naturales, Agroecología y Desarrollo Rural, Río Negro, Argentina. 13Department of Physical and Environmental Sciences and Department of Global Development Studies, University of Toronto,
Toronto, Ontario, Canada. 14Global Science, The Nature Conservancy, Fort Collins, CO, USA. 15Centre for Sustainable Food Systems, University of British Columbia, Vancouver, British Columbia, Canada.
16
Institute for Resources, Environment and Sustainability, University of British Columbia, Vancouver, British Columbia, Canada. 17Lendület Landscape and Conservation Ecology, Institute of Ecology and
Botany, HUN-REN Centre for Ecological Research, Vácrátót, Hungary. 18Department of Plant Protection, Bogor Agricultural University, Jalan Kamper, Kampus Darmaga, Bogor, Indonesia. 19Centro
Agronómico Tropical de Investigación y Enseñanza (CATIE), Turri Alba, Costa Rica. 20Center for Research on Sustainable Agricultural Systems (CIPAV), Cali, Colombia. 21Department of Entomology,
Washington State University, Pullman, WA, USA. 22INRAE, EFNO Nogent-sur-Vernisson, France. 23Department of Environmental Science, Policy, and Management, University of California, Berkeley, CA,
USA. 24Department of Entomology, University of Kentucky, Lexington, KY, USA. 25Federación Colombiana de Ganaderos (FEDEGAN), Bogotá, Columbia. 26Department of Plant Protection, IPB University,
Bogor, Indonesia. 27Federal University of Bahia (UFBA), Biology Institute, Salvador, Brazil. 28Universidade Federal de Viçosa, Conselho de Ensino, Pesquisa e Extensão, Universidade Federal de Viçosa,
Campus Universitário, Viçosa, MG, Brazil. 29Brazil Instituto Nacional de Pesquisas da Amazônia, INPA, Manaus, AM, Brazil. 30Environmental Change Institute, School of Geography and the Environment,
University of Oxford, Oxford, UK. 31Urban and Regional Planning Program, University of Michigan, Ann Arbor, MI, USA. 32Department of Agricultural Economics and Business, University of Dar es Salaam,
Dar es Salaam, Tanzania. 33School of Public Health, University of Michigan, Ann Arbor, MI, USA. 34Department of Wildlife, Fish, and Conservation Biology, University of California-Davis, Davis, CA, USA.
35
Eco&Sols, Université de Montpellier, IRD, CIRAD, INRAE, Institut Agro, Montpellier, France. 36University of Toronto, Toronto, Ontario, Canada. 37Department of Sociology, University of Victoria, Victoria,
British Columbia, Canada. 38Department of Plant Ecology and Ecosystems Research, University of Göttingen, Göttingen, Germany. 39Biodiversity, Macroecology & Biogeography, University of Göttingen,
Göttingen, Germany. 40Consejo de Salud Rural Andino, La Paz, Bolivia. 41Department of Environmental Geography, Institute for Environmental Studies, Vrije Universiteit Amsterdam, Amsterdam,
Netherlands. 42Thünen Institute of Biodiversity, Johann Heinrich von Thünen Institute - Federal Research Institute for Rural Areas, Forestry, and Fisheries, Braunschweig, Germany. 43The Nature
Conservancy, Latin America North Andes and Central America Region, Bogota, Columbia. 44Applied Ecology Graduate Program, Luiz de Queiroz College of Agriculture, University of São Paulo, Piracicaba,
São Paulo, Brazil. 45Department of Geography & the Environment, University of Denver, Denver, CO USA. 46Conservation Science Partners, Truckee, CA, USA. 47Center for Development Research (ZEF),
University of Bonn, Bonn, Germany. 48Department of Animal Ecology and Tropical Biology, Biocenter, Julius-Maximilians-University Würzburg, Würzburg, Germany. 49Leibniz Institute for the Analysis
of Biodiversity Change (LIB), Museum Koenig, Centre for Biodiversity Monitoring and Conservation Science, Bonn, Germany. 50Bonn Institute for Organismic Biology, Faculty of Mathematics and Natural
Sciences, University of Bonn, Bonn, Germany. 51The Organic Center, Washington, DC, USA. 52Sustainable Agrifood Systems, International Maize and Wheat Improvement Center (CIMMYT), El Batan,
Mexico. 53Department of Entomology, University of Georgia, Athens, GA, USA. 54Sustainable Use of Natural Resources Department, Institute of Social Sciences in Agriculture, University of Hohenheim,
Stuttgart, Germany. 55Department of Agroecology, University of Göttingen, Göttingen, Germany. 56Farming Systems Ecology Group, Wageningen University and Research, Wageningen, Netherlands.
57
Department of Environment, Agriculture and Geography at Bishop’s University, Sherbrooke, Quebec, Canada. 58Department of Ecology, Swedish University of Agricultural Sciences, Uppsala, Sweden.
59
Institute for Resources, Environment and Sustainability, Biodiversity Research Centre and Department of Zoology, University of British Columbia, Vancouver, British Columbia, Canada.
*Corresponding author. Email: lr@ign.ku.dk
†These authors contributed equally to this work.
Fig. 3. Synergies and trade-offs among environmental and social outcomes strategies, with arrows pointing in the upper right corner indicating win-win
of agricultural diversification. (A and B) Methodological approach to identify outcomes, and arrows pointing in the other quadrants indicating either trade-offs
synergies and trade-offs based on the example of nonagricultural biodiversity and (top left quadrant, lose-win; bottom right quadrant, win-lose) or lose-lose (bottom
crop yield. (C) Synergies and trade-offs for all pairwise combinations of studied left quadrant) outcomes. (B) shows outcomes by diversification strategy. Circle
environmental and social outcome variables. (A) shows the effects of agricultural size is proportional to arrow length in (A) and indicates the effect strength, that is,
diversification strategies on nonagricultural biodiversity and crop yield. Arrow tips are the conjoined change in nonagricultural biodiversity and crop yield with diversification.
located at the predicted change in nonagricultural biodiversity and crop yield, In (C), colored circles indicate outcome combinations of environmental and
respectively, with an increase of 1 SD of a given diversification strategy. The x-y social outcome variables (black, win-win; orange, win-lose; blue, lose-win; green,
coordinates of the arrow tips are numerically similar to the effect sizes shown in lose-lose). Circle size is proportional to the joint change of the paired environmental
Fig. 2. The resulting arrow directions indicate trade-offs and synergies of diversification and social variables.
in which we kept the total number of observa- results for specific outcomes (e.g., ecosystem egies and practices maximized benefits across
tions constant but artificially up- or down- services) and diversification strategies (e.g., environmental and social outcomes (fig. S9).
weighted studies so that all were equally livestock diversification); however, equalizing We interpreted the results from both model
represented in the model fitting. In the equal- the influence of each study did not alter the sets as suggesting that different outcomes,
ized model, we observed some difference in key finding that multiple diversification strat- trade-offs, and synergies may be more present
Fig. 4. Landscape composition moderates the outcomes of agricultural diversification strategies. (A to F) Shown are standardized effect sizes (means and
95% confidence intervals) of the five diversification strategies and the total number of diversification strategies or practices applied, depending on the proportion of
seminatural habitat in a 3000-m-radius surrounding farms (cleared, <5%; simple, 5 to 20%; complex, >20%). When the confidence interval is not overlapping
with the 0 line, there is a significant effect in a given landscape.
in different contexts, but that the overall best of nonagricultural biodiversity and multiple biodiversity. Conversely, many national strat-
strategy to maximize environmental and so- social outcomes (Fig. 3C). Half of the farms in egies and programs focusing on agricultural
cial outcomes is to apply multiple diversifica- our study practiced some form of livestock intensification did not achieve win-wins (28).
tion strategies and practices in tandem. integration (14 studies surveyed farms with We also observed trade-offs, such as soil con-
livestock integration). Livestock diversifica- servation practices leading to gains in biodi-
Maximizing win-wins, minimizing trade-offs tion had the largest positive effect on food versity but potential yield losses, although the
We further examined the extent to which the security (0.23 ± 0.03), which is more than four latter were not statistically significant (Figs. 2C
five diversification strategies might promote times as large as the effect on yields. Livestock and 3, A to C). For noncrop diversification, we
pairs of environmental and social win-win out- diversification promoted nonagricultural bio- did not observe a consistent positive effect on
comes and also which are particularly prom- diversity, but with a lower effect size than for biodiversity (Fig. 2D), potentially driven by di-
ising for achieving paired benefits. Overall, food security (0.16 ± 0.04) (Fig. 2A). Soil con- vergent responses of taxa. However, positive
applying a high number of diversification strat- servation practices also promoted synergies effects on food security can readily arise from
egies or practices was associated with more through gains in all three environmental out- such practices. For example, having trees on-
potential for win-win outcomes (Fig. 3, B and comes, accompanied by enhanced human well- farm can support peoples’ diets by providing
C). Livestock diversification and soil conser- being and food security (Fig. 3C). Overall, three edible products, including fruits, nuts, and leaves
vation were the two strategies that appeared of the five assessed agricultural diversification (29–33). Five of our studies (corresponding to
to consistently elicit multiple positive outcomes strategies offered potential win-win outcomes 810 farms) used dietary diversity scores (count-
(Fig. 2, A and C), especially win-win outcomes regarding food security and nonagricultural ing the number of food groups consumed) to
proxy dietary quality, a key component of food Outlook lations, trade agreements exacerbating cor-
security and a metric likely to capture effects At a time when the outlook for simultaneous- porate concentration in global food systems,
of having trees on-farm (34). The positive ef- ly improving and protecting the environment and other supply chain pressures (40). Tran-
fect of noncrop diversification on food security and social conditions for farmers often seems sitions to diversified farming systems often
was pronounced (0.21 ± 0.08), almost twice bleak (2), our findings present a promising require financial support because of potential
the size as the negative effect of noncrop di- avenue for shaping global agricultural policy initial yield declines or implementation costs
versification on human well-being (Fig. 2D). by showing how applying a suite of diversifi- (8). Indeed, current policies often lock in sim-
Explanations for this negative effect include cation strategies or practices can create win- plified, conventional farming rather than en-
longer crop rotations or the implementation win scenarios. Our results support the notion abling durable transitions to diversified farming,
of practices such as hedgerows, which could lead that a diversified farming system is often more and investments are needed to develop appro-
to a smaller area planted with cash crops, po- beneficial than specific diversified farming priate seeds, crop mixes and rotations, and equip-
tentially leading to lower production and/or strategies or practices in isolation (13, 14). This ment to promote the profitability of diversified
higher labor demands. Although we did not finding emphasizes the need for more explicit farms.
detect significant effects of single diversifica- evidence about which combinations of diver- Effective policies for encouraging the adop-
tion strategies on ecosystem services, several sification strategies and practices are most com- tion of diversification strategies and practices
large meta-analyses have shown strong pos- plementary in different social and ecological likely vary with cropping system and region
itive effects of diversification practices on many contexts. Most of our present findings, which and include incentives, regulations, and com-
ecosystem services (12, 13). are based on working farm data, support the bined approaches. The use of incentives can be
growing body of literature linking agricultural seen in the European Union, where farmers are
Role of landscape composition in diversification strategies with better outcomes financially compensated on a per-area basis
diversification outcomes for nonagricultural biodiversity and regulat- (41, 42) for some diversification practices such
The “intermediate landscape complexity hy- ing ecosystem services without compromising as noncrop plantings. Also, a recent synthesis
pothesis” states that agricultural diversifica- yields (12, 13). from Ghana shows that incentivizing noncrop
tion is unlikely to result in improvements in Our study advances existing knowledge about diversification has cascading positive effects
cleared landscapes because the regional spe- how diversification affects agricultural system on adoption patterns (43). Regulatory mech-
cies pool available to colonize crop fields and sustainability by (i) considering how a multi- anisms can be used for soil or water conserva-
provide ecosystem services is limited (25). Sim- tude of diversification strategies, not just crop tion through policies requiring farmers to use
ilarly, in complex landscapes, diversification may diversification, may affect sustainability out- diversification practices to reduce pollutants
not lead to measurable increases in nonagri- comes (including both individual and com- on their farms (e.g., for water quality) (44).
cultural biodiversity and/or ecosystem services bined effects of diversification strategies); (ii) Finally, the benefits of combining incentives
on farms because sufficient alternative resources examining how diversification influences mul- with regulatory mechanisms can be seen in
and habitats are already present in the farm tiple social and environmental outcomes while California, where increasing adoption of di-
surroundings to support these species and highlighting trade-offs and/or synergies within versification practices on larger farms may
services. Instead, the predicted environmental and between environmental and social out- require supplementing the “pull” of incentives
benefits from agricultural diversification strat- comes; and (iii) examining how landscape com- with the “push” of regulatory mandates (44).
egies are largest in simple landscapes con- position moderates the effects of diversification Our study suggests several contexts in which
taining 5 to 20% of seminatural habitats or on environmental and social outcomes. Thus, desirable local outcomes occur most frequent-
noncrop areas (25). We tested this hypothesis we have moved beyond existing studies that ly, with a key example being the positive effect
and found that landscape composition (the typically assess the effects of diversification on human well-being and food security from
proportion of seminatural habitats in a land- strategies in isolation and on selected output applying a high number of diversification strat-
scape) moderated the outcomes of agricultural variables (35–37), preventing the systemic un- egies in cleared landscapes and on small farms
diversification strategies (see Fig. 4 and table derstanding needed for informing policy de- (table S12). However, researchers have much
S10 for sample sizes). We observed that the bates on how to produce food while maintaining more to discover about the variability of out-
number of diversification strategies applied a safe operating space for humanity. By focus- comes that can occur across different agricul-
had strong positive effects on food security ing on agricultural working landscapes, our tural diversification strategies, landscapes, and
(Fig. 4E) in cleared landscapes, indicating ben- work complements earlier studies examining social contexts, and future work should use
efits even in landscapes lacking natural hab- environmental and social trade-offs and/or syn- quasiexperimental methods that control for
itat. However, we also observed positive effects ergies of protected areas (38) and considers possible underlying differences between farm-
in complex landscapes. For example, livestock working land conservation approaches affect- ers that choose more versus fewer diversifica-
diversification showed the strongest positive ing a broader area (39). Because we include tion strategies.
effects on human well-being in cleared land- diverse datasets representing multiple world The future of agriculture faces great chal-
scapes (Fig. 4D). However, although we found regions, our flexible approach can be replicated lenges: large increases in demand for agricul-
positive social outcomes in cleared landscapes, and expanded to incorporate additional data- tural commodities must be met while at the
diversification practices there did not result sets in the future. same time minimizing agriculture’s negative
in positive environmental outcomes, which is How can and should policy-makers and prac- environmental, health, and social impacts (2).
partially consistent with the hypothesis. Finally, titioners encourage the adoption of specific Our interdisciplinary analysis spanning a wide
we observed that the number of diversification types of diversified farming systems? Although array of regions provides convincing evidence
strategies and practices applied had positive we recognize the benefits of diversification, that agricultural diversification is a promising
effects on biodiversity in both simple and com- it is also critical to acknowledge that many win-win strategy for providing social and en-
plex landscapes (Fig. 4A). These results par- farmers are working “against the odds” (8). vironmental benefits.
tially agree with the intermediate landscape Structural factors are often the main barrier
REFERENCES AND NOTES
complexity hypothesis, but indicate that diver- for adopting diversification practices, and in-
1. J. Rockström, O. Edenhofer, J. Gaertner, F. DeClerck, Nat. Food
sification strategies on farms can be beneficial clude high land rents, the predominance of 1, 3–5 (2020).
for biodiversity even in complex landscapes. short-term leases, stringent food safety regu- 2. B. Carducci et al., Nat. Food 2, 68–70 (2021).
3. Z. Mehrabi, Nat. Sustain. 6, 949 (2023). by the European Union’s Horizon 2020 research and innovation BB/J014427/1) and a Royal Geographic Society postgraduate
4. P. L. Pingali, Proc. Natl. Acad. Sci. U.S.A. 109, 12302–12308 program (Marie Skłodowska-Curie grant 796451 FFSize). J.B., A.J., fieldwork award. Author contributions: L.V.R., I.G., Z.M., and C.K.
(2012). L.H., and R.L. were funded by the Daniel and Nina Carasso designed the study. L.V.R., I.G., Z.M., O.M.S., J.B., L.A.G., M.E.I.,
5. N. Ramankutty et al., Annu. Rev. Plant Biol. 69, 789–815 (2018). Foundation. P.B. was funded by the Hungarian National Research, C.M.K., R.B.K., H.W., C.L., and C.K. developed the code book. I.G.,
6. S. K. Jones, A. C. Sánchez, S. D. Juventia, N. Estrada-Carmona, Development and Innovation Office (NKFIH KKP 133839). M.E.I. Z.M., O.M.S., J.B., L.A.G., M.E.I., C.M.K., R.B.K., H.W., P.B., R.C.,
Sci. Data 8, 212 (2021). was funded by the Canada Research Chairs program. S.R. and J.C., D.C., K.D., K.D.M., K.G., D.G., P.H., J.H., L.H., D.J., I.J., A.J.,
7. C. Kremen, A. Iles, C. Bacon, Ecol. Soc. 17, art44 (2012). I.S.D. were funded by the European Union (FP7 311781). C.K. and D.K., M.K., Y.K., C.B.K., S.K., H.K., R.L., A.L., R.N.M., P.B.M., A.M.,
8. J. Blesh et al., One Earth 6, 479–491 (2023). D.K. were funded by the US Department of Agriculture (USDA S.M., H.N.F., E.M.O., J.P.O., M.Q., S.R., A.S., S.S., W.E.S., I.S.D.,
9. J. Rosa-Schleich, J. Loos, O. Mußhoff, T. Tscharntke, Ecol. Econ. NIFA project 2015- 67019- 23147/1005662) and the CS Fund. A.E.S., J.M.T., V.V., C.V., and C.K. contributed to data collection
160, 251–263 (2019). R.B.K., C.B.K., C.V., and I.S.D. were funded through the 2017-2018 and/or data entry. L.V.R. and I.G. conducted the data cleaning and
10. C. Kremen, A. Miles, Ecol. Soc. 17, art40 (2012). Belmont Forum and BiodivERsA joint call for research proposals, data analyses with assistance from Z.M., C.S., and C.K. L.V.R.
11. C. Kremen, Emerg. Top. Life Sci. 4, 229–240 (2020). under the BiodivScen ERA-NetCOFUND program, and by the and I.G. wrote the first manuscript draft with contributions by all
12. D. Beillouin, T. Ben-Ari, E. Malézieux, V. Seufert, D. Makowski, Natural Sciences and Engineering Research Council of Canada authors. O.M.S. designed Fig. 1. Competing interests: The authors
Glob. Chang. Biol. 27, 4697–4710 (2021). (NSERC grant 523660-2018), National Science Foundation (NSF declare no competing interests. Data and materials availability:
13. G. Tamburini et al., Sci. Adv. 6, eaba1715 (2020). grant 1852587), German Federal Ministry of Education and The dataset that we compiled based on 24 case studies and
14. A. C. Sánchez, H. N. Kamau, F. Grazioli, S. K. Jones, Ecol. Econ. Research (BMBF grant 01LC11804A), and the Research Council of comprising 2655 farms is available in a Dryad repository (46). The
201, 107595 (2022). Norway (grant 295442). C.M.K. and O.M.S. were funded by R code used to generate the results is available at Zenodo and
USDA-NIFA-OREI grant 2015-51300-24155 and C.M.K. through the can also be accessed through Dryad (46). License information:
15. J. Hickel, S. Hallegatte, Dev. Policy Rev. 40, e12584 (2022).
USDA-NIFA-AFRI grant 12679452- 2019-67012-29720. D.J.L., A.M., Copyright © 2024 the authors, some rights reserved; exclusive
16. J. D. van der Ploeg et al., J. Rural Stud. 71, 46–61 (2019).
and H.P. were funded by The World Bank project Colombia licensee American Association for the Advancement of Science. No
17. A. E. Stratton, H. Wittman, J. Blesh, Agron. Sustain. Dev. 41, 35
Mainstreaming Sustainable Cattle Ranching (CMSCR-P104687), claim to original US government works. https://www.science.org/
(2021).
FEDEGAN, CIPAV, TNC, and Action Fund, with financial support about/science-licenses-journal-article-reuse
18. B. Maas et al., Trends Ecol. Evol. 35, 1049–1052 (2020).
from the Department of Business, Energy, and Industrial Strategy
19. Z. Mehrabi et al., One Earth 5, 756–766 (2022).
of the United Kingdom (BEIS) and the Global Environment Facility SUPPLEMENTARY MATERIALS
20. S. Gong et al., Ecol. Lett. 25, 1699–1710 (2022).
(GEF). Data collection in Indonesia was funded by DFG project
21. X. He et al., Nat. Food 4, 788–796 (2023). science.org/doi/10.1126/science.adj1914
192626868 in the framework of the collaborative German–
22. A. A. Rakotomalala, A. M. Ficiciyan, T. Tscharntke, Agric. Materials and Methods
Indonesian research center CRC990. C.S. was funded by the
Ecosyst. Environ. 356, 108617 (2023). Supplementary Text
European Union’s Horizon 2020 research and innovation program
23. S. F. Bender, C. Wagg, M. G. A. van der Heijden, Trends Ecol. Figs. S1 to S10
under agreement 727284 and was partially funded by the DFG
Evol. 31, 440–452 (2016). Tables S1 to S13
under Germany’s Excellence Strategy (EXC 2070–390732324).
24. K. Moser, C. Ahn, G. Noe, Wetlands 27, 1081–1097 (2007). References (47–91)
Z.M. and M.H. and the ECOLIMITS project were funded through the
25. T. Tscharntke et al., Biol. Rev. Camb. Philos. Soc. 87, 661–685 MDAR Reproducibility Checklist
UK NERC-DFID-ESRC Ecosystem Services for Poverty Alleviation
(2012).
(ESPA) program (grant NE/K010379-1) and NERC (grants Submitted 15 June 2023; resubmitted 1 September 2023
26. R. Marja et al., Ecol. Lett. 22, 1493–1500 (2019).
NE/P001092/1 and NE/P00394X/1). Z.M. was also funded by the Accepted 28 February 2024
27. F. Tacconi, K. Waha, J. J. Ojeda, P. Leith, Agron. Sustain. Dev.
Biotechnology and Biological Sciences Research Council (grant 10.1126/science.adj1914
42, 2 (2022).
28. L. V. Rasmussen et al., Nat. Sustain. 1, 275–282 (2018).
29. K. T. Sibhatu, M. Qaim, Food Policy 77, 1–18 (2018).
30. M. M. Kansanga et al., Soc. Sci. Med. 288, 113550 (2021).
31. R. Bezner Kerr et al., Glob. Food Secur. 29, 100540 (2021).
32. S. S. Snapp, M. J. Blackie, R. A. Gilbert, R. Bezner-Kerr, ANTIMICROBIALS
G. Y. Kanyama-Phiri, Proc. Natl. Acad. Sci. U.S.A. 107,
20840–20845 (2010).
33. M. G. Khonje, J. Ricker-Gilbert, M. Muyanga, M. Qaim, Lancet
Antibacterial activity of nonantibiotics is orthogonal
Planet. Health 6, e391–e399 (2022).
34. E. C. Vansant et al., People Nat. 4, 296–311 (2022).
35. S. Madsen, R. Bezner Kerr, L. Shumba, L. Dakishoni, Agroecol.
to standard antibiotics
Sustain. Food Syst. 45, 197–224 (2021).
36. A. C. Sánchez, S. K. Jones, A. Purvis, N. Estrada-Carmona, Mariana Noto Guillen, Carmen Li, Brittany Rosener, Amir Mitchell*
A. De Palma, Agric. Ecosyst. Environ. 332, 107933 (2022).
37. L. C. Ponisio et al., Proc. Biol. Sci. 282, 20141396 (2015).
Numerous nonantibiotic drugs have potent antibacterial activity and can adversely affect the
38. P. J. Ferraro, M. M. Hanauer, K. R. Sims, Proc. Natl. Acad.
Sci. U.S.A. 108, 13913–13918 (2011). human microbiome. The mechanistic underpinning of this toxicity remains largely unknown. We
39. L. A. Garibaldi et al., Conserv. Lett. 14, e12773 (2021). investigated the antibacterial activity of 200 drugs using genetic screens with thousands of barcoded
40. L. Carlisle et al., Agroecol. Sustain. Food Syst. 46, 1145 (2022).
Escherichia coli knockouts. We analyzed 2 million gene-drug interactions underlying drug-specific
41. P. Batáry, L. V. Dicks, D. Kleijn, W. J. Sutherland, Conserv. Biol.
29, 1006–1016 (2015). toxicity. Network-based analysis of drug-drug similarities revealed that antibiotics clustered into
42. F. A. Boetzl et al., Proc. Natl. Acad. Sci. U.S.A. 118, e2016038118 modules that are consistent with the mode of action of their established classes, whereas nonantibiotics
(2021). remained unconnected. Half of the nonantibiotics clustered into separate modules, potentially revealing
43. M. Isaac, H. Nyantakyi-Frimpong, P. Matouš, E. Dawoe,
L. Anglaaere, Ecol. Soc. 26, art12 (2021). shared and unexploited targets for new antimicrobials. Analysis of efflux systems revealed that they
44. K. E. Esquivel et al., Front. Sustain. Food Syst. 5, 734088 (2021). widely affect antibiotics and nonantibiotics alike, suggesting that the impact of nonantibiotics on
45. V. Ricciardi, N. Ramankutty, Z. Mehrabi, L. Jarvis, antibiotic cross-resistance should be investigated closely in vivo.
B. Chookolingo, Glob. Food Secur. 17, 64–72 (2018).
46. L. V. Rasmussen et al., Data for: Joint environmental and
social benefits from diversified agriculture, Dryad (2023);
O
https://doi.org/10.5061/dryad.1zcrjdfxw. ur microbiomes play a critical role in main- also investigated the reciprocal effect: how non-
ACKN OW LEDG MEN TS
taining our health and can affect the ef- antibiotic drugs might disrupt the microbiome
We thank the anonymous reviewers for comments on a previous
ficacy of various therapeutics. Studies and potentially culminate in dysbiosis, a detri-
version of the manuscript and our stakeholder advisory committee have revealed that the effectiveness of many mental disruption of the host’s microbiome.
chaired by S. Murphy (IISD) and consisting of 19 people drugs not traditionally prescribed for treat- Recent works revealed that multiple nonanti-
representing the following agencies and organizations: MOSES, the
GEF, IPES FOOD, NFU, NFFC, EFAO, FAO, CIFOR, McKnight
ing bacterial infections, called nonantibiotics biotics are associated with shifts in microbiome
Foundation, Bioversity, NSAC, USDA, CATIE, TNC, and CIAT. hereafter, can be affected by drug-microbiome species composition (5, 6). Examples include
Funding: This work was supported by the National Socio- interactions, including bacterial-driven drug antidiabetics (7), proton-pump inhibitors (8, 9),
Environmental Synthesis Center (SESYNC) under funding received
from the National Science Foundation (grant DBI-1639145 to
metabolism (1–3) and bioaccumulation (4). antipsychotics (10), and nonsteroidal anti-
Z.M. and C.K.) and by the UBC Research Excellence Cluster for Complementary studies of nonantibiotics have inflammatory drugs (11). Although the mecha-
Diversified Agroecosystems. L.V.R. was funded by the European nisms underlying these associations are complex,
Research Council under the European Union’s Horizon 2020
direct drug-bacterial interactions are thought to
Research and Innovation Programme (grant 853222 FORESTDIET). Department of Systems Biology, University of Massachusetts
I.G. was partially funded by the German Research Foundation Medical School, Worcester, MA, USA. be one key force mediating these shifts (2). A
(DFG project number 532858005 GR 4844/4-1). C.L. was funded *Corresponding author. Email: amir.mitchell@umassmed.edu recent in vitro screen revealed that roughly a
Results
Drug screen reveals nonantibiotics
with antibacterial activity
To identify drugs with antibacterial activity, we
tested the effects of 1280 drugs (226 antibiotics
and 1054 nonantibiotics) on E. coli growth in
defined media (Fig. 1A). This panel includes
approved and investigational drugs, as well as
26 nutraceutical compounds (table S1A). We
monitored culture density over multiple hours
and used a permissive cutoff for identifying
growth inhibition—1 SD below the median op-
tical density of control cultures. We identified
176 nonantibiotics and 142 antibiotics that were
inhibitory at a 10-mM drug concentration. Ab-
sence of antibacterial activity of some antibiotics
may arise from their restricted phylogenetic
spectrum as reflected by the annotated affected
organism (table S1A). We validated the anti-
Fig. 1. Screening drugs for antibacterial activity. (A) Overview of drug screens. Overnight culture was bacterial activity of the selected drugs in a sim-
diluted into minimal media aliquoted with a panel of 1280 drugs at a single concentration, and toxic drugs were ilar follow-up screen with four replicates and a
identified by reduced growth. (B) Results from antibacterial activity screens with nonantibiotic and antibiotic strict statistical cutoff. We validated that 104
drugs. Putative antimicrobials identified by the first screen were validated with a secondary screen, and top nonantibiotics significantly inhibited growth
validated antimicrobials were used for the genetic screens. (C) Overview of pooled genetic screens. An overnight [one-tailed t test, false discovery rate (FDR)–
pooled collection of 6709 barcoded stains was diluted into minimal media aliquoted with 186 drugs and grown adjusted P value of 0.25]. Figure 1B and tables S1,
until control cultures reached late log phase. DNA was extracted, and barcodes were amplified by polymerase B to D, show a summary of the number of hits
chain reaction (PCR) before sequencing. Screens were performed in triplicates. (D) Pairwise chemical similarity identified in the screens. We focused on the 103
across all toxic drugs. The heatmap shows the Tanimoto coefficient with Morgan2 fingerprints of all drugs most toxic antibiotics and the 83 most toxic
screened, sorted by drug class, and clustered within the class. (E) Histogram of pairwise chemical similarity nonantibiotics for further analysis (supplemen-
across drug groups. Scores below 0.2 were considered chemically dissimilar. tary materials, materials and methods).
We next tested if the antibacterial activity of
nonantibiotics can be attributed to chemical
similarity to standard antibiotics. We measured
quarter of nonantibiotics are potent antibacterials ical implications. Previously unrecognized tox- the chemical similarity of each drug pair by cal-
at physiologically relevant concentrations (12). icity mechanisms may identify unexploited culating the Tanimoto coefficient with Morgan2
A key question that arises from the observa- targets for future antibiotic development, and fingerprints (17, 18). This measurement denotes
tion of widespread antibacterial activity of non- shared toxicity mechanisms can identify drugs the proportion of shared chemical features be-
antibiotics is the underlying mode of action in that could inadvertently select for multidrug tween two molecules (values >0.8 are typically
bacteria. Work on even very well-characterized resistance under chronic administration and considered highly similar). Figure 1D and table
drugs, such as the chemotherapy agent fluoro- may undermine the efficacy of antibiotics. S2 show the coefficients of the top 186 most
uracil (13–16), demonstrated that they can have Motivated by reports of nonantibiotics’ wide- toxic drugs ordered by their classes. Overall,
both shared and distinctive toxicity mecha- spread antibacterial activity, we used an Esche- we observed high chemical similarity between
nisms across phylogenetic kingdoms. Uncov- richia coli lab strain as a platform to uncover antibiotics that target the same cellular process
ering the mode of action across a wide panel the pathways underlying bacterial growth in- (Fig. 1D, dark triangles). We observed low chem-
of nonantibiotic drugs has the potential to hibition. We used a pooled genetic screening ap- ical similarity for drug pairs belonging to dif-
identify both common and distinct toxicity proach to investigate a panel of almost 200 drugs. ferent antibiotic classes or pairs with at least one
mechanisms with standard antibiotic drug Our screens yielded more than 2 million bacte- nonantibiotic (Fig. 1D, light colors). The histo-
classes, both of which have important biomed- rial fitness measurements and permitted analy- grams of chemical similarity coefficients across
Fig. 2. Genetic screens capture antibiotic mechanisms of action and reveal knockout fitness were used as the measurement of resemblance, and the
interclass interactions. (A) Hierarchical clustering of antibiotics by the fitness network was generated with a force-layout algorithm according to the correlation
score of informative knockouts. The colored squares at the bottom row indicate scores. Node and frame colors mark the antibiotic class. Most antibiotics self-
the class of each antibiotic and show that the algorithm successfully grouped grouped into the same class of network modules. The side panels show
most antibiotics into class-specific clusters. The right panels highlight signature functional enrichment analysis by GO biological process terms of each module.
knockouts characteristic to specific clusters. (B) Network representation of Circle size and shading mark the size of gene set and statistical significance,
antibiotic drug resemblance by informative knockouts. Drug-drug correlations in respectively. ncRNA, noncoding RNA; met., metabolic; biosyn., biosynthetic.
different groups revealed highly dissimilar dis- Pooled genetic screen for identifying modulators tify the genes influencing drug-specific sensitiv-
tributions (Fig. 1E). We used these distributions of toxicity ity, we used a pooled collection of 6709 barcoded
to define a lower (permissive) cutoff for chem- We next employed a pooled genetic screening knockout E. coli strains covering 3500 nones-
ical similarity. Hence, antibacterial activity of method that we previously developed (15, 19, 20) sential genes. We determined the concentration
nonantibiotics is not necessarily a result of chem- to identify the cellular pathways modulating needed to achieve 50% growth inhibition (IC50)
ical similarity to standard antibiotics. the antibacterial activity of 186 drugs. To iden- of wild-type E. coli for each drug. Drugs with
low toxicity (IC50 > 50 mM) were used at a con- each knockout. We identified 191,552 statistically Known toxicity mechanisms are captured for
centration of 50 mM. We removed eight drugs significant depleted or enriched knockouts standard antibiotics
from the panel for technical reasons (table relative to the no-drug control [FDR-adjusted We validated our method by verifying that it
S1B). In three replicates, we inoculated the P value < 0.25, Wald test in the DESeq2 prog- correctly identifies the known modes of action
strain collection into minimal media with or ram (21)]. Depleted knockouts reflect gene loss of 90 well-characterized antibiotics. Specifical-
without the drugs and terminated the screen of function that increases drug sensitivity, ly, we chose a subset of knockouts (581 strains)
when controls approached stationary phase whereas enriched knockouts reflect loss of that we expected would be informative for
(Fig. 1C). We sequenced the barcodes from ex- function that increases resistance (all results identifying shared modes of action by filtering
tracted DNA and calculated the frequency of are in table S3). out strains that rarely modulated toxicity (hits
in less than two antibiotics) and strains that Antibacterial activity of nonantibiotics differs bolic processes. This module included fluoro-
modulated toxicity very frequently (hits in from that of standard antibiotics pyrimidine chemotherapeutic drugs, which we
more than 30 antibiotics). We also discarded We sought to identify the toxicity mechanisms previously characterized with similar findings (15).
knockouts of specific biological functions, such of nonantibiotics by adding them to the drug- The second-largest module (10 drugs) was en-
as efflux pumps, that can hinder the analysis of relatedness network that we generated for anti- riched in processes related to lipopolysaccharide
the mode of action (table S4). Figure 2A shows biotics. This analysis adds more nodes (drugs) biosynthesis. The third module was enriched in
the results of hierarchical clustering of the infor- and edges (relatedness interactions) to the net- processes related to DNA damage repair and
mative strains. We observed that clustering was work without altering the topology among anti- the stress response, and the two smallest mod-
highly successful in recapturing antibiotic classes biotics. In this analysis, nonantibiotics that ules included processes related to oxidative
(shown by the colored dendrogram at the top target similar pathways to those of standard stress and siderophore-biosynthesis pathways.
of the heatmap): Seventy-five antibiotics were antibiotic classes would cluster with antibiotic Taken together, results from analyses of non-
grouped into “pure” clusters with a single anti- network modules (Fig. 3A). However, we ob- antibiotic drugs indicated that their antibac-
biotic class. All four cell-membrane targeting served that most nonantibiotics remained com- terial mechanisms are largely distinct from those
antibiotics were in a single mixed cluster. pletely unconnected to the antibiotic modules of standard antibiotics. Moreover, our network
Focusing on top hits of individual clusters (57/77). The notable exceptions were 14 nonanti- analysis showed that nonantibiotics are not
uncovered many genes from the known target biotics that were connected to multiple anti- completely distinct from one another but that
pathways (Fig. 2A, right side). For example, biotic classes (positioned between the colored most of them can be grouped into modules
knockouts disrupting the SOS response and modules). When focusing on 52 chemically sim- reflecting shared toxicity mechanisms. This
DNA repair (e.g., recA and recN) increased sen- ilar antibiotic and nonantibiotic pairs (Tanimoto conclusion is further supported by the func-
sitivity to DNA-targeting antibiotics (red and coefficient >0.2), we did not detect high sim- tional enrichment analysis, which uncovered
pink clusters), whereas knockouts disrupting ilarity in the genetic screen results (fig. S2A). common pathways modulating antibacterial
nucleotide excision repair, such as uvrABC, To substantiate this conclusion, we used an activity. In addition, the modular network struc-
only increased sensitivity to nitrofurans (pink alternative and widely used approach for clas- ture indicates that unexploited yet robust cel-
cluster) that induce DNA adducts requiring dis- sifying the mode of action of nonantibiotics. Spe- lular targets for inhibiting bacterial growth may
tinct repair mechanisms (22). For folic acid– cifically, we trained a random forest classifier exist and provides multiple lead molecules for
targeting antibiotics, deletions in folX and folM that relied on all knockout strains to predict further investigation.
increased resistance, whereas deletions in the whether drugs can be classified as belonging
glycine cleavage system (gcv genes) increased to specific antibiotic classes (this analysis cir- Transport systems affect both antibiotics
sensitivity (23). Knockouts disrupting peptido- cumvented our choice of informative knock- and nonantibiotics
glycan recycling increased sensitivity to cell- outs). In agreement with our network analysis, Our network analyses purposefully disregarded
wall–targeting antibiotics. we observed that although our classifier suc- genes encoding efflux pumps, because they
To explore the relationship between anti- cessfully assigned most antibiotics to their are expected to affect the antibacterial activity
biotic classes, we calculated the Pearson corre- correct class, it failed to classify most nonanti- irrespective of the compound’s target. We sub-
lation between each pair of antibiotics (Fig. 2A, biotics (fig. S2, B and C). Overall, the results in- sequently examined the impact on antibacterial
columns) and generated a graph representing dicate that nonantibiotic drugs likely operate activity of genes involved in membrane perme-
drug relatedness. Figure 2B and fig. S1A show through toxicity mechanisms that are orthog- ability and transport (76 genes from 33 pathways
the drug similarity network (network layout onal to standard classes of antibiotics. of drug transport and membrane permeabil-
was set with a force-directed algorithm by the The network analysis revealed that although ity, table S4). Figure 3D shows interactions
strength of drug-drug correlations). Similarly most nonantibiotics were unconnected to anti- identified across all the drugs tested, as well as
to hierarchical clustering, this analysis grouped biotic modules, some nonantibiotics are re- the pathways that increase bacterial sensitivity
most antibiotics by their mechanism of action. lated to one another (Fig. 3A). This suggested upon deletion. The widest impact on toxicity was
However, the graph-based analysis also high- that relatedness may exist but that it may be observed for mutations affecting the resistance-
lighted interclass relatedness that cannot be occluded by the antibiotic-centered analysis. We nodulation-division (RND) family transporters,
explained by chemical similarity (Fig. 2B, blue therefore repeated the network analysis, relying the Tol-Pal system, and outer membrane pro-
edges). Interclass relatedness inferred from the exclusively on knockouts that were informative teins (OMP). We observed that drugs varied
screens reveals shared pathways modulating for nonantibiotics (using similar filters as before). considerably in the number of interactions that
the drug toxicity that extend beyond the anti- Figure 3B and fig. S1B show the resulting rel- they have with transport systems, with 26 drugs
biotic target. DNA-targeting antibiotics emerged atedness network that is based on drug-drug affected by five or more different transport sys-
as a network hub. Relatedness between classes correlations across 509 informative knockouts. tems (specificity of pumps to antibiotic classes
was also supported by a separate approach— In this analysis, 52 of 77 nonantibiotics had at is shown in fig. S3). Overall, we observed that
functional enrichment (Fig. 2B, margins, and least one neighbor. Many connections were mutation in at least one transport system in-
table S5). This analysis revealed that cellular identified between drugs that are used for creased the antibacterial activity of 93% of anti-
pathways known to be related to one class are treating different human conditions and dis- biotics and 82% of nonantibiotics (Fig. 3D).
also enriched in seemingly unrelated classes, eases (Fig. 3B, different node shapes), and most This observation implies that evolved resistance
such as DNA-related categories enriched in connections could not be explained by chem- against nonantibiotic drugs can potentially,
ribosome-targeting antibiotics. ical similarity (Fig. 3B, blue edges). in some cases, inadvertently lead to multidrug
Taken together, our results verified our ap- We identified nine independent network resistance and undermine the efficacy of anti-
proach for capturing the established mech- modules and found that five modules were biotics. Increased risk of unintentional cross-
anisms of action of standard antibiotics. The enriched for specific biological functions (table resistance may exist for nonantibiotic drugs that
analysis revealed the value of this approach S4). Figure 3C shows the top enriched Gene are affected by multiple transport systems.
in uncovering higher-order interactions between Ontology (GO) biological processes summar-
antibiotic classes that have different cellular ized by REVIGO (reduce and visualize Gene Evolved resistance against nonantibiotic drugs
targets yet affect common pathways that mod- Ontology) (24, 25). The largest module (14 drugs) Our genetic screens, coupled with network anal-
ulate toxicity. was enriched for diverse biosynthetic and meta- ysis, have led to two important conclusions. First,
Fig. 4. Evolved resistance against nonantibiotics exposes unknown drug clones, and dots mark mutations. Dark-pink dots mark gene hits across at
targets and collateral cross-resistance. (A) Overview of evolution experiment least four of the five replicates and likely adaptive mutations. Light-pink
and downstream assays. Independent cultures were serially transferred in dots mark mutations in genes conferring drug resistance according to the
media supplemented with three nonantibiotics over 200 generations. A single genetic screen. The diagram on top represents the inferred mechanism
resistant clone was isolated from each independently evolved population. (B) Drug of resistance. (D) Sertraline-evolved strains are resistant to multiple antibiotics,
resistance evolved across all cultures and drugs. The panels show the drug- whereas triclabendazole- and streptozotocin-evolved strains do not confer
sensitivity curves of ancestor and evolved clones; horizonal lines mark the cross-resistance. The panels show the increased resistance (IC50 fold-change) of
concentration that inhibits growth by 50% (IC50). (C) Results of whole-genome evolved clones against six antibiotics. The bars represent the median across
sequencing uncover three alternative strategies of evolved resistance. The all evolved clones, and dots represent the mean of three biological replicates of
concentric Circa plots represent the bacterial chromosome of individually evolved each clone. The bar color shows the antibiotic class.
the orthogonality in toxicity mechanisms indi- Given the bacterial adaptation strategies mechanisms similar to those of standard anti-
cates that nonantibiotics may potentially un- that we observed, we predicted that sertraline- biotics (Fig. 3A). A subsequent analysis revealed
cover unexploited targets for inhibiting bacterial adapted clones, but not other clones, will emerge that most of the nonantibiotics clustered into
growth. Second, analysis of drug-transport sys- as multidrug resistors. Experiments measur- highly intraconnected modules enriched in
tems indicates that despite this orthogonality, ing the resistance against a panel of antibiotics specific cellular pathways (Fig. 3, B and C). Re-
broad resistance to antibiotics can emerge if confirmed this prediction (Fig. 4D). In addi- sults from almost 200 drugs support the hy-
drug transport–based resistance evolves as a tion, given the hypothesis that triclabendazole pothesis that antibiotics and nonantibiotics
result of selection by nonantibiotics. We tested binds to IF2, we predicted that triclabendazole- target bacteria through orthogonal mecha-
these predictions experimentally by selecting for adapted clones will also be resistant to neigh- nisms of action. Network analysis focusing
drug resistance in evolution experiments (26) boring drugs in the relatedness network (fig. exclusively on nonantibiotics indicated that
(Fig. 4A). We focused on three nonantibiotics se- S4A). Drug-sensitivity measurements confirmed many of them share common toxicity mech-
lected from different modules in the nonanti- this prediction by showing that triclabendazole- anisms (Fig. 3B). Moreover, analysis of genes
biotic network: streptozotocin, triclabendazole, evolved clones were also resistant to lamotrigine involved in drug transport indicated that de-
and sertraline (Fig. 3B). Information about the and pyrimethamine (fig. S4B). Furthermore, spite orthogonality in mode of action, bacterial
toxicity of these drugs across representative compound binding SEC-ASMS assays (size ex- sensitivity to most drugs is increased by inac-
microbiome species and five E. coli strains is clusion chromatography followed by affinity tivation of transport systems and membrane-
provided in table S6. selection-mass spectrometry) confirmed that permeability proteins (Fig. 3D). The extensive
Figure 4B shows the drug sensitivity curves all three drugs bind the IF2 protein (fig. S4C). impact of efflux systems on nonantibiotics is
of the ancestor strain and of five indepen- The results from our evolution experiments consistent with their known role in xenobiotic
dently evolved clones. Streptozotocin- and uncovered instances of three representative detoxification (32).
triclabendazole-adapted clones were fully re- strategies for adaptive resistance: decreased Our screen results showing orthogonality in
sistant to the drugs, whereas sertraline-adapted import, mutation of target, and increased ex- toxicity suggest that nonantibiotics target bac-
clones increased resistance by more than two- port. However, beyond the specific mechanisms terial cellular components not currently ex-
fold. We sequenced the genomes of evolved of action, the evolution experiment substanti- ploited by standard antibiotics. The modular
strains to uncover the mechanisms underly- ated the two main premises arising from our network topology (Fig. 3B) further implies
ing resistance. Figure 4C shows the mutations work: (i) that orthogonality in modes of action that some targets may be common to multiple
that we identified with the breseq tool (27). We can be leveraged to uncover proteins not cur- nonantibiotics. This prediction is supported
used genes mutated across independent clones rently used as targets by standard antibiotics by our observations of triclabendazole-evolved
as evidence for likely driver mutations (Fig. 4C, and (ii) that despite orthogonality, adaptation resistance by in-frame mutations in IF2, a pro-
dark-pink dots). Further support was provided against nonantibiotics can emerge through tein not targeted by standard antibiotics. Further-
by overlap of mutations with genetic-screening changes in drug efflux and therefore inadver- more, confirming our network topology, we
hits (Fig. 4C, light-pink dots). tently culminate in antibiotic cross-resistance. found that the two neighboring drugs in our net-
We found three representative adaptation work also bind IF2 (fig. S4). Another prediction,
strategies for resistance. We identified numer- Discussion stemming from our finding of widespread
ous mutations in streptozotocin-evolved clones Human health and microbiome integrity are sensitivity to inactivation of drug-transport sys-
with a clear location bias toward the origin of tightly interlinked. This fundamental realiza- tems, suggested that evolved resistance against
replication. Streptozotocin, a naturally occur- tion is the basis of current efforts to identify nonantibiotics may confer broad drug resistance;
ring alkylating agent that is used to treat pan- correlations between a myriad of environmen- this prediction was validated with sertraline-
creatic neuroendocrine tumors (28), repeatedly tal factors—such as diet and medication—and adapted strains (Fig. 4D). Although estimates
selected for inactivation of nagE, which en- microbiome species composition. Although the of drug concentrations in the human digestive
codes for glucose-6-phosphate transporter. This widespread antibacterial activity of nonanti- tract suggest that many nonantibiotics may
transporter is likely central for drug import biotics is thought to be a key force underlying apply a selective pressure on the gut micro-
and was previously observed to confer strep- these associations, the modes of action of non- biome, it remains to be determined whether
tozotocin resistance (29). Clones adapted to antibiotics in bacteria remain largely unknown. adaptions will influence efflux pumps and in-
triclabendazole, a drug used for treating worm We addressed this gap in knowledge by devel- advertently affect antibiotic resistance. Our
infections, had only a handful of mutations. oping a high-throughput genetic screening ap- results allow the gauging of nonantibiotics in-
However, all independently evolved clones proach for systematically mapping interactions teracting with multiple efflux systems (Fig. 3D).
shared an in-frame mutation in a specific re- between dozens of antibacterial drugs and single- This ranking can inform future studies fo-
gion of the essential gene infB, which encodes gene knockouts. Although a well-characterized cusing on specific interactions in the human
the translation initiation factor IF2. Previous lab strain of E. coli was an ideal platform for a microbiome.
work on lamotrigine, a drug that we identi- high-throughput investigation of mode of ac- The emergence and spread of antibiotic-
fied as a triclabendazole neighbor in the drug- tion, additional work is needed to extend our resistant pathogens is a fundamental challenge
relatedness network (Fig. 3B), found that it study’s conclusions to the human microbiome. for global health (33). Decreasing investment
binds to the same region of IF2 (30). Last, all Specifically, testing is necessary to determine in development of new antibiotics in recent dec-
sertraline-adapted clones had mutations in whether nonantibiotics operate with similar ades is only expected to compound the prob-
key regulators of efflux pumps (marR, acrR, modes of action in microbiome isolates and lem (34). Although technological advancements
and lon). Mutations targeting these regula- whether drug adaptation is observed in treated in robotic automation greatly increase the ca-
tors were very recently reported in sertraline- individuals. pacity to screen ever-growing libraries of small
adapted strains and were shown to increase The network we constructed using drug-drug molecules (35), rediscovery of antibacterial
expression of multiple efflux systems (31). Our correlations recaptured the known antibiotic compounds that belong to established antibi-
genetic screens also showed that sertraline, classes (Fig. 2B). When we added nonanti- otic classes remains a bottleneck for innova-
more than triclabendazole and streptozotocin, biotics to the antibiotic-centered clustering tion (36). This obstacle is common to both
is sensitive to inactivation of transport systems analysis, most of them remained unconnected, traditional natural product–based screening
(Fig. 3D). indicating that they do not inhibit E. coli through (37) and compounds proposed by artificial
intelligence (38). Our methodology can be used Visualization: A.M. and M.N.G. Formal analysis: A.M., M.N.G., claim to original US government works. https://www.science.org/
to address this challenge because it can com- and C.L. Funding acquisition: A.M. Project administration: A.M. about/science-licenses-journal-article-reuse
Supervision: A.M. Writing – original draft: A.M. and M.N.G. Writing –
pare the pathways underlying antibacterial editing: A.M. and M.N.G. Writing – review: A.M., M.N.G., B.R., SUPPLEMENTARY MATERIALS
activity between hundreds of uncharacter- and C.L. Competing interests: The authors declare that they have
science.org/doi/10.1126/science.adk7368
ized compounds and established antibiotics. no competing interests. Data and materials availability: Raw
Materials and Methods
sequencing reads from barcoded knockout library screen and whole
As we showed, the results of these screens genomes of evolved bacteria are available on the NCBI Sequence
Figs. S1 to S4
can be used for training classifiers to identify Tables S1 to S6
Read Archive under bioprojects PRJNA986181 and PRJNA998361.
MDAR Reproducibility Checklist
toxicity profiles that match established anti- Cytoscape networks and the computer code (R and MATLAB) used
in this article are deposited at Zenodo (39). License information: Submitted 13 September 2023; accepted 4 March 2024
biotic classes. These classifiers can therefore Copyright © 2024 the authors, some rights reserved; exclusive Published online 14 March 2024
also identify compounds operating through licensee American Association for the Advancement of Science. No 10.1126/science.adk7368
alternative, nonstandard modes of action.
RE FE RENCES AND N OT ES
1. M. Zimmermann, M. Zimmermann-Kogadeeva, R. Wegmann,
A. L. Goodman, Nature 570, 462–467 (2019). MICROBIOLOGY
2. M. Zimmermann, K. R. Patil, A. Typas, L. Maier, Mol. Syst. Biol.
17, e10116 (2021).
3. P. Spanogiannopoulos, E. N. Bess, R. N. Carmody,
Prophage terminase with tRNase activity sensitizes
P. J. Turnbaugh, Nat. Rev. Microbiol. 14, 273–287 (2016).
4. M. Klünemann et al., Nature 597, 533–538 (2021).
5. G. Falony et al., Science 352, 560–564 (2016).
Salmonella enterica to oxidative stress
6. A. Vich Vila et al., Nat. Commun. 11, 362 (2020).
7. K. Forslund et al., Nature 528, 262–266 (2015). Siva Uppalapati1†, Sashi Kant1†‡, Lin Liu1, Ju-Sim Kim1, David Orlicky2,
8. F. Imhann et al., Gut 65, 740–748 (2016). Michael McClelland3, Andres Vazquez-Torres1,4*
9. M. A. Jackson et al., Nat. Commun. 9, 2655 (2018).
10. S. A. Flowers, S. J. Evans, K. M. Ward, M. G. McInnis,
V. L. Ellingrod, Pharmacotherapy 37, 261–267 (2017). Phage viruses shape the evolution and virulence of their bacterial hosts. The Salmonella enterica genome
11. M. A. M. Rogers, D. M. Aronoff, Clin. Microbiol. Infect. 22, 178. encodes several stress-inducible prophages. The Gifsy-1 prophage terminase protein, whose canonical function
e1–178.e9 (2016). is to process phage DNA for packaging in the virus head, unexpectedly acts as a transfer ribonuclease
12. L. Maier et al., Nature 555, 623–628 (2018).
13. A. Tomasz, E. Borek, Proc. Natl. Acad. Sci. U.S.A. 46, 324–327
(tRNase) under oxidative stress, cleaving the anticodon loop of tRNALeu. The ensuing RNA fragmentation
(1960). compromises bacterial translation, intracellular survival, and recovery from oxidative stress in the vertebrate
14. A. Tomasz, E. Borek, Biochemistry 1, 543–552 (1962). host. S. enterica adapts to this transfer RNA (tRNA) fragmentation by transcribing the RNA repair Rtc system.
15. B. Rosener et al., eLife 9, e59831 (2020).
The counterintuitive translational arrest provided by tRNA cleavage may subvert prophage mobilization and
16. A. P. García-González et al., Cell 169, 431–441.e8 (2017).
17. H. L. Morgan, J. Chem. Doc. 5, 107–113 (1965). give the host an opportunity for repair as a way of maintaining bacterial genome integrity and ultimately
18. D. Rogers, M. Hahn, J. Chem. Inf. Model. 50, 742–754 survival in animals.
(2010).
19. M. Noto Guillen, B. Rosener, S. Sayin, A. Mitchell, Cell Syst. 12,
1064–1078.e7 (2021).
O
20. S. Sayin et al., eLife 12, e83140 (2023). xidative stress imposes tremendous se- with the repression of ribosomal RNA (rRNA)
21. M. I. Love, W. Huber, S. Anders, Genome Biol. 15, 550 (2014). lective pressures on all domains of life. and tRNA genes undermines the translational
22. K. R. Ona, C. T. Courcelle, J. Courcelle, J. Bacteriol. 191,
4959–4965 (2009).
Intracellular bacterial pathogens, such capacity of phylogenetically diverse micro-
23. R. J. Nichols et al., Cell 144, 143–156 (2011). as S. enterica, that reside in effector cells organisms undergoing the stringent response
24. M. Ashburner et al., Nat. Genet. 25, 25–29 (2000). of the mammalian innate immune sys- during periods of oxidative stress. In this study,
25. F. Supek, M. Bošnjak, N. Škunca, T. Šmuc, PLOS ONE 6,
e21800 (2011).
tem, are exposed to high concentrations of we describe a lysogenic prophage that represses
26. J. E. Barrick, R. E. Lenski, Nat. Rev. Genet. 14, 827–839 reactive oxygen species (ROS) synthesized translation in Salmonella exposed to oxidative
(2013). by the respiratory burst of the phagocyte stress generated by the phagocyte NADPH
27. J. E. Barrick et al., BMC Genomics 15, 1039 (2014).
NADPH (nicotinamide adenine dinucleotide oxidase through the moonlighting transfer
28. J. Capdevila et al., Neuroendocrinology 112, 1155–1167 (2022).
29. J. Lengeler, Mol. Gen. Genet. 179, 49–54 (1980). phosphate) oxidase (1). The cytotoxic concen- ribonuclease (tRNase) function of a deoxy-
30. J. M. Stokes, J. H. Davis, C. S. Mangat, J. R. Williamson, trations of ROS produced by the respiratory ribonuclease (DNase) whose canonical role is
E. D. Brown, eLife 3, e03574 (2014). burst damage nucleotides, metal cofactors, and viral-genome processing and capsid packaging.
31. Y. Wang et al., Proc. Natl. Acad. Sci. U.S.A. 120, e2208344120
(2023). sulfur-containing amino acids (2). Oxidation
32. P. J. F. Henderson et al., Chem. Rev. 121, 5417–5478 (2021). of biosynthetic enzymes induces functional Results
33. C. J. L. Murray et al., Lancet 399, 629–655 (2022). auxotrophies for aromatic and branched-chain Translational halts in S. enterica after
34. M. A. Cook, G. D. Wright, Sci. Transl. Med. 14, eabo7793 oxidative stress
(2022).
amino acids (3, 4). The resulting drops in ami-
35. L. Martínez-Fructuoso et al., ACS Infect. Dis. 9, 1245–1256 no acids are followed by a surplus of deacyl- To measure de novo translation in S. enterica
(2023). ated transfer RNAs (tRNAs) that trigger an in vitro in response to oxidative stress gen-
36. M. Miethke et al., Nat. Rev. Chem. 5, 726–749 (2021).
37. T. M. Privalsky et al., J. Am. Chem. Soc. 143, 21127–21142
adaptation in bacteria commonly known as erated by application of hydrogen peroxide
(2021). the stringent response (5–7). The combination (H2O2), we used a Western blot version of
38. J. M. Stokes et al., Cell 180, 688–702.e13 (2020). of a shrinking pool of charged tRNAs together SUnSET (surface sensing of translation) (8).
39. M. N. Guillen, A. Mitchell, Mitchell-SysBiol./2024notoetal:
v1.0.0, Zenodo (2024); https://zenodo.org/doi/10.5281/
S. enterica experienced a substantial (P <
zenodo.10525157. 1
University of Colorado School of Medicine, Department of
0.0001) inhibition of de novo protein syn-
Immunology and Microbiology, Aurora, CO, USA. 2University thesis upon exposure to 400 mM H2O2 (Fig. 1,
ACKN OW LEDG MEN TS of Colorado School of Medicine, Department of Pathology, A and B). Under the high density of bacterial
We thank N. London and A. Orlov for their advice on measuring Aurora, CO, USA. 3Department of Microbiology and Molecular cultures used in these experiments, 400 mM
chemical similarity. We thank H. Youk and C. Navarro for their Genetics, University of California Irvine School of Medicine,
comments on the manuscript. Funding: This research was Irvine, CA, USA. 4Veterans Affairs Eastern Colorado Health H2O2 did not affect bacterial viability (Fig. 1C),
supported by National Institute of General Medical Sciences grant Care System, Denver, CO, USA. indicating that S. enterica can tolerate mod-
R35GM133775 (A.M.) and by National Institute of Allergy and *Corresponding author. Email: andres.vazquez-torres@cuanschutz.edu erate levels of peroxide stress by repressing
Infectious Diseases grant R01AI170722 (A.M.). Author †These authors contributed equally to this work.
contributions: Conceptualization: A.M. and M.N.G. Methodology: ‡Present address: Bacterial Pathogenesis, Boehringer Ingelheim Animal translation. S. enterica resumed translation
A.M., M.N.G., B.R., and C.L. Investigation: M.N.G., B.R., and C.L. Health USA, Ames, IA, USA. 90 min after exposure to H2O2 (Fig. 1, D and E).
Fig. 1. Transcriptional adaptations to the repression of de novo translation 30 min of treatment. Data are shown as mean ± SD (n = 4); ****P ≤ 0.0001 and
in S. enterica undergoing oxidative stress. (A and D) Immunoblots with ***P ≤ 0.001 as determined by one-way ANOVA with Dunnet’s multiple
Ponceau S–stained controls and (B and E) a corresponding densitometric comparison test. (F) Differentially expressed genes in S. enterica after 30 min of
analysis examining puromycin incorporation by S. enterica 30 min after H2O2 400 mM H2O2 treatment as assessed by RNA-seq (30). Each circle represents the
treatment. Densitometry was normalized against Ponceau S–stained lanes average log2-fold change of four biological replicates. Genes with a log2-fold
with ImageJ. Data are shown as mean ± SD (n = 3 independent observations); change >0.8 or <−0.8 are depicted on the top and bottom, respectively. Orange
****P ≤ 0.0001 as determined by one-way analysis of variance (ANOVA) with and green boxes represent pathogenicity islands, and the yellow box represents
Dunnet’s multiple comparison test. (C) Killing of S. enterica by H2O2 in PBS after S. enterica plasmid genes.
To identify responses that may underlie sis of lipids, amino acids, and aminoacyl tRNAs to a 2′, 3′- cyclic phosphate intermediate (11),
recovery of translational activity during oxida- were down-regulated. Quantitative reverse tran- which serves as a substrate for the phosphodi-
tive stress, we investigated the genome-wide scription polymerase chain reaction (RT-PCR) esterase RtcB. The RtcB protein also catalyzes
transcriptional responses of S. enterica ex- confirmed the up-regulation of katG and sufABC the GTP-driven ligation of 3′-phosphate and 5′-
posed to H2O2. Hierarchical clustering analy- genes and the down-regulation of oxidative hydroxyl RNA fragments (12, 13). The activation
sis of RNA-sequencing (RNA-seq) data revealed phosphorylation genes (fig. S1, B and C). of rtc genes during oxidative stress prompted us
that 497 and 569 genes were up-regulated and The patterns of gene expression that we to explore the role of this operon in adaptation
down-regulated, respectively, in H2O2-treated found are consistent with the idea that of S. enterica to the inhibition of de novo pro-
S. enterica compared with untreated controls S. enterica adapts to oxidative stress by up- tein synthesis through oxidative stress.
(fig. S1A) [false discovery rate (FDR) P value < regulating antioxidant defenses, while favor-
0.05; tables S1 to S4]. This analysis showed ing overflow metabolism and undergoing the Role of the RNA repair Rtc system during
transcriptional up-regulation of katG, trxC, stringent response (2, 5, 9). We also noted that oxidative stress
and dps of the OxyR regulon, of mntH and the rsr-yrlBA-rtcBA operon, which mediates We tested the capacity of S. enterica rtcBA and
sitABCD involved in manganese transport, repair of fragmented RNA molecules, was up- rtcR deletion mutants to synthesize protein
and of genes encoding the Salmonella patho- regulated in H2O2-treated S. enterica (Fig. 1F after exposure to H2O2. As reported above,
genicity island-2 type III secretion system, gly- and fig. S1D). The rtc operon encodes for the wild-type (WT) S. enterica recovered de novo
colysis, iron acquisition, and [Fe-S] repair. By RNA cyclase RtcA, the RNA ligase RtcB, and the protein synthesis after 90 min of H2O2 treat-
contrast, genes encoding oxidative phosphor- s54 transcriptional activator RtcR (10). RtcA ment. In comparison, the DrtcBA and DrtcR
ylation; ribosomal proteins; or the biosynthe- converts 3′ and 5′-phosphate RNA fragments isogenic controls undergoing oxidative stress
showed delayed (P < 0.01) de novo protein DrtcBA S. enterica after H2O2 treatment (fig. S3D). ceiving intraperitoneal (i.p.) inoculation with
translation during the recovery phase (Fig. 2, Sequencing of 3′ RACE products of tRNALeuPQTV WT controls (Fig. 2H) and the reduced burden
A and B, and fig. S2, A to C). These data in- and tRNALeuW fragments recovered from H2O2- and number of microabscesses recorded in
dicate that the RNA repair Rtc system plays a treated S. enterica identified cleavage at posi- infected livers (Fig. 2I) and spleens (fig. S4E)
role in the adaptation of S. enterica to transla- tions U34 and G37 of the anticodon loop (Fig. when inoculated as a single culture in mice.
tional halts that occur during oxidative stress. 2E and fig. S3E). The DrtcBA mutant was also attenuated in a
Coinciding with rtcB gene transcription (fig. WT and rtc S. enterica mutants were equally streptomycin-treated oral model of S. enterica
S3A), the addition of H2O2 to WT S. enterica susceptible to the bacteriostatic activity of H2O2 infection (fig. S4F). The DrtcBA S. enterica
resulted in fragmentation of tRNA Tyr and (fig. S4A); however, compared with WT con- strain was as fit as WT S. enterica in Cybb−/−
tRNALeuPQTV (Fig. 2, C and D), so we explored trols, a higher percentage (P < 0.0001) of DrtcR mice lacking the catalytic gp91phox subunit of
how the Rtc system aids the recovery of de and DrtcBA S. enterica mutants were killed by the phagocyte NADPH oxidase (Fig. 2, H and I,
novo translation through RNA repair (10). Fur- H2O2 (Fig. 2F and fig. S4B). The DrtcBA mu- and fig. S4E). Cumulatively, these experiments
ther analysis revealed that S. enterica under- tant became resistant to H2O2 killing upon show that the RNA repair Rtc system assists
going oxidative stress specifically experienced complementation with the rtcBA operon (fig. S. enterica to adapt to ROS cytotoxicity in mac-
fragmentation of all major leucine tRNA iso- S4C). These data indicate that the RNA repair rophages and mice.
acceptors tRNALeuPQTV, tRNALeuX, and tRNALeuZ, Rtc system protects S. enterica from the bac-
but not tRNAGln, tRNAHis, or tRNAMet (fig. S3B). tericidal effects of H2O2. The rtc S. enterica tRNA fragmentation after SOS induction of a
S. enterica DrtcBA and DrtcR mutants exper- mutants were hypersusceptible to ROS produced temperate prophage
ienced higher levels of tRNALeu and tRNATyr by the respiratory burst of bone marrow– We sought the molecular mechanism underly-
fragmentation after H2O2 treatment than did derived macrophages (Fig. 2G and fig. S4D). ing the cleavage of tRNAs in S. enterica after
WT controls (Fig. 2C and fig. S3C). Expression The DrtcBA S. enterica mutant was also at- H2O2 treatment. Gene expression in S. enterica
of the rtcBA operon in trans reversed the ab- tenuated in C57BL/6 mice, as measured by exposed to H2O2 showed that the SOS re-
normally high tRNALeuPQTV fragmentation of both a decreased competitive index when re- sponse was activated (Fig. 3A) (10, 14). Thus, we
this putative toxin-antitoxin system was not (fig. S6D). Cumulatively, these findings sup- and resulted in 3′-phosphate or 2′, 3′ cyclic
involved. port the role of Gifsy-1 terminase in tRNA phosphate and 5′-hydroxyl fragments amena-
Systematic mapping of the Gifsy-1 genome cleavage after oxidative stress. ble for RtcB repair (fig. S7D). The oxidized
(Fig. 4A) associated a gene in the capsid- and To further test the idea that the Gifsy-1 GpA protein retained the canonical DNase
DNA-packaging region with cleavage of S. enterica terminase has endoribonuclease activity, we activity (fig. S7E). Reduced and oxidized re-
tRNALeu during oxidative stress (Fig. 4B and expressed the GpA protein with an arabinose- combinant GpA differentially migrated in non-
fig. S6, A and B). The terminase encoded in the inducible promoter. In the absence of H2O2, reducing SDS gels (fig. S7F), indicating that the
Gifsy-1 gpA gene is the only determinant in the induction of the gpA gene in S. enterica did Gifsy-1 GpA terminase is redox active. Recom-
capsid- and DNA-packaging region that has not result in fragmentation of tRNA (fig. S7A). binant GpA preparations further purified by
canonical endonuclease activity. Terminases hy- However, heterologous expression of gpA si- size-exclusion chromatography retained tRNase
drolyze linear concatemers of double-stranded multaneously with the addition of 400 mM H2O2 activity upon oxidation (fig. S7, G to I), demon-
viral DNA into single genomes ready for pack- resulted in the fragmentation of tRNALeu (Fig. strating that the tRNase activity maps to GpA.
aging into mature virions. The combination 4E). These findings suggest that oxidative stress Recombinant GpA cleaved tRNALeuPQTV at the
of phylogenetic analysis and the AlphaFold either promotes endoribonuclease activity in G37 wobble position in the anticodon loop (Fig.
structure of the C-terminal domain of the the GpA terminase or that the tRNA cleavage 4G and fig. S7J). Our bioinformatic analysis
Gifsy-1 GpA revealed intriguing similarities be- is the result of the synergistic actions of GpA of the projected AlphaFold structure of the
tween this terminase and colicin family mem- with another factor induced by oxidative stress. C-terminal domain of GpA revealed a Walker
bers (Fig. 4C and fig. S6C), some of which are In support of the former model, recombinant A motif close to the predicted colicin tRNase
endoribonucleases with specificity against the GpA treated with H2O2—but not GpA treated domain (Fig. 4H). Mutagenesis of Lys497 and
anticodon loop of tRNA molecules (16). Deletion with dithiothreitol (DTT)—induced cleavage of His481 in the Walker A motif and predicted
of the gpA gene from Gifsy-1 averted tRNALeu tRNALeu in a reconstituted biochemical system tRNase domain, respectively, prevented cleav-
fragmentation after exposure of S. enterica to (Fig. 4F and fig. S7B). The tRNase activity of the age of both tRNALeuPQTV and a DNA template
H2O2 (Fig. 4D), a phenotype that could be oxidized recombinant GpA was potentiated by containing the cosN site by the oxidized GpA
complemented with the gpA gene in trans the addition of ATP in the reactions (fig. S7C) protein (Fig. 4I and fig. S7, K to M). Hence, the
oxidized form of the redox-sensitive GpA Gifsy-1 about 20% of all codons in the genome of 11. U. Das, S. Shuman, RNA 19, 1355–1362 (2013).
terminase can moonlight as a tRNase with spec- S. enterica encode for leucine residues, the 12. M. R. Manwar et al., Sci. China Life Sci. 63, 251–258
(2020).
ificity to the anticodon loop without losing its concerted effects of oxidative stress on dehy- 13. H. Temmel et al., Nucleic Acids Res. 45, 4708–4721
canonical DNase activity. droxyacid dehydratase and the tRNALeu cleav- (2017).
A strain of S. enterica lacking the Gifsy-1 age mediated by Gifsy-1 terminase will have a 14. C. A. Juan, J. M. Pérez de la Lastra, F. J. Plou, E. Pérez-Lebeña,
Int. J. Mol. Sci. 22, 4642 (2021).
gpA gene was hyperresistant to H2O2 and hy- profound effect on the nascent proteome (fig. S8). 15. J. E. Kurasz et al., J. Bacteriol. 205, e0026222
pervirulent in C57BL/6 mice when tested in The importance of the RNA repair Rtc sys- (2023).
oral and systemic models of infection (Fig. 4, tem in resistance of S. enterica to the phagocyte 16. K. Tomita, T. Ogawa, T. Uozumi, K. Watanabe,
H. Masaki, Proc. Natl. Acad. Sci. U.S.A. 97, 8278–8283
J and K, and fig. S8A). The expression of gpA NADPH oxidase indicates that this intracellular
(2000).
in trans reversed the hypersusceptibility of the pathogen benefits from recycling damaged 17. P. L. Wagner, M. K. Waldor, Infect. Immun. 70, 3985–3993
DgpA mutant to H2O2 (fig. S8B). Moreover, tRNA. Given the complex modifications involved (2002).
the presence of a functional gpA gene slowed in maturation of tRNAs (27), with most con- 18. M. A. De Groote et al., Proc. Natl. Acad. Sci. U.S.A. 94,
13997–14001 (1997).
the recovery of de novo protein translation centrated at the wobble position of the anti- 19. T. D. Ho et al., J. Bacteriol. 184, 5234–5239 (2002).
after peroxide stress (fig. S8, C and D). These codon stem loop, de novo tRNA generation 20. M. Fasnacht, N. Polacek, Front. Mol. Biosci. 8, 671037
findings indicate that when acting as an during oxidative stress must be taxing to an (2021).
21. N. Shcherbik, D. G. Pestov, Cells 8, 1379 (2019).
RNAse, the prophage GpA protein predisposes already energy- and resource-constrained cell. 22. C. L. Simms, B. H. Hudson, J. W. Mosior, A. S. Rangwala,
S. enterica to oxidative stress generated in the Recycling of damaged tRNA molecules be- H. S. Zaher, Cell Rep. 9, 1256–1264 (2014).
innate host response. comes critical in the resistance of S. enterica 23. L. I. Leichert et al., Proc. Natl. Acad. Sci. U.S.A. 105, 8197–8202
(2008).
to oxidative stress as shown by the Cybb- 24. D. H. Flint, J. F. Tuminello, M. H. Emptage, J. Biol. Chem. 268,
Discussion dependent attenuation of rtc mutants. 22369–22376 (1993).
The pathogenesis of Vibrio cholerae, enter- Why would a pathogen maintain an appar- 25. L. Macomber, J. A. Imlay, Proc. Natl. Acad. Sci. U.S.A. 106,
8344–8349 (2009).
ohemorrhagic Escherichia coli, Shigella dysenteriae, ently debilitating locus within its genome? 26. L. Yang et al., Proc. Natl. Acad. Sci. U.S.A. 116, 14368–14373
or Clostridium botulinum is intimately linked A possible answer to this paradox might lie (2019).
to the toxins encoded within their temperate in the fact that prokaryotic cells couple tran- 27. V. de Crécy-Lagard, M. Jaroch, Trends Microbiol. 29, 41–53
(2021).
bacteriophages (17). The virulence of S. enterica scription and translation. The induction of
28. S. Proshkin, A. R. Rahmouni, A. Mironov, E. Nudler, Science
has also been chiseled by the horizontal acqui- Gifsy-1–mediated tRNA cleavage through the 328, 504–508 (2010).
sition of prophages. The Gifsy-2 prophage SOS response in S. enterica suggests that the 29. B. K. Bharati et al., Nature 604, 152–159 (2022).
promotes S. enterica virulence through vacu- observed halts in translation are caused by 30. S. Kant et al., PLOS Biol. 21, e3002051 (2023).
31. S. Uppalapati et al., Data from: Prophage terminase with
olar remodeling by an effector of the pathoge- double-stranded DNA damage. The concerted tRNase activity sensitizes S. enterica to oxidative stress, Dryad
nicity island-2 type III secretion system and inhibition of leucine biosynthesis and cleav- (2024); https://doi.org/10.5061/dryad.xpnvx0knt.
the antioxidant properties of a Cu/Zn super- age of tRNALeu must slow ribosomes through
AC KNOWLED GME NTS
oxide dismutase (18, 19). In contrast to the the nascent transcript. Decelerating ribosomes
We thank J. Jones-Carson for kindly providing the mice. We
archetypical lysogenic bacteriophages that permits backtracking of RNA polymerase thank J. Hesselberth for technical discussions on tRNA Northern
provide virulence traits to their bacterial hosts, (28), which would enable access of repair en- blotting and for helpful discussions with the manuscript. We thank
our investigations show that the Gifsy-1 pro- zymes to DNA lesions (29). Therefore, Gifsy-1– H. V. Batra for valuable discussions on protein expression and
purification. We also thank A. Margolis for comments on the
phage reduces fitness of S. enterica in the mediated inhibition of translation in H2O2- manuscript, and we thank the Mass Spectrometry Proteomics
context of ROS-induced oxidative stress pro- treated S. enterica may help maintain the in- Shared Resource Facility, University of Colorado Anschutz Medical
duced by phagocyte NADPH oxidase in mice. tegrity of the genome. In this context, S. enterica Campus, for protein identification. Funding: This work was
supported by VA Merit Grants BX0002073 and IK6BX005384 and
Thus, the redox-responsive terminase encoded may hijack host cell–derived ROS to activate a by NIH grants R01AI54959 and R01AI136520. M.M. was funded in
in the capsid- and DNA-packaging region of lysogenic phage determinant that halts bacte- part by NIH grant R03AI139557. Author contributions:
the Gifsy-1 prophage gains endoribonuclease rial translation and growth during intense Conceptualization: S.U. and A.V.T. Data curation: S.U., L.L., S.K.,
D.O., and A.V.T. Formal analysis: S.U., L.L., S.K., D.O., and A.V.T.
activity upon oxidation, sensitizing S. enterica periods of oxidative stress, providing oppor- Funding acquisition and resources: A.V.T. and M.M. Investigation:
to oxidative killing. tunities for repair and survival. In addition, S.U., L.L., S.K., D.O., and A.V.T. Methodology: S.U., L.L., S.K., and
ROS damage almost every cogwheel of the the moonlighting tRNase function of GpA D.O. Supervision and validation: J.S.K. and A.V.T. Writing – original
draft, review, and editing: S.U. and A.V.T. Competing interests:
translational machinery, including ribosomes, precludes phage-genome processing and cap-
The authors declare that they have no competing interests.
elongation factors, mRNA, tRNA, and tRNA sid maturation, protecting the bacterial host Data and materials availability: All data needed to evaluate
synthetases [reviewed in (20)]. ROS oxidize from the lytic cycle at times of exposure to the conclusions in the paper are present in the paper and/or in the
cysteine and methionine residues in small oxidative stress exerted by the innate immune supplementary materials. RNA-seq data are available in the GEO
database under accession no. GSE203342. Bacterial cultures
(S17) and large (L7/L12, L14) ribosomal sub- response. and plasmids generated in the study are available upon request
unit proteins and damage ribonucleotides that to the corresponding author. All the data and unprocessed images
can then be misincorporated into the transcrip- utilized to draw figures in the manuscript are available at Dryad
RE FERENCES AND NOTES (31). License information: Copyright © 2024 the authors, some
tome (21–23). Moreover, the [4Fe-4S] cluster of rights reserved; exclusive licensee American Association for the
1. P. Mastroeni et al., J. Exp. Med. 192, 237–248 (2000).
the ilvD-encoded dihydroxyacid dehydratase 2. F. C. Fang, Nat. Rev. Microbiol. 2, 820–832 (2004). Advancement of Science. No claim to original US government
in the biosynthesis of branched-chain amino 3. S. Jang, J. A. Imlay, J. Biol. Chem. 282, 929–937 works. https://www.science.org/about/science-licenses-journal-
(2007). article-reuse
acids is especially vulnerable to the toxic ef-
4. B. Ezraty, A. Gennaris, F. Barras, J. F. Collet, Nat. Rev. Microbiol.
fects of ROS (24–26). Cleavage of tRNALeu iso- 15, 385–396 (2017).
acceptors should be included in the mechanism 5. J. S. Kim et al., Proc. Natl. Acad. Sci. U.S.A. 115, SUPPLEMENTARY MATERIALS
by which oxidative stress negatively impacts E11780–E11789 (2018). science.org/doi/10.1126/science.adl3222
6. L. E. Leiva et al., Front. Genet. 11, 856 (2020). Materials and Methods
translation in S. enterica. In contrast to the Figs. S1 to S9
7. V. N. Fritsch et al., Free Radic. Biol. Med. 161, 351–364
direct damage caused by ROS to ribosomes, (2020). Tables S1 to S7
biosynthetic enzymes, and ribonucleotides, 8. C. A. Goodman, T. A. Hornberger, Exerc. Sport Sci. Rev. 41, References (32–39)
107–115 (2013). MDAR Reproducibility Checklist
the cleavage of tRNA under oxidative stress
9. S. Chakraborty et al., Nat. Commun. 11, 1783 (2020).
is an indirect result of the SOS-mediated ex- 10. K. J. Hughes, X. Chen, A. M. Burroughs, L. Aravind, S. L. Wolin, Submitted 11 October 2023; accepted 28 February 2024
pression of the Gifsy-1 terminase. Because Cell Rep. 33, 108527 (2020). 10.1126/science.adl3222
PROTEIN DESIGN this process, COMBS uses van der Mers (vdMs)
to search for preferred spatial positions where
De novo design of drug-binding proteins with chemical groups can interact with an amino
acid (3). vdMs are similar to rotamers (16),
predictable binding energy and specificity which define favorable positions to arrange an
amino acid’s sidechain atoms relative to its
Lei Lu1, Xuxu Gou2, Sophia K. Tan1, Samuel I. Mann1,3, Hyunjun Yang1, Xiaofang Zhong4, backbone atoms. However, vdMs instead define
Dimitrios Gazgalis5,6, Jesús Valdiviezo5,6, Hyunil Jo1, Yibing Wu1, Morgan E. Diolaiti2, Alan Ashworth2, favorable positions of an interacting chemical-
Nicholas F. Polizzi5,6*, William F. DeGrado1* group fragment relative to a residue’s backbone
atoms. Each amino acid type can adopt mul-
The de novo design of small molecule–binding proteins has seen exciting recent progress; however, high- tiple rotamers, which are widely used in design
affinity binding and tunable specificity typically require laborious screening and optimization after algorithms to position sidechains onto preexist-
computational design. We developed a computational procedure to design a protein that recognizes a ing backbones. In vdMs, we track the positions
common pharmacophore in a series of poly(ADP-ribose) polymerase–1 inhibitors. One of three designed of chemical groups that can interact with a
proteins bound different inhibitors with affinities ranging from <5 nM to low micromolar. X-ray crystal given type of amino acid. For example, a
structures confirmed the accuracy of the designed protein-drug interactions. Molecular dynamics GlnCONH2 vdM consists of a glutamine residue
simulations informed the role of water in binding. Binding free energy calculations performed directly on that contacts a carboxamide group. Also like
the designed models were in excellent agreement with the experimentally measured affinities. We rotamers, vdMs can be clustered into similar
conclude that de novo design of high-affinity small molecule–binding proteins with tuned interaction groups with associated probabilities based on
energies is feasible entirely from computation. their occurrence in the Protein Data Bank (PDB).
The COMBS algorithm then finds multiple posi-
tions on a given protein backbone that can
olecular recognition underlies catalyt- structure and sequence as well as scoring simultaneously form favorable van der Waals,
M
ic activity of enzymes and drives sig- functions that rely on a mix of statistical and aromatic, and/or hydrogen-bonded interactions
naling of protein receptors. Although physical terms without explicit representation with the chemical groups of a target small mol-
we have an advanced understanding of dynamics, conformational entropy, or water ecule. The remainder of the sequence is then
of both protein design and molecular (3–7). This dependence leaves open the funda- filled in through flexible backbone sequence
interactions, the rational design of de novo mental question of whether our understand- design.
proteins that specifically bind small molecules ing is grounded in physical forces or limited COMBS showed promise in identifying pro-
with low nanomolar to picomolar affinity is exclusively to advanced pattern recognition tein backbones and creating sites capable of
a major challenge (1, 2) that has not been (15). In this study, we asked whether adherence binding the drug apixaban. In previous work,
achieved in de novo proteins (3, 4) without to simple rules based on physical principles we identified one protein that bound with
experimental screening of large libraries of might increase the success rate for designing Kd = 500 nM and a second with Kd = 5 mM
variants (5–7). Even with the application of drug-binding proteins and whether all-atom after screening only six designed sequences
recent advances in artificial intelligence to molecular dynamics (MD) simulations with (3). In the process, we learned lessons that could
facilitate de novo design (8–10), it has been explicit water might be able to recapitulate improve the use of COMBS in design pipelines.
necessary to screen thousands of independent the experimentally determined binding affini- Firstly, during the final steps of sequence de-
designs to discover binders with low-micromolar ties of high-affinity binders, starting with design sign and backbone optimization, sidechains
to high-nanomolar dissociation constants (Kd) models that come directly from computation. were sometimes introduced that ultimately
directly from design algorithms (3, 11–14). Pro- did not form their intended favorable vdMs.
teins with higher affinity are often desirable. Rationale for design of high-affinity Structural analysis showed that they were not
Given the low success rate, screening large drug-binding proteins in optimal orientations to interact with the
numbers of designs often relies on biotinylated The recognition of polar groups presents a drug, and site-directed mutagenesis showed
or fluorescently labeled versions of their small- challenge in de novo design of binders because that they made small or no contributions to
molecule targets, which restricts the region of the polar groups must lose most or all of their the free energy of binding. We reasoned that
the molecule available for binding. The cost of highly favorable interactions with water mol- it should be possible to improve binding by
synthesizing thousands of genes and the ne- ecules upon binding to the protein. To com- using backbone ϕ- or y-dependent vdMs and
cessity of synthetic chemistry for conjugation pensate for this loss in hydration, the polar alternating rounds of COMBS with Rosetta
places a practical limitation on the access of chemical groups of the small molecule must flexible-backbone sequence design. Secondly,
these methods to many groups. Moreover, de form highly directional and distance-dependent binding affinity can be optimized by pre-
novo design methods rely heavily on struc- hydrogen bonds and electrostatic interactions organizing a receptor’s conformation so that it
tural informatics to guide sampling of protein with atomic groups in the protein. These polar loses minimal conformational entropy upon
interactions are not only required for affinity, but binding. Originally, we analyzed preorganiza-
they also provide the specificity of proteins for tion using Rosetta ab initio folding; now we
1
Department of Pharmaceutical Chemistry & Cardiovascular their substrates over other similarly shaped can more reliably determine whether the se-
Research Institute, University of California, San Francisco,
CA 94158, USA. 2Helen Diller Family Comprehensive Cancer molecules. quence adopted the desired reorganized struc-
Center, University of California, San Francisco, CA 94158, We recently developed a design method ture using the AlphaFold2 program (17).
USA. 3Department of Chemistry, University of California, known as Convergent Motifs for Binding Sites Another important feature is the need to
Riverside, CA 92521, USA. 4Department of Cellular and
Molecular Pharmacology, University of California, San
(COMBS) to enable sampling of only sequences consider the energetically unfavorable loss of
Francisco, CA 94158, USA. 5Department of Cancer Biology, and structures capable of forming such in- hydrogen bonds between water molecules and
Dana-Farber Cancer Institute, Boston, MA 02215, USA. teractions prior to searching for less specific both the drug as well as the protein’s bind-
6
Department of Biological Chemistry and Molecular
Pharmacology, Harvard Medical School, Boston, MA 02215, USA.
and directional, but energetically favorable, ing site, which occurs when the drug binds
*Corresponding author. Email: nicholasf_polizzi@dfci.harvard. van der Waals and hydrophobic interactions the pocket. Although COMBS and other algo-
edu (N.F.P.); william.degrado@ucsf.edu (W.F.D.) that complete the binding site (3). To facilitate rithms (12) consider the need to form hydrogen
pLDDT
difference test scores (pLDDTs) concurred with 80
the trend of RMSD difference of the design model. Binding site
70
For example, the N terminal, C terminal, and the includes residues
17 to 31, 48 to 62, 60
middle loop with low pLDDTs (<90) showed higher 88 to 102, 121 to 135
Ca RMSD. (C) The design model showing that 50
the polar groups of rucaparib are all hydrogen 0 50 100 150
bonded. (D and E) A fluorescence titration showed Residue
that PiB and PiB′ bind rucaparib with Kd < 5 nM. C
The fluorescence emission intensity at 420 nm Binding site Asp131
Asp29
of rucaparib (excitation wavelength, 355 nm) Cα RMSD 0.41 Å
was measured after titrating aliquots of PiB (D) Upper fold
or PiB′ (E) to a final concentration indicated in the Cα RMSD 0.49Å
All polar groups are
Middle loop Overall
abscissa, in which [PIB]T/[ruc]T or [PIB′]T/[ruc]T hydrogen-bonded.
Cα RMSD 0.67Å
refers to the ratio of the molar concentrations PiB Alphafold model
of PiB or PiB′ to rucaparib, respectively. The data vs Phe123
are well described by a single-site protein-ligand PiB Design model
(N terminal to C terminal) Gln54
binding model, and a nonlinear least-squares fit to
the data returned Kd values of 2.2 ± 0.9 nM for
PIB and 0.37 ± 0.29 nM for PiB′. Although the fitting
D 0.1 nM 1 nM 5 nM 10 nM E 0.1 nM 1 nM 5 nM 10 nM
30000 PiB Kd = 2.2 nM ± 0.9 nM PiB´ Kd = 0.37 nM ± 0.29 nM
Emission @ 420nM (104 a.u.)
30000
error was relatively small, a sensitivity analysis in
should serve to bind a wide range of com- (Figs. 1E and 2, A and B), as a third round of total vdM cluster scores); (ii) satisfaction of all
pounds. We targeted three chemical groups COMBS showed that the vdMs were now opti- potential buried H-bond donors and acceptors
in rucaparib’s structure: the indole NH and mal. The final designs included numerous CH-p in the protein and ligand; (iii) low Rosetta en-
the C=O and NH2 groups of its carboxamide and hydrophobic interactions interspersed with ergy (lowest 50 of the 1000 total designs); and
(Fig. 1C). Additionally, COMBS identified Asp58 specific polar interactions, including four hydro- (iv) avoidance of clashes with the three other
as a second-shell interaction to the carboxa- gen bonds (a H-bond donor to the drug’s car- PARP inhibitors, which show structural varia-
mide of rucaparib (Fig. 1C). It is important to boxamide oxygen as well as three H-bond bility near the amine end of the molecule. Top-
design binding interactions with these groups acceptors to the drug’s carboxamide NH, indole scoring designs selected for expression were
with sub-Ångstrom accuracy to engender spec- NH, and charged ammonium group), as well as designated PARPi binder (PiB) and its variant
ificity and a favorable free energy of association. second-shell interactions. (Fig. 2C and fig. S2). PiB′ (fig. S3 and table S1). PiB′ differs from PiB
Next, the remainder of the sequence was de- Throughout the design process, we ensured by the substitution of five solvent-exposed
signed by using Rosetta flexible backbone de- that the designs would also retain favorable charged residues with alanine to encourage
sign (Fig. 1D) (3, 20) while retaining the identity interactions with most of the common phar- crystallization. The other two proteins chosen
of the keystone residues (identified in the macophores of the three other drugs (materials for expression (PiB-1 and PiB-2) were less closely
COMBS step). The mainchain moved 1-Å root and methods). However, we predicted that the related to PiB in structure (RMSD = 0.93 and
mean square deviation (RMSD) during this protein would have lower affinity for niraparib 0.79 for PiB-1 and PiB-2 to PiB, respectively) and
step (fig. S1), so a second round of vdM sam- and mefuparib because they lack the H-bonding sequence (41 and 42% identity for PiB-1 and PiB-
pling was performed on the relaxed backbone. group indole NH of rucaparib. Additionally, we 2 to PiB, respectively) (fig. S4). Circular dichro-
This procedure identified three mutants involv- expected veliparib to bind weakly because it ism spectroscopy showed that all four had
ing drug-contacting residues, including N29D, lacks a hydrophobic phenyl group and the posi- substantial a-helical character (fig. S5). How-
W90L, and N131D (Fig. 1E). A second round of tion of its charged ammonium group differs ever, PiB-1 and PiB-2 failed to induce large
flexible backbone sequence design with this from that found in the other three drugs. changes in the fluorescence emission spectrum
backbone and the newly fixed vdMs resulted The final models were chosen based on of rucaparib (fig. S6); therefore, we focused our
in converged sequence-structure combinations multiple criteria: (i) Favorable vdMs (highest efforts on PiB and PiB′ (figs. S7 to S11).
0 0
0 2 4 6 8 10 0 2 4 6 8
[PiB]T / [ruc] T [mef] T / [PiB] T
Kd = 630 ± 16 nM Kd = 14 ± 5 µM
0 0
0 2 4 6 8 0 2 4 6 8
[nir] T / [PiB] T [vel]T / [PiB] T
50 50 50 50
0 0 0 0
-1 0 1 2 3 4 -1 0 1 2 3 4 -1 0 1 2 3 4 -1 0 1 2 3 4
Rucaparib (log10 nM) Mefuparib (log10 nM) Niraparib (log10 nM) Veliparib (log10 nM)
Equivalents of
100
PIB added
0 1 2.5 5 10
Olaparib
Cell viability (%)
O
NH O Rucaparib 161 597 >104 n/a n/a
N N N
O 50 Mefuparib 1278 3192 >104 n/a n/a
F
Niraparib 201 233 417 >104 >104
Fig. 3. Spectral titrations and cell viability assay of PiB with PARPi. (A) The rucaparib, mefuparib, niraparib, and veliparib toxicity in a dose-dependent manner.
Kd values of various drugs for PiB as obtained from a global fit of a single-site The PARP inhibitors were preincubated with PiB in media at room temperature for
binding model to the fluorescence or absorbance changes as a function of the 5 min at multiple concentration ratios (protein:ligand) of 0:1, 1:1, 2.5:1, 5:1, and 10:1.
concentration of PiB or drugs. The indicated wavelengths for the titration were chosen (C) Cell viability assay as in Fig. 3B showing that PiB had no effect on the olaparib
to maximize the difference in absorption for the free versus bound drug. (B) Growth dose response. (D) Table showing IC50 values for the inhibition of cell proliferation by
assays in DLD-1 BRCA2-mutated cells showed that PiB alleviates the effects of PARPi drugs in the presence of increasing mole ratios of added PiB protein.
Spectral titrations showed that PiB and PiB′ showed that it folded into a well-defined struc- lowest experimentally feasible rucaparib con-
bound the PARPi drugs with high affinity. ture and that the addition of a single equiv- centration, the binding isotherms showed a
Incubation of PiB with equimolar concen- alent of rucaparib led to a new set of peaks, linear increase in intensity with respect to
trations of rucaparib led to a marked blue which was consistent with a stoichiometric, protein concentration until a single equivalent
shift and an increase in intensity of its fluo- specific complex (figs. S10 and S11). Fluores- was added, followed by an abrupt leveling at
rescence spectrum, as would be expected if cently monitored titrations of protein into a higher protein concentrations. This behavior
its indole core were bound in a rigid, solvent- solution of rucaparib showed that PiB and is indicative of a dissociation constant that is
inaccessible site (figs. S6 and S7). Nuclear PiB′ bound with very low to subnanometer af- much lower than the total rucaparib concen-
magnetic resonance spectroscopy of PiB finity (Figs. 2, D and E, and 3A). Even at the tration. A nonlinear least-squares fit to the data
returned a Kd of 2.2 nM for PiB and 0.37 nM. niraparib (Kd = 600 and 550 nM). The cor- of cells with certain DNA repair deficiencies
For PiB′, and a sensitivity analysis showed that responding Kd values of PiB and PiB′ were including loss-of-function mutations in BRCA2
the Kd was <5 nM for both proteins (Fig. 2, D 14 and 24 mM, respectively, for the structurally (18). To determine whether PiB and PiB′ could
and E). Achieving single-digit nanomolar to divergent drug veliparib, and no binding was attenuate the lethal effects of PARPi drugs, we
picomolar binding affinity for de novo–designed detected for the most divergent drug, olaparib measured their effects on the growth of BRCA2-
proteins has previously required extensive ex- (fig. S13). This observed trend in binding affin- mutated DLD-1 cells and SUM149 cells (22).
perimental optimization (5). ity matched the order expected from the struc- Dose-response curves were first established in
Ultraviolet-visible absorption titrations showed tural differences mentioned previously. the absence of PiB, then the titration was
that PiB and PiB′ also bound to the remaining We next examined the in vitro stability and repeated with PiB or PiB′ at varying protein/
ligands with affinities that grew increasingly potency of PiB and PiB′ in serum and cellular drug ratios for each PARPi drug concentration.
weaker as the drugs’ structures diverged from assays. PiB and PiB′ were highly stable in human The addition of a single equivalent of PiB or
rucaparib (Fig. 3A and fig. S12). PiB retained serum, as are other de novo proteins designed PiB′ resulted in a fourfold increase in the half-
submicromolar affinity for mefuparib (Kd = 190 for medical applications (figs. S14 and S15) maximal inhibitory concentration (IC50) value
and 350 nM for PiB and PiB′, respectively) and (3, 21). PARPi drugs potently inhibit the viability for rucaparib. Thus, PiB competes effectively
for binding of rucaparib to human PARP1, an
enzyme reported to bind rucaparib with a Kd
of 0.1 to 1 nM in biochemical assays (23, 24)
(Fig. 3, A, B, and D, and figs. S16 to S18). The
A B Asp131 Ruc-bound PiB´ potencies of PiB and PiB′ in the cellular assay
Asp131
vs were similar and decreased in the progression
Design
of rucaparib to mefuparib and niraparib to
veliparib, which was in good agreement with
the spectroscopic assays (Fig. 3, B and D, and
Asp29 Asp29 figs. S16 to S18). Moreover, PiB and PiB′ did
Phe123 not appreciably change the cellular response
Phe123
to olaparib (Fig. 3, C and D, and figs. S16 to
Gln54 Gln54 S18), which agreed with spectroscopic data
that indicated that PiB and PiB′ do not bind
C Open, water-filled binding pocket in drug-free PiB´ this drug.
Asp131 Structural and mutational validation
of designs
The crystallographic structures of PiB′ were
Asp131 Asp29
Asp29 solved in the absence and presence of the four
Ruc-bound PiB´ active compounds at 1.3- to 1.6-Å resolution
Phe123 (table S2). The protein’s conformation was in ex-
vs
Phe123
Design Gln54 cellent agreement with the predicted Alpha-
Gln54 Fold2 model, particularly near the binding
site [a carbon (Ca) RMSD of the 60 sur-
D E rounding residues was 0.2 to 0.5 Å] (Figs. 2A
Relative binding affinity
2.0 4 and 4, A and B; fig. S19; and table. S3). A
of PiB alanine mutants
similar degree of agreement (<0.5 Å RMSD)
1.5 3 was observed comparing the experimental
structure and the designed model. An import-
(kcal/mol)
Relative binding
(kcal/mol)
affinity of PiB ant aspect of the design was that the binding
1.0 2
mutants before site should be preorganized. Indeed, the bind-
vdM optimization. ing pocket of PiB′ was nearly identical be-
0.5 1 tween the experimentally determined drug-free
and drug-bound structures (0.2 to 0.5 Å Ca
0.0 0 RMSD) (Fig. 4, B and C, and fig. S20). Moreover,
the sidechains interacted precisely as predicted
N
0W
1N
A
A
D A
1A
4
29
54
23
29
L2
13
13
L9
Direct H-bond
Water-mediated Computational prediction of binding
0.5 H-bond thermodynamics
To provide insight into the experimental re-
sults for PiB, PiB′, PiB-1, and PiB-2, we next
conducted 2-ms all-atom MD simulations to
NA NA compare their structural dynamics on this
0
time scale. The simulations were performed
uc
ef
ir
uc
ef
ir
uc
ef
ir
l
Ve
Ve
Ve
N
N
M
M
on the designed models (instead of the crys-
R
R
tallographic structures) to assess the use of
B C MD as a predictor of experimental success.
GROMACS G Experimental G P value = 0.008 The protein backbone conformations were very
Protein Ligand (kcal/mol) (kcal/mol) stable for all four complexes. However, rucaparib’s
at 298 K at 298 K
designed binding pose was stable only in PiB and
Computational G (kcal/mol)
-4
PiB Rucaparib -12.34 -11.82 PiB′ (fig. S25A) (because PiB and PiB′ behave
PiB Mefuparib -9.76 -9.18 similarly in MD, we only discuss PiB): it re-
PiB Niraparib -8.77 -8.45 -8 tained its bivalent hydrogen bonding interac-
PiB Veliparib -6.21 -6.26 tion to Gln54 (Fig. 5A), and Asp29 and Asp131
PiB´ Rucaparib -14.02 -12.86 showed stable interactions with rucaparib’s
-12
PiB´ Mefuparib -11.82 -8.8
indole NH and ammonium groups through
direct and water-mediated hydrogen bonds,
PiB´ Niraparib -11.18 -8.53
-16 respectively (Fig. 5A). By contrast, PiB-1 and
PiB´ Veliparib -9.25 -6.3
-16 -12 -8 -4 PiB-2 simulations deviated from rucaparib’s
PiB-D29N Rucaparib -10.19 -10.72 designed pose, and their key buried H bonds
Experimental
PiB-Q54A Rucaparib -9.11 -9.86 G (kcal/mol) to Gln54 were broken within 50 ns in each of
PiB-D131A Rucaparib -8.056 -9.15 three independent calculations (fig. S25). More-
PiB-F123A Rucaparib -7.54 -8.82 over, PiB shielded the apolar atoms in rucaparib
more efficiently than PiB-1 and PiB-2, as de-
Fig. 5. The MD simulations of PiB, PiB′, and mutants. (A) By using unbiased MD simulations in Amber, we termined from solvent-accessible surface area
calculated (in triplicate) the frequency with which the intermolecular hydrogen bonds formed between the calculations within individual MD trajectories
protein scaffold and the bound drug molecule. PiB was found to form a hydrogen bond between Gln54 and the (fig. S26). Furthermore, MD simulations of PiB
targeted drug carboxamide in 100% of all simulations for each drug complex. The charged ammonium in complex with niraparib, veliparib, and mefu-
groups of rucaparib and mefuparib interacted with Asp29 through a combination of direct and water-mediated parib showed similar binding-site conforma-
hydrogen bonds, totaling to more than half of the full simulation time, which contrasts niraparib and tional stability as PiB-rucaparib over 2 ms (Fig.
veliparib’s inabilities to form equivalent hydrogen bonds (owing to changes in chemical structure around 5A and fig. S27). MD can thus help rationalize
the ammonium tail of the ligand). In a small fraction of each rucaparib and veliparib trajectory, Asp131 how interactions contribute to stability in pre-
engaged in water-mediated hydrogen bonds to the drugs. (B) By using biased simulations in GROMACS, we dicted complexes and may be a useful tool in
calculated the binding free energy (DG) for each ligand and found that ranked affinity for each drug was design.
consistent with experimental results. (C) Comparison of DG from the GROMACS calculation with the We next turned to alchemical and physical
experimental value from spectral titrations. pathway methods to determine whether these
methods could predict the absolute binding
free energies for PiB and PiB′ directly from
MD simulations, using the computational de-
The structures of mefuparib and niraprib whether these substitutions indeed increased signs (rather than experimental structures) as
bound to PiB′ showed a similar set of inter- affinity, we evaluated mutants with the second- the starting models. The alchemical transfer
actions as rucaparib did when bound to PiB′ round substitutions reverted to their identities method (26, 27) was carried out by one of the
(fig. S22). However, their aromatic five-membered in the first round of design. These changes each authors who had no knowledge of the experi-
azole rings lacked a H-bonding group to interact led to one to two orders of magnitude–weaker mental results. This method has been shown
with Asp131, explaining their decreased affinity binding affinity for rucaparib D29N (Kd = to be comparable to other alchemical methodol-
for the protein. As expected from its divergent 13 nM), L90W (Kd = 24 nM), and D131N (Kd = ogies, such as Schrodinger’s Free Energy Pertur-
structure, veliparib had a less favorable fit with 50 nM) (Fig. 4D and fig. S24A). Thus, iterative bation plus (FEP+) (26) or Amber’s thermodynamic
PiB′s binding site, and it lacked a salt bridge to vdM selection successfully identified interac- integration (27), given comparable sampling
its ammonium group as in other complexes tions with improved binding affinity. We sug- of the configurational space. An initial abso-
(figs. S22 and S23). gest that vdM-guided amino acid optimization lute binding free energy calculation was used
Three residues were changed to improve might provide a useful alternative to other to evaluate the energetic contributions of the
binding during the second round of COMBS methods of affinity optimization. We also con- fused-ring cores of the drugs and to ensure
design of PiB (after the first flexible-backbone ducted an alanine scan to probe the energetic convergence of the calculations. An additional
sequence-design step; no experiments were contribution of each of the residues that lined relative binding free energy calculation was per-
performed between rounds). To determine the pocket in the rucaparib complex (Fig. 4E formed to transform each core into the target
ligand to estimate the contribution from non- tained excellent quantitative agreement between 19. A. Koehl, M. Jagota, D. D. Erdmann-Pham, A. Fung, Y. S. Song,
core regions (fig. S28). Universally, the alche- computed and experimentally measured binding Pac. Symp. Biocomput. 27, 22–33 (2022).
20. S. I. Mann, A. Nayak, G. T. Gassner, M. J. Therien,
mical transfer method tended to overestimate free energies using only the designed models W. F. DeGrado, J. Am. Chem. Soc. 143, 252–259 (2021).
the binding energy, possibly owing to having as the input structures. These data demon- 21. D.-A. Silva et al., Nature 565, 186–191 (2019).
two sets of restraint potentials. However, this strate the possibility of designing proteins 22. T. Hucl et al., Cancer Res. 68, 5023–5030 (2008).
23. J. Rudolph, G. Roberts, K. Luger, Nat. Commun. 12, 736 (2021).
procedure correctly predicted the relative affin- with high affinity (Kd < 5 nM) to small mol- 24. H. D. Thomas et al., Mol. Cancer Ther. 6, 945–956 (2007).
ities of the four ligands (table S4). ecules by using fully rational criteria for design 25. A. Fersht, Structure and Mechanism in Protein Science:
We next used potential of mean force cal- and “physics-based” force fields to evaluate the A Guide to Enzyme Catalysis and Protein Folding (Macmillan,
1999).
culations, an orthogonal physical-pathway complexes. 26. L. Wang et al., J. Am. Chem. Soc. 137, 2695–2703 (2015).
methodology (28), to compute absolute binding Rucaparib binds to the human PARP1 enzyme 27. T.-S. Lee et al., J. Chem. Inf. Model. 60, 5595–5623 (2020).
energies (figs. S29 to S33), and the results were with a Kd ranging from 0.1 to 1 nM depending 28. C. Chipot, Annu. Rev. Biophys. 52, 113–138 (2023).
29. E. P. Barros et al., J. Chem. Theory Comput. 15, 5703–5715
in remarkably good agreement with those of on the experimental conditions (23, 24), which (2019).
the experiment (Fig. 5, B and C, and table. S5). is close to the range observed for PiB and PiB′. 30. G. Ulas, T. Lemmin, Y. Wu, G. T. Gassner, W. F. DeGrado,
The RMS error between the predicted and ex- Ligand efficiency is often used as a guiding rule Nat. Chem. 8, 354–359 (2016).
31. M. Gill, M. E. McCully, Protein Eng. Des. Sel. 32, 317–329 (2019).
perimental values is 1.3 kcal/mol, and the in drug discovery to determine whether the
32. A. Chevalier et al., Nature 550, 74–79 (2017).
correct rank order of affinities was observed. affinity of a molecule of a given size is within a 33. M. C. Childers, V. Daggett, Mol. Syst. Des. Eng. 2, 9–33
This error was close to the experimental error range typically seen in highly optimized small- (2017).
34. A. L. Hopkins, G. M. Keserü, P. D. Leeson, D. C. Rees,
in the measurement of Kd for rucaparib. We molecule drugs and natural organic ligands C. H. Reynolds, Nat. Rev. Drug Discov. 13, 105–121 (2014).
also obtained very good agreement between for proteins (34, 35). As ligands become larger, 35. I. D. Kuntz, K. Chen, K. A. Sharp, P. A. Kollman, Proc. Natl.
computation and experimental results for a they have more opportunities to form favorable Acad. Sci. U.S.A. 96, 9997–10002 (1999).
36. L. Lu, Zenodo (2024). https://doi.org/https://doi.org/
set of four mutants of PiB (Fig. 5, B and C, interactions with their target proteins. Thus, 10.5281/zenodo.10653015.
and table S5). This agreement bodes well for the maximal affinity possible roughly scales
the use of alchemical and physical pathway– with the size of a small molecule, and the li- AC KNOWLED GME NTS
based binding free energy calculations to gand efficiency is defined as the free energy of We thank members of the DeGrado lab for useful discussions and
G. Meigs from Lawrence Berkeley National Laboratory for helping
evaluate potential binding energies of de novo binding (1 M standard state) divided by the with data collection. Funding: W.F.D. acknowledges research
small-molecule binding proteins. number of heavy atoms in the ligand. Most support from grants from the National Institutes of Health
drugs have ligand efficiency of around 0.3 kcal/ (2R35GM122603) and the National Science Foundation (grant NSF
Discussion (mol × heavy atom count) (34, 35), although 2108660 and 2306190 NSF MCB BSF). N.F.P. acknowledges
support from the NIH (R00GM135519). A.A. acknowledges research
In this study, we used structural bioinformatics higher values are observed for highly optimized support from SPARC. We collected x-ray data on beamline 8.3.1 of the
to drive the de novo design of a high-affinity drugs such as rucaparib, which has a ligand Advanced Light Source, which is supported by the DOE Office of
Science User Facility under contract no. DE-AC02-05CH11231.
drug-binding protein. Molecular dynamics efficiency of 0.5 kcal/(mol × heavy atom count).
Beamline 8.3.1 is also supported in part by the ALS-ENABLE program of
simulations accurately captured the trend in The ligand efficiency of a drug is similarly a the National Institutes of Health, National Institute of General Medical
binding affinity for the series of drugs and good measure of how well optimized a de novo Sciences grant P30 GM124169-01. Author contributions: L.L. and
predicted the role of water in binding. Our results protein is for binding to a small molecule. The W.F.D. conceived the idea for the project; L.L. and N.F.P. developed
computer code; L.L. performed the computational design; L.L.,
portend a future in which artificial intelligence– 0.5 kcal/(mol × heavy atom count) ligand effi- X.G., S.I.M., X.Z., and H.J. performed the experiments; L.L., H.Y.,
enabled sampling of conformational and se- ciency of PiB is a considerable improvement X.G., and Y.W. performed the data analysis; N.F.P., S.K.T., D.G.,
quence spaces are seamlessly interfaced with over the 0.21 to 0.26 ligand efficiency of the and J.V. ran and analyzed MD stimulations; X.G., M.E.D., A.A., and
W.F.D. designed the cell assay experiments; and L.L., N.F.P., and
physics-based models to design proteins and first COMBS-designed apixaban binders, which W.F.D. wrote the paper with input from all authors. Competing
biomimetic polymers with precisely predefined demonstrates the importance of incorporating interests: A.A. is a cofounder of Tango Therapeutics, Azkarra
binding affinity, specificity, and, given the inti- the design principles discussed above. With Therapeutics, Ovibio Corporation, and Kytarro; a member of the
board of Cytomx and Cambridge Science Corporation; a member
mate relation between binding and catalysis, these and similar refinements, it should be in- of the scientific advisory boards of Genentech, GLAdiator, Circle,
catalytic properties. To achieve binding of creasingly possible to design high-affinity small Bluestar, Earli, Ambagon, Phoenix Molecular Designs, Yingli, ProRavel,
complex polar molecules, sampling favorable molecule–binding proteins with predicable bind- Oric, Hap10, and Trial Library; a consultant for SPARC, ProLynx,
Novartis, and GSK; and holds patents on the use of PARPi held
geometries for ligand-protein interactions is ing energetics for a variety of practical appli- jointly with AstraZeneca from which he has benefited financially (and
essential. In this work, vdMs and COMBS were cations in sensing and pharmaceuticals. may do so in the future). Data and materials availability:
key to this achievement. Although they were Coordinates and structure files have been deposited to the PDB
with accession codes: 8TN1 (drug-free PiB′), 8TN6 (rucaparib-bound
used in conjunction with Rosetta sequence RE FERENCES AND NOTES
PiB′), 8TNB (mefuparib-bound PiB), 8TNC (niraparib-bound PiB′),
design, they could be easily adapted to improve 1. B. Schreier, C. Stumpp, S. Wiesner, B. Höcker, Proc. Natl. Acad. and 8TND (veliparib-bound PiB′). Computational code and design
sampling and/or scoring of a variety of recently Sci. U.S.A. 106, 18491–18496 (2009). scripts are available in the supplementary materials and at Zenodo
2. W. Yang, L. Lai, Curr. Opin. Struct. Biol. 45, 67–73 (2017). (36). All other relevant data are available in the main text or the
developed protein-design algorithms based on 3. N. F. Polizzi, W. F. DeGrado, Science 369, 1227–1233 (2020). supplementary materials. License information: Copyright © 2024
diffusion models (3, 8–10). MD simulations of 4. F. Thomas et al., ACS Synth. Biol. 7, 1808–1816 (2018). the authors, some rights reserved; exclusive licensee American
de novo proteins have only been occasionally 5. C. E. Tinberg et al., Nature 501, 212–216 (2013). Association for the Advancement of Science. No claim to original
6. J. Dou et al., Protein Sci. 26, 2426–2437 (2017). US government works. https://www.science.org/about/science-
used to provide insight into de novo protein 7. J. Dou et al., Nature 561, 485–491 (2018). licenses-journal-article-reuse
design (29–33). However, by using this method, 8. J. L. Watson et al., Nature 620, 1089–1100 (2023).
we were able to differentiate successful versus 9. J. B. Ingraham et al., Nature 623, 1070–1078 (2023). SUPPLEMENTARY MATERIALS
10. J. Dauparas et al., Science 378, 49–56 (2022).
unsuccessful designs, suggesting that it should 11. R. Krishna et al., bioRxiv, 2023.10.09.561603 [Preprint] (2023).
science.org/doi/10.1126/science.adl5364
be helpful for prioritizing designs. Although we Materials and Methods
12. G. R. Lee et al., bioRxiv, 2023.11.01.565201 [Preprint]
Supplementary Text
ran simulations for 2 ms, their essential features (2023). http://biorxiv.org/lookup/doi/10.1101/2023.11.01.
Figs. S1 to S33
565201
could be gleaned after 100 ns, suggesting that Tables S1 to S5
13. L. An et al., bioRxiv, 2023.12.20.572602 [Preprint] (2023).
simulations on this timescale should be useful. References (37–93)
14. J. Dauparas et al., bioRxiv, 2023.12.22.573103 [Preprint] (2023).
MDAR Reproducibility Checklist
Free energy calculations have not previously 15. P. Gainza et al., Nat. Methods 17, 184–192 (2020).
Data S1
been applied to designed proteins. Although 16. M. V. Shapovalov, R. L. Dunbrack Jr., Structure 19, 844–858
(2011). Submitted 24 October 2023; resubmitted 23 December 2023
they are more computationally intensive and 17. J. Jumper et al., Nature 596, 583–589 (2021). Accepted 28 February 2024
require more user-specified parameters, we ob- 18. C. J. Lord, A. Ashworth, Science 355, 1152–1158 (2017). 10.1126/science.adl5364
N
forming reaction, originated in the 19th can mediate regioselective quaternization of C reoxidize Fe(II) to (III). Below, we identify
century as one of the earliest examples using ubiquitous chemical feedstocks: olefins conditions to reduce this design to practice
of C–N bond formation in organic syn- and carboxylic acids (Fig. 1B). An inexpensive and achieve a versatile C quaternization that
thesis (Fig. 1A) (1). Its inherent simplicity Fe-based catalyst in the presence of simple unites ubiquitous bench-stable starting mate-
and the widespread availability of the necessary reducing agents can effectively achieve this rials (olefins and acids) in a simple and easily
starting materials stands in stark contrast to powerful transformation in a chemoselective scalable fashion.
the way chemists go about forging quaternary fashion with broad substrate scope through a
C’s (2). Specifically, most C quaternizations similar mechanistic pathway. It is particularly Reaction development
entail polar transformations such as conjugate notable that a single catalyst can both activate The realization of this plan is summarized in
addition, enolate chemistry, and pericyclic re- distinct functional groups [redox active esters Fig. 1B using RAE 5 with either olefin 7 or
actions (3–5). Such processes are complicated (RAEs) and olefins] and mediate bond forma- RAE 8 leading to the same product 6. As il-
by the differing regioisomers that can result tion between them. An illustrative demonstra- lustrated by entries 1–8,15 (olefin-RAE coupling)
depending on the substrates and conditions tion of the power of such a reaction is depicted and 9–14 (RAE-RAE coupling), each reaction
used. For instance, conjugate additions can in Fig. 1C, in which the 10-step synthesis of a variant required simple and similar condi-
often lead to undesired 1,2-addition products, seemingly trivial structure (4) (10) can be trun- tions: the Fe(TPP)Cl complex, a base, a reduc-
enolate C alkylation often competes with O cated into a single step by coupling a primary tant and a 1,2-dichloroethane (DCE)/acetone
alkylation, and pericyclic processes can suffer tosylate-containing RAE 3 with either RAE solvent mixture. The choice of base proved
from regioisomeric mixtures depending on the 1 or olefin 2 with high chemoselectivity in crucial as exemplified in entries 2 and 3, with
innate sterics and electronics of the reactants. A 47% and 42% yield, respectively. CsOAc emerging as optimal. Increased yields
simple and predictable method to form quater- The mechanistic framework outlined in Fig. were observed after a simple degassing (bub-
nary C’s akin to classic N quaternization would 1B was inspired by key literature precedents bling Ar or N2) of the reaction mixture, although
be an appealing transformation. Our previous outlined in Fig. 2A. Seminal studies by Hirobe the presence of air was not detrimental (entry 4).
studies have demonstrated how radical meth- and co-workers showed that Drago-Mukaiyama- Exclusion of acetone led to lower yields, pre-
ods can be leveraged to cross-couple olefins type olefin hydration could be achieved using sumably due to catalyst insolubility (entry 5).
with other olefins or alkyl/aryl halides (6, 7). Fe porphyrins (11). Indeed, the pioneering Addition of other Fe sources such as Fe(acac)3
In many cases, quaternary C synthesis can be studies of Setsune et al. identified sec-alkyl iron led to diminished conversion (entry 6) as well.
achieved through such methods. The cross- porphyrins formed upon exposure of 1-alkenes As observed in other MHAT transformations,
coupling between two different redox active to iron(tetraphenylporphyrinato) chloride [Fe the use of preformed Ruben’s silane proved
esters has also been reported but is not gen- (TPP)Cl] and NaBH4 (12). Fe–H formation and beneficial (entries 7 and 8). For RAE-RAE cou-
erally applicable to quaternary C synthesis metal-hydride hydrogen atom transfer (MHAT) pling, the optimized conditions above only led
explain the branched-selective hydrometalation to 9% yield of the desired adduct 6 (entry 9).
(12). Brault and Neta depicted alkyl-Fe3+(TPP) Increasing the Fe loading to 20% had a mea-
1
Department of Chemistry, Scripps Research, La Jolla, CA complexes as mesomers with single-electron surable positive effect on conversion (entry 10).
92037, USA. 2College of Chemistry and Molecular density localized on C, which explained its fer- The greatest improvement emerged when
Engineering, Peking University, Beijing 100871, China.
3 rous phorphyrin-like behavior (13). Shenvi and switching the reductant from a silane to Zn
Oncology Medicinal Chemistry Department, Pfizer
Pharmaceuticals, San Diego, CA 92122, USA. 4Pfizer co-workers recently used this same iron com- metal (entry 11). A slight change in solvent
Medicine Design, Groton, CT 06340, USA. 5Biogen Inc., plex, along with a silane, to catalyze C–C bond composition to DCE/acetone 7:4 along with
Cambridge, MA 02142, USA. formation (14) between benzyl bromides and KOAc in place of CsOAc led to a further in-
*Corresponding author. Email: yukawama@scripps.edu (Y.K.);
pbaran@scripps.edu (P.S.B.); rshenvi@scripps.edu (R.S.) olefins to form quaternary centers through crease in yield (entries 12 and 13). The inclu-
†These authors contributed equally to this work. bimolecular homolytic substitution (SH2) reac- sion of a weak acid (Et3N•HCl) likely served
Fig. 1. C quaternization
overview and application.
(A) N quaternization
compared with methods
for C quaternization.
(B) Overview of method
reported here. (C) Simplifi-
cation imparted by this
cross-coupling strategy
compared with traditional
multistep synthesis. Fsp3,
fraction of sp3 C atoms;
PCC, pyridinium
chlorochromate.
to activate Zn(0) and resulted in a 61% isolated resultant radical: 1H NMR studies demonstrated likely fulfilled different roles: Silanes can re-
yield of 6 (entry 14). The final optimized con- that Fe(TPP) will complex n-pentyl-CO(NHPI) quire Lewis base addition to increase hydridic
ditions for RAE-RAE coupling could not be before the formation of n-pentyl–Fe(TPP) (see strength (19) and/or accelerate metal hydride
applied to the closely related olefin-RAE coupling the supplementary materials). Fe(TPP) can formation (20) en route to MHAT with alkenes,
(entry 15) and vice versa (entry 9). Neverthe- be formed from its corresponding hydride by and bases can presumably sequester Lewis
less, both conditions used a simple combina- either hydrogen evolution or alkene MHAT acidic metals, e.g., ZnCl2, that might accrue as
tion of commercially available reagents and (see the supplementary materials) (18). As a an RAE is reduced by Zn(0).
the same Fe(TPP)Cl catalyst. result, Fe(TPP)Cl mimics the polyfunctional-
Neither Fe-catalyzed decarboxylative cou- ity of precious metals in inner-sphere (coor- Substrate scope exploration
pling (conditions A or B) necessitated O2 to dinative) alkene functionalization catalysis, With a streamlined set of conditions in hand
turn over the catalytic cycle (Fe2+ to Fe3+), but appears to do so in an outer-sphere (non- for the radical quaternization of olefins and
whereas recent work from our group required coordinative) series of elementary steps. In the carboxylates, the scope was examined as illus-
an air atmosphere in conjunction with an Fe case of RAE-RAE coupling, the use of Zn metal trated in Fig. 3. In general, these highly chemo-
b-diketonate MHAT catalyst or co-catalyst is well known to reductively decompose NHPI selective conditions are compatible with a range
(15, 17). In the present work, the single iron(II) esters into C-centered radicals. of functional groups, including carbamates,
porphyrin turned over to iron(III) in the ab- The different optimal bases in conditions amides, alkyl halides, epoxides, aryl halides, es-
sence of oxygen by reaction with the RAE or its A and B were determined empirically and ters, alcohols, nitriles, tertiary amines, ketones,
ureas, and electron-rich or electron-deficient ternization despite their notorious reactivity untouched despite their common use in radical
heterocycles. It is particularly noteworthy that in both transition metal–catalyzed couplings cross-couplings. In the case of olefins, preferential
alkyl bromides, terminal alkynes as well as an and radical chemistry. Similarly, aryl iodides reactivity can be achieved during the RAE-RAE
alkyl boronate, remain intact during this qua- and electron-deficient aryl chlorides are left coupling, whereas less reactive olefins are
Fig. 3. Exploration of scope. Carboxylic acids (tan) and olefins (pink) substrates couple with quaternizing carboxylic acid reagents (gray). Complex fragment
couplings to form quaternary C’s under simple conditions are now possible (white).
spared when using the RAE-olefin coupling. lows an intuitive coupling from readily acces- neutral process (Fig. 2A). These features bode
Stabilized radicals derived from a-heteroatom sible heterocyclic fragments (10, 11, 12, 15, well for adoption in a parallel or combinatorial
or a-aryl carboxylic acids can also be coupled. and 16). The reactions described above are synthesis campaign in which ubiquitous, bench-
In the case of RAE-RAE couplings of such sub- extremely simple and practical to perform stable precursors can combine to deliver long-
strates, FeOEP led to a substantially increased using inexpensive reagents, include a rapid sought-after complex libraries rich in sp3
yield. Complex pharmaceutically relevant setup time, and prove amenable to gram-scale content, including quaternary C’s.
structures can be coupled efficiently, resulting preparation (see 41 and 34). The absence of
in quaternary C-containing analogs 36 to 40. oxygen in conditions A/B allow facile scal- Simplification of quaternary C synthesis
The direct installation of quaternary centers ing to gram quantities with no change in yield The capacity for these new reactions to
within saturated heterocycles is a particularly (Schlenk or glovebox conditions are not re- markedly simplify quaternary C–containing
challenging motif to construct by conventional quired, just a simple Ar balloon) and point to molecules has been exemplified through 10
polar bond analysis; this quaternization al- internal oxidative turnover, i.e., an overall redox different case studies (see the supplementary
Fig. 4. Examples of simplified syntheses. Lengthy routes to quaternary C–containing targets are compressed into short sequences using radical
quaternization.
materials for a complete listing), of which conventional routes to these mostly simple diates sidesteps these inefficient tactics
six examples are illustrated in Fig. 4. The structures rely on extensive redox manipu- through modular LEGO-like transformations
common theme among the prior syntheses lations, functional group interconversions, of commercially available building blocks
is a nearly exclusive reliance on polar bond and a variety of pyrophoric reagents for var- (olefins and acids).
analysis during the planning stages, espe- ious carbonyl reductions. By contrast, direct For example, the simple alkyne 53 was
cially carbonyl chemistry. Consequently, the C quaternization through radical interme- previously prepared (en route to prostaglandin
analog synthesis) in an eight-step route, of for the modulation of microglial activation, 17. A. L. Balch, R. L. Hart, L. Latos-Grazynski,
which only one step forged a C–C bond and required a 10-step route reliant on classic T. G. Traylor, J. Am. Chem. Soc. 112, 7382–7388
(1990).
the key quaternary C was purchased in the polar bond analysis (28). In a complete de- 18. S. W. M. Crossley, C. Obradors, R. M. Martinez, R. A. Shenvi,
form of a dimethyl cyclohexanone 52 with a parture from this strategy, commercially Chem. Rev. 116, 8912–9000 (2016).
skeleton that was tediously edited (21). By available Trolox (75) could be used to effici- 19. C. R. Howie, J. K. Lee, R. L. Schowen, J. Am. Chem. Soc. 95,
5286–5288 (1973).
contrast, commercially available olefin 54 ently incorporate the desired alkanol side
20. D. Kim, S. M. W. Rahaman, B. Q. Mercado, R. Poli,
was coupled with the RAE derived from the chain through radical quaternization (58% P. L. Holland, J. Am. Chem. Soc. 141, 7473–7485
simple commercial alkyne-containing acid yield) with RAE 72 to deliver analog 74 in (2019).
(55) under radical quaternization conditions only three steps. 21. L. Heslinga, R. van Der Linde, H. J. J. Pabon,
D. A. van Dorp, Recl. Trav. Chim. Pays Bas 94, 262–273
to deliver 53 in only two steps (48% isolated Quaternary C’s can now be dissected into (1975).
yield). Ketoaldehyde 56, a useful building block feedstock precursors, alkenes, and carboxylic 22. J. Lu et al., Synthesis (Stuttg.) 51, 3964–3972
for steroid analog synthesis, was previously acids through radical quaternization. Num- (2019).
accessed by classic carbonyl chemistry, C–C erous high Fsp3 materials are now available 23. J. Walker, J. K. Wood, J. Chem. Soc. Trans. 89, 598–604
(1906).
homologation, and a variety of undesirable with a fraction of the prior synthetic burden. 24. J. Sicher, M. Svoboda, J. Závada, Collect. Czech. Chem. Commun.
reagents (e.g., Br2, Mg, and OsO4) (22). Alter- Indeed, the approach outlined herein repre- 30, 421–437 (1965).
natively, the same structure was accessed sents a complete departure from the most- 25. J. Sicher, M. Svoboda, J. Závada, R. B. Turner, P. Goebel,
Tetrahedron 22, 659–671 (1966).
directly after radical quaternization of olefin ly carbonyl-focused retrosynthetic analysis
26. C. Blaszykowski et al., Org. Lett. 6, 3771–3774
57 with RAE 58, followed by acidic workup in (Table 2) historically used to forge such (2004).
45% isolated yield. Substrate 60, containing systems (3). In this simple protocol, two sub- 27. S. Proto et al., Org. Lett. 14, 3916–3919 (2012).
two quaternary centers, was previously accessed strate classes converge into the same catalytic 28. T. Muller, D. Coowar, M. Hanbali, P. Heuschling,
B. Luu, Tetrahedron 62, 12025–12040
through a laborious 11-step sequence with low cycle in which an Fe(III) catalyst undergoes (2006).
ideality commencing from glutaric anhydride reduction to Fe(II) by either Zn or a silane/ 29. S. H. Bertz, J. Am. Chem. Soc. 104, 5801–5803
59, again relying on polar bond analysis and alkene combination, and oxidative capture by (1982).
carbonyl chemistry (23–25). Diene 61 could be a redox-active ester returns the Fe(III). The 30. M. Tomanik, I. T. Hsu, S. B. Herzon, Angew. Chem.
Int. Ed. 60, 1116–1150 (2021).
enlisted in a sequential radical cross-coupling RAE cross-coupling exhibits perfect hetero- 31. J. R. Coombs, J. P. Morken, Angew. Chem. Int. Ed. 55,
first by using olefin-olefin Fe-catalyzed MHAT selectivity. By contrast, cross-couplings that 2636–2649 (2016).
cross-coupling with methylacrylate (72% yield), involve weak alkene-metal center complexes 32. L. T. Kliman, S. N. Mlynarski, G. E. Ferris, J. P. Morken,
Angew. Chem. Int. Ed. 51, 521–524 (2012).
followed by radical quaternization of 62 with to form sp3-sp3 linkages (e.g., Heck, Kulinkovich,
33. A. W. Hird, A. H. Hoveyda, J. Am. Chem. Soc. 127,
RAE 63 to deliver the same product, 60, in 1,4-addition) often do not allow chemoselec- 14988–14989 (2005).
58% isolated yield, thus obviating the need tive merger of complex fragments. It has been 34. T. S. Mei, H. H. Patel, M. S. Sigman, Nature 508, 340–344
for nine inconvenient steps and toxic and/or widely recognized that fragment coupling (2014).
35. G. Pratsch, G. L. Lackner, L. E. Overman, J. Org. Chem. 80,
pyrophoric reagents (e.g., LiAlH4, HBr, and simplifies retrosynthetic analysis by increas- 6025–6036 (2015).
KCN). Enyne 65, used as a substrate in a cyclo- ing convergency (29, 30). Simple tools such
isomerization study, was previously accessed as the reactions described herein point to a
through a 15-step sequence relying on alkyne distinctive conceptual platform (31–35) to AC KNOWLED GME NTS
hydrometallation, Wittig, carbonyl chem- decisively solve this vexing problem in or- We gratefully acknowledge J. S. Chen and the Scripps
istry, and acetylide addition as the key C–C ganic synthesis. Automated Synthesis facility (ASF) for help with analysis,
L. Pasternack for assistance with NMR spectroscopy,
bond-forming steps (26). This lengthy sequence and Y. Wang and J. Sun for helpful discussions and early
required the use of protecting groups and ex- contributions. Funding: This work was supported by the
tensive redox manipulations (three instances National Institutes of Health (grant GM122606 to R.A.S. and
RE FERENCES AND NOTES grant GM118176 to P.S.B.), the National Science Foundation
of LiAlH4 reduction). A more intuitive “LEGO”- (grant CHE1955922 to R.A.S.), Nanjing King-Pharm Co.,
1. N. Menschutkin, Z. Phys. Chem. 5, 589–600
like approach was enabled through radical (1890). Ltd. (X.G.), Pfizer (P.S.B. and Y.K.), and Biogen (P.S.B. and
retrosynthesis. Thus, an Ag-Ni-facilitated de- 2. T. T. Talele, J. Med. Chem. 63, 13291–13315 Y.K.). Author contributions: Conceptualization: X.G., B.Z.,
(2020). N.D., Y.K., P.S.B., R.S.; Funding acquisition: Y.K., P.S.B.,
carboxylative alkenylation of commercial acid R.S.; Investigation: X.G., B.Z., N.D., C.B., M.P., L.K., M.R.C.,
3. E. J. Corey, X.-M. Cheng, The Logic of Chemical Synthesis
66 with vinyl iodide 67, followed by radical (Wiley, 1995). C.C.T., P.N.B., M.N., Y.K.; Methodology: X.G., B.Z., N.D.,
quaternization with alkyne-containing RAE Y.K., P.S.B., R.S.; Project administration: Y.K., P.S.B., R.S.;
4. S. F. Martin, Tetrahedron 36, 419–460 (1980).
Supervision: Y.K., P.S.B., R.S.; Writing – original draft: N.D.,
68 afforded 65 in only five steps without any 5. S. Kotha, N. R. Panguluri, R. Ali, Eur. J. Org. Chem. 2017,
Y.K., P.S.B., R.S.; Writing – review and editing: N.D., Y.K.,
redox manipulations. Radical quaternization 5316–5342 (2017).
P.S.B., R.S. Competing interests: The authors declare no
6. S. Kotesova, R. A. Shenvi, Acc. Chem. Res. 56, 3089–3098
could also be used to construct scaffolds ap- (2023).
competing interests. Data and materials availability: All data
are available in the main text or the supplementary materials.
pearing in natural products. For example, 7. J. M. Smith, S. J. Harwood, P. S. Baran, Acc. Chem. Res. 51, License information: Copyright © 2024 the authors, some
piperidine 70, featuring a quaternary C at the 1807–1817 (2018). rights reserved; exclusive licensee American Association for the
C-3 position, serves as a pivotal scaffold with- 8. B. Zhang et al., Nature 606, 313–318 (2022). Advancement of Science. No claim to original US government
9. B. Zhang et al., Nature 623, 745–751 (2023). works. https://www.science.org/about/science-licenses-
in the madangamine alkaloids. In the previ- 10. F. Nagatsugi, F. Inoue, S. Sasaki, M. Maeda, Chem. Pharm. Bull. journal-article-reuse
ous seven-step synthesis, the construction of 43, 607–615 (1995).
the quaternary center again relied on carbonyl 11. T. Santa, T. Mori, M. Hirobe, Chem. Pharm. Bull. (Tokyo) 33,
2175–2178 (1985).
chemistry commencing from 2-piperidone (27). 12. J. Setsune, Y. Ishimaru, A. Sera, Chem. Lett. 21, 377–380
Of those seven steps, two forged C–C bonds (1992). SUPPLEMENTARY MATERIALS
with the remainder comprising concession 13. D. Brault, P. Neta, J. Am. Chem. Soc. 103, 2705–2710 science.org/doi/10.1126/science.adn5619
(1981). Materials and Methods
steps (redox, FGIs, and PG manipulations). Figs. S1 to S6
14. W. Liu, M. N. Lavagnino, C. A. Gould, J. Alcázar,
Conversely, starting from commercial acid D. W. C. MacMillan, Science 374, 1258–1263 NMR Spectra
71, a simple alkylation, followed by radical (2021). References (36–73)
quaternization with free alcohol containing 15. X. C. Gan et al., J. Am. Chem. Soc. 145, 15714–15720
(2023).
RAE 70, enables a three-step synthesis. Fi- 16. F. Toriyama et al., J. Am. Chem. Soc. 138, 11132–11135 Submitted 15 December 2023; accepted 28 February 2024
nally, a-tocopherol analog 74, which is useful (2016). 10.1126/science.adn5619
Removal Stalling
B C D E
100% 100%
10 minutes
15 minutes
10 kb 5% 5% 5% 5%
9% 10 kb
20% 25%
20% 33%
18% 3%
6%
66% 67%
57% 56%
Fig. 1. Replisomes push cohesin during DNA replication. (A) Cartoon replication fork. (C) Comparison of primary cohesin fate after collision with replication
showing DNA replication from a single origin on surface-tethered l DNA. forks in extracts. Two independent experiments are shown. (D) A representative
Replication is performed in X. laevis egg extracts while collisions between kymogram showing LD555-GINS collision with JF646-cohesin. Examples of cohesin
replisomes and cohesins are visualized. (B) Collisions between replication sliding and removal are marked using the symbols shown in (A). (E) Proportions of
forks, labeled with Fen1-mKikGR (red), and preloaded JF646-cohesin (magenta) are cohesin fates after collision by labeled replisomes. Two independent experiments are
visualized. An example showing cohesin sliding ahead of a Fen1-mKikGR–labeled shown. The n values in (C) and (E) indicate the total number of events analyzed.
PCNA has been implicated in cohesion estab- replisome onto the replicated sister DNAs? We origin firing experiments were performed with
lishment (23), we omitted Fen1-mKikGR to speculated that cohesin pushed ahead of the fluorescent CMG. As expected, replisomes were
exclude the possibility of inefficient cohesin replisome could generate cohesion when meet- disassembled shortly after fork convergence (26)
transfer resulting from improper PCNA re- ing a converging replisome. To determine the (fig. S9A), even when labeled cohesin persisted
tention. We visualized the replisome directly fate of cohesin during fork convergence, repli- on DNA (Fig. 2C; movie S4; and fig. S9, B to G).
using a method previously described (24). cation was started from multiple origins (fig. In 51 to 58% of cases, cohesin remained at the site
Purified fluorescent GINS was used to rescue S7A). When visualizing converging replication of replisome disassembly (Fig. 2D). Cohesin sig-
DNA replication in GINS-depleted extracts forks using Fen1-mKikGR, we observed that nal lifetime on DNA after replication termi-
(fig. S5). During replication of l DNA from most cohesin pushed ahead of replication nation varied, on average lasting for 43.7 min
single origins, fluorescent CMG moved at the forks remained in positions where converging (±17.8 min) (fig. S8, G to I). We conclude that
tip of Fen1-mKikGR tracts (24–26), at an av- replication forks met (Fig. 2, A and B; movie cohesin complexes are pushed by advancing
erage speed consistent with previous work S3; and fig. S7, B to D). In some cases, fork replisomes to sites of fork convergence and re-
(426 base pairs/minute; fig. S6, A to C). We convergence was accompanied by either cohe- main at these sites after replisome disassembly.
next visualized the outcomes of collisions be- sin eviction or further cohesin movement after The key question therefore is: Do the cohesin
tween fluorescent replisomes and preloaded fork convergence (Fig. 2C and fig. S7, E to G). rings that persist on DNA after replication ter-
JF646-cohesin (Fig. 1D; movie S2; and fig. S6, Cohesin remaining on DNA after fork conver- mination in our assay mediate cohesion?
D to G). Under these conditions, cohesin trans- gence was resistant to a high-salt wash that
fer was still very rare (5% of events; Fig. 1E), and removed Fen1 from DNAs (Fig. 2A and fig. S7, Cohesin complexes retained at replication
cohesin sliding ahead of the replisome domi- B to D), suggesting that this cohesin popula- termination sites can tether sister DNAs
nated (56 to 67% of events). Therefore, we tion was topologically bound to DNA (27). On To assess whether cohesin molecules at repli-
concluded that the low frequency of cohesin fully replicated DNAs, we observed multistep cation termination sites provide cohesion, we
transfer observed in our previous experiments photobleaching of cohesin in diffraction-limited developed an assay to measure sister DNA co-
was not caused by Fen1-mKikGR. spots (mean = 1.90 steps; fig. S8, A to F), sug- hesion. Previous experiments tethering linear
gesting that multiple cohesins are pushed to DNAs to the surface only through 3′ biotins
Cohesins relocalize to DNA replication sites of replication fork convergence. have shown that the new sister DNA that does
termination sites We envisaged that upon fork convergence, not contain biotin collapses from the surface
How does conversion generate cohesion if pre- replisomes disassemble while cohesin traps both after the replisome reaches the 5′ end (fig.
loaded cohesin is not transferred behind the daughter strands together. To test this, multiple- S10A) (24). Using this knowledge, we designed
A B 100% C D 100%
15 minutes
10 minutes
10 kb 13%
21% 22%
32%
21%
27%
5%
87%
63% 58%
51%
Fig. 2. Cohesin is pushed to positions of DNA replication termination. replication fork was measured. Two independent experiments are shown. (C) Kymogram
(A) Kymogram showing replication forks colliding with JF646-cohesin complexes under example showing replisome (LD555-GINS) progression on DNA from multiple origins and
conditions of high origin firing. After a period of replication, a high-salt wash (HSW) was colliding with JF646-cohesin. (D) Quantification of JF646-cohesin fate at sites where
performed, and the same DNA was imaged. (B) Quantification of cohesin fates at converging replisomes (LD555-GINS) are removed, with two independent experiments
converging replication forks. The fate of cohesin that was pushed to a converging shown. The n values in (B) and (D) indicate the total number of events analyzed.
a DNA template to visualize the interaction both ends (fig. S10E). When both new sister that cohesion establishment by cohesin con-
between sister DNAs after replication (Fig. 3A). DNAs collapsed, they colocalized together for version occurs during replication termination.
Binding of Alexa Fluor 488–labeled LacI (LacI- varying lengths of time before separating, giving Our model also predicts that cohesive cohesin
AF488, fig. S10B) to 48 lacO repeats at each DNA a measure of cohesion. The critical question is complexes should be dragged by the collapsing
end (fig. S10C) blocked replisome progres- whether colocalization of sister DNAs resulted sister DNAs and remain associated with both
sion (Fig. 3A). Subsequent removal of LacI- from cohesin-mediated cohesion. sister DNAs (fig. S13A, right panel, top sce-
AF488 by isopropyl-b-D-thiogalactopyranoside To test whether cohesin complexes physically nario). If cohesin does not tether sister DNAs
(IPTG) addition resulted in synchronous com- tethered collapsing sister DNAs in the experi- together, we would expect half of the cohesin
pletion of replication and collapse of both ments described above, the assay was performed molecules to remain associated with stretched
sister DNAs (Fig. 3A). This setup enabled us in either mock- or cohesin-depleted extracts (fig. DNA after sister DNA collapse (fig. S13A, right
to measure cohesion between the replicated S11A), and we compared the length of time that panel, bottom scenario). To test this, sister DNA
sister DNAs. Lack of cohesion between the collapsed sister DNAs colocalized (Fig. 3, B and C; collapse experiments were performed with
sister DNAs would result in sister DNAs im- movie S5; and fig. S11B). The collapsed sister JF549-cohesin preloaded onto parental DNA
mediately separating (Fig. 3A, right panel, top DNAs remained associated reproducibly longer and imaged simultaneously with AF647-dUTP.
scenario). If, however, the replicated DNAs in mock-depleted extracts than in cohesin-depleted Cohesin was observed to move when both sister
were held together, the two collapsed sister extracts (Fig. 3C). When purified cohesin was DNAs collapsed (Fig. 3D, movie S6, and fig.
DNAs would colocalize (Fig. 3A, right panel, preloaded onto DNAs before replication in S13B) and when a single sister DNA collapsed
bottom scenario). cohesin-depleted extracts, collapsed sister DNAs (fig. S13C). Between 73 and 85% of cohesin mol-
We initiated replication from multiple ori- remained together for periods of time com- ecules moved and colocalized with collapsed
gins in the presence of LacI-AF488, and Alexa parable to that in mock-depleted extracts (Fig. sister DNAs (Fig. 3E), whereas only 15 to 27%
Fluor 647–deoxyuridine triphosphate (AF647- 3C). In a bulk replication assay, formation of remained on stretched DNAs (fig. S13D). We
dUTP) was incorporated into nascent DNA for replicated supercoiled plasmid was indistin- estimate that 50 to 70% of cohesins retained
visualization. Excess LacI-AF488 and AF647- guishable between mock- and cohesin-depleted on DNA after replication termination bound
dUTP were washed away, and replication ex- extracts (fig. S12, A to C), eliminating the pos- both sister DNAs (see methods in the supple-
tract containing IPTG was added to release sibility that tethering between sister DNAs may mentary materials). The cohesin colocaliza-
LacI-AF488 from DNA ends. Under these con- be due to cohesin delaying replication com- tion with collapsed sister DNAs reaffirms that
ditions, regions of AF647-labeled replicated pletion at converging forks. Furthermore, most cohesion establishment by preloaded cohesin
DNA could be visualized during synchronous colocalizing sister DNAs were disrupted complexes happens during replication termi-
collapse of new sister DNAs from DNA ends. when treated with 0.1% SDS (fig. S12, E and nation. A crucial event during replication ter-
The assay is intrinsically validated by mole- F), indicating that sister DNAs are coupled mination is the removal of the CMG helicase by
cules with partially replicated DNA, where through DNA–protein interactions rather than the adenosine triphosphatase p97 (28). We next
the labeled nascent DNA approaches only one DNA–DNA interactions. These results show asked whether a timely replisome disassembly
DNA end. In these instances, after removal of that cohesin complexes preloaded onto paren- during termination could be required for proper
LacI, the collapsed sister DNA moved with the tal DNA physically tether sister DNAs after cohesion establishment.
replisome toward the opposite end of the DNA replication. Taken together with our previous
(fig. S10D), as observed previously (24). On mol- observation that preloaded cohesin is predomi- Cohesion establishment during replication termination
ecules where the DNA template was fully rep- nantly pushed to sites of replication termina- To assess whether CMG disassembly during
licated up to the LacI barrier at both ends, LacI tion and remains at these sites after replisome replication termination is important for proper
removal caused the collapse of sister DNAs from disassembly, our data provide strong evidence cohesion, we inhibited CMG disassembly in
A B LacI-AF488
LacO repeats
5'
10 minutes
10 minutes
3' t
Replication extract
LacI-AF488
AF647-dUTP
Separation
t
Colocalization
Replication extract
IPTG
26 kb
AF647-dUTP
C D LacI-AF488 E
p>0.01 100%
10 minutes
p<10-4
p<10-4 Cohesin moving with +IPTG
60
both collapsed sister
DNAs
50
85% 73%
Time (minutes)
40
30
27%
15%
20
Exp. 1 Exp. 2
n = 66 n = 93
10
Cohesin colocalizing
0 with collapsed DNA
Mock Pre-loaded Cohesin remaining on
-
Cohesin AF647-dUTP JF549-Cohesin stretched DNA
Merge
Cohesin
Fig. 3. Cohesin holds newly replicated surface-tethered sister DNAs mock-depleted (n = 127), cohesin-depleted (n = 106), and rescue (n = 110) extracts.
together. (A) Diagram showing sister DNA collapse experiments. Replication Data are shown mean value with 95% confidence intervals. Three independent
forks are paused at DNA ends by LacI. Upon IPTG addition, replication forks experiments were performed. P values were calculated using a two-tailed Mann-
reach DNA ends, and collapsing sister DNAs are visualized. (B) Example Whitney U test. (D) Example kymogram showing JF549-cohesin associating with
kymograms where both sister DNA strands collapse (see cartoon). Both examples both collapsed sister DNAs. The red arrow indicates cohesin at the end of DNA tethers,
are from mock depletion with different sister DNA separation times. The time that which is excluded from the analysis. (E) Quantification of cohesin position after sister
collapsed DNA strands colocalized before separating is indicated. (C) Individual data DNA collapse of either one strand or both strands, with data from two independent
points showing the time that collapsed sister DNAs colocalized before separation for experiments shown. The n values indicate the total number of events analyzed.
Saccharomyces cerevisiae and measured co- To determine the fate of cohesin at replication vergence (Fig. 4C and fig. S15D) and cohesin
hesion using the well-established dot assay (13). termination sites when CMG disassembly is persisting between bypassing replisomes
Depletion of the p97 homolog Cdc48 eliminates blocked in egg extracts, we repeated single- were rare (Fig. 4C and fig. S15E). Because
CMG disassembly without affecting DNA rep- molecule assays with labeled CMG and cohesin cohesin demonstrated stability on DNA when
lication in yeast (fig. S14, A to C) (29). Cdc48 in the presence of a p97 inhibitor (p97i) (30). replisome unloading was inhibited, we ex-
depletion through the auxin-inducible degron In p97i-supplemented extracts, replisomes re- plored whether cohesin could contribute to
(AID) system before the initiation of DNA mained on DNA and occasionally bypassed cohesion in the presence of p97i. Repeating
replication resulted in a significant cohesion each other after fork convergence, as seen sister DNA collapse assays (fig. S15, F and G)
defect, as shown by the appearance of two previously (26). In instances where cohesin and comparing the time at which collapsed
separate URA-GFP dots in large-budded cells preceded one or both replisomes before sister DNAs colocalized revealed no signifi-
with high securin levels (Fig. 4A), while Cdc48 fork convergence, most replisomes by- cant difference in the presence of p97i (fig.
depletion in post-replicative cells did not af- passed each other, with cohesin colocalizing S15H). These results indicate that, in our
fect maintenance of preestablished cohesion with one of the replisomes (Fig. 4, B and C; single-molecule assay, inhibiting replisome
(fig. S14D). This result suggests that CMGs movie S7; and fig. S15, A and B). In some cases, disassembly allows remaining cohesins to
must be removed from DNA after replication replisomes stalled, failing to bypass each other, still provide cohesion between sister DNAs.
termination for proper cohesion establish- resulting in cohesin retention (Fig. 4C and fig. However, given that our in vivo data under-
ment in vivo. S15C). Cohesin removal during replisome con- score the importance of CMG disassembly
A B C D
+p97i
15 minutes
10 kb 7%
22%
4%
67%
?
n=27
Cohesin removal
Fig. 4. Cohesin dynamics during replication termination when replisome were labeled directly using LD555-GINS, and replisome disassembly upon fork
disassembly is inhibited. (A) Cohesion of the URA3 locus was measured convergence was inhibited with 200 mM p97i (NMS-873). In this example,
in wild-type and Cdc48-AID yeast cells that were synchronized in G1 and released converging replisomes bypass one another, and one replisome continues pushing
into G2 arrest in the presence of 5 mM indole-3-acetic acid (IAA). Cohesion a labeled cohesin (white square). (C) Quantification of cohesin and replisome
was also measured in asynchronously growing wild-type, dia2D, and dia2-13A fates observed during replication termination when replisome disassembly
mutant cells. For each experimental condition, at least 100 mitotic cells is inhibited. The n value indicates the total number of events analyzed. (D) Model
were scored for one or two GFP dots, and each experiment was repeated three of cohesion establishment during replication termination. Possible scenarios
times. The graphs show the mean and standard deviation. (B) Replisomes describing termination-coupled cohesion establishment are shown in fig. S16.
for cohesion, it is likely that the single- one replication origin. Consequently, owing to Discussion
molecule sister DNA collapse assay may the presence of two replication origins, rep- In contrast to prevailing models, which place
not be sensitive enough to measure mod- lication termination and subsequent CMG dis- preloaded cohesin behind replication forks, we
erate levels of cohesion defects. assembly exclusively occur in the URA+ and find that cohesin rings are pushed along the
To further investigate the interplay between not in the MET+ minichromosome. We noted DNA by the advancing replisome. Cohesin rings
cohesion establishment and replisome disas- that wild-type cells retained the two-origin pushed to positions of fork convergence are
sembly in vivo, we explored the role of Dia2, URA+ minichromosome more efficiently than retained on replicated DNA even after replisome
the ubiquitin ligase responsible for ubiquity- the single-origin MET+ counterpart (fig. S14E). disassembly and are capable of tethering sister
lating Mcm7 in yeast and marking replisomes The less-efficient retention of the MET+ mini- DNAs together. The notion of replisomes push-
for removal by Cdc48 (29). Deleting the DIA2 chromosome could either stem from potential ing cohesin is supported by transcription re-
gene as well as mutating it to prevent Dia2’s cohesion defects due to the absence of repli- positioning cohesin on yeast chromosomes (14)
association with the replisome (31) resulted in cation termination events or to less-efficient and cohesin being pushed ahead of T7 RNA
significant cohesion defects (Fig. 4A), further replication from its single origin compared polymerase and DNA translocase FtsK in vitro
supporting the notion that defective CMG dis- with the dual-origin URA+ minichromosome. (32, 33). A previous single-molecule study re-
assembly adversely affects cohesion. Although To test these scenarios, we assessed the reten- ported that a substantial fraction of cohesin was
these findings demonstrate Dia2’s role in ro- tion efficiency of minichromosomes in dia2D incorporated into replicated DNA in Xenopus
bust cohesion, they do not address whether cells. We reasoned that if Dia2-mediated CMG egg extracts (34). We suggest that the cohesin
the cohesion defects in dia2 mutants result removal during replication termination is es- transfer events observed therein primarily re-
from perturbations in CMG disassembly dur- sential for cohesion, dia2 deletion would se- sulted from cohesin being incorporated dur-
ing replication termination. lectively impair retention of the URA+, but not ing replication termination, as the study used
To ascertain the specificity of Dia2’s role in the MET+, minichromosome. Consistent with firing from multiple origins. In our system,
cohesion at replication termination, we engi- our hypothesis, the dia2D mutant displayed conversion of preloaded cohesin to cohesive
neered a yeast strain harboring two small comparable efficiency to wild-type cells in re- structures through bona fide cohesin transfer
linear chromosomes of similar lengths but dif- taining the MET+ minichromosome but ex- behind the replication fork might occur, albeit
fering in the number of replication origins and hibited significant defects in maintaining the rarely. Removal and stalling events might re-
selection markers (fig. S14E). The URA+ mini- URA+ minichromosome (fig. S14E). Together, flect differences in cohesin binding onto DNA
chromosome, which enables cells to grow in the in vivo data reinforce single-molecule results before replication. We envisage that cohesin re-
absence of uracil in the growth medium (uracil and support the idea that cohesion establish- moval helps prevent the accumulation of many
auxotrophy), has two high-efficiency replication ment is intricately connected to DNA replica- cohesins sliding ahead of the fork, which could
origins, whereas the MET+ minichromosome, tion termination and subsequent CMG complex impede replisome interaction with other DNA-
conferring methionine auxotrophy, has only disassembly. bound proteins, such as histones.
We envisage that CMG complexes that are 30. J. M. Dewar, E. Low, M. Mann, M. Räschle, J. C. Walter, minichromosomes: J.B. All other experiments: G.C. Supervision:
retained on DNA after fork convergence may Genes Dev. 31, 275–290 (2017). M.S. and H.Y. Writing – original draft: G.C. and H.Y. Writing – review &
31. M. Jenkyn-Bedford et al., Nature 600, 743–747 (2021). editing: G.C., D.T.G., Ç.K., K.A.N., M.S., and H.Y. Competing
continue pushing cohesin complexes along 32. J. Stigler, G. Ö. Çamdere, D. E. Koshland, E. C. Greene, interests: The authors declare that they have no competing
DNA, as seen in our single-molecule assays, Cell Rep. 15, 988–998 (2016). interests. Data and materials availability: All data needed to
and eventually evict them from chromatin in 33. I. F. Davidson et al., EMBO J. 35, 2671–2685 (2016). evaluate the conclusions in the paper are present in the paper
34. M. Kanke, E. Tahara, P. J. Huis In’t Veld, T. Nishiyama, EMBO J. and/or the supplementary materials. Raw data are available
cells, resulting in cohesion impairment. We 35, 2686–2698 (2016). in Figshare (35). All materials produced in this study, such as
propose several potential models for cohesion 35. G. Cameron et al., Raw data for: Sister chromatid cohesion plasmids and cell lines, can be provided upon request. License
establishment at termination sites (Fig. 4D establishment during DNA replication termination, Figshare information: Copyright © 2024 the authors, some rights
(2024); https://doi.org/10.25418/crick.25062113. reserved; exclusive licensee American Association for the
and fig. S16). Converging replisomes could Advancement of Science. No claim to original US government works.
pull the final stretches of unreplicated DNA ACKN OWLED GMEN TS https://www.science.org/about/science-licenses-journal-article-
through cohesin rings (fig. S16A). This would We thank J. Diffley for permission to use the yeast strain reuse. This research was funded in whole or in part by the
containing linear minichromosomes, S. Yardimci for helping with Wellcome Trust (CC2133, 223235/Z/21/Z, 226494/Z/22/Z), a
result in replisome disassembly and cohesion
extract preparations, and A. Kaushik for technical assistance. We cOAlition S organization. The author will make the Author
between new sister DNAs. This model negates thank the Francis Crick Institute Aquatics Facility for X. laevis Accepted Manuscript (AAM) version available under a CC BY
the requirement for replisomes to pass through husbandry and egg collection and Chemical Biology Facility for public copyright license.
cohesin rings or for transient ring opening. Two synthesis of fluorescent peptides for GINS labeling. Funding:
This work was supported by the Francis Crick Institute, which
alternative possibilities are that (i) cohesin is SUPPLEMENTARY MATERIALS
receives core funding from Cancer Research UK, the UK Medical
transferred behind one replication fork during Research Council, and the Wellcome Trust grant CC2133 (G.C., science.org/doi/10.1126/science.adf0224
termination (fig. S16B) or that (ii) a terminat- D.T.G., S.X., Ç.K., and H.Y.); a Boehringer Ingelheim Fonds PhD Materials and Methods
Fellowship (G.C.); a Sir Henry Dale Fellowship jointly funded by the Figs. S1 to S16
ing replisome bypasses the cohesin ring (fig. Table S1
Wellcome Trust and the Royal Society grant 223235/Z/21/Z
S16C). Bypass of the cohesin ring could use a (D.T.G.); the Wellcome Trust grant 226494/Z/22/Z (M.S.); and References (36–42)
mechanism similar to bypass of DNA–protein Cancer Research UK grant 26747 (K.A.N.). Author contributions: MDAR Reproducibility Checklist
Purified Xenopus cohesin complexes and performed cohesin Movies S1 to S7
cross-links by CMG helicase (24). Factors re-
loading in bulk: M.S. Experiments with Fen1-mKikGR comparing
quired for cohesin conversion (13) could aid undepleted and cohesin-depleted extracts: D.T.G. Analysis of Submitted 27 September 2022; resubmitted 16 June 2023
pushing of cohesin by the replisome, facilitate single-molecule data: G.C., D.T.G., and Ç.K. Xenopus GINS cloning Accepted 27 February 2024
CMG removal, or be involved in the molecular and expression: S.X. Experiments with S. cerevisiae and data Published online 14 March 2024
analysis: R.G. Construction of the yeast strain containing 10.1126/science.adf0224
transactions between DNA, replisomes, and
cohesin during replication termination.
RE FE RENCES AND N OT ES
PLANT SCIENCE
1. S. Yatskevich, J. Rhodes, K. Nasmyth, Annu. Rev. Genet. 53,
445–482 (2019).
2. I. F. Davidson et al., Science 366, 1338–1345 (2019).
HD-Zip proteins modify floral structures
3. Y. Kim, Z. Shi, H. Zhang, I. J. Finkelstein, H. Yu, Science 366,
1345–1349 (2019).
for self-pollination in tomato
4. R. Ciosk et al., Mol. Cell 5, 243–254 (2000).
5. T. S. Takahashi, P. Yiu, M. F. Chou, S. Gygi, J. C. Walter, Minliang Wu1†, Xinxin Bian1†, Benben Huang1†, Yadi Du1,2, Shourong Hu1, Yanli Wang1,2,
Nat. Cell Biol. 6, 991–996 (2004).
6. G. Zheng, M. Kanchwala, C. Xing, H. Yu, eLife 7, e33920 (2018).
Jingyuan Shen1, Shuang Wu1*
7. S. P. Bell, K. Labib, Genetics 203, 1027–1067 (2016).
8. J. T. P. Yeeles, A. Janska, A. Early, J. F. X. Diffley, Mol. Cell 65, Cleistogamy is a type of self-pollination that relies on the formation of a stigma-enclosing floral
105–116 (2017).
structure. We identify three homeodomain-leucine zipper IV (HD-Zip IV) genes that coordinately promote
9. A. Gambus et al., Nat. Cell Biol. 8, 358–366 (2006).
10. M. Srinivasan et al., Cell 173, 1508–1519.e18 (2018). the formation of interlocking trichomes at the anther margin to unite neighboring anthers, generating
11. F. Uhlmann, K. Nasmyth, Curr. Biol. 8, 1095–1101 a closed anther cone and cleistogamy (flower morphology necessitating strict self-pollination). These
(1998).
HD-Zip IV genes also control style length by regulating the transition from cell division to endoreduplication.
12. H. Xu, C. Boone, G. W. Brown, Genetics 176, 1417–1429 (2007).
13. M. Srinivasan, M. Fumasoni, N. J. Petela, A. Murray, The expression of these HD-Zip IV genes and their downstream gene, Style 2.1, was sequentially
K. A. Nasmyth, eLife 9, e56611 (2020). modified to shape the cleistogamy morphology during tomato evolution and domestication. Our results
14. A. Lengronne et al., Nature 430, 573–578 (2004). provide insights into the molecular basis of cleistogamy in modern tomato and suggest targets for
15. B. Hu et al., Nucleic Acids Res. 43, e132 (2015).
16. J. Walter, L. Sun, J. Newport, Mol. Cell 1, 519–529 (1998). improving fruit set and preventing pollen contamination in genetically modified crops.
17. A. B. Loveland, S. Habuchi, J. C. Walter, A. M. van Oijen,
Nat. Methods 9, 987–992 (2012).
18. H. Yardimci, A. B. Loveland, A. M. van Oijen, J. C. Walter,
C
leistogamy favors genetic stabilization was thought to have been achieved by the
Methods 57, 179–186 (2012).
19. W. S. Hoogenboom, D. Klein Douwel, P. Knipscheer, Dev. Biol. of desired agronomic traits as well as a elimination of self-incompatibility (SI) (7–9)
428, 300–309 (2017). higher fruiting rate (1–3). As a result, va- and stigma shortening (mutations in the
20. A. Losada, M. Hirano, T. Hirano, Genes Dev. 12, 1986–1997 rieties with cleistogamy were often re- Style2.1 and SE3.1 genes) (10, 11). However,
(1998).
21. H. Yardimci, A. B. Loveland, S. Habuchi, A. M. van Oijen,
tained during crop domestication (4, 5). as shown in other Solanaceae species such as
J. C. Walter, Mol. Cell 40, 834–840 (2010). Cultivated tomatoes display complete cleis- peppers and eggplants, stigma-shortening and
22. D. T. Gruszka, S. Xie, H. Kimura, H. Yardimci, Sci. Adv. 6, togamy and thus have considerably higher self-incompatibility release alone are not suffi-
eabc0330 (2020).
23. H. W. Liu et al., Mol. Cell 78, 725–738.e4 (2020).
fruit set, compared with wild species that re- cient for the establishment of cleistogamy
24. J. L. Sparks et al., Cell 176, 167–181.e21 (2019). quire cross-pollination (6). The transition from (12–14). Tomato flowers develop a morpho-
25. K. B. Vrtis et al., Mol. Cell 81, 1309–1318.e6 (2021). anisogamy to complete cleistogamy in tomato logical feature known as the “anther cone”,
26. E. Low, G. Chistol, M. S. Zaher, O. V. Kochenova, J. C. Walter,
Genes Dev. 34, 1534–1545 (2020).
in which five anthers join together to form a
27. M. Minamino, T. L. Higashi, C. Bouchoux, F. Uhlmann, Life Sci. 1
College of Horticulture, Haixia Institute of Science and hollow tube-like cone around the style. In this
Alliance 1, e201800143 (2018). Technology, Fujian Agriculture and Forestry University, structure, the stigma is surrounded and sealed
28. J. M. Dewar, J. C. Walter, Nat. Rev. Mol. Cell Biol. 18, 507–516 Fuzhou 350002, China. 2College of Life Sciences, Fujian by the joined anthers (15). As tomato flowers are
(2017). Agriculture and Forestry University, Fuzhou 350002, China.
29. M. Maric, T. Maculins, G. De Piccoli, K. Labib, Science 346, *Corresponding author. Email: wus@fafu.edu.cn pendant, released pollen grains fall onto the
1253596 (2014). †These authors contributed equally to this work. stigma, ensuring self-pollination. The anther
cone is most likely required for cleistogamy disruption of Wo function in woW106R (26), the in tomato (25), we found that Wo was able to
because only tomato varieties with this mor- CRISPR/Cas9–generated mutant of Wo (cr-wo, interact with several similar HD-Zip IV pro-
phological trait reproduce by cleistogamy (15). a loss of function mutation) had no obvious teins, including HD7L and HD7 (table S1). Using
However, it remains unknown how anther fu- phenotype in the style and anther cone (Fig. Co-immunoprecipitation (Co-IP) and bimolec-
sion is achieved to form the anther cone. 2E and fig. S2B). Consistent with woW106R, the ular fluorescence complementation (BiFC) assays,
HD-Zip IV family proteins are transcriptional style and interlocking trichome phenotypes we demonstrated that the woW106R mutant-form
regulators that are often involved in epidermal of the CRISPR/CAS9 mutant of woW106R (cr- protein can also interact with these proteins
cell development in plants (16–19). In Arabidopsis, woW106R) were rescued (Fig. 2E and fig. S2, B (Fig. 2F and fig. S2H).
epidermal differentiation is regulated by two to G). We thus speculate that the W106R mu- Using the promoter of MX1 (MIXTA-like 1)—
HD-Zip IV genes, ARABIDOPSIS THALIANA tation generates a dominant negative effect on a previously identified downstream target
MERISTEM LAYER1 (AtML1) and PROTODERMAL Wo function. of Wo (25)—as a reporter, we found that the
FACTOR2 (PDF2) (20–24). In tomato, Woolly Dominant negative effects generally oc- woW106R protein not only lost its ability to ac-
(Wo), a homolog of AtML1 and PDF2, con- cur where protein regulation is mediated by tivate MX1 expression but also inhibited the
trols the initiation and differentiation of the multimers (28). In Immunoprecipitation-Mass activation ability of several other HD-Zip IV
trichomes on the surface of leaves and stems, Spectrometry (IP-MS) analysis of the Wo protein proteins, including wild-type Wo, as well as
acting as a part of the defense against her-
bivores (25). Here we show that three HD-Zip
IV members including Wo, HD7, and HD7L
coordinate the development of interlocking
trichomes and the style to form a stigma-
enclosing anther cone, leading to cleistogamy
in cultivated tomato.
HD7L and HD7 (Fig. 2G and fig. S2I). The HD7L had the highest sequence similarity to and HD7L play a dominant role in anther cone
W106R mutation disrupts the ability of Wo Wo proteins (fig. S3C). Consistent with this, formation.
to bind the L1-box element of downstream luciferase reporter analysis showed that MTR1
genes (26). We thus hypothesized that the also inhibited the transcriptional function of HD7, HD7L, and Wo coordinately regulate
woW106R protein forms dimers with other HD7 and HD7L (fig. S3, D to E). Therefore, we reproductive development
HD-Zip IV proteins, thereby interfering with speculate that HD7 and HD7L play a dominant To further test the function of HD7 and HD7L,
their function of binding downstream targets role in the regulation of interlocking trichomes we generated loss-of-function mutants using
(Fig. 2H). of the anther cone. CRISPR/Cas9 (fig. S4, A and B). Single mutants
We constructed a promoter-driven GUS re- of Wo, HD7, or HD7L had no phenotype in the
Repression of HD-Zip IV genes splits porter for Wo and its homologs HD7 and HD7L. anther cone or style development (fig. S4, C
the anther cone Histological staining showed that all three to H). In line with our hypothesis, cr-hd7/hd7l
We found that the overexpression of MTR1 HD-Zip IV genes were specifically expressed double mutants had almost no anther trichomes;
(OE-MTR1), a repressor of the HD-Zip IV genes, in the upper region of the style, whereas Wo we also observed that the anther cones split
resulted in a shortened style, similar to those expression was higher than that of HD7 and apart (fig. S5, A to D). By means of SEM we
in woW106R (fig. S3A). In addition, we observed a HD7L (Fig. 3A). In support of their function found that the initiation of anther trichomes
disrupted anther cone in OE-MTR1, which was in anthers, the three HD-Zip IV genes were in cr-hd7/hd7l was less than that in the WT at
not seen in woW106R (fig. S3B and Fig. 2A). One highly expressed at the inter-anther junction, 10 days before anthesis (Fig. 3, C to D). Con-
possible explanation for this phenotype is that with the expression of HD7 and HD7L higher than sistently, the nuclear size in mature anther
MTR1 overexpression simultaneously represses that of Wo (Fig. 3B and fig. S3F). The spatio- trichomes was considerably reduced (Fig. 3,
Wo and its functionally redundant homologs. temporal expression pattern suggests that Wo E to F). cr-hd7/hd7l double mutants had no
Phylogenetic analysis revealed that HD7 and may function mainly in the style whereas HD7 obvious style shortening phenotype but style
Fig. 2. Dominant negative mutation of Wo inhibits interlocking trichomes Wo, HD7, and HD7L proteins in Co-IP. (G) The woW106R protein inhibits the ability of
and style length. (A) Interlocking trichomes are moderately affected in the the interacting HD-Zip IV proteins to activate downstream gene MX1 in Luciferase
woW106R mutant. (B) Mutation site and style phenotype of woW106R mutant. (LUC) reporter assay. Data are mean ± SD (n = 3 biological replicates) (**P < 0.01,
(C) Phenotype of style epidermal cells and (D) quantification of cell size. Note ***P < 0.001). (H) Working model of HD-Zip IV and interference by woW106R. Wo
that cell size is smaller but cell number is considerably larger in woW106R. Data are forms dimers with other HD-Zip IV proteins to bind to DNA through the HD domain
mean ± SD (n > 8) (****P < 0.0001). (E) Style phenotype in different backgrounds. during the activation of downstream genes. The dimers formed with woW106R
Note that Wo knockout results in no style shortening phenotype and knockout protein disrupt this DNA binding ability, resulting in inactivation of downstream
of woW106R mutant allele partially rescues style phenotype. (F) woW106R interacts with genes. Unpaired t tests were used for all statistical analyses.
shortening was most evident in cr-hd7/wo Cells in the upper region of the style (near the E). Using the H2B marker, we found that the
double mutants (Fig. 3G). Together with the stigma) were all smaller whereas the cells in transition from cell division to endoredupli-
specific tissue expression patterns of HD7, Wo, the middle and lower regions were larger (fig. cation was blocked in the cr-hd7/wo double
and HD7L, we propose that these three HD- S6A). Using a histone (H2B-GFP) marker, we mutant, which showed active cell division
Zip IV proteins jointly regulate anther cone found that cell division occurred specifically in throughout the whole style at both early and
formation and coordinated style growth in the upper region of the style. Away from the late stages of development (Fig. 4F). This ulti-
tomato (Fig. 3, G to I), albeit with different division region, the style cells became progres- mately resulted in the style cells having smaller
levels of involvement (Fig. 3, A and B). sively endoreduplicated resulting in a gradual nuclei in the cr-hd7/wo double mutant com-
In addition to the functional indispensabil- increase in cell size in the middle and bottom pared with the WT (Fig. 4G). The three HD-Zip
ity of interlocking trichomes for anther cones, parts. The polar distribution of division, elon- IV genes are likely to jointly promote the tran-
we found that the misexpression of the domi- gation, and maturation zones in the style is sition from cell division to endoreduplication
nant allele of Wo (WoV) in the petal caused similar to that of plant root development (Fig. (Fig. 4H) though the style phenotype varied in
dispersed trichomes to become interlocked, 4A and fig. S6B). In the mature zone of the style, their respective mutants. The reduction in endo-
resulting in a closed petal structure (fig. S5, E nuclei were highly polyploid with many reach- reduplication was most severe in cr-wo/hd7
to F). This result indicates that interlocking of ing 10C (C represents the DNA amount of the double mutants and was least severe in cr-hd7/
trichomes between adjacent separate organs unreplicated haploid genome) (Fig. 4B). hd7l double mutants (Fig. 4, C to E). The distinct
can generate organ fusion. Among different combinations of double involvement of the three HD-Zip IV genes in
mutants of Wo, HD7, and HD7L, cr-hd7/wo style development may be due to their different
HD-Zip IV promotes endoreduplication in the style had the smallest style cells (Fig. 4C). How- expression patterns in style tissue (Fig. 3A).
Style development exhibits a spatially polar ever, the number of style cells in this mutant Based on our results, we propose that the
distribution of cell division and cell expansion. was increased compared with WT (Fig. 4, D to three HD-Zip IV genes function redundantly
Fig. 3. HD7, HD7L, and Wo coordinate the development of anther cone and trichomes in different backgrounds. (F) Quantification of nuclear size. Nuclear
style. (A) Promoter-driven GUS assay shows that HD7, WO, and HD7L are DNA content is normalized to that in guard cells. Data are mean ± SD
specifically expressed in the upper region of the style. −8d and −4d are 8 and (n > 10). (G) Style phenotype in different backgrounds and the quantification
4 days before flower opening, respectively. (B) GUS staining of the cross sections of mature style length. Data are mean ± SD (n > 5). Unpaired t tests were used
of -10d anthers. Note that HD7, WO, and HD7L are specifically expressed at for statistical analysis (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).
the junction between adjacent anthers. (C) Trichome phenotype in the -10d (H) Model depicting the roles of HD7, WO, and HD7L in the initiation of interlocking
anther of cr-hd7/wo, cr-wo/hd7l, and cr-hd7/hd7l. (D) Quantification of trichome trichomes. (I) Illustration showing that the neighboring anthers are united
number. Data are mean ± SD (n > 5). (E) DAPI staining of the mature interlocking together by the intertwining interlocked trichomes to form an anther cone.
Fig. 4. HD-Zip IV regulators promote the endoreduplication of style mutants (n > 5). (F) Endoreduplication of style cells in cr-hd7/wo (marked by
giant cells. (A) (top) Tomato style exhibiting zone-specific features. (Middle) H2B-GFP). The mitotic activity of cr-hd7/wo is substantially higher than that of
SEM images showing that style cell size progressively increases from top to the WT at both the early (-10 d) and late stage (−2d) of style development.
bottom. The H2B-GFP marker indicates that division occurs only in the (G) Quantification of nuclear size in mature style cells of the WT and cr-hd7/wo
upper region of the style. (B) Quantification of the nuclei size in mature style cells (n > 10). (H) Model illustrating the spatial expression of HD-Zip IV genes and
expressing H2B-GFP. Nuclear DNA content is normalized to that in guard cells polar distribution of mitosis and endoreduplication in tomato style development.
(n > 10). (C) SEM images of style cells in cr-hd7/wo, cr-wo/hd7l, and cr-hd7/hd7l. Data are mean ± SD for all quantifications. Unpaired t tests were used for all
(D and E) Quantification of cell size (n > 10) and cell number of the style in the statistical analyses (*P < 0.05, ****P < 0.0001).
in anther cone formation and style length reg- tication process we examined the transcription exsertion resulting from an elongated style
ulation. However, we were unable to obtain of Wo, HD7, and HD7L in wild tomato relatives (fig. S8, A to G).
triple mutants likely because of embryonic and wild tomato species, including Solanum The expression of Style 2.1, a previously iden-
lethality. To overcome this we made a chimeric lycopersicoides, Solanum pennellii, and Solanum tified key factor for stigma shortening, was 1.5
fusion protein of Wo and HD7 proteins with habrochaites (29–32). Trichome-containing to 2.5 times lower in the modern tomato than
an SRDX transcriptional repressor (pWo:Wo- anthers did not evolve in S. lycopersicoides that of modern tomato ancestors including
SRDX and pHD7:HD7-SRDX). We found that and the anther of S. pennellii only forms stunt S. lycopersicoide (Fig. 5C). In addition, we
all transgenic lines expressing these chimeric trichomes without interlocking capacity (Fig. 5A). found that the expression Style 2.1 was con-
repressive proteins showed phenotypes of Consistent with these differences, our transcrip- siderably decreased in the styles of cr-hd7/wo
shortened styles with smaller cells, similar to tional analyses indicated that the expression double mutants but considerably increased in
cr-hd7/wo; they also showed disappearance of of Wo, HD7, and HD7L in the anthers of S. the styles of WoP635R mutants (Fig. 5D and fig.
interlocking trichomes on the anthers, result- lycopersicoides and S. pennellii was 2 to 4 times S8H). In both wild tomato S. pennellii and the
ing in split anther cones (fig. S7). These results lower than that of the modern tomato (Fig. 5B). modern tomato MT, the spatial expression
suggest that HD7, Wo, and HD7L are HD-Zip IV It is possible that domestication enhanced pattern of Style 2.1 in the style resembles that
regulators that jointly promote interlock- the expression of the HD-Zip IV regulators in of the HD Zip IV genes (Fig. 5E and Fig. 3A). In
ing trichome formation and style cell endo- modern varieties, leading to the formation of addition, the expression of all Style 2.1 genes
reduplication (Fig. 3, H and I, and Fig. 4H). anther cones and an exserted style. To test this from different tomato species and relatives
possibility we observed the floral organs of the was repressed by woW106R (fig. S9, A and B).
Cleistogamy was selected for during WoP635R mutant, a functionally enhanced mu- Raising Style 2.1 expression level in the woW106R
tomato domestication tation that substantially increases Wo pro- background considerably mitigated the cell
Although the modern tomato is a cleistoga- tein stability (25). In this mutant, we observed division phenotype in the woW106R mutant (Fig.
mous species, wild tomato species reproduce an increase in interlocking trichomes on the 5F and fig. S9, C and F). These results suggest
by crosspollination. To understand the domes- anthers compared with the WT and stigma that Style 2.1 is downstream of the pathways
Fig. 5. Domestication of anther cone is putatively mediated by enhanced (F) Misexpression of Style 2.1 driven by the Wo promoter partially rescued the
expression of HD-Zip IV genes. (A) SEM images showing the incomplete reduced endoreduplication phenotype in the style of woW106R mutants.
anther cone in tomato close relatives and wild tomato varieties. (B) Transcription (G) Predicted model of the transition from crosspollination to cleistogamy in wild
of HD7, HD7L, and Wo is considerably enhanced in the anther of MT. tomatoes. The expression of HD-Zip IV gene was increased first, resulting in a
(C) However, Style 2.1 expression in the style of MT is substantially reduced cylinder-shaped anther and exposed stigma in wild tomatoes, followed by the
compared with S. lycopersicoides, S. pennellii, and S. habrochaites (n = 3 recessed stigma which appeared after removal of self-incompatibility. (H) Fruit
biological replicates). (D) Expression of Style 2.1 in a cr-hd7/wo mutant (n = 3 set and (I) quantification for the woW106R mutant. Data are mean ± SD for all
biological replicates). (E) GUS reporter of the Style 2.1 promoters from wild quantifications. Unpaired t tests were used for all statistical analyses (ns, not
tomato (S. Pennellii) and cultivated tomato (MT). SP, S. Pennellii; MT, micro-Tom. significant) (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).
regulated by HD7 and Wo. Based on our results crosspollination in wild tomato species before formation of giant cells on Arabidopsis sepals
we propose that the increased expression of the self-incompatibility trait was lost. Simulta- through a dose-dependent manner (23, 24).
HD-Zip IV genes led to an exserted stigma and neous promotion of anther fusion and style Additionally, style cells of different sizes were
another cone structure in wild tomato during elongation seems to ensure the exserted style formed in different combinations of cr-hd7/
its evolution. Subsequently, the domestication for the pollination by insects or other pollina- wo, cr-wo/hd7l, etc., suggesting that the dosage
of stigma shortening occurred only after the loss tors. Upon the loss of self-incompatibility, the effect may also exist in the development of in-
of self-incompatibility (Fig. 5G). These stepwise exserted stigma was not necessary, the style terlocking trichomes and style giant cells.
processes may have eventually produced cleis- evolved to be shorter, and the stigma was con- However, the threshold of functional doses
togamy in the modern tomato. verted to an inserted one, eventually leading seems to be tissue-specific. High levels of Wo
to cleistogamy in the modern tomato. In sup- appear to promote cell division in trichomes
Discussion port of cleistogamy favoring fruit set, we found (34), while inhibiting cell division in the style.
It has been previously reported that the con- that the dominant negative mutant woW106R A similar phenomenon has also been observed
version from exserted to inserted stigma in had a 15 to approximately 30% higher fruit set with another HD Zip IV protein, GL2, which
modern tomatoes is due to the mutations in rate than the WT (Fig. 5, H and I). This could has opposite effects on trichomes and root hairs
the Style 2.1 promoters (11). When the HD-Zip provide an advantage during evolution or be in the shoot and root, respectively (35, 36).
regulators promote anther cone formation deliberately selected in domestication. The Dominant negative effects often occur during
they also promote style elongation, which ap- mechanism identified in tomato could also regulation that requires the formation of homo-
pears to be detrimental to the formation of a be applied to many other plant species with or heterodimers (37, 38). The Arabidopsis
fully closed cleistogamous flower. However, anther cones in nature (14, 33). gene GL2, also a HD-Zip IV protein, binds DNA
this mechanism is likely to be evolutionarily AtML1 and PDF2 are homologs of HD7, Wo, through its HD domain whereas the START
advantageous because it may have facilitated and HD7L in Arabidopsis. They promote the domain of GL2 is required to form a dimer (39).
The dominant mutation or deletion mutation 19. M. Abe, H. Katsumata, Y. Komeda, T. Takahashi, Development scanning. Wild tomato species were obtained from the Tomato
of HD7 and HD7L in the HD domain could lead 130, 635–643 (2003). Genetics Resource Center (TGRC). Funding: This work is
20. E. Ogawa et al., Plant Cell Physiol. 56, 1183–1192 (2015). supported by the grant from the National Natural Science
to the formation of non-functional dimers sim- 21. B. Rombolá-Caldentey, P. Rueda-Romero, R. Iglesias-Fernández, Foundation of China (32370354), and Outstanding Talent Fund
ilar to woW106R with other HD-Zip IV proteins P. Carbonero, L. Oñate-Sánchez, Plant Cell 26, 2905–2919 and Science Promotion Fund of Fujian Agriculture and Forestry
(fig. S10, A to C). Because the HD and START (2014). University (125-118992201A and KSYLX003, to S.W.). Author
22. N. Kamata, H. Okada, Y. Komeda, T. Takahashi, Plant J. 75, contributions: M.L.W. and S.W. conceived and designed the
domains are highly conserved structural do- 430–440 (2013). experiments; M.L.W., X.X.B., B.B.H., Y.D.D., S.R.H., J.Y.S., and
mains in HD-Zip IV proteins (39), the dominant- 23. A. H. K. Roeder, A. Cunha, C. K. Ohno, E. M. Meyerowitz, Y.L.W. performed most of the experiments and analyzed the data;
negative effect caused by mutations in the HD Development 139, 4416–4427 (2012). X.X.B., B.B.H., S.R.H., and J.Y.S. performed tomato stable
24. H. M. Meyer et al., eLife 6, e19131 (2017). transformation; M.L.W., X.X.B., and S.W. wrote the article.
domain is likely to be a common mechanism. 25. M. Wu et al., Dev. Cell 58, 278–288.e5 (2023). Competing interests: The authors declare no competing interests.
26. B. Hua et al., Plant Biotechnol. J. 19, 375–393 (2021). Data and materials availability: All data are available in the main
RE FE RENCES AND N OT ES
27. C. Yang et al., Proc. Natl. Acad. Sci. U.S.A. 108, 11836–11841 text or the supplementary materials. Tomato mutants and
1. M. M. Strelin, P. K. Diggle, M. A. Aizen, Trends Plant Sci. 28, (2011). constructs generated in this study are all available under a material
1360–1369 (2023). 28. I. Herskowitz, Nature 329, 219–222 (1987). transfer agreement from the Fujian Agriculture and Forestry
2. D. W. Schemske, Ecology 59, 596–613 (1978). 29. P. A. Bedinger et al., Sex. Plant Reprod. 24, 171–187 (2011). University. Accession Numbers: Woolly, Solyc02 g080260; HD7,
3. D. J. Schoen, D. G. Lloyd, Biol. J. Linn. 23, 303–322 (1984). 30. M. A. Mutschler, B. E. Liedl, Genetic Control of Self- Solyc10 g005330; HD7-LIKE, Solyc03 g031760; HD1, Solyc06
4. G. L. Stebbins, Am. Nat. 91, 337–354 (1957). Incompatibility and Reproductive Development in Flowering g050160; HD2, Solyc06 g035940; HD3, Solyc03 g026070; HD4,
5. S. C. H. Barrett, Nat. Rev. Genet. 3, 274–284 (2002). Plants, E. G. Williams, A. E. Clarke, R. B. Knox, Eds. (Springer, Solyc03 g120620; HD5, Solyc07 g041850; HD6, Solyc12 g005830;
6. C. M. Rick, W. H. Dempsey, Bot. Gaz. 130, 180–186 (1969). 1994), pp. 164–188. HD8, Solyc03 g098200; HD9, Solyc06 g072310; HD10, Solyc08
7. S. Takayama, A. Isogai, Annu. Rev. Plant Biol. 56, 467–489 31. I. E. Peralta, D. M. Spooner, S. Knapp, Syst. Bot. Monogr. 84, g076370; HD11, Solyc10 g018000; HD12, Solyc09 g066030; HD13,
(2005). 1–186 (2008). Solyc11 g011940; CD2, Solyc01 g091630; Style 2.1, Solyc02
8. W. Li, R. T. Chetelat, Science 330, 1827–1830 (2010). g087860; MX1, Solyc01 g010910; MTR1, Solyc10 g083140; ACTIN2,
32. N. Li et al., Nat. Genet. 55, 852–860 (2023).
9. J. W. Busch, D. J. Schoen, Trends Plant Sci. 13, 128–136 Solyc11 g005330. License information: Copyright © 2024 the
33. N. Holstein, M. Gottschling, Flora 240, 7–15 (2018).
(2008). authors, some rights reserved; exclusive licensee American
34. M. L. Wu et al., Plant Cell koae077 (2024).
10. L. Shang et al., Plant Cell 33, 3293–3308 (2021). Association for the Advancement of Science. No claim to original
35. W. G. Rerie, K. A. Feldmann, M. D. Marks, Genes Dev. 8,
11. K.-Y. Chen, B. Cong, R. Wing, J. Vrebalov, S. D. Tanksley, US government works. https://www.science.org/about/science-
1388–1399 (1994).
Science 318, 643–645 (2007). licenses-journal-article-reuse
36. J. D. Masucci et al., Development 122, 1253–1260 (1996).
12. R. Subramanya, H. Y. Ozaki, Euphytica 33, 13–16 (1984).
37. E. D. Shpak, M. B. Lakeman, K. U. Torii, Plant Cell 15,
13. G. Kowalska, Acta Sci. Pol. Hortorum Cultus 7, 63–76
1095–1110 (2003). SUPPLEMENTARY MATERIALS
(2008).
38. Q. Xin et al., Plant J. 103, 1723–1734 (2020).
14. R. Hilgenhof et al., Taxon 72, 811–847 (2023). science.org/doi/10.1126/science.adl1982
39. T. Mukherjee et al., Plant Physiol. 190, 2315–2334
15. B. J. Glover, S. Bunnewell, C. Martin, Gene 331, 1–7 Figs. S1 to S10
(2022).
(2004). Tables S1 and S2
16. P. Lu, R. Porat, J. A. Nadeau, S. D. O’Neill, Plant Cell 8, References (40–47)
ACKN OWLED GMEN TS
2155–2168 (1996). MDAR Reproducibility Checklist
17. A. Sessions, D. Weigel, M. F. Yanofsky, Plant J. 20, 259–263 We thank Z. Yang and J. Chang for critical comments, A. Wang
Movie S1
(1999). for sharing S. habrochaites (LA1777), K. Shi for sharing
18. M. Abe, T. Takahashi, Y. Komeda, Plant J. 26, 487–494 S. lycopersicoides, and C. Li for sharing VF36 and the introgression Submitted 4 October 2023; accepted 4 March 2024
(2001). lines. We thank B. Li and L. Xu for assistance with microCT 10.1126/science.adl1982
Electronically submit your new product description or product literature information! Go to www.science.org/about/new-products-section for more information.
Newly offered instrumentation, apparatus, and laboratory materials of interest to researchers in all disciplines in academic, industrial, and governmental organizations are featured in this
space. Emphasis is given to purpose, chief characteristics, and availability of products and materials. Endorsement by Science or AAAS of any products or materials mentioned is not
implied. Additional information may be obtained from the manufacturer or supplier.
Visit ScienceCareers.org
today — all resources are free
opportunities in china
capabilities, excellent young scholars, postdoctoral researchers, and other high-level talents. We provide generous remunera-
tion, scientific research commencement fees, settling-in allowances, and housing subsidies. Also, we offer access to related
research platforms and supportive research conditions. Talents joining us will be eligible for preferential policies and benefits
based on national, provincial, and university-level plans (projects), such as talent application, children's enrollment, household
settlement, and talent apartments.
CDUT sincerely invites talents worldwide to join us and achieve academic results at new heights.
Scan the QR Code to follow CDUT on WeChat
For more job requirements and conditions, please visit CDUT's talent recruitment website at http://hr.cdut.edu.cn.
IT’S NOT
JUST A JOB.
IT’S A CALLING.
Find your next job at ScienceCareers.org
WORKING LIFE
By Denis Meuthen
Beyond words
W
hen I started my faculty position, I was excited to be leading my own lab—and nervous.
I’m legally deaf and rely on lip-reading for verbal communication. I had managed fine as a
graduate student and postdoc, though not without misunderstandings and challenges. But
leading a team was different. I worried about whether I would be able to communicate ef-
fectively with my lab members, and also whether they would respect me. My pronunciation
is sometimes off because of my disability, which leads some people to judge my intelligence
as lacking. I set out unsure how to navigate these uncertain waters. But after almost 2 years in the
position, I’ve come up with a set of solutions for how best to communicate with my trainees. Many
would be useful for any lab head, disabled or not.
go along. And time with the students allows me to get to Fostering multiple lines of communication can benefit all
know them better and carefully explain the rationale be- mentors and trainees, preventing misunderstandings that
hind each experimental step. conversation alone can lead to. j
I also established written communication as the most
important way to exchange information. Through instant Denis Meuthen is a Freigeist fellow and a research group leader at Biele-
messaging services, I can quickly reach out to my students feld University. Send your career story to SciCareerEditor@aaas.org.