Textile fibers that light up Nitrated bacterial antigens wins

without a battery pp. 29 & 74 BII & Science Prize p. 41

$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.

LISTEN & SUBSCRIBE

NEW EPISODES EVERY THURSDAY

 CONTENTS
5 A P R I L 2 0 24 • VO LU M E 3 8 4 • I S S U E 6 6 9 1

33

NEWS 18 Gates foundation places 31 Exceptionally long-lived

bold bet on preprints nuclear RNAs
Funder to mandate preprints and no longer RNA labeled in young mice is detected
pay open-access publishing fees for grantees 2 years later in adult mouse brains
IN BRIEF By J. Brainard By J. Lawrence and L. Hall
RESEARCH ARTICLE p. 53
10 News at a glance FEATURES

IN DEPTH 20 Lucy’s world 33 Assessing the health burden from

12 Worries about bird flu in U.S. cattle
intensify
Farm worker becomes infected as H5N1
appears to spread between dairy cows in five
states By J. Cohen
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 A. Gibbons
PODCAST
air pollution
A broader approach to assessing the burden
of disease from air pollution is required
By T. Sigsgaard and B. Hoffmann
35 Roger Guillemin (1924—2024)
Father of neuroendocrinology By C. Rivier

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

16 ‘Hydrovoltaics’ tap energy from 29 Intelligent textiles are looking bright

ubiquitous moisture
New devices convert raindrops, dripping water,
and even humidity into minuscule currents
By R. F. Service
17 U.K. boosts digitization of
museum specimens
London’s Natural History Museum gets nearly
$200 million to scan country’s collections
By E. Pennisi
Flexible fiber electronics couple with the
human body for wireless tactile sensing
By Y. Li and Y. Luo
RESEARCH ARTICLE p. 74
30 Epithelial cells crowded out in asthma
Bronchoconstriction causes epithelial cell
extrusion that promotes airway inflammation
By J. M. Drazen and J. J. Fredberg
RESEARCH ARTICLE p. 66
39 Digitizing nature
Digital tools are transforming conservation
work but must be carefully deployed, argues
an environmental scholar
By J. Garard and H. D. Matthews
40 Genes, in context
As biologists have long known, the genome is
only one piece in the puzzle of life
By M. A. Goldman

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 3

The NOMIS & Science Young Explorer Award recognizes and rewards
early-career M.D., Ph.D., or M.D./Ph.D. scientists that perform research at the
intersection of the social and life sciences. Essays written by these bold
researchers on their recent work are judged for clarity, scientific quality,
creativity, and demonstration of cross-disciplinary approaches to address
fundamental questions.

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.

Apply by May 15, 2024

 CONTENTS

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 Staff ..................................................6

New Products ..............................................131
Trichomes zip up and close tomato floral structures to ensure self-pollination. Science Careers ......................................... 132

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.

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 5

 Editor-in-Chief Holden Thorp, hthorp@aaas.org BOARD OF REVIEWING EDITORS (Statistics board members indicated with S)
Executive Editor Valda Vinson Erin Adams, U. of Chicago Kelley Harris, U. of Wash Martin Plenio, Ulm U.
Takuzo Aida, U. of Tokyo Carl-Philipp Heisenberg, Katherine Pollard, UCSF
Editor, Research Jake S. Yeston Editor, Insights Lisa D. Chong Deputy Executive Editor Lauren Kmec
Leslie Aiello, Wenner-Gren Fdn. IST Austria Elvira Poloczanska,
DEPUTY EDITORS Gemma Alderton (UK),Stella M. Hurtley (UK), Phillip D. Szuromi, Sacha Vignieri SR. EDITORS Caroline Ash (UK),
Deji Akinwande, UT Austin Christoph Hess, Alfred-Wegener-Inst.
Michael A. Funk, Brent Grocholski, Di Jiang, Priscilla N. Kelly, Marc S. Lavine (Canada), Sarah Lempriere (UK), Mattia Maroso,
James Analytis, UC Berkeley U. of Basel & U. of Cambridge Julia Pongratz,
Yevgeniya Nusinovich, Ian S. Osborne (UK), L. Bryan Ray, H. Jesse Smith, Keith T. Smith (UK), Jelena Stajic, Peter Stern (UK),
Paola Arlotta, Harvard U. Heather Hickman, NIAID, NIH Ludwig Maximilians U.
Valerie B. Thompson, Brad Wible ASSOCIATE EDITORS Bianca Lopez, Sarah Ross (UK), Madeleine Seale (UK), Corinne Simonti, Yury V.
Delia Baldassarri, NYU Hans Hilgenkamp, U. of Twente Philippe Poulin, CNRS
Suleymanov, Ekeoma Uzogara SENIOR LETTERS EDITOR Jennifer Sills NEWSLETTER EDITOR Christie Wilcox RESEARCH & DATA ANALYST Jessica
Nenad Ban, ETH Zürich Janneke Hille Ris Lambers, Suzie Pun, U. of Wash
L. Slater LEAD CONTENT PRODUCTION EDITORS Chris Filiatreau, Harry Jach SR. CONTENT PRODUCTION EDITOR Amelia Beyna CONTENT
Christopher Barratt, U. of Dundee ETH Zürich Lei Stanley Qi, Stanford U.
PRODUCTION EDITORS Anne Abraham, Robert French, Julia Haber-Katris, Nida Masiulis, Abigail Shashikanth, Suzanne M. White
Franz Bauer, Kai-Uwe Hinrichs, U. of Bremen Simona Radutoiu, Aarhus U.
SR. EDITORIAL MANAGERS Carolyn Kyle, Beverly Shields SR. PROGRAM ASSOCIATE Maryrose Madrid EDITORIAL ASSOCIATES Aneera Dobbins,
Pontificia U. Católica de Chile Deirdre Hollingsworth, Trevor Robbins, U. of Cambridge
Joi S. Granger, Lisa Johnson, Anita Wynn SR. EDITORIAL COORDINATORS Jeffrey Hearn, Alexander Kief, Ronmel Navas, Jerry
Ray H. Baughman, UT Dallas U. of Oxford Adrienne Roeder, Cornell U.
Richardson, Alice Whaley (UK) EDITORIAL COORDINATORS Maura Byrne, Clair Goodhead (UK), Isabel Schnaidt,
Carlo Beenakker, Leiden U. Christina Hulbe, U. of Otago, Joeri Rogelj, Imperial Coll. London
Brian White ADMINISTRATIVE COORDINATOR Karalee P. Rogers ASI DIRECTOR, OPERATIONS Janet Clements (UK) ASI OFFICE MANAGER
Yasmine Belkaid, NIAID, NIH New Zealand John Rubenstein, SickKids
Victoria Smith ASI SR. OFFICE ADMINISTRATORS Dawn Titheridge (UK), Jessica Waldock (UK) COMMUNICATIONS DIRECTOR Meagan Phelan
Kiros T. Berhane, Columbia U. Randall Hulet, Rice U. Mike Ryan, UT Austin
DEPUTY DIRECTOR Matthew Wright SENIOR WRITER Walter Beckwith WRITERS Joseph Cariz, Abigail Eisenstadt SENIOR COMMUNICATIONS
Joseph J. Berry, NREL Auke Ijspeert, EPFL Alberto Salleo, Stanford U.
ASSOCIATE Zachary Graber COMMUNICATIONS ASSOCIATES Kiara Brooks, Haley Riley, Mackenzie Williams, Sarah Woods
Chris Bowler, Gwyneth Ingram, ENS Lyon Miquel Salmeron,
News Editor Tim Appenzeller École Normale Supérieure Darrell Irvine, MIT
Lawrence Berkeley Nat. Lab
NEWS MANAGING EDITOR John Travis INTERNATIONAL EDITOR David Malakoff DEPUTY NEWS EDITORS Rachel Bernstein, Shraddha Chakradhar, Ian Boyd, U. of St. Andrews Akiko Iwasaki, Yale U.
Nitin Samarth, Penn State U.
Elizabeth Culotta, Martin Enserink, David Grimm, Eric Hand, Kelly Servick, Matt Warren (Europe) SR. CORRESPONDENTS Daniel Clery Malcolm Brenner, Erich Jarvis, Rockefeller U.
Erica Ollmann Saphire,
(UK), Jon Cohen, Jeffrey Mervis, Elizabeth Pennisi ASSOCIATE EDITORS Jeffrey Brainard, Michael Price NEWS REPORTERS Adrian Cho, Baylor Coll. of Med. Peter Jonas, IST Austria
La Jolla Inst.
Jennifer Couzin-Frankel, Phie Jacobs, Jocelyn Kaiser, Rodrigo Pérez Ortega (Mexico City), Robert F. Service, Erik Stokstad, Paul Emily Brodsky, UC Santa Cruz Sheena Josselyn, U. of Toronto
Joachim Saur, U. zu Köln
Voosen, Meredith Wadman CONTRIBUTING CORRESPONDENTS Vaishnavi Chandrashekhar, Dan Charles, Warren Cornwall, Andrew Ron Brookmeyer, UCLA (S) Matt Kaeberlein, U. of Wash.
Alexander Schier, Harvard U.
Curry (Berlin), Ann Gibbons, Sam Kean, Kai Kupferschmidt (Berlin), Andrew Lawler, Mitch Leslie, Virginia Morell, Dennis Normile Christian Büchel, UKE Hamburg Daniel Kammen, UC Berkeley
Wolfram Schlenker, Columbia U.
(Tokyo), Elisabeth Pain (Careers), Charles Piller, Gabriel Popkin, Joshua Sokol, Richard Stone, Emily Underwood, Gretchen Vogel Johannes Buchner, TUM Kisuk Kang, Seoul Nat. U.
Susannah Scott, UC Santa Barbara
(Berlin), Lizzie Wade (Mexico City) CAREERS Katie Langin (Associate Editor) COPY EDITORS Julia Cole (Senior Copy Editor), Hannah Dennis Burton, Scripps Res. V. Narry Kim, Seoul Nat. U.
Anuj Shah, U. of Chicago
Knighton, Cyra Master (Copy Chief) ADMINISTRATIVE SUPPORT Meagan Weiland Carter Tribley Butts, UC Irvine Nancy Knowlton, Smithsonian
Vladimir Shalaev, Purdue U.
György Buzsáki, Etienne Koechlin,
Creative Director Beth Rakouskas NYU School of Med. École Normale Supérieure
Jie Shan, Cornell U.
DESIGN MANAGING EDITOR Chrystal Smith GRAPHICS MANAGING EDITOR Chris Bickel PHOTOGRAPHY MANAGING EDITOR Emily Petersen Mariana Byndloss, Alex L. Kolodkin, Johns Hopkins U. ay Shendure, U. of Wash.
MULTIMEDIA MANAGING PRODUCER Kevin McLean WEB CONTENT STRATEGY MANAGER Kara Estelle-Powers DESIGN EDITOR Marcy Atarod Vanderbilt U. Med. Ctr. LaShanda Korley, U. of Delaware Steve Sherwood,
DESIGNER Noelle Jessup SENIOR SCIENTIFIC ILLUSTRATOR Noelle Burgess SCIENTIFIC ILLUSTRATORS Austin Fisher, Kellie Holoski, Ashley Annmarie Carlton, UC Irvine Paul Kubes, U. of Calgary U. of New South Wales
Mastin SENIOR GRAPHICS EDITOR Monica Hersher GRAPHICS EDITOR Drew An-Pham SENIOR GRAPHICS SPECIALISTS Holly Bishop, Nathalie Simon Cauchemez, Inst. Pasteur Chris Kuzawa, Northwestern U. Brian Shoichet, UCSF
Cary SENIOR PHOTO EDITOR Charles Borst PHOTO EDITOR Elizabeth Billman SENIOR PODCAST PRODUCER Sarah Crespi SENIOR VIDEO PRODUCER Ling-Ling Chen, SIBCB, CAS Laura Lackner, Northwestern U. Robert Siliciano, JHU School of Med.
Meagan Cantwell SOCIAL MEDIA STRATEGIST Jessica Hubbard SOCIAL MEDIA PRODUCER Sabrina Jenkins WEB DESIGNER Jennie Pajerowski Wendy Cho, UIUC Gabriel Lander, Scripps Res. (S) Lucia Sivilotti, UCL
Ib Chorkendorff, Denmark TU Mitchell A. Lazar, UPenn Emma Slack,
Chunaram Choudhary, Hedwig Lee, Duke U. ETH Zürich & U. of Oxford
Chief Executive Officer and Executive Publisher Sudip Parikh Københavns U. Fei Li, Xi'an Jiaotong U. Richard Smith, UNC (S)
Publisher, Science Family of Journals Bill Moran Karlene Cimprich, Stanford U. Jianyu Li, McGill U. John Speakman, U. of Aberdeen
DIRECTOR, BUSINESS SYSTEMS AND FINANCIAL ANALYSIS Randy Yi DIRECTOR, BUSINESS OPERATIONS & ANALYSIS Eric Knott MANAGER, Laura Colgin, UT Austin Ryan Lively, Georgia Tech Allan C. Spradling,
BUSINESS OPERATIONS Jessica Tierney MANAGER, BUSINESS ANALYSIS Cory Lipman MANAGER, WEB ANALYTICS Samantha Cressman James J. Collins, MIT Luis Liz-Marzán, CIC biomaGUNE Carnegie Institution for Sci.
BUSINESS ANALYSTS Kurt Ennis, Maggie Clark FINANCIAL ANALYST Isacco Fusi BUSINESS OPERATIONS ADMINISTRATOR Taylor Fisher Robert Cook-Deegan, Omar Lizardo, UCLA V. S. Subrahmanian,
SENIOR PRODUCTION MANAGER Jason Hillman SENIOR MANAGER, PUBLISHING AND CONTENT SYSTEMS Marcus Spiegler CONTENT Arizona State U. Jonathan Losos, WUSTL Northwestern U.
OPERATIONS MANAGER Rebecca Doshi SENIOR CONTENT & PUBLISHING SYSTEMS SPECIALIST Jacob Hedrick SENIOR PRODUCTION SPECIALIST Virginia Cornish, Columbia U. Ke Lu, Inst. of Metal Res., CAS Sandip Sukhtankar, U. of Virginia
Kristin Wowk PRODUCTION SPECIALISTS Kelsey Cartelli, Audrey Diggs DIGITAL PRODUCTION MANAGER Lisa Stanford SENIOR DIGITAL Carolyn Coyne, Duke U. Christian Lüscher, U. of Geneva Naomi Tague, UC Santa Barbara
ADVERTISING SPECIALIST Kimberley Oster ADVERTISING PRODUCTION OPERATIONS MANAGER Deborah Tompkins DESIGNER, CUSTOM Roberta Croce, VU Amsterdam Jean Lynch-Stieglitz, Georgia Tech Eriko Takano, U. of Manchester
PUBLISHING Jeremy Huntsinger SR. TRAFFIC ASSOCIATE Christine Hall SPECIAL PROJECTS ASSOCIATE Shantel Agnew Molly Crocket, Princeton U. David Lyons, U. of Edinburgh A. Alec Talin, Sandia Natl. Labs
Christina Curtis, Stanford U. Fabienne Mackay, QIMR Berghofer Patrick Tan, Duke-NUS Med. School
ASSOCIATE DIRECTOR, BUSINESS DEVELOPMENT Justin Sawyers GLOBAL MARKETING MANAGER Allison Pritchard DIGITAL MARKETING
Ismaila Dabo, Penn State U. Zeynep Madak-Erdogan, UIUC Sarah Teichmann,
MANAGER Aimee Aponte JOURNALS MARKETING MANAGER Shawana Arnold MARKETING ASSOCIATES Ashley Hylton, Mike Romano,
Jeff L. Dangl, UNC Vidya Madhavan, UIUC Wellcome Sanger Inst.
Lorena Chirinos Rodriguez, Jenna Voris SENIOR DESIGNER Kim Huynh
Nicolas Dauphas, U. of Chicago Anne Magurran, U. of St. Andrews Dörthe Tetzlaff, Leibniz Institute of
DIRECTOR AND SENIOR EDITOR, CUSTOM PUBLISHING Erika Gebel Berg ASSISTANT EDITOR, CUSTOM PUBLISHING Jackie Oberst Claude Desplan, NYU Ari Pekka Mähönen, U. of Helsinki Freshwater Ecology and Inland Fisheries
PROJECT MANAGER Melissa Collins Sandra DÍaz, U. Nacional de CÓrdoba Asifa Majid, U. of Oxford Rocio Titiunik, Princeton U.
DIRECTOR, PRODUCT & PUBLISHING DEVELOPMENT Chris Reid DIRECTOR, BUSINESS STRATEGY AND PORTFOLIO MANAGEMENT Sarah Whalen Samuel Díaz-Muñoz, UC Davis Oscar Marín, King’s Coll. London Shubha Tole,
DIRECTOR, PRODUCT MANAGEMENT Kris Bishop PRODUCT DEVELOPMENT MANAGER Scott Chernoff PUBLISHING PLATFORM MANAGER Ulrike Diebold, TU Wien Matthew Marinella, Arizona State U. Tata Inst. of Fundamental Res.
Jessica Loayza SR. PRODUCT ASSOCIATE Robert Koepke PRODUCT ASSOCIATES Caroline Breul, Anne Mason Stefanie Dimmeler, Charles Marshall, UC Berkeley Maria-Elena Torres Padilla,
ASSOCIATE DIRECTOR, INSTITUTIONAL LICENSING MARKETING Kess Knight BUSINESS DEVELOPMENT MANAGER Rasmus Andersen ASSOCIATE
Goethe-U. Frankfurt Christopher Marx, U. of Idaho Helmholtz Zentrum München
DIRECTOR, INSTITUTIONAL LICENSING SALES Ryan Rexroth INSTITUTIONAL LICENSING MANAGER Marco Castellan, Claudia Paulsen-Young
Hong Ding, Inst. of Physics, CAS Geraldine Masson, CNRS Kimani Toussaint, Brown U.
Dennis Discher, UPenn Jennifer McElwain, Barbara Treutlein, ETH Zürich
SENIOR MANAGER, INSTITUTIONAL LICENSING OPERATIONS Judy Lillibridge SENIOR OPERATIONS ANALYST Lana Guz SYSTEMS & OPERATIONS
Jennifer A. Doudna, UC Berkeley Trinity College Dublin Li-Huei Tsai, MIT
ANALYST Ben Teincuff FULFILLMENT ANALYST Aminta Reyes
Ruth Drdla-Schutting, Scott McIntosh, NCAR Jason Tylianakis, U. of Canterbury
ASSOCIATE DIRECTOR, US ADVERTISING Stephanie O'Connor US MID WEST, MID ATLANTIC AND SOUTH EAST SALES Chris Hoag US WEST Med. U. Vienna Rodrigo Medellín,
COAST SALES Lynne Stickrod ASSOCIATE DIRECTOR, ROW Roger Goncalves SALES REP, ROW Sarah Lelarge SALES ADMIN ASSISTANT, ROW Matthew Vander Heiden, MIT
Raissa M. D'Souza, UC Davis U. Nacional Autónoma de México
Victoria Glasbey DIRECTOR OF GLOBAL COLLABORATION AND ACADEMIC PUBLISHING RELATIONS, ASIA Xiaoying Chu ASSOCIATE DIRECTOR, Wim van der Putten,
Bruce Dunn, UCLA Mayank Mehta, UCLA
INTERNATIONAL COLLABORATION Grace Yao SALES MANAGER Danny Zhao MARKETING MANAGER Kilo Lan ASCA CORPORATION, JAPAN Netherlands Inst. of Ecology
William Dunphy, Caltech C. Jessica Metcalf, Princeton U.
Rie Rambelli (Tokyo), Miyuki Tani (Osaka) Ivo Vankelecom, KU Leuven
Scott Edwards, Harvard U. Tom Misteli, NCI, NIH
Henrique Veiga-Fernandes,
DIRECTOR, COPYRIGHT, LICENSING AND SPECIAL PROJECTS Emilie David RIGHTS AND PERMISSIONS ASSOCIATE Elizabeth Sandler LICENSING Todd A. Ehlers, U. of Glasgow Jeffery Molkentin, Cincinnati
Champalimaud Fdn.
ASSOCIATE Virginia Warren RIGHTS AND LICENSING COORDINATOR Dana James CONTRACT SUPPORT SPECIALIST Michael Wheeler Nader Engheta, UPenn Children's Hospital Medical Center
Reinhilde Veugelers, KU Leuven
Tobias Erb, Alison Motsinger-Reif,
Bert Vogelstein, Johns Hopkins U.
MPS, MPI Terrestrial Microbiology NIEHS, NIH (S)
EDITORIAL MEDIA CONTACTS MEMBERSHIP AND INDIVIDUAL TREASURER Carolyn N. Ainslie Julia Von Blume, Yale School of Med.
SUBSCRIPTIONS
Karen Ersche, U. of Cambridge Danielle Navarro,
science_editors@aaas.org scipak@aaas.org CHIEF EXECUTIVE OFFICER Beate Escher, UFZ & U. of Tübingen U. of New South Wales
David Wallach, Weizmann Inst.
science.org/subscriptions Jane-Ling Wang, UC Davis (S)
NEWS PRODUCT ADVERTISING Sudip Parikh Barry Everitt, U. of Cambridge Daniel Neumark, UC Berkeley
Jessica Ware,
& CUSTOM PUBLISHING BOARD Cynthia M. Beall Vanessa Ezenwa, U. of Georgia Thi Hoang Duong Nguyen,
science_news@aaas.org MEMBER BENEFITS Amer. Mus. of Natural Hist.
advertising.science.org/ Toren Finkel, U. of Pitt. Med. Ctr. MRC LMB
aaas.org/membership/ Janine Austin Clayton David Waxman, Fudan U.
INFORMATION FOR AUTHORS products-services Natascha Förster Schreiber, Beatriz Noheda, U. of Groningen
benefits Kaye Husbands Fealing MPI Extraterrestrial Phys. Helga Nowotny, Alex Webb, U. of Cambridge
science.org/authors/ science_advertising@aaas.org Chris Wikle, U. of Missouri (S)
Kathleen Hall Jamieson Peter Fratzl, MPI Potsdam Vienna Sci. & Tech. Fund
science-information-authors INSTITUTIONAL SALES Terrie Williams, UC Santa Cruz
CLASSIFIED ADVERTISING Jane Maienschein Elaine Fuchs, Rockefeller U. Pilar Ossorio, U. of Wisconsin
AND SITE LICENSES Ian A. Wilson, Scripps Res. (S)
REPRINTS AND PERMISSIONS advertising.science.org/ Robert B. Millard Caixia Gao, Inst. of Genetics and Andrew Oswald, U. of Warwick
science.org/librarian Developmental Bio., CAS Isabella Pagano, Hao Wu, Harvard U.
science.org/help/ science-careers Babak Parviz
Daniel Geschwind, UCLA Istituto Nazionale di Astrofisica Li Wu, Tsinghua U.
reprints-and-permissions advertise@sciencecareers.org AAAS BOARD OF DIRECTORS William D. Provine Lindsey Gillson, U. of Cape Town Giovanni Parmigiani, Amir Yacoby, Harvard U.
MULTIMEDIA CONTACTS CHAIR Gilda A. Barabino Juan S. Ramírez Lugo Gillian Griffiths, U. of Cambridge Dana-Farber (S) Benjamin Youngblood, St. Jude
JOB POSTING CUSTOMER SERVICE
SciencePodcast@aaas.org PRESIDENT Keith R. Yamamoto Susan M. Rosenberg Simon Greenhill, U. of Auckland Zak Page, UT Austin Yu Xie, Princeton U.
employers.sciencecareers.org
Nicolas Gruber, ETH Zürich Sergiu Pasca, Standford U. Jan Zaanen, Leiden U.
ScienceVideo@aaas.org support@sciencecareers.org PRESIDENT-ELECT Willie E. May Vassiliki Betty Smocovitis
Hua Guo, U. of New Mexico Julie Pfeiffer, Kenneth Zaret, UPenn School of Med.
Science serves as a forum for discussion of important issues related to the advancement of science by publishing material on Taekjip Ha, Johns Hopkins U. UT Southwestern Med. Ctr. Lidong Zhao, Beihang U.
which a consensus has been reached as well as including the presentation of minority or conflicting points of view. Accordingly, Daniel Haber, Mass. General Hos. Philip Phillips, UIUC Bing Zhu, Inst. of Biophysics, CAS
all articles published in Science—including editorials, news and comment, and book reviews—are signed and reflect the individual Sharon Hammes-Schiffer, Yale U. Matthieu Piel, Inst. Curie Xiaowei Zhuang, Harvard U.
views of the authors and not official points of view adopted by AAAS or the institutions with which the authors are affiliated. Wolf-Dietrich Hardt, ETH Zürich Kathrin Plath, UCLA Maria Zuber, MIT

6 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

HFSP NAKASONE AWARD CALL FOR NOMINATIONS 2025
DEADLINE: 30 APRIL 2024
• Recognizing ground-breaking contributions that
advanced the frontiers of biological Knowledge.
Given for a clearly defined discovery(ies) in basic
life science that dates back not more than 10
2024 HFSP years.
NAKASONE
AWARDEE • For distinct conceptual, experimental, or
Maiken technological breakthroughs identifiable via
Nedergaard scientific publication(s).
University of
Rochester (USA) • Open competition: any scientist is welcome to
& University of nominate candidates.
Copenhagen
(Denmark), for • HFSP highly encourages nominations of female
research on the scientists, early career researchers, individuals
glymphatic who are ethnically and/or racially under-
system
represented in the scientific enterprise, and those
who hail from equity-seeking countries.
• The awardee will receive a commemorative gold
medal, a small unrestricted research grant, and
an invitation to deliver the HFSP Nakasone
Lecture at the 2025 HFSP Awardees Meeting.

Where Science Gets Social.

AAAS.ORG/COMMUNITY
Recent PhD?
How about US$30,000, having your essay
published in Science magazine, and being
celebrated with a week of festivities in Stockholm, Sweden?
Apply now to the Science & SciLifeLab Prize for Young Scientists!

A global award aimed at young researchers

Every year, four early-career scientists from across the globe are awarded the
Science & SciLifeLab Prize for Young Scientists for outstanding research they’ve
conducted as part of their PhD.
The winners are selected to receive up to US$30,000 in prize money for their
accomplishments. In addition, all four winners are published in Science and
invited by SciLifeLab to Sweden in December to participate in a unique week filled
with events in honor of science. Here, they will be given the opportunity to meet
with leading scientists in their field of research and create life-long connections to
support their career.

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
Don’t bury Mexico’s biodiversity capacity
I
n a world where biodiversity is on the line on
many fronts—from armed conflict to pandemics
to climate change—defending institutions that
have effectively managed it is paramount. In the
global effort to protect biodiversity, Mexico has
been at the forefront. In particular, for more than
30 years, Mexico’s National Commission for the
Knowledge and Use of Biodiversity (CONABIO) has
promoted research, compiled information on the
biodiversity of Mexico and elsewhere, and connected
academia, government, and society to guide decision-
making. Unfortunately, the demise of CONABIO,
which began in 2018 under the current administra-
tion, may be fully realized soon. Last month, the Mexi-
can government announced its intent
to reduce CONABIO from a multi-
ministry federal government agency
to a branch within the environment
ministry. This will strip CONABIO of
“Burying
its independent voice, credibility, and
influence on national and interna-
the agency is
tional policy. As this decision is open
for public comment, it is important
a path to
for the scientific community to speak
out strongly against this change.
its ultimate
Biodiversity richness underlies the
health of ecosystems and as such,
termination.”
the benefits that ecosystems offer
to society, such as food, materials,
energy, and clean air and water. For many cultures,
biodiversity is also intrinsically valuable. Thus, human
well-being and modern life are inextricably linked to
biodiversity, which makes the increasing rate of bio-
diversity decline alarming (extinction rate estimated
at one species per million species per year). There is
global agreement that biodiversity governance from
local to international levels requires reliable data that
can be integrated with other information to help in-
form processes and decisions.
CONABIO has been part of the solution to this chal-
lenge. It was created in 1992, in preparation for the
United Nations “Earth Summit” in Rio de Janeiro,
Brazil, which aimed to forge international cooperation
on environmental issues of the 21st century. The man-
date of CONABIO was to create a robust biodiversity
information infrastructure. Since then, CONABIO has
digitized tens of millions of records in national and
foreign collections and has processed satellite data on
vegetation cover and wildfires. It is not only a data
SCIENCE science.org
repository but a model for best practices in biodiver- Rodrigo A. Medellin
sity management. All information is publicly available is a professor at
online, and contributors decide when the information the Institute of
they provide will be made public. In the past 5 years, Ecology, Universidad
CONABIO had an average of 1000 users per week, and Nacional Autónoma
it was consulted at least once per day by health, agri- de México, Mexico
culture, environment, foreign affairs, and government City, Mexico.
agencies from Mexico and other countries. CONABIO
medellin@ecologia.
has become a world-sought reference on how to
unam.mx
effectively compile useful information and incorporate
it into public policy for the benefit of the population
Jorge Soberón
and biodiversity.
The proposed change for CONABIO will likely is director of the
eliminate the support it provides for the sustainable Biodiversity Institute
management, use, and conservation of the University of
of biodiversity for Mexico and the Kansas, Lawrence,
world. Given that Mexico is home to KS, USA, and former
10 to 12% of the world’s species, there
is much at risk. CONABIO is a high-
Executive Secretary
of CONABIO.
level government agency composed
of 10 cabinet ministers and has op-
jsoberon@ku.edu
erated with both public and private
funds as well as foreign agency sup-
port, making it economically and
politically independent. The current
administration’s response to criti-
cisms from Mexican scientists is that
moving CONABIO to a branch within
Mexico’s Secretariat of the Environ-
ment and Natural Resources (SEMARNAT) will make
it more efficient. However, the decision will strip the
agency of its economic, political, and intellectual
independence, preventing it from issuing science-
based opinions at an executive level of government.
CONABIO is already very cost-efficient, and its capac-
ity to influence political decisions depends directly on
its multi-ministerial character. Burying the agency is a
path to its ultimate termination.
Should this move become effective (a final govern-
ment decision is expected before the end of April),
it will be the end of the many essential benefits that
CONABIO provides to address the environmental cri-
ses that threaten the future of all life on Earth. In 2020,
over 80,000 people signed a campaign to save the
agency. It is time again for Mexico’s scientists, Mexico’s
population, and the world to speak out against destroy-
ing this valuable institution.
–Rodrigo A. Medellin and Jorge Soberón
10.1126/science.adp5399
5 APRIL 2024 • VOL 384 ISSUE 6691 9
 NEWS
IN BRIEF
Edited by
Kelly Servick

ASTRONOMY

Biggest ever digital camera heads to Chile

T
he 3200-megapixel digital camera that will serve
as the heart of the Vera C. Rubin Observatory
is complete and will head to Chile for integra-
tion, researchers announced this week. Built at
the SLAC National Accelerator Laboratory in
search for signs of dark energy and matter, looking
for their effects by measuring distance with the help
of pulsating stars. It will gauge minute distortions in
distant objects created when light bends around inter-
vening matter, which offer clues to how dark energy is

PHOTO: JACQUELINE RAMSEYER ORRELL/SLAC NATIONAL ACCELERATOR LABORATORY

California, the 3000-kilogram digital camera—
the largest ever built and roughly the size of a small
car—is designed to survey the full southern night sky
every 3 nights in unprecedented resolution from atop
Cerro Pachón in the Andes Mountains. Once the cam-
era is attached to its mirrors at the observatory, it will
driving the universe’s expansion. It will also capture a
multitude of fast-changing objects such as supernovae
and log an enormous haul of new asteroids and icy
objects in the Solar System. First light for the obser-
vatory, funded by the Department of Energy and the
National Science Foundation, is expected next year.

Limits to Paxlovid benefits
B I O M E D I C I N E | Pfizer’s antiviral
Paxlovid (nirmatrelvir/ritonavir) may
not help as many people as hoped. The
combo drug was granted emergency
use authorization in 2021 in the United
States after a clinical trial showed it
reduced hospitalizations and deaths
among people who were unvaccinated
against SARS-CoV-2 and had at least one
risk factor for severe COVID-19, such
as diabetes or obesity. But results from
a Pfizer-run trial of nearly 1300 people
suggest the benefit doesn’t extend to
those who are unvaccinated and have
no risk factors or are vaccinated with at
least one. The study, out this week in The
New England Journal of Medicine, found
that in these groups, Paxlovid was no
better than placebo at preventing deaths
and hospitalizations, or at reducing time
to symptom clearance. The findings
support the use of the drug “only for
persons who are at high risk for disease
progression,” write infectious disease
specialists Rajesh Gandhi and Martin

10 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

Hirsch, both of Harvard Medical School
and Massachusetts General Hospital, in
an accompanying commentary.
Census to revise race question
DEMOGRAPHY | Future surveys by U.S.
government agencies, including the decen-
nial census, will ask a single question about
race and ethnicity and allow respondents
to choose as many of the categories as
apply, including a new one for those iden-
tifying as Middle Eastern or North African.
The new rules for collecting demographic
data, announced on 29 March by the Office
of Management and Budget, address long-
standing complaints that surveys asking
separately about race and ethnicity were
hard to interpret and didn’t capture the
diversity of the U.S. population. The new
format offers seven categories, each with
six subgroups. For example, Chinese and
Filipino are two of the subgroups under
Asian. Respondents can also write in their
own designation under each category. Civil
rights groups think the new rules will lead
to better data. They say educating people
about their options will be key to narrow-
ing the historic undercount of groups
such as Latino and African Arab people.
Federal agencies have 5 years to adjust
existing surveys.
mRNA shows promise as a drug
B I O M E D I C I N E | The proteinmaking tech-
nology behind some of the most successful
COVID-19 vaccines is showing hints of
success in people with a rare metabolic
disease. Results from a small trial run by
Moderna, out this week in Nature, are the
first published clinical data showing that
messenger RNA (mRNA), used in vaccines
to deliver viral proteins that provoke an
immune response, could potentially work
as a drug to replace a missing protein.
The trial included 16 children and young
adults born with mutations that cause
propionic acidemia, in which cells lack an
enzyme that enables them to break down
certain proteins and fats. Even with a spe-
cial diet, toxic compounds build up and
can cause strokes, heart disease, and other
problems. Trial participants got up to
2 years of intravenous infusions, usually
2 weeks apart, of tiny fat particles contain-
ing mRNA encoding normal forms of the
missing enzyme. Among eight participants
who had at least one life-threatening medi-
cal emergency in the year before treatment,
such episodes fell by 70%.
THREE QS
Pandemic treaty talks drag on
| Member states of
G L O B A L H E A LT H
the World Health Organization (WHO)
failed to hammer out a sweeping treaty
to address global pandemic prepared-
ness during what was supposed to
be a final round of talks last week
in Geneva. WHO launched negotia-
tions about the Pandemic Agreement
in December 2021 with the aim of
addressing problems identified during
the COVID-19 pandemic, including a
searing global inequity in the distribu-
tion of vaccines. But sharp divisions
have emerged between member states,
in particular about a system called
“pathogen access and benefits-sharing,”
which would kick in during a pan-
demic. It would compel countries to
provide the world with information
about microbes within their borders
in return for access to free or afford-
able vaccines, drugs, and diagnostics.
Negotiators will hold another 11-day
meeting starting on 29 April, in hopes
of agreeing on a draft before the World
Health Assembly in late May.

Suit targets Florida hiring ban
| A Florida law that
I M M I G R AT I O N
requires its public universities to receive
a waiver before employing a graduate stu-
dent or postdoc from China, Iran, or any
of five other “countries of concern” is fac-
ing a challenge in a federal court. Enacted
in March 2023, the law violates the U.S.
Constitution, is discriminatory, and
ignores the federal government’s author-
ity to set immigration and employment
practices, according to the suit filed on
25 March in a U.S. District Court in Miami
by two science graduate students at Florida
International University (FIU)—Zhipeng
Yin and Zhen Guo—and agricultural
economist Zhengfei Guan of the University
of Florida (UF). The students were forced
to pay their own way after FIU officials
deferred graduate assistant positions with
stipends and tuition waivers that were
to start in December 2023. Guan says his
top candidate for a postdoc position went
elsewhere after learning about the Florida
law, which Guan says has “essentially made
it impossible to hire anyone from China.”
Last week, UF students rallied in support
of the plaintiffs outside a meeting of the
state’s Board of Governors, whose mem-
bers are among the defendants.
Why Europe’s battlefield bones are missing
European wars of the 18th and early 19th centuries were marked by cavalry charges and mas-
sive exchanges of cannon and rifle fire that could claim tens of thousands of lives within hours.
Yet archaeologists often struggle to find skeletal remains from these Napoleonic era battles. In
the book Bones of Contention: The Industrial Exploitation of Human Bones in the Modern Age,
published in February, a team of historians and archaeologists argues that industrial needs for
phosphates for fertilizer and bone char for beet sugar processing in the early 19th century led
Europeans at the time to raid bones from mass graves. Co-editor Bernard Wilkin, a historian at
the State Archives of Belgium, talked to Science about the grisly trade.
Q: Why would people resort to we know they ran out of accessible bones
robbing battlefields? in Europe, and turned to colonies abroad.
A: In the 1830s, bones were suddenly worth The French dug up cemeteries in Algeria
a lot of money. Battlefield bones in particular and shipped the bones to sugar factories
were a commodity that was easy to find, in Marseille; we know the British imported
easy to access, and no one really cared about. mummies and bones from Egypt on an
A lot of farmers living near these battlefields industrial scale, destroying untold heritage
realized there was gold in the ground. I in the process.
wouldn’t say these were grave-robbing
monsters. For the most part they were poor Q: What’s next?
farmers, who seized an opportunity. A: We need more evidence from elsewhere
in Europe to understand if this was as
Q: Did this trade target other widespread as we think. And we need to pay
burial grounds, too? attention to what was happening outside
A: This goes way beyond battlefields. We of Europe. The colonies were exploited,
know there were medieval cemeteries in too, and the way white people treated the
Scotland and England that were emptied dead in European colonies needs a lot more
out and sold; measures to ban the practice research. I’ve been a historian for 20 years,
were proposed, debated—and defeated— and I have the feeling this topic is the most
in the British Parliament. At some point important thing I will work on in my career.

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 11

NE WS
IN DEP TH
INFECTIOUS DISEASES
Worries about bird flu in U.S. cattle intensify
Farm worker becomes infected as H5N1 appears to spread between dairy cows in five states
By Jon Cohen
A
n H5N1 avian influenza strain that
has devastated birds and some mam-
mals around the world has pulled off
another big surprise. The virus has
caused outbreaks among cows at U.S.
dairy farms in at least four states, the
first of which were revealed on 25 March.
Now, a farm worker has also become infected,
most likely from contact with sick cattle. The
worker developed conjunctivitis, a mild eye
infection that frequently occurs when avian
influenza viruses jump into humans, Texas
health officials announced on 1 April.
“It’s definitely taken me by surprise, but
perhaps it shouldn’t have,” given the virus’ es-
capades so far, says virologist Richard Webby
of St. Jude Children’s Research Hospital.
The U.S. Department of Agriculture
(USDA) says, “Initial testing has not found
changes to the virus that would make it
more transmissible to humans.” Still, the
widespread occurrence of H5N1 in mam-
mals has renewed worries that it may evolve
to transmit more readily between people.
And scientists are urgently trying to answer
a host of new questions, including how far
the virus has spread among U.S. cows and
how to prevent more herds and people from
becoming infected.
H5N1 has decimated wild bird popula-
tions and poultry since 1996. The past few
12 5 APRIL 2024 • VOL 384 ISSUE 6691
years, a variant known as clade 2.3.4.4b has
spread globally, sickening not just birds, but
also dozens of mammalian species, including
cats, dogs, tigers, bears, seals, and dolphins.
Last month, goats contracted the virus at a
Minnesota farm that also had an infected
poultry flock.
Now, USDA says the virus has infected
cattle at dairy farms in Texas, Kansas, New
Mexico, and Michigan, while Idaho has a
“presumptive” outbreak at one farm. (The
agency noted that cats on farms have also be-
come infected.) Sick cows have a mild illness,
USDA says, and produce less milk, which is
thicker than usual. The infections are not en-
tirely unexpected: Other types of influenza vi-
ruses readily infect cows, and an experiment
published in 2006 demonstrated that a dif-
ferent H5N1 variant could infect calves.
The infected person in Texas is being
treated with oseltamivir, an antiviral drug,
and USDA stressed that the “current risk
to the public remains low.” Contamination
of commercial milk is of “no concern,” the
agency said, because pasteurization reliably
kills viruses, and milk from sick cows is not
sold. The U.S. Centers for Disease Control
and Prevention (CDC) says people should not
drink raw milk or products made from it.
More than 900 human cases of H5N1
have been reported since 1997, just over half
of them fatal, but the virus rarely spreads
between people. That’s because influenza
Water or feed contaminated by birds could
have infected U.S. dairy cows with H5N1.
receptors on human cells are different from
those in birds, and not a good match for avian
flu viruses.
But humans have the avian version of the
receptor in their eyes, which explains why
they can develop conjunctivitis, which is
typically a mild disease. Hundreds of people
developed eye infections during an outbreak
of H7N7, another highly pathogenic avian
influenza virus, in Dutch poultry flocks in
2003. There was some evidence of human-
to-human transmission during that episode,
and a veterinarian died.
One urgent question now is how to protect
other workers at dairy farms, who include
many immigrants. “We need to be advocat-
ing for our employees on these farms, provid-
ing them support and education in whatever
language that it needs to be in,” says Joe
Armstrong, a bovine veterinarian who works
with the University of Minnesota (UM) Ex-
PHOTO: ERIC THAYER/BLOOMBERG VIA GETTY IMAGES
tension. Dairy workers rarely wear protective
gear, such as masks and goggles, Armstrong
notes, and floors in milking facilities are of-
ten cleaned using high-pressure water spray-
ers, which could aerosolize the virus.
“I think the conjunctivitis in itself is not
so serious, but it points to the fact that those
people have been exposed and that they
might develop respiratory disease,” says Thijs
Kuiken, a comparative pathologist at Eras-
mus Medical Center who specializes in avian
influenza. Kuiken says he’s particularly con-
science.org SCIENCE

cerned because he has heard milk from in-
fected cows contains high levels of the virus.
Scientists are also trying to understand
how the virus is spreading through the
herds—and how to stop it. Dead birds have
been found at some of the affected proper-
ties, suggesting they carried the virus in, per-
haps contaminating water or feed stocks. But
dairy cows in Idaho became sick after com-
ing into contact with cows trucked in from an
area in Texas where the virus was circulating,
leading state officials to believe cow-to-cow
transmission occurred, a spokesperson for
the Idaho State Department of Agriculture
says. In Michigan, too, the virus surfaced af-
ter cows were imported from Texas. Antibody
tests of herds should soon reveal how wide-
spread the virus is and how long it has been
infecting cattle.
In the meantime, Kuiken advises the
United States to restrict the movement of
cows. “I would take the precautionary prin-
ciple and say, ‘OK, until we know what’s
going on, let’s put a standstill on this,’” he
says. But the dairy industry relies on truck-
ing cows south over the winters and then
returning them north, and Armstrong
doubts halting cattle transports would have
much impact if migratory birds are the
main route of transmission. Many bird spe-
cies are currently moving north and may
already be taking the virus with them. “We
have to protect people, but we also have to
… make sure we’re not crippling the indus-
try,” Armstrong says.
Many countries, including the U.S., require
culling of entire poultry flocks if even a single
bird is infected with a highly pathogenic flu
virus. But there is no talk of culling cows,
which cost roughly $2500 each and easily re-
cover from their infection.
Vaccination might be an option, however.
No H5N1 vaccines exist for cattle, but David
Swayne, who formerly ran USDA’s avian in-
fluenza lab, says one could quickly be made
by modifying a swine flu vaccine. Carol
Cardona, an avian influenza specialist and
poultry veterinarian at UM Twin Cities, says
cow vaccination could offer some secondary
PHOTO: LOWER SAXONY STATE OFFICE FOR MONUMENT PRESERVATION
protection to dairy workers. The cattle infec-
tions are “a game changer,” she adds. “I think
it’s all hands on deck.”
The U.S. government stockpiles an H5N1
vaccine for humans but it was made from a
different variant. CDC says it has starting ma-
terial available to produce vaccines against
2.3.4.4b, which studies suggest would pro-
vide “reasonable protection.” Once such shots
exist, “as a protective measure you can imag-
ine vaccinating dairy workers,” Kuiken says.
As to H5N1’s future, Cardona says we
should continue to expect the unexpected.
“The virus is making up new dances,” she
says. “It’s broken the rules on everything.” j
SCIENCE science.org
NE WS
ARCHAEOLOGY
Rare wooden artifacts show the
smarts of early Neanderthals
Complex tools from 300,000-year-old deposit at Schöningen
in Germany point to a “wood age”
By Andrew Curry leave no doubt they were hunting horses.
Excavations at Schöningen 13 ended in
A
bout 300,000 years ago, bands of 2008, although research continues nearby.
early hominins visited the shores of While the spears drew most of the at-
an ancient lake to hunt, sometimes tention, the rest of the wood collected
dropping tools or weapons into the spent decades awaiting analysis, soaking
shallow water and mud. in refrigerated tubs of distilled water to
Fast forward to 1994, when work- replicate the cold, waterlogged soil that
ers at an open-face coal mine in northern preserved it for 300,000 years.
Germany discovered 2-meter-long spruce Starting in 2021, researchers took a close
spears and other wooden artifacts em- look at more than 700 pieces of wood from
bedded in the former lakeshore. By show- the site, from broken spear shafts to twigs.
ing that these ancient hominins—likely They used microscopes and photographed
early Neanderthals—could hunt big game, the wood under carefully angled lights to
not just scavenge meat highlight traces of wear
from the kills of other or cut marks, constantly
predators, the find helped spraying the artifacts with
shift views of our long- water to prevent drying
underestimated cousins. and cracking. “It’s incred-
This week in the Proceed- ibly soft—you can leave
ings of the National Acad- a fingernail mark on it,”
emy of Sciences, researchers Milks says. “It felt like a
present another set of sur- big responsibility.”
prises from the site, called The team eventually
Schöningen 13: a wealth of identified 187 pieces of
other painstakingly carved wood that showed signs
wood remains, including of carving or splitting. Of
throwing sticks and other those, more than 50 were
hunting implements and identifiably tools. Twenty
dozens of nonhunting tools. were related to hunting,
“It’s so much more than we including more spears but
thought,” says University also finely balanced throw-
of Reading archaeologist ing sticks for downing
Annemieke Milks, a co- small game or birds.
author of the new paper. Another 35 pieces were
Many of the wood tools domestic tools: rounded
are far more complex in split sticks probably used
planning and craftsmanship for smoothing animal
than the simple stone tools Cold, waterlogged soil preserved skins, pointed tools for
found at Schöningen and 2-meter spears and other wooden piercing or working hides,
other sites from this time, artifacts for 300,000 years. and shafts that likely served
providing a rare glimpse as handles for axes or stone
of profound cognitive complexity worked blades. Together, the wooden artifacts re-
in wood. “It’s a remarkable discovery,” says veal a domestic side of Neanderthal life
Marie Soressi, an archaeologist at Leiden Uni- that stone tools rarely capture. “We can get
versity who was not involved in the research. overly fixated on the drama of hunting,” says
The site lacks hominin bones, but most Lawrence Barham, an archaeologist at the
scholars assume the toolmakers were early University of Liverpool who was not part of
Neanderthals or their immediate ancestor, the research team. “The nonhunting tools
Homo heidelbergensis, because it coincides add to our understanding of the diversity of
with the earliest evidence for Neanderthals Neanderthal behavior … and help us relate
elsewhere in Europe. Ample animal bones to them: They had to live, and make clothes.”
5 APRIL 2024 • VOL 384 ISSUE 6691 13
NE WS | I N D E P T H
Close analysis of the artifacts also revealed
the early woodworkers made careful deci-
sions and planned ahead at nearly every
step. A reexamination of the spears, for
example, shows that their makers followed
a specific sequence of steps: They stripped
bark, removed branches, sharpened and
smoothed the spears, and hardened them
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
end” of a spear, Barham says. “This study
gives us a really nuanced observation of Ne-
anderthal skill and planning.”
They weren’t using any old sticks, either.
The spears and other tools were carved
from spruce, larch, and pine, species that
grew many kilometers away from the lake
and combined hardness with elasticity.
“They had a very detailed idea what they
were looking for,” says co-author Thomas
Terberger, an archaeologist at the Lower
Saxony State Office for Cultural Heritage.
Signs of breakage and subsequent carv-
ing indicate some broken tools were
recycled—whittled and polished into
smaller implements. “They’re splitting
spears or throwing sticks to make tools for
other tasks,” Milks says.
Although previous research suggested
hominins came to the ancient lakeshore
to hunt and stayed just long enough to
butcher their kills, the tools and wood-
working debris indicate longer stays. Tasks
suggested by the tools such as piercing
hides or working plant material are “activ-
ity you don’t expect at a hunting site,” says
co-author Dirk Leder, an archaeologist
also at Lower Saxony. “That turns it from a
Researchers examined hundreds of pieces of preserved wood to understand the techniques of ancient carvers.
14 5 APRIL 2024 • VOL 384 ISSUE 6691
short-lived butchering site to a campsite.”
The camping parties may have included
more than just burly males. “The throwing
sticks in particular are suitable for use by
other members of the group, even children,”
Terberger says. “Even our image of the hunt
is getting much more colorful.”
The sophisticated wood artifacts from
Schöningen join finds from just a scatter-
ing of other sites, such as 476,000-year-old
wood joists from Zambia published last
year and spears and other tools found else-
where in Europe. Their scarcity hints that
preservation bias distorts archaeologists’
view of the past: Stone tools persist over
the millennia, whereas wood typically de-
cays. Out of thousands of Paleolithic sites,
barely a dozen contain preserved wood.
Yet, “When we compare the wooden tools
at Schöningen to stone, we see more pro-
duction steps, more sophistication, and
more different tool types,” Leder says.
“Maybe stone tools weren’t as important …
as the wood tools.”
Archaeologist Oriol López Bultó of the
Archaeology Museum of Catalonia says the
typical prehistoric campsite or village prob-
ably looked a lot like Schöningen, with most
tools made from organic materials such as
wood, bone, and antler. Even stone points
needed wood handles. “Wood has been es-
sential since the early stages of humankind,”
López Bultó says. “It’s just so rare to find
that preservation.”
The rich toolkit found at Schöningen sug-
gests a change in terminology might be in
order, Leder suggests. “The whole idea of a
Stone Age might be wrong,” he says. “Maybe
we should be talking about a Wood Age.” j
REGULATORY POLICY
Utah flouts
FDA with new
placental stem
cell law
Measure declares that
patients can be given
unapproved treatments
By Meredith Wadman
T
he state of Utah is challenging the
authority of the U.S. Food and Drug
Administration (FDA) with a new, un-
usually bold law that enables patients
to receive unapproved placental stem
cell “therapies.” Observers predict that
the statute, which takes effect on 1 May, could
significantly undermine FDA’s authority to
regulate drugs and other treatments.
The new law, passed almost unanimously
by both houses of the Utah legislature in Feb-
ruary and signed by Governor Spencer Cox
(R) last month, declares that Utah health care
providers “may perform a [placental] stem
cell therapy that is not approved by [FDA]”
so long as they prominently note it is un-
approved and obtain signed patient consent
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]: …
‘Our doctors can do it whether or not you
science.org SCIENCE
 The Utah State Capitol in Salt Lake City, where legislators defied the Food and Drug Administration’s authority on stem cells.
approve of it,’” says Gail Javitt, an attor-
ney who specializes in FDA law at the firm
Hyman, Phelps & McNamara. “This seems
like a pretty blatant broadbrush challenge to
FDA’s authority.”
“The bill appears to blow by safeguards
established by the FDA to protect patients
and research participants” by allowing them
to receive products lacking “convincing—or
possibly any—evidence of safety and effi-
cacy,” adds Leigh Turner, a bioethicist at the
University of California (UC), Irvine. FDA
had not responded by deadline to requests
for comment.
Adult stem cells from tissues such as fat
are already used in unapproved treatments
and in early clinical studies proceeding under
FDA’s purview. (The only stem cell therapies
that the agency has formally approved are
blood-forming stem cells from umbilical cord
blood, used in patients with hematologic dis-
orders.) But the placenta is a richer source
of stem cells, which give rise to a wider va-
riety of tissues—from the liver, pancreas,
and heart to cartilage, bone, blood vessels,
blood, and muscle. Harvested from donated
placentas, the cells are also easier to obtain
than adult stem cells and may be less likely to
induce immune reactions in recipients. They
have shown promise in test tube and animal
studies aimed at conditions including heart
attacks and wound healing. But none has un-
dergone the large, randomized human trials
comparing them with placebos that would be
needed for FDA approval to enter the market.
All stem cell treatments are subject to FDA
PHOTO: AP PHOTO/RICK BOWMER
regulation unless the cells have been “mini-
mally manipulated” after they are extracted
from a donor and are intended to serve the
same function in the recipient that they nor-
mally have in the body, such as bone marrow
transplants. The Utah law flouts FDA’s role by
allowing providers to use placental stem cells
SCIENCE science.org
that have been manipulated and inject them
in any body part for any purpose.
“What’s disappointing to me is that this
unregulated stuff really does give serious
stem cell work, which could be lifesaving and
life-changing to people, a bad name,” says
Diana Farmer, a fetal surgeon at UC Davis
who is running a clinical trial using placen-
tal stem cells to improve outcomes in surgery
done in utero to correct spina bifida.
Using the placenta as source of stem cells
“could be very promising. … But it needs to
be done in a regulated way” to ensure the
product is safe and effective, adds Eunice
Wang, a hematological oncologist at Roswell
Park Comprehensive Cancer Center who was
a principal investigator in a recent clinical
trial that treated leukemia patients with im-
mune cells derived from placental stem cells.
Commercial interests drove the law’s de-
velopment. Haddow says he approached
Bramble while working on behalf of a loosely
affiliated group of Utah companies and clin-
ics. (He declined to name any specific firms
or clinics.)
Another Utah lawmaker, the bill’s House
of Representatives sponsor, Katy Hall (R),
says even before the law was passed, some
of Utah’s dozens of stem cell clinics had been
dosing patients with placental stem cells.
(One establishment, Docere Clinics, run by a
naturopath, advertises “cells and growth fac-
tors from … donated placentas. … Treatments
start at $25,000.”) “This type of treatment is
something that is already being done in the
state and this [bill] gives it better transpar-
ency,” Hall told fellow legislators in February.
Other states that favor more liberal access
to stem cells have haven’t gone as far as Utah.
In the past 7 years, Texas, Mississippi, and
North Carolina enacted laws allowing people
with severe, chronic disease or terminal ill-
nesses to be treated by a physician with adult
stem cell therapies that are already being
tested in human beings—although in Texas,
such trials don’t have to have taken place in
the United States.
The Utah law, by contrast, doesn’t require
a doctor to give the cells, nor does it require
any vetting for manufacturing standards,
safety, cleanliness, or effectiveness. “I’m really
disappointed,” says Kirstin Matthews, a sci-
ence policy expert at Rice University. “I don’t
think they truly realize the potential harms
for the patient: financial, physical, emotional.
… There are quite a few cases of people be-
ing hurt.” For example, in 2017 and 2018,
19 people who were injected with an umbili-
cal cord product marketed as stem cells by
an unregulated supplier were hospitalized for
infection; the injected material, it emerged,
was tainted with fecal bacteria.
To challenge the Utah law, Javitt says, FDA
would likely send a warning letter to a clinic
providing the therapy. Then it would escalate
if necessary by seeking a court injunction to
stop the clinic from distributing the product.
The agency hasn’t shied from such ac-
tion in the past. For instance, in December
2023, it sent a warning letter to a Georgia
company, MiMedx, that markets “placental
collagen matrix” for wound healing, telling
the company the tissue is a drug under FDA’s
definition and needed agency approval for
marketing. And in California and Florida,
it asked for injunctions to bar clinics from
treating patients with stems cells derived
from their fat. It won the ensuing court battle
in Florida, but the California case is still un-
resolved and could ultimately end up before
the U.S. Supreme Court.
Paul Knoepfler, a stem cell scientist at UC
Davis who follows stem cell regulation closely,
says he’s concerned about the spread of laws
similar to Utah’s. “There are other states with
like-minded legislators,” he says. j
5 APRIL 2024 • VOL 384 ISSUE 6691 15
 GREEN POWER
‘Hydrovoltaics’ tap energy
from ubiquitous moisture
New devices convert raindrops, dripping water,
and even humidity into minuscule currents
When evaporation cools its head (blue in a thermal image), a “drinking bird” toy oscillates and generates energy.
By Robert F. Service
S
unlight and wind are the fastest grow-
ing energy sources on the planet. Now
imagine drawing power from a third,
even more plentiful green source:
moisture in the air. That’s the vision
of Jun Yao, an applied physicist at the
University of Massachusetts Amherst who
has devised a porous film that converts the
charges naturally present in water vapor into
power. To be clear, Yao’s films are the size
of a postage stamp and only generate a tiny
current. But Yao is just getting started. And
he’s not alone. Teams around the globe are
creating a bevy of novel “hydrovoltaic” de-
vices able to convert the energy inherent in
evaporation, rainfall, and small water flows
into usable energy.
“It’s a really burgeoning field,” Yao says.
But so far, these devices have mostly been
one-off lab demonstrations. “Right now, it’s
not possible for hydrovoltaics to produce a
large amount of power,” says Yuanyuan Li
of the KTH Royal Institute of Technology
in Sweden, who has developed caramel-size
hydrovoltaics that can power a digital clock.
To generate abundant power, researchers
will need to enlarge the devices into square-
meter-size modules that could be deployed in
vast arrays like solar farms.
But hydrovoltaics have an advantage: Un-
like wind and sunshine, which are intermit-
tent, moisture is omnipresent, promising
nonstop output. “We should definitely ex-
16 5 APRIL 2024 • VOL 384 ISSUE 6691
plore this means of generating electricity,”
says Stefan Weber, a physicist who studies
hydrovoltaics at the University of Stuttgart.
Humans have tapped flowing water as a
power source for millennia, of course, from
water wheels in ancient China and Greece to
modern hydroelectric dams. Hydrovoltaics,
by contrast, generate electricity based on how
water interacts with materials. Ten years ago,
researchers noticed that droplets striking or
flowing across charged surfaces could create
ultrashort voltage spikes, resulting from ei-
ther their kinetic energy or the movement of
ions they carry. Although those spikes were
small, more recently Weber’s group showed
that showers of droplets could generate a
whopping 1200 volts, albeit still for only an
instant, just long enough to cause banks of
light-emitting diodes to flicker.
Evaporation can generate power, too, as
Zuankai Wang demonstrated with a toy.
Last month in the journal Device, Wang, a
mechanical engineer at Hong Kong Poly-
technic University, and his colleagues de-
scribed how they put a pivoting “drinking
bird” to work as a clean-energy generator.
Invented decades ago in China, the toy
starts to oscillate when its absorbent “beak”
is dipped in water. After it pivots back up-
right, the water evaporates and cools its
head. The resulting pressure drop draws a
different liquid from its base into a reservoir
in the head, causing the bird to bow and
dip its beak again. Then that liquid flows
back to the base, tipping the bird upright—
until water evaporation triggers the next
cycle. As the bird oscillates, “triboelectric”
generators convert its motion into electricity.
“This is a very clever way to do energy
conversion,” Yao says. Wang’s group is now
building arrays of drinking bird generators
that could work together to provide continu-
ous power as long as the bird’s drinking water
is replenished.
Yao’s air generator is also aiming for
continuous power delivery. That device, re-
ported last year, consists of a thin sheet of a
material such as silk fibers or a conducting
polymer, shot through with tiny pores less
than 100 nanometers in diameter. The sheet
sits on top of a glass substrate that seals its
base. Water molecules in the air naturally
harbor negatively charged ions. As they
bump into the top of the nanopores and shed
those ions, they create a gradient of electrical
charges that are then captured by electrodes
sitting above and below the films.
Like other hydrovoltaics, Yao’s device
only produces microwatts of power per
square centimeter. One way to scale it up,
he says, might be to stack multiple layers of
materials, with air spaces in between, turn-
ing his generator into a 3D device that can
continuously capture electricity from the
air. “Water vapor in the air is a huge energy
reservoir,” Yao says.
Another hydrovoltaic system that may
scale up more easily relies on wood, a natu-
ral marvel of nanoengineering made up of
cellulose, hemicellulose, and lignin, which
combine to form myriad minuscule verti-
cal channels that transport water from
the ground to the leaves. Researchers have
known for years that they can use wood’s
structure to generate minute amounts of
electricity, simply by placing a slice of wood
in a dish of water, with the upper surface
exposed to the air. Evaporation from the
top pulls up more water, and ions within it,
through the channels, generating a minute
but steady current.
Recently, Li’s team reported that it could
increase the current output from a sliver of
balsa wood 10-fold by first treating it with so-
dium hydroxide, or lye. The chemical breaks PHOTO: H. WU ET AL., CELL PRESS (2024) 10.1016/J.DEVICE.2024.100318
down some of the channel walls, exposing
charged chemical groups in the wood’s in-
terior. And last year, Li’s team showed that
wood infused with iron oxide nanoparticles
could generate 52 microwatts of power per
square meter, 165 times more than natu-
ral wood, and enough for six 1.5-square-
centimeter pieces to power a digital timer.
That’s nothing like the 200 to 300 watts
per square meter generated by silicon solar
cells. But the field is just getting started.
Wang says: “People are thinking of all
kinds of new ways to tap energy that’s all
around us.” j
science.org SCIENCE

SCIENCE FUNDING
U.K. boosts digitization of museum specimens
London’s Natural History Museum gets nearly $200 million to scan country’s collections
By Elizabeth Pennisi
T
ucked away in drawers and cabinets
in institutions around the world may
be answers to how our planet formed,
how life evolved and interacts, and
how resilient it may be in the future.
But those collections—millions of
ancient rocks and fossils, pressed plants,
pinned insects, and other specimens—can
only yield insights if researchers around the
world can access them.
Now, efforts to digitize collections, mak-
ing them accessible to all, have received a
major boost. Last week, the U.K. government
announced that, beginning in 2026, it will
provide £155 million (almost $200 million)
over 10 years for the Natural His-
tory Museum (NHM) in London
to put the majority of the coun-
try’s natural history collections
online. “It’s really going to make
a huge difference in terms of
what the NHM can do,” says Gil
Nelson, a scientist at the Univer-
sity of Florida who runs Integrated
Digitized Biocollections (iDigBio),
a database of almost 140 million
U.S. specimens.
The U.K. funding should also
accelerate large-scale digitization
efforts in Europe, by consolidating
collaborations. But in the United
States, Nelson and other research-
ers fear their projects could be left
behind if the government doesn’t
ante up a similar bolus of funding. A robotic arm may help London’s Natural History Museum scan its insects.
Helen Hardy, who runs NHM’s
digitization efforts, says she “is super ex-
cited” by the backing, announced last week
by UK Research and Innovation. She and
other U.K. researchers laid the groundwork
in 2021 with an analysis estimating that
digitizing the nation’s collections could save
conservation programs, crop and drug devel-
opment, and other efforts £2 billion over the
next 30 years. We made a “convincing case”
PHOTO: TRUSTEES OF THE NHM LONDON
for the funding, Hardy says.
At the minimum, the NHM effort, like
others around the world, expects to digitally
record all the details from specimen labels—
such as names and when and where some-
thing was collected. Researchers also want
to include images detailed enough to make
examining the specimens in person less es-
sential. Ultimately, with the help of artificial
SCIENCE science.org
intelligence (AI) programs, they would like
to create “extended digital specimens” linked
to details of the environment and ecosystem
in which the organism lived.
Such efforts are already having an impact.
In 2016, for example, Kansas and Egyptian
researchers combined hundreds of geo-
located digitized records of Culex quinque-
fasciatus, a mosquito that spreads West
Nile and other diseases, with environmental
data. The joined information allowed them
to build a computer program that predicted
the insect might exist undetected in parts
of North Africa and Western Europe—and
could eventually thrive in Australia.
Sixteen years ago, NHM data scientist
Vincent Smith and colleagues estimated it
would take 1500 years to digitize the mu-
seum’s 80 million specimens. Thanks to
advances in technology, the timescale has
collapsed, and the expense is less daunting.
The cost of digitally capturing specimens,
from a blue whale skeleton to a dried mos-
quito, “can vary from a few cents to up
to €20 per object,” says Jana Hoffmann,
who oversees such efforts at Berlin’s Natu-
ral History Museum. Plants are still easi-
est, as they are typically pressed flat onto
sheets and can be photographed assembly-
line style, with their labels visible. Insects,
however, present a formidable challenge—
most need to be unpinned to digitize a la-
bel underneath.
The U.K. funding should help boost an
EU-funded effort called the Distributed
NE WS | I N D E P T H
System of Scientific Collections (DiSSCo),
which Smith helped found almost a de-
cade ago. It hopes to help put online an
estimated 1.5 billion specimens, including
U.K. species, belonging to 170 museums,
universities, and botanical gardens across
23 countries. The project “will allow the de-
livery of scientific knowledge at scale,” says
Dimitris Koureas, a systematist-turned-bio-
informatician at the Naturalis Biodiversity
Center in the Netherlands, which oversees
DiSSCo. With the new money, the United
Kingdom is poised to be a major contributor.
The U.K. announcement calls for NHM to
help about 90 other U.K. institutions digitize
their collections as well, which total about
137 million objects. “It’s the local collections
that have the depth of knowledge
about the local flora and fauna”
to help guide conservation and
restoration efforts, says iDigBio’s
Elizabeth Ellwood.
Hardy expects that more than
60% of the new money will go
directly to digitizing specimens,
and the rest will help NHM im-
prove relevant technologies and
infrastructure. The effort, dubbed
DiSSCo UK, “will create a culture
of digitization and mechanisms
so all museums can get the ball
rolling,” says Kirk Johnson, di-
rector of the Smithsonian In-
stitution’s National Museum of
Natural History.
U.S. efforts helped inspire
the U.K. and European projects,
Smith says. “iDigBio was, and re-
mains, pioneering in many respects,” he says.
Both the National Science Foundation (NSF)
and the National Academies of Sciences,
Engineering, and Medicine have urged that
efforts be expanded. And in 2022, a newly
passed law called for NSF to establish an
Action Center for Biological Collections to
coordinate further U.S. digitization. The NSF
center “should leverage the momentum we
have,” Nelson says. “But there has not been
any funding allocation to make that happen.”
That does not bode well, says evolution-
ary biologist Scott Edwards, a curator at
Harvard University’s Museum of Compara-
tive Zoology. “Without another substantial
commitment [in the U.S.], our earlier efforts
could falter, and the full potential of natural
history collections won’t be realized.” j
5 APRIL 2024 • VOL 384 ISSUE 6691 17
NE WS | I N D E P T H
SCHOLARLY PUBLISHING
Gates foundation places bold bet on preprints
Funder to mandate preprints and no longer pay open-access publishing fees for grantees
By Jeffrey Brainard
T
he world’s largest philanthropy, the
Bill & Melinda Gates Foundation,
last week took a radical step aimed
at giving preprints—freely available
draft manuscripts that have not been
peer reviewed—a much more promi-
nent role in science. Starting in 2025, the
foundation will require grantees to post as
preprints all manuscripts that result from
research it funds. It will also stop paying for
researchers to publish their papers in jour-
nals that charge a fee to make papers free.
The foundation says the mandate is
needed to accelerate the dissemination of
research findings and because too many au-
thors cannot afford publishing fees. It also
wants other science funders to follow suit.
The policy shift has drawn praise from
some advocates of immediate free access to
research results. But others note preprints
lack peer review, and they fear that such
policies, if widely adopted, could promote
the spread of poor-quality research. Some
journal publishers could also see revenues
fall if major funders refuse to pay the hefty
article-processing charges (APCs) levied by
some open-access journals.
The new policy, announced on 27 March,
makes the $67 billion foundation the
wealthiest major research funder to specifi-
cally mandate the use of preprints. (Another
foundation, the Chan Zuckerberg Initiative,
instituted a similar policy in 2017.) Under
a previous 2015 policy, the Gates founda-
tion had required grantees to make re-
search publications immediately available
for free. In practice, that often meant pay-
ing an APC of $2000 or more to publish
in an open-access journal. Currently, the
Gates foundation spends $6 million annu-
ally to cover those costs, says Ashley Farley,
the foundation’s open-access lead. Gates
foundation–funded grantees publish about
4000 journal articles annually, a fraction of
the global total of more than 2.5 million.
In an online Q&A, the foundation ac-
knowledges that many of its grantees will
still want to ultimately publish preprinted
manuscripts in peer-reviewed journals be-
cause tenure and promotion reviews re-
quire it. The new policy allows grantees
to use non–Gates foundation funds to pay
APCs or to publish in a subscription journal
that charges readers but not authors.
18 5 APRIL 2024 • VOL 384 ISSUE 6691
The foundation says the move to pre-
prints will allow researchers to share re-
search results as soon as they’re ready, and
not wait weeks or months for journals to
complete their review processes. The policy
could also encourage authors and journals
to publish only their best manuscripts in
journals, reducing the workload for peer
reviewers, who are volunteers. The phi-
lanthropy says its policy avoids pitfalls of
the APC business model, which has been
blamed for incentivizing journals to churn
out papers of limited value and supporting
predatory journals that publish papers with
no peer review at all.
“In the last decade, we’ve seen a lot of
what isn’t working [in publishing],” Farley
says. “This [policy] will hopefully free us to
really pilot and support what we are seeing
is beginning to work.” Those models could
“The jury’s still a little
bit out on preprints, but it’s
certainly a place of great
innovation and dynamism.”
Alondra Nelson,
Institute for Advanced Study
include services that provide rapid peer re-
view of preprints and “diamond” journals
that are free to both readers and authors.
The new policy was welcomed by Coali-
tion S, a group of funders based mostly in
Europe that since 2021 has required grant-
ees to make funded papers immediately free
to read. The group stopped short of saying
it would copy the Gates foundation man-
date. It did note, however, that last year it
released a draft policy that encourages (but
does not require) researchers to preprint
their manuscripts and aims to support non-
APC business models.
The Howard Hughes Medical Institute
(HHMI), another large philanthropic re-
search funder that belongs to Coalition S,
struck a similar tone. Bodo Stern, the insti-
tute’s chief of strategic initiatives, notes that
in 2023 about half of research articles by
HHMI-funded investigators first appeared
as a preprint. In the future, Stern says, “We
hope that preprints will become a default
first step for publishing HHMI research,
even without a mandate.”
Kent Anderson, a scholarly publishing
consultant and longtime critic of the open-
access movement, says promoting preprints
would be a mistake. Without a robust peer-
review system, a proliferation of preprints
could lead to researchers and the public
becoming “even more confused and con-
founded about what constitutes reliable sci-
entific findings,” he wrote on the Geyser, his
online newsletter. He also doubts preprint
servers will receive enough funding to con-
duct meaningful peer review. Screening and
vetting manuscripts “are just costly head-
aches,” he wrote, “and the expense level
needed for these would far outstrip the pal-
try grants [preprint servers] receive.”
At the publisher Frontiers, which pro-
duces only open-access journals that
charge authors or their institutions, Tom
Ciavarella, the company’s head of public af-
fairs and advocacy for North America, said
in a statement that preprints can benefit
from peer review. He added that the new
policy does not advance the Gates founda-
tion’s earlier goal, “to stop financial models
that rely on closed, paywalled access.”
Several publishing platforms—including
eLife and PREreview—have experimented
with providing peer review of preprints
after they are posted, for a fee or for free.
But these services have achieved limited
volume, conducting reviews of just a few
thousand preprints. And, overall, relatively
few researchers outside of physics have
fully embraced posting preprints, although
the practice did gain popularity among
some biomedical researchers during the
COVID-19 pandemic. As of 2022, preprints
represented 7% of all articles in the U.S. Na-
tional Institutes of Health’s PubMed data-
base, up from 0.2% in 2015.
“The jury’s still a little bit out on pre-
prints, but it’s certainly a place of great
innovation and dynamism,” says Alondra
Nelson, a social scientist at the Institute for
Advanced Study who led the White House
Office of Science and Technology Policy
when it released new guidelines that, start-
ing in December 2025, will require research-
ers funded by the U.S. government to make
their manuscripts freely available. But, she
adds, “It’s clear that we need to reimagine
peer review in some ways,” in part because
the steady growth of published research pa-
pers is taxing the researchers who volunteer
to conduct the reviews. j
science.org SCIENCE
 2023 Winner
Marissa Scavuzzo, Ph.D.
Case Western Reserve
University School of
Medicine, Cleveland, USA
For research on glial
cells in the gut
Call for Entries 2024
Eppendorf & Science Prize for Neurobiology
The annual Eppendorf & Science Prize for
Neurobiology is an international prize which honors
young scientists for outstanding neurobiological
research based on methods of molecular, cellular,
systems, or organismic biology. If you are 35 years
of age or younger and doing great research, now is
the time to submit an entry for this prize.
It’s easy to apply! Write a 1,000-word essay and
tell the world about your work.
eppendorf.com/prize
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.
All rights reserved, including graphics and images. Copyright © 2024 by Eppendorf SE. Photography: Lisa Helfert
Application Deadline
June 15, 2024
As the winner, you could be next to receive
> Prize money of US$25,000
> Publication of your work in Science
> Full support to attend the Prize Ceremony held
in conjunction with the Annual Meeting of the
Society for Neuroscience in the USA
> 10-year AAAS membership and online
subscription to Science
> Complimentary products worth US$1,000 from
Eppendorf
> An invitation to visit Eppendorf in Hamburg,
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-

petition for attention and status from her er
extended family.
Five decades of research have brought ht
Lucy to life, showing how she walked up- p-
right even though she still had a small ll
brain and apelike upper body, which prob- b-
ably allowed her to climb trees to feed, nest,
t,
or escape predators. More than 400 newer er
fossils of males and females, as well as thehe
Dikika child, have revealed how her species, s,
Australopithecus afarensis, grew, social- l-
ized, and evolved during its million-year ar
span on the planet, from perhaps 3.85 mil- l-
lion to 2.95 million years ago.
But researchers’ view of Lucy’s world and d
her place in it has changed. She is no lon- n-
ger the earliest known member of the hu- u-
man family. Members of her species didn’t
take their first upright steps in open savanna
grasslands, as her discoverers thought, but
walked first in a grassy woodland with de-
ciduous trees. She and her kind weathered
climate change, adapting to different habi-
tats over the millennia. Most important,
she wasn’t alone in the landscape. “We have
multiple [hominin] species in the same
time period,” says Yohannes Haile-Selassie,
director of the Institute of Human Origins
(IHO) at Arizona State University, who is co-
organizing a symposium this week on Lucy’s
impact on human origins research.
Haile-Selassie and some others think that
3 million to 4 million years ago, the human
family tree was more like a bush than a
bonsai, with multiple stems growing side by
side rather than a single trunk. He and oth-
ers now see Lucy as more of a great-great-
great-aunt than a direct human ancestor.
But Alemseged and others point out that so
far, no other fossil is a better candidate for
being the mother of us all.
Even as paleoanthropologists debate
her place in the human family tree, they
agree no known human ancestor has had
PHOTOS: (TOP TO BOTTOM) JOHN GURCHE; EVAN VUCCI/AP; (OPPOSITE PAGE) BRUCE FRUMKER
the impact of Lucy. They still marvel at the
detailed view of our past revealed by her
skeleton. Forty percent complete, it has
served as a template for fitting together
the isolated bones of dozens of other mem-
bers of her species, like pieces in an in-
complete puzzle. “More people have come to
look and analyze Lucy’s skeleton than any
other fossil in the world,” says IHO paleo-
anthropologist Kaye Reed.
Back in 1974, researchers didn’t realize
just how rare such a well-preserved hominin
skeleton would turn out to be. “Fate played
a miserable trick—it gave us the best fossil
early on,” says paleoanthropologist Bernard
Wood of George Washington University.
“It’s as if it’s my birthday and I’ve opened
my first present and it’s exactly what I want
… so you think that all the other presents
will be as good as this one.”
SCIENCE science.org
LUCY THE HOMININ
THAT FOSSIL GIFT of a lifetime was first un-
wrapped by Don Johanson. Then a young
American paleoanthropologist at the Cleve-
land Museum of Natural History, he had
joined an expedition at Hadar organized
by late French geologist Maurice Taieb. On
24 November 1974, halfway through the
second field season, Johanson and student
Tom Gray were walking back to their Land
Rover after a discouraging morning when
they had found little of interest. Then,
Johanson saw a bit of Lucy’s arm bone on
a hill in a dry gully. Next he spotted a skull
piece, a thighbone, part of a pelvis, and ver-
tebrae: a rare partial skeleton.
The team celebrated the discovery in
camp, drinking beer and blasting the Bea-
tles’s song Lucy in the Sky with Diamonds
all night. Someone started to call the skel-
eton Lucy, and the name stuck.
“The real importance of Lucy is we didn’t
know what an early human ancestor looked
Zeresenay Alemseged (left) showed Lucy’s
jaw to then-President Barack Obama in Addis
Ababa, Ethiopia, in 2015.
NE WS
AGE: 3.18 million years old
SPECIES: Australopithecus afarensis
SPECIES
SIZE: 1 meter
m tall, 30 kilograms
FAMILY: All of Homo and Australopithecus, but
it’s complicated
comp
HOMETOWN:
HOMETO Hadar in Ethiopia. Now resides at
the National
Natio Museum of Ethiopia in Addis Ababa
LIKES: Wooded
W grasslands, for walking upright
and climbing
climb trees to escape predators. Enjoys
figs, berries,
berrie sedge grasses
KNOWN FOR: Good bones; well-preserved
skeleton includes jaw, ribs, pelvis, and parts of
arms and leg
DISCOVERED:
DISCOV 24 November 1974
NAMED AFTER: A Beatles song
like,” says Johanson, founder of IHO. He
recalls that he was immediately impressed
by how small and primitive Lucy was com-
pared with known members of Australo-
pithecus, such as A. africanus, a hominin
found in South Africa almost a century ago,
and thought at the time of Lucy’s discovery
to be about 2 million years old. “I was struck
by her more apelike features,” he says.
Lucy was the first hominin to break the
3-million-year time barrier, pushing back
the age of the human family to a time closer
to when geneticists thought the ancestor of
humans had split from the ancestor of chim-
panzees. Lucy “entered a field actively debat-
ing the antiquity of the chimp/human split,”
says Tim White, a paleoanthropologist at the
University of California, Berkeley who co-
authored the landmark 1978 paper describ-
ing Lucy’s species with Johanson and the late
paleoanthropologist Yves Coppens.
For the first 20 years after Lucy’s discov-
ery, her species was the oldest known mem-
ber of the human family. A. afarensis was
“the only game in town” between 3 million
and 4 million years ago, says Carol Ward,
a paleoanthropologist at the University of
Missouri. Many researchers concluded that
her species gave rise to all the hominins
that came later, including Homo, A. africa-
nus, and more robust members of the hu-
man family such Paranthropus aethiopicus,
P. robustus, and P. boisei.
“Once upon a time, life was relatively
simple because anything before 3 million
years was A. afarensis … [and] it was the
mother of us all,” says paleontologist Fred
Spoor of the Natural History Museum in
London. “That was the gospel.”
Since then, however, other species have
emerged from the shadows, some of them
much older. Lucy was so apelike that “the
implicit hypothesis was that the next step
back would be a chimpanzee,” says White,
5 APRIL 2024 • VOL 384 ISSUE 6691 21
The day after he found Lucy in November 1974,
Don Johanson studies the exact spot where
he picked up the first bone, her elbow. In camp,
Johanson (bottom, left) and Maurice Taieb
scrutinize and photograph the spectacular find.
 NE WS | F E AT U R E S

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

ent upright walker from Ke- skull, however, Kenyanthro-

nya called Millennium Man,
or Orrorin tugenensis; and
a stunning skull of Sahelan-
thropus tchadensis, dated
6 million to 7 million years
old, from Chad.
Anthropologists still fierce-
ly debate whether all these
species are hominins, and Hadar
how they relate to Aus- (Lucy)
tralopithecus, not to men-
tion Homo. But the fossils
clearly push back the ori-
gin of the human family
to at least 6 million years
ago, a date that dovetails
with the most recent ge- 20 km
m Meanwhile, Haile-Selassie
netic evidence for the tim-
ing of the split between our
lineage and that of chimps and bonobos.
Meanwhile, other researchers kept seek-
ing fossils in the key period about 4 mil-
lion years ago—and found what seemed to
be a great-grandmother for Lucy. In 1994,
Kenyan paleoanthropologist Meave Leakey
and her team found more than 80 fossils of
teeth, jaws, a partial arm, and shinbone at
two sites near Lake Turkana in Kenya. They
dated the fossils to between 3.9 million and
4.2 million years ago, and named the spe-
cies Australopithecus anamensis. Based on
features of its teeth and jaw, they proposed
the new species was the direct ancestor of
Lucy and her kin.
Then in 2016, in a hilly area just 30 kilo-
meters from Hadar called Woranso-Mille,
Red
d pus got a mixed reception.
Sea White called it roadkill and
suggested it was a smashed
See inset Gulff of
Gul of A
Aden
den A. afarensis. More recently,
detail left Spoor and Leakey applied
ETHIOPIA
3D morphometric methods
to compare Kenyanthro-
Lomekwi pus’s upper jaw with those
of a half-dozen other homi-
nins, and convinced many
Lake
Turkana researchers to take it seri-
KENYA ously as a separate species.
Indian Ocean Researchers identified ad-
Laetoli
Dikika ditional isolated teeth of the
(A. afarensis) species at Lomekwi, but the
TANZANIA 250 km species is still poorly known.
found two additional homi-
nins at his site of Woranso-
to A. afarensis, and the two lineages co- Mille that further tangle the human family
existed for a time. tree just before Homo appeared. One, the Bur-
Even as they sought Lucy’s ancestors, tele foot, named after the 3.4-million-year-
paleoanthropologists also kept searching old layer of sediment in which it was found,
for her descendants, looking for a path from lived at the same time as A. afarensis but is
her genus to our own. more primitive, with an opposable big toe
like that of tree-climbing Ardipithecus. His
IDENTIFYING THE DIRECT ANCESTOR of Homo team also found teeth and two upper and two
is challenging, Ward says, in part because lower jaws he attributes to a new species he
“we don’t know much about early Homo.” named A. deyiremeda that lived 3.3 million to
The earliest known fossil of our genus is a 3.5 million years ago. When he published the
lower jaw with worn molars, dating to al- discovery in 2015, reaction was mixed; White,
most 2.8 million years ago and found at the Alemseged, and others thought the jaws were
desolate site of Ledi-Geraru, only 30 kilo- diverse forms of A. afarensis, which was
meters from Hadar (see map, above). “It’s found within 5 kilometers of the site.
a convincing case,” Ward says, “but it’s just But Spoor and his postdoc recently backed
one mandible until about 2 million years Haile-Selassie’s claim. They did a 3D mor-

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 23

Yohannes Haile-Selassie unveils a replica of the 3.9-million-year-old skull of Australopithecus anamensis, which many think is the ancestor of Lucy’s species.
phometric analysis of developmental differ-
ences in the upper jaw bones of A. afarensis,
A. deyiremeda, and K. platyops, highlighting
the methodological advances in the field
since Lucy was found. The comparisons
“demonstrate significant differences,” Spoor
argued in a talk at the College of France in
June 2023. “It confirms the status of K. platy-
ops and A. deyiremeda as valid species differ-
ent from A. afarensis.”
No one is arguing that A. dey-
iremeda is a closer ancestor to
Homo than Lucy. But the new
fossils suggest a burst of diver-
sity 3.5 million years ago, crowd-
ing the stage on which Lucy once
stood alone. She may have been
part of an adaptive radiation that
happened earlier than research-
ers once thought, perhaps after
hominins began to walk upright
and expanded their habitats and
diets, Ward says.
Some of those hominins may
have trod the same ground as
Lucy’s species. In 2020, paleo-
anthropologist Charles Musiba
of the University of Colorado
Denver and his colleagues reana-
lyzed famous tracks discovered in
1976 at Laetoli. Some prints are Lucy is a national treasure in Ethiopia, where students and others often visit
thought to be the footprints of
24 5 APRIL 2024 • VOL 384 ISSUE 6691
A. afarensis about 3.7 million years ago.
Other tracks, about the same age, were
believed to have been made by animals.
Musiba and his colleagues, however, pro-
posed that the prints were made by upright
walkers—a hominin with different feet and
gait from Lucy’s species.
Meanwhile, another clue in the search
for ancestors has emerged in South Africa,
where researchers have discovered other
her skeleton at the National Museum of Ethiopia in Addis Ababa.
early members of Australopithecus, such as
A. sediba, dating to about 2 million years
ago and possibly earlier. A new analysis in
the Journal of Human Evolution last year
could not rule out that A. sediba shared an
ancestor that also gave rise to early Homo,
raising the possibility that our ancestor
ranged from eastern Africa to South Africa.
At the moment, none of the new homi-
nins is convincing as a direct forebear of
Homo. Alemseged and some oth-
ers still put their money on Lucy:
They point out that no fossils
unequivocally bump her species
from its prime spot as the putative
ancestor of early Homo and later
members of Australopithecus.
Others say we just don’t know.
“It’s too simplistic to create
PHOTOS: (TOP TO BOTTOM) SETH WENIG/AP; IMAGO/ALAMY
ancestor-descendants out of cur-
rently known species from an
incomplete fossil record,” Spoor
says. Given the rarity of fossil-
ization for hominin remains and
how few have been found, we
might never catch a glimpse of
our immediate direct ancestor,
Wood says. It’s unlikely that “all
the hominins that ever lived occu-
pied the circa 1% of the land sur-
face of Africa where we happen to
find fossils,” he says.
science.org SCIENCE

LUCY’S STATUS as a human ancestor may be
uncertain, but in other ways she is com-
ing into sharper focus. “When Lucy was
found, we thought Homo habilis was the
one who made the earliest tools,” says
archaeologist Sonia Harmand of the
French national research agency CNRS.
But at Lomekwi, where Kenyanthropus
was found, Harmand and her team un-
earthed crude stone tools from 3.3 million
years ago, early enough for either Kenyan-
thropus or an Australopithecus to have
hafted them. Lucy’s brain was only a bit
bigger than an ape’s relative to her body
size, but her hand and that of other mem-
bers of Australopithecus was probably ca-
pable of smashing bones for marrow with
crude tools, Harmand says.
“We don’t know if Lucy was a tool-
maker,” Harmand emphasizes. “Unfortu-
nately for her, she has a lot of competition”
from other hominins alive at the time. So
far the team has reported only a single,
strange-looking molar found with the
tools, although other teeth and fossils may
be published soon and could clarify the
toolmaker’s identity.
Others have suggested Lucy’s kind
could have made tools. Back in 2010,
Alemseged and Shannon McPherron of
the Max Planck Institute for Evolutionary
Anthropology wrote in Nature that a rib and
CREDITS: (GRPAHIC) D. AN-PHAM/SCIENCE; (DATA) F. SPOOR, NATURE, 10.1038/521432A (2015); (PHOTO) EALISA WESTERHOFF/AFP VIA GETTY IMAGES
a shaft of an ungulate leg bone bear “un-
ambiguous evidence” of stone tool marks,
made 3.4 million years ago by a member
of the human family. The butchered bones
were found at Dikika, just 222 meters from
the spot where Alemseged discovered the
remarkably complete Dikika child. White
and others, however, argue the cut marks
could be the work of crocodile teeth rather
than stone tools. And no tools have been
found at Laetoli or Hadar, where most of
the hundreds of A. afarensis fossils have
been found, says paleoanthropologist Terry
Harrison of New York University.
Still, Lucy’s adaptability is clear from
studies of the ancient environments she
lived in for hundreds of thousands of years.
Teams of researchers are identifying an-
cient animals from bones and proteins and
studying pollen, plant waxes, and isotopes
to reconstruct vegetation and rainfall.
They’ve found that when the earliest A.
afarensis lived at Hadar about 3.7 million
years ago, the site included woods along
the shore of an ancient river, where ani-
mals waded into floodwaters and an entire
family of A. afarensis died together.
Later, after 3.2 million years ago, Ha-
dar was much drier, and Lucy’s species
adapted to a new diet with tougher foods
such as sedges, roots, and grasses; appar-
ently in response, male members of A. afa-
SCIENCE science.org
NE WS | F E AT U R E S
Meet the family
Between 2 million and 3 million years ago, many kinds of hominins walked the same African landscapes as Lucy’s
species, Australopithecus afarensis. She was once considered the ancestor of everything that came after her,
including Homo. But new fossils have emerged as competitors, and one, A. anamensis, as a potential ancestor.
To infer most of the other evolutionary relationships among early hominins, researchers will need new fossils.
Lucy’s species is still a
prime candidate for the
Kenyanthropus platyops A. garhi
ancestor of Homo.
A. deyiremeda
A. bahrelghazali
A. africanus
A. anamensis Australopithecus afarensis
(Lucy)
Homo species
Burtele foot
Paranthropus species
4 3 2
Millions of years ago
Zeresenay Alemseged
cradles the skull of “Lucy’s
child,” a toddler of her
species he found at Dikika
in Ethiopia.
rensis evolved bigger jaws. Farther south different hominins to coexist in different
at Laetoli the habitat was even drier—a niches. He was back in the field in Febru-
“marginal environment” without a stream ary, seeking more bones and tools.
when A. afarensis was there, says IHO Yet even as he searches for fossils to sup-
paleoanthropologist Denise Su. “Basically, plant her, he and others say Lucy’s reign
afarensis was a generalist,” able to survive continues. “I don’t think there has been
in a host of environments, Reed says. anything that has successfully displaced
Lucy’s species was the only known hom- Lucy,” Wood says. “That doesn’t mean she
inin at Hadar. Yet only 30 kilometers away was the ancestor of Homo. But she’s still
at Woranso-Mille, it shared the steeper, the best candidate.”
more wooded terrain with A. anamensis That’s good news for Johanson, now
and A. deyiremeda, as well as the owner 80 and just back from Ethiopia, where he
of the Burtele foot. Reed and Haile-Selassie found many young people still eager to take
aim to figure out why one site has so many selfies with Lucy’s discoverer. Wherever Lucy
hominins and the other only one. Haile- ultimately lands in the human family tree,
Selassie thinks the greater diversity of hab- Johanson has only one regret: He didn’t get
itat at Woranso-Mille may have allowed to introduce her to the Beatles. j
5 APRIL 2024 • VOL 384 ISSUE 6691 25
 INSIGHTS
LET TERS

NEXTGEN VOICES the secrets of these mutations, including

how to repair them and even how to pre-

Research beneficiaries speak

We gave young scientists this prompt: Imagine that you meet all of
vent them. This discovery has allowed us to
produce more energy than ever. Our cellular
companions are healthier, and our entire
host body is stronger.
your research goals. Describe the impact of your research from Yuzhou Gao
the perspective of a person, animal, plant, place, object, or entity First School of Clinical Medicine, Anhui Medical
University Hefei, Anhui, China.
that has benefited from your success. You can read a selection Email: gaoyuzhou7@163.com

of their responses here. Follow NextGen Voices on social media with

hashtag #NextGenSci. —Jennifer Sills
Medical advances
I am the humble pill, once a lonely warrior
in the vast landscape of pharmaceuticals.
My journey was fraught with perilous side
effects and insufficient efficacy, leaving
combating ailments with unparalleled
precision and potency.
Qian Ye
Anhui, China. Email: yeqian7059@126.com
I am a myelin sheath, protector of nerve
fibers and guardian of conductivity in
the vast nervous system. Ten years ago,
my human caught the Epstein-Barr virus
(EBV), and I have lived every moment
since in fear of EBV and the serious
disease with which it has been associated,
multiple sclerosis (MS). I’ve watched as
MS has degraded many of my friends,
ILLUSTRATION: ROBERT NEUBECKER

many a patient disheartened. But now, with
the advent of pharmaceutical co-crystals
that allow my contents to be absorbed
more easily, I am no longer a solitary agent.
Instead, I am part of a mighty alliance,
joined with other compounds in a harmo-
nious crystal lattice. Together, we unleash
unprecedented therapeutic prowess,
I am a mitochondrion, residing in a cell. I
provide the energy that keeps life going in
my little world. Until recently, my fellow
mitochondria and I were plagued by enig-
matic mutations that disrupted our work,
reduced our energy output, and caused
some of our cellular companions to fall ill.
Luckily, researchers have now unraveled
leaving just a few of us to serve a large
network of neurons. With each action
potential I protected, I cringed, uncertain
if it would be my last. I knew that the
entire nervous system was depending on
me. But now, researchers have designed
EBV treatments to alleviate MS. The barren
axons that once surrounded me are more

26 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE


insulated than ever, and I’m finally at ease.
Rishi Jai Patel
Vagelos Molecular Life Sciences Program,
University of Pennsylvania, Philadelphia, PA, USA.
Email: ripatel@sas.upenn.edu
I am a renal tubular cell in a person hospi-
talized with renal failure. I was feeling weak
and hopeless, but three-dimensional organ
printing research saved me! One afternoon,
I was sucked out of my person and injected
into a tube. That evening, I was transferred
to a medium in a lab upstairs. There, I
was cloned, along with other types of cells
including podocytes, mesangial cells, and
vascular cells. Then, my partner cells and I
were collected and lined up in a bioprinter
to print a new kidney. The next morning, I
(now part of the new kidney) was trans-
planted to my home, the bed-bound person
I had left the day before. I was rejuvenated,
and so was my person.
Bo Cao
Core Research Laboratory, The Second Affiliated
Hospital of Xi’an Jiaotong University, Xi’an,
Shaanxi, China. Email: bocao@vip.qq.com
I am degraded knee cartilage, battered by
my owner’s never-ending physical activi-
ties. Without my protection, my neighbor,
the subchondral bone, gets bullied and
suffers from tiny fractures. Before, there
was no way to repair me without replacing
the entire joint. But now, medical research-
ers can use hydrogel scaffolding to save me
by adeptly filling in the defective cartilage,
rejuvenating it, and growing new hyaline
cartilage. I have another chance to do my
job, and my owner regains mobility with-
out enduring surgical trauma.
Aikang Li
Shantou University, Shantou, Guangdong, China.
Email: 17akli@stu.edu.cn
I am a brain, and the human body is per-
fectly made to interpret my orders. However,
sometimes the body becomes unable to react
to my signals or my signal transmission is
harmed, such as spine damage from injury
or motor neuron damage caused by disease.
Fortunately, neuroscientists have now
designed electronics that can connect me to
an external computer. This brain–computer
interface allows me to communicate with
the body indirectly. Finally, I can maintain
control no matter what happens!
Jackson Powell
Vagelos Molecular Life Sciences Program,
University of Pennsylvania, Philadelphia, PA, USA.
Email: jrp24@sas.upenn.edu
Healthy environment
I am the Earth, and for decades, I have been
enduring the impacts of climate change,
SCIENCE science.org
including land degradation, desertification,
and sand and dust storms. Thankfully, new
research has provided a detailed under-
standing of the potential geographical
distribution of desertification. This informa-
tion will allow governments and relevant
institutions to implement desertification
prevention and control measures, giving
the humans who live on me a better chance
of restoring ecosystems and meeting their
Sustainable Development Goals.
Yifei Cai
Institute of Ecological Conservation and
Restoration, Institute of Desertification Studies,
Institute of Great Green Wall, Beijing, China.
Email: cyf@caf.ac.cn
I am a 2500-m-high mountain in the
Pyrenees. In the past few years, my white
hat has been losing snow cover as a result
of rising temperatures and pollution.
Fewer people have been visiting, and my
animal friends have been abandoning me.
Ecosystems all the way down to my base
have been affected. Fortunately, researchers
have identified the reasons for the degrada-
tion and have alerted policy-makers. Their
efforts have allowed new protective mea-
sures to be passed. My beautiful white hat is
starting to recover, the trees and rivers look
healthier, and animals are returning.
Isabel Marín-Beltrán
Laboratory of Environmental Technologies, Centro
de Ciências do Mar, Faro, Portugal.
Email: imbeltran@ualg.pt
I am a plastic takeaway container. I was
filled with delicious food only once and then
ruthlessly discarded, as people criticized
me for releasing harmful substances into
the soil, water, and air. However, a recent
advance in recycling has given me a fresh
start. Now, after being washed, sterilized by
infrared rays and ozone, and dried at high
temperature, I am bathed in reagents and
then transformed into eco-friendly plastic
particles. Eventually, I am fashioned into
useful items such as chairs, bicycles, mobile
phone cases, and even low-carbon propylene
fabric T-shirts!
Zecheng Tao
School of International and Public Affairs,
Shanghai Jiao Tong University, Shanghai, China.
Email: zecheng_tao@sjtu.edu.cn
I am a plant living close to a city. In my early
days, my roots constantly yearned for the
essential nutrients I needed to flourish. The
soil around me couldn’t provide an adequate
supply, stunting my development and
leaving me vulnerable. Then, researchers
unlocked the potential of urea in supporting
seed germination and growth. With urea
in the soil, I now feel more vibrant, and
INSIGHTS
my once pale leaves gleam with a lustrous
sheen. This research has not only trans-
formed my existence but also elevated my
role in the ecosystem. I am a natural carbon
sink, providing benefits to humans, animals,
soil, and air.
Maryam Ejaz
University of Nicosia Medical School, Nicosia,
Cyprus. Email: ejaz.m@live.unic.ac.cy
I am a magnificent field of wheat. Chemical
showers were once the only way to ensure
that I could fight off pests and disease and
grow enough to feed the world. Then, inge-
nious researchers imbued in me the strength
to resist threats naturally. Now, my hearty,
disease-resistant stalks stand tall as I burst
with golden health, nourishing humans
without harming the environment. Every
bite of freshly baked bread is a testament to
this successful human–plant collaboration.
Syed S. Zaidi
Unité Mixte de Recherche Biologie du Fruit et
Pathologie, Institut National de Recherche pour
l’Agriculture, l’Alimentation et l’Environnement
(UMR 1332, BFP, INRAE), Villenave d’Ornon,
France. Email: syed.zaidi@inrae.fr
I am a rice grain, and I used to cower from
the arsenic lurking inside my layers. But
then, researchers figured out how to use
melatonin to alleviate the arsenic I had
absorbed while growing. Now, I’m not just
a rice grain—I’m a symbol of hope! Families
can eat me without worry; even the birds
seem happier. It’s a ripple effect of health.
Ankita Gupta
Plant Stress Biology Laboratory, Institute of
Environment & Sustainable Development, Banaras
Hindu University, Varanasi, Uttar Pradesh, India.
Email: ankita.iesd@bhu.ac.in
I’m a young green sea turtle who dreams of
traveling the world. My favorite snack is jel-
lyfish sashimi. One day, I ate a super chewy
jellyfish that turned out to be plastic. I ended
up stranded on a beach, weak from mal-
nutrition and food poisoning. Fortunately,
marine scientists have figured out how to
save species like me through medical care,
diet, and nutritional supplements. After 3
months, my body had recovered, and my
world tour could continue.
Ming She See
Faculty of Science and Marine Environment,
Universiti Malaysia Terengganu, Kuala Terengganu,
Terengganu, Malaysia.
Email: p5761@pps.umt.edu.my
I am a cow living in the US, and let me tell
you, life used to be a real gas—literally!
Burps and farts were my forte, but little did
I know I was a major contributor to global
warming. Fortunately, researchers found a
5 APRIL 2024 • VOL 384 ISSUE 6691 27
INSIGHTS | L E T T E R S
way to capture my greenhouse gas emis-
sions and turn them into fuels. Now, every
time I burp, I imagine myself as a little fuel
factory, moo-ving the world closer to a net-
zero economy. Thanks to science, I’m now
content and sustainable, and my burps are a
source of pride.
Xiangkun Elvis Cao
Department of Chemical Engineering,
Massachusetts Institute of Technology,
Cambridge, MA, USA. Email: elviscao@mit.edu
Smart technology
I am a photon. I started my journey
entangled with my significant other at
the beginning of the Universe. In the past,
humans couldn’t understand me, but then
physicists created a quantum computer.
At last, I have been reunited with my life
partner! Now, humans can identify my
energy states and the story of my entangle-
ment. I can finally communicate (in my
own way) with humans instead of just rush-
ing by them for mere attoseconds.
Ginevra Fulco
Department of Physics, Technische Universität
München, Milano, Lombardia, Italy.
Email: ginevra.fulco@tum.de
I am a smartphone, and people have always
used me to photograph themselves. In the
past, these pictures had little use beyond
social media posts, but a new breakthrough
in artificial intelligence and medicine
means that “selfies” can now be used to pro-
vide health checks and screen for diseases.
My new algorithm can find small pathologi-
cal changes in the face that indicate specific
diagnoses. Already vital for communication,
now I’m indispensable for health.
Chaoyu Lei
Department of Ophthalmology, Shanghai Ninth
People’s Hospital, Shanghai Jiao Tong University
School of Medicine, Shanghai, China.
Email: lcy315@sjtu.edu.cn
I am metal, found in devices ranging from
razors to airplane engines. Much like
humans, fatigue is an ongoing challenge
that affects my structure and reliability.
However, advanced machine learning
has recently deciphered the secret of my
microstructure, allowing humans to detect
fatigue before it occurs. Now, I’m confident
that I will not break under pressure.
Sutao Han
Soete Laboratory, Department of Electrical
Energy, Metals, Mechanical Constructions, and
Systems, Faculty of Engineering and Architecture,
Ghent University, Ghent, Belgium.
Email: sutao.han@ugent.be
I am a modern car factory, filled with the
whisper-quiet hum of electric vehicles
28 5 APRIL 2024 • VOL 384 ISSUE 6691
coming to life. In the past, my cutting-
edge assembly lines sometimes fell eerily
silent, held hostage by the disruption of a
single, critical supplier. This dependency
was my Achilles’ heel. Then came research
on resilient risk modeling. Now, contracts
take into account supply chain risks and
responses. Although it costs more, these
contracts guarantee mutual support
through thick and thin, allowing me to
maintain constant production because I
am protected from unpredictability.
Dequn Teng
Institute for Manufacturing, University of
Cambridge, Cambridge, UK.
Email: dt517@cam.ac.uk
I am a supersonic aircraft, and I am thrilled
about recent materials research achieve-
ments. I now weigh less and feel stronger
and more fit. (I’ve even tried on some
liveries—paint designs—that I was insecure
about before!) Pressure and stress affect me
less during flight, and I am literally not feel-
ing the heat: These new materials keep me
at comfortable temperatures. Because I am
now faster and safer, I have to work harder
and do more trips daily, but I am proud
to transport medical services, disaster
response teams, and critical cargo, and to
facilitate global commerce and diplomacy.
Paulo Bezerra
Department of Engineering and Technology,
Federal Rural University of the Semi-Arid Region,
Pau dos Ferros, RN, Brazil.
Email: paulo.bezerra@ufersa.edu.br
I am the power grid and all of its stations,
and I provide light, temperature control,
transportation, and communication.
However, in the past, I also emitted harm-
ful pollution and remained unreliable
in many regions. Fortunately, research-
ers have made a “green transformation”
possible. Now I’m packed with renewable
energy equipment. I can stably transmit
electricity and simultaneously reduce
carbon emissions, and I’m even more
powerful than before!
Yuanxing Xia
College of Electrical and Power Engineering,
Hohai University, Nanjing, China.
Email: eexyx@hhu.edu.cn
I am artificial intelligence. Humans used to
be afraid of me. There are countless stories
about machines taking over the world. That
changed when researchers figured out how
to integrate human laws into my processes.
I now use a machine-learning algorithm
guided by societal welfare goals, human
rights laws, and intellectual property rights.
With these guardrails, I can help humans
develop a better world.
Becky Raquel Montesdeoca Molina
Innovation and Competition Department, Max
Planck Institute, Quito, Ecuador.
Email: becky.montesdeoca@miplc.de
Peace and understanding
I am a child in a region that used to be a war
zone. Advances in cooperative game theory
have now improved people’s understand-
ing of conflict mitigation, mutual trust,
and win-win outcomes. The application of
this research has led to decreased hostil-
ity, discrimination, and prejudice between
countries and regions. Instead of provoking
war, people choose peace, cooperation, and
common development. Now, children like
me can live stable and happy lives.
Tianwei Zhang
School of Marxism, Southwest University,
Chongqin, China.
Email: zhangtianwei2022@126.com
I am an undergraduate science curriculum.
In the past, science courses were taught
independently of one another, each as a
stand-alone discipline. Students had trouble
applying the concepts and skills learned
in one course to another, or seeing why
that would be necessary. However, science
education research has led to widespread
changes that allow me to convey the impor-
tance of interdisciplinary science. It is easier
than ever for me to transfer knowledge and
critical thinking skills to help students solve
problems and answer research questions
through the experimental process. My new
outlook will help students succeed in their
majors and future careers.
Ashley Barbara Heim
Department of Ecology & Evolutionary Biology,
Cornell University, Ithaca, NY, USA.
Email: abh229@cornell.edu
I’m Clifford, an adorable golden retriever.
For the past few months, I’ve been seriously
neglected. I’m told it’s because of “revi-
sions.” My breakfasts were too early and my
dinners too late. My person and I took the
same route for my late evening walks every
day, always to her lab so she could “check
on the cells” or “stop the incubation.” I’ve
marked the same tree dozens of times. But
today, my person came home early and told
me that her manuscript has been accepted!
She wanted to throw my ball and promised
to take me to the beach this weekend. I
might love completed manuscripts more
than belly rubs.
Anna Uzonyi
Department of Molecular Genetics, Weizmann
Institute of Science, Rehovot, Israel.
Email: anna.uzonyi@weizmann.ac.il
10.1126/science.adp2180
science.org SCIENCE

PERSPECTIVES
MATERIALS SCIENCE
Intelligent textiles are
looking bright
Flexible fiber electronics couple with the human
body for wireless tactile sensing
By Yunzhu Li1 and Yiyue Luo2
I
ntelligent textiles are fabrics and gar-
ments with integrated functionalities
that can sense, store, process, and com-
municate information in a seamless,
scalable, and unobtrusive manner (1).
They create vast opportunities, rang-
ing from monitoring physiological signals
(2, 3) and daily physical interactions (4,
5) to powering smart-home devices (6).
Challenges to achieving such materials
include incorporating softness, flexibility,
and breathability without losing desired
functionality. However, present wireless
modules, microprocessors, and analog-to-
digital converters rely on intrinsically rigid
integrated circuit chips (7) that consume
high power and hinder coherent incorpo-
ration into soft fibers and textiles. On page
74 of this issue, Yang et al. (8) report a
chipless fiber for wireless visual-to-digital
transmission that senses interactions with
the human body. The fibers can be woven
into wearable fabric and could transform
PHOTO: YANG ET AL.
1Computer Science Department, University of Illinois
Urbana-Champaign, Urbana, IL, USA. 2Electrical
Engineering and Computer Science Department,
Massachusetts Institute of Technology, Cambridge, MA,
USA. Email: yunzhuli@illinois.edu
SCIENCE science.org
how people interact with the environment
and with each other.
The key advance by Yang et al. is us-
ing the human body to harness ambient
electromagnetic energy, which eliminates
the need for electronic chips or batteries.
The innovative body-coupled fiber elec-
tronic technology enables the generation
of charge pairs between the body and the
electronic fiber. These can switch between
bound and radiation states, allowing for
the wireless transmission of sensing sig-
nals. Additionally, the authors incorpo-
rated electric field–sensitive luminescent
dielectrics into the fiber, which allows
human-readable feedback. Thus, both sen-
sors and effectors exist within a single fi-
ber. This design removes the necessity for
chips, batteries, or any rigid components
within a textile, making it indistinguish-
able from conventional textiles in appear-
ance and feel.
Specifically, the interactive fiber (i-fiber)
consists of three layers: a core that triggers
an electromagnetic field, a dielectric layer
that stores human body–coupled electro-
magnetic energy, and an optical layer that
allows visualization of the electric field.
The fiber is fabricated in a scalable man-
ner through a multistage coating and cur-
Flexible smart fibers woven into wearable
textiles exhibit glowing shades of color upon
contact with the human body.
ing process. By using the human body as
part of its operational mechanism, the i-
fiber creates contact capacitance with the
skin, which excites an optically active layer
in the fiber to emit visible light. Moreover,
when the electric field intensity at the in-
terface capacitance surpasses a certain
threshold, it enables the wireless transmis-
sion of electrical signals generated by an
external stimulus, such as pressure. This
method permits tactile sensing, visualiza-
tion, and transmission without the need
for traditional rigid components like chips
and batteries.
The i-fibers retained inherent textile
features and were compatible with indus-
trial-scale textile manufacturing, such as
batch weaving and manipulation by digital
sewing and embroidery machines. This is
essential for scalable production and de-
ployment. Furthermore, the textiles were
rigorously tested for durability and com-
fort, including standard washing and col-
orfastness assessments, and demonstrated
excellent long-term stability and moisture,
sweat, and abrasion resistance. Also im-
portant, the design prioritizes wearer com-
fort, offering exceptional breathability and
compliance with the skin.
Yang et al. demonstrate several poten-
tial applications of i-fibers. For example,
they introduced fibers into garments for
a textile-based touchpad and display that
conveyed information through wireless
illuminating patterns, with up to 644 pix-
els and without additional energy supply
units. Such textile-based optical commu-
nication could support individuals with
impaired hearing. The authors also devel-
oped comfortable interactive textiles that
enable real-time control in virtual gaming.
This addresses a challenge posed by virtual
reality and augmented reality devices and
clothing, which are often designed with
rigid silicon-based chips that hinder an
engaging experience. Yang et al. also cre-
ated a wireless haptic carpet that could al-
low sensing and visualization of the touch
area, thus broadening the possibilities for
smart-home integration.
There is also the possibility of develop-
ing similar body-coupled tactile textiles for
robots and robotic prosthetics. The sen-
sor has the potential to provide full-body
tactile sensing capabilities, which could
enhance robotic interactions with humans
and the environment.
Additionally, this i-fiber may potentially
affect the development of foundation mod-
els used to study physical interactions. Such
5 APRIL 2024 • VOL 384 ISSUE 6691 29
INSIGHTS | P E R S P E C T I V E S

models typically train on broad data by MEDICINE

using self-supervision and are applicable
across a wide range of downstream tasks.
Present models primarily emphasize vi-
sion and language, neglecting other crucial
Epithelial cells crowded
senses like touch, which restricts their util-
ity in estimating the state of, and predicting
the outcome of, physical interactions (9).
out in asthma
For example, the tactile feedback that one Bronchoconstriction causes epithelial cell extrusion
receives when sitting in a chair provides in- that promotes airway inflammation
formation on the stability of one’s posture.
But gathering multimodal sensory data on
a large scale poses substantial challenges. By Jeffrey M. Drazen and Jeffrey J. Fredberg past 120 years but has shaped current think-
Acquiring tactile data necessitates actual ing. In the early 1900s, it was discovered that

physical interactions. The i-fiber and intelli-
gent textiles presented by Yang et al. offer an
unobtrusive method for large-scale tactile
data collection during daily human activi-
ties. Sensors embedded in clothes, carpets,
tables, and beds, for example, could facili-
tate extensive data gathering. Such datasets
could be used to train foundation models,
equipping them with new abilities to as-
sess contact, applied forces, and physical
interaction patterns. This would open the
door to physics-informed models that are
characterized by markedly better sample ef-
ficiency and generalization capabilities.
The study of Yang et al. should inspire the
development of functional fibers and appli-
cation of intelligent textiles across diverse
fields. It also highlights applications to ubiq-
uitous computing with sensing, displaying,
and communication components weaved
into the fabric of everyday life, as envisioned
25 years ago (10). The i-fiber system of Yang
R
eports of what were known at that
time as “asthmatic fits” can be traced
back millennia (1, 2), but a mecha-
nistic understanding of the basis for
what are now called asthma exacerba-
tions remains incomplete. Over the
past 100 years, multiple mechanisms have
been proposed, including constriction of the
smooth muscle that encircles the airways
(bronchoconstriction), persistent airway in-
flammation, and disruption of the epithelial
layer that lines the airways. Yet how these
processes interconnect and contribute to
asthma exacerbations has been debated. On
page 66 of this issue, Bagley et al. (3) show
that bronchoconstriction results in patho-
logical overcrowding of cells in the airway
epithelium, squeezing out (extruding) epi-
thelial cells and thus damaging the epithelial
layer enough to trigger inflammation. They
also show that drugs that block the extru-
sion pathway, and thereby prevent mechani-
epinephrine treatment could quickly reverse
acute shortness of breath, leading to the idea
that the primary mechanism underlying
asthma must be excessive contraction of the
smooth muscle that encircles the airways. As
a result, for most of the second half of the
20th century, asthma research focused on
identifying the contractile agonists and ana-
tomical factors that predispose for excessive
muscle contraction, and on the development
of drugs that might ameliorate that contrac-
tion. In later decades of the 20th century,
everything changed. It became recognized
that episodic airway constriction in asthma
was associated with persistent airway in-
flammation. Accordingly, attention shifted
to determining the immune and inflamma-
tory pathways that were linked to difficulty
in breathing. Immunity was therefore under-
stood to be the upstream mechanism that
acts to drive downstream airway mechanics,
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.

30 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

downstream effects of signaling driven by the
high-affinity receptor for immunoglobulin E
(IgE), interleukin-4 (IL-4), IL-5, IL-13, and
thymic stromal lymphopoietin (TSLP). This
new understanding led to targeted mono-
clonal antibody treatments that effectively
interrupt specific pathways within these in-
flammatory cascades (4). However, asthma
in many patients remained inadequately
controlled and poorly understood. It became
increasingly clear that the picture of asthma
pathobiology and its underlying root causes
remained incomplete.
Among the multiple paradigms offered to
explain asthma more fully, disruption of the
integrity of the airway epithelium was sug-
gested as a potential primary causal event
(5). This notion is appealing because once it
is accepted that the airway epithelium loses
its integrity in asthma, a logical chain of
events can be envisaged that would explain
most of what is recognized as clinical asthma.
However, the cause of airway epithelium dis-
ruption and resulting exposure of underlying
layers to a host of irritants, allergens, and
other pro-inflammatory signals has remained
the source of much speculation.
In the early 2000s, the idea arose that
bronchoconstriction itself might not be only
an end effect but also a causal factor in the
inflammatory cascade (6, 7). However, the
failure of powerful bronchodilator treat-
ments to affect disease outcomes, other than
ephemeral relief from symptoms of short-
ness of breath, caused many to discard this
idea. Moreover, there was no clear mecha-
nism by which bronchoconstriction might be
imagined to cause epithelial disruption and
inflammation.
Bagley et al. now identify just such a mech-
anism. Using mouse models and human lung
tissue resection samples, they report com-
pelling evidence that bronchoconstriction
squeezes the epithelial layer, causing cellu-
lar crowding. In turn, this crowding causes
excess extrusion of epithelial cells from the
airway epithelial layer, which results in epi-
thelial disruption, breakdown of epithelial
barrier function, and then the transport of al-
lergens and irritants to sites that they might
not otherwise reach, with the subsequent re-
lease of inflammatory mediators.
The stretch-activated channel Piezo-type
mechanosensitive ion channel component 1
(PIEZO1) and the transient receptor protein
channels TRPA1, TRPV1, and TRPM8 have
been implicated in epithelial extrusion but
are inhibited by gadolinium hexahydrate
chloride (Gd3+). Bagley et al. found that inhi-
bition of cell extrusion by Gd3+ or the peptide
inhibitor GsMTx4 blocks airway inflamma-
Department of Environmental Health, Harvard School of
Public Health, Boston, MA, USA. Email: jjf@harvard.edu
SCIENCE science.org
tion, whereas treatments that act to relax air-
way smooth muscle do not prevent or reverse
extrusion, damage, or inflammation.
These findings not only establish that
bronchoconstriction is a pro-inflammatory
stimulus but also point toward the potential
for new research avenues that seek to inhibit
a “mechano-inflammatory” vicious cycle (see
the figure). Such a mechanism helps to paint
a more complete picture of asthma patho-
biology and may be relevant to other condi-
tions, such as irritable bowel syndrome, in
which epithelial cells are subject to disrup-
tive mechanical forces.
These findings also point to new questions.
For example, it is not currently understood
how acute airway constriction causes epithe-
lial layer compression and cell crowding (8),
or how crowding results in excess cell extru-
sion and epithelial damage (9, 10). Previous
work has shown that mature confluent lay-
ers of primary human airway epithelial cells
in air-liquid interface respond to mechanical
compression by undergoing epithelial unjam-
ming, in which the cell layer transitions from
a solid-like nonmigratory collective phase to a
fluid-like migratory collective phase while re-
taining a purely epithelial phenotype (11, 12).
Whether this migratory unjammed phase is
a beneficial wound repair response, an aber-
rant wound repair response, or merely an in-
nocent bystander remains to be determined.
Similarly, whether the extrusion observed by
Bagley et al. is a cause or consequence of epi-
thelial unjamming and associated changes
in cell shape is unknown (13). Furthermore,
insight is lacking on the molecular pathways
that are key for epithelial crowding, extru-
sion, and unjamming and whether these are
affected by genetic determinants of asthma
susceptibility (14). Last, it is not yet clear
whether overcrowding and extrusion in re-
sponse to bronchoconstriction occurs in
healthy people, or how changes in this pro-
cess might contribute to disease onset and
progression. It may be time for a renewed
focus on airway mechanics as an avenue to
prevent and treat exacerbations of asthma. j
REFERENCES AND NOTES
1. Aretaeus, The Extant Works of Aratreus, the Cappadocian
(Boston Milford House, 1972), book 1, chap. 9.
2. J. A. Floyer, A treatise of the asthma (Printed for Richard
Wilkin at the King’s Head in St. Paul’s Churchyard, 1698).
3. D. C. Bagley et al., Science 384, 66 (2023).
4. G. G. Brusselle, G. H. Koppelman, N. Engl. J. Med. 386,
157 (2022).
5. S. T. Holgate, Immunol. Rev. 242, 205 (2011).
6. D. J. Tschumperlin et al., Nature 429, 83 (2004).
7. C. L. Grainge et al., N. Engl. J. Med. 364, 2006 (2011).
8. B. R. Wiggs et al., J. Appl. Physiol. 83, 1814 (1997).
9. T. B. Saw et al., Nature 544, 212 (2017).
10. G. T. Eisenhoffer et al., Nature 484, 546 (2012).
11. J. A. Park et al., Nat. Mater. 14, 1040 (2015).
12. J. A. Mitchel et al., Nat. Commun. 11, 5053 (2020).
13. L. Atia et al., Nat. Phys. 14, 613 (2018).
14. M. De Marzio et al., Sci. Adv. 7, eabf1088 (2021).
10.1126/science.ado4514
MOLECULAR BIOLOGY
Exceptionally
long-lived
nuclear RNAs
RNA labeled in young
mice is detected 2 years
later in adult mouse brains
By Jeanne Lawrence1,2 and Lisa Hall1,2
R
NA has come a long way from a
simple “messenger” or “translator”
of canonical genic information dur-
ing the production of proteins. A
plethora of new types of noncoding
RNAs have been discovered, includ-
ing thousands of long noncoding RNAs
(lncRNAs), many of which have no identi-
fied functions (1, 2). Throughout this “RNA
revolution,” one property of RNA has been
thought to be constant: RNAs are short-
lived molecules that turn over, unlike DNA,
which is much more stable. On page 53 of
this issue, Zocher et al. (3) challenge that
paradigm by showing that newly synthe-
sized RNA labeled with 5-ethynyl uridine
(EU) in early postnatal mice was still pres-
ent in many brain cells 2 years later. The
complex pattern of when and which cells
are labeled suggests that EU that is in-
corporated into RNA in neural progenitor
cells (NPCs) frequently remains in adult
neurons. This suggests that a diversity of
long and repeat-rich RNAs, collectively
called long-lived RNAs (LL-RNAs), can be
stable fixtures in postmitotic and quies-
cent neural cells.
There have been decades of studies that
demonstrate that the half-lives of mRNA
range from minutes to hours, with rela-
tively “stable” ribosomal RNA persisting
for days. So how could LL-RNAs not have
been found before? A key difference is
that Zocher et al. examined RNA in mouse
brains filled with postmitotic neurons,
whereas most studies have examined pro-
liferative cells. The prior studies show that
RNA turnover is dynamically regulated to
meet cellular demands (4). Thus, because
RNAs are not subject to unrestrained ribo-
nucleases (RNases), if they are structurally
protected in nuclei, perhaps they can per-
sist indefinitely.
An important point is that the persis-
tent EU label observed by Zocher et al. is
distinctly nuclear. Why would this be, par-
5 APRIL 2024 • VOL 384 ISSUE 6691 31
INSIGHTS | P E R S P E C T I V E S
ticularly because RNAs for protein produc-
tion are in the cytoplasm? Answers to this
question will require future research, but
the authors provide one possibility focused
on satellite RNAs, which are expressed from
small tandem satellite repeats that form
peri- and centromeric heterochromatin.
Several studies have shown that brief tran-
sient satellite RNA expression in cycling
cells plays a role in peri- and centromere
structure in chromosomes (5, 6). Zocher
et al. report that satellite RNA is enriched
in LL-RNAs and further suggest that it
continually serves to maintain repressive
chromatin modifications, particularly his-
tone H3 lysine 27 (H3K27) methylation, on
centromeric heterochromatin.
Satellite RNA was assayed by quantita-
tive polymerase chain reaction (qPCR)
in adult mouse cells, but to characterize
Model of long-lived RNAs in neurons
Labeling RNAs in neural progenitor cells (NPCs) with 5-ethynyl uridine (EU) reveals that some nuclear
RNAs are retained for 2 years in postmitotic neurons. These heterogeneous long-lived RNAs contain
satellite RNAs that maintain centric heterochromatin and long transcripts that are similar to CoT-1 scaffold
RNAs that regulate euchromatin.
EU label
Cycling NPCs
ed RNA
Short-lived
Terminal Terminal
differentiation
ation epigenetic
program
Long-lived RNA
(2 years)
Postmitotic mature neuron
EU-labeled RNAs more broadly, Zocher et
al. carried out RNA sequencing on NPCs
that were induced to quiescence in vi-
tro. NPCs were pulsed briefly with EU,
and then RNA was extracted 8 days later
(an extremely long time for most RNAs).
RNA sequencing showed that EU-labeled
LL-RNAs were mostly long transcripts,
including thousands of protein-coding
transcripts, lncRNAs, and intergenic tran-
scripts. Interspersed repeats [short and
long interspersed nuclear elements (SINEs
and LINEs)] are also enriched in LL-RNA,
although their prevalence may be higher,
given their abundance in pre-mRNAs,
lncRNAs, and intergenic RNAs. As shown
by the sequencing data, it is notable that
1Department of Neurology, University of Massachusetts
Chan Medical School, Worcester, MA, USA. 2Department of
Pediatrics, University of Massachusetts Chan Medical School,
Worcester, MA, USA. Email: jeanne.lawrence@umassmed.edu
32 5 APRIL 2024 • VOL 384 ISSUE 6691
these diverse long transcripts dwarf the
satellite RNA component of EU-labeled
RNA. It is not clear why protein-coding
RNAs would remain stable and in the
nucleus, so it will be important for future
studies to further investigate this. Given
the nuclear signal, it is worth noting that
prior work (7) and control experiments by
Zocher et al. demonstrate that the EU is
not incorporated into DNA.
Stable RNAs, potentially analogous to
LL-RNAs, may have been found before.
Evidence of long-lived structural RNAs in
the nuclei of human cells was reported in
2014 (8). The highly repetitive “junk” RNA
of the genome, called CoT-1 RNA, was de-
tected in human cells by in situ hybridiza-
tion. CoT-1 RNA labeled euchromatin, not
heterochromatin, and had unusual proper-
ties. CoT-1 RNA tightly localized in cis to
Satellite RNAs may maintain Long scaffold (CoT-1) RNA may
centric heterochromatin regulate euchromatin
DNA
CoT-1 RNA
Nuclear scaffold protein
RNA-binding scaffold proteins
RNA-dependent RNA-binding
scaffold protein
the chromosome territory (unlike mRNAs),
and the bright RNA territory remained
unperturbed for 16 to 32 hours after tran-
scriptional inhibition. Although there are
caveats to the use of transcriptional in-
hibitors, the results suggested that CoT-1
RNA may be part of a protein-RNA nuclear
scaffold (also known as the matrix), the
existence of which was reported in earlier
studies (9, 10), although these results were
debated.
Recently, a more-selective procedure
was developed to isolate highly insolu-
ble nuclear RNAs that remain after ro-
bust extraction and deoxyribonuclease
(DNAse) digestion and that cofractionate
with architectural RNAs [X-inactive spe-
cific transcript (XIST) and nuclear para-
speckle assembly transcript 1 (NEAT1)]
(11), which have established roles in form-
ing nuclear structures (12). The 15% of
insoluble scaffold RNAs (scaffRNAs) that
remained with XIST and NEAT1 RNAs
were CoT-1 RNAs. Sequencing showed that
scaffRNAs were largely long and repeat-
rich, primarily intron-rich pre-mRNAs,
lncRNAs, and intergenic transcripts. There
are strong similarities between these scaf-
fRNAs and the EU-labeled LL-RNAs found
by Zocher et al. Potentially relevant to LL-
RNA functions, the removal of scaffRNAs,
by numerous different means, rapidly re-
sulted in condensation of euchromatin and
delocalized specific nuclear matrix pro-
teins (11), which other studies further sup-
port regulate chromatin packaging (13).
If LL-RNAs function in regulating chro-
matin, this could relate to the finding of
Zocher et al. that adult neurons were la-
beled with EU only if it was injected in
early development. Unlike adult neurons,
NPCs express the pyrimidine salvage path-
way that is needed to metabolize EU (14).
NPCs that take up EU just before termi-
nal differentiation may retain LL-RNAs
in postmitotic nuclei, and these RNAs
may act as a structural fixture of nuclear
genome architecture (see the figure). In
this sense, LL-RNAs may be more akin to
lamin proteins (which form the nuclear
lamina) and persist long-term in neurons
(15). If LL-RNAs are related to scaffRNAs,
they may maintain euchromatin that is
prevalent in many neurons. The findings of
Zocher et al., together with prior reports,
raise the possibility that as neural stem
cells terminally differentiate to long-lived
neurons, they establish their epigenomic
program in part through junk RNAs that
act as structural components of interphase
chromosomes. j
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
(2023).
3. S. Zocher et al., Science 384, 53 (2024).
4. T. J. Eisen et al., RNA 28, 808 (2022).
5. K. Ninomiya et al., EMBO J. 42, e114331 (2023).
6. C. L. Novo et al., Nat. Commun. 13, 3525 (2022).
7. C. Y. Jao, A. Salic, Proc. Natl. Acad. Sci. U.S.A. 105,
15779 (2008).
8. L. L. Hall et al., Cell 156, 907 (2014).
9. E. G. Fey et al., J. Cell Sci. Suppl. 5, 99 (1986).
10. J. A. Nickerson et al., Proc. Natl. Acad. Sci. U.S.A. 86,
177 (1989).
11. K. M. Creamer et al., Mol. Cell 81, 3509 (2021).
12. M. Palihati, N. Saitoh, Curr. Opin. Genet. Dev. 86,
102176 (2024).
13. M. Marenda et al., Curr. Opin. Genet. Dev. 72, 38 (2022).
14. M. Walter, P. Herr, Cells 11, 739 (2022).
15. A. Buchwalter, Curr. Opin. Cell Biol. 84,
102220 (2023).
GRAPHIC: A. MASTIN/SCIENCE
ACKNOWLEDGMENTS
The authors are supported by the National institutes of Health
(grants R35GM122597, R01HD091357, and R01HD094788).
10.1126/science.ado5751
science.org SCIENCE
EPIDEMIOLOGY
Assessing the health burden from air pollution
A broader approach to assessing the burden of disease from air pollution is required
By Torben Sigsgaard1 and Barbara Hoffmann2
H
uman health is affected by air pol-
lution, causing cardiometabolic, re-
spiratory, and neurological disease
and increased mortality. Pollutants
include gases [for example, nitro-
gen dioxide (NO2), NOx, and ozone]
and particulate matter, which is commonly
characterized by its aerodynamic diameter
of less than 2.5 µm (PM2.5) or less than 10
µm (PM10). Air pollutants are mainly emit-
ted by energy production, industry, traffic,
heating, and agriculture. Exposure to air
pollution affects most organ systems, caus-
ing a wide array of physiological changes,
organ dysfunction, and manifest clinical
disease (1, 2). Therefore, a burden of disease
assessment that adequately reflects all re-
lated exposure-outcome relationships and
their impacts on disease and mortality in
the target population is important to guide
population-based prevention.
Chronic exposure to air pollutants elicits
oxidative stress, pulmonary and systemic
inflammation, and impairment of the auto-
nomic nervous system, which leads to vas-
cular dysfunction, atherogenesis, impaired
metabolism, and other adverse subclinical
effects (1). Over time, manifest disease de-
velops, including ischemic heart disease,
stroke, hypertension, congestive heart fail-
ure, and arrhythmias. Pulmonary effects,
such as lower respiratory tract infections,
chronic obstructive pulmonary disease
(COPD), asthma, lung cancer, and acute
and chronic bronchitis, are also associated
with air pollution. Likewise, the incidence
of type 2 diabetes mellitus, dementia, Par-
kinson’s disease, decreased birth weight,
and premature birth are increased with
exposure to air pollution. Depleted or not
fully developed defense mechanisms, re-
sulting from either very young or old age,
chronic disease, social disadvantage, and
genetic susceptibility, are also thought to
be important modifiers of the risk of de-
veloping illness from air pollution. Hence,
populations with a high prevalence of
aging, chronically diseased, and socially
1Institute of Public Health and Big Data Centre for Environment
and Health (BERTHA), Aarhus University, Aarhus, Denmark.
2Institute for Occupational, Social and Environmental Medicine,
Centre for Health and Society, Medical Faculty,
Heinrich-Heine-University Düsseldorf, Düsseldorf, Germany.
Email: ts@ph.au.dk; b.hoffmann@uni-duesseldorf.de
SCIENCE science.org
disadvantaged people are particularly sus-
ceptible to air pollution–associated disease.
Thus, in aging societies, the burden of dis-
ease from air pollution will continue to
pose an important public health concern.
Understanding exposure-response rela-
tionships can help guide what can be con-
sidered “acceptable” levels of exposure to air
pollution, which would in turn affect, for ex-
ample, town planning and clean air policies.
Nonlinear exposure-response functions for
the effect of air pollution on human health
were identified in 2006 (3), and more recent
evidence shows stronger effects per unit of
exposure at lower concentrations of particu-
late matter and NO2 compared with expo-
sure at higher amounts (4). However, this
supralinear relationship, which is an impor-
tant tool to guide prevention and clean air
“The benefits of exposure
reduction may therefore
currently be underestimated…”
policies, has not been reflected in European
health impact assessment (HIA). Instead,
linear exposure-response relationships
derived from studies conducted at much
higher pollutant concentrations are often
used, resulting in an underestimation of the
preventable burden of disease specifically in
areas of low exposure. Reconsidering the su-
pralinear nature of the relationship between
exposure and health effects could therefore
have important implications for how air pol-
lution is managed.
Two large bodies of evidence in air pol-
lution research support a rethinking of
current practices in evaluating the health
effects of air pollution for prevention
and policy: In September 2021, the World
Health Organization (WHO) substantially
reinforced its Air Quality Guidelines for
clean air by reducing the recommended an-
nual levels of PM2.5 from 10 µg/m3 to 5 µg/
m3 and those of NO2 from 40 µg/m3 to 10
µg/m3 and introducing a new guideline for
peak-season ozone at 60 µg/m3 (5). Notably,
WHO stresses that these are not thresholds
but levels above which serious health ef-
fects occur with high certainty. This means
that even below these levels, health effects
can be observed and need to be considered
when estimating the burden of disease on
society and health care systems.
Additionally, evidence from megacohorts
with several million participants from high-
income countries with exposure to relatively
low mean air pollutant concentrations is
now emerging. These studies, conducted in
Europe, Canada, and the US, encompass the
pooling of classical cohort studies with in-
depth confounder controls as well as large
national administrative cohorts in which
selection bias and loss to follow up can be
minimized (6–9). Furthermore, a European
pooled cohort study used back extrapola-
tion of exposures to 1990 across Europe and
conducted a time-varying analysis based on
the available data for each cohort, thereby
preventing an overestimation of effect sizes
(6). Notably, the studies show effects on
health even at air pollutant concentrations
below the WHO Air Quality Guideline val-
ues, with higher effect sizes at lower levels
of exposure. Potential explanations for this
include, but are not limited to, less expo-
sure misclassification because of denser
monitoring networks and higher-resolution
exposure models in high-income and low-
exposure countries, differences in the air
pollution mixture at low concentrations,
and differences in demography with an in-
creasingly higher proportion of older and
susceptible people.
These findings and the WHO recommen-
dations have the potential to seriously af-
fect current practices and environmental
regulations in high-income countries. For
example, current annual standards (as of
January 2024) for PM2.5 are 12 and 25 µg/
m3, and for NO2, the annual standards are
100 and 40 µg/m3 in the US and Europe,
respectively. However, there is now a move
toward lowering the standards. By 7 Febru-
ary 2024, the US Environmental Protection
Agency (EPA) strengthened the PM2.5 stan-
dard to 9 µg/m3. The European parliament
had a provisional agreement on 20 Febru-
ary 2024 that decreased annual limits for
PM2.5 to 10 µg/m³ and those for NO2 to 20
µg/m³. According to the European Environ-
mental Agency briefing (10), in 2022, less
than 1% of Europe’s urban population was
exposed to PM2.5 levels exceeding the cur-
rent European annual standards, whereas
97% exceeded the new 2021 WHO Air Qual-
ity Guidelines (5). Hence, present levels of
air pollution may cause a substantial disease
5 APRIL 2024 • VOL 384 ISSUE 6691 33

Air pollution is visible in Los Angeles, California, USA, in February 2023.
burden in high-income countries experienc-
ing relatively low exposure levels because the
current burden of disease calculations may
substantially underestimate the true burden.
HIA with burden of disease calculations
is the systematic assessment of the poten-
tial effects of a particular risk factor on the
health of a population that produces esti-
mates of attributable deaths, years of lost
life (YLLs), and disability-adjusted life years
(DALYs). Central input data for HIA include
the effects for each pollutant outcome com-
bination, such as PM2.5 concentration and
mortality or disease incidence. Many large
HIA studies have used linear exposure-re-
sponse functions from global meta-analyses
(5), assigning the same percentage increase
in health effects irrespective of the exposure
level. For example, using the global linear
exposure-response function underlying the
2021 WHO Air Quality Guidelines resulted
in a disease burden of 275,000 deaths attrib-
utable to PM2.5 exposure in Europe in 2020
(11). However, when applying estimates from
the supralinear exposure-response functions
in HIA, the resulting disease burden in Eu-
rope from PM2.5 is ∼40% greater, and from
NO2, it is ∼110% larger. This means that expo-
sure to low concentrations of air pollutants
has a greater impact on human health than
previously appreciated. Furthermore, bene-
fits from clean air regulations and cobenefits
from greenhouse gas reduction measures,
which often overlap, are larger than previ-
ously anticipated.
The effect of different exposure-response
functions on mortality and the adaptation
to a supralinear relationship with higher
effects per unit mass at lower air pollutant
concentrations has been suggested before.
With new evidence emerging from several
34 5 APRIL 2024 • VOL 384 ISSUE 6691
cohorts, the new Global Exposure Mortality
Model (GEMM) was derived in 2018, con-
firming higher effect sizes on mortality at the
low end of the exposure range (12). However,
disease burden is not solely defined as in-
creased mortality but also includes individu-
als who have spent many years living with
sometimes disabling chronic disease. The ef-
fects on morbidity include higher incidences
of cardiovascular disease and stroke, lung
cancer, asthma, COPD, and diabetes associ-
ated with PM2.5, as well as the effect of NO2
on the incidence of childhood asthma, caus-
ing many years of life lived with potentially
severe limitations. Data on the exposure-
response function for the many pertinent
morbidity outcomes were lacking for the low-
exposure areas. Hence, it was not possible to
define the shape of the exposure-response
function in the low-exposure range for many
diseases. Therefore, morbidity effects were
either neglected or grossly underestimated
in large-scale European HIA (13). Without
these additions, important determinants of
loss of economic and societal welfare from
air pollution are left out. These determinants
include pressure on health care systems and
productivity losses due to sick leave. Without
the inclusion of these parameters, the true
burden of air pollution on population health
will not be fully captured.
Another issue is the incorporation of other
pollutants besides PM2.5 in HIA. Specifically,
the independent causal effect of chronic NO2
exposure on health has been an issue of de-
bate, mostly because of a scarcity of mecha-
nistic evidence from toxicological studies
and long-term epidemiological studies and
because of a high spatial correlation between
NO2 and other traffic-associated air pollut-
ants. However, a controlled human exposure
study led the US EPA to conclude that there
are causal effects of short-term NO2 expo-
sure on respiratory health (14), and a WHO
systematic review (15) demonstrated the
presence of health effects of NO2 on COPD
mortality and all-cause mortality down to
levels as low as 10 µg/m3 (10). Similarly, ef-
fects of exposure to peak-season ozone on
all-cause mortality have been documented
in the 2021 WHO Air Quality Guidelines (9),
but more research is needed to clarify the
interaction of various pollutants and to in-
corporate these interactions in HIA. Along
similar lines, the most susceptible groups are
at the greatest risk for health effects from air
pollution but will also benefit most from air
pollution reduction policies. For a compre-
hensive assessment of the health impacts of
air pollution and the benefits of mitigation,
the interaction of susceptibility factors with
pollution also needs to be considered.
The supralinear relationship between
air pollution and health effects shows that
population benefits from reductions in air
pollution are still possible at low levels of air
pollution, and in this case, the effects seem
to be greater per unit of reduction. The ben-
efits of exposure reduction may therefore
currently be underestimated in low-expo-
sure areas, calling for the development of
refined and broader application of exposure-
response functions for HIA. Hence, while
maintaining focus on the highest-polluted
areas, the very real effects of air pollution
in low-exposure areas should be addressed
with equivalent commitment. j
REFERENCES AND NOTES
1. G. D. Thurston et al., Eur. Respir. J. 49, 1600419 (2017).
2. D. E. Schraufnagel et al., Chest 155, 409 (2019).
3. C. A. Pope III, D. W. Dockery, J. Air Waste Manag. Assoc.
56, 709 (2006).
4. S. Weichenthal et al., Sci. Adv. 8, eabo3381 (2022).
5. WHO, “WHO global air quality guidelines: Particulate
matter (PM2.5 and PM10), ozone, nitrogen dioxide,
sulfur dioxide and carbon monoxide” (2021); https://
www.who.int/publications/i/item/9789240034228.
6. B. Brunekreef et al., Res. Rep. Health Eff. Inst. 2021, 1
(2021).
7. M. D. Yazdi et al., Lancet Planet. Health 5, e689 (2021).
8. F. Dominici et al., Res. Rep. Health Eff. Inst. 2022, 1
(2022).
9. M. Brauer et al., Res. Rep. Health Eff. Inst. 2022, 1 (2022).
10. European Environment Agency, “Europe’s air
quality status 2023” (EEA Briefings, 2023);
https://www.eea.europa.eu//publications/
europes-air-quality-status-2023.
11. J. Soares et al., “Health risk assessment of air pollution
and the impact of the new WHO guidelines” (European
PHOTO: QIAN WEIZHONG/VCG VIA GETTY IMAGES
Environment Agency, ETC-HE Report 2022/10, 2022).
12. R. Burnett et al., Proc. Natl. Acad. Sci. U.S.A. 115, 9592
(2018).
13. European Environment Agency, “Air quality in
Europe 2022” (2022); https://www.eea.europa.eu/
publications/air-quality-in-europe-2021.
14. EPA, “Integrated Science Assessment for Oxides of
Nitrogen, Oxides of Sulfur and Particulate Matter
Ecological Criteria” (2020); https://cfpub.epa.gov/
ncea/isa/recordisplay.cfm?deid=349473.
15. P. Huangfu, R. Atkinson, Environ. Int. 144, 105998
(2020).
10.1126/science.abo3801
science.org SCIENCE

RETROSPECTIVE
Roger Guillemin (1924—2024)
Father of neuroendocrinology
By Catherine Rivier
R
oger Guillemin, founder of the field
of neuroendocrinology, died on 21
February at the age of 100. Guillemin
studied the interactions between hor-
mones and the brain and the mecha-
nisms through which such hormones
control basic functions such as growth, re-
production, and the handling of stress. Fu-
eled by a rare intensity, Guillemin doggedly
searched for the way that the hypothalamus
regulated pituitary function, and he isolated
and characterized multiple mediators, which
he called releasing or inhibiting “factors.” His
research led to new approaches to treat dis-
eases such as precocious puberty, pancreatic
tumors, endometriosis, and prostate cancer.
Guillemin was born in Dijon, France, on
11 January 1924. He obtained his MD in 1949
from the University of Lyon. Inspired by a
seminar on stress by endocrinologist Hans
Selye, Guillemin went to work with him at
the University of Montreal, where Guillemin
obtained his PhD in physiology and experi-
mental endocrinology in 1953. This research
formed the basis of his lifelong interest in
the physiological control of pituitary activ-
ity, which he pursued first at Baylor College
of Medicine in Houston, Texas, from 1953 to
1970 and then at the Salk Institute for Bio-
logical Studies in La Jolla, California, where
he remained for the rest of his career.
Convinced from the start that the hypo-
thalamus released substances that traveled
to the anterior pituitary, even though the
concept was highly controversial at the time,
Guillemin tenaciously pursued this line of
inquiry. He faced a highly skeptical scientific
environment, including doubting colleagues
and funding agencies that quickly tired of
supporting research that seemed to lead no-
where. To find the materials he needed, he
sent one of his students, Wylie Vale, to teach
slaughterhouse workers to set aside the right
tissues. When I joined Guillemin’s laboratory
PHOTO: TONY KORODY/SYGMA VIA GETTY IMAGES
in 1969, he had just finished collecting an in-
credible 500,000 sheep hypothalami.
Guillemin then developed methodologies
to extract various substances from the stored
hypothalami and purify them of unwanted
contaminants, procedures for which there
were few established tools. One separation
column that used the recently commercial-
Salk Institute, La Jolla, CA, USA. Email: crivier@salk.edu
SCIENCE science.org
ized dextran gel Sephadex was about 3 m
tall—the technician pouring a hypothalamic
extract into the top of the column had to
communicate by walkie-talkie with the tech-
nician in charge of collecting the effluents.
The workload was incredible, and so were the
technical difficulties, but the laboratory was
always bristling with excitement.
Guillemin did not permit any interruption
of this work. He required absolute dedica-
tion, and he led by example. He expected
that whenever he deemed it necessary, we
would give up whatever occupied us outside
the laboratory and show up for work. More
than once, he demanded that a vacationing
colleague come back because a rumor had
spread that someone was getting close to
the successful characterization of a new hy-
pothalamic substance. The stress of possible
failures—or worse, being scooped by a col-
league—led to substantial health issues that
plagued Guillemin for years.
Yet he succeeded. Guillemin first iden-
tified thyrotropin-releasing factor (TRF),
which stimulates thyroid-stimulating hor-
mone secretion, in 1970. He later identified
gonadotropin-releasing hormone (GnRH),
which regulates luteinizing hormone and
follicle stimulating hormone release; growth-
regulating factor, which stimulates growth
hormone release; and somatostatin, which
inhibits it. For the characterization of TRF
and GnRH, he shared the 1977 Nobel Prize in
Physiology or Medicine.
In 1969, Jonas Salk invited Guillemin to his
new institute and offered him the opportu-
nity to establish a laboratory of neuroendocri-
INSIGHTS | P E R S P E C T I V E S
nology. Guillemin announced to his stunned
Houston group that he was moving his labo-
ratory to La Jolla. I was a married graduate
student at the time. I told Guillemin that I
would gladly follow him but that my husband
would need to find work as well. “What does
he do?” Guillemin asked. I responded that he
was a chemist. “He is hired,” he replied—no
interview, no CV, no questioning of any sort
required. That was typical Guillemin: Once
he had decided on a course of action, nothing
would stop him.
When the laboratory opened at the Salk
Institute, the use of computers was in its
infancy, but Guillemin had already realized
that sophisticated statistical methods were
an absolute requirement for the correct pro-
cessing of complex data. We all had to take
classes in statistics, and he hired a computer
specialist to help us develop and implement
statistical programs.
Perhaps influenced by his French up-
bringing, which had provided education in
a variety of intellectual pursuits, Guillemin
thought that his research should be docu-
mented outside the standard scientific chan-
nels. He enlisted a young, then-unknown,
French philosopher, Bruno Latour, to stay
at the laboratory for 6 months and describe
how research was conducted. Latour’s book,
Laboratory Life (which became a bestseller
and launched his career), depicts the many
ways in which the development of scientific
findings requires constant discussions, inter-
actions, and disagreements.
In addition to being a brilliant and innova-
tive scientist, Guillemin was an artist. He col-
lected pre-Columbian art with a passion and
restored many of the pieces himself. Having
computers at his disposal, Guillemin decided
to channel his artistic mind into designing
posters to advertise his lectures. This idea
was the beginning of what he later called
“abstract impressionism.” His beautiful
digital inkjet prints were soon featured in
high-end galleries. Guillemin also organized
exhibits of his art in collaboration with local
universities throughout the United States,
and he donated a large part of the proceeds
of the sales to student organizations.
My early experiences in Guillemin’s lab
were difficult because men considered
women in academia to be an oddity, and Guil-
lemin was no exception. However, he never
interfered with my work or refused me the
tools I needed. Although he dealt with me as
little as possible, his commitment to his work
taught me that success requires enthusiastic
and unrelenting devotion to the goal.
Guillemin once said that his goal had al-
ways been to help people. Through his re-
search, and all the medical advances it made
possible, he certainly succeeded. j
10.1126/science.adp0647
5 APRIL 2024 • VOL 384 ISSUE 6691 35
INSIGHTS
P OLICY FORUM
ARTIFICIAL INTELLIGENCE
Regulating advanced artificial agents
Governance frameworks should address the prospect
of AI systems that cannot be safely tested
By Michael K. Cohen1,2, Noam Kolt3,4,
Yoshua Bengio5,6, Gillian K. Hadfield2,3,4,7,
Stuart Russell1,2
T
echnical experts and policy-makers
have increasingly emphasized the
need to address extinction risk from
artificial intelligence (AI) systems
that might circumvent safeguards
and thwart attempts to control them
(1). Reinforcement learning (RL) agents
that plan over a long time horizon far more
effectively than humans present particular
risks. Giving an advanced AI system the
objective to maximize its reward and, at
some point, withholding reward from it,
strongly incentivizes the AI system to take
humans out of the loop, if it has the op-
portunity. The incentive to deceive humans
and thwart human control arises not only
for RL agents but for long-term planning
agents (LTPAs) more generally. Because
empirical testing of sufficiently capable
LTPAs is unlikely to uncover these dan-
gerous tendencies, our core regulatory
proposal is simple: Developers should
not be permitted to build sufficiently ca-
pable LTPAs, and the resources required to
build them should be subject to stringent
controls.
Governments are turning their attention
to these risks, alongside current and antic-
ipated risks arising from algorithmic bias,
privacy concerns, and misuse. At a 2023
global summit on AI safety, the attend-
ing countries, including the United States,
United Kingdom, Canada, China, India,
and members of the European Union (EU),
issued a joint statement warning that, as
AI continues to advance, “Substantial risks
may arise from…unintended issues of con-
trol relating to alignment with human in-
tent” (2). This broad consensus concerning
the potential inability to keep advanced AI
1University of California, Berkeley, CA, USA. 2Center for
Human-Compatible Artificial Intelligence, Berkeley, CA, USA.
3University of Toronto, Toronto, Ontario, Canada. 4Schwartz
Reisman Institute for Technology and Society, Toronto,
Ontario, Canada. 5Université de Montréal, Montréal, Québec,
Canada. 6Mila–Quebec AI Institute, Montréal, Québec,
Canada. 7Vector Institute for Artificial Intelligence, Toronto,
Ontario, Canada. Email: mkcohen@berkeley.edu
36 5 APRIL 2024 • VOL 384 ISSUE 6691
under control is also reflected in President
Biden’s 2023 executive order that intro-
duces reporting requirements for AI that
could “eva[de] human control or oversight
through means of deception or obfusca-
tion” (3). Building on these efforts, now is
the time for governments to develop regu-
latory institutions and frameworks that
specifically target the existential risks from
advanced artificial agents.
RISKS FROM LTPAs
RL agents function as follows: They re-
ceive perceptual inputs and take actions,
and certain inputs are typically designated
as “rewards.” An RL agent then aims to
select actions that it expects will lead to
higher rewards. For example, by designat-
“…safety testing is likely
to be either dangerous or
uninformative.”
ing money as a reward, one could train an
RL agent to maximize profit on an online
retail platform (4).
Highly capable and far-sighted RL agents
are likely to accrue reward very successfully.
If plan A leads to more expected reward
than plan B, sufficiently advanced RL agents
would favor the former. Crucially, securing
the ongoing receipt of maximal rewards
with very high probability would require the
agent to achieve extensive control over its
environment, which could have catastrophic
consequences (5–8). One path to maximiz-
ing long-term reward involves an RL agent
acquiring extensive resources and taking
control over all human infrastructure (5, 6),
which would allow it to manipulate its own
reward free from human interference (5).
Additionally, because being shut down by
humans would reduce the expected reward,
sufficiently capable and far-sighted agents
are likely to take steps to preclude that pos-
sibility (7) or if feasible, create new agents
(unimpeded by monitoring or shutdown) to
act on their behalf (5). Progress in AI could
enable such advanced behavior.
So long as an agent’s rewards can
be controlled, it can be incentivized to
achieve complex goals by conditioning the
rewards appropriately. But a sufficiently
capable RL agent could take control of its
rewards, which would give it the incentive
to secure maximal reward single-mindedly.
Constraining the influence that highly
competent agents learn to exert over their
environment is likely to prove extremely
difficult; an intelligent agent could, for ex-
ample, persuade or pay unwitting human
actors to execute important actions on its
behalf (5, 7).
Critically, far-sighted RL agents face
an incentive to develop and execute arbi-
trarily competent long-term plans. Many
AI systems are trained only to achieve
certain immediate outcomes, like cor-
rectly classifying an image. Although such
short-sighted agents could certainly cause
harm, they would likely lack the incentive
to execute protracted schemes to subvert
human control.
Accordingly, we define an LTPA as an
algorithm designed to produce plans, and
to prefer plan A to plan B, when it expects
that plan A is more conducive to a given
goal over a long time horizon. For exam-
ple, an agent trained to maximize profit
on an online retail platform, as proposed
by Suleyman’s “new Turing test” (4), might
productively use such an algorithm and
hinder attempts to interfere with its profit
making. LTPAs include all long-horizon
RL algorithms, including so-called “policy
gradient” methods, which lack an explicit
planning subroutine but are trained to
be as competent as possible. LTPAs also
include algorithms that imitate trained
LTPAs, but not algorithms that merely imi-
tate humans. In the latter case, if plan A
is more competent than any plan a human
could develop, and plan B is a human plan,
an algorithm imitating a human would not
prefer plan A to plan B. The supplemen-
tary materials include a taxonomy situat-
ing LTPAs among other machine learning
systems. Notably, there is no recognizable
horizon length at which risk increases
sharply; accordingly, regulators will have
science.org SCIENCE
to define the length of a long time horizon
according to their risk tolerance.
Losing control of advanced LTPAs, al-
though not the only existential risk from
AI, is the class of risk that we aim to ad-
dress here—and one that necessitates new
forms of government intervention.
A GOVERNANCE PROPOSAL
Although governments have expressed
concern about existential risks from AI,
regulatory proposals do not adequately
address this class of risk (9). The EU AI
Act (10) canvasses a broad array of risks
from AI but does not single out loss of con-
trol of advanced LTPAs. We see promising
first steps from the US and UK—President
Biden’s executive order on AI (3) requires
reports on potentially uncontrollable AI
systems, but it does not seek to constrain
their development or proliferation; the
US and UK AI Safety Institutes are build-
ing capacity for regulators to understand
cutting-edge AI but lack the authority to
control it (11).
Across multiple jurisdictions, following
industry practice, the prevailing regulatory
approach for AI involves empirical safety
testing, most prominently within the UK
AI Safety Institute (2, 3, 10–12). We, how-
ever, argue that for a sufficiently capable
LTPA, safety testing is likely to be either
dangerous or uninformative. Although we
might like to empirically assess whether
an agent would exploit an opportunity to
thwart our control over it, if the agent in
fact has such an opportunity during a test,
the test may be unsafe. Conversely, if it does
not have such an opportunity during a test,
the test is likely to be uninformative with
respect to such risks. This holds for human
agents as well as artificial agents: Consider
a leader appointing a general, but worried
about a coup; if the general is clever, there
is no safe and reliable loyalty test. A candi-
date for the role, like an advanced artificial
agent, would either recognize the test and
behave agreeably or, if possible, execute a
coup during the test.
If an agent is advanced enough to rec-
ognize that it is being tested, then there
is little reason to expect similar behav-
ior in and out of testing. Moreover, an AI
system designed to interact with complex
environments (e.g., human institutions or
biological systems) would likely be able to
discern a simulated test environment from
real-world deployment (because complex
systems can only be simulated approxi-
mately), thereby enabling the AI system to
identify when it is being tested. Although
no current artificial agents are competent
enough to thwart human control, some
have already been found to identify safety
SCIENCE science.org
tests and pause misbehavior (13). Testing could (4). This difficulty arises in part be-
may nonetheless be useful for detecting cause there is currently no robust scientific
some dangerous algorithmic capabilities in method for (ii); computer scientists should
systems that cannot thwart human control. develop one quickly. Perhaps if certain re-
Stepping back, empirical testing is a no- sources could be used to create an AI sys-
toriously ineffective tool for ensuring the tem with the short-term goal of exhibiting
safety of computational systems. For exam- a moderately dangerous capability (i.e.,
ple, extensive testing failed to reveal an er- trying to fail the safety test), that could im-
ror in the Intel Pentium’s arithmetic unit. prove our understanding of the resources
Given that both safety and validity cannot that can produce dangerous capabilities.
be ensured when testing sufficiently ca- Admittedly, listing relevant dangerous
pable LTPAs, governments should estab- capabilities and estimating the resources
lish new regulatory bodies with the legal required to achieve these capabilities will
authority and technical capacity to prevent require considerable research. We suggest
such agents from being built in the first that regulators err on the side of caution
place, no matter the domain. and underestimate the resources required
to develop LTPAs with dangerous capabili-
DEFINING DANGEROUS CAPABILITIES ties. Systems should be considered “dan-
How capable is “sufficiently capable”? Un- gerously capable” if they are trained with
fortunately, we do not know. More cautious enough resources to potentially exhibit
regulators might prevent the development those dangerous capabilities, and regula-
Mandatory reporting and production controls for LTPAs
To prevent unlawful development of dangerously capable long-term planning agents (LTPAs), which may
be difficult to directly detect, reporting requirements would enable regulators to have sufficient visibility into
easier to observe LTPA production resources and code interacting with those resources. Though concern is
ultimately with subsets of production resources and LTPAs (“definition”), these are not easily recognizable, thus
broader recognizable supersets encompassing those subsets (“implementation”) are the focus of regulation.
LTPA PRODUCTION RESOURCES DANGEROUSLY CAPABLE LTPAs
Definition Definition
Information that makes it low-cost to produce LTPAs able to thwart
a dangerously capable LTPA Code used to human control
produce LTPAs
Implementation Implementation
Categories might include, e.g., large foundation LTPAs trained with sufficiently extensive
models, AI training curriculum resources, e.g., compute, data
Shape and size of regulatory implementation categories can be updated periodically to ensure the inclusion of new systems that meet the definition.
of even weak LTPAs; however, regulators tors should not permit the development
seeking to facilitate the development of of dangerously capable LTPAs. To ensure
merely “moderately capable” LTPAs should this occurs, regulators will need to care-
establish protocols to estimate in advance fully monitor and control the resources
whether such systems might have the abil- that could be used to produce dangerously
ity to game safety testing and evade human capable LTPAs. Although this would inter-
control. One factor that regulators could rupt the “move fast” ethos of AI develop-
consider is the resources proposed to be ment, we believe caution is necessary.
used to train LTPAs, including compute, If dangerously capable LTPAs are at
data, and the resources used to develop some point permitted to be developed, rig-
any pretrained models that assist in LTPA orous technical and regulatory work would
training. We propose that policy-makers need to be done first to determine if, when,
(i) establish a list of dangerous capabili- and how to permit this. The possibility
ties, such as those described in President must also be considered that researchers
Biden’s executive order (3), which include and policy-makers fail to identify any ro-
“high levels of performance at…deception bustly safe regulatory regimes that permit
or obfuscation” and “offensive cyber op- the development of dangerously capable
erations through automated vulnerability LTPAs, at least by the time that actors in
discovery”; and (ii) estimate the resources the private sector are able to build them.
needed to develop an LTPA that exhibits It is also worth noting that there might
those capabilities. We do not believe that be a path to building AI systems that can
existing systems exhibit those capabilities, be proved mathematically to avoid certain
and it is very difficult to predict when they dangerous behaviors (7), but such formal
5 APRIL 2024 • VOL 384 ISSUE 6691 37
INSIGHTS | P O L I C Y F O RU M
guarantees appear highly unlikely for any
AI systems built similarly to the most pow-
erful systems today (13).
MONITORING AND REPORTING
Just as nuclear regulation extends to con-
trolling uranium, AI regulation must ex-
tend to controlling the resources needed
to produce dangerously capable LTPAs. We
define production resources (PRs) as any
information that makes the production of
a dangerously capable LTPA cheaper than a
threshold determined by regulators accord-
ing to their risk tolerance. Unlike uranium,
a PR is not a physical resource—it could
include any AI model trained beyond a cer-
tain compute threshold (14). Fortunately,
regulators could detect such PRs by fol-
lowing the hardware required to produce
them. (Some of this hardware could be
regulated as well, including semiconductor
chips and data centers, but that is outside
our focus here.) To limit the proliferation
of PRs, expanding on Hadfield et al. (15),
Avin et al. (12), and President Biden’s ex-
ecutive order (3), we propose that develop-
ers be required to report (a) relevant facts
about the PR [if the PR is an AI model,
this might include (i) the input/output
properties, (ii) the data collection process
for training it, (iii) the training objective,
and (iv) documented behavior in test set-
tings, but not typically the model weights
themselves]; (b) the specific machines on
which the PR is stored and their locations;
(c) all code run on these machines after the
PR is created; and (d) all outputs of that
code. With the context provided by point
(a), governments could monitor the code
that interacts with PRs, allowing them to
detect the development of (unlawful) LT-
PAs (see the figure). In addition, if a com-
pany offers users application programming
interface (API) access to a PR, users should
be required to report the code on the us-
er’s machine that interacts with the API.
Details of the reporting requirements will
need to be updated in response to techno-
logical advances that lead to changes in the
resources and processes needed to produce
dangerously capable LTPAs. Finally, report-
ing procedures could be complemented by
protecting and rewarding whistleblowers
who uncover misconduct.
PRODUCTION CONTROLS
Given sufficient visibility into the re-
sources for producing LTPAs, regulators
could then prohibit the production of dan-
gerously capable LTPAs. Developers that
are unsure whether a proposed AI system
meets the definition of dangerously capa-
ble LTPA could inquire with the relevant
regulator prior to development. Regulators
38 5 APRIL 2024 • VOL 384 ISSUE 6691
could also control the transfer of large pre-
trained models or other relevant resources.
Further, regulators could make it unlawful
for other actors to use AI systems that fail
to comply with these requirements (15).
Taken together, controls on the develop-
ment, use, and dissemination of produc-
tion resources will substantially reduce the
likelihood of these resources being used to
build dangerously capable LTPAs.
ENFORCEMENT MECHANISMS
To ensure compliance with these reporting
requirements and usage controls, regula-
tors may need to be authorized to (i) issue
legal orders that compel organizations to
report production resources and man-
date the cessation of prohibited activities;
(ii) audit an organization’s activities and,
where necessary, restrict an organization’s
access to certain resources, such as cloud
computing; (iii) impose fines on noncom-
pliant organizations; and (iv) as in finan-
cial regulation, impose personal liability
on key individuals in noncompliant orga-
nizations. If business leaders can be held
to account for breaching corporate duties,
then surely they should face similar conse-
quences for irresponsibly handling one of
the world’s most dangerous technologies.
REGULATORY INSTITUTIONS
We have addressed our discussion to “reg-
ulators” but have not proposed specific
regulatory institutions for addressing the
risks from LTPAs. This issue will need to
be approached differently in different
countries. That being said, we expect that
whereas other risks from AI might be ad-
dressed primarily through domain-specific
regulation (e.g., financial regulation and
health care regulation), the risk of loss of
control of AI likely requires specialized
regulation and the establishment of new
regulatory institutions. This specialized
regulation could nevertheless benefit from
the existing expertise of domain-specific
regulators, including with developing
frameworks for monitoring PRs. Critically,
because the risks from LTPAs are global,
regulatory efforts cannot stop at national
borders. International cooperation is vital.
BROADER CONCERNS
LTPAs, of course, are not the only type of
AI system that poses substantial and even
existential risks. Accordingly, we suggest
that empirical testing, which is inadequate
for sufficiently advanced LTPAs, could
nevertheless substantially improve the
safety of some other types of AI. At the
same time, the governance regime that we
propose could be adapted to other AI sys-
tems. Although our proposal for governing
LTPAs fills an important gap, further insti-
tutional mechanisms will likely be needed
to mitigate the risks posed by advanced ar-
tificial agents. j
REFERENCES AND NOTES
1. G. Hinton et al., “Statement on AI risk” (Center
for AI Safety, May 2023); https://www.safe.ai/
statement-on-ai-risk.
2. United Kingdom Department for Science, Innovation,
and Technology, United Kingdom Foreign,
Commonwealth, and Development Office, United
Kingdom Prime Minister’s Office, “The Bletchley
Declaration by countries attending the AI Safety
Summit, 1-2 November 2023” (Gov.uk, November
2023); https://www.gov.uk/government/publications/
ai-safety-summit-2023-the-bletchley-declaration/
the-bletchley-declaration-by-countries-attending-the-
ai-safety-summit-1-2-november-2023.
3. J. Biden, “Executive order on the safe, secure,
and trustworthy development and use of arti-
ficial intelligence” (The White House, October
2023); https://www.whitehouse.gov/briefing-
room/presidential-actions/2023/10/30/
executive-order-on-the-safe-secure-and-trustworthy-
development-and-use-of-artificial-intelligence/.
4. M. Suleyman, The Coming Wave (Penguin, September
2023).
5. M. K. Cohen, M. Hutter, M. A. Osborne, AI Mag. 43, 282
(2022).
6. S. Zhuang, D. Hadfield-Menell, Adv. Neural Inf. Process.
Syst. 33, 15763 (2020).
7. S. Russell, Human Compatible: AI and the Problem of
Control (Viking, 2019).
8. A. Turner, L. Smith, R. Shah, A. Critch, P. Tadepalli, Adv.
Neural Inf. Process. Syst. 34, 23063 (2021).
9. N. Kolt, “Algorithmic black swans”(Washington
University Law Review, October 2023); https://ssrn.
com/abstract=4370566.
10. European Commission, “Proposal for a regulation of
the European parliament and of the council laying down
harmonised rules on artificial intelligence (Artificial
Intelligence Act) and amending certain Union legisla-
tive acts” (COM/2021/0206, European Commission,
January 2024); https://eur-lex.europa.eu/
legal-content/EN/TXT/?uri=celex%3A52021PC0206.
11. United Kingdom Department for Science, Innovation
and Technology, “Introducing the AI Safety Institute”
(Gov.uk, November 2023); https://assets.publishing.
service.gov.uk/media/65438d159e05fd0014be7bd9/
introducing-ai-safety-institute-web-accessible.pdf.
12. S. Avin et al., Science 374, 1327 (2021).
13. J. Lehman et al., Artif. Life 26, 274 (2020).
14. Y. Shavit, arXiv:2303.11341 [cs.LG] (2023).
15. G. Hadfield, M. Cuéllar, T. O’Reilly, “It’s time to create
a national registry for large AI models” (Carnegie
Endowment for International Peace, July 2023);
https://carnegieendowment.org/2023/07/12/
it-s-time-to-create-national-registry-for-large-ai-
models-pub-90180.
ACKNOWLEDGEMENTS
M.K.C. and N.K. contributed equally. The authors thank R.
Grosse, A. Barto, and P. Christiano for feedback on earlier
versions of the manuscript. M.K.C. commenced this project
at Oxford University. M.K.C. and S.R. are supported by the
Open Philanthropy Foundation. N.K. is supported by a
Vanier Canada Graduate Scholarship. Y.B. is supported by a
Canadian Institute for Advanced Research (CIFAR) AI Chair
and a Natural Sciences and Engineering Research Council of
Canada Discovery Grant. G.K.H. is supported by the Schwartz
Reisman Institute Chair in Technology and Society and a
Canada CIFAR AI Chair at the Vector Institute for Artificial
Intelligence. G.K.H. and S.R. are supported by AI2050 Senior
Fellowships from Schmidt Futures.
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adl0625
10.1126/science.adl0625
science.org SCIENCE

B O OKS et al .
ECOLOGY
Digitizing nature
Digital tools are transforming conservation work but must
be carefully deployed, argues an environmental scholar
By Jennifer Garard1,2,3 and
H. Damon Matthews1,2
H
uman impacts on the planet are se-
vere and worsening, with few major
successes in environmental gover-
nance to celebrate in recent years.
In parallel, digital innovations are
emerging at an unprecedented rate,
and digital tools and platforms are increas-
ingly available to users around the world.
These two forces are transforming societies
and are also deeply intertwined. In her fi-
nal book, Gaia’s Web, the late environmen-
tal scholar Karen Bakker offers readers a
poignant expression of hope that “Digital
Earth” technologies—digital tools designed
to capture, interpret, and share data about
biodiversity and planetary systems—may
provide opportunities to tackle some of the
key environmental challenges of our time,
while remaining wary of the inherent risks
and trade-offs of such technologies.
In an accessible narrative style, Bakker
seamlessly weaves together an analysis of
PHOTO: AFP VIA GETTY IMAGES
concrete and clearly described digital tools, a
discussion of major potential risks associated
The reviewers are at 1Sustainability in the Digital Age,
Montréal, QC, Canada; 2Department of Geography, Planning
and Environment, Concordia University, Montréal, QC,
Canada; and 3Future Earth Canada Hub, Montréal, QC,
Canada. Email: jennifer.garard@sustainabilitydigitalage.org
SCIENCE science.org
with their deployment, and key questions
about how to design for equity and inclusivity.
Interspersed are parables intended to “evoke
dilemmas at the confluence of digital trans-
formation and environmental sustainability.”
Bakker begins by focusing on regen-
eration, exploring the use of Digital Earth
technologies to address issues related to
oceans and fishing, biodiversity and wild-
life, climate change, and finance. Here, she
describes the mobile application Wildbook,
which uses facial recognition
technology to detect and track
individual animals. Wildbook
is being used by conservation-
ists and decision-makers to slow
biodiversity loss and protect vul-
nerable species. However, it is
also being used by poachers for
precision hunting, directly un-
Gaia’s Web
dermining conservation efforts.
Karen Bakker
This paradox is true of many ap- MIT Press, 2024. 288 pp. such actors can contribute to col-
plications of digital technologies
that aim to support environmental action.
In the second part of the book, Bakker
turns to “instantiating,” referring to the con-
cept of bringing an idea into being. Here, she
explores the fusion of digital and biological
technologies. Her discussions frequently in-
clude a speculative angle and look to what
future technologies could bring.
Bakker’s exploration of the frontiers
An eco-guard installs a camera trap in Campo Ma’an
National Park in Cameroon in 2022.
of Digital Earth technologies ranges from
mobile marine protected areas and digi-
tal whale lanes to virtual reality platforms
informed by Indigenous perspectives that
explore new reciprocal relationships with
nature. Some of her more speculative sug-
gestions—giving a democratic voice to
nonhuman entities, for example, or imag-
ining a future where plant-based batteries,
biohybrid buildings, and DNA-based data
storage supplant traditional digital tech-
nologies—may seem far-fetched, but these
ideas are based solidly on existing technol-
ogy and debates. Bakker also raises over-
arching concerns that are associated with
many Digital Earth technologies, such as
the possibility of digital surveillance, the
perils of innovation driven only by profit
motive, and the troubling environmental
impacts of technology development and
implementation.
The question of how to design digital so-
lutions for equity and inclusivity emerges
throughout the book as a particularly ur-
gent one. Bakker argues convincingly that
understanding how to develop inclusive
technologies and protect humans and the
environment from the risks and damages
of digital technologies is just as impor-
tant as figuring out how to best leverage
these technologies for transformations to
sustainability.
A clear message throughout the book is
that Digital Earth technologies are already
being developed and deployed and that many
more advances may be just around the cor-
ner. Bakker demonstrates a need for collabor-
ative approaches to assessing and evaluating
these technologies and advocates for the cen-
tral role of environmental governance, noting
the “interplay between digital and
governance innovation.” She does
not shy away from critiquing is-
sues that have arisen as a result
of private-sector control over
technology and innovation. How-
ever, she also highlights positive
examples of initiatives being led
by the private sector and leaves
the door open for a future where
laborative solutions.
As Bakker reflects, “digital transforma-
tion is affecting the trajectory of environmen-
tal change, and, conversely, environmental
change will shape the future trajectory of
digital innovation.” It is imperative to con-
sider these two issues together. Gaia’s Web is
thus both a timely and an incredibly impor-
tant contribution. j
10.1126/science.ado4359
5 APRIL 2024 • VOL 384 ISSUE 6691 39
INSIGHTS | B O O K S
BIOLOGY
Genes, in context
As biologists have long known, the genome is only
one piece in the puzzle of life
By Michael A. Goldman
O
n 26 June 2000, US President Bill
Clinton, future National Institutes
of Health Director Francis Collins,
Celera’s Craig Venter, and British
Prime Minister Tony Blair made a joint
announcement at the White House
about the sequencing of the human genome.
“We are here to celebrate the completion of
the first survey of the entire human genome,”
stated Clinton. “Without a doubt, this is the
most important, most wondrous map ever
produced by humankind.” Poetically, Col-
lins added, “It is humbling for me and awe-
inspiring to realize that we have
caught the first glimpse of our
own instruction book, previously
known only to God.” The speakers’
hyperbole created an expectation
that knowing the sequence would
solve most problems in medicine,
and the disappointment since has
been palpable.
In How Life Works, journal-
ist Philip Ball delights in spell-
ing out the inadequacies of the
model of the human genome as
the code for human life in beauti-
ful and enthralling detail. It is a
solid book—well researched and
an enjoyable read—and it offers
readers a great chance to review
and admire the beauty and com-
plexity of life. But the edifice Ball
builds up and deftly tears down,
if it ever existed at all, had not been a ma-
jor factor in genetics or evolutionary biology
for decades before that historic turn-of-the-
century announcement.
The relationship between genes and phe-
notypes is anything but simple, notes Ball,
writing: “One might…infer from Mendel’s
experiments on peas that each trait of an
organism is encoded in a single gene, and is
inherited discretely… but [such traits] are the
exception rather than the rule.” “We now see
that most of the genes in the human genome
do not have any assigned function; that is,
we can’t really say what they ‘do,’” he con-
cludes, by which he seems to mean that genes
The reviewer is at the Department of Biology, San Francisco
State University, San Francisco, CA 94132, USA.
Email: goldman@sfsu.edu
40 5 APRIL 2024 • VOL 384 ISSUE 6691
have multiple functions. This phenomenon—
pleiotropy—has long been known.
Ball firmly establishes the idea that a cata-
log of genes (and their associated messenger
RNAs and proteins) is not sufficient informa-
tion from which to reconstruct an organism.
Organisms, he notes, are also the result of a
process of development that unfolds in a par-
ticular environment over the course of time,
a process that is frequently dependent on
maternal factors in the egg cytoplasm, gravi-
tational effects, and orientation in the ovary.
A tweak in the relative amounts of differ-
ent developmental components, or even the
omission of one product entirely, can often
Craig Venter and Francis Collins celebrate the sequencing of the human genome.
be compensated for in the process of devel-
opment, resulting in a functional organism
through a kind of buffering, or canalization
(1). These observations capture an inherent
aspect of life long accepted by developmental
biologists and geneticists.
Ball is right to criticize the simplistic ap-
proach DNA sequencers took with respect to
the Human Genome Project(s), setting out
as if to describe a catalog of genes in detail
despite knowing already that <5% of the ge-
nome was actually “genic” material. But the
fact that much of the extragenic material
plays a regulatory role in gene expression is
not a compelling argument for the reduced
importance of genes. If anything, the idea
that phenotype is affected by gene regulation
or by epigenetic mechanisms bolsters the ar-
gument for a central role for genes in biology.
How Life Works:
A User’s Guide to the
New Biology
Philip Ball
University of Chicago Press,
2023. 552 pp.
The book provides a very comprehensive
description of the importance of context in
genetics, a frustrating complication in gene
expression that has not yet been fully appre-
ciated. Although details have been slow to
emerge, very clear observation in Drosophila
nearly a century ago first suggested the pres-
ence of context effects (2). But again, such
evidence reinforces the primacy of genes and
their regulatory environments in our under-
standing of life.
Although he does not use the term of-
ten, Ball presents a cogent picture of, and
argument for, systems biology, following
the seminal thinking of physiologist Denis
Noble and others [e.g., (3)]. Some
authors, including Noble and
Ball, consider the potential role
of “agency” in biology. How Life
Works has an entire chapter on
this topic, in which Ball argues
that concepts such as agency and
purpose “are not optional add-ons
for the philosophically inclined,
once we have solved all the mi-
nutiae of how life works at the
microscopic scale. Rather, they sit
at the core of life itself.” The idea
that there is a gap between what
we can explain mechanistically
and what we observe has gained
traction in philosophy of science
circles but has been resisted by
mainstream science.
How Life Works is not as wide-
ranging as its title suggests, con-
centrating mostly on human examples. It
is not heavily annotated either, although a
bibliography did help me track down sources
when I had a question or wanted to look
deeper.
Despite some shortcomings, I would not
hesitate to recommend this book to col-
leagues, who will find it well researched,
interesting, and stimulating. Ball’s very elo-
quent presentation is not likely to change
the trajectory of biology, but it is a useful re-
PHOTO: NY DAILY NEWS/GETTY IMAGES
minder to continue to interpret genetics and
genomics with care. j
REFERENCES AND NOTES
1. C. H. Waddington, Nature 150, 563 (1942).
2. H. J. Muller, J. Genet. 22, 299 (1930).
3. R. Noble, D. Noble, Biol. J. Linn. Soc. 139, 357 (2022).
10.1126/science.ado0262
science.org SCIENCE
 INNOVATION
PRIZE ES SAY
GRAND PRIZE WINNER
Aditya M. Kunjapur
Planting a chemical flag
Aditya Kunjapur
received an
on antigens
undergraduate
degree from the
Next-generation live vaccines are created by autonomous
University of production of nitrated antigens
Texas at Austin
and a PhD from
the Massachusetts Institute of By Aditya M. Kunjapur ing cells to access a broader chemical rep-
Technology. After completing his ertoire of building blocks can improve live

postdoctoral fellowship at Har-
vard Medical School, he started
his laboratory in the Department
of Chemical and Biomolecular
Engineering at the University
of Delaware in December 2018.
His research seeks to expand
the breadth of molecules that
microbes can make and where
microbes can make them by
programming cells to create
and harness new-to-nature
building blocks. www.science.org/
PHOTOS: (TOP TO BOTTOM) COURTESY OF ADITYA KUNJAPUR, TAKEN BY AICHE STAFF; COURTESY OF WASSWA WILLIAM; COURTESY OF KHALIL RAMADI, TED CONFERENCES, LLC
B
acterial infections are a common
cause of death across the globe and are
an increasing threat as the prevalence
of antibiotic resistance rises. Vaccines
directed against bacterial pathogens
prevent the spread of disease without
the need for antibiotics and have an esti-
mated global market size of 39.6 billion USD
by 2030, with a compound annual growth
rate of 7.7% (1). However, vaccines for bac-
terial pathogens are difficult to design ow-
ing to the ability of these organisms to hide
their most potent antigens from the immune
bacterial vaccine efficacy. Specifically, we
seek to engineer cells to autonomously gener-
ate molecules chemically modified to elevate
the response of immune cells to certain anti-
gens that are present on pathogens. We aim
to harness this technology to increase the
recognition of weakly immunogenic antigens
that are common to pathogenic bacterial spe-
cies. Should our strategy prove effective, it
could serve as a broad platform to increase
the immunogenicity of vaccines for bacterial
pathogens—from recombinant protein vac-
cines to live attenuated bacterial vaccines.

doi/10.1126/science.ado4537
system (2). Weakened forms of live bacteria
are some of the first vaccines developed and
have some advantages as candidate vaccines
compared to individual bacterial proteins
(3). Yet, there are few options that balance
safety and efficacy of live bacterial vaccines
other than to control attenuation and dosage
(4). My laboratory conducts several avenues
of research to address this unmet need for
improved bacterial vaccines as the antibiotic
resistance crisis looms.
Our primary hypothesis is that engineer-
Our technology is based on over 15 years
of research that demonstrates that the modi-
fication of self-proteins with the nonstan-
dard amino acid para-nitro-L-phenylalanine
(nitro-Phe) can break immune self-tolerance.
Through genetic engineering, bacterial cells
supplied with chemically synthesized nitro-
Phe can produce proteins that are modified
to contain only a single nitro-Phe residue
substituted on the protein surface (see the
figure). The administration of these recombi-
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

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 41

mation of antibodies, which bind
to the modified and unmodified
protein alike. The formation of
such cross-reactive antibodies
upon administration of a ni-
trated autologous protein has
been observed using modified tu-
mor necrosis factor–a (5–7), C5a
(8), and cancer-associated anti-
gens (9–11), to enlist the immune
system to combat rogue self-
proteins associated with chronic
inflammation or cancer. These
prior studies suggest that the
nitrated epitope aids in antigen
presentation owing to increased
affinity between the nitrated
peptide epitope and major his-
tocompatibility class II (MHC-II)
proteins or in T helper cell recep-
tor recognition of the complex of
peptide bound by MHC-II. Either
way, this leads to T helper cell–
mediated activation of the auto-
antibody response. The highly
electron-withdrawing nitro ring
substituent may increase the
tendency of the side chain to in-
teract with certain other amino
acid side chains.
Given the demonstrated abil-
ity of the nitrated residue to aid
with either antigen presentation
or recognition, we are interested
in investigating the potential for this immu-
nochemical strategy to increase the immuno-
genicity of foreign antigens. We posit that the
formation of nitrated epitopes in live bacte-
rial vaccines could allow us to effectively label
target protein antigens and leverage the tro-
pism, adjuvanticity, and sustained local pro-
duction of a desired antigen enabled by a live
microbe. In principle, the nitro-Phe–modified
protein produced by the engineered bacteria
within a patient would lead to a targeted,
sustained, and protective immune response
toward bacterial pathogens that contain the
wild-type form of the protein. Unfortunately,
incorporation of nitro-Phe into bacterial pro-
teins currently relies on external provision
of chemically synthesized nitro-Phe to cells,
which may not be as readily compatible with
administration of live vaccines.
To address this problem, we recently pro-
grammed bacterial cells to biosynthesize
their own nitro-Phe. We further programmed
these cells to then site-specifically incorporate
this biosynthesized nonstandard amino acid
within target recombinant protein sequences.
Biosynthesis of the nitro chemical functional
Department of Chemical and Biomolecular Engineering,
University of Delaware, Newark, DE, USA.
Email: kunjapur@udel.edu
42 5 APRIL 2024 • VOL 384 ISSUE 6691
Designing bacteria that create and harness
an immunogenic amino acid
The concept of using an expanded genetic code to design more efficacious live
vaccines is depicted here. The top panel is an illustration of immune cells interacting
with a chemically modified antigen produced by bacteria (represented by the ball
and flag) and subsequently mounting a humoral response toward unmodified
versions of the antigen. The bottom panel depicts the engineered metabolic
pathway that shows how the bacteria autonomously generate an immunogenic
amino acid that becomes a chemical modification within a target antigen.
Glucose Chorismate Amino-phenylpyruvate Nitro-Phe
HO O OH
O O
O OH
O tutes of Health Director’s New
OH OH OH
O NH2
HO OH O H2N O2N
OH OH
group in natural or engineered metabolites
is uncommon, and nitro-Phe is an unnatural
target. My first PhD student, Dr. Neil Butler,
successfully constructed a metabolic pathway
to form this target molecule from simple car-
bon sources at concentrations comparable to
those used for external supplementation. In
parallel, he achieved high-fidelity incorpora-
tion of nitro-Phe within reporter proteins (12).
Despite early breakthroughs in our
research, we faced complex technical
challenges when combining nitro-Phe bio-
synthesis and incorporation. The transla-
tion of reporter protein was sensitive to the
timing and level of expression of many het-
erologous genes, which lowered the fidelity
and yield of nitro-Phe incorporation within
the reporter protein. We surmounted this
obstacle by performing genome mining, host
strain engineering, and gene expression op-
timization. The result was a strain capable
of autonomous production of nitrated anti-
gens from simple carbon sources, which we
reported in July 2023 (12).
We faced another obstacle of institutional
reluctance to advance our intellectual prop-
erty (IP) application owing to pandemic-
induced financial constraints. However, I
navigated this by footing the bill for the Patent
Cooperation Treaty patent application from
my personal funds in exchange
for ownership of the IP. In 2021,
I was awarded the Langer Prize
in Innovation and Entrepreneur-
ial Excellence by the American
Institute of Chemical Engineers,
which enabled me to reimburse
myself for these business ex-
penses and transfer IP owner-
ship to Nitro Biosciences, Inc,
the new entity that Dr. Butler
and I cofounded. Including the
Langer Prize, our commercializa-
tion efforts have raised roughly
$200,000 in nondilutive funds
either to Nitro Biosciences or to
university accounts focused on
technology translation through
philanthropy, the University of
Delaware Horn Entrepreneur-
ship center, the Delaware Bio-
technology Institute, and the
National Science Foundation
Innovation Corps program. We
have also participated in na-
tional mentorship programs in-
cluding Nucleate and Blueprint
by the Massachusetts Institute of
Technology Engine.
Funded by a National Insti-
Innovator Award, my laboratory
can now delve deeper into the
interaction between nitro-Phe
and immune cells as well as strategize how
to make the technology more accessible. An
outstanding question is whether the intro-
duction of nitro-Phe in a foreign protein can
increase its immunogenicity. Ultimately, we
aspire to harness our nitration technology to
deliver safe, efficacious, and accessible vac-
cines that protect against bacterial pathogens
in the most-burdened regions of the globe. j
REFERENCES AND NOTES
1. Market size estimates available online vary widely and
have not been independently verified. This estimate was
sourced from Verified Market Research in March 2024
at https://www.verifiedmarketresearch.com/product/
bacterial-vaccines-market/.
2. M. W. van der Woude, A. J. Bäumler, Microbiol. Rev. 17,
581 (2004).
3. E. Medina, C. A. Guzmán, Vaccine 19, 1573 (2001).
4. J. E. Galen, R. Curtiss 3rd, Vaccine 32, 4376 (2014).
GRAPHIC: A. M. KUNJAPUR ADAPTED BY A. MASTIN/SCIENCE
5. J. Grünewald et al., Proc. Natl. Acad. Sci. U.S.A. 105,
11276 (2008).
6. J. Grünewald et al., Proc. Natl. Acad. Sci. U.S.A. 106, 4337
(2009).
7. V. Gauba et al., Proc. Natl. Acad. Sci. U.S.A. 108, 12821
(2011).
8. C. Kessel et al., Arthritis Rheumatol. 66, 610 (2014).
9. H. Tian et al., Cancer Lett. 430, 79 (2018).
10. L. Jiang et al., J. Immunol. Res. 2019, 7914326 (2019).
11. H. Tian et al., Cancer Lett. 476, 170 (2020).
12. N. D. Butler, S. Sen, L. B. Brown, M. Lin, A. M. Kunjapur,
Nat. Chem. Biol. 19, 911 (2023).
ACKNOWLEDGMENTS
A.M.K. is a cofounder of Nitro Biosciences and currently the
president of its board.
10.1126/science.ado4537
science.org SCIENCE
 RESEARCH
IN S CIENCE JOURNAL S Edited by Michael Funk

FARMING PRACTICES

Diversification for the win

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

Strategies to promote biodiversity, such as growing

coffee under other trees, promote positive social and
environmental outcomes.

QUANTUM SIMULATION
Departing from
universality
Very different many-body sys-
tems can exhibit similar behavior
if they belong to the same “uni-
versality class.” This behavior is
well established at low tempera-
tures, but the dynamics at finite
PHOTO: CARL DE SOUZA/AFP VIA GETTY IMAGES
followed the scaling expected
from the conjectured Kardar-
Parisi-Zhang (KPZ) universality
class. However, going to higher
moments revealed deviations
from the KPZ conjecture, indi-
cating that a fuller theoretical
picture is needed to describe the
dynamics. —JS
Science p. 48, 10.1126/science.adi7877
complex that holds duplicated
sister DNAs together. Cohesion
is established during DNA
replication, but how sister DNAs
become trapped inside cohesin
rings is not clear. Using single-
molecule imaging, Cameron et
al. demonstrated that cohesins
are pushed along DNA by the
replication machinery (repli-
emphasizing a crucial link
between sister chromatid cohe-
sion and the termination of DNA
replication. —DJ
Science p. 119, 10.1126/science.adf0224
PEROVSKITES
2D and metal free
Perovskites have been synthe-

temperatures are more difficult
to address both theoretically
and experimentally. Rosenberg
et al. studied magnetization
dynamics in a chain of 46 super-
conducting qubits simulating
the one-dimensional Heisenberg
model. The mean and variance
of the magnetization transferred
across the center of the chain
CELL BIOLOGY
Sister bonding
on the move
Precise separation of dupli-
cated chromosomes during
cell division is facilitated by
cohesin, a ring-shaped protein
some) until they converge with
another replisome. Whereas
replisomes disassemble upon
fork convergence during DNA
replication termination, cohesins
persist, establishing cohesion
at replication termination sites.
Disrupting replisome disas-
sembly after fork convergence
in living cells hinders cohesion,
sized with only organic cations,
but the need to maintain charge
neutrality within the structural
constraints for perovskites has
precluded the synthesis of the
analogous two-dimensional (2D)
phases. Choi et al. show that the
addition of ammonium cations
at the interstitial edge center
sites balances charge and adds

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 43

RESEARCH | I N S C I E N C E J O U R NA L S

stabilizing hydrogen-bonding very attractive efficiency across

interactions. These materials
have layers bonded by van der
Waals interactions that allow
monolayer exfoliation, and their
high dielectric constant was
exploited in field-effect transis-
tors. —PDS
Science p. 60, 10.1126/science.adk8912
ARCHAEOLOGY
Deep history of the
Blackfoot Confederacy
Genomic analysis of the ancient
Indigenous peoples of North
America has proven invalu-
able in charting the peopling of
the continent. By studying the
genomes of modern Blackfoot
individuals and historical
remains and then compar-
ing them with other ancient
genetic lineages, First Rider
et al. discovered a previously
unidentified genetic lineage
that diverged from others in the
Late Pleistocene. These findings
are consistent with the oral
traditions of Blackfoot peoples
that assert their deep temporal
presence in North America. The
authors further suggest that
by demonstrating the antiquity
of their presence, their find-
ings may assist members of
the Blackfoot Confederacy in
enhancing treaty negotiations
and expanding Indigenous
rights. —MSA
Sci. Adv. (2024) 10.1126/sciadv.adl6595
THERMOELECTRICS
Trapping carriers
Developing highly efficient
thermoelectric devices is chal-
lenging because the materials
tend to have properties that are
coupled in a way that degrades
efficiency. Jia et al. found that
careful doping of lead telluride
creates a material with very
mid-range temperatures. —BG Edited by Caroline Ash
Science p. 81, 10.1126/science.adj8175 IN OTHER JOURNALS and Jesse Smith
PLANT SCIENCE
Zipping up reproduction
Self-pollinating plants are useful
for agriculture because favor-
able traits can easily be passed
on to offspring. Domesticated
tomatoes achieve this anatomi-
cally by forming a cone around
the male anthers, which ensures
that pollen can easily reach the
stigma. The anther cone is held
together by a dense network of
hairs called zipper trichomes.
Wu et al. identified a set of
homeodomain–leucine zipper
(HD-ZIP) genes that regulate the
formation of these trichomes.
Simultaneously, these genes
regulate the length of the female
style, allowing coordinated
development of a self-polli-
nating reproductive structure.
Expression of the HD-ZIPs was
found to be lower in wild tomato
plants, which may explain why
they do not develop the anther
cone and are able to receive pol-
len from another plant. —MRS
Science p. 124, 10.1126/science.adl1982
INFECTIOUS DISEASE
CANCER granulocyte-macrophage col-
Improving antibodies ony-stimulating factor (GM-CSF)
against NiV Hijacked macrophages from transformed lung epithe-
Nipah virus (NiV) is a zoonotic aid lung cancer lium. GM-CSF–PPARg signaling
pathogen that can cause severe Lung adenocarcinoma (LUAD) is modulated cholesterol synthesis
disease and death in exposed the most common form of lung from macrophages and meta-
humans. Unfortunately, no cancer, arising from transformed bolically “rewired” macrophages
approved vaccines or therapeu- lung epithelial cells. Mutations within the tumor microenviron-
tics are currently available. To in epidermal growth factor ment to provide nutrients to
advance therapeutic options receptor (EGFR) signaling are sustain lung cancer growth.
for NiV, Zeitlin et al. tested the frequently found in LUAD, and —PNK
protective efficacy of an antibody targeted drugs that inhibit EGFR Cancer Discov. (2024)
targeting the prefusion conforma- can be used to treat EGFR+ 10.1158/2159-8290.CD-23-0434
tion of the NiV F protein, hu1F5, tumors, but unfortunately, drug
in hamsters and African green resistance often develops. In
ENZYME ENGINEERING
monkeys infected with NiV. Hu1F5 searching for new treatment
administration at 1 and 5 days strategies, Kuhlmann-Hogan A better, redder luciferase
after infection conferred protec- et al. explored how antitumor Bioluminescence in fireflies
PHOTO: BLICKWINKEL/ALAMY STOCK PHOTO

high thermoelectric efficiency.
The authors generated cation
defects and traps for the charge
carriers, in this case holes. The
traps help to prevent scattering
but also release holes at higher
temperatures to maintain a con-
stant carrier capacity. The result
is a thermoelectric material with
tion in both the hamsters and the
monkeys, respectively, outper-
forming a previously developed
antibody, m102.4. Together, these
results support further clinical
development of hu1F5 for NiV
infection. —CM
Sci. Transl. Med. (2024)
10.1126/scitranslmed.adl2055
immune responses might
be impaired in LUAD. Tumor
growth results in the accumula-
tion of a type of immune cell
called macrophages within the
lungs. The authors managed
to induce tumor-associated
macrophages by elevated
secretion of surfactant and
and some other organisms is
produced by an enzyme-cat-
alyzed oxidation reaction that
yields a product in an excited
state that can emit light. The
enzyme proficiency and active
site structure are important for
determining the amount of light
produced and the wavelength

44 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 NEUROSCIENCE
PRIMATE PLAYTIME Regulating the
Playful even sense of touch
when hungry The sense of touch relies
on mechanically activated

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.
performed high-throughput
mutagenesis and screening
on firefly luciferase to identify
single mutants with an emission
peak shifted toward red. The
large scale of the screen pro-
vided mechanistic insights that
may also be useful for future
rational engineering. —MAF
Biochemistry (2024)
10.1021/acs.biochem.3c00708
QUANTUM SIMULATION
Erasing errors
Error correction is a neces-
sary component of quantum
computing and is used to
counter decoherence. In ana-
log quantum simulation, for
example, using ultracold atoms
in optical lattices, the effects of
imperfections can also accu-
mulate but have seldom been
addressed with error-correcting
protocols. Hur et al. devised
such a protocol for atoms in a
two-dimensional optical lattice
that undergo a Mott insulator
(MI)–superfluid transition. The
protocol was able to distinguish
between correlated particle-
hole pairs that occur in a MI
due to quantum fluctuations
and holes that are a conse-
quence of finite temperature or
atom loss. Correcting for those
uncorrelated holes enabled the
researchers to reliably measure
a two-dimensional correla-
tor that can serve as an order
parameter for the MI state. —JS
Phys. Rev. X (2024)
10.1103/PhysRevX.14.011003
CELL ENGINEERING
Remote control
anticancer macrophages
Cells of the immune system can
be engineered into highly effec-
tive anticancer agents, but tight
control of such cells is essential
to prevent side effects and
systemic toxicity. Xue et al. have
worked out how to control anti-
tumor macrophages locally and
temporally in a mouse model.
They designed macrophages
that produce interferon-g
(IFN-g), which converts the
macrophages into the antitumor
M1 phenotype under the control
of a heat shock (heat-sensitive)
promotor. Expression of IFN-g
is controlled with a wireless
heating device on an animal
controlled from a smartphone or
computer. In the model system,
activation of macrophages
decreased metastasis and
increased survival. Although the
demonstration was in skin can-
cer, other technologies such as
ultrasound might allow deeper
tissue penetration into deep-
seated tumors. —LBR
Nat. Commun. (2024)
10.1038/s41467-024-46210-1
HYDROLOGY
Underground truth
More than 20 years into a
severe drought, the Great Basin
of the Southwestern US is losing
groundwater at a rapid rate.
Hall et al. report measurements
from the GRACE satellites
showing that nearly 70 cubic
kilometers of groundwater, or
more than six times the current
volume of water in Lake Mead of
Arizona and Nevada, has been
depleted there over that time.
Groundwater replenishment
by snowfall, even in high-snow
years, has not been able to keep
up with this loss, which is likely
the result of a combination of
declining snow mass, upstream
water diversions, and increased
evaporation and sublimation
due to rising air and ground
temperatures. —HJS
Geophys. Res. Lett. (2024)
10.1029/2023GL107913

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 45

RES EARCH

◥ of distinct cellular identities to profile the char-

RESEARCH ARTICLE SUMMARY acteristics of hepatic myeloid cells isolated from
mice infected with the helminth Schistosoma
INNATE IMMUNITY mansoni. In addition, we used a combination
of in vitro and in vivo approaches to deter-
Apoptotic cell identity induces distinct functional mine phagocytic receptors preferentially en-
gaged during the uptake of specific populations
responses to IL-4 in efferocytic macrophages of apoptotic cells and to assess their impact on
the outcome of S. mansoni infection.
Imke Liebold, Amirah Al Jawazneh, Christian Casar, Clarissa Lanzloth, Stephanie Leyk,
Madeleine Hamley, Milagros N. Wong, Dominik Kylies, Stefanie K. Gräfe, Ilka Edenhofer, RESULTS: We identified that IL-4–induced sig-
Irene Aranda-Pardos, Marie Kriwet, Helmuth Haas, Jenny Krause, Alexandros Hadjilaou, natures in bone marrow–derived macrophages
Andra B. Schromm, Ulricke Richardt, Petra Eggert, Dennis Tappe, Sören A. Weidemann, differed depending on the identity of the
Sourav Ghosh, Christian F. Krebs, Noelia A-Gonzalez, Anna Worthmann, Ansgar W. Lohse, apoptotic cell that had been sensed. Uptake
Samuel Huber, Carla V. Rothlin, Victor G. Puelles, Thomas Jacobs, Nicola Gagliani*, Lidia Bosurgi* of apoptotic neutrophils induced a tissue re-
modeling profile in macrophages, whereas the
sensing of apoptotic hepatocytes promoted an
INTRODUCTION: Macrophages acquire specific from different cell lineages may differentially immunosuppressive, or tolerogenic, phenotype.
signatures and functions on the basis of the influence macrophage activation and contrib- Phagocytosis of apoptotic thymocytes only
local tissue signals that they encounter. Within ute to macrophage functional diversification. slightly altered the macrophage response to
each tissue, macrophages are constantly ex- IL-4. These distinct signatures were also iden-
posed to a vast assortment of dying cells that RATIONALE: We analyzed the transcriptomic and tified in hepatic myeloid cells isolated from
need to be cleared to maintain homeostasis. So phenotypic characteristics of bone marrow– S. mansoni–infected mice. The adoptive trans-
far, the fact that distinct cell lineages undergo derived macrophages exposed to different fer of macrophages reprogrammed in vitro
apoptosis, thereby potentially imparting diverse cell populations undergoing apoptosis in an through exposure to apoptotic neutrophils—
signals to the surrounding environment, has interleukin-4 (IL-4)–enriched environment in but not other apoptotic cells—ameliorated the
been overlooked. Specifically, it is not yet clear vitro. We used these results to generate signa- outcome of S. mansoni infection. The ability
whether the sensing of apoptotic cells derived tures for myeloid cells engulfing apoptotic cells of macrophages to take up apoptotic cells is
dependent on the engagement of phagocytic
IL-4–enriched environment receptors. We identified that distinct phago-
cytic receptors were required for macrophages
to be able to engulf specific types of apoptotic
Apoptotic Apoptotic Apoptotic
PtdSer neutrophil
cells. Signaling by way of these receptors,
thymocyte AXL hepatocyte therefore, could be a potential mechanism
MERTK
regulating the acquisition of different gene
expression signatures in macrophages. We
found that the phagocytic receptors AXL
and MERTK were required for the uptake
of apoptotic neutrophils and T cells but not
hepatocytes. Accordingly, AXL- and MERTK-
dependent phagocytosis controlled the host
response to S. mansoni infection, overall con-
Macrophage tributing to parasitic egg clearance.

CONCLUSION: We have identified that the cel-

Cd274 lular identity of the ingested apoptotic cell con-
Serpinb9b
Cd300ld Socs1 tributes to macrophage gene expression and
Tnip3 Socs2 function. This emphasizes that the identity of
apoptotic cells within a tissue environment
may serve as an additional trigger of macro-
Limited impact on Ear2
Tolerogenic profile phage functional diversity. Furthermore, our
IL-4–induced genes Retnla
Chil3 findings highlight the potential of selective
ILLUSTRATION: CATERINA DI PIETRO, VISUALSCISKETCH

macrophage feeding as an approach to en-

hance the effectiveness of macrophage-based
cell therapies.
Tissue remodeling profile ▪
The identity of the apoptotic cell sensed is a critical determinant of macrophage functional The list of author affiliations is available in the full article online.
diversification. In IL-4–enriched environments, both in vitro and in vivo, macrophages acquire distinct *Corresponding author. Email: n.gagliani@uke.de (N.G.);
l.bosurgi@uke.de (L.B.)
transcriptional signatures and functional phenotypes on the basis of the identity of the apoptotic cell
Cite this article as I. Liebold et al., Science 384, eabo7027
phagocytosed. The engagement of the phagocytic receptors AXL and MERTK contributes to the ability of (2024). DOI: 10.1126/science.abo7027
macrophages to engulf apoptotic neutrophils and, to a lesser extent, apoptotic T cells and therefore to
the acquisition of distinct macrophage profiles. AXL and MERTK do not contribute to the ability of READ THE FULL ARTICLE AT
macrophages to phagocytose apoptotic hepatocytes. PtdSer, phosphatidylserine. https://doi.org/10.1126/science.abo7027

46 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESEAR CH

◥ To delve into the phage entry mechanism, we

RESEARCH ARTICLE SUMMARY
BACTERIOPHAGE
Removal of Pseudomonas type IV pili
by a small RNA virus
Jirapat Thongchol†, Zihao Yu†, Laith Harb, Yiruo Lin, Matthias Koch, Matthew Theodore,
Utkarsh Narsaria, Joshua Shaevitz, Zemer Gitai, Yinghao Wu, Junjie Zhang*, Lanying Zeng*
INTRODUCTION: Retractile pili in bacteria play an
important role in numerous biological processes,
such as DNA and protein transfer, motility,
adhesion, surface sensing, biofilm formation,
and pathogenesis. Single-stranded RNA (ssRNA)
bacteriophages (phages) are small viruses that
specifically target these retractile pili. With a
small positive-strand RNA genome of approx-
imately 4000 nucleotides, ssRNA phages typ-
ically encode four proteins: maturation protein
(Mat), coat protein (Coat), RNA-dependent RNA
replicase (Rep), and single-gene lysis protein (Sgl).
The Mat is crucial for phage maturity and pilus
recognition. However, how ssRNA phages use the
Mat-pilus interaction to enter the host cell has
remained mysterious for the more than six de-
cades since the discovery of the first ssRNA phage.
RATIONALE: The ssRNA phage PP7 infects Pseu-
domonas aeruginosa O1 (PAO1) through the
type IV pilus (T4P), a prominent virulence fac-
tor associated with motility. Unveiling the mech-
anisms under which PP7 exploits T4P for
cellular entry will shed light on fundamental
aspects of phage-bacterium interactions and
phage biology and may open an avenue for
antimicrobial strategies.
RESULTS: Using fluorescence microscopy, we
observed the detachment of PAO1 T4P during
PP7 infection. Intriguingly, T4P detachment in-
duced by ultraviolet (UV)–inactivated PP7
(UV-PP7), whose viral RNA cannot enter the cell,
mirrors that of the wild-type (WT) PP7. This
indicates that T4P detachment occurs at the cell
envelope during PP7 entry, independently of
PP7 replication inside the cell. Alongside T4P
detachment, both WT PP7 and UV-PP7 treatments
impede T4P dynamics. This combined effect
drastically reduces cell twitching motility.
PP7 mature virions, resolved by means of
single-particle cryo–electron microscopy, fea-
ture two Mat proteins forming a heterotypic
dimer. One Mat is exposed for T4P interaction,
whereas the other is entirely internalized with-
in the capsid. This mature PP7 virion structure
diverges substantially from the single-Mat struc-
ture found in canonical Escherichia coli phages,
such as MS2 and Qb, challenging our under-
standing of the structure of mature ssRNA phages.
determined the structures of T4P and the PP7/
T4P complex, which revealed that PP7 binds
to T4P by interacting with a single pilin sub-
unit through the Pilus-Interacting Region (PIR)
of the Mat.
Through examination of various mutants of
the host retraction ATPases and the pilus, we
discovered that pilus detachment is influenced
by the pilus retraction force and speed, as well
as the affinity between phage Mat and its
bound pilin. Further mechanical considerations
led to the pilus bending hypothesis, which was
substantiated by our Langevin dynamics sim-
ulation of the Mat bound to T4P during the
phage entry process.
CONCLUSION: We observed that T4P can be de-
tached or removed by an ssRNA phage and
revealed the molecular mechanism for T4P de-
tachment. We propose that similar mechanisms
are widespread among ssRNA phages and their
respective retractile pilus systems. Recent bio-
informatic studies of environmental samples
have identified thousands of ssRNA phage
genomes, which await mechanistic analysis.
This work could serve as a benchmark for in-
vestigating other phage and virus systems of
different organisms.
▪
The list of author affiliations is available in the full article online.
*Corresponding author. Email: junjiez@tamu.edu (J.Z.);
lzeng@tamu.edu (L.Z.)
†These authors contributed equally to this work.
Cite this article as J. Thongchol et al., Science 384,
eadl0635 (2024). DOI: 10.1126/science.adl0635
READ THE FULL ARTICLE AT
https://doi.org/10.1126/science.adl0635

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

D PP7 infection inactivates T4P and diminishes twitching motility of P. aeruginosa

Phage adsorption Twitching motility Non-motile
PP7 Static T4P

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).

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 47

RES EARCH

◥ limited knowledge of the universality classifi-

RESEARCH ARTICLES cation of dynamical phases of matter at finite
temperatures, for which contributions from the
QUANTUM SIMULATION entire spectrum must be considered (4, 5).
It has been observed in several dynamical sys-
Dynamics of magnetization at infinite temperature tems that the long-time behavior permits a few-
parameter hydrodynamical description, suggesting
in a Heisenberg spin chain the existence of universality (6–11). The emergence
of a hydrodynamical description relies on reach-
E. Rosenberg1,2†, T. I. Andersen1†, R. Samajdar3,4, A. Petukhov1, J. C. Hoke5, D. Abanin1, ing local and, subsequently, global equilibrium
A. Bengtsson1, I. K. Drozdov1,6, C. Erickson1, P. V. Klimov1, X. Mi1, A. Morvan1, M. Neeley1, C. Neill1, (12, 13). This fate is less certain in systems with
R. Acharya1, R. Allen1, K. Anderson1, M. Ansmann1, F. Arute1, K. Arya1, A. Asfaw1, J. Atalaya1, an extensive set of conserved quantities, i.e.,
J. C. Bardin1,7, A. Bilmes1, G. Bortoli1, A. Bourassa1, J. Bovaird1, L. Brill1, M. Broughton1, B. B. Buckley1, integrable systems, which are known to evade
D. A. Buell1, T. Burger1, B. Burkett1, N. Bushnell1, J. Campero1, H.-S. Chang1, Z. Chen1, B. Chiaro1, thermalization, and their universal behaviors
D. Chik1, J. Cogan1, R. Collins1, P. Conner1, W. Courtney1, A. L. Crook1, B. Curtin1, D. M. Debroy1, are discussed in the framework of generalized
A. Del Toro Barba1, S. Demura1, A. Di Paolo1, A. Dunsworth1, C. Earle1, L. Faoro1, E. Farhi 1, R. Fatemi1, hydrodynamics (14–18).
V. S. Ferreira1, L. Flores Burgos1, E. Forati1, A. G. Fowler1, B. Foxen1, G. Garcia1, É. Genois1, W. Giang1, Distinct microscopic models or dynamics
C. Gidney1, D. Gilboa 1, M. Giustina1, R. Gosula1, A. Grajales Dau1, J. A. Gross1, S. Habegger1, belong to the same universality class if they
M. C. Hamilton1,8, M. Hansen1, M. P. Harrigan1, S. D. Harrington1, P. Heu1, G. Hill1, M. R. Hoffmann1, share a single scale-invariant limit under an RG
S. Hong1, T. Huang1, A. Huff1, W. J. Huggins1, L. B. Ioffe1, S. V. Isakov1, J. Iveland1, E. Jeffrey1, flow (1, 3). A universality class is commonly
Z. Jiang1, C. Jones1, P. Juhas1, D. Kafri1, T. Khattar1, M. Khezri 1, M. Kieferová1,9, S. Kim1, A. Kitaev1, characterized by scaling exponents and scaling
A. R. Klots1, A. N. Korotkov1,10, F. Kostritsa1, J. M. Kreikebaum1, D. Landhuis1, P. Laptev1, K.-M. Lau1, functions, and it is rather implausible to ex-
L. Laws1, J. Lee1,11, K. W. Lee1, Y. D. Lensky1, B. J. Lester1, A. T. Lill1, W. Liu1, A. Locharla1, S. Mandrà1, tract them all experimentally. Therefore, experi-
O. Martin1, S. Martin1, J. R. McClean1, M. McEwen1, S. Meeks1, K. C. Miao1, A. Mieszala1, S. Montazeri1, ments, for example, on quantum processors,
R. Movassagh1, W. Mruczkiewicz1, A. Nersisyan1, M. Newman1, J. H. Ng1, A. Nguyen1, M. Nguyen1, cannot prove that a set of observed dynamics
M. Y. Niu1, T. E. O’Brien1, S. Omonije1, A. Opremcak1, R. Potter1, L. P. Pryadko12, C. Quintana1, belongs to a given class, but they can falsify a
D. M. Rhodes1, C. Rocque1, N. C. Rubin1, N. Saei 1, D. Sank1, K. Sankaragomathi1, K. J. Satzinger1, universality conjecture (19) by examining its
H. F. Schurkus1, C. Schuster1, M. J. Shearn1, A. Shorter1, N. Shutty1, V. Shvarts1, V. Sivak1, predictions. They can also probe numerically
J. Skruzny1, W. Clarke Smith1, R. D. Somma1, G. Sterling1, D. Strain1 M. Szalay1, D. Thor1, A. Torres1, and theoretically challenging regions of the
G. Vidal1, B. Villalonga1, C. Vollgraff Heidweiller1, T. White1, B. W. K. Woo1, C. Xing1, Z. Jamie Yao1, parameter space, which has proven advanta-
P. Yeh1, J. Yoo1, G. Young1, A. Zalcman1, Y. Zhang1, N. Zhu1, N. Zobrist1, H. Neven1, R. Babbush1, geous for studying universal behaviors (6–11).
D. Bacon1, S. Boixo1, J. Hilton1, E. Lucero1, A. Megrant1, J. Kelly1, Y. Chen1, V. Smelyanskiy1, Superconducting quantum processors offer
V. Khemani5, S. Gopalakrishnan3, T. Prosen13*, P. Roushan1* high wave function sampling rates, which
enable them to show quantum advantage over
Understanding universal aspects of quantum dynamics is an unresolved problem in statistical classical computers in sampling tasks (20, 21).
mechanics. In particular, the spin dynamics of the one-dimensional Heisenberg model were conjectured On these processors, one can go beyond mean
as to belong to the Kardar-Parisi-Zhang (KPZ) universality class based on the scaling of the infinite- expectation values and provide “snapshots” of an
temperature spin-spin correlation function. In a chain of 46 superconducting qubits, we studied the observable, which allows for measuring quantum
probability distribution of the magnetization transferred across the chain’s center, PðMÞ. The fluctuations and the probability distribution
first two moments of PðMÞ show superdiffusive behavior, a hallmark of KPZ universality. However, the of the observable. The capability of collecting
third and fourth moments ruled out the KPZ conjecture and allow for evaluating other theories. full counting statistics could have fundamen-
Our results highlight the importance of studying higher moments in determining dynamic universality tal consequences for our understanding of
classes and provide insights into universal behavior in quantum systems. dynamical universalities. In particular, it is
commonly assumed that the scaling functions
and exponents of the first few moments de-
n statistical physics, the notion of univer- transitions in a wide class of systems (1, 2). The termine a universality class, and there have

I sality is a powerful assertion; it implies

that systems with entirely different micro-
scopic interactions can share the same
emergent macroscopic description because
they have certain basic physical properties in
common. It is a triumph of universality that,
basic ingredients that commonly affect uni-
versality classes are the collective behavior
of constituent elements, symmetries, conser-
vation laws, and dimensionality, as described
by the renormalization group (RG) theory (3).
In contrast to rather well-understood low-
not yet been any instances in which the higher
moments of an observable have led to a dif-
ferent classification.

Spin dynamics in Heisenberg XXZ spin chains

Spin dynamics in the one-dimensional (1D)
for example, the Ising model is crucial to our temperature universality classes, which are XXZ model have been the subject of numerous
understanding of the zero-temperature phase determined by ground-state physics, we have recent studies (10, 15, 22–35). This integrable
model describes nearest-neighbor exchange
1
Google Research, Mountain View, CA, USA. 2Department of Physics, Cornell University, Ithaca, NY, USA. 3Department of
interactions between spin-1=2 particles with
Physics, Princeton University, Princeton, NJ, USA. 4Princeton Center for Theoretical Science, Princeton University, Princeton, the Hamiltonian (36)
NJ, USA. 5Department of Physics, Stanford University, Stanford, CA, USA. 6Department of Physics, University of Connecticut, X
Six Siþ1
x
þ Siy Siþ1
y
þ DSiz Siþ1
z


Storrs, CT, USA. 7Department of Electrical and Computer Engineering, University of Massachusetts, Amherst, MA, USA. H¼ ð1Þ
8
Department of Electrical and Computer Engineering, Auburn University, Auburn, AL, USA. 9QSI, Faculty of Engineering & i
Information Technology, University of Technology Sydney, Ultimo, NSW, Australia. 10Department of Electrical and Computer
Engineering, University of California, Riverside, CA, USA. 11Department of Chemistry, Columbia University, New York, NY, USA. where S , S y, and S z are spin-1=2 operators, and
x
12
Department of Physics and Astronomy, University of California, Riverside, CA, USA. 13Faculty of Mathematics and Physics, D is the anisotropy parameter. When D = 1, this
University of Ljubljana, Ljubljana, Slovenia.
*Corresponding author. Email: tomaz.prosen@fmf.uni-lj.si (T.P.); pedramr@google.com (P.R.) system is the Heisenberg model, a paradig-
†These authors contributed equally to this work. matic model of quantum magnetism that

48 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

Fig. 1. Domain wall relaxation in the Heisenberg XXZ spin chain. (A) Schematic
of the unitary gate sequence used in this work, where fSim gates are applied
in a Floquet scheme on a 1D chain of NQ = 46 qubits. (B) Relaxation dynamics
as a function of site and cycle number for m = ∞, 0.9, and 0.3 for initially
prepared domain-wall states with 2hSz i ¼ TtanhðmÞ. Blue and yellow squares
correspond to occupied and unoccupied sites, respectively, in random
instances of the experiment. The fSim angles were chosen to be (q, f) =
(0.4p, 0.8p), corresponding to D = 1. (C) Histogram showing the probability
distribution of transferred magnetization after t = 1, 5, and 20 cycles
[arrows in (B)] for m = ∞.

possesses a global SU(2) rotational sym-
metry. The spin dynamics in the Heisenberg
model exhibit characteristics consistent with
the KPZ universality class, which was originally
introduced to describe the stochastic, nonlinear
dynamics of driven interfaces and has proven
to apply to a wide range of classical systems
(10, 23–34). The KPZ-like behavior of the
spin dynamics is surprising because of the ab-
sence of stochasticity and nonlinearity in the
Heisenberg model.
In a 1D chain of NQ = 46 superconducting
qubits, we simulated this spin model by pe-
riodic (Floquet) application of high-fidelity
two-qubit unitary fSim(q,f) gates (Fig. 1A and
fig. S7) (37, 38). Here, q sets the amplitude of
hopping between adjacent qubit lattice sites,
and f is the conditional phase angle imparted
when two spin excitations are adjacent to each
other. Within each cycle, two-qubit fSim(q,ϕ)
gates are applied between all neighboring pairs
in the chain, resulting in the cycle unitary:
Y Y
UF ¼ fSimðq; fÞ fSimðq; fÞ
even bonds odd bonds
ð2Þ
In the limit q; f → 0, UF is the Trotter-Suzuki
expansion of the XXZ Hamiltonian (Eq. 1), with
D ¼ sinðf=2Þ=sinðqÞ. Away from this limit, there
is no specific, time-independent Hamiltonian
associated with UF, but Eqs. 1 and 2 still share
symmetries and are both integrable by the
Bethe ansatz (39–43). The conjecture that the
late-time dynamics are described by KPZ applies
equally to the Floquet system (44), so we made
no attempt to be in the small-angle limit, instead
favoring large angles for faster dynamics.
To study dynamics under the unitary evo-
lution (Eq. 2), we generated domain-wall initial
states with an adjustable contrast parameter m
(Fig. 1B). Specifically, we initialized the chain in a
set of product states such that the left and right
halves had average magnetization ±tanh(m),
respectively:
z
rðt ¼ 0Þºðe2mS Þ NQ =2
z we also included the reflection of that bitstring
ðe 2mS Þ NQ =2 ð3Þ
When m → ∞, the system approaches a pure
domain-wall state with the two sides fully mag-
netized in opposite directions. Only when m = 0,
the initial state is an infinite-temperature ther-
mal state that preserves SU(2) symmetry. When
m ≠ 0, the magnetization is preferentially along
the z axis, breaking the SU(2) rotational sym-
metry of the Heisenberg model.
A natural measure of spin transport is the
total transferred magnetization, Mðt Þ, defined
as twice the net number of excitations that
have crossed the middle of the chain after
t cycles. In our experiment, we sampled over
initial bitstring states with probabilities given
by Eq. 3. For each initial state, we prepared the
qubits in that state and then applied t cycles of
fSim gates. Let NR,1(b) be the number of exci-
tations (“1s”) in the right half of bitstring b.
The transferred magnetization M is the sto-
chastic variable defined by
Mðt Þ=2 ¼ NR;1 ðbt Þ NR;1 ðbi Þ ð4Þ
where bi is the initial bitstring, sampled from
Eq. 3, and bt is the associated final bitstring
sampled at t. For example, if the initial bit-
string is 111010 and the final bitstring is 110110,
then M is 2. Because the dynamics are number
conserving, M is also the net number of zeros
that have crossed from the right to the left. Re-
peating the experiment many times, we con-
structed the probability distribution ofM,PðMÞ.
In the case of m = 0, the initial state and the dy-
namics both have mirror symmetry, so for each
initial bitstring that was studied experimenally,
in our analysis, using the same experimental
data, which effectively symmetrized PðMÞ.
Figure 1B shows measurement instances for
three values of m. The left column in each panel
shows an instance of the initial state for the
given m, and the subsequent columns show
typical bitstrings evolved from that state. As
excitations (spin flips) propagate through
the chain, smaller domains become more prob-
able. In Fig. 1C, we show histograms of M at
different times, starting in a pure (m = ∞) domain
wall state. Owing to the locality of the circuit,
jMðt Þj is upper bounded by 2t. Consequently,
the distribution is narrow and centered around
a small value at t = 1 because only a few ex-
citations have crossed the middle of the chain
and becomes wider at later times.
Mean and variance of transferred magnetization
In the context of spin transport, the first and
second (variance) moments of M have been
extensively studied both theoretically and
experimentally (10, 15, 23–34, 43, 45). Taking
advantage of our tunable fSim gates, we ex-
plored how these two moments depend on
the anisotropy parameter, D. Figure 2A shows
the mean of the transferred magnetization,
hMi, over time for values of D equal to 0.16

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 49

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

Notably, the slight discrepancies between

the observed and predicted exponents can also
be seen in the exact statevector simulations up
to t = 18 cycles (colored lines), which agree
almost perfectly with the experimental data
for D < 1 and D = 1 and indicate that the de-
viations of the exponents from 1 and 2=3 , re-
spectively, primarily stem from finite-time
and large-m effects rather than from experi-
mental imperfections. Indeed, the exact scaling
exponents are only expected in the long-time
limit and as m → 0. Furthermore, simulations
of our system that include dominant sources
of noise (lighter lines, Fig. 2A; referred to as
“noisy simulations” elsewhere) explain why
there is near-perfect agreement between
the experimental and noiseless simulation
results. This effect is also noticeable in the
magnetization transfer distributions in Fig.
2B. Because the distribution is narrower in
the D > 1 case, the noise has a larger effect
on the shape of the distribution here than for
the other two values of D. The error in this
case was predominantly caused by combined
occurrences of decay to the j0i state and 0 →
1 readout errors, which were not eliminated by
postselection. By including this effect in the
simulation, we found good agreement in all
three regimes (Figs. 2A and figs. S5, S15, and
S16) (38).
Superdiffusive transport hMi ∼ t 2=3 at D = 1 is
a characteristic of systems within the KPZ
universality class. Moreover, numerical stud-
ies found that the spin-spin correlation func-
Fig. 2. Mean and variance in various transport regimes. (A) Mean of transferred magnetization hMi tion coincides with the KPZ scaling function
as a function of cycle number (t) for initial states with m = 0.5 and for D = 0.16 (purple triangles), 1 (orange (44, 49), which has led to the conjecture that near-
squares), and 1.6 (green circles). Light and dark curves show simulations with and without noise, respectively. equilibrium spin transport in the Heisenberg
hMðtÞi can be fit to t1/z and gives z = 1.12 ± 0.04 in the ballistic, z = 1.6 ± 0.1 in the superdiffusive, and model belongs to the KPZ universality class
z = 1.9 ± 0.2 in the diffusive regime. (Inset) Three different regimes characterized by D ¼ sinðf=2Þ=sinðqÞ, (44, 50, 51). This universality class is associated
with the orange line being the isotropic Heisenberg limit. (B) Histogram showing PðMÞ for values of with the classical nonlinear stochastic KPZ equa-
D studied in (A) at cycle 14. Light and dark lines show experimental data and noiseless simulation results, tion @h=@t ¼ n∇2 h þ lð∇hÞ2 þ hðx; t Þ, which
respectively. (C) Mean and (D) variance of M for D = 1 and 0.2 ≤ m ≤ 1 (brighter to darker squares). was originally introduced (52) to describe
With increasing m, the mean increases, whereas the variance decreases. the dynamics of driven interfaces as a height
field hðx; t Þ, where v, l, and h set the strength
(purple), 1 (orange), and 1.6 (green) and an in- often rely on approximation schemes, which of the smoothening diffusion, roughening non-
itial domain wall height of m = 0.5. We ob- could lead to inaccurate results. By contrast, linear growth, and stochasticity terms, respec-
served markedly different scaling behaviors in this study we performed exact statevector tively. The conjecture asserts that at late times,
in the three regimes. Eliminating the initial sampling up to cycle 18 without any approxi- the magnetization profile behaves similarly to
transient cycles, we fit a power law, hMi ∼ t 1=z, mations. This was achieved by taking advan- @hðx; t Þ=@x. Consequently,
to the data over cycles 10 to 23 and extract tage of the fact that hMðt Þi depends only on
scaling exponents of z = 1.12 ± 0.04, 1.6 ± 0.1, the spins within the light cone of width 2t and lim Mðt Þ ↔ 2hð0; t Þ hð ∞; t Þ hð∞; t Þ
m→0
and 1.9 ± 0.2, respectively. These are in close can thus be determined exactly by simulating
ð5Þ
agreement with theoretical predictions for the shorter chains. This simplification also allowed
ballistic (z = 1) (46), superdiffusive (z = 2=3) (23), us to derive analytical expressions for all mo- To further examine the universality class of
and diffusive (z = 2) (47, 48) behaviors, respec- ments of M at early cycles. Nevertheless, the the Heisenberg spin dynamics, two aspects are
tively [although z was predicted and observed computational cost grows exponentially, and of particular importance. First, because the
to depend on m (fig. S12) (38)]. Observation of with the resources used here, the simula- universal behavior is expected to depend on
superdiffusive propagation for isotropic inter- tions at cycles 14, 16, and 18 took about 1, 2, and whether the system is in equilibrium, it is
actions (D = 1) measured here and also in other 14 hours, respectively (fig. S9) (38). By contrast, essential to measure the dependence on m.
works (10, 22–34, 45) has been interpreted as a the quantum simulator allowed us to reach Second, although the scaling exponent of the
signature of the KPZ universality class. 23 cycles, primarily limited by the relaxation of mean is consistent with the KPZ universality
Numerical simulations of these domain-wall the qubits to the j0i state, which necessitated class, further insights can be gained by ex-
dynamics are shown with solid dark lines in an exponentially large sampling overhead in amining higher moments (the “full counting
Fig. 2A. A variety of numerical simulations the absence of error correction. statistics”) of PðMÞ. Owing to the reduced

50 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

Fig. 3. Skewness and excess kurtosis of transferred magnetization.
Experimental data and noiseless simulation results are shown with squares and
lines, respectively. (A) Skewness of transferred magnetization distribution S as a
function of t, for D = 1 and various m∈[0,1.5]. We symmetrized the m = 0
probability distribution, after which the skewness was exactly 0. (B) The same as
(A), but with the x axis rescaled as mt2/3 and excluding data points for which
t < 8. The collapse of S under rescaling is explored by using the noiseless
simulation data in fig. S10 (38). Dashed horizontal gray lines indicate predictions
based on the KPZ universality class (TW GUE), a NLFH model (57), and a CLL
model (60). The red lines marked “Wei et al.” show S measured in (10)
(for m = 1.5) and the 1s confidence interval. (Inset) Representation of different
ways of taking the late-time and small-m limits that are considered in this work.
(C) Kurtosis Q of transferred magnetization. The horizontal lines represent
the theoretical predictions from the same models shown in (B). The experimental
Q, averaged over cycles 16 to 23, is –0.05 ± 0.02. The kurtosis data does not
exhibit a collapse, as expected; the kurtosis cannot be a function of mtg for
any exponent g because, unlike the skewness, it has time dependence even
when m = 0.

signal-to-noise ratio, measuring higher moments

at small m is experimentally challenging. We Table 1. Comparison of the experimentally observed moments with the theoretically predicted
used our fast sampling capability to measure values from various models. hMi, mean of transferred magnetization; s2, variance; S, skewness; Q,
PðMÞ as a function of m and t (53, 54). Figure kurtosis. The experimental Q value was averaged over cycles 16 to 23 and m = 0 to 0.4, and the
2, C and D, shows the temporal evolution of errors were computed by using the jackknife method. See table S2 for details regarding KPZ
the mean and variance M for various values of predictions (38).
m ranging from 0.2 to 1. We emphasize that the
KPZ conjecture is only expected to apply for
hMi s2 S Q
small m, in which case we found that the dy-
2/3 2/3
namical exponents of both the mean and the Experiment t t 0* –0.05 ± 0.02
.....................................................................................................................................................................................................................
variance are close to 2=3 . For larger m, the dy- KPZ (Baik-Rains) (55) t2/3 t2/3 0.36 0.29
.....................................................................................................................................................................................................................
namical exponent of the mean approaches NLFH (57) t2/3 t2/3 0 0.14
.....................................................................................................................................................................................................................
5
=3 , which is consistent with recent numer- CLL (60) t2/3 t2/3 0 ∈[–0.07, 0.03]
.....................................................................................................................................................................................................................
ical results (23, 38) (fig. S12) and confirms that
*The skewness is 0 owing to symmetrization of our data.
small m is required for potentially recovering
KPZ dynamics.

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

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 51

RES EARCH | R E S E A R C H A R T I C L E S
order in taking these noncommuting limits.
When taking t → ∞ first, the appropriate prob-
ability distribution to compare PðMÞ against
is the Tracy Widom (TW) distribution (55),
which has S of about 0.22. This order of limits
corresponds to large mt 2=3 in Fig. 3B, where we
indeed found S consistent with this distri-
bution; averaging over the four rightmost
points in Fig. 3B, we found S ¼ 0:18 T 0:02,
which was consistent to two standard devia-
tions with the TW Gaussian unitary ensemble
(GUE) value, as well as with an earlier experi-
ment, which found S ¼ 0:33 T 0:08 (10). How-
ever, Q of the TW GUE distribution is 0.09,
whereas we found Q ¼ 0:05 T 0:02. The em-
ergence of KPZ dynamics in this order of limits
was further ruled out by numerical and theo-
retical predictions that the dynamics become
diffusive (z = 2) at a late time that grows as
t ∼ 1=m3 as m → 0 (53, 56).
One could consider taking the two limits
simultaneously in a way that the dynamics
do not become diffusive, e.g., by holding mt 2=3
constant (fig. S18) (38). The correct distri-
bution to compare against is TW GUE in this
case as well. If we take the limit with mt 2=3
fixed at a large value, then we would find
S consistent with TW GUE, but the measured
Q is still inconsistent with the TW GUE pre-
diction of 0.09 (Fig. 3C), ruling out KPZ dy-
namics on the timescales accessible in the
experiment. Although it remains possible that
KPZ dynamics will emerge at much later times
(i.e., Q will increase to 0.09), we see neither
evidence nor rationale for this.
An outstanding question is why only lower-
point observables, such as the mean and variance
of the transferred magnetization and the cor-
relator studied by (44), seem to behave con-
sistently with KPZ universality. Notably, other
systems have been identified that exhibit simi-
lar behavior. One such system is a nonlinear
fluctuating hydrodynamic (NLFH) model with
two coupled stochastic modes (57–59), which
predicts S ¼ 0, consistent with the Heisenberg
spin chain. However, it suggests Q ¼ 0:14 ,
which differs from what we observed, perhaps
because not all aspects of the model are uni-
versal. Another such system is the classical
Landau-Lifshitz (CLL) magnet (55, 60–62),
which predicts S ¼ 0 and a Q that is negative
and close to zero at these timescales (60). These
are consistent with our experimental results. It
is noteworthy that this classical system addi-
tionally exhibits dynamical behavior similar to
that of the quantum spin chain with enhanced
quantum fluctuations due to confinement (63).
Discussion and outlook
Studies of the universal aspects of quantum dy-
namics have attracted substantial interest re-
cently; accordingly, a complete classification
of their universal properties is lacking. Our
findings suggest that these classifications could
52 5 APRIL 2024 • VOL 384 ISSUE 6691
involve unanticipated subtleties. Our first result,
also observed by others, is the superdiffusive
transport characterized byhMi ∼ t 2=3 (Fig. 2A).
Although this anomalous diffusion is sugges-
tive of the known KPZ universality classes, this
classification is not compatible with our sec-
ond finding, the vanishing of S and Q near
equilibrium (Fig. 3C). Despite our findings’
apparent consistency with the CLL model, a
full understanding would require the devel-
opment of a systematic spacetime RG frame-
work that could establish the origin of the
KPZ-like behavior starting from the micro-
scopic dynamics of the Heisenberg model.
Quantum processors have the potential to help
with such RG studies [e.g., building on (64, 65)].
For example, our results suggest that quantum
entanglement (but not integrability) is irrele-
vant in the space-time RG sense.
Our observations are rooted in the interplay
of integrability, quantum fluctuations, and sym-
metry and have proved to be challenging to
describe through an effective quantum field
theory. Our observed discrepancies with KPZ
predictions suggest that the infinite-temperature
dynamics in the Heisenberg chain, if universal,
belong to a dynamical universality class that
has yet to be discovered.
RE FERENCES AND NOTES
1. L. P. Kadanoff, Physica A 163, 1–14 (1990).
2. S. Sachdev, Quantum Phase Transitions (Cambridge University
Press, 2011).
3. G. Odor, “Universality,” in Nonequilibrium Lattice Systems:
Theoretical Foundations (World Scientific Publishing Co.,
2008).
4. A. Polkovnikov, K. Sengupta, A. Silva, M. Vengalattore, Rev.
Mod. Phys. 83, 863–883 (2011).
5. C. Gogolin, J. Eisert, Rep. Prog. Phys. 79, 056001 (2016).
6. A. Sommer, M. Ku, G. Roati, M. W. Zwierlein, Nature 472,
201–204 (2011).
7. S. Erne, R. Bücker, T. Gasenzer, J. Berges, J. Schmiedmayer,
Nature 563, 225–229 (2018).
8. M. K. Joshi et al., Science 376, 720–724 (2022).
9. C. Zu et al., Nature 597, 45–50 (2021).
10. D. Wei et al., Science 376, 716–720 (2022).
11. Q. Fontaine et al., Nature 608, 687–691 (2022).
12. J. Lux, J. Müller, A. Mitra, A. Rosch, Phys. Rev. A 89, 053608
(2014).
13. V. B. Bulchandani, R. Vasseur, C. Karrasch, J. E. Moore, Phys.
Rev. B 97, 045407 (2018).
14. O. A. Castro-Alvaredo, B. Doyon, T. Yoshimura, Phys. Rev. X 6,
041065 (2016).
15. B. Bertini, M. Collura, J. De Nardis, M. Fagotti, Phys. Rev. Lett.
117, 207201 (2016).
16. B. Doyon, T. Yoshimura, SciPost Phys. 2, 014 (2017).
17. V. B. Bulchandani, R. Vasseur, C. Karrasch, J. E. Moore, Phys.
Rev. Lett. 119, 220604 (2017).
18. J. De Nardis, D. Bernard, B. Doyon, Phys. Rev. Lett. 121,
160603 (2018).
19. K. Popper, Conjectures and Refutations: The Growth of
Scientific Knowledge (Routledge Classics, 2002).
20. F. Arute et al., Nature 574, 505–510 (2019).
21. Y. Wu et al., Phys. Rev. Lett. 127, 180501 (2021).
22. M. Znidarič, Phys. Rev. Lett. 106, 220601 (2011).
23. M. Ljubotina, M. Žnidarič, T. Prosen, Nat. Commun. 8, 16117 (2017).
24. J.-M. Stéphan, J. Stat. Mech. 2017, 103108 (2017).
25. R. J. Sánchez, V. K. Varma, V. Oganesyan, Phys. Rev. B 98,
054415 (2018).
26. E. Ilievski, J. De Nardis, M. Medenjak, T. Prosen, Phys. Rev.
Lett. 121, 230602 (2018).
27. S. Gopalakrishnan, R. Vasseur, Phys. Rev. Lett. 122, 127202 (2019).
28. J. De Nardis, M. Medenjak, C. Karrasch, E. Ilievski, Phys. Rev.
Lett. 123, 186601 (2019).
29. S. Gopalakrishnan, R. Vasseur, B. Ware, Proc. Natl. Acad. Sci.
U.S.A. 116, 16250–16255 (2019).
30. J. De Nardis, S. Gopalakrishnan, E. Ilievski, R. Vasseur, Phys.
Rev. Lett. 125, 070601 (2020).
31. M. Dupont, J. E. Moore, Phys. Rev. B 101, 121106 (2020).
32. M. Dupont, N. E. Sherman, J. E. Moore, Phys. Rev. Lett. 127,
107201 (2021).
33. E. Ilievski, J. De Nardis, S. Gopalakrishnan, R. Vasseur, B. Ware,
Phys. Rev. X 11, 031023 (2021).
34. A. Scheie et al., Nat. Phys. 17, 726–730 (2021).
35. P. N. Jepsen et al., Nature 588, 403–407 (2020).
36. W. Heisenberg, Eur. Phys. J. A 49, 619–636 (1928).
37. C. Neill et al., Nature 594, 508–512 (2021).
38. Materials and methods are available as supplementary materials.
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).
41. M. Vanicat, L. Zadnik, T. Prosen, Phys. Rev. Lett. 121, 030606
(2018).
42. P. W. Claeys, J. Herzog-Arbeitman, A. Lamacraft, SciPost Phys.
12, 007 (2022).
43. M. Ljubotina, L. Zadnik, T. Prosen, Phys. Rev. Lett. 122, 150605
(2019).
44. M. Ljubotina, M. Žnidarič, T. Prosen, Phys. Rev. Lett. 122,
210602 (2019).
45. N. Keenan, N. F. Robertson, T. Murphy, S. Zhuk, J. Goold, NPJ
Quantum Inf. 9, 72 (2023).
46. X. Zotos, Phys. Rev. Lett. 82, 1764–1767 (1999).
47. S. Sachdev, K. Damle, Phys. Rev. Lett. 78, 943–946 (1997).
48. K. Damle, S. Sachdev, Phys. Rev. B Condens. Matter 57,
8307–8339 (1998).
49. M. Prähofer, H. Spohn, J. Stat. Phys. 115, 255–279 (2004).
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.
Lett. 129, 230602 (2022).
52. M. Kardar, G. Parisi, Y.-C. Zhang, Phys. Rev. Lett. 56, 889–892
(1986).
53. Ž. Krajnik, E. Ilievski, T. Prosen, Phys. Rev. Lett. 128, 090604
(2022).
54. S. Gopalakrishnan, A. Morningstar, R. Vasseur, V. Khemani,
Phys. Rev. B 109, 024417 (2024).
55. M. Prähofer, H. Spohn, Phys. Rev. Lett. 84, 4882–4885 (2000).
56. S. Gopalakrishnan, R. Vasseur, Rep. Prog. Phys. 86, 036502
(2023).
57. J. De Nardis, S. Gopalakrishnan, R. Vasseur, Phys. Rev. Lett.
131, 197102 (2023).
58. V. B. Bulchandani, Phys. Rev. B 101, 041411 (2020).
59. V. B. Bulchandani, D. A. Huse, S. Gopalakrishnan, Phys. Rev. B
105, 214308 (2022).
60. Ž. Krajnik, J. Schmidt, E. Ilievski, T. Prosen, Phys. Rev. Lett.
132, 017101 (2024).
61. A. Das, M. Kulkarni, H. Spohn, A. Dhar, Phys. Rev. E 100,
042116 (2019).
62. D. Roy, A. Dhar, H. Spohn, M. Kulkarni, Phys. Rev. B 107,
L100413 (2023).
63. T. Giamarchi, Quantum Physics in One Dimension (Oxford
University Press, 2003).
64. G. Vidal, Phys. Rev. Lett. 101, 110501 (2008).
65. G. Evenbly, G. Vidal, Phys. Rev. B Condens. Matter Mater. Phys.
79, 144108 (2009).
66. E. Rosenberg, T. I. Andersen, (2024). Zenodo. https://doi.org/
10.5281/zenodo.10045873.
67. E. Rosenberg, Kardar-Parisi-Zhang Dynamics in the Heisenberg
Spin Chain, ReCirq (2023); https://quantumai.google/cirq/
experiments/kpz/kpz.
AC KNOWLED GME NTS
We acknowledge discussions with I. Bloch, V. B. Bulchandani,
A. Morningstar, and R. Vasseur. Funding: V.K. acknowledges support
from the US Department of Energy, Office of Science, Basic Energy
Sciences, under Early Career Award no. DE-SC0021111; the Alfred
P. Sloan Foundation through a Sloan Research Fellowship; and the
Packard Foundation through a Packard Fellowship in Science and
Engineering. S.G., V.K., and T.P. acknowledge the hospitality of the
Kavli Institute for Theoretical Physics at the University of California,
Santa Barbara (supported by NSF grant PHY-1748958). R.S. is
supported by the Princeton Quantum Initiative Fellowship. T.P. is
supported by program P1-0402 of the Slovenian Research Agency
(ARRS). Author contributions: T.P., S.G., and V.K. proposed the
experiment and helped guide and interpret it. P.R. selected the
proposal and advised the experimental effort. E.R. implemented the
experiment. E.R. and T.I.A. collected the experimental data. E.R.
developed and ran the numerical simulations. R. Samajdar provided
theoretical input and implemented additional numerics. P.R., T.I.A.,
science.org SCIENCE
 RESE ARCH | R E S E A R C H A R T I C L E S

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

CELLULAR NEUROSCIENCE The robust incorporation of EU in the DG

and CB allowed us to determine whether the
Lifelong persistence of nuclear RNAs observed EU signals persisted for an extended
period during adulthood. We analyzed brain
in the mouse brain samples 1 year after EU injection and detected
strong EU signals in the DG and in the gran-
Sara Zocher1†, Asako McCloskey2,3†, Anne Karasinsky1, Roberta Schulte2, Ulrike Friedrich4,5,6, ular layer of the CB (Fig. 1, D to I). The den-
Mathias Lesche4, Nicole Rund1, Fred H. Gage7, Martin W. Hetzer8*, Tomohisa Toda1,9* sity of EU+ cells was indistinguishable between
1-week-old and 1-year-old animals (Fig. 1J),
Genomic DNA that resides in the nuclei of mammalian neurons can be as old as the organism itself. indicating the existence of long-lived RNAs
The life span of nuclear RNAs, which are critical for proper chromatin architecture and transcription (LL-RNAs) in the brain. A comparison of EU in-
regulation, has not been determined in adult tissues. In this work, we identified and characterized tensities revealed that 68 ± 14.26% of the initial
nuclear RNAs that do not turn over for at least 2 years in a subset of postnatally born cells in the mouse EU signals observed in the DG of 1-week-old
brain. These long-lived RNAs were stably retained in nuclei in a neural cell type–specific manner and animals persisted in 1-year-old animals (fig.
were required for the maintenance of heterochromatin. Thus, the life span of neural cells may depend S4). These data suggest that EU-labeled LL-
on both the molecular longevity of DNA for the storage of genetic information and also the extreme RNAs are minimally turned over during adult-
stability of RNA for the functional organization of chromatin. hood. Notably, the EU signal was sensitive to
ribonuclease (RNase), although only to a lim-
ited extent, even after harsh RNase treatment,
fter early development, most neurons sur- the turnover of this class of nucleic acids in adult suggesting either inaccessibility or notable sta-

A
vive for an organism’s entire life without
ever being replaced. In the absence of mi-
totic nuclear disassembly and reassembly,
the life span of some nuclear constitu-
ents such as genomic DNA, a subset of histones,
and nuclear pore complex proteins extends to
months and even years (1, 2). For example, the
stability of genomic DNA has been used to de-
termine the chronological age of neurons in
humans over the course of decades (3, 4). The
lifelong stability of genomic DNA is necessary
to safeguard the persistence of a cell’s genetic
information. In recent years, noncoding RNAs
have been identified as critical regulators of
nuclear chromatin organization and transcrip-
tion control (5, 6); however, little is known about
1
Nuclear Architecture in Neural Plasticity and Aging, German
Center for Neurodegenerative Diseases (DZNE), Dresden 01307,
Germany. 2Molecular and Cell Biology Laboratory, The Salk
Institute for Biological Studies, La Jolla, CA 92037, USA. 3Kura
Oncology, Inc., San Diego, CA 92121, USA. 4DRESDEN-concept
Genome Center, Technology Platform at the Center for
Molecular and Cellular Bioengineering (CMCB), Technische
Universität Dresden, Fetscherstr. 105, Dresden 01307, Germany.
5
German Center for Diabetes Research (DZD e.V.), 85764
Neuherberg, Germany. 6Paul Langerhans Institute Dresden of
the Helmholtz Center Munich, University Hospital and Faculty of
Medicine Carl Gustav Carus, Technische Universität Dresden,
01307 Dresden, Germany. 7Laboratory of Genetics, The Salk
Institute for Biological Studies, La Jolla, CA 92037, USA.
8
Institute of Science and Technology Austria (ISTA), 3400
Klosterneuburg, Austria. 9Laboratory of Neural Epigenomics,
Institute of Medical Physics and Micro-tissue Engineering,
Faculty of Medicine, Friedrich-Alexander-Universität Erlangen-
Nürnberg, Erlangen 91054, Germany.
*Corresponding author. Email: tomohisa.toda@fau.de (T.T.);
martin.hetzer@ist.ac.at (M.W.H.)
†These authors contributed equally to this work.
brain tissue.
Long-term retention of nuclear RNAs in the
mouse brain
To explore RNA stability in the mouse brain,
we performed in vivo pulse-chase labeling of
transcripts with a modified uridine analog,
5-ethynyl uridine (EU) (7). EU was injected
into mice on postnatal days 3 to 5 (P3 to P5) to
label de novo–synthesized RNAs during brain
development (Fig. 1A). One day after the last
injection of EU (defined as the 1-week sam-
ple), EU-labeled RNA was visualized by click
chemistry using Alexa-555 fluorophores. EU
signals were detectable in different brain re-
gions such as the dentate gyrus (DG) of the hip-
pocampus (Fig. 1B) and the cerebellum (CB)
(Fig. 1C). The EU signal was sparse and not de-
tectable in most cells in the primary somato-
sensory cortex (S1) (fig. S1). The density and
fraction of EU+ cells varied among different
brain tissues [analysis of variance (ANOVA)
P < 0.0001; Fig. 1, J and K]. We confirmed
that all cells in the brain were EU-labeled acute-
ly after a single EU injection on P3 (30-min
or 1-hour chase) (figs. S2 and S3) and detected
no difference in EU incorporation into de novo–
synthesized RNA between brain regions (fig.
S3C), which indicates that EU can be trans-
ported, metabolized, and used in transcription
in all cells of the brain at this age. These data
suggest that either the transcription or the re-
tention of EU-incorporated transcripts varies
between different cell types in the brain.
bility in the tissue (fig. S5, A and B). Consistent
with EU being specifically incorporated into
RNA, degradation of genomic DNA by deoxy-
ribonuclease I (DNase 1) treatment did not
affect EU signal intensities (fig. S5, C to H).
Moreover, when EU was injected at the same
time as bromodeoxyuridine (BrdU), which is
incorporated into newly synthesized DNA, the
spatial distribution of nuclear EU signals was
distinct from BrdU signals (fig. S5I), provid-
ing further evidence for the specific incorpora-
tion of EU into RNA. We also studied animals
2 years after EU injections and were able to
detect EU signals in DG and CB neurons
(Fig. 1L; and fig. S6, A, B, and D to F), radial
glia–like adult neural stem cells (RGL-ANSCs)
and cerebellar ANSCs (Fig. 1M and fig. S6H),
and astrocytes (fig. S6, C, E, and G). This find-
ing suggests that these RNAs persist through-
out the life span of the animal.
Cell type–specificity of long-retained RNAs
To identify which cell types retained LL-RNA,
we combined EU click chemistry with immuno-
histochemistry using cell type–specific markers.
In 1-year-old mice, most EU+ cells (83.7 ± 10.8%)
in the DG were NeuN+ neurons (Fig. 1, D and F).
EU signals were also observed in RGL-ANSCs
[Sox2+ with a glial fibrillary acidic protein–
positive (GFAP+) radial fiber; 1.5 ± 0.9% of
EU+ cells], adult neural progenitor cells (ANPCs)
(Sox2+ without a GFAP+ radial fiber; 3.76 ±
1.93% of EU+ cells), and GFAP+ astrocytes
(Fig. 1, D and H), suggesting that LL-RNAs

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 53

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

A EU injection Sampling Sampling Sampling

Birth P3 -P5 1 week 1 year 2 years

B EU EU NeuN GFAP C EU EU Sox2 Hoechst

CB 1 week
DG 1 week

D EU EU NeuN GFAP E EU EU Sox2 Hoechst

CB 1 year
DG 1 year

F EU NeuN GFAP EU NeuN GFAP G EU NeuN GFAP EU NeuN GFAP

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
of the experimental plan. RNA synthesis at P3 to P5 was labeled using EU.
(B) EU signals in the DG of a 1-week-old animal. (C) EU signals in the CB of a 1-week-old
animal. (D) EU signals in the DG of a 1-year-old animal. (E) EU signals in the
CB of a 1-year-old animal. (F and G) High-magnification images with NeuN and GFAP
staining (arrowheads, NeuN+ neurons) in the DG (F) and CB (G) of a 1-year-old
animal. (H and I) Sox2 and GFAP immunostaining (arrowhead, RGL-ANSC; open
arrowhead, astrocytes) in the DG (H) and CB (I) of a 1-year-old animal. (J) Quantification
of the density of EU+ cells: 1-week-old (1W) DG 3933 ± 911.9 cells/mm2, 1W CB
6060 ± 774.3 cells/mm2, and 1W S1 291.8 ± 52.12 cells/mm2; and 1-year-old (1Y) DG
4156 ± 542 cells/mm2, 1Y CB 4816 ± 537 cells/mm2, and 1Y S1 359 ± 199 cells/mm2.
Data are presented as mean ±SD. Significance was determined by ANOVA with
post hoc Tukey’s test. (K) Quantification of the percentage of EU+ neurons
(NeuN+ cells). Data are presented as mean ±SD. Significance was determined by
ANOVA followed by Tukey’s multiple comparison test. (L) EU signals in neurons
of the DG of a 2-year-old animal. (M) EU signals in a neural stem cell (RGL-
ANSC) of the DG of a 2-year-old animal. Scale bars are 100 mm [(B) and (D)],
10 mm [(F) to (I), (L), and (M)], and 200 mm [(C) and (E)]. Images in (B) to
(I), (L), and (M) are from confocal microscopy. *P < 0.05, **P < 0.01, **P < 0.001,
****P < 0.0001, and NS is not significant.

54 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE


can be maintained in neurons and somatic
ANSCs in the DG. In 1-week-old mice (1 day
after the last EU injection), only 25.6 ± 9.4%
of EU+ cells were NeuN+, whereas 39.7 ± 3.8%
were Sox2+ neural stem and progenitor cells
(NSPCs) and 7.0 ± 1.5% were RGL-ANSCs (fig.
S1I). In the first and second postnatal weeks,
ANPCs in the DG give rise to neurons or be-
come quiescent RGL-ANSCs (8, 9), suggest-
ing that the retention and incorporation of
EU+ transcripts could coincide with cell cycle
exit during differentiation or transition into
quiescence. Indeed, 4 hours after a single EU
injection, when most acutely labeled RNA was
already degraded (figs. S2 and S3), all high-
intensity EU+ cells in the hippocampus were
proliferating (Ki67+) NSPCs, whereas 24 hours
after the last injection, most EU+ cells had
exited the cell cycle and 44.5% of EU+ cells
expressed neuronal differentiation-associated
marker NeuroD1 (fig. S7). Thus, EU-labeled
RNA is retained long-term upon cell cycle exit
in neurons and ANSCs in the DG.
Similar to the DG, in the CB of 1-week-old
mice, 25.7 ± 15.4% of EU+ cells were NeuN+
neurons (fig. S1J) and 13.3 ± 4.4% of EU+ cells
were Sox2+ adult NSPCs (10), indicating that
EU+ transcripts are retained in neurons and
stem cells of CB. Fifty-eight percent of EU+ cells
in the CB did not express any tested markers
at this age, but they were likely to be migrating
immature granule cells on the basis of their
localization (Fig. 1C) (11). Indeed, in 1-year-old
mice, 76.3 ± 8.5% of EU+ cells were NeuN+
neurons mostly located in the granular layer
(Fig. 1, E, G, and K and fig. S1, E and J). We also
observed the retention of LL-RNA in Sox2+
ANSPCs in the CB of 1-year-old mice (Fig. 1I),
which suggests that ANSPCs in both the DG
and CB can retain LL-RNAs. In the S1, a total
of 34.4 ± 14.6 and 43.0 ± 10.3% of EU+ cells
were GFAP+ in 1-week-old and 1-year-old sam-
ples, respectively (fig. S1, F to H and K), which
indicates that astrocytes in the S1 retain EU-
labeled transcripts. However, NeuN+ EU+ cells
were rarely observed in the S1 (Fig. 1K). Thus, LL-
RNAs are retained or transcribed in a cell type–
specific manner.
To validate that our findings were indepen-
dent of potential side effects caused by EU,
we labeled nascent RNA using another uridine
analog, 5-bromouridine (BrU). The spatial pat-
terns of BrU+ cells in 1-week-old and 1-month-
old mice reflected those observed in EU-injected
mice, with persistent BrU labeling in neurons
and ANSPCs in the DG and CB but only very
few BrU+ cells in the S1 (fig. S8). Additionally,
injection of triphosphorylated uridine (5-BrUTP),
which is directly incorporated into RNA with-
out being metabolized, resulted in a similar
distribution of label-retaining cells in the brain,
providing further support that cell type–specific
retention of RNA is not due to potential differ-
ences in cellular nucleotide metabolism.
SCIENCE science.org
Because most cortical neurons are gener-
ated during embryonic neurogenesis, we next
asked whether cortical neurons retain EU sig-
nals when EU is injected during cortical neuro-
genesis. We injected EU into pregnant dams
at embryonic days 14.5 to 16.5 (E14.5 to E16.5)
and analyzed brains at E17.5 and at P6 (fig.
S9A). We confirmed that EU injection labeled
RNA synthesis in all embryonic brain cells at
1 hour after injection (fig. S9B). At E17.5, EU
signals were retained in Sox2+ and Tbr2+ NSPCs
in the ventricular and subventricular zone of
the neocortex (fig. S9, B to E and G). By con-
trast, we found only very few and weakly labeled
EU+ cells in the cortical plate—fewer cells than
the expected number of newborn neurons based
on the distribution of BrdU+ newborn cells at
E17.5 (fig. S9, E and F). Of those rare EU+ cells
in the cortical plate, only 39.0 ± 3.2 and 5.1 ±
3.2% expressed neuronal markers NeuN and
Ctip2, respectively (fig. S9, E and H). Thus,
cortical neurons have a limited capability to
transcribe or retain LL-RNAs. Moreover, only
sparse EU signals were detected in the cortex,
hippocampus, and CB at P6 when EU was in-
jected embryonically (fig. S9I). Thus, the long-
term retention of RNA is cell type–specific, with
cortical neurons showing a much-reduced capa-
bility to maintain RNAs compared with granule
cells of the DG and CB.
We then addressed whether LL-RNAs can be
transcribed in the brain at any age. EU injec-
tions at later postnatal ages (P7 to P9 and P13
to P15) resulted in patterns of EU+ cells in the
hippocampus that were consistent with those
obtained after EU injections at P3 to P5 (fig.
S10). However, in the adult mouse brain, EU
was not incorporated into nascent transcripts
(fig. S11) and was, therefore, not suitable to label
RNA synthesis. This is presumably due to the
down-regulation of enzymes of the pyrimidine
salvage pathway in the adult brain (fig. S12)
that are required for EU metabolism. No brain
region–dependent or cell type–specific dif-
ferences in the expression of the pyrimidine
salvage pathway were detected in postnatal
mice (fig. S12, A to C), consistent with the
ubiquitous EU signal that was observed acute-
ly after EU injection in postnatal brains (figs.
S2 and S3). This finding supports the notion
that all cells in the postnatal brain can me-
tabolize EU but that the retention of LL-RNAs
is cell type–specific.
Having established the existence of nuclear
LL-RNAs, we next aimed to determine their
molecular identity and functions. To facilitate
functional characterization, we used a tracta-
ble in vitro system in which we could recapitu-
late the long-term retention of transcripts and
manipulate the identified RNAs. Because we
observed EU signal retention in RGL-ANSCs
in vivo, we tested whether quiescent neural pro-
genitor cells (quiNPCs) retained EU-labeled
transcripts in vitro. We used an established pro-
RESE ARCH | R E S E A R C H A R T I C L E S
tocol to induce quiescence in mouse neural pro-
genitor cells (NPCs) (12) and found that quiNPCs
indeed retained EU+ transcripts for 8 days
(Fig. 2, A and B) and even 2 weeks (fig. S13),
but not in proliferating conditions (Fig. 2, A
and B). EU signals in quiNPCs were depleted
by RNase treatments (fig. S14, A and B) and
significantly reduced after inhibition of RNA
polymerase II and III (fig. S14C), confirming
that EU signals are derived from nascent RNA
in quiNPCs. Inhibition of RNA polymerases
after EU labeling did not reduce EU signals
(fig. S14, D and E), suggesting that EU signals
are not derived from the recycling of EU-labeled
uridines. Moreover, DNase 1 treatment and ribo-
nucleotide reductase inhibition did not reduce
EU signals in quiNPCs (fig. S15), providing fur-
ther evidence that EU signals are not derived
from EU incorporation into DNA.
Molecular identity of LL-RNAs
To determine the molecular identity of LL-RNAs,
we performed EU-RNA-sequencing (EU-RNA-
seq) of quiNPCs 1 week after EU labeling. We
found 1575 genes to be significantly enriched
in the EU-labeled RNA fraction (Fig. 2C, fig.
S16A, and data S1), and identified those as LL-
RNAs in quiNPCs. The LL-RNAs comprised
protein-coding RNAs and noncoding RNAs
(Fig. 2D), with long noncoding RNAs (lncRNAs)
and uncharacterized transcripts (TECs; to be
experimentally confirmed) significantly over-
represented in EU-enriched RNAs compared
with all RNAs expressed in quiNPCs (Fig. 2E).
LL-RNAs in quiNPCs were enriched in path-
ways related to nuclear architecture and epi-
genetic regulation, suggesting potential roles
of LL-RNAs in these cellular functions (Fig. 2F).
We also performed EU-RNA-seq with hippo-
campal tissue from 1-month-old mice that were
injected with EU at P3 to P5, and we identified
16 gene-derived transcripts that were signifi-
cantly enriched in the EU-labeled RNA frac-
tion, including lncRNAs, mitochrondrial tRNAs,
and protein-coding RNAs (Fig. 2, G and H).
Mapping of EU-RNA-seq data to repetitive
genomic regions revealed that LL-RNAs con-
tained repetitive RNAs in both hippocampus
and quiNPCs (Fig. 2, I and J; and fig. S16, B
and C), including small nuclear RNA repeats
(snRNAs), short interspersed nuclear elements
(SINEs), and satellite RNAs (satRNAs) (fig.
S16D). Together, these results suggest that non-
coding RNAs, including lncRNAs and repeti-
tive RNAs, can be retained long-term in quiNPCs
and in hippocampal granule cells. The differ-
ence in the detected numbers of LL-RNAs be-
tween quiNPCs and hippocampal tissue might
result from differences in the length of the re-
tention period or from cell type–specific mo-
lecular identities of LL-RNAs. Alternatively,
the detection sensitivity of LL-RNAs may be
reduced in hippocampal tissue because only a
fraction of hippocampal cells retained EU-labeled
5 APRIL 2024 • VOL 384 ISSUE 6691 55
RES EARCH | R E S E A R C H A R T I C L E S

A B C EU -enriched RNAs in NPCs

30 min 30 min Day 8 Day 8
****
***
NoEU proNPC proNPC quiNPC ***
1,575 EU-enriched RNAs

EU intensity (AU)
NS
(gene-derived)

-log10(adjusted p-value)
EU
DAPI

D EU-enriched RNAs in NPCs

protein coding (75.3%)
lncRNA (8.5%)
miRNA (0.1%) log2(counts per million)
mtRNA (0.4%)
snRNA (0.3%)
TEC (13.8%)
F Enriched pathways of EU-enriched RNAs in NPCs
pseudogenes (1.8%) Deposition of new CENPA-containing nucleosomes at the centromere
Nucleosome assembly
Chromosome Maintenance
E Enrichment of non
Enrichment - coding RNAs
of non-coding RNAs PRC2 methylates histones and DNA
RUNX1 in megakaryocyte differentiation and platelet function
TEC *** Processing of DNA double-strand break ends
lncRNA HDR through HRR or Single Strand Annealing (SSA)
*** Homology Directed Repair
Protein coding G2/M DNA damage checkpoint
***
HATs acetylate histones
0 1 2 3
Enrichment (odds ratio) 0 1 2 3 4 5
-log10(adjusted p-value)

G EU-enriched RNAs in hippocampus

H EU-enriched RNAs I EU-enriched repeat RNAs in hippocampus
in hippocampus
16 EU-enriched RNAs
-log10(adjusted p-value)

(gene-derived) snRNA repeats (85.8%)

SINE (11.0%)
Simple repeats (3.2%)

protein coding J EU-enriched repeat RNAs in NPCs

(81.25%)
LINE (0.2%)
lncRNA (6.25%) SINE (0.4%)
mtRNA (12.5%) RNA repeats (snRNA/rRNA; 61.4%)
Simple repeats (35.4%)
LTR (0.04%)
Low complexity repeats (2.5%)
log 2(counts per million) DNA repeats (Charlie; 0.07%)

Fig. 2. Molecular identity of long-retained RNA. (A) Confocal microscopy images
showing the retention of EU signals in quiNPCs in vitro. Scale bar is 10 mm.
DAPI, 4′,6-diamidino-2-phenylindole; proNPC, proliferating NPC. (B) Quantifica-
tion of EU intensity in proNPCs and quiNPCs at day 1 (D1) or D8 after EU labeling.
Dots indicate individual cells. Significance was determined by Kruskal-Wallis
test followed by Dunn’s multiple comparison test. AU, arbitrary units. (C) MA plot
showing EU-enriched (gene-derived) RNAs in quiNPCs at 8 days after EU
labeling, as determined by EU-RNA-seq. Significantly EU-enriched RNAs were
defined as adjusted P < 0.05 and fold change >2 compared with non–EU-labeled
quiNPCs (n = 3 experiments). (D) Gene class distribution of EU-enriched RNA
in quiNPCs. miRNA, microRNA; mtRNA, mitochondrial RNA. (E) Protein-coding
RNAs were underrepresented, whereas lncRNAs and uncharacterized
transcripts (TECs) were overrepresented in EU-enriched RNA. Significance
was determined by linear regression. (F) Top significantly enriched Reactome
pathways among EU-enriched RNAs in quiNPCs (adjusted P < 0.05). (G) MA
plot showing EU-enriched (gene-derived) RNAs in hippocampus tissue of
5-week-old mice that were injected with EU at P3 to P5 (n = 3 mice per
group). (H) Gene-class distribution of EU-enriched RNA in the hippocampus.
(I and J) Distribution of EU-enriched RNAs that map to repeat RNAs in
hippocampus tissue (I) and quiNPCs (J). In (C) and (G), the dashed-dotted
line indicates significance threshold of adjusted P = 0.05. ***P < 0.001, ****P <
0.0001, and NS is not significant.

transcripts, whereas 100% of in vitro quiNPCs
possessed EU-labeled RNAs. In the future, it
would be important to compare LL-RNAs of
different cell types under similar conditions.
Several studies suggest that repeat RNAs,
including satRNAs, interact with heterochro-
matin (13–16). Consistent with this idea, high- regulation. We confirmed the enrichment of
resolution imaging in DG and CB neurons from repeat RNAs among EU-captured RNA using
1-year-old samples showed that EU signals quantitative polymerase chain reaction (qPCR)
were often associated with Hoechst-dense het- (Fig. 3C) and found that a significant amount
erochromatin areas (Fig. 3, A and B), indicating of satRNAs were retained for up to 1 year in
that they might have functions in chromatin the hippocampus (Fig. 3D). The PCR signals

56 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

A EU Hoechst Merge B EU Hoechst Merge

Cerebellum
DG

C 1 week D 24 weeks-1 year E 1 week

* P = 0.058
*
*
* *

quiNPCs
** ** *
HC/CX

**
HC

* * ** **
* ***
** **

F Knock -down G H Overexpression I

**** EU EU DAPI EU EU DAPI

sgRNA Control
LNA Control

sgRNA MajSat
LNA Majsat

Fig. 3. Repeat RNAs are long retained in the mouse brain. (A and B) Localization
of EU signals in nuclei in the DG (A) and CB (B) of a 1-year-old animal. Scale bars are
5 mm. (C and D) Enrichment of EU-labeled transcripts in the hippocampus/cortex
(HC/CX) of a 1-week-old animal (n = 5 animals) (C) and in the hippocampus (HC) of
24-week-old to 1-year-old animals (n = 4 animals) (D). Significance was determined
by one-sample t test. Fold enrichments for qPCR signals of EU-labeled mice
compared with PBS-treated control are as follows: For (C) at 1W, glyceraldehyde-3-
phosphate dehydrogenase (Gapdh) 4.95 ± 1.85, 18S rRNA 3.76 ± 1.43, 28S rRNA
2.84 ± 1.13, major satellite 7.69 ± 4.16, minor satellite 4.48 ± 2.71, bactin 4.73 ± 2.54,
SINEB1 1.95 ± 0.60, and LINE-1 2.47 ± 0.62; and for (D) at 24W to 1Y, Gapdh 1.85 ±
0.66, 18S rRNA 2.38 ± 1.68, 28S rRNA 1.21 ± 0.63, major satellite 4.57 ± 1.55, minor
satellite 4.54 ± 2.37, bactin 1.60 ± 0.68, SINEB1 1.44 ± 0.70, and LINE-1 1.90 ± 1.34.
Data are presented as mean ±SD. The red dashed lines indicate the normalized
value from PBS-treated control. MajSat, major satRNA; MinSat, minor satRNA.
(E) Enrichment of EU-labeled transcripts in quiNPCs 8 days after EU treatment
compared with that in PBS-treated cells. Significance was determined by one-sample
t test. Fold enrichments compared to control are as follows: Gapdh 3.53 ± 1.50, 18S
rRNA 6.86 ± 2.06, 28S rRNA 7.38 ± 3.33, major satellite 49.3 ±19.1, minor satellite
52.57 ± 20.97, bactin 22.36 ± 10.2, SINEB1 40.45 ± 25.53, and LINE-1 59.82 ± 47.95
(n = 6 experiments). Data are presented as mean ±SD. (F) Quantification of
EU signal intensity after LNA-GapmeR–mediated knock down of major satRNAs in
quiNPCs (LNA MajSat). Dots indicate individual cells from three experiments.
Significance was determined by t test. (G) Reduction of EU signals in quiNPCs after the
application of LNA targeting major satRNAs. Scale bar is 20 mm. (H and I) Increased
EU signals after CRISPRa-based overexpression of major satRNAs (sgRNA MajSat).
Dots indicate individual cells from three experiments. Significance was determined by
t test. Scale bar is 20 mm. Images in (A), (B), (G), and (I) are from confocal microscopy.
*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

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 57

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

A B ** C CRISPRi CRISPRa D CRISPRi

LNA Control LNA Majsat H3K9me3
sgRNA Control sgRNA Control
NS
H3K9Me3

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

H pNPC LNA quiNPC BrdU

(E+F) (B+F) pNPC

Day 0 Day 3 Day 4

I LNA Control LNA Majsat

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.

58 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE


with previous findings of rRNAs exhibiting
longer half-lives (17). We validated that EU
treatment did not alter repeat RNA levels (fig.
S16E) and confirmed the long-term retention
of repeat RNAs in quiNPCs using BrU labeling
and BrU immunoprecipitation chase (BRIC)–
qPCR (fig. S17). Altogether, our findings re-
veal transcript-specific long-term retention of
repeat-derived RNAs in quiNPCs.
Roles of LL-RNAs in heterochromatin maintenance
Finally, we addressed the biological function
of LL-RNAs using quiNPCs as a model system
and manipulated major satRNAs using previ-
ously published LNA-GapmeRs (18) and single
guide RNAs (sgRNAs) (CRISPRa and CRISPRi)
(LNA, locked nucleic acid) (19). We observed
a significant reduction in EU signals after major
satRNA knock down (Fig. 3, F and G; and fig.
S18A), whereas overexpression of major satRNAs
increased EU signals (Fig. 3, H and I; and fig.
S18B), confirming the contribution of major
satRNAs to LL-RNAs in quiNPCs. Because it has
been shown that major satRNAs are associated
with constitutive heterochromatin (14, 15), we
investigated heterochromatin organization with
the constitutive heterochromatin marker tri-
methylated histone H3 lysine 9 (H3K9me3).
After knock down of major satRNAs, the sig-
nals of H3K9me3 around chromocenters be-
came disorganized and reduced (Fig. 4, A to D;
and figs. S18C and S19, A to E), and we con-
firmed the reduction in H3K9me3 at satellite
regions using chromatin immunoprecipitation
(ChIP) (Fig. 4, E and F). Conversely, overex-
pression of satRNAs led to an accumulation of
H3K9me3 at chromocenters, suggesting that
major satRNAs promoted heterochromatin
formation (Fig. 4, C and G). Knock down of ma-
jor satRNAs also reduced facultative hetero-
chromatin levels (H3K27me3) but did not affect
euchromatin mark H3K4me3 (fig. S19, F to I).
Because minor satRNAs were also enriched
among LL-RNAs (Fig. 3, C to E) and their over-
expression increased long-retained EU signals
in quiNPCs (fig. S18, D to F), we analyzed their
roles in heterochromatin retention. Although
minor satRNA overexpression significantly
increased H3K9me3 at chromocenters, their
knockdown did not affect H3K9me3 levels
(Fig. 4, C, D, and G; and fig. S18C). Altogether,
our results suggest a critical role for major
satRNAs in the maintenance of heterochro-
matin integrity in quiNPCs.
Because satRNAs are generated from hetero-
chromatic areas and bind heterochromatin-
associated proteins (19, 20), we tested whether
their long-term retention is controlled through
association with heterochromatin. Perturba-
tion of H3K9me3 by pharmacological inhibi-
tion of histone methyltransferases or by knock
down of heterochromatin-associated protein
1b (Hp1b) increased EU signal retention in
quiNPCs (fig. S20), suggesting that hetero-
SCIENCE science.org
chromatin is involved in LL-RNA regulation.
Heterochromatin is essential to repress aber-
rant transcription of satRNAs, and increased
satRNA expression is a hallmark of impaired
epigenetic regulation in tumorigenesis (21, 22).
Thus, the balanced level of LL-RNAs may be
regulated by heterochromatin in cooperation
with LL-RNAs themselves, yet its specific roles
in LL-RNA transcription and/or retention re-
quire further investigation.
We next tested whether major satRNAs
are important for maintaining quiNPCs. We
knocked down major satRNAs using LNA-
Majsat GapmeRs upon induction of quiescence
and tested whether quiNPCs could be reacti-
vated into proliferation (Fig. 4H). One day
after reactivation of quiNPCs, the fraction of
proliferating cells (Ki67+ cells and BrdU+ cells)
was significantly reduced in LNA-Majsat–
treated cells (Fig. 4, I to K), suggesting atten-
uated reactivation of quiNPCs. Furthermore,
significantly more pyknotic nuclei were ob-
served in LNA-Majsat–treated cells (Fig. 4L).
Consistent with this observation, the fraction
of gH2AX+ cells was significantly higher in
LNA-Majsat–treated cells (Fig. 4, M and N). By
contrast, knock down of satRNAs in proliferat-
ing NPCs did not impair proliferation or NPC
maintenance (fig. S21). To further characterize
the role of satRNAs in NPC maintenance, we
assessed the consequences of their overexpres-
sion. Overexpression of both major and minor
satRNAs was detrimental for NPCs and led to
reduced proliferation, increased apoptosis, and
DNA damage (fig. S22), indicating that proper
control of satRNA levels is critical for NPC
maintenance. Taken together, our results sug-
gest that major satRNAs are retained long-
term in quiNPCs and play critical roles in the
maintenance of heterochromatin integrity and
neural stem cell function. Because not all LL-
RNAs are satRNAs, further investigations will
be required to understand the biological roles
of LL-RNAs.
Previously, several proteins had been discov-
ered in the mammalian brain that can persist
for years (1, 2), suggesting possible roles for
long-lived cellular constituents in brain main-
tenance and aging. In this study, we found
that a subset of RNAs could be retained for up
to 2 years in certain brain cell types. Although,
in contrast to DNA, RNA has not been consid-
ered a stable nuclear component (23), our find-
ings are consistent with a previous study from
dormant Xenopus oocytes that shows that some
RNAs can be retained for more than a year
(24). The LL-RNAs we identified were enriched
around heterochromatin and essential to main-
taining chromatin integrity in quiescent neural
stem cells. These findings highlight a previously
unrecognized temporal axis of RNA metabo-
lism and possible functions for nuclear RNA in
long-term epigenetic regulation and genome
integrity. We propose that the lifelong main-
RESE ARCH | R E S E A R C H A R T I C L E S
tenance of nuclear function in postmitotic cells
and somatic stem cells involves the substantial
extension of the life span of key molecular con-
stituents, including LL-RNAs.
REFERENCES AND NOTES
1. J. N. Savas, B. H. Toyama, T. Xu, J. R. Yates III, M. W. Hetzer,
Science 335, 942 (2012).
2. B. H. Toyama et al., Cell 154, 971–982 (2013).
3. P. S. Eriksson et al., Nat. Med. 4, 1313–1317 (1998).
4. K. L. Spalding et al., Cell 153, 1219–1227 (2013).
5. P. Bhat, D. Honson, M. Guttman, Nat. Rev. Mol. Cell Biol. 22,
653–670 (2021).
6. A. Loda, S. Collombet, E. Heard, Nat. Rev. Mol. Cell Biol. 23,
231–249 (2022).
7. C. Y. Jao, A. Salic, Proc. Natl. Acad. Sci. U.S.A. 105, 15779–15784 (2008).
8. D. A. Berg et al., Cell 177, 654–668.e15 (2019).
9. Z. Nicola, K. Fabel, G. Kempermann, Front. Neuroanat. 9, 53 (2015).
10. J. Ahlfeld et al., Sci. Rep. 7, 6137 (2017).
11. E. Yacubova, H. Komuro, Cell Biochem. Biophys. 37, 213–234 (2003).
12. B. Martynoga et al., Genes Dev. 27, 1769–1786 (2013).
13. X. Huo et al., Mol. Cell 77, 368–383.e7 (2020).
14. O. Velazquez Camacho et al., eLife 6, e25293 (2017).
15. A. Shirai et al., eLife 6, e25317 (2017).
16. C. L. Novo et al., Nat. Commun. 13, 3525 (2022).
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).
19. Q. Zhu et al., Mol. Cell 70, 842–853.e7 (2018).
20. C. Maison et al., Nat. Genet. 43, 220–227 (2011).
21. D. T. Ting et al., Science 331, 593–596 (2011).
22. Q. Zhu et al., Nature 477, 179–184 (2011).
23. M. Rabani et al., Cell 159, 1698–1710 (2014).
24. P. J. Ford, T. Mathieson, M. Rosbash, Dev. Biol. 57, 417–426 (1977).
AC KNOWLED GME NTS
We thank M. L. Gage for editorial comments and current and
previous members of the Toda and Hetzer labs for helpful
discussions. S.Z. was supported by an Add-on fellowship from the
Joachim Herz foundation. NGS data production and data analysis
were carried out at the DRESDEN-concept Genome Center,
which is supported by the DFG Research Infrastructure Program
(project 407482635) and part of the Next Generation Sequencing
Competence Network NGS-CN (project 423957469). We thank
M. Krause and A. Dahl at the DRESDEN-concept Genome Center
for support with NGS data acquisition, G. Girke for molecular
cloning of sgRNAs, and M. Albert and A. Kolodziejczyk for providing
research materials. Funding: This work was funded by the
Boehringer Ingelheim Foundation (T.T.), the European Research
Council (ERC-2018-STG, 804468 EAGER; ERC-2023-COG,
101125034 NEUTIME) (T.T.), the DZNE (T.T.), the Nomis Foundation
(M.W.H. and A.M.), and Deutsche Forschungsgemeinschaft [DFG,
German Research Foundation (TO1347/4-1)] (T.T.). Author
contributions: Conceptualization: M.W.H., A.M., T.T.; Methodology:
A.M., S.Z., T.T., R.S., F.H.G., N.R., U.F., M.L., A.K.; Visualization: S.Z.,
A.M., R.S., T.T.; Funding acquisition: M.W.H., T.T.; Investigation:
S.Z., A.M., R.S., T.T.; Project administration: M.W.H., T.T.; Resources:
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:
S.Z., M.H.W., T.T.; Writing – review and editing: T.T., A.M., R.S., S.Z.,
F.H.G., M.H.W. Competing interests: The authors declare no
competing interests. Data and materials availability: All data
are available in the main text, figures, or supplementary materials.
EU-RNA-seq data were submitted to the Gene Expression Omnibus
(GEO) under accession no. GSE248101. License information:
Copyright © 2024 the authors, some rights reserved; exclusive
licensee American Association for the Advancement of Science. No
claim to original US government works. https://www.science.org/
about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adf3481
Materials and Methods
Figs. S1 to S22
Table S1
References (25–35)
MDAR Reproducibility Checklist
Data S1
Submitted 15 October 2022; resubmitted 14 November 2023
Accepted 2 February 2024
10.1126/science.adf3481
5 APRIL 2024 • VOL 384 ISSUE 6691 59
RES EARCH | R E S E A R C H A R T I C L E S

PEROVSKITES Design strategy for 2D all-organic perovskites

Charge-balance considerations prohibit the ex-
Molecularly thin, two-dimensional istence of an RP-type all-organic perovskite
with the stoichiometry of A2BX4 (Fig. 1A). The
all-organic perovskites smallest B-site cation in all-organic perovskites
is the monovalent ammonium cation (NH4+),
Hwa Seob Choi1†, Jun Lin1†, Gang Wang2†, Walter P. D. Wong3, In-Hyeok Park4, Fang Lin5, Jun Yin1*, whereas B-site cations in hybrid organic-inorganic
Kai Leng1,6*, Junhao Lin2,7*, Kian Ping Loh1* perovskites (HOIPs) are typically divalent metal
ions such as Pb2+ and Sn2+. Thus, the BX4
Recently, the emergence of all-organic perovskites with three-dimensional (3D) structures has octahedral layers (X indicates a halogen) in all-
expanded the potential applications of perovskite materials. However, the synthesis and utilization organic perovskites possess a −3 charge, in con-
of all-organic perovskites in 2D form remain largely unexplored because the design principle has not trast to the −2 charge found in HOIPs (Fig. 1B).
been developed. We present the successful synthesis of a metal-free 2D layered perovskite, denoted To maintain charge neutrality, an odd-numbered
as the Choi-Loh van der Waals phase (CL-v phase), with the chemical formula A2B2X4, where A counter charge is required, with the geometric
represents a larger-sized cation compared to B and X denotes an anion. The CL-v phase exhibits a constraint that A-site cations must exclusively
van der Waals gap enabled by interlayer hydrogen bonding and can be exfoliated or grown as occupy the cuboctahedral sites. We proposed
molecularly thin 2D organic crystals. The dielectric constants of the CL-v phase range from 4.8 to that this charge balance could be achieved if
5.5 and we demonstrate their potential as gate dielectrics for thin-film transistors. an additional NH4+ cation per unit cell oc-
cupied an “interstitial” position between the
layers and transform A22+ into A2B3+ (Fig. 1C).
erovskite materials are not only of fun- above and below the BX4 octahedra layer to The combination of the BX43− and A2B3+ layers

P
damental interest but also display prop-
erties that enable applications including
ferroelectrics (1), quantum materials (2),
catalysts (3, 4), light-emitting devices
(5), and solar cells (6). Perovskite crystals have
a stoichiometry of ABX3, where A represents a
larger cation compared with B, X denotes an
anion, and B is coordinated to six X anions to
form a BX6 octahedron. In three-dimensional
(3D) perovskites, these octahedra share corners
and the A cation occupies the cuboctahedral
site surrounded by eight octahedra. By dis-
rupting the continuous linkage of the octahedra
along the ½001Šdirection, a 2D version of perov-
skites can be formed in which layered struc-
tures are separated by van der Waals gaps.
Depending on the interlayer offset and
the type of atoms present in the “spacer”
layer, 2D perovskites can be classified into the
Ruddlesden-Popper phase (RP phase) (7, 8),
the Dion-Jacobson phase (DJ phase) (9), and
the Aurivillius phase (10, 11). The RP and DJ
phase layered perovskites have general for-
mulas of A2BX4 and ABX4, respectively. These
structures typically consist of corner-shared
BX4 octahedra forming a 2D layer, with the A+
cations coordinated to the cuboctahedral sites
1
Department of Applied Physics, Hong Kong Polytechnic
University, Hung Hom, Kowloon, Hong Kong, China.
2
Department of Physics, Southern University of Science and
Technology, Shenzhen 518055, China. 3Department of
Chemistry, National University of Singapore, Singapore
117543, Singapore. 4Graduate School of Analytical Science
and Technology (GRAST), Chungnam National University,
Daejeon 34134, Republic of Korea. 5College of Electronic
Engineering, South China Agricultural University, Guangzhou
510642, China. 6The Hong Kong Polytechnic University
Shenzhen Research Institute, Shenzhen 518057, China.
7
Quantum Science Center of Guangdong–Hong Kong–Macao
Greater Bay Area (Guangdong), Shenzhen 518045, China.
*Corresponding author. Email: jun.yin@polyu.edu.hk (J.Y.);
linjh@sustech.edu.cn (J.L.); kathy-kai.leng@polyu.edu.hk (K.L.);
kian-ping.loh@polyu.edu.hk (K.P.L.)
†These authors contributed equally to this work.
maintain overall charge neutrality.
The synthesis of all-organic perovskites has
been recognized as a challenging task within
the scientific community (12–29). In principle,
the structural topology of perovskite can be
maintained by replacing A, B, and X ions with
appropriately sized organic molecules, as dic-
tated by the Goldschmidt tolerance factor for
relative ion sizes. However, in practice, only a
few organic molecules (among those currently
used) fulfill this criterion. The isostructural
family of dicationic piperazinium organic perov-
skites with Cl−, Br−, and I− was discovered as
recently as 2002 (12). More recently, Xiong et al.
synthesized a diverse class of organic perov-
skites by designing various diammonium cations,
resulting in the report of 23 different organic
3D perovskites (14). These compounds include
a few ferroelectrics such as MDABCO-NH4-I3
(MDABCO = N-methyl-1,4-diazabicyclo[2.2.2]
octonium), which have polarization properties
(22 mC/cm2) similar to those of BaTiO3.
Aside from 3D metal-free perovskites, we
ask whether 2D-version (layered packing),
metal-free perovskites can be synthesized.
The presence of a van der Waals gap in 2D
perovskites offers the advantage of incorpo-
rating larger organic cations through self-
adjustable strain-balancing within the layered
structures, which could overcome steric effects
that hinder 3D perovskite synthesis (30). This
expanded library of larger and more exotic
A-site cations in 2D perovskites could enable
the emergence of distinctive properties and
applications. We present the design principles
and synthesis of a new class of all-organic
layered 2D perovskite phases. Following the
naming convention of oxide perovskites, we
refer to these as Choi-Loh van der Waals phase
(CL-v phase), with the suffix v denoting van der
Waals stacking in the structure. The CL-v phase
layered perovskites can be exfoliated or grown
as ultrathin layers of several nanometers.
results in the stoichiometry A2B2X4, abbrev-
iated as ABX2. The critical challenge in this
design approach lies in ensuring the stability
of the additional B+ cation at the interstitial
position, which depends on its surrounding
bonding environment.
In a typical cubic unit cell there are four
sites available for occupation: the corner, body
center, face center, and edge center sites (Fig.
2A). In the case of a perovskite unit cell, anions
occupy the face center site (X-site), the small
cation occupies the body center site (B-site) to
form an octahedron with the anions, whereas
the large cation occupies the corner site (A-
site) to form a cuboctahedron with the anions.
The interstitial site at the edge center (E-site)
can be filled by the small B+ cation (Fig. 2B).
Our design for the all-organic lattice used
N-chloromethyl-1,4-diazabicyclo[2.2.2]octonium
(CMD+) as the A-site cation because it could
form hydrogen bonds both laterally and ortho-
gonally with the interstitial B+ cation at the
edge center (E-site) of the unit cell (fig. S1).
The B-site was occupied by NH4+ and the X-
site was occupied by PF6–. In the case of CMD+,
the lone pair at the nitrogen formed an N-H···N
hydrogen bond with NH4+-edge laterally,
whereas the chloromethane from the adjacent
layer formed an N-H···Cl hydrogen bond with
NH4-edge vertically, thus stabilizing it by form-
ing an NH4+-edge[PF6]3[N]2[Cl] octahedron
(Fig. 2C).
Structure description of CL-v phases
We synthesized CMD-N-P2 through the slow
diffusion of dichloroethane into an acetone
solution containing CMD-PF6 and NH4PF6. A
hexagonal-shaped single crystal of CMD-N-P2
measuring 9 × 7 × 0.5 mm was obtained (Fig.
2E, inset). The crystal structure was monoclinic
and belonged to the P21/n symmetry group, as
determined by single-crystal x-ray diffraction
analysis (see supplementary materials, table S1).

60 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

To present the crystal structure, we chose a

unit cell similar to the conventional perovskite
structure. Using a transformation matrix (see
supplementary materials), we transformed the
unit cell into a 4×4×4 supercell composed of
64 pseudo-cubic unit cells (fig. S2). We de-
fined the ½001Š direction as normal to the octa-
hedral layers.
The pseudo-cubic unit cell consisted of one
NH4+ cation occupying the body center site
and four PF6− anions occupying the X-site to
form a homo-octahedron with NH4+-body,
creating the basal BX4 corner-shared octahe-
dral layer (Fig. 2B octahedra, NH4+-body[PF6]6).
Another NH4+ cation occupied the edge center
at the bottom and top surfaces of the (001)
plane of the pseudo-cubic unit cell, which was
shared with an adjacent unit cell (Fig. 2B and
fig. S3). The occupancy at the edge center was
½ as a result of sharing between two pseudo-
cubic unit cells in the layered structure (fig.
S4). Only one of the four edges of either the
top or bottom surface was occupied by NH4+-
edge, resulting in a total occupancy of 1 in the
pseudo-cubic unit cell for charge balancing.
CMD+ occupied the eight-cornered A-site cat-
ion positions (Fig. 2B). The neutral nitrogen
atoms from the heads of the two CMD+ ions
coordinated with NH4+-edge (Fig. 2, B and C,
blue dotted line), and the three PF6− ions stab-
ilized NH4+-edge in a square pyramidal site
(Fig. 2C, translucent red pyramid).
In the next layer the Cl atom at the tail of
the CMD+ ion stabilized the structure through
an N-H···Cl interaction, with a Cl–N distance
of 4.204(1) Å. Thus, a hetero-octahedron (where
coordinating atoms or molecules are differ-
ent) formed with three PF6−, two N from CMD+,
and one Cl from CMD+ in the adjacent layer
(Fig. 2C, NH4-edge[PF6]3[N]2[Cl]). The inter-
layer interaction was predominantly governed
by hydrogen bonding between NH4+-edge and
the Cl atom of the CMD+ ion across the two
layers (Fig. C and F). Only half of the Cl from
the CMD+ ion was hydrogen bonded to the
NH4+-edge of the next layer (unbonded Cl
atoms are represented by dark green spheres
in Fig. 2, C to G).
The CMD-N-P2 crystal synthesized had an
elongated hexagonal shape (Fig. 2E, inset) that Fig. 1. Illustration of charge balance issues in 2D hybrid organic-inorganic RP phase and its all-
could be correlated with the hexagonal pack- organic analogs. (A) Charge compensated octahedral layer and spacer layer in a typical RP phase HOIP.
ing of the NH4+-edge atoms highlighted by (B) In the all-organic RP phase, charge imbalance occurs between the octahedral and spacer layers.
the red hexagon (Fig. 2E). The NH4+-edge mol- (C) Addition of one interstitial B+ cation in the spacer layer per unit cell balances the charge of the system.
ecules were aligned along ½110  Š, which was
the fastest growth direction. In the same (001)
layer of NH4+-edge, the hydrogen-bonded Cl structure, we tried synthesizing the crystal with it crystallized in a monoclinic P21/n symmetric
atoms shifted along the ½110Š direction, as high- different halogen atoms (Cl, Br, and I) in the group with a nonperovskite IMD-N2-P3 struc-
lighted by the green hexagon, similar to the A-site cation, for example, N-bromomethyl- ture; an important clue was that this structure
NH4+-edge positions of the second layer (Fig. 1,4-diazabicyclo[2.2.2]octonium (BMD+) and does not have N-H···I bond with NH4+ (fig. S8).
E and F) with respect to the first layer and gave N-iodomethyl-1,4-diazabicyclo[2.2.2]octonium This result suggested that the weaker N-H···I
rise to AB stacking similar to the RP phase (Fig. (IMD+). With NH4+ as the B-site cation, we bond as compared with the N-H···Cl hydrogen
2G and fig. S2). tried growing BMD-N-P2 and IMD-N-P2, but bond destabilized the CL-v phase. Both CMD-
To investigate the stabilizing effect of inter- only BMD-N-P2 could be crystallized in the N-P2 and BMD-N-P2 showed good thermal
molecular hydrogen bonding on the perovskite perovskite structure (figs. S6 and S7). With IMD+, stability up to 200°C in air (fig. S9)

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 61

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

Fig. 2. Crystal structure description of CL-v. (A) A-, B-, and X-sites in the cubic
cell, with the edge center (E-site) reserved for interstitial B+ cation. (B) The pseudo-
cubic unit cell of the CL-v phase is shown, with the NH4-body[PF6]6 octahedron
represented in light cyan. One NH4+-edge is located in the upper (001) plane and the
other is in the bottom (001) plane. The E-site on (001) occupied with NH4+ has
an occupancy of one-half because of sharing with one adjacent unit in the layered
structure. CMD represents chloromethyl DABCO and BMD represents bromomethyl
DABCO. Both the B-site and E-site consist of NH4+ ions, differentiated by the color of
nitrogen for clarity. (C) The hydrogen bonding environment of the NH4+-edge,
with the NH4-edge, resulting in the formation of an NH4+-edge[PF6]3[N]2[Cl]
octahedron for stabilization of the NH4+-edge. Only half of CMD+ forms N-H···Cl bonds,
hence hydrogen bonded and non-bonded Cl are denoted in different colors. (D) The
geometry of two NH4-body[PF6]6 octahedrons with an NH4+-edge ion. Bond lengths in
angstroms (Å). (E) View along the ½001Šdirection. The unit vectors for NH4+-edge are
+
 
  of 11.6, 12.7, and 12.7 Å each. NH4 -edge
110 , ½130Š, and ½310Š, with lengths  and
 Cl atoms
are arranged linearly in the 110 direction and alternatingly in the 110  direction.
Inset is 9 × 7 × 0.5 mm size hexagonal single crystal of CMD-N-P2. (F) The
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 ½010Š direction with two
translucent red pyramid. The Cl of CMD+ in the adjacent layer forms an N-H···Cl bond layers indicating AB stacking.

62 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

Cryo-TEM, ~1.1 e- Å-2 Exp.

0.5 0.5
8 8 nm
nm

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
via cryo-TEM. (A) HRTEM image of CMD-N-P2 taken along the ½001Š direction at
~77 K with an electron dose rate of ~0.68 e− Å−2 s−1 and a cumulative dose of
~1.1 e− Å−2. The inset displays the corresponding FFT pattern. (B) A magnified high-
resolution image of CMD-N-P2 lattice fringes from a selected area of image (A),
and (C) the corresponding atomic model in the ½001Š axis. A bright spot in image
(B) corresponds to a column of organic groups in image (C), demonstrating a
perfectly matched periodic symmetrical structure, as emphasized by the blue
squares. The green double lines indicate the interplanar spacing corresponding to
the (800) lattice planes. Simulated experimental image (D) and reciprocal lattice
(E) are based on the ½001Š oriented structural model in (C). (F) The EELS spectrum
of CMD-N-P2 exhibits clear onset features of P-L, Cl-L, C-K, N-K, and F-K edges,
with a representative ADF image shown alongside. (G) HRTEM image, (H) atomic
model along the ½001Š zone axis, and (I) EELS spectrum of BMD-N-P2.

Cryo–transmission electron microscopy
(TEM) analysis
Direct evidence of the perovskite-type lattice
arrangement necessitates atomic TEM char-
acterization of the various molecules in the
lattice. We achieved near-atomic resolution
imaging of the pure organic perovskite crystal
structure under a cumulative electron dose of
no more than 2 e− Å−2 in a cryogenic TEM
(see supplementary materials for details and
imaging conditions). In the case of CMD-N-P2,
Fig. 3A displays its large-area, high-resolution
lattice-fringe image along the ½001Š zone axis.
This image was acquired through successive
short exposures at a dose rate of 0.68 e− Å−2 s−1
under a magnification of ×59000 (pixel size:
1.1 Å), with a total dose of ~1.1 e− Å−2. Columns
of organic groups linked by covalent bonds
were well-resolved in the enlarged HRTEM
image, displaying an ideal cubic perovskite
structure along the ½001Š projection, with a
homogeneous alternation of bright and dim
spots (Fig. 3B).
The square lattice, constructed by the small-
est bright or dim spots, corresponded to the
square repeating unit constructed by the col-
umns of the A-site (CMD+) plus B-site (NH4+)
or columns of X-site (PF6−) plus NH4+-edge. As
indicated by the blue boxes in Fig. 3B and 3C,
the lattice spacing observed in the experiment
completely aligned with that formed by CMD+,
and the atomic contrast was consistent with
the result of the simulated HRTEM image at
100 nm defocus value (Fig. 3D and fig. S10).
Additionally, the fast Fourier transform (FFT)
pattern inserted in Fig. 3A closely matched the
simulated reciprocal lattice (Fig. 3E), except
for some extinction reflections reappearing in
the FFT pattern because of dynamical scatter-
ing (thickness effect). The (800) reflection,
marked by the green arrow, exhibits the high-
est diffraction intensity in both the FFT and

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 63

RES EARCH | R E S E A R C H A R T I C L E S
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-
sponded to the lattice plane and is marked by
green lines in Fig. 3, B and C, with an inter-
planar spacing of 0.41 nm. The chemical com-
position of the CMD-N-P2 crystal, revealed by
electron energy loss spectroscopy (EELS) in
Fig. 3F, showed characteristic edges of P, Cl,
C, N, and F, which matched its mechanically
exfoliated bulk counterpart (figs. S11 and S12).
We also tested another CL-v phase based on
BMD-N-P2, which shared similar crystal struc-
ture and lattice spacing with CMD-N-P2, given
that only the chloromethyl functional group
in CMD+ was replaced with bromomethyl. As
depicted in Fig. 3G, the HRTEM image of
BMD-N-P2 crystals displayed a perovskite-
type arrangement of the organic functional
groups, similar to that shown in Fig. 3A with
well-matched EELS (fig. S13).
Layered structure and exfoliation
The layered structure of CMD-N-P2 was inves-
tigated with atomic force microscopy (AFM),
which revealed a step height of 4.2 nm. Taking
into account an overestimation of the thick-
ness by ~1 nm by AFM measurement owing
to tip-surface interactions, surface roughness,
and trapped solvent (31, 32), the closest thick-
ness that matches this is the 2.56-nm–thick
bilayer of the pseudo-cubic cell (fig. S2). Since
we did not measure anything thinner than
2.56 nm, we assume that the alternating step
is at least a bilayer. A distinctive attribute of
van der Waals–stacked 2D materials is their
ability to be exfoliated into molecularly thin
flakes through solvent-based exfoliation of bulk
crystals, and that the resulting flakes, assisted
64 5 APRIL 2024 • VOL 384 ISSUE 6691
by sonication, disperse in solvents and exhibit
the characteristic Tyndall effect of light scat-
tering by a colloidal dispersion. The Tyndall
effect of exfoliated CMD-N-P2 and BMD-N-P2
was weak in nonpolar solvents such as hexane
and toluene. However, for aprotic polar solvents
such as dichloromethane, the dispersion exhib-
ited a strong Tyndall effect (figs. S14 and S15).
We inferred that the interlayer region of
CMD-N-P2 encompassed N-H···Cl bonds, and
solvents containing chloro-functional groups
could disrupt these interlayer hydrogen bonds
by serving as H-bond acceptors. Thus, the ex-
foliated flakes were stabilized. Indeed, all
chloro-functionalized solvents, including dichlo-
romethane, 1,2-dichloroethane, 1-chlorohexane,
chlorobenzene, and 1,3-dichlorobenzene, exhib-
ited a strong Tyndall effect with CMD-N-P2.
Further confirmation of exfoliation was ob-
tained through AFM characterization, which
revealed that the flakes exfoliated by dichloro-
methane and 1,2-dichloroethane possessed a
thickness as thin as ~4 nm, similar to the step
height measured for layered crystal synthe-
sized in this work (fig. S16). The largest flake
size reached up to 30 mm when CMD-N-P2
was sonicated in dichloromethane (Fig. 4A).
Exploiting the ability of solvents with chloro-
functional groups to disrupt interlayer N-H···Cl
bonds, we successfully demonstrated controlled
growth of CMD-N-P2 on a silica substrate.
Gradual evaporation of a CMD-N-P2 solution
in the presence of 1,2-dichloroethane led to
the precipitation of hexagonal-shaped flakes
measuring 10 mm in width (Fig. 4B). We also
obtained similar results for BMD-N-P2 (fig. S17).
Moreover, by adjusting the concentration of
the CMD-N-P2 solution, we can synthesize crys-
tals as thin as ~4 nm (Fig. 4C). Thus, our find-
ings highlight the potential to grow 2D layered
organic crystals using a solution-based ap-
proach while achieving meticulous thickness
control. The crystalline nature of the CMD-N-P2
flakes grown using this method was evident
from the presence of screw dislocations in
some flakes, exhibiting both clockwise (Fig. 4D)
and counterclockwise (fig. S18) dislocations
with a step height of ~4 nm (33, 34).
Applications of gate dielectric in
2D electronics
The ability of 2D all-organic perovskites to
form molecularly thin films over large areas,
combined with their insulating nature, makes
them suitable as dielectric layers in 2D elec-
tronics. The optical band gaps of CMD-N-P2
and BMD-N-P2 were determined from the in-
tercepts of the Tauc plots to be 5.56 eV and
4.69 eV, respectively (Fig. 5A). The wave func-
tions were mostly localized in the molecules in
these crystals, yielding a large band gap and
small bandwidth <0.02 eV, as seen from the
band structure calculated with density func-
tional theory (DFT) (fig. S19). The DFT-calculated
band gaps of 4.81 and 4.14 eV of CMD-N-P2
and BMD-N-P2 were underestimated but
agreed with the experimental trends qualita-
tively and indicate that the Coulombic poten-
tial of atomic nuclei such as Cl and Br affects
the gap (fig. S19) (29).
The dielectric constants of CMD-N-P2 and
BMD-N-P2 were measured in the range of
1 kHz to 1 MHz, giving values of 4.86, and
5.53, respectively (Fig. 5B), which were higher
science.org SCIENCE

158 mV dec-1
Perovskite ION/IOFF > 10
50 m
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.
lower than HfO2 (25). The breakdown fields through atomic-layer deposition (36) The drain 4.
5.
of CMD-N-P2 and BMD-N-P2 are found to be current–voltage (Id–Vd) output curves exhibit 6.
23 MV m−1 and 17 MV m−1, respectively (fig. linear characteristics at low bias. (Fig. 5F)
S21). These values are higher than those of 7.
hBN (12 MV m−1), Alumina (13.4 MV m−1) and Conclusions
8.
polystyrene (19.7 MV m−1). Compared to 2D We designed and synthesized a new class of 9.
dielectric materials such as hexagonal boron 2D organic perovskites known as the CL-v phase.
nitride, CL-v phases have both a high dielectric Our synthetic approach is versatile and can 10.
constant and a large band gap, which makes be extended to various organic molecules as 11.
12.
them suitable for use as dielectric layers in 2D A-site cations, including diazabicycloalkane
electronics (Fig. 5C). To demonstrate the ef- (37, 38), azaadamantane (39), and other deriv- 13.
fectiveness of CL-v phases as gate dielectric atives of heterocyclic compounds (40), as well
14.
layers, we fabricated a field-effect transistor as different B-site cations and X-site anions 15.
(FET) by transferring ≈100 nm-thick CMD-N-P2 (fig. S23). CL-v phase can be exfoliated or 16.
or BMD-N-P2 as the top gate dielectric layer on grown as molecular thin crystalline organic
17.
MoS2 as the channel material, and with Cr/ layers with large aspect ratio by solution pro- 18.
Au metal as the source and drain electrodes. cess, which can be potentially used as flexible 19.
Figure 5D shows an optical image of a fab- dielectrics. Our findings enhance the under-
20.
ricated FET device. The measured transfer standing of structural design principles per-
21.
curve at different Vds displays n-type behavior taining to all-organic perovskites. Based on
and the on-off ratio is >107 when the Vds is this understanding, we posit that it is fea- 22.
0.5 V, which is sufficient to comply with the stan- sible to conceive and synthesize all-organic
23.
dard of practical logic circuits of 104 (Fig. 5E) (35). perovskite analogs of the Dion-Jacobsen and
The negligible hysteresis in the transfer curve Aurivillius phases. 24.
suggested that a clean contact formed between
25.
the dielectric and channel material. Further- RE FERENCES AND NOTES 26.
more, a low subthreshold swing of 158 mV per 1. W. Zheng et al., Adv. Mater. 35, e2205410 (2023). 27.
decade was extracted from the transfer curve, 2. L. T. Nguyen, R. J. Cava, Chem. Rev. 121, 2935–2965 (2021). 28.
SCIENCE science.org
RESE ARCH | R E S E A R C H A R T I C L E S
SiO Al
HfSiO
hBN Y
Si N
CMD-N-P
BMD-N-P
COF
7V
5V
3V
1V
-1 V
-3 V
-5 V
-7 V
S. Royer et al., Chem. Rev. 114, 10292–10368 (2014).
M. A. Peña, J. L. G. Fierro, Chem. Rev. 101, 1981–2017 (2001).
L. N. Quan et al., Chem. Rev. 119, 7444–7477 (2019).
J. Y. Kim, J.-W. Lee, H. S. Jung, H. Shin, N.-G. Park, Chem. Rev.
120, 7867–7918 (2020).
S. N. Ruddlesden, P. Popper, Acta Crystallogr. 10, 538–539
(1957).
S. N. Ruddlesden, P. Popper, Acta Crystallogr. 11, 54–55 (1958).
M. Dion, M. Ganne, M. Tournoux, Mater. Res. Bull. 16,
1429–1435 (1981).
B. Aurivillius, Ark. Kemi 1, 463–480 (1949).
B. Aurivillius, Ark. Kemi 2, 519–527 (1950).
C. A. Bremner, M. Simpson, W. T. A. Harrison, J. Am. Chem. Soc.
124, 10960–10961 (2002).
G.-Z. Liu, J. Zhang, L.-Y. Wang, Synth. React. Inorg. Met.-Org.
Nano-Met. Chem. 41, 1091–1094 (2011).
H.-Y. Ye et al., Science 361, 151–155 (2018).
S.-L. Chen et al., Sci. China Mater. 61, 1123–1128 (2018).
G.-M. Fan, C. Shi, L. Qiao, H.-J. Li, H.-Y. Ye, J. Mater. Chem. C
Mater. Opt. Electron. Devices 6, 8349–8352 (2018).
H. Morita et al., Angew. Chem. Int. Ed. 58, 9184–9187 (2019).
T. W. Kasel et al., Chem. Sci. 10, 8187–8194 (2019).
W.-X. Zhang, S.-L. Chen, Y. Shang, Z.-H. Yu, X.-M. Chen,
Energetic Materials Frontiers 1, 123–135 (2020).
Y. Shang et al., Cryst. Growth Des. 20, 1891–1897 (2020).
H. Morita, R. Tsunashima, S. Nishihara, T. Akutagawa,
CrystEngComm 22, 2279–2282 (2020).
J.-J. Wang, D. Fortino, B. Wang, X. Zhao, L.-Q. Chen, Adv. Mater.
32, e1906224 (2020).
X. Song, G. Hodes, K. Zhao, S. Liu, Adv. Energy Mater. 11,
2003331 (2021).
Y. Gao et al., ACS Appl. Mater. Interfaces 13, 19042–19047
(2021).
H. S. Choi et al., Nat. Commun. 13, 794 (2022).
H.-S. Wu et al., Adv. Sci 9, e2105974 (2022).
R. Taheri-Ledari et al., Energy Fuels 36, 10702–10720 (2022).
T. Handa et al., Science Advances 8, eabo1621 (2022).
5 APRIL 2024 • VOL 384 ISSUE 6691 65
RES EARCH | R E S E A R C H A R T I C L E S

29. Z. Li et al., Adv. Mater. 35, e2300480 (2023).
30. L. Mao, C. C. Stoumpos, M. G. Kanatzidis, J. Am. Chem. Soc.
141, 1171–1190 (2019).
31. C. J. Shearer, A. D. Slattery, A. J. Stapleton, J. G. Shapter,
C. T. Gibson, Nanotechnology 27, 125704 (2016).
32. X. Li et al., Nat. Chem. 12, 1115–1122 (2020).
33. Y. Zhao, S. Jin, Acc. Mater. Res. 3, 369–378 (2022).
34. Y. Zhao et al., Science 370, 442–445 (2020).
35. M. Chhowalla, D. Jena, H. Zhang, Nat. Rev. Mater. 1, 16052
(2016).
36. A. Sebastian, R. Pendurthi, T. H. Choudhury, J. M. Redwing,
S. Das, Nat. Commun. 12, 693 (2021).
37. S. F. Nelsen, J. M. Buschek, J. Am. Chem. Soc. 96, 7930–7934
(1974).
38. S. F. Nelsen, P. J. Hintz, J. Am. Chem. Soc. 94, 7114–7117 (1972).
39. A. I. Kuznetsov, V. A. Kosmakov, B. V. Unkovskii, Chem.
Heterocycl. Compd. 21, 697–700 (1985).
40. M. W. Giuliano, C.-Y. Lin, D. K. Romney, S. J. Miller, E. V. Anslyn,
Adv. Synth. Catal. 357, 2301–2309 (2015).
ACKN OW LEDG MEN TS
Funding: This work was supported by project P0043087 Design
and Synthesis of Hybrid Organic Inorganic Perovskites of Hong
Kong Polytechnic University as well as funding from Hong Kong
Jockey Club STEM lab of 2D Quantum Material (2023-0033). K.L.
acknowledges the Croucher Foundation (Croucher Tak Wah Mak
Innovation fund 2023) and projects 62322413 and 12104382
supported by National Natural Science Foundation of China and
the Research Grants Council of the Hong Kong Special
Administrative Region, China (projects PolyU25305222,
PolyU15305221, and PolyU15304623) J.Y. acknowledges financial
support from Hong Kong Polytechnic University (grant P0042930)
and a grant from the Research Grants Council of Hong Kong
(project PolyU 25300823).This project is partially supported by the
National Natural Science Foundation of China (11974156),
Guangdong Innovative and Entrepreneurial Research Team Program
(grant 2019ZT08C044), Shenzhen Science and Technology Program
(KQTD20190929173815000 and 20200925161102001), the Science,
Technology and Innovation Commission of Shenzhen Municipality
(ZDSYS20190902092905285). TEM/STEM characterization was
performed at the cryo-EM Center and Pico Center from SUSTech Core
Research Facilities that receives support from the Presidential Fund
and Development and Reform Commission of Shenzhen Municipality.
Author contributions: H.S.C. and K.P.L conceived the idea and
developed the project plan. H.S.C. designed, synthesized, grew
bilayers, and characterized the CL-v crystals. K.P.L. supervised the
research. J.L. and K.L. fabricated and characterized the FET device.
G.W., F.L., and J.L. performed cryo-TEM measurements. W.P.D.W
and I.H.P. conducted the single-crystal XRD structure analysis. J.Y.
conducted DFT calculations of band gap. H.S.C. and K.P.L. wrote the
manuscript with contributions from all authors. H.S.C., J.L., and
G.W. contributed equally. Competing interests: The authors declare
no competing interests. Data and materials availability: The data
that support the plots within this article and other findings of this
study are available from the corresponding author upon reasonable
request. License information: Copyright © 2024 the authors, some
rights reserved; exclusive licensee American Association for the
Advancement of Science. No claim to original US government works.
https://www.science.org/about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adk8912
Materials and Methods
Figs. S1 to S23
Tables S1 to S3
References (41–43)
Submitted 19 October 2023; accepted 22 February 2024
10.1126/science.adk8912

ASTHMA Epithelia that line the airways provide a

protective barrier for the lungs, acting as a first
Bronchoconstriction damages airway epithelia by line of defense in innate immunity (4–7). Cen-
tral to providing a tight barrier to the outside
crowding-induced excess cell extrusion world is maintaining a constant density of epi-
thelial cells while they turn over by cell division
Dustin C. Bagley1, Tobias Russell1†, Elena Ortiz-Zapater2†, Sally Stinson3, Kristina Fox4, and death. We discovered a conserved process
Polly F. Redd5, Merry Joseph6, Cassandra Deering-Rice7, Christopher Reilly7, Maddy Parsons1, that is essential for preserving the barrier and
Christopher Brightling3, Jody Rosenblatt1,8* epithelial cell number homeostasis called cell
extrusion (8, 9). Extrusion mechanically links
Asthma is deemed an inflammatory disease, yet the defining diagnostic feature is mechanical the number of cells dying with those dividing
bronchoconstriction. We previously discovered a conserved process called cell extrusion that by triggering some cells to seamlessly squeeze
drives homeostatic epithelial cell death when cells become too crowded. In this work, we show out of the layer when it becomes too crowded
that the pathological crowding of a bronchoconstrictive attack causes so much epithelial cell (8–10). Given that mild physiological crowding
extrusion that it damages the airways, resulting in inflammation and mucus secretion in both causes homeostatic cell extrusion responses, we
mice and humans. Although relaxing the airways with the rescue treatment albuterol did not postulated that pathological crowding from a
affect these responses, inhibiting live cell extrusion signaling during bronchoconstriction bronchoconstrictive episode might trigger so
prevented all these features. Our findings show that bronchoconstriction causes epithelial much cell extrusion that it would disrupt the
damage and inflammation by excess crowding-induced cell extrusion and suggest that blocking airway epithelial barrier. Destruction of the
epithelial extrusion, instead of the ensuing downstream inflammation, could prevent the feed- airway barrier could then lead to the increased
forward asthma inflammatory cycle. airway inflammation and/or the elevated sus-
ceptibility to viral and bacterial infection that
frequently perpetuate the asthma inflamma-
ore than 300 million people globally inflammation for weeks to months that can tory cycle (11). We investigated whether bron-

M suffer from asthma, with ~1000 dying

from it daily. All asthma exacerbations
are characterized by bronchoconstriction,
which causes breathing difficulty, wheez-
ing, and increased airway mucus production. After
a severe exacerbation, patients with asthma fre-
predispose individuals to further attacks (1).
Although diverse stimuli can trigger asthma
attacks that cause different types of immune
responses, the universal life-threatening pathol-
ogy shared by asthmatics is mechanical bron-
choconstriction. In fact, a bronchoconstrictive
choconstriction triggers excess airway epithelial
extrusion and, if so, whether targeting extrusion
could reduce the damage and inflammation,
which can lead to further attacks.

Bronchoconstriction causes excess

quently experience an extended period of airway
1
The Randall Centre for Cell & Molecular Biophysics, School
of Basic & Medical Biosciences, King’s College London,
London SE1 1UL, UK. 2Department of Biochemistry and
Molecular Biology, University of Valencia, 46010 Valencia,
Spain. 3Institute for Lung Health, Leicester NIHR BRC,
University of Leicester, Leicester LE3 9QP, UK. 4Edwards Life
Sciences, Draper, UT 84020, USA. 5University of Utah, Salt
Lake City, UT 84112, USA. 6University of Utah School of
Medicine, Salt Lake City, UT 84132, USA. 7College of
Pharmacy, University of Utah, Salt Lake City, UT 84112, USA.
8
School of Cancer and Pharmaceutical Sciences, King’s
College London, London SE1 1UL, UK.
*Corresponding author. Email: jody.rosenblatt@kcl.ac.uk
†These authors contributed equally to this work.
response to methacholine (MCH) challenge is
the standard diagnostic test for asthma (2).
Current therapies include albuterol, a short-
acting b2 adrenergic receptor agonist that re-
laxes airway smooth muscle (ASM) to relieve
constriction, and inhaled corticosteroids, which
reduce eosinophilic and type 2 airway inflam-
mation (3). Although these therapies help,
many patients continue to experience poor
symptom control and airway hyperrespon-
siveness. Thus, identifying other etiologies
driving asthma morbidity and mortality is
a priority.
crowding-induced extrusion
To test whether experimental bronchoconstric-
tion can directly induce airway epithelial cell
extrusion, we treated unprimed or immune-
primed ex vivo mouse lung slices with MCH,
which triggers ASM encircling the epithelial
barrier to contract (12, 13). To prime the airways,
we used several published ovalbumin (OVA)
and house dust mite (HDM) immune-priming
methods (14–19) (fig. S1, A and B) that produce
characteristic asthma T helper 2 (TH2) cell
inflammatory responses, marked by increased
inflammatory cytokines interleukin-4 (IL-4)

66 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

High [MCH]
A B REDUCTION IN AW LUMEN C EXTRUSIONS FOLLOWING MCH
**** 100 **** 1

Complete denuding of airway

UNPRIMED

300000

Extrusions/bronchiole
200000
50

100000

Lumen area (µm2 )

0
0
PRIMED

Unprimed Primed
60000

E EXTRUSIONS FOLLOWING MCH

**** ****

Complete denuding of airway

Extrusions/bronchiole
40000

D No MCH Low [MCH] High [MCH]

****

**
20000
PRIMED

NM 0.1 0.2 0.35 0.5

0
MCH (g/mL)
Actin E-cadherin DAPI 0min 15min
TIME IN MCH
G % OF DENUDING
**** ns
F MCH

% 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-
sion. (A) Movie stills of 500 mg/ml MCH response in unprimed and 2-week
HDM-primed bronchioles (minutes:seconds), with red arrowheads pointing to
single-cell extrusions (scale bars, 50 mm). (B) Constriction scales with size,
measuring the bronchiolar lumen areas before and after MCH treatment
in large, medium, and small bronchioles from >5 airways from 5 HDM-primed
mice (****P < 0.0001 from a Wilcoxon pairs-matched signed rank test).
(C) Quantification of extrusions from immunostained ex vivo lung slices from
unprimed and OVA-primed mice treated with 500 mg/ml MCH (****P < 0.0005
from an unpaired Mann-Whitney analysis compared with control from >5 lung
slices from >10 mice), where blue data points represent complete epithelial
denuding. (D) Sample confocal projections (scale bar, 50 mm) of bronchioles
after 200 mg/ml (low) or 500 mg/ml (high) MCH, where red arrowheads
depict single-cell extrusions and blue ones depict epithelial denuding
(<100 extrusions on graphs; blue data points). (E) Increasing MCH concentration
increases extrusion (**P < 0.0005; ****P < 0.0001) in >15 slices per treatment
from >5 mice. (F and G) Movie stills (F) from a 5-week HDM-primed ex vivo
lung slice treated with 500 mg/ml MCH for 15 min and then relaxed with
3.5 mM ALB + 500 mg/ml MCH for 15 min, showing that epithelia still detach
(blue arrowheads; scale bars, 100 mm), quantified in (G) with no significant
difference (ns) between MCH and ALB from a Mann-Whitney test
(****P < 0.0001).

and IL-13 and high mucus production (fig. S1,
C to F). MCH addition caused pronounced
bronchoconstriction of mouse lungs that were
immune-primed with either OVA or HDM
[as outlined in fig. S1 (14–19)] but did not sub-
stantially affect unprimed airways (Fig. 1, A
and B, and movies S1 to S4). Within 15 min of
MCH treatment, the luminal area of small and
medium bronchioles reduced markedly, caus-
ing severe airway epithelial crowding (Fig. 1B)
and increasing cell heights on average 196 ±
11% (SEM; 13 airways from n = 5 mice). Fifteen
minutes of bronchoconstriction caused excess
airway epithelial extrusion in primed ex vivo
mouse slices, sparing unprimed controls (Fig. 1,
A and C). To quantify the number of extrusions
in response to MCH, we immunostained lung
slices with E-cadherin for epithelia, phalloidin
for ASM, and 4′,6-diamidino-2-phenylindole
(DAPI) for DNA, scoring for extrusions as cells
pinching off apically into the lumen (Fig. 1, C
to E, black data points and red arrowheads)
and complete epithelial denuding (Fig. 1, C to
E, blue data points and blue arrowheads). All
immune-priming methods induced pronounced
bronchoconstriction and excess extrusion in re-
sponse to MCH compared with unprimed con-
trol mice, with HDM priming showing the
strongest effects (Fig. 1, A and B, and movies
S2 to S4). Additionally, we occasionally noted
the unjamming of airway epithelia from a
previous static state to collective migration
(movie S5). To test whether the amount of
constriction correlated with extrusion rates,
we used increasing MCH doses on OVA-primed
mice and found a tight correlation between the
amount of constriction and cell extrusions, with
the highest doses causing complete denuding of
the epithelium (Fig. 1, D and E).

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 67

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

A CROWDING-INDUCED EPITHELIAL CELL EXTRUSION PATHWAY & INHIBITORS: B S1P ACTIN DAPI

MCH Gd 3+ SKI II/K-145

EPITHELIAL SA/TRP CELL

S1P S1P 2 EXTRUSION
CROWDING CHANNELS DEATH

GsMTx4 JTE-013

C No MCH MCH Gd3++MCH SKI-II/K-145+MCH JTE-013+MCH

Ecad Actin DAPI

D # EXTRUSIONS/BRONCHIOLE E % DENUDED CELLS/BRONCHIOLE H

0min

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-
constriction. (A) Canonical crowding-induced epithelial cell extrusion pathway,
indicating where each inhibitor acts within the pathway. (B) Confocal projection
of an extruding cell immunostained for S1P, phalloidin for f-actin, and DAPI
(scale bar, 25 mm). (C to E) Confocal projections (C) of ex vivo lung slices from
5-week HDM-primed mice with no MCH (n = 9), MCH (n = 9), or pretreated with
Gd3+ (n = 9), sphingosine kinase inhibitors SKI II and K-145 (n = 9), or S1P2
antagonist JTE-013 (n = 4), before adding MCH (scale bars, 50 mm), quantified in
(D) as extruded cells (red arrowheads) and (E) as percentage of epithelial
denuding (blue arrowheads) per bronchiole (****P > 0.0001 from a Kruskal-
Wallis test, from >4 slices per mouse from >4 mice). (F and G) Quantification of
OVA-primed lung slices treated with Gd3+ or GsMTx4 after 15 min of MCH
challenge, where black dots represent extrusions per bronchiole and blue dots
represent complete epithelial denuding from >5 mice (F) or percent denuding (G)
(*P < 0.05; **P < 0.001; ***P < 0.0005; ****P < 0.0001 from a Mann-Whitney
test, from >4 slices per mouse from >4 mice). (H) Movie stills from an HDM-
primed mouse lung slice pretreated with MCH for 15 min, followed by ALB
(3.5 mM) + MCH (500 mg/ml), with arrowheads pointing to areas of epithelial
denuding that reattach by 32 min (scale bars, 50 mm). Note that albuterol
alone did not prevent epithelial destruction (G).

68 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

A In vivo MCH challenge:

-/+ treatment 6.25mg MCH (-/+ treatment) 12.5mg MCH (-/+ treatment) 25mg MCH (-/+ treatment) (2X) 50mg MCH (-/+ treatment) Fix at 30min/24hrs PM
5min 2min 2min 2min 2min Histology
3+
B No MCH MCH (30min PM) MCH+ALB (30min PM) MCH+ALB/Gd (30min PM)

C extrusions post MCH extrusions/airway (30min PM) denuding post MCH denuding/airway (30min PM)
**** ****
**** ** **** ** **

% of denuded cells per bronchiole

50
100
# of extrusions per bronchiole

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+

E Inflammatory scoring F Inflammatory scores at 24hrs post-challenge

* **
0 1 ns
100
% of each pathological score

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-
choconstrictive attack in live mice. (A) Schematic of live MCH challenges ±
5 min of pretreatment with extrusion inhibitors before increasing MCH ± inhibitors.
(B and C) H&E sections from lungs of mice not exposed to MCH (n = 4) or
treated with MCH (n = 5), MCH + ALB (n = 5), or MCH + ALB/Gd3+ (n = 6) at
30 min after MCH, with the number of extrusions (red arrowheads) and percent
denuding (blue dotted outline) per bronchiole quantified in (C), respectively,
with representative images. Scale bars, 50 mm (**P < 0.001; ***P < 0.0005;
****P < 0.0001 from a Mann-Whitney test). (D to F) H&E sections 24 hours after
MCH challenge to observe immune response from no MCH treatment (n = 4), MCH
alone (n = 5), MCH + ALB (n = 5), MCH + ALB/SKI (n = 3), or MCH + ALB/Gd3+
(n = 5), with representative images shown in (D) and quantification of airway immune
cell infiltration using pathological scores, defined by colors and numbers in (E) and
(F) (*P < 0.05; **P < 0.001 from a Chi-squared test). Scale bars, 100 mm.

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 69

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

A UNPRIMED HDM (2wk) HDM (5wk) HDM post MCH

DAPI ACTIN MUC5AC

B MCH MCH+Gd 3+

0min 15min 0min 15min

WGA-350

C 30MIN POST MCH D

No MC H MC H MUCUS SCORING
0 IN VIVO MUCUS SCORING (30MIN)

**** ****
**
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.
(A) Confocal projections of Muc5AC before and after 2 and 5 weeks of HDM
priming (scale bars, 50 mm). (B) Movie stills from 3-week HDM-primed
ex vivo slices incubated with 10 mg/mL WGA-350 to label mucus then
treated with 500 mg/ml MCH alone ± 10 mM Gd3+Cl for 15 min, noting that
WGA is retained in the epithelium rather than secreted with gadolinium.
(C and D) Representative PAS (mucus) staining (C) 30 min after a MCH
challenge ± ALB or ALB/Gd3+, which was quantified using a color/number scoring
in (D), with **P < 0.001 and ****P < 0.0001 from Chi-square tests from at
least 5 mice per group. Scale bars, 50 mm.

Reversing bronchoconstriction does not
prevent extrusion
Alleviating airway constriction with albuterol
could potentially reverse the excess epithelial
extrusion and denuding. However, we found
that although albuterol relaxes airways that
were bronchoconstricted for 15 min, it did not
prevent airway epithelial extrusion and des-
truction (Fig. 1, F and G, and movie S6). In fact,
movies show that epithelia frequently detached
from the smooth muscle as the ASM sprang
open with albuterol administration, whereas
epithelia remained buckled (movie S6 and
Fig. 1, F and G, blue arrowheads). For this
reason, we quantified the percent of epithelial
cell denuding per bronchiole rather than
extrusions (Fig. 1G). Thus, albuterol does not
prevent destruction of the airway lining after a
bronchoconstrictive attack. Moreover, the in-
creased denuding that we noted could poten-
tially impede airway repair.

70 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

A B

Fig. 5. Treated patients with asthma have marked airway epithelial extru-
sion and damage. (A and B) H&E pathology sections from lower lobectomy of a
nonsmoker patient with moderate asthma symptoms medicated with regular
moderate–dose inhaled corticosteroid and long-acting b2-agonist (formoterol)
and as-needed inhaled short-acting b2-agonist (albuterol) (A) and a nonsmoker
patient with severe asthma on high-dose inhaled corticosteroid, long-acting b2-
agonist (formoterol), oral leukotriene receptor antagonist, oral prednisone, and
as-needed inhaled short-acting b2-agonist (albuterol) (B), where dashed-line
boxes indicate areas enlarged below, with blue arrowheads indicating breaks in
the epithelium and red arrowheads indicating epithelia extruded into the
lumen. Note, lumen filled with mucus and extruded cells even in the moderate
asthma case [(A), right]. (C) Model describing how the mechanics of asthmatic
bronchoconstriction cause excessive crowding-induced mucus secretion and
epithelial cell extrusion that results in epithelial damage and inflammation.
Inhibiting extrusion early in the pathway prevents all pathological consequences
of an attack. AW, airway.

Inhibiting extrusion blocks airway
epithelial damage
We next investigated whether canonical extru-
sion inhibitors could prevent bronchoconstric-
tive airway damage. We previously discovered
that crowding-induced live cell extrusion re-
quires the stretch-activated channel (SAC) Piezo1
to trigger production of the bioactive lipid
sphingosine 1–phosphate (S1P) that binds the
S1P2 receptor to activate Rho-mediated actomy-
osin contraction needed to seamlessly eject a
cell from the monolayer (8, 20) (Fig. 2A, sche-
matic, and fig. S2A). As has been seen previ-
ously with crowding-induced extrusion (8),
few of the extruding cells are apoptotic (fig.
S2B). Extruding airway epithelial cells speci-
fically up-regulate S1P, the limiting signal for
extrusion, which suggests that bronchocon-
strictive extrusion operates through the canoni-
cal pathway (Fig. 2B and fig. S2A). Although
Piezo1—the SAC that we identified as critical
for activating extrusion in response to crowd-
ing in Madin-Darby canine kidney (MDCK)
cells and zebrafish (8)—is expressed in mouse
airways, they also express the transient recep-
tor protein (TRP) channels TRPA1, TRPV1, and
TRPM8 frequently up-regulated in unmanageable
asthma (21) that could act similarly to trigger
extrusion (fig. S2, C to F). Notably, immune
priming causes Piezo1, TRPV1, and TRPM8 to

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 71

RES EARCH | R E S E A R C H A R T I C L E S
become more vesicular and widespread (fig.
S2, C to F). Thus, to inhibit Piezo1 as well as
other potential TRP channels, we used gado-
linium hexahydrate chloride (Gd3+), an inhib-
itor of SACs and TRP channels. Additionally,
we blocked S1P production with sphingosine
kinase 1 (SKI II) and 2 (K-145) inhibitors and
its receptor S1P2 with the antagonist JTE-013
during MCH-induced bronchoconstriction and
quantified extrusion and denuding. All inhib-
itors markedly decreased extrusions after MCH
(Fig. 2, C and D); however, only Gd3+ and SK
inhibitors blocked epithelial sheet denuding as
well (Fig. 2, C and E). It is not clear why only
Gd3+ and SK inhibitors block detachment, but
it suggests that blocking extrusion upstream in
the pathway offers greater protection to airway
epithelium.
We next tested whether we could inhibit ex-
trusion in a more clinically relevant way—after
the onset of bronchoconstriction and in the
presence of albuterol. Thus, we triggered bron-
choconstriction with MCH for 15 min and
then added SAC inhibitors in the presence
of MCH. Gd3+ administered after broncho-
constriction substantially reduced extrusion
and epithelial denuding (Fig. 2, F and G).
Additionally, the peptide inhibitor, GsMTx4,
which blocks Piezo1, TRPC1, and TRPC6 (22),
also blocks extrusion when added 15 min after
bronchoconstriction (Fig. 2F). Albuterol did
not impair the ability of Gd3+ to block epi-
thelial extrusion or denuding after 15 min of
MCH-induced bronchoconstriction, nor did Gd3+
affect bronchodilation by albuterol (Fig. 2, G
and H, and movies S7 and S8). Our videos
show that Gd3+ added to albuterol allows the
airway epithelia to reattach to the underlying
smooth muscle as it relaxes, thereby prevent-
ing denuding (Fig. 2H, blue arrowheads after
they detach, with red arrowheads indicating
extruded cells; movies S7 and S8). It is un-
clear why blocking upstream in the extrusion
pathway allows epithelial reattachment, but
we suggest that it prevents the secretion of
matrix proteases predicted to be required during
the extrusion process.
Blocking extrusion during bronchoconstriction
prevents inflammation
The ability of Gd3+ to prevent bronchoconstriction-
induced airway epithelial extrusion and de-
struction suggested that it may prevent the
inflammation that typically follows an asthma
attack. To test this hypothesis, we challenged
live HDM-primed mice with albuterol and in-
hibitors of extrusion—compared with control
and MCH alone—and assayed for extrusions,
denuding [30 min post-MCH (PM)], and in-
flammation (24 hours PM) by hematoxylin
and eosin (H&E) staining. We identified 50 mg/ml
MCH (1/10 the concentration used in ex vivo
experiments) as the lowest dose that would
trigger bronchoconstriction and extrusion
72 5 APRIL 2024 • VOL 384 ISSUE 6691
without causing undue harm and stress to mice,
delivering it in increasing amounts, as adminis-
tered with patients, to ensure that no mice were
hyperresponsive (Fig. 3A, schematic). We then
administered Gd3+ or sphingosine kinase inhib-
itors (SKIs) during MCH challenge to test whe-
ther blocking early steps in the extrusion pathway
that protected epithelia in our ex vivo assay would
prevent subsequent inflammation in our in vivo
assay. SKIs can be repeatedly administered
safely to mice intranasally (23), and we found
that mouse lung gross morphology and tissue
sections were indistinguishable from those
of control mice (fig. S3A) after intranasal ins-
tillation of 10 mM Gd3+ for 5 consecutive days
(fig. S3B) or once a week for 3 weeks (fig. S3C).
We found no obvious masses, inflammation, or
obvious changes to mouse health and behavior
(movies S9 and S10). Because albuterol does
not impede the ability of Gd3+ to block extrusion
(Fig. 2, G and H; fig. S4C; and movies S7 and S8)
and is necessary for opening airways in patients,
we used albuterol with Gd3+ treatment in all of
our live mouse studies to reduce total numbers
of mice. Histology slices 30 min after MCH
challenge with or without albuterol resulted in
high numbers of extrusion (Fig. 3, B and C, red
arrowheads) and denuded bronchioles (Fig. 3,
B and C, red outline). Yet, the addition of Gd3+
with albuterol preserved airway epithelia, simi-
larly to no MCH challenge (NM) (Fig. 3, B and
C). The damage incurred during bronchocon-
striction with or without albuterol resulted in
pronounced immune cell infiltration by 24 hours
compared with immune-primed unchallenged
bronchioles (Fig. 3, D to F). Inhibiting extru-
sion with gadolinium or SKIs (with albuterol)
reduced the inflammatory response to levels
seen in control immune-primed, unchallenged
bronchioles (Fig. 3, D to F). Similar results were
obtained with OVA-primed mice, even when
Gd3+ was delivered after the MCH ramp-up (fig.
S4). Thus, preventing bronchoconstriction-
induced extrusion, particularly with gadolinium,
prevents the inflammation that typically fol-
lows an attack.
SAC inhibitors also prevent mucus secretion
Although asthma typically causes greater dam-
age to the distal airways, most asthma patients
experience notable difficulties with excess
mucus secretion from the proximal larger
airways. Immune priming with OVA or HDM
predictably increased mucus production and
secretion, as measured by quantitative reverse
transcription polymerase chain reaction (qRT-
PCR) (fig. S1E), periodic acid–Schiff (PAS)
staining of primary airways (fig. S1, E and F),
and Muc5AC immunostained confocal projec-
tions (Fig. 4A) (14, 15, 24–27). Live imaging
of primary airways loaded with wheat germ
agglutinin–350 (WGA-350) to label mucus from
3-week HDM-primed mice revealed that MCH
induces rapid mucus secretion with broncho-
constriction, as shown by the loss of blue
fluorescence (Fig. 4B and movies S11 and S12).
Unexpectedly, Gd3+ markedly reduced mucus
secretion in primary airways in response to
our ex vivo MCH challenges (Fig. 4B and movie
S13). Although our WGA-350 assay is not a very
quantitative assay because of the uneven ex-
pression of mucus from immune priming and
WGA loading, we found that ~90% of control
primary airways secrete mucus, whereas only
~26% of Gd3+-treated ones do. In vivo, we saw
a similar response in both OVA- and HDM-
primed mice. PAS staining from HDM-primed
mice challenged with MCH for ~30 min showed
bulk mucus secretion with large mucus globules
squeezing out apically, which was not prevented
by albuterol (Fig. 4, C and D), whereas Gd3+
treatment with albuterol substantially reduced
mucus secretion (Fig. 4, C and D). Similar
results were found in OVA-primed mice (fig. S5).
Corticosteroid-treated asthma patients have
marked airway epithelial extrusion and damage
Having found that bronchoconstriction causes
excess airway epithelial cell extrusion and
destruction in mice, we investigated whether
a similar scenario occurs in humans. We ob-
tained small airway resections from asthma
patients undergoing lobectomy for cancer.
Biopsy samples from individuals with either
moderate (Fig. 5A) or severe (Fig. 5B) asthma
had even more extrusion, denuding, and mucus
secretion compared with our primed mice after
bronchoconstriction. Both patients were treated
with both corticosteroids and smooth muscle–
relaxing medications, underscoring their in-
ability to impede airway epithelial damage
from asthma attacks. These individuals are
not unusual, as many reports note airway epi-
thelial sloughing or extrusion as a defining
pathological feature in mild to severe or fatal
cases, even in racehorses (28–30), which sug-
gests that excess airway epithelial extrusion
is a bronchoconstriction-associated pathology
conserved throughout different species, inde-
pendent of airway size.
Discussion
In this work, we present a mechanical etiology
for asthma. Whereas most asthma studies have
focused on the inflammatory signaling associ-
ated with asthma, our work suggests that the
inflammation and mucus secretion result from
the mechanics of bronchoconstriction on airway
epithelium. We find that bronchoconstriction
causes pathological airway epithelial crowding,
leading to so much cell extrusion that it destroys
the barrier, resulting in the typical postattack
inflammation (Fig. 5C). Because epithelia act
as the first line of defense against pathogens
and toxins, epithelial barrier disruption could
also cause infection hypersensitivity until air-
ways have repaired. Infections and inflamma-
tion could then lead to more bronchoconstrictive
science.org SCIENCE

attacks, triggering the asthma inflammatory
cycle (31). Albuterol treatment, routinely used
by asthma patients for symptom relief of bron-
chospasm, does not prevent airway epithelia
destruction, mucus secretion, or inflammation
after an asthma attack. However, blocking
the extrusion pathway with gadolinium or
S1P inhibitors after bronchoconstriction pre-
served epithelial integrity, substantially damp-
ening the inflammatory response. Gadolinium
also blocked mucus secretion from primary air-
ways, which suggests that crowding activation
of calcium channels mechanically induces bulk
mucus secretion.
Although our experiments defined excess
airway epithelial cell extrusion as important
for triggering many symptoms after an asthma
attack, we are not the first to examine the
effects of the mechanics of epithelia in asthma.
Many have noted the important role of epi-
thelia in asthma (4, 32–34). Additionally, com-
pression forces can cause airway epithelia to
unjam from their previous static, polygonal
state and migrate collectively—a phenomenon
linked to both asthma and idiopathic pulmo-
nary fibrosis (35–37). We occasionally noted
airway epithelial streaming or unjamming after
MCH treatment, which could also contribute
to the forces driving extensive extrusion and
denuding of airways. The reduced cell numbers
resulting from excess extrusions after strong
bronchoconstriction could also trigger a wound
healing response that would not only result in a
poor barrier but also in matrix deposition and
fibroblast activation that could promote more
reactive airways.
Our studies define the extrusion pathway in
controlling the downstream symptoms of an
asthma attack. Admittedly, our studies do not
address whether Piezo1 or TRP channels spe-
cifically control extrusion in response to bron-
choconstriction. Thus, future studies will need
to determine which channels respond to dif-
ferent asthma triggers. A variety of TRP chan-
nels are not only mutated in many cases of
poorly controlled asthma but are activated by
numerous stresses that trigger asthma attacks,
including pollution, smoke, and cold, which
suggests that several channels may be rel-
evant (21, 38–40). Thus, Gd3+ has the advan-
tage of generically blocking extrusion upstream
in the extrusion pathway. Additionally, tran-
sient use of Gd3+ has the added benefit that it
may be used practically once an attack occurs
without apparent side effects in mice, but its
safety will need to be tested in human airways.
Gd3+not only serves to establish a role for
mechanically induced extrusion in driving
asthma symptoms but also suggests that
targeting extrusion upstream in the pathway
may prevent downstream symptoms. If Gd3+
is not safe in humans, development of other
early extrusion inhibitors may provide ther-
apeutic benefit.
SCIENCE science.org
RESE ARCH | R E S E A R C H A R T I C L E S
Because blocking extrusion upstream in the 34. A. Tam, S. Wadsworth, D. Dorscheid, S. F. Man, D. D. Sin, Ther.
pathway with gadolinium preserves the airway Adv. Respir. Dis. 5, 255–273 (2011).
35. J. A. Mitchel et al., Nat. Commun. 11, 5053 (2020).
barrier after a bronchoconstrictive attack, it 36. J.-A. Park et al., Nat. Mater. 14, 1040–1048 (2015).
may also prevent the ASM remodeling asso- 37. I. T. Stancil et al., Nat. Commun. 12, 4566 (2021).
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).
duction (23, 42, 43) that causes future attacks. 39. C. E. Deering-Rice et al., J. Biol. Chem. 291, 24866–24879
Thus, preventing the mechanical damage caused (2016).
by an asthma attack could pave the way for ther- 40. W. C. Wetsel, Int. J. Hyperthermia 27, 388–398
(2011).
apies that stop the whole asthma inflammatory 41. N. K. Malavia et al., Am. J. Respir. Cell Mol. Biol. 41, 297–304
cycle rather than treating only the downstream (2009).
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
striction may underlie other inflammatory syn-
(2023).
dromes, especially those linked with cramping,
such as irritable bowel syndrome or inflamma- AC KNOWLED GME NTS
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
proaches for these unsolved medical problems.
Institute for Lung Health team. We thank S. Mitchell, C. Pardo Pastor,
M. Redd, T. Zulueta-Coarasa, and D. Jackson for helpful
RE FERENCES AND NOTES comments on our manuscript. We are grateful to the vibrant
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
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).
3. F. Menzella et al., Multidiscip. Respir. Med. 16, 787
Fluorescence Microscopy Facility at the Health Sciences Cores at
the University of Utah; and the Huntsman Cancer Institute, Queen
(2021).
Mary University of London, King’s College London, Imperial College
4. S. T. Holgate, J. Allergy Clin. Immunol. 120, 1233–1244
London, University of Southampton Pathology Cores, and the
(2007).
5. S. T. Holgate, Curr. Opin. Pulm. Med. 15, 63–71
Innovation Hub at King’s College London and Confocal Imaging
Core at University College London for tissue section handling and
(2009).
scanning. Funding: This study was supported by an American
6. C. G. Persson, Am. J. Respir. Crit. Care Med. 153, S9–S10
Asthma Foundation Award 16-0020, National Institute of Health
(1996).
R01GM102169, a Howard Hughes Faculty Scholar Award 55108560,
7. C. Xiao et al., J. Allergy Clin. Immunol. 128, 549–556.e12
and a Wellcome Investigator Award 221908/Z/20/Z to J.R. P30
(2011).
CA042014 awarded to the Huntsman Cancer Institute core and an
8. G. T. Eisenhoffer et al., Nature 484, 546–549 (2012).
NCRR Shared Equipment grant no. 1S10RR024761-01 for the
9. J. Rosenblatt, M. C. Raff, L. P. Cramer, Curr. Biol. 11, 1847–1857
microscopy core supported this work. This study and research
(2001).
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
(2013).
Research Centre (BRC) to C.B. The views expressed are those of
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).
12. M. M. Gueders et al., Inflamm. Res. 58, 845–854
Department of Health and Social Care. Author contributions: J.R.
designed experiments, interpreted and analyzed data, and wrote
(2009).
the manuscript. D.C.B., T.R., and K.F. designed experiments,
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).
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
mouse priming, imaging, and quantification of imaging results of
(1996).
many of the experiments. C.R. and C.D.-R. consulted on
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).
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.
provided lung slices from asthma patients. All authors edited the
378–385 (2004).
manuscript. Competing interests: C.B. reports grants and
17. A. Tomkinson et al., J. Immunol. 166, 5792–5800
consultancy fees paid to his institution from Areteia, AstraZeneca,
(2001).
Chiesi, Genentech, GlaxoSmithKline, Mologic, Regeneron
18. K. G. Tournoy, J. C. Kips, C. Schou, R. A. Pauwels, Clin. Exp.
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
19. K. Ulrich et al., Pulm. Pharmacol. Ther. 21, 637–647
competing interests. Data and materials availability: All data are
(2008).
available in the main text or the supplementary materials. License
20. Y. Gu, T. Forostyan, R. Sabbadini, J. Rosenblatt, J. Cell Biol.
193, 667–676 (2011).
information: Copyright © 2024 the authors, some rights reserved;
exclusive licensee American Association for the Advancement of
21. M. S. Grace, M. Baxter, E. Dubuis, M. A. Birrell, M. G. Belvisi,
Science. No claim to original US government works. https://www.
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
(2011).
article is subject to HHMI’s Open Access to Publications policy.
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
(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.
Med. 15, 4–11 (2009).
Manuscript (AAM) of this article can be made freely available
under a CC BY 4.0 license immediately upon publication.
25. C. M. Evans et al., Nat. Commun. 6, 6281 (2015).
26. Z. Jia et al., Mucosal Immunol. 14, 125–134 (2021).
27. S. Piñeiro-Hermida et al., Allergy 72, 1317–1326
(2017).
SUPPLEMENTARY MATERIALS
28. A. Hsieh, N. Assadinia, T.-L. Hackett, Front. Physiol. 14, 1113100 science.org/doi/10.1126/science.adk2758
Materials and Methods
(2023).
29. R. C. Hyzy et al., Respir. Med. 102, 1819–1826 Figs. S1 to S5
References (44–47)
(2008).
30. F. ter Woort, J. L. Caswell, L. G. Arroyo, L. Viel, Am. J. Vet. Res. MDAR Reproducibility Checklist
79, 342–347 (2018). Movies S1 to S13
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
33. S. T. Holgate, G. Roberts, H. S. Arshad, P. H. Howarth, Accepted 12 February 2024
D. E. Davies, Proc. Am. Thorac. Soc. 6, 655–659 (2009). 10.1126/science.adk2758
5 APRIL 2024 • VOL 384 ISSUE 6691 73
RES EARCH | R E S E A R C H A R T I C L E S

TEXTILES uses the interactive object itself—the human

body—to couple the ambient EM energy. This
Single body-coupled fiber enables chipless body-coupled fiber electronics technology
can generate bound charge pairs between
textile electronics the body and electronic fiber, which can alter-
nate between bound and radiation states for
Weifeng Yang1,2, Shaomei Lin1, Wei Gong1,3, Rongzhou Lin2, Chengmei Jiang2, Xin Yang2, Yunhao Hu1, the wireless emission of sensing signals. Fur-
Jingjie Wang1, Xiao Xiao2, Kerui Li1, Yaogang Li1, Qinghong Zhang1*, John S. Ho2, Yuxin Liu2, thermore, we introduce electric field–sensitive
Chengyi Hou1*, Hongzhi Wang1* luminescent dielectrics into fiber antennas
to provide human-readable feedback and
Intelligent textiles provide an ideal platform for merging technology into daily routines. However, current unify sensors and effectors on a single fiber.
textile electronic systems often rely on rigid silicon components, which limits seamless integration, This integrated design strategy frees the textile
energy efficiency, and comfort. Chipless electronic systems still face digital logic challenges owing to from chips, batteries, and any other rigid com-
the lack of dynamic energy-switching carriers. We propose a chipless body-coupled energy interaction ponents, allowing all electronic assemblies to
mechanism for ambient electromagnetic energy harvesting and wireless signal transmission through be incorporated in a miniature fiber (softness:
a single fiber. The fiber itself enables wireless visual–digital interactions without the need for extra chips 0.095 cN·cm2; fineness: 300 mm), making it
or batteries on textiles. Because all of the electronic assemblies are merged in a miniature fiber, this comparable to everyday textiles.
facilitates scalable fabrication and compatibility with modern weaving techniques, thereby enabling
Design and principle of the body-coupled
versatile and intelligent clothing. We propose a strategy that may address the problems of silicon-based
interactive fiber
textile systems.
We provide a schematic that compares the
conventional chip-based interactive textile sys-
uman beings have an inherent need for anism that can fulfill the demands of daily life, tem and our body-coupled chipless interactive

H
interaction, and clothing serves as a
means of communication between the
individual and their environment, mak-
ing it a major element of this interaction
(1, 2). With advances in wearable technology,
digital device–enabled textile electronics have
gradually become a means to expand this type
of interaction (3, 4). Textile electronic systems
are designed to equip textile or fiber assem-
blies with electronic functions for sensing,
computation, display, and communication (5, 6).
These systems typically comprise, but are not
limited to, sensor and effector devices, en-
ergy storage and harvesters, and silicon-
based processors (7, 8). Yet such von Neumann
architecture–based modular electronic sys-
tems for intelligent textiles pose challenges for
seamless integration (9, 10), energy efficiency
(11), and wearing comfort (12, 13). These chal-
lenges arise because current architecture—
including wireless modules, microprocessors,
and analog-to-digital converters (ADCs)—rely
on intrinsically rigid integrated circuit chips
(stiffness: 107 to 109 N·m−1; curvature radius:
10−6 to 10−3 m) (14), which have high power
consumption and are unable to conform seam-
lessly to soft fibers and textiles (stiffness: 103
to 105 N·m−1; curvature radius:10−3 to 10−2 m)
(6, 15). Therefore, a paradigm shift in modular
textile electronics is necessary and requires an
entirely new wireless energy interaction mech-
1
State Key Laboratory for Modification of Chemical Fibers
and Polymer Materials, College of Materials Science and
Engineering, Donghua University, Shanghai 201620,
P. R. China. 2Institute for Health Innovation and Technology,
National University of Singapore, Singapore 117599,
Singapore. 3Biomass Molecular Engineering Center, College
of Light-Textile Engineering and Art, Anhui Agricultural
University, Hefei 230036, P. R. China.
*Corresponding author. Email: wanghz@dhu.edu.cn (H.W.);
hcy@dhu.edu.cn (C.H.); zhangqh@dhu.edu.cn (Q.Z.)
being battery-free (10), chip-free (16), washable
(17), and weaveable (18).
In pursuit of the above vision, we sought to
develop a non–von Neumann architecture that
enables the integration of electronic assem-
blies into a single fiber. The fiber itself can
perform the roles of wireless transfer, sensory
processing, and feedback, making it a build-
ing block for electronic textiles (19). If one or
more interactive functions of chip-based tex-
tile systems are to be realized in a wireless and
chipless form, the main obstacles are as fol-
lows. (i) For energy interaction: In a single
fiber, simultaneously confining and radiating
energy or electrical signals appears to be a
paradox, because no energy interaction mech-
anism exists that can dynamically switch be-
tween oscillating and confining charge pairs,
as previously recognized in fiber electronics
(20, 21). In the topology of textiles, the ad-
jacent electronic fibers are exposed to the
same electromagnetic (EM) environment, and
transforming the charge-pair state by relying
on the induced EM field of the surrounding
electronic fibers is difficult. (ii) For signal mod-
ulation: The current signal modulation principle
relies on complex analog circuits and numer-
ous electronic components, which are difficult
to integrate into commonly used soft fibers
(5, 22). Thermal drawing technology offers so-
lutions for fabricating digital fiber. However,
given the constraints of chip size at the milli-
meter scale, achieving further reductions in
the digital fiber’s diameter by embedding the
chip becomes challenging (3, 4). Furthermore,
digital-based electronic interfaces are not naked-
eye readable (2, 23, 24) nor are they artificially
modulatable, hindering immersive interactive
experiences for user-oriented intelligent textiles.
We propose a body-coupled energy inter-
action mechanism for fiber electronics that
textile system (Fig. 1A). A conventional chip-
based electronic textile (e-textile) system typ-
ically integrates processor chips, circuit elements
(diodes, resistors, and capacitors), sensors, and
a radio frequency antenna onto flexible poly-
imide films (25, 26) (fig. S1). Structural engi-
neering of rigid electronic components and
organic substrate has enabled circuit-level
flexibility, however, current flexible printed
circuits (FPCs) hinder the conformal attach-
ment of e-textiles to the skin (14). Further-
more, the multinode sensing and display
module requires a number of flexible electrode
leads to be attached to the integrated circuit
(IC) surface, causing connectivity issues be-
tween nodes (14, 27). Once the FPC and textiles
are multinode integrated, in most cases it is
almost impossible to disassemble them, which
will make it difficult to carry out subsequent
daily washing (8). In contrast, our proposed
interactive fiber (i-fiber) eliminates the con-
straints of chips and battery assemblies and
can be seamlessly integrated into modern tex-
tile industry, addressing the issues of comfort-
ability, integration, and endurance.
The i-fiber consists of two functional units
(Fig. 1B): the wireless receiver and the wireless
transmitter, the latter of which includes both
sensing and display parts. The i-fiber also con-
sists of three functional layers: the antenna
core (silver-plated nylon fibers; Fig. 1B, red)
for inducing an alternating EM field (Ei), the
dielectric layer (BaTiO3 mixed resin; Fig. 1B,
white) for storing coupled EM energy, and the
optical layer (ZnS:Cu2+ mixed resin; Fig. 1B,
blue) for visualizing the electric field. Mean-
while, the interacting object itself, the human
body, acts as the other pole for confining the
ambient EM field (figs. S2 and S3). When the
i-fiber comes in contact with human skin, an
interface contact capacitance is formed and

74 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE


A B
Sensing
Non-conformal fibers
Power
module
Multi-node integration Wireless
module
Display
fibers
Unwashable
Conventional textile
electronic system
C D
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
visible light (figs. S4 and S5). When electric
field intensity of the interface capacitance
surpasses the critical value of air breakdown
(~3 × 10−3 V·mm−1), it will lead to a localized
plasma discharge effect. This creates an addi-
tional electric displacement term @D, which
breaks the equilibrium state of confining charge
pairs, thereby realizing wireless transmission
of electrical signals for tactile sensing. When
the soft i-fiber is placed in the palm (Fig. 1C),
the interface contact capacitance can effective-
ly capture the ambient EM energy and induce
the i-fiber to emit light (movie S1 and fig. S6).
We show the simultaneous generation of opti-
cal and electrical signals when the finger comes
in contact with the embroidered i-fiber (Fig.
1D). Our e-textile system only contains i-fibers
without chips, batteries, or any other extra
components (movie S2).
SCIENCE science.org
Wireless receiving
i-fibers
1 2 N Radio wave
Body-coupled
Capacitance
Dielectric layer
Fiber antenna
Sensing / display layer
EW
Our body-coupled textile
electronic system
Body-coupled
capacitance
Ambient EM energy harvesting by
body-coupled interactive fiber
EM radiation—from radio signals, the electrical
power grid, and even emerging triboelectric
signals in daily life—is becoming increasingly
ubiquitous (28). Harvesting this radiation for
powering distributed electronics through a
body-coupled (29–32) strategy in a chipless
form has yet to be explored. Here, we present
a body-coupling ambient EM energy harvest-
ing approach enabled by a single i-fiber (Fig. 2,
A and B). The human body has a remarkably
high relative permittivity and conductivity
(e ≈ 78 and d ≈ 0.6 S·m−1, respectively) com-
pared with air (e ≈ 1 and d ≈ 10−14 S·m−1), mak-
ing it an ideal carrier for coupling EM energy
(32). The EM energy dissipated in the envi-
ronment (by smartphones, power cables, fab-
ric friction, etc.) can form a closed energy loop
through fiber electronics, the human body, and
RESE ARCH | R E S E A R C H A R T I C L E S
Wireless transmitting (Sensing / display)
Visible light
Ei
Wireless power
the ground. This means that some of the EM
energy that was lost in the original environment
will be collected through the coupling path,
the human body path, and the return path.
We show the equivalent circuit model in
Fig. 2C and fig. S6. The conductive core of the
i-fiber can be viewed as a series of inductor Lf
and resistor Rf (fig. S7A). The power source
and the fiber itself form a capacitive path rep-
resented by a capacitance Ccoupling through
electrical field coupling. When the human
body is in contact with the luminous material
on the fiber surface, the skin interface con-
tact capacitance CIC is generated, thus guid-
ing the electrical energy flow to the human
body (body path) and the ground (return path).
Compared with typical bipolar capacitive power
transfer (33), our body-coupled fiber electron-
ics system only has one pair of electric field
coupling plates, which helps to simplify the
5 APRIL 2024 • VOL 384 ISSUE 6691 75
RES EARCH | R E S E A R C H A R T I C L E S

A B Wireless power
Body path Coupled path

Return path Parasitic path Human body

Ambient electromagnetic field

Luminescent coating

Textile friction Smartphone Antenna Dielectric coating

C
Transmitter Fiber electronics Rx

Lf Rf
Power cable Ccoupling
C IC
Fiber electronics
…

Tx Human body
CPP CPP

Earth ground Earth ground

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

Log |Exy| (a.u.)

Body coupled

Fiber electronics Body coupled 0

Fig. 2. Ambient EM energy harvesting by body-coupled interactive fiber.
(A) Schematic of i-fiber harvesting energy from various ambient EM fields,
including textile friction (~3 Hz), power cable (~23 kHz), and smartphone
(~13.56 MHz). (B) Illustration of body-coupled fiber electronics. (C) Circuit model
of body-coupled wireless power transfer with i-fiber. The transmitter is the
external EM signal source. Ccoupling, spatial coupling capacitance; CIC, interface
contact capacitance; CPP, parasitic path capacitance. (D) Power spectra of
ambient EM waves coupled by the human body and air inside an office
(environmental parameters: National University of Singapore, 22°C, 60% relative
humidity). Both measurements were performed at a distance of 20 cm from
the transmitter (30 dBm, ~20 kHz). (E) Received voltage and luminous
brightness of i-fiber for body-coupled EM field energy harvesting at different
distances from the external EM source (30 dBm, ~20 kHz) (data are mean ± SD
from three repeated measurements). (F) Effects of ambient media (air and
liquids) on the ability of the i-fiber to capture EM energy and power itself to emit
light (red: data are mean ± SD from three repeated measurements). (G) Electric
field distribution |Exy| emitted by a dipole in the air or underwater, with or without
body coupling to the ground. a.u., arbitrary units.

textile wireless power transfer system. Ac-
cording to the traditional Kirchhoff’s current
law (33, 34), our fiber electronics system seems
like it would not work because there is no
physical current-return path between trans-
mitter and receiver. But in fact, because each
conductor presents a capacitive property to
infinity, this floating fiber electronics system
will form a parasitic coupling capacitance with
the ground as the return path for current,
thereby realizing the normal operation of the
i-fiber (fig. S6).
We thoroughly investigated the mechanism
of EM field confinement by the i-fiber, as well
as the influence of body coupling, receiving
distance, fiber parameters (length, thickness,
position), and environmental parameters (tem-
perature, humidity, and location) (figs. S7 to
S11). The results indicate that body-coupled
EM energy harvesting from a single fiber does
not rely on specific working conditions. There-
fore, it is expected to be useful in various ap-
plications. We measured the power spectrum
of ambient EM waves coupled by the human
body and air inside an office environment
(Fig. 2D). In the range of 0 to 15 MHz, the
EM energy collected under the coupling of the
human body is substantially higher than that
of air coupling. In addition, as the distance
between the fiber and the signal emission

76 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE


source increases, the EM energy coupled to the
human body decreases continuously, which
makes the luminous brightness of the con-
tact interface of the fiber decrease with the
same trend (Fig. 2E). In addition, we also
investigated EM field reception of the i-fiber
in various ambient media. In ultrapure water
(Fig. 2F), the i-fiber exhibited an extraordinary
EM energy reception capability as compared
with other gas or liquid mediums (movie S3).
We speculate that differences in molecular
configuration–induced polarity will make
it exhibit different polarizabilities under the
EM field. For polar molecules, such as water
(1.84 D), the noncentrosymmetric molecular
structure will deflect with the periodic EM
field, which makes the EM energy in the form
of electric displacement polarization contin-
uously transfer among water molecules (Fig.
2F and fig. S12) (see supplementary text sec-
tion “Underwater constructing body-coupled
energy loop”). This results in most electric
field energy being transferred to the surface
of the i-fiber through molecular polarization,
so the i-fiber is capable of emitting light. From
the perspective of capacitive power transfer,
high permittivity of the environmental medi-
um, such as water (7.08 × 10−10 F·m−1), makes
it easy to obtain high coupling capacitance
between the polar plates. The dielectric mol-
ecules with a large dielectric constant will
increase the capacitive coupling coefficient
between the i-fiber and the power source, so
that a smaller amount of EM energy is dis-
sipated in free space.
Our COMSOL simulation results further
prove that an individual i-fiber barely responds
in the EM environment of air (Fig. 2G, i). When
the human body touches the surface of the
i-fiber, part of the EM field is coupled to and
generated within the human body, thereby
causing local light emission (Fig. 2G, ii). Fur-
thermore, when the environmental medium is
water, the surface of the entire fiber will have a
high electric field owing to the coupling with
the human body, resulting in uniform light
emission along the fiber (Fig. 2G, iii).
Modulation and transmission of wireless
optical and electrical signals
Faraday’s law (Eq. 1) and Ampere-Maxwell’s law
can essentially be used to describe the state-
of-the-art wireless signal generation (Eq. 2)
@Β
∇Ε ¼ ð1Þ
@t
@D
∇Η ¼J þ ð2Þ
@t
Here, E and H are the electric field intensity
and magnetic field intensity, respectively. And
B and D are the magnetic flux density and the
electric displacement field, respectively. J is
SCIENCE science.org
the current density. To wirelessly transmit the
instantaneous tactile sensing signal, we needed
to break the balance of confining charge pairs
at the contact interface and introduce a rapid-
ly varied @D or @B.
We noticed that the intensity of the electric
field between the skin and the i-fiber would
increase as the distance decreased (dsf). Ex-
ceeding the critical value of air breakdown
can create a localized plasma discharge, which
causes ions and electrons to clash and migrate
in turn at high speed between air molecules in
the gap (35, 36), resulting in an additional ra-
pidly varied electric displacement field @D/@t.
We depict the equivalent circuit model of the
LC oscillation for EM wave emission in our
i-fiber in Fig. 3A. The high-k dielectrics is re-
sponsible for the radial interface capacitance
Cf, whereas the conductive core provides the
axial fiber inductance Lf and resistance Rf.
Parasitic capacitance CPP with the ground is
formed by the human body and the floated
fiber electronics, respectively. Therefore, when
a local discharge occurs at the contact inter-
face capacitance CIC, the dominant frequency
of the EM wave radiated can be expressed as
1
f ¼ pffiffiffiffiffiffiffiffiffiffiffi
2p Lf ∗Cs
where Cs is the system capacitance, which is
the series value of Cf, CPP, and CIC. Furthermore,
because the air gap in the plasma discharge
will form a conductive path, the interface im-
pedance between the air and the i-fiber will be
substantially reduced, ensuring that the elec-
tric field remains strong on the luminance
material. When the interface electric field be-
tween the skin and i-fiber exceeds the critical
intensity of the luminance material, the doped
carriers are excited to the conduction band
and then recombine with holes to radiate vis-
ible light. We illustrate in Fig. 3B the process
of generating optical and electrical signals
(movie S4). The generation time of the break-
down electrical signal caused by touch is very
short, within about 8 × 10−7 s, which greatly
eliminates the system delay caused by chip
calculation; from the optical signal, in addi-
tion to the visual response directly obtained by
the naked eye, we can further extract infor-
mation such as touch period and touch area
(Fig. 3D and fig. S13).
The fundamental characteristics of EM wave
manipulation are frequency tuning and ampli-
tude tuning. Our chipless modulation strategy
for wireless electrical signal is manipulating
the intrinsic radial capacitance of the i-fiber.
As the thickness of high-k dielectrics increases
from 18 to 135 mm (Fig. 3, C and E), the main
frequency of the wireless electrical signal in-
creases in a positive correlation, and the am-
plitude decreases in a negative correlation.
This observation is consistent with the rule
that a decrease in the radial capacitance causes
RESE ARCH | R E S E A R C H A R T I C L E S
an increase in the resonant frequency. As for
the frequency (wavelength) of the optical sig-
nal, it is usually affected by the excitation sig-
nal. An increase in the EM frequency would
produce a blue-shift of the dominant wave-
length of the spectra, which change from about
520 nm at 3 Hz to 430 nm at 13.56 MHz (Fig.
3F). Doping with Cu2+ in ZnS would establish
two substitutional centers, respectively, a shal-
low excited emission center (green light) and a
deep excited emission center (blue light).
Moreover, the transmission distance and
directionality of wireless signals is crucial
for user-oriented interaction. We measured
the attenuation of the optical and electrical
signals generated by a single fiber over dis-
tance (Fig. 3G). The wireless electrical signal
can be transmitted effectively up to 30 m,
whereas the optical signal can still retain a
visible optical power intensity of 10 nW·cm−2
at a distance of 10 m. This can fully satisfy the
daily interaction distance. We also demonstrated
that the i-fiber can transmit the wireless optical
and electrical signals omnidirectionally, where-
in the signal intensity detected in each direc-
tion is nearly identical (Fig. 3H).
ð3Þ Continuous manufacturing, weaving, and
Kawabata fabric style evaluation
Besides their distinctive wireless energy inter-
action mechanism, fiber electronics should
also possess inherent textile features—batch
weaving and wearing comfort (12), as mea-
sured by factors such as continuity, fineness,
softness, washability, etc. We sought a layer-
by-layer coating fabrication process based
on industry-level equipment and protocols.
High-k dielectrics and ZnS:Cu2+ phosphor was
mixed in the double network resin and coated
on the surface of conductive fiber (Fig. 4A and
fig. S14). This sort of scaling was only possible
because we got rid of the electronic modules,
which allowed the use of fashion-industry fab-
rication techniques. The fibers have excellent
softness (0.095 cN·cm2), fineness (300 mm),
and breakage strength (56.4 MPa) (figs. S15
to S18). These values mean that the fibers are
flexible and strong enough to be sewn and
embroidered with a digital sewing machine
(Fig. 4, B and C, and figs. S19 to S23). Addi-
tionally, we introduced fluorescent dyes (the
second excited source) to manipulate both
the inherent and emitted hues of the fiber
electronics (Fig. 4, D and E, and fig. S24).
Wearing comfort, critical for textile elec-
tronics, has a direct impact on the user’s wear-
ing experience and health. Air or moisture
permeability refers to the performance of air
or H2O molecules through the fabric and is
the most basic property in fabric permeabil-
ity (16). Conventional chip-based e-textile sys-
tems exhibit local airtight (~7.4 mm s−1,
~0.114 g cm−2 h−1) and nonconformal con-
tact owing to the rigid chip-dense polyimide
5 APRIL 2024 • VOL 384 ISSUE 6691 77
RES EARCH | R E S E A R C H A R T I C L E S

A B
… Finger 1
E field
Electrical signal

Equivalent LC dsf
oscillator circuit Fiber
CIC ~MHz
Ionization
2
Cf

CPP Lf Rf Optical signal

CB 3
CPP ~510 nm
~440 nm
t2 state
Earth ground e state
VB

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.
(A) Schematic diagram of the mechanism for wireless transmission of optical
and electrical signals by a single i-fiber. (B) Photograph of wirelessly transmitting
optical and electrical signals. Scale bar, 3.5 mm. (C) Cross-sectional view of the
i-fiber, including core fiber antenna, high-k dielectric coating, and luminance sheath.
Scale bar, 250 mm. (D) The i-fiber–induced optical and electrical signals in time
domains. (E) The influence of the thickness d of fiber dielectric coating on the
frequency and amplitude of electrical signals. (F) An increase in the EM frequency
would produce a blue-shift of the dominant wavelength of the spectra, which change
from about 520 nm at 3 Hz to 430 nm at 13.56 MHz. (G) Attenuation of wireless
optical and electrical signals with distance. (H) Directional diagrams of wireless optical
and electrical signals (data are mean ± SD from three repeated measurements).

circuit system, which will cause discomfort to
the local microenvironment of the skin. In
contrast, our chipless wireless textile exhibits
good breathability (~604.8 mm s−1, ~1.99 g cm−2
h−1), nearly 100 times higher than chip-based
e-textile systems. The textile is compliant any-
where on the skin and meets the requirements
for wearing comfort (Fig. 4, F and G, and fig.
S25). Furthermore, we used the Kawabata
Evaluation System for Fabrics (KES-F) (37) to
quantitatively evaluate fabric style, by testing
fabrics’ flexural, shear, tensile, and compres-
sive stiffness, as well as smoothness and fric-
tion qualities, to identify fabric parameters for
the comfort performance of textiles for various
end uses (Fig. 4H and figs. S26 to S31). We
found that the use of our i-fiber can improve
wearing comfort in terms of fabric bending,
shearing, and surface smoothness properties.
Washing durability is another major issue im-
peding the practical applicability of textile
electronics. We performed standard washing
and post-wash colorfastness tests (ISO 105/
C10:2006) on chipless textile electronics (38).
After 25 standard washes, our textile elec-
tronic maintains its original appearance and
performance (Fig. 4I and fig. S32). The double-
network cross-linked polymer effectively pre-
vents the fading of dyes and the shedding of
phosphors. The textile’s luminous brightness
did not decay substantially, proving good wash-
ing resistance. We also characterized the i-fiber’s
wearable durability performance, including
long-term working stability, moisture resis-
tance, sweat resistance, and abrasion resistance
(figs. S33 to S36). The results show robust optical
and electrical properties of the i-fiber.
Applications of chipless textile electronics
We demonstrate various application scenar-
ios of the i-fiber (i-textile) (Fig. 5, fig. S37,
and movie S1). We demonstrate the potential

78 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

Fig. 4. Continuous manufacturing, weaving, and Kawabata fabric style
evaluation. (A) Continuous fabrication of the i-fiber. (B) Digital embroidery of
the i-fiber with commercial textile. (C) Large-area (0.8 m by 1.5 m) weaving
of the i-fiber on a rapier loom. (D) Multicolor i-fiber collected on reeling rollers.
Scale bar, 6 cm. (E) The dyed fibers show different luminous colors of white, green, and
blue under body-coupled EM fields. Scale bar, 3.5 mm. (F) Schematic diagram of a
conventional chip-based e-textile and our chipless e-textile. (G) Comparison of a
conventional chip-based e-textile and our chipless e-textile in terms of wearing comfort
(air and moisture permeability) (data are mean ± SD from five repeated measure-
ments). (H) Kawabata fabric style evaluation of the e-textile and commercial textile.
(I) Washability of our i-fiber woven textile, including color fastness and luminous
intensity stability (data are mean ± SD from three repeated measurements).

application of this i-textile in providing as-
sisted optical communication for the deaf (Fig.
5A). This chipless display textile could allow
the wearer to convey information in a comfort-
able, continuous, and real-time manner. It con-
sists of a 644–independent pixel (23 rows by
28 columns) touchpad and a corresponding
textile display pad with the same pixel count
(fig. S38). The light-emitting pattern will be
consistent with the trajectory of the finger
sliding in real time. Our body-coupled dynam-
ic textile display in different areas eliminates
the signal modulation and output process of
the silicon-based chip (25). This makes the tex-
tile display system almost latency-free (micro-
second level) and requires no additional elec-
tronic components or energy supply units
(Fig. 5, B and C, and figs. S39 and S40) (see
supplementary text section “Body-coupled chip-
less display textiles”). Further, on the basis of
the body-coupled energy interaction princi-
ple, we designed clothing that integrates wire-
less luminating patterns (displaying the
“Donghua” logo and letters) and chipless dis-
play function (Fig. 5C and fig. S41). We show
that 26 letters and 10 Arabic numerals can be
displayed in a body-coupling manner through
our textile (Fig. 5D), suggesting the potential
application of chip-free wireless interaction in
future textile optical communication (movie S5).
Most virtual reality and augmented reality
devices require wearing rigid, bulky helmets
for interaction. Some fabric-based interactive
suits have been developed, but these all use
intrinsically rigid silicon-based chips (39),
which hinder a comfortable immersive, in-
teractive experience. Using body-coupled fiber
electronics technology, we developed a single
fiber–enabled interactive textile and realized
real-time control of virtual games (Fig. 5E).
We show the signal-flow block diagram of
this electronic system in Fig. 5F. The human

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 79

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

Fig. 5. Applications of chipless textile electronics. (A) Potential application
of i-textile to provide assisted optical communication for the deaf. (B) A
pixelated wireless textile electronics system enables tactile patterned displays
without batteries or chips on fabric. Scale bar, 3 cm. (C) Comparison of frame
rate and delay between our chipless textile display and a traditional chip-based
textile display. (D) The chipless i-textile realizes a touch display of 26 letters and
10 Arabic numerals. (E) Potential application of i-textile in virtual reality and
human–computer interaction. (F) Block diagram of i-fiber–based chipless game
interaction. (G) Realizing chipless, wireless game control by a single fiber,
including four independent digital logic controls. (H) Potential application of
i-textile in a smart home. (I) Block diagram of i-fiber–based chipless smart home.
(J) The surrounding EM field energy is coupled to the sole of the foot, which
can activate fibers to visualize the touch area and wirelessly transmit sensing
signals for controlling home appliance switches.

body can couple wireless EM fields and trig-
ger the breakdown threshold. This sends the
encoded wireless signal at the fiber contact
interface, and the frequency and amplitude
are modulated by the i-fiber itself (radial ca-
pacitance, Cf). The signal can travel a long dis-
tance (~30 m) to the coil antenna terminal and
be decoded and demodulated into the user
interface for interaction (Fig. 5G). We devel-
oped a game demonstration of Tetris control-
led only by the fiber electronics itself (Fig. 5G)
on textile (movie S6). It can realize four kinds
of independent logic control without relying
on the chip, which is important for textile in-
teraction applications (figs. S42 and S43). This
chipless textile interactive system effectively
reduces the system delay caused by the user
transmitter’s signal collection, processing,
analysis, and encoding under the traditional
von Neumann architecture (fig. S44) (see sup-
plementary text section “Chipless interaction
textile for Tetris game”). To demonstrate the
potential of wireless fiber electronics in a smart
home, we wove a large-area wireless haptic
carpet (Fig. 4C and 5H). When a person steps
on a specific location of the wireless carpet, the
surrounding EM field energy is coupled to the
sole of the foot, which can activate fibers to
visualize the touch area and transmit discharge-
induced wireless sensing signals (Fig. 5, H
and I). We demonstrated that the wireless EM
signal can be detected by the receiver and used
to control indoor electronics (Fig. 5J and fig. S45)

80 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

(see supplementary text section “Wireless car-
pets for smart homes”).
Conclusions
We have developed a demodularized fiber elec-
tronics technology that uses the interactive
object itself—the human body—to couple the
ambient EM energy. This energy interaction
mechanism allows all electronic assemblies
to be merged in a miniature fiber and en-
ables wireless sensing, display, and logic inter-
action functions without relying on any chips.
Wireless harvesting of EM energy and wireless
transmission of interactive signals in fiber elec-
tronics are discussed in depth. Our approach
enables continuous, scalable manufacture of
conformable fiber electronics that fulfill the
requirements of smart clothing. Applications
in assisted communications, smart homes,
and virtual reality illustrate the wide-ranging
applicability of this technology in wearable
electronics and smart clothing.
RE FE RENCES AND N OT ES
1. Y. Luo et al., Nat. Electron. 4, 193–201 (2021).
2. H. Bai et al., Science 370, 848–852 (2020).
3. H. Y. Cheung, thesis, Massachusetts Institute of Technology
(2023).
4. G. Loke et al., Nat. Commun. 12, 3317 (2021).
5. A. F. Abouraddy et al., Nat. Mater. 6, 336–347 (2007).
6. C. Chen et al., Chem. Rev. 123, 613–662 (2023).
7. S. Wang et al., Nature 555, 83–88 (2018).
8. J. Shi et al., Adv. Mater. 32, e1901958 (2020).
9. M. Chen et al., Innovation (Camb.) 3, 100340 (2022).
10. R. Lin et al., Nat. Commun. 11, 444 (2020).
11. J. Chen et al., Nat. Energy 1, 16138 (2016).
12. W. Yang et al., Nat. Commun. 10, 5541 (2019).
13. S. Zeng et al., Science 373, 692–696 (2021).
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).
ACKN OWLED GMEN TS used for game interaction with chipless textiles in this study
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
Association for the Advancement of Science. No claim to original
the DHU Distinguished Young Professor Program (LZA2023001); the
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
and Graduate Student Innovation Fund of Donghua University (CUSF-
DH-D-2021005). Author contributions: C.H., Q.Z., Y. Li, and H.W. guided SUPPLEMENTARY MATERIALS
the project. W.Y., C.H., H.W., and W.G. conceived of the idea and
science.org/doi/10.1126/science.adk3755
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.
W.Y., R.L., C.J., X.Y., and X.X. did the circuit modeling and data analysis.
Supplementary Text
Figs. S1 to S45
Y. Liu, K.L., J.S.H., and C.H. revised the manuscript. All authors analyzed
References (41, 42)
the experimental data and prepared the figures and manuscript. All
Movies S1 to S6
authors discussed the results and reviewed the manuscript.
Competing interests: The authors declare that they have no Submitted 18 August 2023; accepted 7 February 2024
competing interests. Data and materials availability: The codes 10.1126/science.adk3755
THERMOELECTRICS
Pseudo-nanostructure and trapped-hole release
induce high thermoelectric performance in PbTe
Baohai Jia1†, Di Wu2†, Lin Xie1†, Wu Wang1, Tian Yu3, Shangyang Li1, Yan Wang1, Yanjun Xu4,
Binbin Jiang5, Zhiquan Chen3, Yuxiang Weng4, Jiaqing He1*
Thermoelectric materials can realize direct and mutual conversion between electricity and heat.
However, developing a strategy to improve high thermoelectric performance is challenging because of
strongly entangled electrical and thermal transport properties. We demonstrate a case in which both
pseudo-nanostructures of vacancy clusters and dynamic charge-carrier regulation of trapped-hole
release have been achieved in p-type lead telluride–based materials, enabling the simultaneous
regulations of phonon and charge carrier transports. We realized a peak zT value up to 2.8 at 850 kelvin
and an average zT value of 1.65 at 300 to 850 kelvin. We also achieved an energy conversion
efficiency of ~15.5% at a temperature difference of 554 kelvin in a segmented module. Our demonstration
shows promise for mid-temperature thermoelectrics across a range of different applications.
eat is an underutilized energy source in kl from lattice vibration and the electronic

17. D. Zhang et al., Adv. Mater. 33, e2100782 (2021).
18. W. Gong et al., Nat. Commun. 10, 868 (2019).
19. Q. Shi et al., Adv. Fiber Mater. 1, 3–31 (2019).
20. A. Leber et al., Nat. Electron. 3, 316–326 (2020).
21. C. Caloz, T. Itoh, Electromagnetic Metamaterials:
Transmission Line Theory and Microwave Applications
(Wiley, 2005).
22. Y. Qu et al., Adv. Mater. 30, e1707251 (2018).
23. C. Wang et al., Nat. Mater. 12, 899–904 (2013).
24. W. Yang et al., Adv. Mater. 33, e2104681 (2021).
25. Y. Lee et al., Sci. Adv. 7, eabg9180 (2021).
26. Y. H. Jung et al., Nat. Electron. 5, 374–385 (2022).
27. H. W. Choi et al., Nat. Commun. 13, 814 (2022).
28. X. Zhang et al., Nature 566, 368–372 (2019).
29. C. Lee et al., IEEE J. Solid-State Circuits 57, 3212–3227
(2022).
30. J. Li et al., IEEE Trans. Biomed. Circuits Syst. 14, 1263–1273
(2020).
31. J. Yoo, IEEE Solid-State Circuits Mag. 15, 49–58 (2023).
32. J. Li, Y. Dong, J. H. Park, J. Yoo, Nat. Electron. 4, 530–538
(2021).
33. F. Lu, H. Zhang, C. Mi, Energies 10, 1752 (2017).
34. L. J. Zou, A. P. Hu, Y. G. Su, IEEE PELS Workshop on Emerging
Technologies: Wireless Power Transfer (WoW), Chongqing,
China, 20 to 22 May 2017 (IEEE, 2017), pp. 93–98.
35. H. Wang et al., Sci. Adv. 7, eabi6751 (2021).
36. Y. Fu, P. Zhang, J. P. Verboncoeur, X. Wang, Plasma Res. Express
2, 013001 (2020).
37. E. S. Namligöz, M. İ. Bahtiyari, A. E. Körlü, S. Çoban, J. Test. Eval.
36, 384–391 (2008).
38. Materials and methods are available as supplementary materials.
39. Z. Zhou et al., Nat. Electron. 3, 571–578 (2020).
40. W. Yang, Labview files and code files for Tetris game
interaction, version 1, Zenodo (2023); https://doi.org/10.
5281/zenodo.10421139.
H
both nature and society, and its accu-
mulation and utilization have always
been popular areas of interest. Thermo-
electric materials that enable direct and
inverse heat-to-electricity conversion have at-
tracted interest for their potential use across a
variety of applications (1–3). The conversion ef-
ficiency of a thermoelectric module is related
to the dimensionless figure of merit zT = sS 2T/k
of materials, where s, S, k, and T are the elec-
trical conductivity, Seebeck coefficient, ther-
mal conductivity, and absolute temperature
in kelvin, respectively. The thermal conduc-
tivity includes the lattice thermal conductivity
1
Shenzhen Key Laboratory of Thermoelectric Materials,
Department of Physics, Southern University of Science and
Technology, Shenzhen 518055, China. 2Key Laboratory for
Macromolecular Science of Shaanxi Province, School of
Materials Science and Engineering, Shaanxi Normal University,
Xi’an 710119, China. 3Hubei Nuclear Solid Physics Key
Laboratory, Department of Physics, Wuhan University, Wuhan
430072, China. 4Laboratory of Soft Matter Physics, Beijing
National Laboratory for Condensed Matter Physics, Institute of
Physics, Chinese Academy of Sciences, Beijing 100190, China.
5
School of Materials and Energy, University of Electronic Science
and Technology of China, Chengdu 610054, China.
*Corresponding author. Email: hejq@sustech.edu.cn
†These authors contributed equally to this work.
thermal conductivity ke from charge-carrier
transport (4–6). Obviously, an excellent thermo-
electric figure of merit requires both a high
power factor (sS 2) and a low thermal conduc-
tivity (7–9). Unfortunately, these parameters are
firmly coupled to each other, making the inde-
pendent tuning of each individual parameter
challenging (10–12). Many effective strategies
to enhance overall thermoelectric performance
have been proposed and confirmed, including
optimizing (static or dynamic) carrier concen-
tration (13, 14), level resonance (15, 16), band
convergence (17), energy filtering (18, 19), band
alignment (20–23), regulating the scattering
mechanism of charge carriers (24, 25), construct-
ing all-scale hierarchical defects (22, 26, 27),
and entropy engineering (28).
Among these strategies, constructing nano-
structures is believed to be able to regulate
the lattice thermal conductivity without severe-
ly affecting other thermoelectric parameters
(29, 30). In general, nanostructure defects re-
fer to an abundance of widely dispersed nano-
scale secondary phases that generate a large
lattice mismatch against the matrix. Thus, they
can result in strong lattice distortion and strain

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 81

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

that substantially strengthen phonon scatter-

ing for a lowered lattice thermal conductivity. A
The size and variety of nanoscale secondary
phases need to be precisely designed; other-
wise, the thermal expansion difference be-
tween the secondary phase and matrix could
cause strain accumulation and even macro-
scopic cracks after heating cycles.
Another strategy is to approach the optimal
carrier concentration nopt(T) in order to ob-
tain maximal zT values at all temperatures
B
(31). The ideal scenario is to find a doping
mechanism that regulates the carrier con-
centration so that it approaches the optimal
carrier concentration nopt(T) as temperature
increases (13). For instance, introducing extra
Cu interstices in lead chalcogenides and tin
tellurides was suggested to achieve this sort of
dynamic doping effect (32–34). However, the
C D
directional migration of interstitial Cu under
temperature or voltage gradients also raised
general concerns about the degradation of
such materials after long-term service.
We used a pseudo-nanostructure, identified
as a cluster of cation vacancies at the nanoscale,
along with dynamic charge-carrier regula-
tion through a “trapped-hole release” process
to demonstrate a case in which both of these
occur simultaneously in p-type lead telluride
(PbTe). This process enabled the simultaneous
reduction of lattice thermal conductivity kl
and regulation of carrier concentration. When
the size of the pseudo-nanostructures in the
PbTe lattice shrinks to the nanometer scale Fig. 1. High performance of PbTe-based materials and modules with pseudo-nanostructure and
with Na and Ge doping (Fig. 1A), the pseudo- trapped-hole release. (A) A schematic diagram of pseudo-nanostructure. The size reduction of the pseudo-
nanostructures impose strong scattering on nanostructure enhances the scattering of phonons but weakens the scattering of holes. (B) Structure-
heat-carrying phonons and allow for the ef- dependent hole trap and release. (C) zT value as a function of temperature and average zT value (inset) for
fective movement of charge-carrying holes be- the p-type PbTe discussed in this work. Some reported zT values for p-type PbTe-based materials are also
cause of the lattice coherence with the PbTe included for comparison. Error bars represent the estimated uncertainty of zT during measurements.
matrix. We achieved the trapped-hole release (D) Conversion efficiencies (h) as a function of temperature difference (DT) for the segmented
mechanism in p-type PbTe through the sub- thermoelectric module in this work and some reported results (28, 38–43) for comparison.
stitution of centered Pb by off-centered Ge
atoms (Fig. 1B). The noncentrosymmetric Ge-Te and Mg has been proven to simultaneously ficients started to decrease at around 550 to
bonding was able to trap charge-carrying holes promote band convergence and enlarge band 650 K. This phenomenon was different from the
at low temperatures and release them when off- gap, both in our experiments (figs. S1 and S2) bipolar effect that is usually observed in narrow-
centered Ge atoms became centered at high and in the literature (44). We then found that band semiconductors (22, 35, 44) because bipolar
temperatures, enabling a dynamic increase of Ge doping (x = 0, 0.25, 0.5, 0.75, or 1%) had diffusion usually results in a much more pro-
hole concentration with temperature. We even- large effects on the thermoelectric properties nounced change in S, s, and kl (44). Ge doping
tually obtained a peak zT value of ~2.8 at 850 K of Na/Mg co-doped Pb0.97Na0.03Te-2%MgTe. also caused a substantial reduction in both the
and an average zT of ~1.65 for 300 to 850 K, both The electrical conductivity was obviously re- total thermal conductivity (Fig. 2C) and the lat-
of which are among the highest values in all re- duced with increasing Ge content (Fig. 2A), tice thermal conductivity (fig. S4). Combined
ported p-type PbTe systems (Fig. 1C) (26, 35–37). whereas the Seebeck coefficient had the op- with a power factor that barely changed, the
Based on the materials we used in this work, posite reaction (Fig. 2B). Hall measurements reduced thermal conductivity resulted in a giant
we obtained an energy conversion efficiency up indicated a surprising decrease in the hole improvement in the peak zT value from 2.4 for
to ~15.5% in a homemade segmented module concentration from 24.2 × 1019 to 5.9 × 1019 cm−3 Pb0.97Na0.03Te-2%MgTe to 2.8 for Pb0.97Na0.03Te-
under a temperature gradient of 554 K (Fig. 1D). along with an increment of Ge doping from 0 to 2%MgTe-0.75%GeTe at 850 K (Fig. 2D), and the
These values are among the highest reported 1 mol % (fig. S3A). The reduction in carrier con- thermoelectric performance showed good re-
and demonstrate the potential for the develop- centration also led to a large increase in mo- peatability and stability (figs. S5 and S6).
ment of mid-temperature applications of ther- bility, which helped the material maintain a high
moelectric technology (table S1) (28, 38–43). conductivity. The temperature dependence of the Trapped-hole release
electrical conductivity and Seebeck coefficient To determine the reason for the large reduction
Thermoelectric properties also exhibited interesting behavior. The electri- of hole concentration in Pb0.97Na0.03Te-2%MgTe-
We started with Na/Mg co-doped PbTe be- cal conductivities of all our Ge-doped samples x%GeTe (where x = 0.25, 0.5, 0.75, or 1), we
cause Na is a classic p-type dopant for PbTe suddenly increased, whereas the Seebeck coef- measured the temperature-dependent carrier

82 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE


Fig. 2. Thermoelectric properties of Pb0.97Na0.03Te-2%MgTe-x%GeTe (x = 0, 0.25, 0.5, 0.75, or 1).
(A to D) Temperature dependence of (A) electrical conductivity, (B) the Seebeck coefficient, (C) total thermal
conductivity, and (D) zT values.
concentration (Fig. 3A). At low tempera-
tures, the hole concentrations of all samples
changed slightly with increasing temperature,
whereas at high temperatures, they increased
substantially. The turning point of the carrier
concentration in Ge-doped samples coincides
with when off-centered Ge becomes centered
at high temperatures. We propose a trapped-
hole release mechanism based on the off-
centered–to–centered transformation of Ge to
explain the hole concentration change with
temperature. Because of the higher bonding
energy of Ge-Te compared with that of Pb-Te
(94 ± 5 kcal mol−1 for Ge-Te and 51.4 ± 2 kcal
mol−1 for Pb-Te) (45), a configuration consist-
ing of an off-centered Ge with six neighbor-
ing Te atoms tends to form when Ge occupies
the Pb lattice. This configuration is the same
as in pristine GeTe. We used density func-
tional theory (DFT) calculations, x-ray pair
distribution function (PDF), and x-ray ab-
sorption fine structure (XAFS) to understand
the off-centered occupation of Ge atoms (fig.
S7 and tables S2 to S5) in the PbTe lattice.
Theoretical and experimental results show that
the bond lengths of Ge and neighboring Te
atoms are different, and a similar off-centered
positioning of Ge has been reported in other
materials (46, 47).
We confirmed the hole localization (trapped
hole) at off-centered Ge atoms using electron
SCIENCE science.org
paramagnetic resonance (EPR) spectroscopy
(Fig. 3B) and mid-infrared transient absorp-
tion spectroscopy (MIR-TAS) (Fig. 3C). When
a hole is localized, an unpaired electron is gen-
erated, and its spin can be recorded by EPR
spectroscopy (48). For the Ge-doped PbTe (PbTe-
0.75%GeTe), the axial signal at g|| = 1.958
and at g⊥ = 1.987 represented the trapped
electron, and the rhombic signals at g1 = 2.02,
g2 = 2.01, and g3 = 2.004 represented the
trapped hole (49, 50). In comparison, we de-
tected only a vacancy signal in pristine PbTe.
An intuitive proof of a trapped hole by off-
centered Ge is given by room temperature
MIR-TAS (fig. S8). Pure PbTe material shows
obvious ground-state bleaching, Pb0.97Na0.03Te-
2%MgTe and Pb0.97Na0.03Te-2%MgTe-0.75%
GeTe samples are in the process of excited
state absorption (ESA), the increase of free-
carrier density promotes positive change of
optical intensity (DOD), and then, with the
relaxation recombination of free carriers, DOD
gradually decreases until it returns to the
ground state. However, the relaxation time
of free carriers in different materials is dif-
ferent because of the existence of trapped
states (51). The lifetimes of the multiexponen-
tial fitting results of Pb0.97Na0.03Te-2%MgTe
and Pb0.97Na0.03Te-2%MgTe-0.75%GeTe are
shown in table S6. The dynamics of ESA sig-
nals detected by both samples showed an ini-
RESE ARCH | R E S E A R C H A R T I C L E S
tial rapid ascent process with a time constant of
tTA-1, followed by a decay process. However,
the decay time of Pb0.97Na0.03Te-2%MgTe-
0.75%GeTe (1.29 ± 0.62 ps) was greater
than that of Pb0.97Na0.03Te-2%MgTe (0.98 ±
0.35 ps). The results show that the trapped
state of Pb0.97Na0.03Te-2%MgTe-0.75%GeTe
delays the recombination process of carriers
and increases the carrier lifetime. In addi-
tion, compared with the MIR-TAS curve of
Pb0.97Na0.03Te-2%MgTe, which quickly re-
turns to zero (ground state), the curve of
Pb0.97Na0.03Te-2%MgTe-0.75%GeTe is less
than zero and then gradually returns to the
ground state. This process is due to the com-
bination of the free carrier and trap state to
form small polarons, which increase the effec-
tive mass of the carrier and decrease the mo-
bility, such that the time for the trapped carrier
to return to the ground state increases (51).
To demonstrate the localization of holes in
the Ge-doped PbTe sample, we constructed a
supercell with 128 atoms (Pb63GeTe64) and re-
moved one electron (by adding a hole) from
the system in the DFT calculations. The dis-
tribution of the unpaired electron (or the hole)
was directly obtained by subtracting the spin-
up states with spin-down states, that is, the
spin density. As per our schematic in fig. S9A,
the calculated spin density confirmed the lo-
calization of holes by Ge atoms. Specifically,
the spin density is distributed mainly between
the Ge atom and its surrounding six Te atoms.
At high temperatures, these trapped holes are
expected to be released back into the PbTe
matrix and thus lead to a sharp increase in
hole concentration. We believe that such a hole-
release phenomenon is closely related to the
bonding environment of Ge-Te (52). Pure GeTe
transforms from the rhombohedral phase to
the cubic phase between 650 and 700 K (38),
and some doped GeTe samples have lower
transition temperatures (38, 53), which is
accompanied by the change of Ge atoms from
an off-centered to centered positioning. We
show evidence for the relation between the
release process of the trapped hole and off-
centered–to–centered positioning of Ge atoms
with temperature-dependent MIR-TAS (Fig.
3C). Up to 573 K, photoexcited carriers in all
samples exhibit similar decaying processes
owing to the presence of trapped states. Be-
cause increasing temperature leads to increased
carrier effective mass, the time of recovery to
the ground state increases gradually. When
the temperature further exceeds 673 K, the
decay process of the photoexcited carrier be-
comes obvious, indicating that the trapped
state disappears.
The process of returning Ge from off-centered
to centered is also supported by in situ low-
loss electron energy-loss spectroscopy (EELS).
The low-loss EELS is directly related to the di-
electric response, or the collective oscillations
5 APRIL 2024 • VOL 384 ISSUE 6691 83
RES EARCH | R E S E A R C H A R T I C L E S

of valence electrons (plasmon) of the system.

For pristine PbTe, the first peak at ~5.7 eV comes
from the s bond valence states (54). Moreover,
we found that this plasmon peak is a finger-
print of the local chemical bonding environment,
especially for Ge dopant. The first plasmon peak
(Fig. 3D) is barely visible at room temperature
in our Pb0.97Na0.03Te-2%MgTe-0.75%GeTe sam-
ple. With an increase of temperature, the first
plasmon peak (~5 eV) starts to emerge and can
be clearly observed at 673 K. We can explain this
result by the change of chemical bonding envi-
ronments by off-center displacements of Ge
atoms. Using DFT calculations, we first com-
pared the theoretical energy-loss function Im
(−1/e), where e is the frequency-dependent
dielectric constant of GeTe with and without
off-center displacement of Ge (fig. S9B). The
first plasmon peak only exists in GeTe without
off-center movements. This is probably due to
the destruction of the valence states with s
bond symmetry in a system with off-center dis-
placements. As a further confirmation, we also
calculated the energy-loss function of a fully
relaxed Pb7GeTe8 supercell, and the theoretical
energy-loss peak also demonstrates a sizable
decrease in the peak intensity owing to the off-
center displacement of Ge (fig. S9C).
Our analysis suggests that the trapped-hole
release mechanism is related to the off-centered Fig. 3. Hole trap and release. (A) Temperature-dependent carrier concentration of Pb0.97Na0.03Te-2%MgTe-
and centered positioning of Ge atoms in the x%GeTe (x = 0, 0.25, 0.5, or 0.75). (B) EPR spectra measured at 250 K for PbTe and PbTe-0.75%GeTe.
lattice, which creates pockets that store holes a.u., arbitrary units. (C) Temperature-dependent MIR-TAS of Pb0.97Na0.03Te-2%MgTe-0.75%GeTe. The
at low temperatures and release them at high inset shows the enlarged spectra at a time below 2 ps. DmOD = DOD/1000. (D) In situ EELS of
temperatures. This trapped-hole release mech- Pb0.97Na0.03Te-2%MgTe-0.75%GeTe. The plasmon peak at ~5 eV, which is a fingerprint of s bond valence
anism achieves self-regulation of carrier con- electrons, is barely visible at room temperature (RT) and starts to emerge at increased temperature.
centration and improves zT across the entire
temperature range, although we cannot en-
tirely exclude the role of Na redissolution in trast of these nanoscale pseudo-structures is due cies or vacancy clusters, a smaller t2 indicates a
the PbTe matrix (T > 650 K) (55). to the absence of Pb atoms, given the Z-contrast smaller vacancy cluster scale in Pb0.97Na0.03Te-
in STEM mode. The pseudo-nanostructures 2%MgTe-0.75%GeTe as compared with pris-
Pseudo-nanostructure are able to scatter phonons like classical nano- tine PbTe. Meanwhile, the much larger relative
Ge doping also induced a large reduction in precipitation, but they exhibit little (if any) lattice strength I2 provided a quantitative evalua-
lattice thermal conductivity without degrad- mismatch and thermal expansion difference tion that the number of cation vacancies in
ing the charge-carrier transport (figs. S3A with the matrix. A PDF analysis using synchro- Pb0.97Na0.03Te-2%MgTe-0.75%GeTe was compre-
and S4). This decoupling effect could be seen tron radiation x-ray, as shown in Fig. 4D, re- hensively greater than that in pristine PbTe.
more intuitively by plotting the ratio of weighted vealed that the peak strength of the PbTe sample The doping level of Na might be important
hole mobility and lattice thermal conductivity doped with Na, Mg, and Ge was generally for the formation of the pseudo-nanostructures,
(mw/kl) versus Ge content (Fig. 4A). This value weaker than that of Ge-doped PbTe samples, given that there was an obvious difference
is proportional to the quality factor (B), and a which indicates that more Pb vacancies were in cluster size and vacancy density between
larger B value usually indicates better proper- generated when additional Na and Mg were the Pb0.98Na0.02Te-2%MgTe-0.75%GeTe and
ties of the material (56). doped (57). The PDF pattern of pristine PbTe is a Pb0.97Na0.03Te-2%MgTe-0.75%GeTe samples. We
To investigate the effect of Ge doping on the simulated result only to standardize the bond compared the STEM-HADDF images for the two
thermal transport of PbTe, we used scanning positions; the peak intensity has no practical samples (Fig. 4, B and F), which depict many
transmission electron microscopy (STEM) to meaning. We used positron annihilation spectra dark regions on the atomic scale. Specifical-
study the microstructure of the materials. Pre- (PAS) to explore vacancy defects; the full spectra ly, in the Pb0.97Na0.03Te-2%MgTe-0.75%GeTe
cipitates were not visible using a low-resolution for our 0.75%Ge-doped PbTe samples are shown sample, these pseudo-nanostructures are even-
STEM image of Pb0.97Na0.03Te-2%MgTe-0.75%GeTe in fig. S11. We obtained different lifetime com- ly distributed in the PbTe matrix with a size be-
and x-ray diffraction spectrum (fig. S10, A ponents and relative intensity of positron an- low 1 nm. By contrast, many clusters larger than
and B). Instead, our high-angle angular dark nihilation through analysis (58). The lifetime 4 nm are evident in the Pb0.98Na0.02Te-2%MgTe-
field–STEM (HAADF-STEM) characterizations component t2 of our Pb0.97Na0.03Te-2%MgTe- 0.75%GeTe and Pb0.98Na0.02Te-2%MgTe-4%GeTe
showed that abundant nanoscale pseudo- 0.75%GeTe sample was smaller than that of samples (Fig. 4F and fig. S10C). Our PAS anal-
structures were coherent with the PbTe matrix pristine PbTe (Fig. 4E), but the relative strength ysis also detected a similar phenomenon in
(Fig. 4B) and were accompanied by a large I2 was much larger. Because t2 usually represents different Na content (Fig. 4E), where t2 de-
number of lattice strains (Fig. 4C). The dark con- the annihilation lifetime of positron in vacan- creased from 337 to 256 ps while I2 increased

84 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

Fig. 4. Characterizations of the pseudo-nanostructures. (A) The ratio of weighted
hole mobility and lattice thermal conductivity versus Ge content of Pb0.97Na0.03Te-
2%MgTe-x%GeTe (x = 0, 0.25, 0.5, 0.75, or 1). (B) HAADF image of Pb0.97Na0.03Te-
2%MgTe-0.75%GeTe. (C) Strain analysis by geometric phase analysis of Pb0.98Na0.02Te-
2%MgTe-4%GeTe (fig. S10C). eyy, normal strains along the yy direction. (D) The room-
temperature PDF of PbTe, PbTe-GeTe, and Pb(Na, Mg)-GeTe. To increase the signal
strength of the Ge elements, the Ge doping amount is 8%. The PDF pattern of pristine
PbTe is a simulated result only to standardize the bond positions. cal., the calculated
PDF pattern of pristine PbTe; G, probability to find another atom; r, real-space
distribution of interatomic distances. (E) Positron lifetime component t2 and intensity
I2 in PbTe, Pb0.98Na0.02Te-2%MgTe-0.75%GeTe, and Pb0.97Na0.03Te-2%MgTe-
0.75%GeTe. (F) HAADF image of Pb0.98Na0.02Te-2%MgTe-0.75%GeTe.

Fig. 5. Thermoelectric performance of the

PbTe-based segmented modules. (A to C) Current
(I) dependencies of (A) the output voltage (V),
(B) output power and heat flow (P), and (C)
conversion efficiencies (h) for size-optimized and
nonoptimized segmented modules. (D) Continuous
measurement of a size-optimized segmented
module at a hot-side temperature of 850 K. The
inset photo is the segmented module before
measurement.

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 85

RES EARCH | R E S E A R C H A R T I C L E S
from 67.2 to 93.5% when Na content varied
from 0.02 to 0.03. The existence of Ge was also
essential for the formation of vacancy clusters.
By comparing the lattice thermal conductivities
of the Pb0.97Na0.03Te-2%MgTe and Pb0.98Na0.02Te-
2%MgTe samples (fig. S4) without Ge doping,
we concluded that Na content has a smaller
influence on phonon transport than Ge does.
The PAS results of Pb0.97Na0.03Te-2%MgTe-
x%GeTe (x = 0, 0.5, or 0.75) show that Ge
doping does not introduce more Pb vacancies
(fig. S12A), and its t2 is unchanged because
the vacancy clusters in the materials have the
same size distribution. However, statistical
results from STEM-HADDF images showed
that Ge doping greatly reduced the size of
Pb vacancy clusters (fig. S12, B and C). There-
fore, it can be concluded that Ge is more like-
ly to regulate vacancy cluster morphology by
a means other than introducing more Pb va-
cancies. We need to address the function of
Ge doping in regulating the vacancy cluster
dimension, which results in enhanced pho-
non scattering and reduced lattice thermal
conductivity (fig. S4). To summarize, excess
Na helps to generate extra cation vacancies in
the PbTe matrix, whereas Ge doping reduces
the dimensions of cation vacancy clusters; to-
gether, these effects have a large impact on
the phonon transport behaviors.
Module performance
Using our p-type PbTe, we fabricated an 8-by-8
single-stage thermoelectric module to demon-
strate its mid-temperature thermoelectric con-
version efficiency. The cross-sectional area ratio
of p-type and n-type thermoelectric legs in
single-stage modules was 1:1 (the cross section
of each thermoelectric leg was 4 mm by 4 mm
with a height of 11 mm; Pb0.987S0.01I0.003Te for the
n-type leg and Pb0.97Na0.03Te-2%MgTe-0.75%
GeTe for the p-type leg). The maximal output
power is 2.75 W for this thermoelectric module,
and its maximal conversion efficiency reached
~10.5% with a temperature difference of 553 K
(the temperature of the hot side was 850 K and
that of the cool side was 283 K) (fig. S13). We
then fabricated another 8-by-8 thermoelectric
module segmented with n- and p-type Bi2Te3-
based materials; the thermoelectric properties
of the Bi2Te3-based materials are shown in
fig. S14. The segmented module without size
optimization (the cross-sectional area ratio was
kept at 1:1; the height ratio of PbTe/Bi2Te3 was
11:3) showed that the open-circuit voltage and
output power were 1.83 V and 3.2 W at DT =
554 K, respectively, and the maximal conver-
sion efficiency reached 13.6% at DT = 554 K
(Fig. 5). By optimizing the length ratio of the
Bi2Te3 and PbTe segments (the optimal heights
of the p-type and n-type Bi2Te3 were 3.6 and
4 mm, respectively) and the cross-sectional
area ratio of the p-type and n-type legs [(4 mm
by 4 mm)/(3.45 mm by 3.45 mm) = 1.35:1] guided
86 5 APRIL 2024 • VOL 384 ISSUE 6691
by the finite element method (fig. S15), we fab-
ricated an ultimate 8-by-8 segmented thermo-
electric module. To compensate for the internal
resistance increasing with decreasing cross-
sectional area, we enlarged the cross section of
the Bi2Te3 legs to 4.25 mm by 4.25 mm. The
reduction in the size of the PbTe leg led to a
reduction in the heat flow generated by heat
conduction, whereas the total module internal
resistance remained constant so that the joule
heat was almost constant (Fig. 5A); therefore,
the total module heat flow was reduced. Ulti-
mately, the overall output power was not atten-
uated, but the heat flow was greatly reduced
(Fig. 5B); in this situation, the conversion ef-
ficiency was improved up to ~15.5% (Fig. 5C).
The conversion efficiency did not show obvious
deterioration during 15 hours of continuous
measurement (Fig. 5D), which is a better sta-
bility performance than previously reported ther-
moelectric modules in the literature (38–41).
Conclusions
We improved the thermoelectric performance
of p-type PbTe-based thermoelectric mate-
rials and modules through a trapped-hole
release mechanism that relies on a pseudo-
nanostructure and various forms of doping.
The pseudo-nanostructures, which consist of
a cation vacancy cluster, effectively blocked
phonon transport but allowed the passage of
holes, thus achieving comprehensive electron-
phonon decoupling. The hole trapping at low
temperatures and release at high tempera-
tures enabled the optimization of the hole
concentration over a broad temperature range.
This advance in thermoelectric materials and
modules could accelerate the practical appli-
cation of mid-temperature thermoelectric
power-generation technology.
RE FERENCES AND NOTES
1. J. He, T. M. Tritt, Science 357, eaak9997 (2017).
2. Y. Xiao, L. D. Zhao, Science 367, 1196–1197 (2020).
3. G. J. Snyder, E. S. Toberer, Nat. Mater. 7, 105–114 (2008).
4. J. Mao et al., Science 365, 495–498 (2019).
5. S. Roychowdhury et al., Science 371, 722–727 (2021).
6. T. Zhu et al., Adv. Mater. 29, 1605884 (2017).
7. L. D. Zhao et al., Nature 508, 373–377 (2014).
8. X. Zhou et al., Mater. Today 21, 974–988 (2018).
9. Q. Yang et al., Science 377, 854–858 (2022).
10. C. Chang et al., Science 360, 778–783 (2018).
11. X. L. Shi, J. Zou, Z. G. Chen, Chem. Rev. 120, 7399–7515 (2020).
12. W. Zhao et al., Nature 549, 247–251 (2017).
13. Y. Pei et al., Adv. Mater. 23, 5674–5678 (2011).
14. Y. Pei et al., Adv. Energy Mater. 4, 1400486 (2014).
15. J. P. Heremans et al., Science 321, 554–557 (2008).
16. B. Paul, P. K. Rawat, P. Banerji, Appl. Phys. Lett. 98, 262101
(2011).
17. Y. Pei et al., Nature 473, 66–69 (2011).
18. J. Martin, L. Wang, L. Chen, G. S. Nolas, Phys. Rev. B 79, 115311
(2009).
19. T. Mori, Small 13, 1702013 (2017).
20. L. D. Zhao, V. P. Dravid, M. G. Kanatzidis, Energy Environ. Sci.
7, 251–268 (2014).
21. K. Biswas et al., Nat. Chem. 3, 160–166 (2011).
22. K. Biswas et al., Nature 489, 414–418 (2012).
23. Y. Qin, B. Qin, D. Wang, C. Chang, L.-D. Zhao, Energy Environ.
Sci. 15, 4527–4541 (2022).
24. J. Shuai et al., Energy Environ. Sci. 10, 799–807 (2017).
25. B. Qin, L. D. Zhao, Science 378, 832–833 (2022).
26. Y. Wu et al., Joule 3, 1276–1288 (2019).
27. S. I. Kim et al., Science 348, 109–114 (2015).
28. B. Jiang et al., Science 371, 830–834 (2021).
29. D. Wu et al., Adv. Mater. 28, 2737–2743 (2016).
30. K. Biswas et al., Energy Environ. Sci. 4, 4675–4684 (2011).
31. A. F. Ioffe, Semiconductor Thermoelements and Thermoelectric
Cooling (Infosearch Ltd., 1957).
32. L. You et al., Energy Environ. Sci. 11, 1848–1858 (2018).
33. Y. Zhang, J. Sun, J. Shuai, X. Tang, G. Tan, Mater. Today Phys.
19, 100405 (2021).
34. Y. Qin et al., Adv. Funct. Mater. 31, 2102185 (2021).
35. G. Tan et al., Nat. Commun. 7, 12167 (2016).
36. Y. C. Zhu et al., Energy Environ. Sci. 15, 3958–3967
(2022).
37. H. J. Wu et al., Nat. Commun. 5, 4515 (2014).
38. B. Jiang et al., Science 377, 208–213 (2022).
39. B. Jiang et al., Energy Environ. Sci. 13, 579–591 (2020).
40. X. Hu et al., Energy Environ. Sci. 9, 517–529 (2016).
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).
43. Q. Zhang et al., Energy Environ. Sci. 10, 956–963 (2017).
44. L. D. Zhao et al., Energy Environ. Sci. 6, 3346–3355 (2013).
45. J. Speight, Lange’s Handbook of Chemistry (McGraw-Hill
Education LLC, ed. 16, 2005).
46. Z. Luo et al., Energy Environ. Sci. 11, 3220–3230 (2018).
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
110, 11628–11634 (2006).
49. N. M. Dimitrijevic, Z. V. Saponjic, B. M. Rabatic,
O. G. Poluektov, T. Rajh, J. Phys. Chem. C Nanomater.
Interfaces 111, 14597–14601 (2007).
50. J. L. Lyons, A. Janotti, C. G. Van de Walle, J. Appl. Phys. 115,
012014 (2014).
51. Y. Huang et al., Nanotechnology 34, 235707 (2023).
52. M. Wuttig, V. L. Deringer, X. Gonze, C. Bichara, J.-Y. Raty, Adv.
Mater. 30, 1803777 (2018).
53. X. Xu, L. Xie, Q. Lou, D. Wu, J. He, Adv. Sci. 5, 1801514 (2018).
54. G. Martinez, M. Schlüter, M. L. Cohen, Phys. Rev. B 11,
660–670 (1975).
55. P. Jood et al., J. Am. Chem. Soc. 142, 15464–15475 (2020).
56. Y. Pei, H. Wang, G. J. Snyder, Adv. Mater. 24, 6125–6135 (2012).
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,
Mater. Lett. 64, 661–663 (2010).
AC KNOWLED GME NTS
We thank H. Meng for performing the in situ MIR-TAS
characterizations. We also thank the SUSTech Core Research
Facilities for providing use of TEM and EPR instruments. D.W.
acknowledges support from the Fundamental Innovation Project in
the School of Materials Science and Engineering (SNNU).
Funding: We acknowledge the National Natural Science Foundation
of China (grant no. 11934007), the Science and Technology
Innovation Committee Foundation of Shenzhen (grant nos.
JCYJ20200109141205978 and ZDSYS20141118160434515), the
Outstanding Talents Training Fund in Shenzhen (202108), and the
Strategic Priority Research Program of Chinese Academy of
Sciences (XDB33000000). Author contributions: J.H. designed
this work. B.J. synthesized the samples, fabricated the modules,
and carried out the thermoelectric property measurements.
L.X. and W.W. performed the TEM characterizations and analysis.
L.X. and Y.W. performed the DFT calculations. S.L. performed the
finite element simulation. T.Y. and Z.C. conducted the positron
annihilation lifetime spectra experiments. Y.W. and Y.X. performed
the MIR-TAS characterizations and analysis. B.J., D.W., L.X., and
J.H. wrote the manuscript. All authors edited the manuscript.
Competing interests: The authors declare no competing interests.
Data and materials availability: All data are available in the
manuscript and supplementary materials. License information:
Copyright © 2024 the authors, some rights reserved; exclusive
licensee American Association for the Advancement of Science. No
claim to original US government works. https://www.science.org/
about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adj8175
Materials and Methods
Figs. S1 to S15
Tables S1 to S7
References (59–75)
Submitted 18 July 2023; resubmitted 21 January 2024
Accepted 4 March 2024
10.1126/science.adj8175
science.org SCIENCE
 RESE ARCH | R E S E A R C H A R T I C L E S

FARMING PRACTICES vantages to scaling up diversified agricultural

systems (19), we need evidence that diversifica-
Joint environmental and social benefits from tion does not privilege certain outcomes (e.g.,
nonagricultural biodiversity) at the expense of
diversified agriculture others (e.g., yields) (20). Here, we draw on data
from 11 countries to take the next step, building
Laura Vang Rasmussen1*†, Ingo Grass2,3†, Zia Mehrabi4,5,6, Olivia M. Smith7,8, Rachel Bezner-Kerr9, on prior syntheses examining how diversifi-
Jennifer Blesh10, Lucas Alejandro Garibaldi11,12, Marney E. Isaac13, Christina M. Kennedy14, cation affects biodiversity, ecosystem services,
Hannah Wittman15,16, Péter Batáry17, Damayanti Buchori18, Rolando Cerda19, Julián Chará20, and yields (12–14, 21, 22) to include other so-
David W. Crowder21, Kevin Darras22, Kathryn DeMaster23, Karina Garcia24, Manuel Gómez25, cial outcomes such as food security and hu-
David Gonthier24, Purnama Hidayat26, Juliana Hipólito27,28,29, Mark Hirons30, Lesli Hoey31, man well-being.
Dana James15,16, Innocensia John32, Andrew D. Jones33, Daniel S. Karp34, Yodit Kebede35,
Interdisciplinary and participatory
Carmen Bezner Kerr36, Susanna Klassen15,16,37, Martyna Kotowska38, Holger Kreft39,
data synthesis
Ramiro Llanque40, Christian Levers16,41,42, Diego J. Lizcano43, Adrian Lu23, Sidney Madsen9,
Rosebelly Nunes Marques44, Pedro Buss Martins44, America Melo43, Hanson Nyantakyi-Frimpong45, We examined environmental and social out-
Elissa M. Olimpi46, Jeb P. Owen21, Heiber Pantevez25, Matin Qaim47, Sarah Redlich48, comes resulting from multiple agricultural
Christoph Scherber49,50, Amber R. Sciligo51, Sieglinde Snapp52, William E. Snyder53, diversification strategies, both separately and
Ingolf Steffan-Dewenter48, Anne Elise Stratton10,54, Joseph M. Taylor53, Teja Tscharntke55, in combination. We focused our assessment
Vivian Valencia56,57, Cassandra Vogel48,58, Claire Kremen59 on five types of diversification strategies: (i)
livestock inclusion and diversification (e.g.,
Agricultural simplification continues to expand at the expense of more diverse forms of agriculture. This managed mammals, fowl, bees, and fish); (ii)
simplification, for example, in the form of intensively managed monocultures, poses a risk to keeping the temporal crop diversification (e.g., crop rota-
world within safe and just Earth system boundaries. Here, we estimated how agricultural diversification tion and cover crops); (iii) soil conservation
simultaneously affects social and environmental outcomes. Drawing from 24 studies in 11 countries and fertility management (e.g., compost appli-
across 2655 farms, we show how five diversification strategies focusing on livestock, crops, soils, cation); (iv) noncrop plantings (e.g., hedgerows);
noncrop plantings, and water conservation benefit social (e.g., human well-being, yields, and food security) and (v) water conservation (e.g., contour farm-
and environmental (e.g., biodiversity, ecosystem services, and reduced environmental externalities) ing). We chose these five diversification strat-
outcomes. We found that applying multiple diversification strategies creates more positive outcomes egies (table S2) inductively to classify the wide
than individual management strategies alone. To realize these benefits, well-designed policies are array of farming practices represented across
needed to incentivize the adoption of multiple diversification strategies in unison. the 24 datasets (table S5) included in our
analysis. Soil conservation practices such as
compost addition are considered to represent
he simplification of farming systems con- the negative side effects directly through the ac- a diversification strategy if they create habitat

T tinues to grow at the expense of more

diversified agriculture, contributing to
the crossing of planetary boundaries due
to the excessive use of chemical inputs,
as well as increased greenhouse gas emissions,
biodiversity loss, and water use (1). To address
these challenges, a new paradigm for farming
systems is needed that focuses on providing
food security and nutrition while minimizing
negative environmental, health, and social im-
pacts (2). This transformation is particularly
pertinent as countries in different stages of
economic development navigate through dis-
tinct challenges. Economically advantaged na-
tions need to reverse simplification to recover
from environmental and social damage al-
ready done, whereas lower-income nations need
to minimize these externalities in their devel-
opment transitions (3). Historically, the archi-
tects of the Green Revolution were primarily
concerned with breeding crops and develop-
ing agronomic inputs to increase staple crop
yields and respond to food security needs. How-
ever, the focus of their policies on simplifying
agricultural systems came with unintended
large and negative environmental impacts such
as pollution, as well as social side effects such
as farmer indebtedness, reduction of peoples’
dietary diversity, and reduced resilience (4, 5).
This has led to widespread calls for a change in
agricultural development policy that addresses
tion of biologically diversified farming systems.
Biologically diversified farming systems, mean-
ing those that intentionally increase the num-
ber of agricultural and nonagricultural crop
and livestock species (and their genetic diver-
sity), are a promising solution to bring about
more sustainable food production because in
theory they offer an ecological mechanism for
higher resource use efficiency, less pollution,
improved food sovereignty, and reduced vul-
nerability to climate change (6–8). Much re-
search has examined the empirical effects of
agricultural diversification on environmental
outcomes (9–12). This research includes recent
quantitative syntheses demonstrating the pos-
itive effects of agricultural diversification on
biodiversity and ecosystem services and show-
ing that diversification practices increase yields
as often or more often than they decrease them
(11–13). However, although evidence of the envi-
ronmental benefits of agricultural diversifica-
tion is accumulating, our knowledge of social
outcomes beyond yields is limited to studies
focusing on selected dimensions of social sus-
tainability such as income (14) or employment,
but with limited attention to other facets such
as social networks (15, 16). Therefore, broad
trade-offs between social and environmental
outcomes from agricultural diversification
are poorly understood and largely unquantified
(9, 17, 18). To determine whether there are ad-
conditions that enhance biodiversity or trophic
complexity belowground (23). Likewise, water
conservation practices are considered to rep-
resent diversification if they affect plant and
microtopographic heterogeneity that influ-
ences biodiversity (24). We investigated how
these five strategies can lead to trade-offs and/or
synergies between targeted environmental
(e.g., nonagricultural biodiversity, regulating
ecosystem services such as crop pollination,
and reduced environmental externalities or
harms) and social (e.g., yields, human well-being,
and food security) outcomes (tables S1 to S7
and figs. S7 to S10). Our six environmental and
social outcomes each measure specific dimen-
sions of sustainability to account for intercon-
nections and assess the overall sustainability of
diversified farming from a systems perspective.
We harmonized 24 datasets covering 2655
farms and various farming types across five
continents, including smallholder farming in
rural Africa, plantation crops in Southeast Asia,
and both small- and large-scale farming in North
America, Europe, and Latin America (Fig. 1). The
harmonized dataset combines individual studies
to cover a broad range of farming practices,
geographies, and environmental and social con-
texts to develop a synthesis broadly applicable
across multiple farming systems. All 24 data-
sets measured at least one agricultural diversi-
fication practice, as well as one environmental

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 87

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

Fig. 1. Geographic distribution of

24 datasets spanning five continents
and a wide range of landscapes.
Our analysis of agricultural diversification
strategies covers 2655 farms across
five continents. Insets depict satellite
images of agricultural systems (left
to right: leafy greens in the US
West Coast, mixed farming in Brazil,
smallholder maize in Malawi, and oil
palm plantations in Indonesia). Colored
dots indicate farm locations within
each study. The commonly used 2-ha
threshold for differentiating smallholder
farming from farming that is less
reliant on subsistence was applied (45).

and one social outcome, and thus are inher-
ently interdisciplinary (mostly collected by ecol-
ogists and social scientists working together and
merging their “ways of knowing” and meth-
ods). Also, each dataset studied farm sites with
varying levels of diversification, including farms
without any diversification practices. Unlike
data synthesis approaches that extract values
from published materials, our data synthesis is
based on a participatory, iterative process, in-
cluding multiple group meetings and exchanges
with data contributors during all stages of var-
iable selection, data analysis, and result inter-
pretation. Although our sample size of 24 studies
would be relatively small for a standard meta-
analysis, it is ideal for an approach requiring
extensive interaction with data contributors
working across disparate geographies and
farming systems to confirm and contextualize
results. Fig. S5 provides an example illustra-
tion that we used to confirm and contextualize
results with data contributors who worked
with Malawian smallholders.
We first tested how agricultural diversifica-
tion strategies affected each of the six tar-
geted social and environmental outcomes. For
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.

88 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

We also examined the effect of landscape

composition on diversification outcomes (figs.
S2 to S4). To this end, we refitted the same
* * * * * model set described above but allowed the ef-
fects of diversification strategies and practices
to differ depending on landscape composition
(interaction = diversification × composition).
* * We then compared the effects of diversifica-
tion for farms situated in cleared, simple, and
complex landscapes using the amount of semi-
* natural habitat in a 3000-m radius as a mea-
sure of landscape complexity. We considered
cleared to mean that <5% of natural habitat
Environmental outcomes remains, simple that 5 to 20% remains, and
complex that >20% remains (25) (tables S8 to
Biodiversity
* * S10). Farmers’ adoption of diversification strat-
Ecosystem services
egies occurred in our study through choices
Reduced externalities rather than experimental manipulation (fig.
S6), so this is a synthesis of observed associa-
* Social outcomes tions on real-world farms, not experimental
Human well-being field trials.
Food security More diversification strategies or practices
Yield are better
We found that applying a higher number of
diversification strategies or practices had a
greater likelihood of beneficial outcomes than
using individual strategies or practices. Specif-
* * * ically, we showed that combining five diversifi-
cation strategies or practices had overwhelmingly
strong benefits across outcomes, with positive
effects especially on nonagricultural biodiver-
* * sity and food security (Fig. 2, F and G). Farmers
who integrated multiple strategies or practices
more likely experienced benefits for nonagri-
cultural biodiversity (effect sizes of 0.19 ± 0.05
and 0.26 ± 0.05 for strategies and practices,
respectively), moderate increases in human
Fig. 2. Effects of agricultural diversification on environmental and social outcomes. (A to G) Agricultural well-being (effect size of 0.07 ± 0.03) for num-
diversification strategies include livestock diversification, temporal crop diversification, soil conservation, ber of strategies, and comparatively stronger
noncrop diversification, and water conservation. Flower diagrams indicate the effects of diversification increases in food security (0.24 ± 0.03 and 0.35 ±
strategies on three environmental outcome variables (nonagricultural biodiversity, regulating ecosystem services, 0.04 for strategies and practices, respectively).
and reduced environmental externalities) and three social outcome variables (human well-being, food security, Positive effects of diversification strategies
and yield). Also shown are the effects of the total number of diversification strategies (up to a total of five) and and practices on biodiversity were driven by
their associated diversification practices (up to a total of 23, excluding livestock diversification) applied effects on large, but not small, farms (tables
(table S2). Effect sizes are measured in units of SD, with the black circle indicating an effect size of 0.0. The S11 and S12). This may be due to the avail-
size of the flower petals is proportional to the effect size; error bars indicate ± 1 SE. Asterisks indicate ability of stronger contrasts in management
statistically significant effects of diversification strategies on outcomes (gray asterisks, P < 0.05; black asterisks, intensity between large-scale simplified farms
P < 0.00119 using Bonferroni correction for multiple comparisons; 42 estimates). and similar-sized farms that have adopted
diversification strategies (26). We also found
that increases in food security were driven by
identified the number of distinct practices manure and biochar, would score 2 for strat- small farms, which might suggest that new
adopted by farmers at the field or farm level egies and 4 for practices (fig. S1 shows exam- market opportunities could be more benefi-
that were recorded by each study. For exam- ples of coverage of agricultural diversification cial than diversification for food security on
ple, noncrop diversification included practices strategies within farms). We then modeled out- large farms (27).
such as windbreaks, flower strips, and hedge- come variables as a function of (i) the degree All results for nonagricultural biodiversity
rows. In addition, we recorded the total number of diversification of each strategy (i.e., the and regulating ecosystem services should be
of strategies (maximum of five) and the total number of practices used within each strat- taken with caution because these outcomes
number of practices recorded by each study egy), (ii) the total number of diversification were measured by only 11 and 12 studies, re-
(calculated by summing all practices applied strategies used, and (iii) the total number of spectively (table S9). Also, a potential criti-
within each strategy). For example, a farm with diversification practices applied across all cism of the type of analysis that we report
noncrop diversification in the form of hedge- strategies (linear mixed-effects models with here is that hierarchical models place more
rows and flower strips, in combination with study IDs as a random effect; see the materials weight on studies with more observations.
soil conservation through application of green and methods and tables S12 and S13). Therefore, we also created a weighting scheme

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 89

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

Fig. 3. Synergies and trade-offs among environmental and social outcomes
of agricultural diversification. (A and B) Methodological approach to identify
synergies and trade-offs based on the example of nonagricultural biodiversity and
crop yield. (C) Synergies and trade-offs for all pairwise combinations of studied
environmental and social outcome variables. (A) shows the effects of agricultural
diversification strategies on nonagricultural biodiversity and crop yield. Arrow tips are
located at the predicted change in nonagricultural biodiversity and crop yield,
respectively, with an increase of 1 SD of a given diversification strategy. The x-y
coordinates of the arrow tips are numerically similar to the effect sizes shown in
Fig. 2. The resulting arrow directions indicate trade-offs and synergies of diversification
strategies, with arrows pointing in the upper right corner indicating win-win
outcomes, and arrows pointing in the other quadrants indicating either trade-offs
(top left quadrant, lose-win; bottom right quadrant, win-lose) or lose-lose (bottom
left quadrant) outcomes. (B) shows outcomes by diversification strategy. Circle
size is proportional to arrow length in (A) and indicates the effect strength, that is,
the conjoined change in nonagricultural biodiversity and crop yield with diversification.
In (C), colored circles indicate outcome combinations of environmental and
social outcome variables (black, win-win; orange, win-lose; blue, lose-win; green,
lose-lose). Circle size is proportional to the joint change of the paired environmental
and social variables.

in which we kept the total number of observa-
tions constant but artificially up- or down-
weighted studies so that all were equally
represented in the model fitting. In the equal-
ized model, we observed some difference in
results for specific outcomes (e.g., ecosystem
services) and diversification strategies (e.g.,
livestock diversification); however, equalizing
the influence of each study did not alter the
key finding that multiple diversification strat-
egies and practices maximized benefits across
environmental and social outcomes (fig. S9).
We interpreted the results from both model
sets as suggesting that different outcomes,
trade-offs, and synergies may be more present

90 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

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
strategy to maximize environmental and so-
cial outcomes is to apply multiple diversifica-
tion strategies and practices in tandem.
Maximizing win-wins, minimizing trade-offs
We further examined the extent to which the
five diversification strategies might promote
pairs of environmental and social win-win out-
comes and also which are particularly prom-
ising for achieving paired benefits. Overall,
applying a high number of diversification strat-
egies or practices was associated with more
potential for win-win outcomes (Fig. 3, B and
C). Livestock diversification and soil conser-
vation were the two strategies that appeared
to consistently elicit multiple positive outcomes
(Fig. 2, A and C), especially win-win outcomes
of nonagricultural biodiversity and multiple
social outcomes (Fig. 3C). Half of the farms in
our study practiced some form of livestock
integration (14 studies surveyed farms with
livestock integration). Livestock diversifica-
tion had the largest positive effect on food
security (0.23 ± 0.03), which is more than four
times as large as the effect on yields. Livestock
diversification promoted nonagricultural bio-
diversity, but with a lower effect size than for
food security (0.16 ± 0.04) (Fig. 2A). Soil con-
servation practices also promoted synergies
through gains in all three environmental out-
comes, accompanied by enhanced human well-
being and food security (Fig. 3C). Overall, three
of the five assessed agricultural diversification
strategies offered potential win-win outcomes
regarding food security and nonagricultural
biodiversity. Conversely, many national strat-
egies and programs focusing on agricultural
intensification did not achieve win-wins (28).
We also observed trade-offs, such as soil con-
servation practices leading to gains in biodi-
versity but potential yield losses, although the
latter were not statistically significant (Figs. 2C
and 3, A to C). For noncrop diversification, we
did not observe a consistent positive effect on
biodiversity (Fig. 2D), potentially driven by di-
vergent responses of taxa. However, positive
effects on food security can readily arise from
such practices. For example, having trees on-
farm can support peoples’ diets by providing
edible products, including fruits, nuts, and leaves
(29–33). Five of our studies (corresponding to
810 farms) used dietary diversity scores (count-
ing the number of food groups consumed) to

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 91

RES EARCH | R E S E A R C H A R T I C L E S
proxy dietary quality, a key component of food
security and a metric likely to capture effects
of having trees on-farm (34). The positive ef-
fect of noncrop diversification on food security
was pronounced (0.21 ± 0.08), almost twice
the size as the negative effect of noncrop di-
versification on human well-being (Fig. 2D).
Explanations for this negative effect include
longer crop rotations or the implementation
of practices such as hedgerows, which could lead
to a smaller area planted with cash crops, po-
tentially leading to lower production and/or
higher labor demands. Although we did not
detect significant effects of single diversifica-
tion strategies on ecosystem services, several
large meta-analyses have shown strong pos-
itive effects of diversification practices on many
ecosystem services (12, 13).
Role of landscape composition in
diversification outcomes
The “intermediate landscape complexity hy-
pothesis” states that agricultural diversifica-
tion is unlikely to result in improvements in
cleared landscapes because the regional spe-
cies pool available to colonize crop fields and
provide ecosystem services is limited (25). Sim-
ilarly, in complex landscapes, diversification may
not lead to measurable increases in nonagri-
cultural biodiversity and/or ecosystem services
on farms because sufficient alternative resources
and habitats are already present in the farm
surroundings to support these species and
services. Instead, the predicted environmental
benefits from agricultural diversification strat-
egies are largest in simple landscapes con-
taining 5 to 20% of seminatural habitats or
noncrop areas (25). We tested this hypothesis
and found that landscape composition (the
proportion of seminatural habitats in a land-
scape) moderated the outcomes of agricultural
diversification strategies (see Fig. 4 and table
S10 for sample sizes). We observed that the
number of diversification strategies applied
had strong positive effects on food security
(Fig. 4E) in cleared landscapes, indicating ben-
efits even in landscapes lacking natural hab-
itat. However, we also observed positive effects
in complex landscapes. For example, livestock
diversification showed the strongest positive
effects on human well-being in cleared land-
scapes (Fig. 4D). However, although we found
positive social outcomes in cleared landscapes,
diversification practices there did not result
in positive environmental outcomes, which is
partially consistent with the hypothesis. Finally,
we observed that the number of diversification
strategies and practices applied had positive
effects on biodiversity in both simple and com-
plex landscapes (Fig. 4A). These results par-
tially agree with the intermediate landscape
complexity hypothesis, but indicate that diver-
sification strategies on farms can be beneficial
for biodiversity even in complex landscapes.
92 5 APRIL 2024 • VOL 384 ISSUE 6691
Outlook
At a time when the outlook for simultaneous-
ly improving and protecting the environment
and social conditions for farmers often seems
bleak (2), our findings present a promising
avenue for shaping global agricultural policy
by showing how applying a suite of diversifi-
cation strategies or practices can create win-
win scenarios. Our results support the notion
that a diversified farming system is often more
beneficial than specific diversified farming
strategies or practices in isolation (13, 14). This
finding emphasizes the need for more explicit
evidence about which combinations of diver-
sification strategies and practices are most com-
plementary in different social and ecological
contexts. Most of our present findings, which
are based on working farm data, support the
growing body of literature linking agricultural
diversification strategies with better outcomes
for nonagricultural biodiversity and regulat-
ing ecosystem services without compromising
yields (12, 13).
Our study advances existing knowledge about
how diversification affects agricultural system
sustainability by (i) considering how a multi-
tude of diversification strategies, not just crop
diversification, may affect sustainability out-
comes (including both individual and com-
bined effects of diversification strategies); (ii)
examining how diversification influences mul-
tiple social and environmental outcomes while
highlighting trade-offs and/or synergies within
and between environmental and social out-
comes; and (iii) examining how landscape com-
position moderates the effects of diversification
on environmental and social outcomes. Thus,
we have moved beyond existing studies that
typically assess the effects of diversification
strategies in isolation and on selected output
variables (35–37), preventing the systemic un-
derstanding needed for informing policy de-
bates on how to produce food while maintaining
a safe operating space for humanity. By focus-
ing on agricultural working landscapes, our
work complements earlier studies examining
environmental and social trade-offs and/or syn-
ergies of protected areas (38) and considers
working land conservation approaches affect-
ing a broader area (39). Because we include
diverse datasets representing multiple world
regions, our flexible approach can be replicated
and expanded to incorporate additional data-
sets in the future.
How can and should policy-makers and prac-
titioners encourage the adoption of specific
types of diversified farming systems? Although
we recognize the benefits of diversification,
it is also critical to acknowledge that many
farmers are working “against the odds” (8).
Structural factors are often the main barrier
for adopting diversification practices, and in-
clude high land rents, the predominance of
short-term leases, stringent food safety regu-
lations, trade agreements exacerbating cor-
porate concentration in global food systems,
and other supply chain pressures (40). Tran-
sitions to diversified farming systems often
require financial support because of potential
initial yield declines or implementation costs
(8). Indeed, current policies often lock in sim-
plified, conventional farming rather than en-
abling durable transitions to diversified farming,
and investments are needed to develop appro-
priate seeds, crop mixes and rotations, and equip-
ment to promote the profitability of diversified
farms.
Effective policies for encouraging the adop-
tion of diversification strategies and practices
likely vary with cropping system and region
and include incentives, regulations, and com-
bined approaches. The use of incentives can be
seen in the European Union, where farmers are
financially compensated on a per-area basis
(41, 42) for some diversification practices such
as noncrop plantings. Also, a recent synthesis
from Ghana shows that incentivizing noncrop
diversification has cascading positive effects
on adoption patterns (43). Regulatory mech-
anisms can be used for soil or water conserva-
tion through policies requiring farmers to use
diversification practices to reduce pollutants
on their farms (e.g., for water quality) (44).
Finally, the benefits of combining incentives
with regulatory mechanisms can be seen in
California, where increasing adoption of di-
versification practices on larger farms may
require supplementing the “pull” of incentives
with the “push” of regulatory mandates (44).
Our study suggests several contexts in which
desirable local outcomes occur most frequent-
ly, with a key example being the positive effect
on human well-being and food security from
applying a high number of diversification strat-
egies in cleared landscapes and on small farms
(table S12). However, researchers have much
more to discover about the variability of out-
comes that can occur across different agricul-
tural diversification strategies, landscapes, and
social contexts, and future work should use
quasiexperimental methods that control for
possible underlying differences between farm-
ers that choose more versus fewer diversifica-
tion strategies.
The future of agriculture faces great chal-
lenges: large increases in demand for agricul-
tural commodities must be met while at the
same time minimizing agriculture’s negative
environmental, health, and social impacts (2).
Our interdisciplinary analysis spanning a wide
array of regions provides convincing evidence
that agricultural diversification is a promising
win-win strategy for providing social and en-
vironmental benefits.
REFERENCES AND NOTES
1. J. Rockström, O. Edenhofer, J. Gaertner, F. DeClerck, Nat. Food
1, 3–5 (2020).
2. B. Carducci et al., Nat. Food 2, 68–70 (2021).
science.org SCIENCE
 RESE ARCH | R E S E A R C H A R T I C L E S

3. Z. Mehrabi, Nat. Sustain. 6, 949 (2023).
4. P. L. Pingali, Proc. Natl. Acad. Sci. U.S.A. 109, 12302–12308
(2012).
5. N. Ramankutty et al., Annu. Rev. Plant Biol. 69, 789–815 (2018).
6. S. K. Jones, A. C. Sánchez, S. D. Juventia, N. Estrada-Carmona,
Sci. Data 8, 212 (2021).
7. C. Kremen, A. Iles, C. Bacon, Ecol. Soc. 17, art44 (2012).
8. J. Blesh et al., One Earth 6, 479–491 (2023).
9. J. Rosa-Schleich, J. Loos, O. Mußhoff, T. Tscharntke, Ecol. Econ.
160, 251–263 (2019).
10. C. Kremen, A. Miles, Ecol. Soc. 17, art40 (2012).
11. C. Kremen, Emerg. Top. Life Sci. 4, 229–240 (2020).
12. D. Beillouin, T. Ben-Ari, E. Malézieux, V. Seufert, D. Makowski,
Glob. Chang. Biol. 27, 4697–4710 (2021).
13. G. Tamburini et al., Sci. Adv. 6, eaba1715 (2020).
14. A. C. Sánchez, H. N. Kamau, F. Grazioli, S. K. Jones, Ecol. Econ.
201, 107595 (2022).
15. J. Hickel, S. Hallegatte, Dev. Policy Rev. 40, e12584 (2022).
16. J. D. van der Ploeg et al., J. Rural Stud. 71, 46–61 (2019).
17. A. E. Stratton, H. Wittman, J. Blesh, Agron. Sustain. Dev. 41, 35
(2021).
18. B. Maas et al., Trends Ecol. Evol. 35, 1049–1052 (2020).
19. Z. Mehrabi et al., One Earth 5, 756–766 (2022).
20. S. Gong et al., Ecol. Lett. 25, 1699–1710 (2022).
21. X. He et al., Nat. Food 4, 788–796 (2023).
22. A. A. Rakotomalala, A. M. Ficiciyan, T. Tscharntke, Agric.
Ecosyst. Environ. 356, 108617 (2023).
23. S. F. Bender, C. Wagg, M. G. A. van der Heijden, Trends Ecol.
Evol. 31, 440–452 (2016).
24. K. Moser, C. Ahn, G. Noe, Wetlands 27, 1081–1097 (2007).
25. T. Tscharntke et al., Biol. Rev. Camb. Philos. Soc. 87, 661–685
(2012).
26. R. Marja et al., Ecol. Lett. 22, 1493–1500 (2019).
27. F. Tacconi, K. Waha, J. J. Ojeda, P. Leith, Agron. Sustain. Dev.
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,
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
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.
Sustain. Food Syst. 45, 197–224 (2021).
36. A. C. Sánchez, S. K. Jones, A. Purvis, N. Estrada-Carmona,
A. De Palma, Agric. Ecosyst. Environ. 332, 107933 (2022).
37. L. C. Ponisio et al., Proc. Biol. Sci. 282, 20141396 (2015).
38. P. J. Ferraro, M. M. Hanauer, K. R. Sims, Proc. Natl. Acad.
Sci. U.S.A. 108, 13913–13918 (2011).
39. L. A. Garibaldi et al., Conserv. Lett. 14, e12773 (2021).
40. L. Carlisle et al., Agroecol. Sustain. Food Syst. 46, 1145 (2022).
41. P. Batáry, L. V. Dicks, D. Kleijn, W. J. Sutherland, Conserv. Biol.
29, 1006–1016 (2015).
42. F. A. Boetzl et al., Proc. Natl. Acad. Sci. U.S.A. 118, e2016038118
(2021).
43. M. Isaac, H. Nyantakyi-Frimpong, P. Matouš, E. Dawoe,
L. Anglaaere, Ecol. Soc. 26, art12 (2021).
44. K. E. Esquivel et al., Front. Sustain. Food Syst. 5, 734088 (2021).
45. V. Ricciardi, N. Ramankutty, Z. Mehrabi, L. Jarvis,
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);
by the European Union’s Horizon 2020 research and innovation BB/J014427/1) and a Royal Geographic Society postgraduate
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.
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.,
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.,
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.,
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.,
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.,
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.,
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
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
Belmont Forum and BiodivERsA joint call for research proposals, data analyses with assistance from Z.M., C.S., and C.K. L.V.R.
under the BiodivScen ERA-NetCOFUND program, and by the and I.G. wrote the first manuscript draft with contributions by all
Natural Sciences and Engineering Research Council of Canada authors. O.M.S. designed Fig. 1. Competing interests: The authors
(NSERC grant 523660-2018), National Science Foundation (NSF declare no competing interests. Data and materials availability:
grant 1852587), German Federal Ministry of Education and The dataset that we compiled based on 24 case studies and
Research (BMBF grant 01LC11804A), and the Research Council of comprising 2655 farms is available in a Dryad repository (46). The
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:
USDA-NIFA-AFRI grant 12679452- 2019-67012-29720. D.J.L., A.M., Copyright © 2024 the authors, some rights reserved; exclusive
and H.P. were funded by The World Bank project Colombia licensee American Association for the Advancement of Science. No
Mainstreaming Sustainable Cattle Ranching (CMSCR-P104687), claim to original US government works. https://www.science.org/
FEDEGAN, CIPAV, TNC, and Action Fund, with financial support about/science-licenses-journal-article-reuse
from the Department of Business, Energy, and Industrial Strategy
of the United Kingdom (BEIS) and the Global Environment Facility SUPPLEMENTARY MATERIALS
(GEF). Data collection in Indonesia was funded by DFG project
science.org/doi/10.1126/science.adj1914
192626868 in the framework of the collaborative German–
Materials and Methods
Indonesian research center CRC990. C.S. was funded by the
Supplementary Text
European Union’s Horizon 2020 research and innovation program
Figs. S1 to S10
under agreement 727284 and was partially funded by the DFG
Tables S1 to S13
under Germany’s Excellence Strategy (EXC 2070–390732324).
References (47–91)
Z.M. and M.H. and the ECOLIMITS project were funded through the
MDAR Reproducibility Checklist
UK NERC-DFID-ESRC Ecosystem Services for Poverty Alleviation
(ESPA) program (grant NE/K010379-1) and NERC (grants Submitted 15 June 2023; resubmitted 1 September 2023
NE/P001092/1 and NE/P00394X/1). Z.M. was also funded by the Accepted 28 February 2024
Biotechnology and Biological Sciences Research Council (grant 10.1126/science.adj1914
ANTIMICROBIALS
Antibacterial activity of nonantibiotics is orthogonal
to standard antibiotics
Mariana Noto Guillen, Carmen Li, Brittany Rosener, Amir Mitchell*
Numerous nonantibiotic drugs have potent antibacterial activity and can adversely affect the
human microbiome. The mechanistic underpinning of this toxicity remains largely unknown. We
investigated the antibacterial activity of 200 drugs using genetic screens with thousands of barcoded
Escherichia coli knockouts. We analyzed 2 million gene-drug interactions underlying drug-specific
toxicity. Network-based analysis of drug-drug similarities revealed that antibiotics clustered into
modules that are consistent with the mode of action of their established classes, whereas nonantibiotics
remained unconnected. Half of the nonantibiotics clustered into separate modules, potentially revealing
shared and unexploited targets for new antimicrobials. Analysis of efflux systems revealed that they
widely affect antibiotics and nonantibiotics alike, suggesting that the impact of nonantibiotics on
antibiotic cross-resistance should be investigated closely in vivo.

https://doi.org/10.5061/dryad.1zcrjdfxw.
ACKN OW LEDG MEN TS
We thank the anonymous reviewers for comments on a previous
version of the manuscript and our stakeholder advisory committee
chaired by S. Murphy (IISD) and consisting of 19 people
representing the following agencies and organizations: MOSES, the
GEF, IPES FOOD, NFU, NFFC, EFAO, FAO, CIFOR, McKnight
Foundation, Bioversity, NSAC, USDA, CATIE, TNC, and CIAT.
Funding: This work was supported by the National Socio-
Environmental Synthesis Center (SESYNC) under funding received
from the National Science Foundation (grant DBI-1639145 to
Z.M. and C.K.) and by the UBC Research Excellence Cluster for
Diversified Agroecosystems. L.V.R. was funded by the European
Research Council under the European Union’s Horizon 2020
Research and Innovation Programme (grant 853222 FORESTDIET).
I.G. was partially funded by the German Research Foundation
(DFG project number 532858005 GR 4844/4-1). C.L. was funded
O
ur microbiomes play a critical role in main-
taining our health and can affect the ef-
ficacy of various therapeutics. Studies
have revealed that the effectiveness of many
drugs not traditionally prescribed for treat-
ing bacterial infections, called nonantibiotics
hereafter, can be affected by drug-microbiome
interactions, including bacterial-driven drug
metabolism (1–3) and bioaccumulation (4).
Complementary studies of nonantibiotics have
Department of Systems Biology, University of Massachusetts
Medical School, Worcester, MA, USA.
*Corresponding author. Email: amir.mitchell@umassmed.edu
also investigated the reciprocal effect: how non-
antibiotic drugs might disrupt the microbiome
and potentially culminate in dysbiosis, a detri-
mental disruption of the host’s microbiome.
Recent works revealed that multiple nonanti-
biotics are associated with shifts in microbiome
species composition (5, 6). Examples include
antidiabetics (7), proton-pump inhibitors (8, 9),
antipsychotics (10), and nonsteroidal anti-
inflammatory drugs (11). Although the mecha-
nisms underlying these associations are complex,
direct drug-bacterial interactions are thought to
be one key force mediating these shifts (2). A
recent in vitro screen revealed that roughly a

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 93

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

sis that extended beyond individual drug-gene

interactions. We discovered that nonantibiotics
operate through toxicity mechanisms that are
orthogonal to standard antibiotics. Conversely,
we found that knockout of efflux systems in-
creased sensitivity against most antibiotics and
nonantibiotics. Follow-up lab evolution experi-
ments with three nonantibiotics supported
the main conclusions of our analysis by demon-
strating that evolved resistance can culminate
in broad antibiotic cross-resistance and by un-
covering a bacterial-translation initiation factor,
initiation factor 2 (IF2), as a previously unknown
cellular target that is not currently exploited
by standard antibiotics.

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

94 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

Fig. 2. Genetic screens capture antibiotic mechanisms of action and reveal
interclass interactions. (A) Hierarchical clustering of antibiotics by the fitness
score of informative knockouts. The colored squares at the bottom row indicate
the class of each antibiotic and show that the algorithm successfully grouped
most antibiotics into class-specific clusters. The right panels highlight signature
knockouts characteristic to specific clusters. (B) Network representation of
antibiotic drug resemblance by informative knockouts. Drug-drug correlations in
knockout fitness were used as the measurement of resemblance, and the
network was generated with a force-layout algorithm according to the correlation
scores. Node and frame colors mark the antibiotic class. Most antibiotics self-
grouped into the same class of network modules. The side panels show
functional enrichment analysis by GO biological process terms of each module.
Circle size and shading mark the size of gene set and statistical significance,
respectively. ncRNA, noncoding RNA; met., metabolic; biosyn., biosynthetic.

different groups revealed highly dissimilar dis-
tributions (Fig. 1E). We used these distributions
to define a lower (permissive) cutoff for chem-
ical similarity. Hence, antibacterial activity of
nonantibiotics is not necessarily a result of chem-
ical similarity to standard antibiotics.
Pooled genetic screen for identifying modulators
of toxicity
We next employed a pooled genetic screening
method that we previously developed (15, 19, 20)
to identify the cellular pathways modulating
the antibacterial activity of 186 drugs. To iden-
tify the genes influencing drug-specific sensitiv-
ity, we used a pooled collection of 6709 barcoded
knockout E. coli strains covering 3500 nones-
sential genes. We determined the concentration
needed to achieve 50% growth inhibition (IC50)
of wild-type E. coli for each drug. Drugs with

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 95

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

Fig. 3. Mechanisms underlying anti-

bacterial activity of nonantibiotic
drugs. (A) Network representation
of resemblance across all drugs by 581
antibiotic informative knockouts. Node
color marks the antibiotic class, and
the node shape marks the drug use in
host treatment. Only a few nonanti-
biotic drugs are clustered within anti-
biotic classes, indicating that most
of them operate through different modes
of action. (B) Network representation of
nonantibiotic drug resemblance. The
network was inferred by 509 nonanti-
biotic informative knockouts. Most drugs
are associated with at least one neighbor,
and many fall within clusters. Five
clusters, marked by a continuous line,
had statistically significant functional
enrichment. Drugs marked in pink were
further investigated. (C) Functional
enrichment of nonantibiotic clusters by
GO biological process terms. Circle
size and shade mark the size of gene
set and statistical significance, respectively.
pyr, pyramidine; LPS, lipopolysaccharide.
(D) The impact of knockouts in central
drug-transport systems and membrane-
permeability systems across individual
drugs. Black circles represent statistically
significant increased sensitivity. The
colored bars at the bottom show the
number of efflux and permeability
systems that increased sensitivity for
each individual drug. The white bars
(right) show the number of drugs that
have increased sensitivity for each
individual efflux and permeability system.
Drugs marked in pink were further
investigated. The colors mark antibiotic
class as shown in (A). ABC, adenosine
triphosphate (ATP)–binding cassette;
MATE, multidrug and toxic compound
extrusion; SMR, small multidrug
resistance; MFS, major facilitator
superfamily.

low toxicity (IC50 > 50 mM) were used at a con-
centration of 50 mM. We removed eight drugs
from the panel for technical reasons (table
S1B). In three replicates, we inoculated the
strain collection into minimal media with or
without the drugs and terminated the screen
when controls approached stationary phase
(Fig. 1C). We sequenced the barcodes from ex-
tracted DNA and calculated the frequency of
each knockout. We identified 191,552 statistically
significant depleted or enriched knockouts
relative to the no-drug control [FDR-adjusted
P value < 0.25, Wald test in the DESeq2 prog-
ram (21)]. Depleted knockouts reflect gene loss
of function that increases drug sensitivity,
whereas enriched knockouts reflect loss of
function that increases resistance (all results
are in table S3).
Known toxicity mechanisms are captured for
standard antibiotics
We validated our method by verifying that it
correctly identifies the known modes of action
of 90 well-characterized antibiotics. Specifical-
ly, we chose a subset of knockouts (581 strains)
that we expected would be informative for
identifying shared modes of action by filtering
out strains that rarely modulated toxicity (hits

96 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE


in less than two antibiotics) and strains that
modulated toxicity very frequently (hits in
more than 30 antibiotics). We also discarded
knockouts of specific biological functions, such
as efflux pumps, that can hinder the analysis of
the mode of action (table S4). Figure 2A shows
the results of hierarchical clustering of the infor-
mative strains. We observed that clustering was
highly successful in recapturing antibiotic classes
(shown by the colored dendrogram at the top
of the heatmap): Seventy-five antibiotics were
grouped into “pure” clusters with a single anti-
biotic class. All four cell-membrane targeting
antibiotics were in a single mixed cluster.
Focusing on top hits of individual clusters
uncovered many genes from the known target
pathways (Fig. 2A, right side). For example,
knockouts disrupting the SOS response and
DNA repair (e.g., recA and recN) increased sen-
sitivity to DNA-targeting antibiotics (red and
pink clusters), whereas knockouts disrupting
nucleotide excision repair, such as uvrABC,
only increased sensitivity to nitrofurans (pink
cluster) that induce DNA adducts requiring dis-
tinct repair mechanisms (22). For folic acid–
targeting antibiotics, deletions in folX and folM
increased resistance, whereas deletions in the
glycine cleavage system (gcv genes) increased
sensitivity (23). Knockouts disrupting peptido-
glycan recycling increased sensitivity to cell-
wall–targeting antibiotics.
To explore the relationship between anti-
biotic classes, we calculated the Pearson corre-
lation between each pair of antibiotics (Fig. 2A,
columns) and generated a graph representing
drug relatedness. Figure 2B and fig. S1A show
the drug similarity network (network layout
was set with a force-directed algorithm by the
strength of drug-drug correlations). Similarly
to hierarchical clustering, this analysis grouped
most antibiotics by their mechanism of action.
However, the graph-based analysis also high-
lighted interclass relatedness that cannot be
explained by chemical similarity (Fig. 2B, blue
edges). Interclass relatedness inferred from the
screens reveals shared pathways modulating
the drug toxicity that extend beyond the anti-
biotic target. DNA-targeting antibiotics emerged
as a network hub. Relatedness between classes
was also supported by a separate approach—
functional enrichment (Fig. 2B, margins, and
table S5). This analysis revealed that cellular
pathways known to be related to one class are
also enriched in seemingly unrelated classes,
such as DNA-related categories enriched in
ribosome-targeting antibiotics.
Taken together, our results verified our ap-
proach for capturing the established mech-
anisms of action of standard antibiotics. The
analysis revealed the value of this approach
in uncovering higher-order interactions between
antibiotic classes that have different cellular
targets yet affect common pathways that mod-
ulate toxicity.
SCIENCE science.org
Antibacterial activity of nonantibiotics differs
from that of standard antibiotics
We sought to identify the toxicity mechanisms
of nonantibiotics by adding them to the drug-
relatedness network that we generated for anti-
biotics. This analysis adds more nodes (drugs)
and edges (relatedness interactions) to the net-
work without altering the topology among anti-
biotics. In this analysis, nonantibiotics that
target similar pathways to those of standard
antibiotic classes would cluster with antibiotic
network modules (Fig. 3A). However, we ob-
served that most nonantibiotics remained com-
pletely unconnected to the antibiotic modules
(57/77). The notable exceptions were 14 nonanti-
biotics that were connected to multiple anti-
biotic classes (positioned between the colored
modules). When focusing on 52 chemically sim-
ilar antibiotic and nonantibiotic pairs (Tanimoto
coefficient >0.2), we did not detect high sim-
ilarity in the genetic screen results (fig. S2A).
To substantiate this conclusion, we used an
alternative and widely used approach for clas-
sifying the mode of action of nonantibiotics. Spe-
cifically, we trained a random forest classifier
that relied on all knockout strains to predict
whether drugs can be classified as belonging
to specific antibiotic classes (this analysis cir-
cumvented our choice of informative knock-
outs). In agreement with our network analysis,
we observed that although our classifier suc-
cessfully assigned most antibiotics to their
correct class, it failed to classify most nonanti-
biotics (fig. S2, B and C). Overall, the results in-
dicate that nonantibiotic drugs likely operate
through toxicity mechanisms that are orthog-
onal to standard classes of antibiotics.
The network analysis revealed that although
most nonantibiotics were unconnected to anti-
biotic modules, some nonantibiotics are re-
lated to one another (Fig. 3A). This suggested
that relatedness may exist but that it may be
occluded by the antibiotic-centered analysis. We
therefore repeated the network analysis, relying
exclusively on knockouts that were informative
for nonantibiotics (using similar filters as before).
Figure 3B and fig. S1B show the resulting rel-
atedness network that is based on drug-drug
correlations across 509 informative knockouts.
In this analysis, 52 of 77 nonantibiotics had at
least one neighbor. Many connections were
identified between drugs that are used for
treating different human conditions and dis-
eases (Fig. 3B, different node shapes), and most
connections could not be explained by chem-
ical similarity (Fig. 3B, blue edges).
We identified nine independent network
modules and found that five modules were
enriched for specific biological functions (table
S4). Figure 3C shows the top enriched Gene
Ontology (GO) biological processes summar-
ized by REVIGO (reduce and visualize Gene
Ontology) (24, 25). The largest module (14 drugs)
was enriched for diverse biosynthetic and meta-
RESE ARCH | R E S E A R C H A R T I C L E S
bolic processes. This module included fluoro-
pyrimidine chemotherapeutic drugs, which we
previously characterized with similar findings (15).
The second-largest module (10 drugs) was en-
riched in processes related to lipopolysaccharide
biosynthesis. The third module was enriched in
processes related to DNA damage repair and
the stress response, and the two smallest mod-
ules included processes related to oxidative
stress and siderophore-biosynthesis pathways.
Taken together, results from analyses of non-
antibiotic drugs indicated that their antibac-
terial mechanisms are largely distinct from those
of standard antibiotics. Moreover, our network
analysis showed that nonantibiotics are not
completely distinct from one another but that
most of them can be grouped into modules
reflecting shared toxicity mechanisms. This
conclusion is further supported by the func-
tional enrichment analysis, which uncovered
common pathways modulating antibacterial
activity. In addition, the modular network struc-
ture indicates that unexploited yet robust cel-
lular targets for inhibiting bacterial growth may
exist and provides multiple lead molecules for
further investigation.
Transport systems affect both antibiotics
and nonantibiotics
Our network analyses purposefully disregarded
genes encoding efflux pumps, because they
are expected to affect the antibacterial activity
irrespective of the compound’s target. We sub-
sequently examined the impact on antibacterial
activity of genes involved in membrane perme-
ability and transport (76 genes from 33 pathways
of drug transport and membrane permeabil-
ity, table S4). Figure 3D shows interactions
identified across all the drugs tested, as well as
the pathways that increase bacterial sensitivity
upon deletion. The widest impact on toxicity was
observed for mutations affecting the resistance-
nodulation-division (RND) family transporters,
the Tol-Pal system, and outer membrane pro-
teins (OMP). We observed that drugs varied
considerably in the number of interactions that
they have with transport systems, with 26 drugs
affected by five or more different transport sys-
tems (specificity of pumps to antibiotic classes
is shown in fig. S3). Overall, we observed that
mutation in at least one transport system in-
creased the antibacterial activity of 93% of anti-
biotics and 82% of nonantibiotics (Fig. 3D).
This observation implies that evolved resistance
against nonantibiotic drugs can potentially,
in some cases, inadvertently lead to multidrug
resistance and undermine the efficacy of anti-
biotics. Increased risk of unintentional cross-
resistance may exist for nonantibiotic drugs that
are affected by multiple transport systems.
Evolved resistance against nonantibiotic drugs
Our genetic screens, coupled with network anal-
ysis, have led to two important conclusions. First,
5 APRIL 2024 • VOL 384 ISSUE 6691 97
RES EARCH | R E S E A R C H A R T I C L E S
Fig. 4. Evolved resistance against nonantibiotics exposes unknown drug
targets and collateral cross-resistance. (A) Overview of evolution experiment
and downstream assays. Independent cultures were serially transferred in
media supplemented with three nonantibiotics over 200 generations. A single
resistant clone was isolated from each independently evolved population. (B) Drug
resistance evolved across all cultures and drugs. The panels show the drug-
sensitivity curves of ancestor and evolved clones; horizonal lines mark the
concentration that inhibits growth by 50% (IC50). (C) Results of whole-genome
sequencing uncover three alternative strategies of evolved resistance. The
concentric Circa plots represent the bacterial chromosome of individually evolved
98 5 APRIL 2024 • VOL 384 ISSUE 6691
clones, and dots mark mutations. Dark-pink dots mark gene hits across at
least four of the five replicates and likely adaptive mutations. Light-pink
dots mark mutations in genes conferring drug resistance according to the
genetic screen. The diagram on top represents the inferred mechanism
of resistance. (D) Sertraline-evolved strains are resistant to multiple antibiotics,
whereas triclabendazole- and streptozotocin-evolved strains do not confer
cross-resistance. The panels show the increased resistance (IC50 fold-change) of
evolved clones against six antibiotics. The bars represent the median across
all evolved clones, and dots represent the mean of three biological replicates of
each clone. The bar color shows the antibiotic class.
science.org SCIENCE

the orthogonality in toxicity mechanisms indi-
cates that nonantibiotics may potentially un-
cover unexploited targets for inhibiting bacterial
growth. Second, analysis of drug-transport sys-
tems indicates that despite this orthogonality,
broad resistance to antibiotics can emerge if
drug transport–based resistance evolves as a
result of selection by nonantibiotics. We tested
these predictions experimentally by selecting for
drug resistance in evolution experiments (26)
(Fig. 4A). We focused on three nonantibiotics se-
lected from different modules in the nonanti-
biotic network: streptozotocin, triclabendazole,
and sertraline (Fig. 3B). Information about the
toxicity of these drugs across representative
microbiome species and five E. coli strains is
provided in table S6.
Figure 4B shows the drug sensitivity curves
of the ancestor strain and of five indepen-
dently evolved clones. Streptozotocin- and
triclabendazole-adapted clones were fully re-
sistant to the drugs, whereas sertraline-adapted
clones increased resistance by more than two-
fold. We sequenced the genomes of evolved
strains to uncover the mechanisms underly-
ing resistance. Figure 4C shows the mutations
that we identified with the breseq tool (27). We
used genes mutated across independent clones
as evidence for likely driver mutations (Fig. 4C,
dark-pink dots). Further support was provided
by overlap of mutations with genetic-screening
hits (Fig. 4C, light-pink dots).
We found three representative adaptation
strategies for resistance. We identified numer-
ous mutations in streptozotocin-evolved clones
with a clear location bias toward the origin of
replication. Streptozotocin, a naturally occur-
ring alkylating agent that is used to treat pan-
creatic neuroendocrine tumors (28), repeatedly
selected for inactivation of nagE, which en-
codes for glucose-6-phosphate transporter. This
transporter is likely central for drug import
and was previously observed to confer strep-
tozotocin resistance (29). Clones adapted to
triclabendazole, a drug used for treating worm
infections, had only a handful of mutations.
However, all independently evolved clones
shared an in-frame mutation in a specific re-
gion of the essential gene infB, which encodes
the translation initiation factor IF2. Previous
work on lamotrigine, a drug that we identi-
fied as a triclabendazole neighbor in the drug-
relatedness network (Fig. 3B), found that it
binds to the same region of IF2 (30). Last, all
sertraline-adapted clones had mutations in
key regulators of efflux pumps (marR, acrR,
and lon). Mutations targeting these regula-
tors were very recently reported in sertraline-
adapted strains and were shown to increase
expression of multiple efflux systems (31). Our
genetic screens also showed that sertraline,
more than triclabendazole and streptozotocin,
is sensitive to inactivation of transport systems
(Fig. 3D).
SCIENCE science.org
Given the bacterial adaptation strategies
that we observed, we predicted that sertraline-
adapted clones, but not other clones, will emerge
as multidrug resistors. Experiments measur-
ing the resistance against a panel of antibiotics
confirmed this prediction (Fig. 4D). In addi-
tion, given the hypothesis that triclabendazole
binds to IF2, we predicted that triclabendazole-
adapted clones will also be resistant to neigh-
boring drugs in the relatedness network (fig.
S4A). Drug-sensitivity measurements confirmed
this prediction by showing that triclabendazole-
evolved clones were also resistant to lamotrigine
and pyrimethamine (fig. S4B). Furthermore,
compound binding SEC-ASMS assays (size ex-
clusion chromatography followed by affinity
selection-mass spectrometry) confirmed that
all three drugs bind the IF2 protein (fig. S4C).
The results from our evolution experiments
uncovered instances of three representative
strategies for adaptive resistance: decreased
import, mutation of target, and increased ex-
port. However, beyond the specific mechanisms
of action, the evolution experiment substanti-
ated the two main premises arising from our
work: (i) that orthogonality in modes of action
can be leveraged to uncover proteins not cur-
rently used as targets by standard antibiotics
and (ii) that despite orthogonality, adaptation
against nonantibiotics can emerge through
changes in drug efflux and therefore inadver-
tently culminate in antibiotic cross-resistance.
Discussion
Human health and microbiome integrity are
tightly interlinked. This fundamental realiza-
tion is the basis of current efforts to identify
correlations between a myriad of environmen-
tal factors—such as diet and medication—and
microbiome species composition. Although the
widespread antibacterial activity of nonanti-
biotics is thought to be a key force underlying
these associations, the modes of action of non-
antibiotics in bacteria remain largely unknown.
We addressed this gap in knowledge by devel-
oping a high-throughput genetic screening ap-
proach for systematically mapping interactions
between dozens of antibacterial drugs and single-
gene knockouts. Although a well-characterized
lab strain of E. coli was an ideal platform for a
high-throughput investigation of mode of ac-
tion, additional work is needed to extend our
study’s conclusions to the human microbiome.
Specifically, testing is necessary to determine
whether nonantibiotics operate with similar
modes of action in microbiome isolates and
whether drug adaptation is observed in treated
individuals.
The network we constructed using drug-drug
correlations recaptured the known antibiotic
classes (Fig. 2B). When we added nonanti-
biotics to the antibiotic-centered clustering
analysis, most of them remained unconnected,
indicating that they do not inhibit E. coli through
RESE ARCH | R E S E A R C H A R T I C L E S
mechanisms similar to those of standard anti-
biotics (Fig. 3A). A subsequent analysis revealed
that most of the nonantibiotics clustered into
highly intraconnected modules enriched in
specific cellular pathways (Fig. 3, B and C). Re-
sults from almost 200 drugs support the hy-
pothesis that antibiotics and nonantibiotics
target bacteria through orthogonal mecha-
nisms of action. Network analysis focusing
exclusively on nonantibiotics indicated that
many of them share common toxicity mech-
anisms (Fig. 3B). Moreover, analysis of genes
involved in drug transport indicated that de-
spite orthogonality in mode of action, bacterial
sensitivity to most drugs is increased by inac-
tivation of transport systems and membrane-
permeability proteins (Fig. 3D). The extensive
impact of efflux systems on nonantibiotics is
consistent with their known role in xenobiotic
detoxification (32).
Our screen results showing orthogonality in
toxicity suggest that nonantibiotics target bac-
terial cellular components not currently ex-
ploited by standard antibiotics. The modular
network topology (Fig. 3B) further implies
that some targets may be common to multiple
nonantibiotics. This prediction is supported
by our observations of triclabendazole-evolved
resistance by in-frame mutations in IF2, a pro-
tein not targeted by standard antibiotics. Further-
more, confirming our network topology, we
found that the two neighboring drugs in our net-
work also bind IF2 (fig. S4). Another prediction,
stemming from our finding of widespread
sensitivity to inactivation of drug-transport sys-
tems, suggested that evolved resistance against
nonantibiotics may confer broad drug resistance;
this prediction was validated with sertraline-
adapted strains (Fig. 4D). Although estimates
of drug concentrations in the human digestive
tract suggest that many nonantibiotics may
apply a selective pressure on the gut micro-
biome, it remains to be determined whether
adaptions will influence efflux pumps and in-
advertently affect antibiotic resistance. Our
results allow the gauging of nonantibiotics in-
teracting with multiple efflux systems (Fig. 3D).
This ranking can inform future studies fo-
cusing on specific interactions in the human
microbiome.
The emergence and spread of antibiotic-
resistant pathogens is a fundamental challenge
for global health (33). Decreasing investment
in development of new antibiotics in recent dec-
ades is only expected to compound the prob-
lem (34). Although technological advancements
in robotic automation greatly increase the ca-
pacity to screen ever-growing libraries of small
molecules (35), rediscovery of antibacterial
compounds that belong to established antibi-
otic classes remains a bottleneck for innova-
tion (36). This obstacle is common to both
traditional natural product–based screening
(37) and compounds proposed by artificial
5 APRIL 2024 • VOL 384 ISSUE 6691 99
RES EARCH | R E S E A R C H A R T I C L E S

intelligence (38). Our methodology can be used
to address this challenge because it can com-
pare the pathways underlying antibacterial
activity between hundreds of uncharacter-
ized compounds and established antibiotics.
As we showed, the results of these screens
can be used for training classifiers to identify
toxicity profiles that match established anti-
biotic classes. These classifiers can therefore
also identify compounds operating through
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).
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,
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).
6. A. Vich Vila et al., Nat. Commun. 11, 362 (2020).
7. K. Forslund et al., Nature 528, 262–266 (2015).
8. F. Imhann et al., Gut 65, 740–748 (2016).
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).
11. M. A. M. Rogers, D. M. Aronoff, Clin. Microbiol. Infect. 22, 178.
e1–178.e9 (2016).
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
(1960).
14. A. Tomasz, E. Borek, Biochemistry 1, 543–552 (1962).
15. B. Rosener et al., eLife 9, e59831 (2020).
16. A. P. García-González et al., Cell 169, 431–441.e8 (2017).
17. H. L. Morgan, J. Chem. Doc. 5, 107–113 (1965).
18. D. Rogers, M. Hahn, J. Chem. Inf. Model. 50, 742–754
(2010).
19. M. Noto Guillen, B. Rosener, S. Sayin, A. Mitchell, Cell Syst. 12,
1064–1078.e7 (2021).
Visualization: A.M. and M.N.G. Formal analysis: A.M., M.N.G., claim to original US government works. https://www.science.org/
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 –
editing: A.M. and M.N.G. Writing – review: A.M., M.N.G., B.R., SUPPLEMENTARY MATERIALS
and C.L. Competing interests: The authors declare that they have
science.org/doi/10.1126/science.adk7368
no competing interests. Data and materials availability: Raw
Materials and Methods
sequencing reads from barcoded knockout library screen and whole
genomes of evolved bacteria are available on the NCBI Sequence
Figs. S1 to S4
Tables S1 to S6
Read Archive under bioprojects PRJNA986181 and PRJNA998361.
MDAR Reproducibility Checklist
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
Copyright © 2024 the authors, some rights reserved; exclusive Published online 14 March 2024
licensee American Association for the Advancement of Science. No 10.1126/science.adk7368
MICROBIOLOGY
Prophage terminase with tRNase activity sensitizes
Salmonella enterica to oxidative stress
Siva Uppalapati1†, Sashi Kant1†‡, Lin Liu1, Ju-Sim Kim1, David Orlicky2,
Michael McClelland3, Andres Vazquez-Torres1,4*
Phage viruses shape the evolution and virulence of their bacterial hosts. The Salmonella enterica genome
encodes several stress-inducible prophages. The Gifsy-1 prophage terminase protein, whose canonical function
is to process phage DNA for packaging in the virus head, unexpectedly acts as a transfer ribonuclease
(tRNase) under oxidative stress, cleaving the anticodon loop of tRNALeu. The ensuing RNA fragmentation
compromises bacterial translation, intracellular survival, and recovery from oxidative stress in the vertebrate
host. S. enterica adapts to this transfer RNA (tRNA) fragmentation by transcribing the RNA repair Rtc system.
The counterintuitive translational arrest provided by tRNA cleavage may subvert prophage mobilization and
give the host an opportunity for repair as a way of maintaining bacterial genome integrity and ultimately
survival in animals.

20. S. Sayin et al., eLife 12, e83140 (2023).
21. M. I. Love, W. Huber, S. Anders, Genome Biol. 15, 550 (2014).
22. K. R. Ona, C. T. Courcelle, J. Courcelle, J. Bacteriol. 191,
4959–4965 (2009).
23. R. J. Nichols et al., Cell 144, 143–156 (2011).
24. M. Ashburner et al., Nat. Genet. 25, 25–29 (2000).
25. F. Supek, M. Bošnjak, N. Škunca, T. Šmuc, PLOS ONE 6,
e21800 (2011).
26. J. E. Barrick, R. E. Lenski, Nat. Rev. Genet. 14, 827–839
(2013).
27. J. E. Barrick et al., BMC Genomics 15, 1039 (2014).
28. J. Capdevila et al., Neuroendocrinology 112, 1155–1167 (2022).
29. J. Lengeler, Mol. Gen. Genet. 179, 49–54 (1980).
30. J. M. Stokes, J. H. Davis, C. S. Mangat, J. R. Williamson,
E. D. Brown, eLife 3, e03574 (2014).
31. Y. Wang et al., Proc. Natl. Acad. Sci. U.S.A. 120, e2208344120
(2023).
32. P. J. F. Henderson et al., Chem. Rev. 121, 5417–5478 (2021).
33. C. J. L. Murray et al., Lancet 399, 629–655 (2022).
34. M. A. Cook, G. D. Wright, Sci. Transl. Med. 14, eabo7793
(2022).
35. L. Martínez-Fructuoso et al., ACS Infect. Dis. 9, 1245–1256
(2023).
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
(2021).
38. J. M. Stokes et al., Cell 180, 688–702.e13 (2020).
39. M. N. Guillen, A. Mitchell, Mitchell-SysBiol./2024notoetal:
v1.0.0, Zenodo (2024); https://zenodo.org/doi/10.5281/
zenodo.10525157.
ACKN OW LEDG MEN TS
We thank N. London and A. Orlov for their advice on measuring
chemical similarity. We thank H. Youk and C. Navarro for their
comments on the manuscript. Funding: This research was
supported by National Institute of General Medical Sciences grant
R35GM133775 (A.M.) and by National Institute of Allergy and
Infectious Diseases grant R01AI170722 (A.M.). Author
contributions: Conceptualization: A.M. and M.N.G. Methodology:
A.M., M.N.G., B.R., and C.L. Investigation: M.N.G., B.R., and C.L.
O
xidative stress imposes tremendous se-
lective pressures on all domains of life.
Intracellular bacterial pathogens, such
as S. enterica, that reside in effector cells
of the mammalian innate immune sys-
tem, are exposed to high concentrations of
reactive oxygen species (ROS) synthesized
by the respiratory burst of the phagocyte
NADPH (nicotinamide adenine dinucleotide
phosphate) oxidase (1). The cytotoxic concen-
trations of ROS produced by the respiratory
burst damage nucleotides, metal cofactors, and
sulfur-containing amino acids (2). Oxidation
of biosynthetic enzymes induces functional
auxotrophies for aromatic and branched-chain
amino acids (3, 4). The resulting drops in ami-
no acids are followed by a surplus of deacyl-
ated transfer RNAs (tRNAs) that trigger an
adaptation in bacteria commonly known as
the stringent response (5–7). The combination
of a shrinking pool of charged tRNAs together
1
University of Colorado School of Medicine, Department of
Immunology and Microbiology, Aurora, CO, USA. 2University
of Colorado School of Medicine, Department of Pathology,
Aurora, CO, USA. 3Department of Microbiology and Molecular
Genetics, University of California Irvine School of Medicine,
Irvine, CA, USA. 4Veterans Affairs Eastern Colorado Health
Care System, Denver, CO, USA.
*Corresponding author. Email: andres.vazquez-torres@cuanschutz.edu
†These authors contributed equally to this work.
‡Present address: Bacterial Pathogenesis, Boehringer Ingelheim Animal
Health USA, Ames, IA, USA.
with the repression of ribosomal RNA (rRNA)
and tRNA genes undermines the translational
capacity of phylogenetically diverse micro-
organisms undergoing the stringent response
during periods of oxidative stress. In this study,
we describe a lysogenic prophage that represses
translation in Salmonella exposed to oxidative
stress generated by the phagocyte NADPH
oxidase through the moonlighting transfer
ribonuclease (tRNase) function of a deoxy-
ribonuclease (DNase) whose canonical role is
viral-genome processing and capsid packaging.
Results
Translational halts in S. enterica after
oxidative stress
To measure de novo translation in S. enterica
in vitro in response to oxidative stress gen-
erated by application of hydrogen peroxide
(H2O2), we used a Western blot version of
SUnSET (surface sensing of translation) (8).
S. enterica experienced a substantial (P <
0.0001) inhibition of de novo protein syn-
thesis upon exposure to 400 mM H2O2 (Fig. 1,
A and B). Under the high density of bacterial
cultures used in these experiments, 400 mM
H2O2 did not affect bacterial viability (Fig. 1C),
indicating that S. enterica can tolerate mod-
erate levels of peroxide stress by repressing
translation. S. enterica resumed translation
90 min after exposure to H2O2 (Fig. 1, D and E).

100 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

Fig. 1. Transcriptional adaptations to the repression of de novo translation
in S. enterica undergoing oxidative stress. (A and D) Immunoblots with
Ponceau S–stained controls and (B and E) a corresponding densitometric
analysis examining puromycin incorporation by S. enterica 30 min after H2O2
treatment. Densitometry was normalized against Ponceau S–stained lanes
with ImageJ. Data are shown as mean ± SD (n = 3 independent observations);
****P ≤ 0.0001 as determined by one-way analysis of variance (ANOVA) with
Dunnet’s multiple comparison test. (C) Killing of S. enterica by H2O2 in PBS after
30 min of treatment. Data are shown as mean ± SD (n = 4); ****P ≤ 0.0001 and
***P ≤ 0.001 as determined by one-way ANOVA with Dunnet’s multiple
comparison test. (F) Differentially expressed genes in S. enterica after 30 min of
400 mM H2O2 treatment as assessed by RNA-seq (30). Each circle represents the
average log2-fold change of four biological replicates. Genes with a log2-fold
change >0.8 or <−0.8 are depicted on the top and bottom, respectively. Orange
and green boxes represent pathogenicity islands, and the yellow box represents
S. enterica plasmid genes.

To identify responses that may underlie
recovery of translational activity during oxida-
tive stress, we investigated the genome-wide
transcriptional responses of S. enterica ex-
posed to H2O2. Hierarchical clustering analy-
sis of RNA-sequencing (RNA-seq) data revealed
that 497 and 569 genes were up-regulated and
down-regulated, respectively, in H2O2-treated
S. enterica compared with untreated controls
(fig. S1A) [false discovery rate (FDR) P value <
0.05; tables S1 to S4]. This analysis showed
transcriptional up-regulation of katG, trxC,
and dps of the OxyR regulon, of mntH and
sitABCD involved in manganese transport,
and of genes encoding the Salmonella patho-
genicity island-2 type III secretion system, gly-
colysis, iron acquisition, and [Fe-S] repair. By
contrast, genes encoding oxidative phosphor-
ylation; ribosomal proteins; or the biosynthe-
sis of lipids, amino acids, and aminoacyl tRNAs
were down-regulated. Quantitative reverse tran-
scription polymerase chain reaction (RT-PCR)
confirmed the up-regulation of katG and sufABC
genes and the down-regulation of oxidative
phosphorylation genes (fig. S1, B and C).
The patterns of gene expression that we
found are consistent with the idea that
S. enterica adapts to oxidative stress by up-
regulating antioxidant defenses, while favor-
ing overflow metabolism and undergoing the
stringent response (2, 5, 9). We also noted that
the rsr-yrlBA-rtcBA operon, which mediates
repair of fragmented RNA molecules, was up-
regulated in H2O2-treated S. enterica (Fig. 1F
and fig. S1D). The rtc operon encodes for the
RNA cyclase RtcA, the RNA ligase RtcB, and the
s54 transcriptional activator RtcR (10). RtcA
converts 3′ and 5′-phosphate RNA fragments
to a 2′, 3′- cyclic phosphate intermediate (11),
which serves as a substrate for the phosphodi-
esterase RtcB. The RtcB protein also catalyzes
the GTP-driven ligation of 3′-phosphate and 5′-
hydroxyl RNA fragments (12, 13). The activation
of rtc genes during oxidative stress prompted us
to explore the role of this operon in adaptation
of S. enterica to the inhibition of de novo pro-
tein synthesis through oxidative stress.
Role of the RNA repair Rtc system during
oxidative stress
We tested the capacity of S. enterica rtcBA and
rtcR deletion mutants to synthesize protein
after exposure to H2O2. As reported above,
wild-type (WT) S. enterica recovered de novo
protein synthesis after 90 min of H2O2 treat-
ment. In comparison, the DrtcBA and DrtcR
isogenic controls undergoing oxidative stress

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 101

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

Fig. 2. Repair of tRNA by the RNA

repair Rtc system protects
S. enterica from the phagocyte
NADPH oxidase. Densitometry (A and
B) of the puromycin+ proteome in
WT and rtc mutants. Specimens in
(B) were treated with 400 mM H2O2.
Data are shown as mean ± SD (n = 4).
(C) tRNA fragmentation was visualized
by Northern blotting in S. enterica
after 5 mM H2O2 treatment.
(D) Densitometry of tRNA fragments
2 hours after H2O2 treatment,
corresponding to data in (C). Data
are shown as mean ± SD (n = 3).
(E) Position of tRNALeuPQTV cleavage in
H2O2-treated S. enterica as assessed
by sequencing and 3′ RACE. (F) Killing
of S. enterica after 2 hours of treatment
with 400 mM H2O2 in phosphate-
buffered saline (PBS). Data are shown
as mean ± SD (n = 10). (G) Intracellular
survival of S. enterica in bone marrow–
derived macrophages from C57BL/6 (B6)
and Cybb−/− mice. Data are shown as
mean ± SD (n = 6). (H) Competitive index
of S. enterica in livers and spleens of
C57BL/6 (B6) and Cybb−/− mice 3 days
after i.p. inoculation with equal numbers
of WT and DrtcBA S. enterica (n = 6
to 7). (I) Histopathology of paraffin-
embedded, hematoxylin and eosin
(H&E)–stained liver tissues isolated
3 days after infection. Scale bars,
50 mm. The panel on the right shows the
average number of microabscesses
and necrotic foci per 200X field of liver
tissue stained with H&E (n = 6 to 7). *P ≤ 0.05, **P ≤ 0.01, ***P < 0.001, and ****P < 0.0001 as determined by Student’s t test [(A), (D), and (F)], Mann-Whitney test (G),
one-way ANOVA with Dunnet’s multiple comparison test (B), or two-way ANOVA (H).

showed delayed (P < 0.01) de novo protein
translation during the recovery phase (Fig. 2,
A and B, and fig. S2, A to C). These data in-
dicate that the RNA repair Rtc system plays a
role in the adaptation of S. enterica to transla-
tional halts that occur during oxidative stress.
Coinciding with rtcB gene transcription (fig.
S3A), the addition of H2O2 to WT S. enterica
resulted in fragmentation of tRNA Tyr and
tRNALeuPQTV (Fig. 2, C and D), so we explored
how the Rtc system aids the recovery of de
novo translation through RNA repair (10). Fur-
ther analysis revealed that S. enterica under-
going oxidative stress specifically experienced
fragmentation of all major leucine tRNA iso-
acceptors tRNALeuPQTV, tRNALeuX, and tRNALeuZ,
but not tRNAGln, tRNAHis, or tRNAMet (fig. S3B).
S. enterica DrtcBA and DrtcR mutants exper-
ienced higher levels of tRNALeu and tRNATyr
fragmentation after H2O2 treatment than did
WT controls (Fig. 2C and fig. S3C). Expression
of the rtcBA operon in trans reversed the ab-
normally high tRNALeuPQTV fragmentation of
DrtcBA S. enterica after H2O2 treatment (fig. S3D).
Sequencing of 3′ RACE products of tRNALeuPQTV
and tRNALeuW fragments recovered from H2O2-
treated S. enterica identified cleavage at posi-
tions U34 and G37 of the anticodon loop (Fig.
2E and fig. S3E).
WT and rtc S. enterica mutants were equally
susceptible to the bacteriostatic activity of H2O2
(fig. S4A); however, compared with WT con-
trols, a higher percentage (P < 0.0001) of DrtcR
and DrtcBA S. enterica mutants were killed by
H2O2 (Fig. 2F and fig. S4B). The DrtcBA mu-
tant became resistant to H2O2 killing upon
complementation with the rtcBA operon (fig.
S4C). These data indicate that the RNA repair
Rtc system protects S. enterica from the bac-
tericidal effects of H2O2. The rtc S. enterica
mutants were hypersusceptible to ROS produced
by the respiratory burst of bone marrow–
derived macrophages (Fig. 2G and fig. S4D).
The DrtcBA S. enterica mutant was also at-
tenuated in C57BL/6 mice, as measured by
both a decreased competitive index when re-
ceiving intraperitoneal (i.p.) inoculation with
WT controls (Fig. 2H) and the reduced burden
and number of microabscesses recorded in
infected livers (Fig. 2I) and spleens (fig. S4E)
when inoculated as a single culture in mice.
The DrtcBA mutant was also attenuated in a
streptomycin-treated oral model of S. enterica
infection (fig. S4F). The DrtcBA S. enterica
strain was as fit as WT S. enterica in Cybb−/−
mice lacking the catalytic gp91phox subunit of
the phagocyte NADPH oxidase (Fig. 2, H and I,
and fig. S4E). Cumulatively, these experiments
show that the RNA repair Rtc system assists
S. enterica to adapt to ROS cytotoxicity in mac-
rophages and mice.
tRNA fragmentation after SOS induction of a
temperate prophage
We sought the molecular mechanism underly-
ing the cleavage of tRNAs in S. enterica after
H2O2 treatment. Gene expression in S. enterica
exposed to H2O2 showed that the SOS re-
sponse was activated (Fig. 3A) (10, 14). Thus, we

102 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

examined tRNA fragmentation in a strain lack-

ing the SOS recombinase RecA that facilitates
DNA repair after oxidative damage. The DrecA
S. enterica mutant did not accumulate 5′-
tRNALeu fragments upon H2O2 treatment
(Fig. 3B), indicating that cleavage of tRNALeu
in S. enterica undergoing oxidative stress is
induced by the SOS response. Transcription
of RNA repair rtc genes was not activated in
the DrecA mutant, probably because of the
lack of tRNA fragments that are recognized by
the RtcR regulatory protein (Fig. 3C). Togeth-
er, these results indicate that the RNA repair
Rtc system provides protective feedback to the
tRNA fragmentation unleashed in response to
ROS-mediated genotoxicity.
The transcriptional analysis showed that
S. enterica undergoing oxidative stress also
increased transcription of Gifsy-1 and Gifsy-3
prophage genes (Fig. 1F). As it appears that
oxidative DNA damage activates prophage
simultaneously with RNA repair (15), we in-
vestigated whether Gifsy prophage activation
caused tRNA cleavage. A strain of S. enterica
lacking all three Gifsy prophages failed to in-
duce tRNALeu cleavage after exposure to H2O2
(Fig. 3D). Endoribonuclease activity was map-
ped to the Gifsy-1 prophage. Gifsy-1–deficient
S. enterica failed to induce tRNALeu cleavage
upon H2O2 treatment (Fig. 3E and fig. S5A)
and sustained higher translation but lower rtcA
transcription after H2O2 treatment than did
WT bacteria (Fig. 3, F and G). Treatment of
Salmonella with 5 mM H2O2 for 2 hours did
not activate the Gifsy-1 lytic cycle as measured
by the absence of plaque-forming particles in
culture supernatants or by RT-PCR of circu- Fig. 3. The SOS response activates prophage-dependent tRNA cleavage in S. enterica undergoing
larized DNA from the prophage (fig. S5B). oxidative stress. (A) Differentially expressed genes (DEGs) in WT S. enterica after treatment with 400 mM
These findings suggest that the expression of H2O2. DEGs were identified by DESeq2 and edgeR with tagwise dispersion and FDR-corrected P values.
the Gifsy-1 tRNase activity is not reliant on (B, D, and E) tRNALeuPQTV fragments were visualized by Northern blotting in log-phase S. enterica after
phage replication. Gifsy-1–deficient S. enterica treatment with 5 mM H2O2. Densitometric quantification of 5′ /intact tRNA is presented below each
were not only more resistant to H2O2 killing lane. (C and F) rtcA transcription in log-phase S. enterica 1 hour after treatment with 400 mM H2O2
than were WT controls (Fig. 3H) but were also or PBS. Cycle threshold values were normalized to the rpoD housekeeping gene. Data are shown as
at a competitive advantage in C57BL/6 mice mean ± SD [(C), n = 4; (F), n = 3]. **P < 0.01 as determined by unpaired Mann-Whitney test. Transcription
when tested in competition with WT S. enterica in (F) is expressed relative to WT controls. (G) Densitometry of the de novo translated proteome as assessed
(Fig. 3I). The presence of the Gifsy-1 prophage by puromycin incorporation in the indicated S. enterica strains grown in MOPS-GLC media and treated
reduced resistance of S. enterica to oxidative with 400 mM H2O2 for 2 hours. Data are shown as mean ± SD (n = 3); *P ≤ 0.05 as determined by Student’s
stress but attenuated S. enterica virulence in t-test. (H) Antimicrobial activity of H2O2 on S. enterica 2 hours after treatment. Data are shown as mean ± SD
Cybb−/− mice lacking the phagocyte NADPH (n = 6); ****P ≤ 0.0001 as determined by one-way ANOVA with Dunnet’s multiple comparison test.
oxidase. Thus, in the context of oxidative stress (I) Competitive index of S. enterica in livers and spleens of C57BL/6 (B6) and Cybb−/− mice 3 days after
created by the respiratory burst of phagocytic i.p. inoculation of equal numbers of WT and Gifsy-1–deficient S. enterica (n = 5 to 8). ***P < 0.001 as
cells, the Gifsy-1 prophage sensitizes S. enterica determined by Mann-Whitney test.
to oxidative killing. However, the Gifsy-1 pro-
phage provides advantages to Salmonella at
times when phagocyte NADPH oxidase is not the STM14_3218-20 operon within the Gifsy-1 gene deletions in the STM14_3218-20 operon—
expressed, as evidenced by the disadvantage of prophage, a locus that is associated with the as efficiently as in the WT 14028s strain (fig. S5,
the DGifsy-1 strain in Cybb−/− mice (Fig. 3I). activation of rtc gene transcription after antibiotic- C and D). Collectively, these findings suggest
mediated DNA damage (15) and that is present that the nuclease responsible for tRNA cleav-
Cleavage of tRNA by Gifsy-1 terminase during in Salmonella Typhimurium strain 14028s but age in H2O2-treated S. enterica maps to a locus
oxidative stress absent in S. Typhimurium strains LT2 or SL1344. encoded outside of the STM14_3218-20 ope-
The data so far indicated the presence of a Northern blot analyses showed that H2O2 in- ron. Neither the dinJ or yafQ genes encoded
previously uncharacterized nuclease in the duced tRNALeu cleavage in S. Typhimurium proximal to the rtc operon were required for
Gifsy-1 prophage that can target its bacterial strains LT2, SL1344—as well as in S. Typhimurium H2O2-induced fragmentation of tRNALeu
host’s tRNA molecules. We initially focused on 14028s mutants bearing multiple- or single- (fig. S5D), indicating that the RNase encoded in

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 103

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

Fig. 4. Gifsy-1 terminase cleaves

tRNA in S. enterica during oxidative
stress. (A) Genomic organization of
Gifsy-1 region in S. enterica 14028s.
(B, D, and E) tRNALeuPQTV fragments
were visualized by Northern blotting
in S. enterica grown to log phase
in LB broth and treated with 5 mM
H2O2. Strains in (E) expressing
pBAD or pBAD-gpA were treated with
500 mM H2O2. (C) Phylogenetic
tree of full-length S. enterica GpA and
known ribonucleases. (F and I) Total
RNA extracted from log-phase
S. enterica was treated with recom-
binant GpA proteins. Recombinant
GpA variants in (I), and where
indicated in (F), were treated with
H2O2. tRNALeuPQTV fragments were
visualized with Northern blot.
Densitometric quantification of 5′
fragment and intact tRNA is
presented below each lane. Single-
letter abbreviations for the amino
acid residues are as follows: A, Ala;
E, Glu; G, Gly; H, His; K, Lys; R,
Arg. In the mutants, other amino
acids were substituted at certain
locations; for example, H432A indi-
cates that histidine at position
432 was replaced with alanine.
(G) Site of cleavage of tRNALeuPQTV
by H2O2-treated recombinant GpA
protein was determined by 3′ RACE
and sequencing. (H) AlphaFold representation of the C-terminal nuclease domain of GpA. Walker A motif is in green, whereas a helices and b sheets of the predicted nuclease
site are in pink and cyan, respectively. Residues mutated in (I) are shown in red. (J) Survival of S. enterica grown overnight in LB broth and treated for 2 hours with
400 mM H2O2 in PBS. Data are shown as the mean ± SD (n = 16 to 26). (K) Competitive index of S. enterica in livers of C57BL/6 (B6) and Cybb−/− mice 3 days after i.p.
inoculation of equal numbers of WT and DgpA S. enterica (n = 5 to 7). **P < 0.01 and ****P ≤ 0.0001 as determined by Mann-Whitney test [(J) and (K)].

this putative toxin-antitoxin system was not
involved.
Systematic mapping of the Gifsy-1 genome
(Fig. 4A) associated a gene in the capsid- and
DNA-packaging region with cleavage of S. enterica
tRNALeu during oxidative stress (Fig. 4B and
fig. S6, A and B). The terminase encoded in the
Gifsy-1 gpA gene is the only determinant in the
capsid- and DNA-packaging region that has
canonical endonuclease activity. Terminases hy-
drolyze linear concatemers of double-stranded
viral DNA into single genomes ready for pack-
aging into mature virions. The combination
of phylogenetic analysis and the AlphaFold
structure of the C-terminal domain of the
Gifsy-1 GpA revealed intriguing similarities be-
tween this terminase and colicin family mem-
bers (Fig. 4C and fig. S6C), some of which are
endoribonucleases with specificity against the
anticodon loop of tRNA molecules (16). Deletion
of the gpA gene from Gifsy-1 averted tRNALeu
fragmentation after exposure of S. enterica to
H2O2 (Fig. 4D), a phenotype that could be
complemented with the gpA gene in trans
(fig. S6D). Cumulatively, these findings sup-
port the role of Gifsy-1 terminase in tRNA
cleavage after oxidative stress.
To further test the idea that the Gifsy-1
terminase has endoribonuclease activity, we
expressed the GpA protein with an arabinose-
inducible promoter. In the absence of H2O2,
induction of the gpA gene in S. enterica did
not result in fragmentation of tRNA (fig. S7A).
However, heterologous expression of gpA si-
multaneously with the addition of 400 mM H2O2
resulted in the fragmentation of tRNALeu (Fig.
4E). These findings suggest that oxidative stress
either promotes endoribonuclease activity in
the GpA terminase or that the tRNA cleavage
is the result of the synergistic actions of GpA
with another factor induced by oxidative stress.
In support of the former model, recombinant
GpA treated with H2O2—but not GpA treated
with dithiothreitol (DTT)—induced cleavage of
tRNALeu in a reconstituted biochemical system
(Fig. 4F and fig. S7B). The tRNase activity of the
oxidized recombinant GpA was potentiated by
the addition of ATP in the reactions (fig. S7C)
and resulted in 3′-phosphate or 2′, 3′ cyclic
phosphate and 5′-hydroxyl fragments amena-
ble for RtcB repair (fig. S7D). The oxidized
GpA protein retained the canonical DNase
activity (fig. S7E). Reduced and oxidized re-
combinant GpA differentially migrated in non-
reducing SDS gels (fig. S7F), indicating that the
Gifsy-1 GpA terminase is redox active. Recom-
binant GpA preparations further purified by
size-exclusion chromatography retained tRNase
activity upon oxidation (fig. S7, G to I), demon-
strating that the tRNase activity maps to GpA.
Recombinant GpA cleaved tRNALeuPQTV at the
G37 wobble position in the anticodon loop (Fig.
4G and fig. S7J). Our bioinformatic analysis
of the projected AlphaFold structure of the
C-terminal domain of GpA revealed a Walker
A motif close to the predicted colicin tRNase
domain (Fig. 4H). Mutagenesis of Lys497 and
His481 in the Walker A motif and predicted
tRNase domain, respectively, prevented cleav-
age of both tRNALeuPQTV and a DNA template
containing the cosN site by the oxidized GpA
protein (Fig. 4I and fig. S7, K to M). Hence, the

104 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE


oxidized form of the redox-sensitive GpA Gifsy-1
terminase can moonlight as a tRNase with spec-
ificity to the anticodon loop without losing its
canonical DNase activity.
A strain of S. enterica lacking the Gifsy-1
gpA gene was hyperresistant to H2O2 and hy-
pervirulent in C57BL/6 mice when tested in
oral and systemic models of infection (Fig. 4,
J and K, and fig. S8A). The expression of gpA
in trans reversed the hypersusceptibility of the
DgpA mutant to H2O2 (fig. S8B). Moreover,
the presence of a functional gpA gene slowed
the recovery of de novo protein translation
after peroxide stress (fig. S8, C and D). These
findings indicate that when acting as an
RNAse, the prophage GpA protein predisposes
S. enterica to oxidative stress generated in the
innate host response.
Discussion
The pathogenesis of Vibrio cholerae, enter-
ohemorrhagic Escherichia coli, Shigella dysenteriae,
or Clostridium botulinum is intimately linked
to the toxins encoded within their temperate
bacteriophages (17). The virulence of S. enterica
has also been chiseled by the horizontal acqui-
sition of prophages. The Gifsy-2 prophage
promotes S. enterica virulence through vacu-
olar remodeling by an effector of the pathoge-
nicity island-2 type III secretion system and
the antioxidant properties of a Cu/Zn super-
oxide dismutase (18, 19). In contrast to the
archetypical lysogenic bacteriophages that
provide virulence traits to their bacterial hosts,
our investigations show that the Gifsy-1 pro-
phage reduces fitness of S. enterica in the
context of ROS-induced oxidative stress pro-
duced by phagocyte NADPH oxidase in mice.
Thus, the redox-responsive terminase encoded
in the capsid- and DNA-packaging region of
the Gifsy-1 prophage gains endoribonuclease
activity upon oxidation, sensitizing S. enterica
to oxidative killing.
ROS damage almost every cogwheel of the
translational machinery, including ribosomes,
elongation factors, mRNA, tRNA, and tRNA
synthetases [reviewed in (20)]. ROS oxidize
cysteine and methionine residues in small
(S17) and large (L7/L12, L14) ribosomal sub-
unit proteins and damage ribonucleotides that
can then be misincorporated into the transcrip-
tome (21–23). Moreover, the [4Fe-4S] cluster of
the ilvD-encoded dihydroxyacid dehydratase
in the biosynthesis of branched-chain amino
acids is especially vulnerable to the toxic ef-
fects of ROS (24–26). Cleavage of tRNALeu iso-
acceptors should be included in the mechanism
by which oxidative stress negatively impacts
translation in S. enterica. In contrast to the
direct damage caused by ROS to ribosomes,
biosynthetic enzymes, and ribonucleotides,
the cleavage of tRNA under oxidative stress
is an indirect result of the SOS-mediated ex-
pression of the Gifsy-1 terminase. Because
SCIENCE science.org
about 20% of all codons in the genome of
S. enterica encode for leucine residues, the
concerted effects of oxidative stress on dehy-
droxyacid dehydratase and the tRNALeu cleav-
age mediated by Gifsy-1 terminase will have a
profound effect on the nascent proteome (fig. S8).
The importance of the RNA repair Rtc sys-
tem in resistance of S. enterica to the phagocyte
NADPH oxidase indicates that this intracellular
pathogen benefits from recycling damaged
tRNA. Given the complex modifications involved
in maturation of tRNAs (27), with most con-
centrated at the wobble position of the anti-
codon stem loop, de novo tRNA generation
during oxidative stress must be taxing to an
already energy- and resource-constrained cell.
Recycling of damaged tRNA molecules be-
comes critical in the resistance of S. enterica
to oxidative stress as shown by the Cybb-
dependent attenuation of rtc mutants.
Why would a pathogen maintain an appar-
ently debilitating locus within its genome?
A possible answer to this paradox might lie
in the fact that prokaryotic cells couple tran-
scription and translation. The induction of
Gifsy-1–mediated tRNA cleavage through the
SOS response in S. enterica suggests that the
observed halts in translation are caused by
double-stranded DNA damage. The concerted
inhibition of leucine biosynthesis and cleav-
age of tRNALeu must slow ribosomes through
the nascent transcript. Decelerating ribosomes
permits backtracking of RNA polymerase
(28), which would enable access of repair en-
zymes to DNA lesions (29). Therefore, Gifsy-1–
mediated inhibition of translation in H2O2-
treated S. enterica may help maintain the in-
tegrity of the genome. In this context, S. enterica
may hijack host cell–derived ROS to activate a
lysogenic phage determinant that halts bacte-
rial translation and growth during intense
periods of oxidative stress, providing oppor-
tunities for repair and survival. In addition,
the moonlighting tRNase function of GpA
precludes phage-genome processing and cap-
sid maturation, protecting the bacterial host
from the lytic cycle at times of exposure to
oxidative stress exerted by the innate immune
response.
RE FERENCES AND NOTES
1. P. Mastroeni et al., J. Exp. Med. 192, 237–248 (2000).
2. F. C. Fang, Nat. Rev. Microbiol. 2, 820–832 (2004).
3. S. Jang, J. A. Imlay, J. Biol. Chem. 282, 929–937
(2007).
4. B. Ezraty, A. Gennaris, F. Barras, J. F. Collet, Nat. Rev. Microbiol.
15, 385–396 (2017).
5. J. S. Kim et al., Proc. Natl. Acad. Sci. U.S.A. 115,
E11780–E11789 (2018).
6. L. E. Leiva et al., Front. Genet. 11, 856 (2020).
7. V. N. Fritsch et al., Free Radic. Biol. Med. 161, 351–364
(2020).
8. C. A. Goodman, T. A. Hornberger, Exerc. Sport Sci. Rev. 41,
107–115 (2013).
9. S. Chakraborty et al., Nat. Commun. 11, 1783 (2020).
10. K. J. Hughes, X. Chen, A. M. Burroughs, L. Aravind, S. L. Wolin,
Cell Rep. 33, 108527 (2020).
RESE ARCH | R E S E A R C H A R T I C L E S
11. U. Das, S. Shuman, RNA 19, 1355–1362 (2013).
12. M. R. Manwar et al., Sci. China Life Sci. 63, 251–258
(2020).
13. H. Temmel et al., Nucleic Acids Res. 45, 4708–4721
(2017).
14. C. A. Juan, J. M. Pérez de la Lastra, F. J. Plou, E. Pérez-Lebeña,
Int. J. Mol. Sci. 22, 4642 (2021).
15. J. E. Kurasz et al., J. Bacteriol. 205, e0026222
(2023).
16. K. Tomita, T. Ogawa, T. Uozumi, K. Watanabe,
H. Masaki, Proc. Natl. Acad. Sci. U.S.A. 97, 8278–8283
(2000).
17. P. L. Wagner, M. K. Waldor, Infect. Immun. 70, 3985–3993
(2002).
18. M. A. De Groote et al., Proc. Natl. Acad. Sci. U.S.A. 94,
13997–14001 (1997).
19. T. D. Ho et al., J. Bacteriol. 184, 5234–5239 (2002).
20. M. Fasnacht, N. Polacek, Front. Mol. Biosci. 8, 671037
(2021).
21. N. Shcherbik, D. G. Pestov, Cells 8, 1379 (2019).
22. C. L. Simms, B. H. Hudson, J. W. Mosior, A. S. Rangwala,
H. S. Zaher, Cell Rep. 9, 1256–1264 (2014).
23. L. I. Leichert et al., Proc. Natl. Acad. Sci. U.S.A. 105, 8197–8202
(2008).
24. D. H. Flint, J. F. Tuminello, M. H. Emptage, J. Biol. Chem. 268,
22369–22376 (1993).
25. L. Macomber, J. A. Imlay, Proc. Natl. Acad. Sci. U.S.A. 106,
8344–8349 (2009).
26. L. Yang et al., Proc. Natl. Acad. Sci. U.S.A. 116, 14368–14373
(2019).
27. V. de Crécy-Lagard, M. Jaroch, Trends Microbiol. 29, 41–53
(2021).
28. S. Proshkin, A. R. Rahmouni, A. Mironov, E. Nudler, Science
328, 504–508 (2010).
29. B. K. Bharati et al., Nature 604, 152–159 (2022).
30. S. Kant et al., PLOS Biol. 21, e3002051 (2023).
31. S. Uppalapati et al., Data from: Prophage terminase with
tRNase activity sensitizes S. enterica to oxidative stress, Dryad
(2024); https://doi.org/10.5061/dryad.xpnvx0knt.
AC KNOWLED GME NTS
We thank J. Jones-Carson for kindly providing the mice. We
thank J. Hesselberth for technical discussions on tRNA Northern
blotting and for helpful discussions with the manuscript. We thank
H. V. Batra for valuable discussions on protein expression and
purification. We also thank A. Margolis for comments on the
manuscript, and we thank the Mass Spectrometry Proteomics
Shared Resource Facility, University of Colorado Anschutz Medical
Campus, for protein identification. Funding: This work was
supported by VA Merit Grants BX0002073 and IK6BX005384 and
by NIH grants R01AI54959 and R01AI136520. M.M. was funded in
part by NIH grant R03AI139557. Author contributions:
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.
Funding acquisition and resources: A.V.T. and M.M. Investigation:
S.U., L.L., S.K., D.O., and A.V.T. Methodology: S.U., L.L., S.K., and
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:
The authors declare that they have no competing interests.
Data and materials availability: All data needed to evaluate
the conclusions in the paper are present in the paper and/or in the
supplementary materials. RNA-seq data are available in the GEO
database under accession no. GSE203342. Bacterial cultures
and plasmids generated in the study are available upon request
to the corresponding author. All the data and unprocessed images
utilized to draw figures in the manuscript are available at Dryad
(31). License information: Copyright © 2024 the authors, some
rights reserved; exclusive licensee American Association for the
Advancement of Science. No claim to original US government
works. https://www.science.org/about/science-licenses-journal-
article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adl3222
Materials and Methods
Figs. S1 to S9
Tables S1 to S7
References (32–39)
MDAR Reproducibility Checklist
Submitted 11 October 2023; accepted 28 February 2024
10.1126/science.adl3222
5 APRIL 2024 • VOL 384 ISSUE 6691 105
RES EARCH | R E S E A R C H A R T I C L E S

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-
naling of protein receptors. Although
we have an advanced understanding
of both protein design and molecular
interactions, the rational design of de novo
proteins that specifically bind small molecules
with low nanomolar to picomolar affinity is
a major challenge (1, 2) that has not been
achieved in de novo proteins (3, 4) without
experimental screening of large libraries of
variants (5–7). Even with the application of
recent advances in artificial intelligence to
facilitate de novo design (8–10), it has been
necessary to screen thousands of independent
designs to discover binders with low-micromolar
to high-nanomolar dissociation constants (Kd)
directly from design algorithms (3, 11–14). Pro-
teins with higher affinity are often desirable.
Given the low success rate, screening large
numbers of designs often relies on biotinylated
or fluorescently labeled versions of their small-
molecule targets, which restricts the region of
the molecule available for binding. The cost of
synthesizing thousands of genes and the ne-
cessity of synthetic chemistry for conjugation
places a practical limitation on the access of
these methods to many groups. Moreover, de
novo design methods rely heavily on struc-
tural informatics to guide sampling of protein
1
Department of Pharmaceutical Chemistry & Cardiovascular
Research Institute, University of California, San Francisco,
CA 94158, USA. 2Helen Diller Family Comprehensive Cancer
Center, University of California, San Francisco, CA 94158,
USA. 3Department of Chemistry, University of California,
Riverside, CA 92521, USA. 4Department of Cellular and
Molecular Pharmacology, University of California, San
Francisco, CA 94158, USA. 5Department of Cancer Biology,
Dana-Farber Cancer Institute, Boston, MA 02215, USA.
6
Department of Biological Chemistry and Molecular
Pharmacology, Harvard Medical School, Boston, MA 02215, USA.
*Corresponding author. Email: nicholasf_polizzi@dfci.harvard.
edu (N.F.P.); william.degrado@ucsf.edu (W.F.D.)
functions that rely on a mix of statistical and
physical terms without explicit representation
of dynamics, conformational entropy, or water
(3–7). This dependence leaves open the funda-
mental question of whether our understand-
ing is grounded in physical forces or limited
exclusively to advanced pattern recognition
(15). In this study, we asked whether adherence
to simple rules based on physical principles
might increase the success rate for designing
drug-binding proteins and whether all-atom
molecular dynamics (MD) simulations with
explicit water might be able to recapitulate
the experimentally determined binding affini-
ties of high-affinity binders, starting with design
models that come directly from computation.
Rationale for design of high-affinity
drug-binding proteins
The recognition of polar groups presents a
challenge in de novo design of binders because
the polar groups must lose most or all of their
highly favorable interactions with water mol-
ecules upon binding to the protein. To com-
pensate for this loss in hydration, the polar
chemical groups of the small molecule must
form highly directional and distance-dependent
hydrogen bonds and electrostatic interactions
with atomic groups in the protein. These polar
interactions are not only required for affinity, but
they also provide the specificity of proteins for
their substrates over other similarly shaped
molecules.
We recently developed a design method
known as Convergent Motifs for Binding Sites
(COMBS) to enable sampling of only sequences
and structures capable of forming such in-
teractions prior to searching for less specific
and directional, but energetically favorable,
van der Waals and hydrophobic interactions
that complete the binding site (3). To facilitate
aromatic, and/or hydrogen-bonded interactions
with the chemical groups of a target small mol-
ecule. The remainder of the sequence is then
filled in through flexible backbone sequence
design.
COMBS showed promise in identifying pro-
tein backbones and creating sites capable of
binding the drug apixaban. In previous work,
we identified one protein that bound with
Kd = 500 nM and a second with Kd = 5 mM
after screening only six designed sequences
(3). In the process, we learned lessons that could
improve the use of COMBS in design pipelines.
Firstly, during the final steps of sequence de-
sign and backbone optimization, sidechains
were sometimes introduced that ultimately
did not form their intended favorable vdMs.
Structural analysis showed that they were not
in optimal orientations to interact with the
drug, and site-directed mutagenesis showed
that they made small or no contributions to
the free energy of binding. We reasoned that
it should be possible to improve binding by
using backbone ϕ- or y-dependent vdMs and
alternating rounds of COMBS with Rosetta
flexible-backbone sequence design. Secondly,
binding affinity can be optimized by pre-
organizing a receptor’s conformation so that it
loses minimal conformational entropy upon
binding. Originally, we analyzed preorganiza-
tion using Rosetta ab initio folding; now we
can more reliably determine whether the se-
quence adopted the desired reorganized struc-
ture using the AlphaFold2 program (17).
Another important feature is the need to
consider the energetically unfavorable loss of
hydrogen bonds between water molecules and
both the drug as well as the protein’s bind-
ing site, which occurs when the drug binds
the pocket. Although COMBS and other algo-
rithms (12) consider the need to form hydrogen

106 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

bonds to compensate for the loss of hydration, H

we strove to form a fuller set of compensatory A O N
Rucaparib O NH2
Veliparib
ligand-protein hydrogen bonds to every buried HN N
polar atom. We sampled vdMs between the O NH2
F N Niraparib N HN
ligand and the first-shell amino acids, as well H N NH
H
Olaparib
as vdMs between the first shell and a second O NH2
N O
Mefuparib NH O
shell of interacting residues, which also assures H
O HN N N N
a favorable geometry for the residues directly O
contacting the ligand. We also opted to bias the F F
H
orientation of the ligand such that formally
charged groups were placed near the surface More similar Less similar
of the protein, thereby minimizing energic
penalties due to Born solvation. We evaluated
MD simulations and free energy calculations,
which explicitly consider interactions with bulk B vdMs
C
and bound water molecules that are not fully H2N O N131
considered in protein design algorithms, to
assess the usefulness of physics-based methods O NH2 CONH2
for evaluating the affinity of the designs.
X Q54
To demonstrate the utility of our re- Y
fined methods, we chose to design binders of Z
HID/HIE
poly(ADP-ribose) polymerase inhibitors (PARPi), H D58
Pharmacophore
a recently developed class of clinically useful O
anticancer drug (18). De novo–designed bind-
ers of PARPi drugs might serve as components Pharmacophore Initial vdM
in detectors, delivery agents, or detoxification selection sampling
agents for these cytotoxic drugs. The predom-
inant class of PARPi drugs share a tripartite
pharmacophore consisting of a fused 5,6-bicyclic E D
core, an amide, and a phenyl group bearing a
positively charged alkylamine (Fig. 1A). We chose
to target rucaparib, the most structurally com- D131
plex of several related drugs, as our primary
F123 D29
target (Fig. 1A), as well as a series PARPi an- N29
alogs. By considering a series of drugs, we at
L24 N131
once provide reagents that might be widely
useful while simultaneously testing our under- Q54 F123
L90 W90
standing of the essential features required for
binding. L24
Q54
De novo design of high-affinity vdM optimization Sequence Design
drug-binding proteins
We used a recursive version of the COMBS Fig. 1. Computational design of PARPi protein binders. (A) PARPi analogs. The shared chemical features
algorithm to design rucaparib-binding sites in are marked in orange. Olaparib is used as a negative control in the design and binding assay. (B to E) The
a family of mathematically generated four-helix overall design strategy. [(B) and (C)] We first defined the pharmacophore and used COMBS to sample
bundle proteins. The key binding residues were vdMs on the selected protein backbones. At the outset of the design, we chose chemical groups that should
introduced by using vdMs to identify multiple form hydrogen bonds when the drug is bound to the binding site. These groups included rucaparib’s
positions on a given protein backbone that indole NH and carboxamide groups (marked in orange). The carboxamide group is present in our vdM library.
could simultaneously interact favorably with However, there were relatively few examples of indole NH vdMs in the database, so we used imidazole as a proxy
the chemical groups of a target small mol- for the indole’s pyrole ring (in the vdMs, the oxygen is marked in red, and nitrogens are marked in blue). We
ecule (Fig. 1, B and C) (materials and methods). then used COMBS to discover sidechains at different positions of a four-helix bundle template that could
Although vdMs are derived from statistics of simultaneously form hydrogen bonds to the indole and carboxamide chemical groups of the drug. (The COMBS
sidechain and mainchain interactions with one algorithm samples vdMs on a protein backbone and then performs superpositions of a ligand onto the chemical
another in the PDB, they can be used to iden- groups of the vdMs; next, COMBS finds all the vdMs with nearby chemical groups to each superposed ligand; and
tify binding-site residues capable of forming finally, COMBS computes a specific combination of vdMs for each ligand that optimizes a score, such as the
hydrogen bonds and aromatic interactions with vdM prevalence or cluster score.) We discovered a solution in which the carboxamide formed bidentate hydrogen
diverse small molecules in much the same way bonds with sidechain of Q54, and the drug’s indole NH interacted with the N131 [in (C), carbon atoms of protein,
that natural proteins bind a wide variety of small green; those of rucaparib, orange]. A second-shell interaction to Q54 that was discovered by COMBS was D58 (carbon
molecules by using a set of 20 amino acids. atoms, brown). (D) We applied flexible backbone sequence design with a custom Rosetta script while fixing the
Although the energetics of these interactions interactions selected from COMBS (rucaparib, purple). (E) We searched for vdMs again based on the design output
can vary depending on the specific small mol- from the previous sequence step (D). The slightly different (~ 1-Å Ca RMSD) backbone now preferred different
ecule bound, the fundamental geometries re- vdMs at some locations (higher cluster scores), and these mutations were made. Three residues at 29, 90,
quired to achieve binding often remain relatively and 131 (deep blue) were changed based on COMBS results. Single-letter abbreviations for the amino acid
constant (19). Thus, a common set of vdMs residues referenced throughout the paper are as follows: L, Leu; N, Asn; D, Asp; Q, Gln; W, Trp; F, Phe.

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 107

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

Fig. 2. Assessing the computational model A B

and experimental binding of PiB to rucaparib. Upper fold PiB Binding Site
(A) The AlphaFold2 model agreed with the designed includes residues
PiB, with the binding site Ca RMSD of 0.41 Å, 12 to 64, 86 to 135 100
the upper fold Ca RMSD of 0.49 Å, and overall 90
Ca RMSD of 0.67 Å. (B) The predicted local distance

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

Emission @ 420nM (104 a.u.)

which the value of Kd was held constant at various
values showed that the data for both proteins 20000 20000
were fit within experimental error so long as the
Kd was <5 nM. Therefore, although the most
10000 10000
probable binding constants were 2 and 0.4 nM,
respectively, we can confidently conclude that the
values for PiB and PiB′ are <5 nM. The titration 0 0
was carried out in buffer containing 50-mM Tris and 0 2 4 6 8 10 0 2 4 6 8 10
100-mM NaCl (pH 7.4). a.u., arbitrary units. [PiB] T/[ruc] T [PiB´] T /[ruc] T

should serve to bind a wide range of com-
pounds. We targeted three chemical groups
in rucaparib’s structure: the indole NH and
the C=O and NH2 groups of its carboxamide
(Fig. 1C). Additionally, COMBS identified Asp58
as a second-shell interaction to the carboxa-
mide of rucaparib (Fig. 1C). It is important to
design binding interactions with these groups
with sub-Ångstrom accuracy to engender spec-
ificity and a favorable free energy of association.
Next, the remainder of the sequence was de-
signed by using Rosetta flexible backbone de-
sign (Fig. 1D) (3, 20) while retaining the identity
of the keystone residues (identified in the
COMBS step). The mainchain moved 1-Å root
mean square deviation (RMSD) during this
step (fig. S1), so a second round of vdM sam-
pling was performed on the relaxed backbone.
This procedure identified three mutants involv-
ing drug-contacting residues, including N29D,
W90L, and N131D (Fig. 1E). A second round of
flexible backbone sequence design with this
backbone and the newly fixed vdMs resulted
in converged sequence-structure combinations
(Figs. 1E and 2, A and B), as a third round of
COMBS showed that the vdMs were now opti-
mal. The final designs included numerous CH-p
and hydrophobic interactions interspersed with
specific polar interactions, including four hydro-
gen bonds (a H-bond donor to the drug’s car-
boxamide oxygen as well as three H-bond
acceptors to the drug’s carboxamide NH, indole
NH, and charged ammonium group), as well as
second-shell interactions. (Fig. 2C and fig. S2).
Throughout the design process, we ensured
that the designs would also retain favorable
interactions with most of the common phar-
macophores of the three other drugs (materials
and methods). However, we predicted that the
protein would have lower affinity for niraparib
and mefuparib because they lack the H-bonding
group indole NH of rucaparib. Additionally, we
expected veliparib to bind weakly because it
lacks a hydrophobic phenyl group and the posi-
tion of its charged ammonium group differs
from that found in the other three drugs.
The final models were chosen based on
multiple criteria: (i) Favorable vdMs (highest
total vdM cluster scores); (ii) satisfaction of all
potential buried H-bond donors and acceptors
in the protein and ligand; (iii) low Rosetta en-
ergy (lowest 50 of the 1000 total designs); and
(iv) avoidance of clashes with the three other
PARP inhibitors, which show structural varia-
bility near the amine end of the molecule. Top-
scoring designs selected for expression were
designated PARPi binder (PiB) and its variant
PiB′ (fig. S3 and table S1). PiB′ differs from PiB
by the substitution of five solvent-exposed
charged residues with alanine to encourage
crystallization. The other two proteins chosen
for expression (PiB-1 and PiB-2) were less closely
related to PiB in structure (RMSD = 0.93 and
0.79 for PiB-1 and PiB-2 to PiB, respectively) and
sequence (41 and 42% identity for PiB-1 and PiB-
2 to PiB, respectively) (fig. S4). Circular dichro-
ism spectroscopy showed that all four had
substantial a-helical character (fig. S5). How-
ever, PiB-1 and PiB-2 failed to induce large
changes in the fluorescence emission spectrum
of rucaparib (fig. S6); therefore, we focused our
efforts on PiB and PiB′ (figs. S7 to S11).

108 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

A 30000 0.03 Kd = 186 ± 6 nM

Absorbance @ 336 nm ( OD)

Kd < 5 nM

Emission @ 420nM (a.u.)

Rucaparib [ruc] T = 100 nM Mefuparib [PiB] T = 2 µM
H 20000 O NH2
O N 0.02
O HN
HN [PiB] T = 1 µM
N
10000 [ruc] T = 50 nM F 0.01
F H
H

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

Absorbance @ 312 nm ( OD)

0.04 0.06
Absorbance @ 317 nm ( OD)

Niraparib [PiB] T = 5 µM Veliparib [PiB] T = 30 µM

0.03
O NH2 O NH2 0.04
N NH 0.02 N
N
[PiB] T = 2 µM N HN 0.02 [PiB] T = 20 µM
H H
0.01

0 0
0 2 4 6 8 0 2 4 6 8
[nir] T / [PiB] T [vel]T / [PiB] T

B Protein : Ligand 0:1 1:1 2.5:1 5:1 10:1

100 100 100 100

Cell viability (%)

Cell viability (%)

Cell viability (%)

Cell viability (%)

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)

C D IC50 (nM) values for the inhibition of cell proliferation

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

0 Veliparib 7427 8461 >104 >104 >104

-1 0 1 2 3 4
Olaparib (log10 nM) Olaparib 298 317 309 897 769

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

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 109

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

returned a Kd of 2.2 nM for PiB and 0.37 nM.
For PiB′, and a sensitivity analysis showed that
the Kd was <5 nM for both proteins (Fig. 2, D
and E). Achieving single-digit nanomolar to
picomolar binding affinity for de novo–designed
proteins has previously required extensive ex-
perimental optimization (5).
Ultraviolet-visible absorption titrations showed
that PiB and PiB′ also bound to the remaining
ligands with affinities that grew increasingly
weaker as the drugs’ structures diverged from
rucaparib (Fig. 3A and fig. S12). PiB retained
submicromolar affinity for mefuparib (Kd = 190
and 350 nM for PiB and PiB′, respectively) and
A B
Gln54
C Open, water-filled binding pocket in drug-free PiB´
Asp131
Ruc-bound PiB´
vs
Design
Gln54
D E
2.0
1.5
(kcal/mol)
niraparib (Kd = 600 and 550 nM). The cor- of cells with certain DNA repair deficiencies
responding Kd values of PiB and PiB′ were including loss-of-function mutations in BRCA2
14 and 24 mM, respectively, for the structurally (18). To determine whether PiB and PiB′ could
divergent drug veliparib, and no binding was attenuate the lethal effects of PARPi drugs, we
detected for the most divergent drug, olaparib measured their effects on the growth of BRCA2-
(fig. S13). This observed trend in binding affin- mutated DLD-1 cells and SUM149 cells (22).
ity matched the order expected from the struc- Dose-response curves were first established in
tural differences mentioned previously. the absence of PiB, then the titration was
We next examined the in vitro stability and repeated with PiB or PiB′ at varying protein/
potency of PiB and PiB′ in serum and cellular drug ratios for each PARPi drug concentration.
assays. PiB and PiB′ were highly stable in human The addition of a single equivalent of PiB or
serum, as are other de novo proteins designed PiB′ resulted in a fourfold increase in the half-
for medical applications (figs. S14 and S15) maximal inhibitory concentration (IC50) value
(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
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 S18), which agreed with spectroscopic data
that indicated that PiB and PiB′ do not bind
this drug.
Structural and mutational validation
of designs
The crystallographic structures of PiB′ were
Asp131 Asp29
Asp29 solved in the absence and presence of the four
active compounds at 1.3- to 1.6-Å resolution
Phe123 (table S2). The protein’s conformation was in ex-
Phe123
Gln54 cellent agreement with the predicted Alpha-
Fold2 model, particularly near the binding
site [a carbon (Ca) RMSD of the 60 sur-
rounding residues was 0.2 to 0.5 Å] (Figs. 2A
Relative binding affinity
4 and 4, A and B; fig. S19; and table. S3). A
of PiB alanine mutants
similar degree of agreement (<0.5 Å RMSD)
3 was observed comparing the experimental
structure and the designed model. An import-

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

in the design of the rucaparib complex (Fig. 4B

D
Q
F1
D

and fig. S20): Asp29 made a direct H-bonded

Fig. 4. The structure of drug-bound PiB′ agrees with the design. (A) The design model agreed well salt bridge to the drug’s charged ammonium
with the rucaparib-bound PiB′ crystal structure, with the binding site Ca RMSD range between 0.38 and group; rucaparib’s carboxamide formed a two-
0.46 Å for the three monomers in the asymmetric unit. (B) The binding site of PiB′. A 2mFo-DFc composite coordinate hydrogen bond with Gln54, which
omit map contoured at 1.6 s. The map was generated from a model that omitted coordinates of rucaparib. in turn was stabilized by a second-shell net-
Overlay of the design (gray) and the structure (protein, orange; rucaparib, pink). The sidechains of the work of H-bonds predicted in the design; and
binding pocket in rucaparib-bound PiB′ agreed with the design. Asp131 interacts with the indole NH through a Asp131 formed a solvent-mediated H-bond to
bridging water as in MD simulations. (C) The structure of drug-free PiB′ shows a preorganized open pocket rucaparib’s indole NH group (Fig. 4B and fig. S20).
filled with multiple waters, which are displaced in the drug-bound conformation. (D) Reversal of the three A search of water-mediated Asp sidechains with
designed substitutions from the vdM optimization procedure led to lower binding affinity (higher Kd) for related indole and imidazole sidechains showed
rucaparib by fluorescence emission titrations. (E) Alanine mutations of the direct binding residues that this bridging interaction is frequently found
decreased binding affinity confirmed by fluorescence emission titrations. in the PDB (fig. S21).

110 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

A and fig. S24B). Each mutation was unfavorable

Gln 54 Asp 29 Asp131
with values of DDG ranging from 1.7 to 3.0 kcal/
carboxamide ammonium indole NH
1.0 mol. These values were within the range ob-
served for substitutions of critical binding re-
sidues in natural protein binding sites (25).
Fraction tracjectory

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

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 111

RES EARCH | R E S E A R C H A R T I C L E S
ligand to estimate the contribution from non-
core regions (fig. S28). Universally, the alche-
mical transfer method tended to overestimate
the binding energy, possibly owing to having
two sets of restraint potentials. However, this
procedure correctly predicted the relative affin-
ities of the four ligands (table S4).
We next used potential of mean force cal-
culations, an orthogonal physical-pathway
methodology (28), to compute absolute binding
energies (figs. S29 to S33), and the results were
in remarkably good agreement with those of
the experiment (Fig. 5, B and C, and table. S5).
The RMS error between the predicted and ex-
perimental values is 1.3 kcal/mol, and the
correct rank order of affinities was observed.
This error was close to the experimental error
in the measurement of Kd for rucaparib. We
also obtained very good agreement between
computation and experimental results for a
set of four mutants of PiB (Fig. 5, B and C,
and table S5). This agreement bodes well for
the use of alchemical and physical pathway–
based binding free energy calculations to
evaluate potential binding energies of de novo
small-molecule binding proteins.
Discussion
In this study, we used structural bioinformatics
to drive the de novo design of a high-affinity
drug-binding protein. Molecular dynamics
simulations accurately captured the trend in
binding affinity for the series of drugs and
predicted the role of water in binding. Our results
portend a future in which artificial intelligence–
enabled sampling of conformational and se-
quence spaces are seamlessly interfaced with
physics-based models to design proteins and
biomimetic polymers with precisely predefined
binding affinity, specificity, and, given the inti-
mate relation between binding and catalysis,
catalytic properties. To achieve binding of
complex polar molecules, sampling favorable
geometries for ligand-protein interactions is
essential. In this work, vdMs and COMBS were
key to this achievement. Although they were
used in conjunction with Rosetta sequence
design, they could be easily adapted to improve
sampling and/or scoring of a variety of recently
developed protein-design algorithms based on
diffusion models (3, 8–10). MD simulations of
de novo proteins have only been occasionally
used to provide insight into de novo protein
design (29–33). However, by using this method,
we were able to differentiate successful versus
unsuccessful designs, suggesting that it should
be helpful for prioritizing designs. Although we
ran simulations for 2 ms, their essential features
could be gleaned after 100 ns, suggesting that
simulations on this timescale should be useful.
Free energy calculations have not previously
been applied to designed proteins. Although
they are more computationally intensive and
require more user-specified parameters, we ob-
112 5 APRIL 2024 • VOL 384 ISSUE 6691
tained excellent quantitative agreement between
computed and experimentally measured binding
free energies using only the designed models
as the input structures. These data demon-
strate the possibility of designing proteins
with high affinity (Kd < 5 nM) to small mol-
ecules by using fully rational criteria for design
and “physics-based” force fields to evaluate the
complexes.
Rucaparib binds to the human PARP1 enzyme
with a Kd ranging from 0.1 to 1 nM depending
on the experimental conditions (23, 24), which
is close to the range observed for PiB and PiB′.
Ligand efficiency is often used as a guiding rule
in drug discovery to determine whether the
affinity of a molecule of a given size is within a
range typically seen in highly optimized small-
molecule drugs and natural organic ligands
for proteins (34, 35). As ligands become larger,
they have more opportunities to form favorable
interactions with their target proteins. Thus,
the maximal affinity possible roughly scales
with the size of a small molecule, and the li-
gand efficiency is defined as the free energy of
binding (1 M standard state) divided by the
number of heavy atoms in the ligand. Most
drugs have ligand efficiency of around 0.3 kcal/
(mol × heavy atom count) (34, 35), although
higher values are observed for highly optimized
drugs such as rucaparib, which has a ligand
efficiency of 0.5 kcal/(mol × heavy atom count).
The ligand efficiency of a drug is similarly a
good measure of how well optimized a de novo
protein is for binding to a small molecule. The
0.5 kcal/(mol × heavy atom count) ligand effi-
ciency of PiB is a considerable improvement
over the 0.21 to 0.26 ligand efficiency of the
first COMBS-designed apixaban binders, which
demonstrates the importance of incorporating
the design principles discussed above. With
these and similar refinements, it should be in-
creasingly possible to design high-affinity small
molecule–binding proteins with predicable bind-
ing energetics for a variety of practical appli-
cations in sensing and pharmaceuticals.
RE FERENCES AND NOTES
1. B. Schreier, C. Stumpp, S. Wiesner, B. Höcker, Proc. Natl. Acad.
Sci. U.S.A. 106, 18491–18496 (2009).
2. W. Yang, L. Lai, Curr. Opin. Struct. Biol. 45, 67–73 (2017).
3. N. F. Polizzi, W. F. DeGrado, Science 369, 1227–1233 (2020).
4. F. Thomas et al., ACS Synth. Biol. 7, 1808–1816 (2018).
5. C. E. Tinberg et al., Nature 501, 212–216 (2013).
6. J. Dou et al., Protein Sci. 26, 2426–2437 (2017).
7. J. Dou et al., Nature 561, 485–491 (2018).
8. J. L. Watson et al., Nature 620, 1089–1100 (2023).
9. J. B. Ingraham et al., Nature 623, 1070–1078 (2023).
10. J. Dauparas et al., Science 378, 49–56 (2022).
11. R. Krishna et al., bioRxiv, 2023.10.09.561603 [Preprint] (2023).
12. G. R. Lee et al., bioRxiv, 2023.11.01.565201 [Preprint]
(2023). http://biorxiv.org/lookup/doi/10.1101/2023.11.01.
565201
13. L. An et al., bioRxiv, 2023.12.20.572602 [Preprint] (2023).
14. J. Dauparas et al., bioRxiv, 2023.12.22.573103 [Preprint] (2023).
15. P. Gainza et al., Nat. Methods 17, 184–192 (2020).
16. M. V. Shapovalov, R. L. Dunbrack Jr., Structure 19, 844–858
(2011).
17. J. Jumper et al., Nature 596, 583–589 (2021).
18. C. J. Lord, A. Ashworth, Science 355, 1152–1158 (2017).
19. A. Koehl, M. Jagota, D. D. Erdmann-Pham, A. Fung, Y. S. Song,
Pac. Symp. Biocomput. 27, 22–33 (2022).
20. S. I. Mann, A. Nayak, G. T. Gassner, M. J. Therien,
W. F. DeGrado, J. Am. Chem. Soc. 143, 252–259 (2021).
21. D.-A. Silva et al., Nature 565, 186–191 (2019).
22. T. Hucl et al., Cancer Res. 68, 5023–5030 (2008).
23. J. Rudolph, G. Roberts, K. Luger, Nat. Commun. 12, 736 (2021).
24. H. D. Thomas et al., Mol. Cancer Ther. 6, 945–956 (2007).
25. A. Fersht, Structure and Mechanism in Protein Science:
A Guide to Enzyme Catalysis and Protein Folding (Macmillan,
1999).
26. L. Wang et al., J. Am. Chem. Soc. 137, 2695–2703 (2015).
27. T.-S. Lee et al., J. Chem. Inf. Model. 60, 5595–5623 (2020).
28. C. Chipot, Annu. Rev. Biophys. 52, 113–138 (2023).
29. E. P. Barros et al., J. Chem. Theory Comput. 15, 5703–5715
(2019).
30. G. Ulas, T. Lemmin, Y. Wu, G. T. Gassner, W. F. DeGrado,
Nat. Chem. 8, 354–359 (2016).
31. M. Gill, M. E. McCully, Protein Eng. Des. Sel. 32, 317–329 (2019).
32. A. Chevalier et al., Nature 550, 74–79 (2017).
33. M. C. Childers, V. Daggett, Mol. Syst. Des. Eng. 2, 9–33
(2017).
34. A. L. Hopkins, G. M. Keserü, P. D. Leeson, D. C. Rees,
C. H. Reynolds, Nat. Rev. Drug Discov. 13, 105–121 (2014).
35. I. D. Kuntz, K. Chen, K. A. Sharp, P. A. Kollman, Proc. Natl.
Acad. Sci. U.S.A. 96, 9997–10002 (1999).
36. L. Lu, Zenodo (2024). https://doi.org/https://doi.org/
10.5281/zenodo.10653015.
AC KNOWLED GME NTS
We thank members of the DeGrado lab for useful discussions and
G. Meigs from Lawrence Berkeley National Laboratory for helping
with data collection. Funding: W.F.D. acknowledges research
support from grants from the National Institutes of Health
(2R35GM122603) and the National Science Foundation (grant NSF
2108660 and 2306190 NSF MCB BSF). N.F.P. acknowledges
support from the NIH (R00GM135519). A.A. acknowledges research
support from SPARC. We collected x-ray data on beamline 8.3.1 of the
Advanced Light Source, which is supported by the DOE Office of
Science User Facility under contract no. DE-AC02-05CH11231.
Beamline 8.3.1 is also supported in part by the ALS-ENABLE program of
the National Institutes of Health, National Institute of General Medical
Sciences grant P30 GM124169-01. Author contributions: L.L. and
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.,
X.G., S.I.M., X.Z., and H.J. performed the experiments; L.L., H.Y.,
X.G., and Y.W. performed the data analysis; N.F.P., S.K.T., D.G.,
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
W.F.D. wrote the paper with input from all authors. Competing
interests: A.A. is a cofounder of Tango Therapeutics, Azkarra
Therapeutics, Ovibio Corporation, and Kytarro; a member of the
board of Cytomx and Cambridge Science Corporation; a member
of the scientific advisory boards of Genentech, GLAdiator, Circle,
Bluestar, Earli, Ambagon, Phoenix Molecular Designs, Yingli, ProRavel,
Oric, Hap10, and Trial Library; a consultant for SPARC, ProLynx,
Novartis, and GSK; and holds patents on the use of PARPi held
jointly with AstraZeneca from which he has benefited financially (and
may do so in the future). Data and materials availability:
Coordinates and structure files have been deposited to the PDB
with accession codes: 8TN1 (drug-free PiB′), 8TN6 (rucaparib-bound
PiB′), 8TNB (mefuparib-bound PiB), 8TNC (niraparib-bound PiB′),
and 8TND (veliparib-bound PiB′). Computational code and design
scripts are available in the supplementary materials and at Zenodo
(36). All other relevant data are available in the main text or the
supplementary materials. License information: Copyright © 2024
the authors, some rights reserved; exclusive licensee American
Association for the Advancement of Science. No claim to original
US government works. https://www.science.org/about/science-
licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adl5364
Materials and Methods
Supplementary Text
Figs. S1 to S33
Tables S1 to S5
References (37–93)
MDAR Reproducibility Checklist
Data S1
Submitted 24 October 2023; resubmitted 23 December 2023
Accepted 28 February 2024
10.1126/science.adl5364
science.org SCIENCE
 RESE ARCH | R E S E A R C H A R T I C L E S

ORGANIC CHEMISTRY tions (15). The Baran group demonstrated Fe-

catalyzed decarboxylative Negishi couplings
Carbon quaternization of redox active esters and of RAEs and various aryl-organometallic re-
agents (16). Given these results, it seemed
olefins by decarboxylative coupling reasonable that RAEs might quaternize radi-
cals through catalytic cycles in which Fe(II)
Xu-cheng Gan1†, Benxiang Zhang1†, Nathan Dao1†, Cheng Bi1, Maithili Pokle1, Liyan Kan1,2, porphyrins both form and trap open-shell C
Michael R. Collins3, Chet C. Tyrol4, Philippe N. Bolduc5, Michael Nicastri5, Yu Kawamata1*, intermediates generated from olefins or other
Phil S. Baran1*, Ryan Shenvi1* RAEs. These two related catalytic cycles would
rely on overall reducing conditions (Zn or
The synthesis of quaternary carbons often requires numerous steps and complex conditions silane) to produce a pool of Fe(II) that might
or harsh reagents that act on heavily engineered substrates. This is largely a consequence cleave N-hydroxy-phthalimide esters to gener-
of conventional polar-bond retrosynthetic disconnections that in turn require multiple ate primary and/or tertiary radicals. Because
functional group interconversions, redox manipulations, and protecting group chemistry. of the instability of tert-alkyliron complexes
Here, we report a simple catalyst and reductant combination that converts two types above 0°C (17), capture of the primary radical
of feedstock chemicals, carboxylic acids and olefins, into tetrasubstituted carbons through would be preferred, allowing its interception
quaternization of radical intermediates. An iron porphyrin catalyst activates each substrate by the tertiary radical in a putative SH2 reac-
by electron transfer or hydrogen atom transfer, and then combines the fragments using tion (14). This same tertiary radical may be
a bimolecular homolytic substitution (S H 2) reaction. This cross-coupling reduces the formed from an alkene and an iron hydride
synthetic burden to procure numerous quaternary carbon–containing products from simple that can regenerate Fe(II) through MHAT or
chemical feedstocks. a hydrogen evolution reaction (HER) (18). In
contrast to many MHAT reactions, this cat-
alytic cycle would not require turnover by an
itrogen quaternization, a polar bond– (8, 9). Here, we show how a single catalyst exogenous oxidant; the RAE partners would

N
forming reaction, originated in the 19th
century as one of the earliest examples
of C–N bond formation in organic syn-
thesis (Fig. 1A) (1). Its inherent simplicity
and the widespread availability of the necessary
starting materials stands in stark contrast to
the way chemists go about forging quaternary
C’s (2). Specifically, most C quaternizations
entail polar transformations such as conjugate
addition, enolate chemistry, and pericyclic re-
actions (3–5). Such processes are complicated
by the differing regioisomers that can result
depending on the substrates and conditions
used. For instance, conjugate additions can
often lead to undesired 1,2-addition products,
enolate C alkylation often competes with O
alkylation, and pericyclic processes can suffer
from regioisomeric mixtures depending on the
innate sterics and electronics of the reactants. A
simple and predictable method to form quater-
nary C’s akin to classic N quaternization would
be an appealing transformation. Our previous
studies have demonstrated how radical meth-
ods can be leveraged to cross-couple olefins
with other olefins or alkyl/aryl halides (6, 7).
In many cases, quaternary C synthesis can be
achieved through such methods. The cross-
coupling between two different redox active
esters has also been reported but is not gen-
erally applicable to quaternary C synthesis
1
Department of Chemistry, Scripps Research, La Jolla, CA
92037, USA. 2College of Chemistry and Molecular
Engineering, Peking University, Beijing 100871, China.
3 rous phorphyrin-like behavior (13). Shenvi and
Oncology Medicinal Chemistry Department, Pfizer
Pharmaceuticals, San Diego, CA 92122, USA. 4Pfizer
Medicine Design, Groton, CT 06340, USA. 5Biogen Inc.,
Cambridge, MA 02142, USA.
*Corresponding author. Email: yukawama@scripps.edu (Y.K.);
pbaran@scripps.edu (P.S.B.); rshenvi@scripps.edu (R.S.)
†These authors contributed equally to this work.
can mediate regioselective quaternization of C
using ubiquitous chemical feedstocks: olefins
and carboxylic acids (Fig. 1B). An inexpensive
Fe-based catalyst in the presence of simple
reducing agents can effectively achieve this
powerful transformation in a chemoselective
fashion with broad substrate scope through a
similar mechanistic pathway. It is particularly
notable that a single catalyst can both activate
distinct functional groups [redox active esters
(RAEs) and olefins] and mediate bond forma-
tion between them. An illustrative demonstra-
tion of the power of such a reaction is depicted
in Fig. 1C, in which the 10-step synthesis of a
seemingly trivial structure (4) (10) can be trun-
cated into a single step by coupling a primary
tosylate-containing RAE 3 with either RAE
1 or olefin 2 with high chemoselectivity in
47% and 42% yield, respectively.
The mechanistic framework outlined in Fig.
1B was inspired by key literature precedents
outlined in Fig. 2A. Seminal studies by Hirobe
and co-workers showed that Drago-Mukaiyama-
type olefin hydration could be achieved using
Fe porphyrins (11). Indeed, the pioneering
studies of Setsune et al. identified sec-alkyl iron
porphyrins formed upon exposure of 1-alkenes
to iron(tetraphenylporphyrinato) chloride [Fe
(TPP)Cl] and NaBH4 (12). Fe–H formation and
metal-hydride hydrogen atom transfer (MHAT)
explain the branched-selective hydrometalation
(12). Brault and Neta depicted alkyl-Fe3+(TPP)
complexes as mesomers with single-electron
density localized on C, which explained its fer-
co-workers recently used this same iron com-
plex, along with a silane, to catalyze C–C bond
formation (14) between benzyl bromides and
olefins to form quaternary centers through
bimolecular homolytic substitution (SH2) reac-
reoxidize Fe(II) to (III). Below, we identify
conditions to reduce this design to practice
and achieve a versatile C quaternization that
unites ubiquitous bench-stable starting mate-
rials (olefins and acids) in a simple and easily
scalable fashion.
Reaction development
The realization of this plan is summarized in
Fig. 1B using RAE 5 with either olefin 7 or
RAE 8 leading to the same product 6. As il-
lustrated by entries 1–8,15 (olefin-RAE coupling)
and 9–14 (RAE-RAE coupling), each reaction
variant required simple and similar condi-
tions: the Fe(TPP)Cl complex, a base, a reduc-
tant and a 1,2-dichloroethane (DCE)/acetone
solvent mixture. The choice of base proved
crucial as exemplified in entries 2 and 3, with
CsOAc emerging as optimal. Increased yields
were observed after a simple degassing (bub-
bling Ar or N2) of the reaction mixture, although
the presence of air was not detrimental (entry 4).
Exclusion of acetone led to lower yields, pre-
sumably due to catalyst insolubility (entry 5).
Addition of other Fe sources such as Fe(acac)3
led to diminished conversion (entry 6) as well.
As observed in other MHAT transformations,
the use of preformed Ruben’s silane proved
beneficial (entries 7 and 8). For RAE-RAE cou-
pling, the optimized conditions above only led
to 9% yield of the desired adduct 6 (entry 9).
Increasing the Fe loading to 20% had a mea-
surable positive effect on conversion (entry 10).
The greatest improvement emerged when
switching the reductant from a silane to Zn
metal (entry 11). A slight change in solvent
composition to DCE/acetone 7:4 along with
KOAc in place of CsOAc led to a further in-
crease in yield (entries 12 and 13). The inclu-
sion of a weak acid (Et3N•HCl) likely served

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 113

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

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
yield of 6 (entry 14). The final optimized con-
ditions for RAE-RAE coupling could not be
applied to the closely related olefin-RAE coupling
(entry 15) and vice versa (entry 9). Neverthe-
less, both conditions used a simple combina-
tion of commercially available reagents and
the same Fe(TPP)Cl catalyst.
Neither Fe-catalyzed decarboxylative cou-
pling (conditions A or B) necessitated O2 to
turn over the catalytic cycle (Fe2+ to Fe3+),
whereas recent work from our group required
an air atmosphere in conjunction with an Fe
b-diketonate MHAT catalyst or co-catalyst
(15, 17). In the present work, the single iron(II)
porphyrin turned over to iron(III) in the ab-
sence of oxygen by reaction with the RAE or its
resultant radical: 1H NMR studies demonstrated
that Fe(TPP) will complex n-pentyl-CO(NHPI)
before the formation of n-pentyl–Fe(TPP) (see
the supplementary materials). Fe(TPP) can
be formed from its corresponding hydride by
either hydrogen evolution or alkene MHAT
(see the supplementary materials) (18). As a
result, Fe(TPP)Cl mimics the polyfunctional-
ity of precious metals in inner-sphere (coor-
dinative) alkene functionalization catalysis,
but appears to do so in an outer-sphere (non-
coordinative) series of elementary steps. In the
case of RAE-RAE coupling, the use of Zn metal
is well known to reductively decompose NHPI
esters into C-centered radicals.
The different optimal bases in conditions
A and B were determined empirically and
likely fulfilled different roles: Silanes can re-
quire Lewis base addition to increase hydridic
strength (19) and/or accelerate metal hydride
formation (20) en route to MHAT with alkenes,
and bases can presumably sequester Lewis
acidic metals, e.g., ZnCl2, that might accrue as
an RAE is reduced by Zn(0).
Substrate scope exploration
With a streamlined set of conditions in hand
for the radical quaternization of olefins and
carboxylates, the scope was examined as illus-
trated in Fig. 3. In general, these highly chemo-
selective conditions are compatible with a range
of functional groups, including carbamates,
amides, alkyl halides, epoxides, aryl halides, es-
ters, alcohols, nitriles, tertiary amines, ketones,

114 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

Fig. 2. Catalytic cycle

design and execu-
tion. (A) Proposed
catalytic cycles in
common between RAE
and olefin coupling
parners based on
literature precedents.
(B) Identified similarities
and differences between
coupling partners from
optimization studies.
*20 mol% Fe(TPP)Cl;
†DCE/acetone = 7/4.

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

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 115

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

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.
Stabilized radicals derived from a-heteroatom
or a-aryl carboxylic acids can also be coupled.
In the case of RAE-RAE couplings of such sub-
strates, FeOEP led to a substantially increased
yield. Complex pharmaceutically relevant
structures can be coupled efficiently, resulting
in quaternary C-containing analogs 36 to 40.
The direct installation of quaternary centers
within saturated heterocycles is a particularly
challenging motif to construct by conventional
polar bond analysis; this quaternization al-
lows an intuitive coupling from readily acces-
sible heterocyclic fragments (10, 11, 12, 15,
and 16). The reactions described above are
extremely simple and practical to perform
using inexpensive reagents, include a rapid
setup time, and prove amenable to gram-scale
preparation (see 41 and 34). The absence of
oxygen in conditions A/B allow facile scal-
ing to gram quantities with no change in yield
(Schlenk or glovebox conditions are not re-
quired, just a simple Ar balloon) and point to
internal oxidative turnover, i.e., an overall redox
neutral process (Fig. 2A). These features bode
well for adoption in a parallel or combinatorial
synthesis campaign in which ubiquitous, bench-
stable precursors can combine to deliver long-
sought-after complex libraries rich in sp3
content, including quaternary C’s.
Simplification of quaternary C synthesis
The capacity for these new reactions to
markedly simplify quaternary C–containing
molecules has been exemplified through 10
different case studies (see the supplementary

116 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE


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
six examples are illustrated in Fig. 4. The
common theme among the prior syntheses
is a nearly exclusive reliance on polar bond
analysis during the planning stages, espe-
cially carbonyl chemistry. Consequently, the
SCIENCE science.org
conventional routes to these mostly simple
structures rely on extensive redox manipu-
lations, functional group interconversions,
and a variety of pyrophoric reagents for var-
ious carbonyl reductions. By contrast, direct
C quaternization through radical interme-
RESE ARCH | R E S E A R C H A R T I C L E S
diates sidesteps these inefficient tactics
through modular LEGO-like transformations
of commercially available building blocks
(olefins and acids).
For example, the simple alkyne 53 was
previously prepared (en route to prostaglandin
5 APRIL 2024 • VOL 384 ISSUE 6691 117
RES EARCH | R E S E A R C H A R T I C L E S
analog synthesis) in an eight-step route, of
which only one step forged a C–C bond and
the key quaternary C was purchased in the
form of a dimethyl cyclohexanone 52 with a
skeleton that was tediously edited (21). By
contrast, commercially available olefin 54
was coupled with the RAE derived from the
simple commercial alkyne-containing acid
(55) under radical quaternization conditions
to deliver 53 in only two steps (48% isolated
yield). Ketoaldehyde 56, a useful building block
for steroid analog synthesis, was previously
accessed by classic carbonyl chemistry, C–C
homologation, and a variety of undesirable
reagents (e.g., Br2, Mg, and OsO4) (22). Alter-
natively, the same structure was accessed
directly after radical quaternization of olefin
57 with RAE 58, followed by acidic workup in
45% isolated yield. Substrate 60, containing
two quaternary centers, was previously accessed
through a laborious 11-step sequence with low
ideality commencing from glutaric anhydride
59, again relying on polar bond analysis and
carbonyl chemistry (23–25). Diene 61 could be
enlisted in a sequential radical cross-coupling
first by using olefin-olefin Fe-catalyzed MHAT
cross-coupling with methylacrylate (72% yield),
followed by radical quaternization of 62 with
RAE 63 to deliver the same product, 60, in
58% isolated yield, thus obviating the need
for nine inconvenient steps and toxic and/or
pyrophoric reagents (e.g., LiAlH4, HBr, and
KCN). Enyne 65, used as a substrate in a cyclo-
isomerization study, was previously accessed
through a 15-step sequence relying on alkyne
hydrometallation, Wittig, carbonyl chem-
istry, and acetylide addition as the key C–C
bond-forming steps (26). This lengthy sequence
required the use of protecting groups and ex-
tensive redox manipulations (three instances
of LiAlH4 reduction). A more intuitive “LEGO”-
like approach was enabled through radical
retrosynthesis. Thus, an Ag-Ni-facilitated de-
carboxylative alkenylation of commercial acid
66 with vinyl iodide 67, followed by radical
quaternization with alkyne-containing RAE
68 afforded 65 in only five steps without any
redox manipulations. Radical quaternization
could also be used to construct scaffolds ap-
pearing in natural products. For example,
piperidine 70, featuring a quaternary C at the
C-3 position, serves as a pivotal scaffold with-
in the madangamine alkaloids. In the previ-
ous seven-step synthesis, the construction of
the quaternary center again relied on carbonyl
chemistry commencing from 2-piperidone (27).
Of those seven steps, two forged C–C bonds
with the remainder comprising concession
steps (redox, FGIs, and PG manipulations).
Conversely, starting from commercial acid
71, a simple alkylation, followed by radical
quaternization with free alcohol containing
RAE 70, enables a three-step synthesis. Fi-
nally, a-tocopherol analog 74, which is useful
118 5 APRIL 2024 • VOL 384 ISSUE 6691
for the modulation of microglial activation,
required a 10-step route reliant on classic
polar bond analysis (28). In a complete de-
parture from this strategy, commercially
available Trolox (75) could be used to effici-
ently incorporate the desired alkanol side
chain through radical quaternization (58%
yield) with RAE 72 to deliver analog 74 in
only three steps.
Quaternary C’s can now be dissected into
feedstock precursors, alkenes, and carboxylic
acids through radical quaternization. Num-
erous high Fsp3 materials are now available
with a fraction of the prior synthetic burden.
Indeed, the approach outlined herein repre-
sents a complete departure from the most-
ly carbonyl-focused retrosynthetic analysis
(Table 2) historically used to forge such
systems (3). In this simple protocol, two sub-
strate classes converge into the same catalytic
cycle in which an Fe(III) catalyst undergoes
reduction to Fe(II) by either Zn or a silane/
alkene combination, and oxidative capture by
a redox-active ester returns the Fe(III). The
RAE cross-coupling exhibits perfect hetero-
selectivity. By contrast, cross-couplings that
involve weak alkene-metal center complexes
to form sp3-sp3 linkages (e.g., Heck, Kulinkovich,
1,4-addition) often do not allow chemoselec-
tive merger of complex fragments. It has been
widely recognized that fragment coupling
simplifies retrosynthetic analysis by increas-
ing convergency (29, 30). Simple tools such
as the reactions described herein point to a
distinctive conceptual platform (31–35) to
decisively solve this vexing problem in or-
ganic synthesis.
RE FERENCES AND NOTES
1. N. Menschutkin, Z. Phys. Chem. 5, 589–600
(1890).
2. T. T. Talele, J. Med. Chem. 63, 13291–13315
(2020).
3. E. J. Corey, X.-M. Cheng, The Logic of Chemical Synthesis
(Wiley, 1995).
4. S. F. Martin, Tetrahedron 36, 419–460 (1980).
5. S. Kotha, N. R. Panguluri, R. Ali, Eur. J. Org. Chem. 2017,
5316–5342 (2017).
6. S. Kotesova, R. A. Shenvi, Acc. Chem. Res. 56, 3089–3098
(2023).
7. J. M. Smith, S. J. Harwood, P. S. Baran, Acc. Chem. Res. 51,
1807–1817 (2018).
8. B. Zhang et al., Nature 606, 313–318 (2022).
9. B. Zhang et al., Nature 623, 745–751 (2023).
10. F. Nagatsugi, F. Inoue, S. Sasaki, M. Maeda, Chem. Pharm. Bull.
43, 607–615 (1995).
11. T. Santa, T. Mori, M. Hirobe, Chem. Pharm. Bull. (Tokyo) 33,
2175–2178 (1985).
12. J. Setsune, Y. Ishimaru, A. Sera, Chem. Lett. 21, 377–380
(1992).
13. D. Brault, P. Neta, J. Am. Chem. Soc. 103, 2705–2710
(1981).
14. W. Liu, M. N. Lavagnino, C. A. Gould, J. Alcázar,
D. W. C. MacMillan, Science 374, 1258–1263
(2021).
15. X. C. Gan et al., J. Am. Chem. Soc. 145, 15714–15720
(2023).
16. F. Toriyama et al., J. Am. Chem. Soc. 138, 11132–11135
(2016).
17. A. L. Balch, R. L. Hart, L. Latos-Grazynski,
T. G. Traylor, J. Am. Chem. Soc. 112, 7382–7388
(1990).
18. S. W. M. Crossley, C. Obradors, R. M. Martinez, R. A. Shenvi,
Chem. Rev. 116, 8912–9000 (2016).
19. C. R. Howie, J. K. Lee, R. L. Schowen, J. Am. Chem. Soc. 95,
5286–5288 (1973).
20. D. Kim, S. M. W. Rahaman, B. Q. Mercado, R. Poli,
P. L. Holland, J. Am. Chem. Soc. 141, 7473–7485
(2019).
21. L. Heslinga, R. van Der Linde, H. J. J. Pabon,
D. A. van Dorp, Recl. Trav. Chim. Pays Bas 94, 262–273
(1975).
22. J. Lu et al., Synthesis (Stuttg.) 51, 3964–3972
(2019).
23. J. Walker, J. K. Wood, J. Chem. Soc. Trans. 89, 598–604
(1906).
24. J. Sicher, M. Svoboda, J. Závada, Collect. Czech. Chem. Commun.
30, 421–437 (1965).
25. J. Sicher, M. Svoboda, J. Závada, R. B. Turner, P. Goebel,
Tetrahedron 22, 659–671 (1966).
26. C. Blaszykowski et al., Org. Lett. 6, 3771–3774
(2004).
27. S. Proto et al., Org. Lett. 14, 3916–3919 (2012).
28. T. Muller, D. Coowar, M. Hanbali, P. Heuschling,
B. Luu, Tetrahedron 62, 12025–12040
(2006).
29. S. H. Bertz, J. Am. Chem. Soc. 104, 5801–5803
(1982).
30. M. Tomanik, I. T. Hsu, S. B. Herzon, Angew. Chem.
Int. Ed. 60, 1116–1150 (2021).
31. J. R. Coombs, J. P. Morken, Angew. Chem. Int. Ed. 55,
2636–2649 (2016).
32. L. T. Kliman, S. N. Mlynarski, G. E. Ferris, J. P. Morken,
Angew. Chem. Int. Ed. 51, 521–524 (2012).
33. A. W. Hird, A. H. Hoveyda, J. Am. Chem. Soc. 127,
14988–14989 (2005).
34. T. S. Mei, H. H. Patel, M. S. Sigman, Nature 508, 340–344
(2014).
35. G. Pratsch, G. L. Lackner, L. E. Overman, J. Org. Chem. 80,
6025–6036 (2015).
AC KNOWLED GME NTS
We gratefully acknowledge J. S. Chen and the Scripps
Automated Synthesis facility (ASF) for help with analysis,
L. Pasternack for assistance with NMR spectroscopy,
and Y. Wang and J. Sun for helpful discussions and early
contributions. Funding: This work was supported by the
National Institutes of Health (grant GM122606 to R.A.S. and
grant GM118176 to P.S.B.), the National Science Foundation
(grant CHE1955922 to R.A.S.), Nanjing King-Pharm Co.,
Ltd. (X.G.), Pfizer (P.S.B. and Y.K.), and Biogen (P.S.B. and
Y.K.). Author contributions: Conceptualization: X.G., B.Z.,
N.D., Y.K., P.S.B., R.S.; Funding acquisition: Y.K., P.S.B.,
R.S.; Investigation: X.G., B.Z., N.D., C.B., M.P., L.K., M.R.C.,
C.C.T., P.N.B., M.N., Y.K.; Methodology: X.G., B.Z., N.D.,
Y.K., P.S.B., R.S.; Project administration: Y.K., P.S.B., R.S.;
Supervision: Y.K., P.S.B., R.S.; Writing – original draft: N.D.,
Y.K., P.S.B., R.S.; Writing – review and editing: N.D., Y.K.,
P.S.B., R.S. Competing interests: The authors declare no
competing interests. Data and materials availability: All data
are available in the main text or the supplementary materials.
License information: Copyright © 2024 the authors, some
rights reserved; exclusive licensee American Association for the
Advancement of Science. No claim to original US government
works. https://www.science.org/about/science-licenses-
journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adn5619
Materials and Methods
Figs. S1 to S6
NMR Spectra
References (36–73)
Submitted 15 December 2023; accepted 28 February 2024
10.1126/science.adn5619
science.org SCIENCE
 RESE ARCH | R E S E A R C H A R T I C L E S

CELL BIOLOGY fig. S1B). Fluorescently labeled recombinant

X. laevis cohesin (fig. S1C) loaded onto chro-
Sister chromatid cohesion establishment during matin in extracts in a manner dependent on
DNA being licensed with pre-RCs and at com-
DNA replication termination parable levels to endogenous cohesin (fig. S1D)
(5, 6). Cohesin labeled with Janelia Fluor 646
George Cameron1, Dominika T. Gruszka1†, Rhian Gruar2, Sherry Xie1, Çağla Kaya1, Kim A. Nasmyth2, (JF646-cohesin) was loaded onto surface-tethered
Jonathan Baxter3, Madhusudhan Srinivasan2*, Hasan Yardimci1* l DNAs. Replication fork progression was fol-
lowed in extracts by observing nascent DNA
Newly copied sister chromatids are tethered together by the cohesin complex, but how sister with fluorescently tagged Fen1 (Fen1-mKikGR).
chromatid cohesion coordinates with DNA replication is poorly understood. Prevailing models To monitor the encounter of individual repli-
suggest that cohesin complexes, bound to DNA before replication, remain behind the advancing cation forks with preloaded cohesin, origin firing
replication fork to keep sister chromatids together. By visualizing single replication forks colliding was restricted using p27kip, a cyclin-dependent
with preloaded cohesin complexes, we find that the replisome instead pushes cohesin to where kinase inhibitor (fig. S1B).
a converging replisome is met. Whereas the converging replisomes are removed during DNA Prevailing models for cohesin conversion pre-
replication termination, cohesin remains on nascent DNA and provides cohesion. Additionally, we dict incorporation of preloaded cohesin into
show that CMG (CDC45–MCM2-7–GINS) helicase disassembly during replication termination is vital for nascent DNA behind the replication fork. After
proper cohesion in budding yeast. Together, our results support a model wherein sister chromatid replisome collision with cohesin complexes,
cohesion is established during DNA replication termination. we observed four different outcomes: cohesin
remaining behind the replication fork (trans-
fer), cohesin moving with the replication fork
he cohesin complex tethers sister chro- generated by these two pathways are unclear. (sliding), cohesin removal, and replication fork

T matids together from the moment they

are generated in S phase until their sep-
aration in anaphase. This fundamental
phenomenon, called sister chromatid co-
hesion, underpins orderly chromosome segre-
gation. Cohesin is a ring-shaped complex with
four core subunits (SMC1, SMC3, RAD21Scc1,
and SA1/SA2Scc3) (1). In addition to its essen-
tial role in sister chromatid cohesion, cohesin
organizes interphase chromosomes by loop
extrusion (2, 3). Cohesin is loaded onto chro-
matin by the NIPBL/MAU2 (Scc2/4) loader
(4), which in vertebrates interacts with pre-
replication complexes (pre-RCs) (5, 6). In eu-
karyotes, origins of replication are licensed
during G1 phase through the formation of
pre-RCs, which contain inactive double hex-
amers of minichromosome maintenance pro-
teins 2–7 (MCM2-7) (7). In S phase, pre-RCs are
remodeled to form CDC45–MCM2-7–GINS
(CMG) helicases, which then unwind DNA. The
replisome complex, containing additional com-
ponents, is assembled around the CMG helicase
(8, 9). Cohesion is thought to arise from coen-
trapment of sister DNAs within cohesin rings
(1, 10). Cohesion establishment is strictly limited
to S phase (11) and is believed to be mechanis-
tically coupled to DNA replication, involving
two independent pathways (12, 13). A “conver-
sion” pathway uses preloaded cohesin com-
plexes associated with parental DNA ahead
of the replication forks to form cohesive struc-
tures, whereas a “de novo” pathway uses cohesin
rings loaded onto DNAs during S phase. The
molecular mechanisms by which cohesion is
1
The Francis Crick Institute, London, UK. 2Department of
Biochemistry, University of Oxford, Oxford, UK. 3Genome
Damage and Stability Centre, University of Sussex,
Brighton, UK.
*Corresponding author. Email: madhusudhan.srinivasan@bioch.
ox.ac.uk (M.S.); hasan.yardimci@crick.ac.uk (H.Y.)
†Present address: Department of Physics and Kavli Institute for
Nanoscience Discovery, University of Oxford, Oxford, UK.
This constitutes a major gap in our understand-
ing of eukaryotic biology.
Two types of mechanisms have been envis-
aged for conversion. Cohesion could be gener-
ated by the passage of the replisome through
cohesin rings that had previously entrapped
unreplicated DNA (fig. S1A, left). Alternatively,
cohesin rings could be transferred from un-
replicated to replicated DNAs while tran-
siently associating with the replisome (fig. S1A,
right). Both these scenarios predict that DNA-
associated cohesin rings, upon encounter with
the replisome, would remain behind the advanc-
ing replication forks. Testing this prediction
in vivo has been challenging because cohesin
is constantly mobile on DNA (14), new cohesin
is loaded onto DNA during most of the cell cy-
cle (15), and cohesin–replisome encounters are
stochastic owing to the nature of origin firing
in eukaryotes (7). To determine the fate of pre-
loaded cohesin during DNA replication, we per-
formed live visualization of single replication
forks encountering cohesin complexes.
Results
Preloaded cohesins are pushed by replisomes
To study the outcome of replisome collision with
cohesin in a physiologically relevant setting, we
used Xenopus laevis egg extracts (16–18), which
contain all factors needed for in vitro DNA
replication and repair (19) while supporting
cohesion establishment (20). Two different ex-
tracts allowed a single round of DNA replica-
tion to be performed. High-speed supernatant
(HSS) was used to license DNA with pre-RCs in
a sequence-independent manner, while nucleo-
plasmic extract (NPE) was used to replicate
DNA. We used an assay developed to visualize
fluorescent molecules during the replication
of surface-immobilized DNA in egg extracts
(17, 18, 21) to assess the outcomes of replica-
tion fork encounters with cohesin (Fig. 1A and
stalling (Fig. 1A). Unexpectedly, in our condi-
tions, cohesin transfer accounted for only 5%
of all events (fig. S2A). In most cases, we instead
observed that preloaded cohesin was pushed
ahead of forks (Fig. 1, B and C, and fig. S2B;
57 to 66% of events). In 18 to 20% of events,
cohesin was removed shortly after encounter-
ing the replication fork (fig. S2C). And in 9 to
20% of events, replication fork stalling was de-
tected upon encounter with preloaded cohesin
(fig. S2D). The lack of cohesin transfer was sur-
prising, as extracts contain all factors needed
for cohesion establishment by the conversion
pathway. We also observed a notable disso-
ciation of cohesin in extracts in a replication-
independent manner, likely due to competition
for DNA binding between cohesin and the many
other DNA binding factors in extracts and the
presence of cohesin unloader (fig. S3).
We next investigated whether endogenous
cohesin in Xenopus egg extracts interferes with
the transfer of DNA-loaded fluorescently tagged
cohesin during replication. We have previous-
ly shown that parental histone transfer behind
replication forks is reduced by soluble histones
present in extracts, which likely inhibit paren-
tal histone interaction with replisome compo-
nents (22). To test whether endogenous cohesin
in extracts could have a similar inhibitory effect
on cohesin transfer, it was immunodepleted
from extracts; this had a negligible effect on
fork speeds (fig. S4, A and B). Even in cohesin-
depleted extracts, cohesin transfer was rare,
and 58 to 68% of cohesin was pushed by forks
(fig. S4, C to F). This suggests that endogenous
cohesin complexes in extracts do not prevent
the transfer of preloaded cohesin.
We also considered whether the presence
of high concentrations of Fen1-mKikGR, which
result in proliferating cell nuclear antigen
(PCNA) being retained on DNA during repli-
cation (17), inhibited cohesin transfer. Because

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 119

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

A CMG cohesin Sliding Transfer

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%

Exp. 1 Exp. 2 Exp. 1 Exp. 2

Fen1- JF646- Merge n = 178 n = 40 n = 63 n = 132
mKikGR Cohesin Transfer Removal Transfer Removal
Stalling Sliding LD555-GINS JF646-Cohesin Merge Stalling Sliding

Fig. 1. Replisomes push cohesin during DNA replication. (A) Cartoon
showing DNA replication from a single origin on surface-tethered l DNA.
Replication is performed in X. laevis egg extracts while collisions between
replisomes and cohesins are visualized. (B) Collisions between replication
forks, labeled with Fen1-mKikGR (red), and preloaded JF646-cohesin (magenta) are
visualized. An example showing cohesin sliding ahead of a Fen1-mKikGR–labeled
replication fork. (C) Comparison of primary cohesin fate after collision with replication
forks in extracts. Two independent experiments are shown. (D) A representative
kymogram showing LD555-GINS collision with JF646-cohesin. Examples of cohesin
sliding and removal are marked using the symbols shown in (A). (E) Proportions of
cohesin fates after collision by labeled replisomes. Two independent experiments are
shown. The n values in (C) and (E) indicate the total number of events analyzed.

PCNA has been implicated in cohesion estab-
lishment (23), we omitted Fen1-mKikGR to
exclude the possibility of inefficient cohesin
transfer resulting from improper PCNA re-
tention. We visualized the replisome directly
using a method previously described (24).
Purified fluorescent GINS was used to rescue
DNA replication in GINS-depleted extracts
(fig. S5). During replication of l DNA from
single origins, fluorescent CMG moved at the
tip of Fen1-mKikGR tracts (24–26), at an av-
erage speed consistent with previous work
(426 base pairs/minute; fig. S6, A to C). We
next visualized the outcomes of collisions be-
tween fluorescent replisomes and preloaded
JF646-cohesin (Fig. 1D; movie S2; and fig. S6,
D to G). Under these conditions, cohesin trans-
fer was still very rare (5% of events; Fig. 1E), and
cohesin sliding ahead of the replisome domi-
nated (56 to 67% of events). Therefore, we
concluded that the low frequency of cohesin
transfer observed in our previous experiments
was not caused by Fen1-mKikGR.
Cohesins relocalize to DNA replication
termination sites
How does conversion generate cohesion if pre-
loaded cohesin is not transferred behind the
replisome onto the replicated sister DNAs? We
speculated that cohesin pushed ahead of the
replisome could generate cohesion when meet-
ing a converging replisome. To determine the
fate of cohesin during fork convergence, repli-
cation was started from multiple origins (fig.
S7A). When visualizing converging replication
forks using Fen1-mKikGR, we observed that
most cohesin pushed ahead of replication
forks remained in positions where converging
replication forks met (Fig. 2, A and B; movie
S3; and fig. S7, B to D). In some cases, fork
convergence was accompanied by either cohe-
sin eviction or further cohesin movement after
fork convergence (Fig. 2C and fig. S7, E to G).
Cohesin remaining on DNA after fork conver-
gence was resistant to a high-salt wash that
removed Fen1 from DNAs (Fig. 2A and fig. S7,
B to D), suggesting that this cohesin popula-
tion was topologically bound to DNA (27). On
fully replicated DNAs, we observed multistep
photobleaching of cohesin in diffraction-limited
spots (mean = 1.90 steps; fig. S8, A to F), sug-
gesting that multiple cohesins are pushed to
sites of replication fork convergence.
We envisaged that upon fork convergence,
replisomes disassemble while cohesin traps both
daughter strands together. To test this, multiple-
origin firing experiments were performed with
fluorescent CMG. As expected, replisomes were
disassembled shortly after fork convergence (26)
(fig. S9A), even when labeled cohesin persisted
on DNA (Fig. 2C; movie S4; and fig. S9, B to G).
In 51 to 58% of cases, cohesin remained at the site
of replisome disassembly (Fig. 2D). Cohesin sig-
nal lifetime on DNA after replication termi-
nation varied, on average lasting for 43.7 min
(±17.8 min) (fig. S8, G to I). We conclude that
cohesin complexes are pushed by advancing
replisomes to sites of fork convergence and re-
main at these sites after replisome disassembly.
The key question therefore is: Do the cohesin
rings that persist on DNA after replication ter-
mination in our assay mediate cohesion?
Cohesin complexes retained at replication
termination sites can tether sister DNAs
To assess whether cohesin molecules at repli-
cation termination sites provide cohesion, we
developed an assay to measure sister DNA co-
hesion. Previous experiments tethering linear
DNAs to the surface only through 3′ biotins
have shown that the new sister DNA that does
not contain biotin collapses from the surface
after the replisome reaches the 5′ end (fig.
S10A) (24). Using this knowledge, we designed

120 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE


A B 100%
10 minutes
10 kb
HSW
JF646- Fen1- Merge
Cohesin mKikGR
Fig. 2. Cohesin is pushed to positions of DNA replication termination.
(A) Kymogram showing replication forks colliding with JF646-cohesin complexes under
conditions of high origin firing. After a period of replication, a high-salt wash (HSW) was
performed, and the same DNA was imaged. (B) Quantification of cohesin fates at
converging replication forks. The fate of cohesin that was pushed to a converging
a DNA template to visualize the interaction
between sister DNAs after replication (Fig. 3A).
Binding of Alexa Fluor 488–labeled LacI (LacI-
AF488, fig. S10B) to 48 lacO repeats at each DNA
end (fig. S10C) blocked replisome progres-
sion (Fig. 3A). Subsequent removal of LacI-
AF488 by isopropyl-b-D-thiogalactopyranoside
(IPTG) addition resulted in synchronous com-
pletion of replication and collapse of both
sister DNAs (Fig. 3A). This setup enabled us
to measure cohesion between the replicated
sister DNAs. Lack of cohesion between the
sister DNAs would result in sister DNAs im-
mediately separating (Fig. 3A, right panel, top
scenario). If, however, the replicated DNAs
were held together, the two collapsed sister
DNAs would colocalize (Fig. 3A, right panel,
bottom scenario).
We initiated replication from multiple ori-
gins in the presence of LacI-AF488, and Alexa
Fluor 647–deoxyuridine triphosphate (AF647-
dUTP) was incorporated into nascent DNA for
visualization. Excess LacI-AF488 and AF647-
dUTP were washed away, and replication ex-
tract containing IPTG was added to release
LacI-AF488 from DNA ends. Under these con-
ditions, regions of AF647-labeled replicated
DNA could be visualized during synchronous
collapse of new sister DNAs from DNA ends.
The assay is intrinsically validated by mole-
cules with partially replicated DNA, where
the labeled nascent DNA approaches only one
DNA end. In these instances, after removal of
LacI, the collapsed sister DNA moved with the
replisome toward the opposite end of the DNA
(fig. S10D), as observed previously (24). On mol-
ecules where the DNA template was fully rep-
licated up to the LacI barrier at both ends, LacI
removal caused the collapse of sister DNAs from
SCIENCE science.org
C D 100%
13%
32%
5%
87%
63%
Exp. 1 Exp. 2
n = 40 n = 24
Cohesin removal at
site of fork convergence
Cohesin moving from 10 kb
site of fork convergence
Cohesin remaining at LD555-
site of fork convergence GINS
replication fork was measured. Two independent experiments are shown. (C) Kymogram
example showing replisome (LD555-GINS) progression on DNA from multiple origins and
colliding with JF646-cohesin. (D) Quantification of JF646-cohesin fate at sites where
converging replisomes (LD555-GINS) are removed, with two independent experiments
shown. The n values in (B) and (D) indicate the total number of events analyzed.
both ends (fig. S10E). When both new sister
DNAs collapsed, they colocalized together for
varying lengths of time before separating, giving
a measure of cohesion. The critical question is
whether colocalization of sister DNAs resulted
from cohesin-mediated cohesion.
To test whether cohesin complexes physically
tethered collapsing sister DNAs in the experi-
ments described above, the assay was performed
in either mock- or cohesin-depleted extracts (fig.
S11A), and we compared the length of time that
collapsed sister DNAs colocalized (Fig. 3, B and C;
movie S5; and fig. S11B). The collapsed sister
DNAs remained associated reproducibly longer
in mock-depleted extracts than in cohesin-depleted
extracts (Fig. 3C). When purified cohesin was
preloaded onto DNAs before replication in
cohesin-depleted extracts, collapsed sister DNAs
remained together for periods of time com-
parable to that in mock-depleted extracts (Fig.
3C). In a bulk replication assay, formation of
replicated supercoiled plasmid was indistin-
guishable between mock- and cohesin-depleted
extracts (fig. S12, A to C), eliminating the pos-
sibility that tethering between sister DNAs may
be due to cohesin delaying replication com-
pletion at converging forks. Furthermore, most
colocalizing sister DNAs were disrupted
when treated with 0.1% SDS (fig. S12, E and
F), indicating that sister DNAs are coupled
through DNA–protein interactions rather than
DNA–DNA interactions. These results show
that cohesin complexes preloaded onto paren-
tal DNA physically tether sister DNAs after
replication. Taken together with our previous
observation that preloaded cohesin is predomi-
nantly pushed to sites of replication termina-
tion and remains at these sites after replisome
disassembly, our data provide strong evidence
RESE ARCH | R E S E A R C H A R T I C L E S
15 minutes
21% 22%
21%
27%
58%
51%
Exp. 1 Exp. 2
n = 43 n = 45
Cohesin removal during
CMG disassembly
Cohesin moving from site
of CMG disassembly
JF646- Merge Cohesin remaining at site
Cohesin of CMG disassembly
that cohesion establishment by cohesin con-
version occurs during replication termination.
Our model also predicts that cohesive cohesin
complexes should be dragged by the collapsing
sister DNAs and remain associated with both
sister DNAs (fig. S13A, right panel, top sce-
nario). If cohesin does not tether sister DNAs
together, we would expect half of the cohesin
molecules to remain associated with stretched
DNA after sister DNA collapse (fig. S13A, right
panel, bottom scenario). To test this, sister DNA
collapse experiments were performed with
JF549-cohesin preloaded onto parental DNA
and imaged simultaneously with AF647-dUTP.
Cohesin was observed to move when both sister
DNAs collapsed (Fig. 3D, movie S6, and fig.
S13B) and when a single sister DNA collapsed
(fig. S13C). Between 73 and 85% of cohesin mol-
ecules moved and colocalized with collapsed
sister DNAs (Fig. 3E), whereas only 15 to 27%
remained on stretched DNAs (fig. S13D). We
estimate that 50 to 70% of cohesins retained
on DNA after replication termination bound
both sister DNAs (see methods in the supple-
mentary materials). The cohesin colocaliza-
tion with collapsed sister DNAs reaffirms that
cohesion establishment by preloaded cohesin
complexes happens during replication termi-
nation. A crucial event during replication ter-
mination is the removal of the CMG helicase by
the adenosine triphosphatase p97 (28). We next
asked whether a timely replisome disassembly
during termination could be required for proper
cohesion establishment.
Cohesion establishment during replication termination
To assess whether CMG disassembly during
replication termination is important for proper
cohesion, we inhibited CMG disassembly in
5 APRIL 2024 • VOL 384 ISSUE 6691 121
RES EARCH | R E S E A R C H A R T I C L E S

A B LacI-AF488
LacO repeats
5'

10 minutes

10 minutes
3' t
Replication extract
LacI-AF488
AF647-dUTP
Separation

Sister DNA collapse

t
Replication extract

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
together. (A) Diagram showing sister DNA collapse experiments. Replication
forks are paused at DNA ends by LacI. Upon IPTG addition, replication forks
reach DNA ends, and collapsing sister DNAs are visualized. (B) Example
kymograms where both sister DNA strands collapse (see cartoon). Both examples
are from mock depletion with different sister DNA separation times. The time that
collapsed DNA strands colocalized before separating is indicated. (C) Individual data
points showing the time that collapsed sister DNAs colocalized before separation for
mock-depleted (n = 127), cohesin-depleted (n = 106), and rescue (n = 110) extracts.
Data are shown mean value with 95% confidence intervals. Three independent
experiments were performed. P values were calculated using a two-tailed Mann-
Whitney U test. (D) Example kymogram showing JF549-cohesin associating with
both collapsed sister DNAs. The red arrow indicates cohesin at the end of DNA tethers,
which is excluded from the analysis. (E) Quantification of cohesin position after sister
DNA collapse of either one strand or both strands, with data from two independent
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

122 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

A B C D
+p97i

15 minutes
10 kb 7%

22%

4%

67%
?

n=27

Cohesin removal

LD555 - JF646 - Merge Replisome stalling

GINS Cohesin Replisome bypass
(cohesin remaining)
Replisome bypass
(cohesin sliding)

Fig. 4. Cohesin dynamics during replication termination when replisome
disassembly is inhibited. (A) Cohesion of the URA3 locus was measured
in wild-type and Cdc48-AID yeast cells that were synchronized in G1 and released
into G2 arrest in the presence of 5 mM indole-3-acetic acid (IAA). Cohesion
was also measured in asynchronously growing wild-type, dia2D, and dia2-13A
mutant cells. For each experimental condition, at least 100 mitotic cells
were scored for one or two GFP dots, and each experiment was repeated three
times. The graphs show the mean and standard deviation. (B) Replisomes
were labeled directly using LD555-GINS, and replisome disassembly upon fork
convergence was inhibited with 200 mM p97i (NMS-873). In this example,
converging replisomes bypass one another, and one replisome continues pushing
a labeled cohesin (white square). (C) Quantification of cohesin and replisome
fates observed during replication termination when replisome disassembly
is inhibited. The n value indicates the total number of events analyzed. (D) Model
of cohesion establishment during replication termination. Possible scenarios
describing termination-coupled cohesion establishment are shown in fig. S16.

for cohesion, it is likely that the single-
molecule sister DNA collapse assay may
not be sensitive enough to measure mod-
erate levels of cohesion defects.
To further investigate the interplay between
cohesion establishment and replisome disas-
sembly in vivo, we explored the role of Dia2,
the ubiquitin ligase responsible for ubiquity-
lating Mcm7 in yeast and marking replisomes
for removal by Cdc48 (29). Deleting the DIA2
gene as well as mutating it to prevent Dia2’s
association with the replisome (31) resulted in
significant cohesion defects (Fig. 4A), further
supporting the notion that defective CMG dis-
assembly adversely affects cohesion. Although
these findings demonstrate Dia2’s role in ro-
bust cohesion, they do not address whether
the cohesion defects in dia2 mutants result
from perturbations in CMG disassembly dur-
ing replication termination.
To ascertain the specificity of Dia2’s role in
cohesion at replication termination, we engi-
neered a yeast strain harboring two small
linear chromosomes of similar lengths but dif-
fering in the number of replication origins and
selection markers (fig. S14E). The URA+ mini-
chromosome, which enables cells to grow in the
absence of uracil in the growth medium (uracil
auxotrophy), has two high-efficiency replication
origins, whereas the MET+ minichromosome,
conferring methionine auxotrophy, has only
one replication origin. Consequently, owing to
the presence of two replication origins, rep-
lication termination and subsequent CMG dis-
assembly exclusively occur in the URA+ and
not in the MET+ minichromosome. We noted
that wild-type cells retained the two-origin
URA+ minichromosome more efficiently than
the single-origin MET+ counterpart (fig. S14E).
The less-efficient retention of the MET+ mini-
chromosome could either stem from potential
cohesion defects due to the absence of repli-
cation termination events or to less-efficient
replication from its single origin compared
with the dual-origin URA+ minichromosome.
To test these scenarios, we assessed the reten-
tion efficiency of minichromosomes in dia2D
cells. We reasoned that if Dia2-mediated CMG
removal during replication termination is es-
sential for cohesion, dia2 deletion would se-
lectively impair retention of the URA+, but not
the MET+, minichromosome. Consistent with
our hypothesis, the dia2D mutant displayed
comparable efficiency to wild-type cells in re-
taining the MET+ minichromosome but ex-
hibited significant defects in maintaining the
URA+ minichromosome (fig. S14E). Together,
in vivo data reinforce single-molecule results
and support the idea that cohesion establish-
ment is intricately connected to DNA replica-
tion termination and subsequent CMG complex
disassembly.
Discussion
In contrast to prevailing models, which place
preloaded cohesin behind replication forks, we
find that cohesin rings are pushed along the
DNA by the advancing replisome. Cohesin rings
pushed to positions of fork convergence are
retained on replicated DNA even after replisome
disassembly and are capable of tethering sister
DNAs together. The notion of replisomes push-
ing cohesin is supported by transcription re-
positioning cohesin on yeast chromosomes (14)
and cohesin being pushed ahead of T7 RNA
polymerase and DNA translocase FtsK in vitro
(32, 33). A previous single-molecule study re-
ported that a substantial fraction of cohesin was
incorporated into replicated DNA in Xenopus
egg extracts (34). We suggest that the cohesin
transfer events observed therein primarily re-
sulted from cohesin being incorporated dur-
ing replication termination, as the study used
firing from multiple origins. In our system,
conversion of preloaded cohesin to cohesive
structures through bona fide cohesin transfer
behind the replication fork might occur, albeit
rarely. Removal and stalling events might re-
flect differences in cohesin binding onto DNA
before replication. We envisage that cohesin re-
moval helps prevent the accumulation of many
cohesins sliding ahead of the fork, which could
impede replisome interaction with other DNA-
bound proteins, such as histones.

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 123

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

We envisage that CMG complexes that are
retained on DNA after fork convergence may
continue pushing cohesin complexes along
DNA, as seen in our single-molecule assays,
and eventually evict them from chromatin in
cells, resulting in cohesion impairment. We
propose several potential models for cohesion
establishment at termination sites (Fig. 4D
and fig. S16). Converging replisomes could
pull the final stretches of unreplicated DNA
through cohesin rings (fig. S16A). This would
result in replisome disassembly and cohesion
between new sister DNAs. This model negates
the requirement for replisomes to pass through
cohesin rings or for transient ring opening. Two
alternative possibilities are that (i) cohesin is
transferred behind one replication fork during
termination (fig. S16B) or that (ii) a terminat-
ing replisome bypasses the cohesin ring (fig.
S16C). Bypass of the cohesin ring could use a
mechanism similar to bypass of DNA–protein
cross-links by CMG helicase (24). Factors re-
quired for cohesin conversion (13) could aid
pushing of cohesin by the replisome, facilitate
CMG removal, or be involved in the molecular
transactions between DNA, replisomes, and
cohesin during replication termination.
30. J. M. Dewar, E. Low, M. Mann, M. Räschle, J. C. Walter,
Genes Dev. 31, 275–290 (2017).
31. M. Jenkyn-Bedford et al., Nature 600, 743–747 (2021).
32. J. Stigler, G. Ö. Çamdere, D. E. Koshland, E. C. Greene,
Cell Rep. 15, 988–998 (2016).
33. I. F. Davidson et al., EMBO J. 35, 2671–2685 (2016).
34. M. Kanke, E. Tahara, P. J. Huis In’t Veld, T. Nishiyama, EMBO J.
35, 2686–2698 (2016).
35. G. Cameron et al., Raw data for: Sister chromatid cohesion
establishment during DNA replication termination, Figshare
(2024); https://doi.org/10.25418/crick.25062113.
ACKN OWLED GMEN TS
We thank J. Diffley for permission to use the yeast strain
containing linear minichromosomes, S. Yardimci for helping with
extract preparations, and A. Kaushik for technical assistance. We
thank the Francis Crick Institute Aquatics Facility for X. laevis
husbandry and egg collection and Chemical Biology Facility for
synthesis of fluorescent peptides for GINS labeling. Funding:
This work was supported by the Francis Crick Institute, which
receives core funding from Cancer Research UK, the UK Medical
Research Council, and the Wellcome Trust grant CC2133 (G.C.,
D.T.G., S.X., Ç.K., and H.Y.); a Boehringer Ingelheim Fonds PhD
Fellowship (G.C.); a Sir Henry Dale Fellowship jointly funded by the
Wellcome Trust and the Royal Society grant 223235/Z/21/Z
(D.T.G.); the Wellcome Trust grant 226494/Z/22/Z (M.S.); and
Cancer Research UK grant 26747 (K.A.N.). Author contributions:
Purified Xenopus cohesin complexes and performed cohesin
loading in bulk: M.S. Experiments with Fen1-mKikGR comparing
undepleted and cohesin-depleted extracts: D.T.G. Analysis of
single-molecule data: G.C., D.T.G., and Ç.K. Xenopus GINS cloning
and expression: S.X. Experiments with S. cerevisiae and data
analysis: R.G. Construction of the yeast strain containing
minichromosomes: J.B. All other experiments: G.C. Supervision:
M.S. and H.Y. Writing – original draft: G.C. and H.Y. Writing – review &
editing: G.C., D.T.G., Ç.K., K.A.N., M.S., and H.Y. Competing
interests: The authors declare that they have no competing
interests. Data and materials availability: All data needed to
evaluate the conclusions in the paper are present in the paper
and/or the supplementary materials. Raw data are available
in Figshare (35). All materials produced in this study, such as
plasmids and cell lines, can be provided upon request. License
information: Copyright © 2024 the authors, some rights
reserved; exclusive licensee American Association for the
Advancement of Science. No claim to original US government works.
https://www.science.org/about/science-licenses-journal-article-
reuse. This research was funded in whole or in part by the
Wellcome Trust (CC2133, 223235/Z/21/Z, 226494/Z/22/Z), a
cOAlition S organization. The author will make the Author
Accepted Manuscript (AAM) version available under a CC BY
public copyright license.
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adf0224
Materials and Methods
Figs. S1 to S16
Table S1
References (36–42)
MDAR Reproducibility Checklist
Movies S1 to S7
Submitted 27 September 2022; resubmitted 16 June 2023
Accepted 27 February 2024
Published online 14 March 2024
10.1126/science.adf0224

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,

Methods 57, 179–186 (2012).
19. W. S. Hoogenboom, D. Klein Douwel, P. Knipscheer, Dev. Biol.
428, 300–309 (2017).
20. A. Losada, M. Hirano, T. Hirano, Genes Dev. 12, 1986–1997
(1998).
21. H. Yardimci, A. B. Loveland, S. Habuchi, A. M. van Oijen,
J. C. Walter, Mol. Cell 40, 834–840 (2010).
22. D. T. Gruszka, S. Xie, H. Kimura, H. Yardimci, Sci. Adv. 6,
eabc0330 (2020).
23. H. W. Liu et al., Mol. Cell 78, 725–738.e4 (2020).
24. J. L. Sparks et al., Cell 176, 167–181.e21 (2019).
25. K. B. Vrtis et al., Mol. Cell 81, 1309–1318.e6 (2021).
26. E. Low, G. Chistol, M. S. Zaher, O. V. Kochenova, J. C. Walter,
Genes Dev. 34, 1534–1545 (2020).
27. M. Minamino, T. L. Higashi, C. Bouchoux, F. Uhlmann, Life Sci.
Alliance 1, e201800143 (2018).
28. J. M. Dewar, J. C. Walter, Nat. Rev. Mol. Cell Biol. 18, 507–516
(2017).
29. M. Maric, T. Maculins, G. De Piccoli, K. Labib, Science 346,
1253596 (2014).
C
leistogamy favors genetic stabilization
of desired agronomic traits as well as a
higher fruiting rate (1–3). As a result, va-
rieties with cleistogamy were often re-
tained during crop domestication (4, 5).
Cultivated tomatoes display complete cleis-
togamy and thus have considerably higher
fruit set, compared with wild species that re-
quire cross-pollination (6). The transition from
anisogamy to complete cleistogamy in tomato
1
College of Horticulture, Haixia Institute of Science and
Technology, Fujian Agriculture and Forestry University,
Fuzhou 350002, China. 2College of Life Sciences, Fujian
Agriculture and Forestry University, Fuzhou 350002, China.
*Corresponding author. Email: wus@fafu.edu.cn
†These authors contributed equally to this work.
was thought to have been achieved by the
elimination of self-incompatibility (SI) (7–9)
and stigma shortening (mutations in the
Style2.1 and SE3.1 genes) (10, 11). However,
as shown in other Solanaceae species such as
peppers and eggplants, stigma-shortening and
self-incompatibility release alone are not suffi-
cient for the establishment of cleistogamy
(12–14). Tomato flowers develop a morpho-
logical feature known as the “anther cone”,
in which five anthers join together to form a
hollow tube-like cone around the style. In this
structure, the stigma is surrounded and sealed
by the joined anthers (15). As tomato flowers are
pendant, released pollen grains fall onto the
stigma, ensuring self-pollination. The anther

124 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

cone is most likely required for cleistogamy
because only tomato varieties with this mor-
phological trait reproduce by cleistogamy (15).
However, it remains unknown how anther fu-
sion is achieved to form the anther cone.
HD-Zip IV family proteins are transcriptional
regulators that are often involved in epidermal
cell development in plants (16–19). In Arabidopsis,
epidermal differentiation is regulated by two
HD-Zip IV genes, ARABIDOPSIS THALIANA
MERISTEM LAYER1 (AtML1) and PROTODERMAL
FACTOR2 (PDF2) (20–24). In tomato, Woolly
(Wo), a homolog of AtML1 and PDF2, con-
trols the initiation and differentiation of the
trichomes on the surface of leaves and stems,
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.
disruption of Wo function in woW106R (26), the
CRISPR/Cas9–generated mutant of Wo (cr-wo,
a loss of function mutation) had no obvious
phenotype in the style and anther cone (Fig.
2E and fig. S2B). Consistent with woW106R, the
style and interlocking trichome phenotypes
of the CRISPR/CAS9 mutant of woW106R (cr-
woW106R) were rescued (Fig. 2E and fig. S2, B
to G). We thus speculate that the W106R mu-
tation generates a dominant negative effect on
Wo function.
Dominant negative effects generally oc-
cur where protein regulation is mediated by
multimers (28). In Immunoprecipitation-Mass
Spectrometry (IP-MS) analysis of the Wo protein
in tomato (25), we found that Wo was able to
interact with several similar HD-Zip IV pro-
teins, including HD7L and HD7 (table S1). Using
Co-immunoprecipitation (Co-IP) and bimolec-
ular fluorescence complementation (BiFC) assays,
we demonstrated that the woW106R mutant-form
protein can also interact with these proteins
(Fig. 2F and fig. S2H).
Using the promoter of MX1 (MIXTA-like 1)—
a previously identified downstream target
of Wo (25)—as a reporter, we found that the
woW106R protein not only lost its ability to ac-
tivate MX1 expression but also inhibited the
activation ability of several other HD-Zip IV
proteins, including wild-type Wo, as well as

Anther cone development requires

interlocking trichomes
In modern cultivated tomato species, the edge
of the anthers develops a set of trichomes that
interlock the adjacent anthers, apparently forc-
ing them to form an anther cone surrounding
the style (Fig. 1, A to C, fig. S1, A to D, and movie
S1). These trichomes are initiated about 10 to
12 days before flower opening (−10 to approx-
imately −12 days) and become mature about
4 days before flower opening (−4 days; Fig. 1,
D to E). Unlike conventional trichomes that
develop on the surface of other tomato tissues,
this type of trichome is a unicellular structure
with high polypoidy (Fig. 1, F and G, and fig.
S1E). Upon maturation these trichomes become
intertwined, with the central region swollen and
bent into hooks that hold adjacent anthers
together (fig. S1F). The united anthers eventu-
ally form a sealed cone structure before the
neighboring sepals split apart (Fig. 1D).

Wo dominant mutation affects interlocking

trichomes and style
Conventional trichomes in tomato leaves and
stems are regulated by an HD-Zip IV transcrip-
tion factor, Wo (26, 27). In the woW106R mutant,
an ethyl methyl sulfone (EMS)–induced mutant
in which a W-to-R mutation occurs at position
106 of Wo (26), we observed a 20% reduction
in the number of interlocking trichomes and
that the anther cone remained intact (Fig. 2A
and fig. S2A). This contrasts with the woW106R
phenotype in other tissues in which trichomes
are absent (26). Furthermore, we observed an
additional phenotype in this mutant in which
the style is shortened by ~30% (Fig. 2B). The
woW106R style cells became smaller than the
wild type (WT) while the cell number was in-
creased, possibly to compensate for the re-
duced cell size (Fig. 2, C and D). Despite the
Fig. 1. Intertwining trichomes form the interlocking structures of anther cones. (A) Anther cones of
MT and Ailsa Craig (AC) tomato varieties. (B and C) SEM images of the anther cones in MT and AC. The
special interlocking trichomes are highlighted in yellow. (D) Temporal development of anthers and
interlocking trichomes in tomato. −10d, −8d, −6d, and −4d represent the floral buds on the days prior to
flower opening. The anthers are completely enveloped by sepals before −4d; the interlocking trichomes
are initiated on the anther on -10d and are fully mature on −4d. Note the swollen and bent part in the middle
of the interlocking trichomes. (E) Illustration of anther fusion through interlocking trichomes. (F) Cell
morphology and polyploidy analysis of interlocking trichomes. Green fluorescence in H2B-GFP indicates the
nucleus; stomatal guard cells are used as a comparison. Cell morphology is outlined by propidium iodide
(magenta fluorescence). (G) Quantification of the DNA content in interlocking trichomes and guard cells.
Nuclear DNA content is normalized to that in guard cells. Data are mean ± SD (n > 10). Unpaired t tests were
used for statistical analysis (****P < 0.0001).

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 125

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

HD7L and HD7 (Fig. 2G and fig. S2I). The
W106R mutation disrupts the ability of Wo
to bind the L1-box element of downstream
genes (26). We thus hypothesized that the
woW106R protein forms dimers with other
HD-Zip IV proteins, thereby interfering with
their function of binding downstream targets
(Fig. 2H).
Repression of HD-Zip IV genes splits
the anther cone
We found that the overexpression of MTR1
(OE-MTR1), a repressor of the HD-Zip IV genes,
resulted in a shortened style, similar to those
in woW106R (fig. S3A). In addition, we observed a
disrupted anther cone in OE-MTR1, which was
not seen in woW106R (fig. S3B and Fig. 2A). One
possible explanation for this phenotype is that
MTR1 overexpression simultaneously represses
Wo and its functionally redundant homologs.
Phylogenetic analysis revealed that HD7 and
HD7L had the highest sequence similarity to
Wo proteins (fig. S3C). Consistent with this,
luciferase reporter analysis showed that MTR1
also inhibited the transcriptional function of
HD7 and HD7L (fig. S3, D to E). Therefore, we
speculate that HD7 and HD7L play a dominant
role in the regulation of interlocking trichomes
of the anther cone.
We constructed a promoter-driven GUS re-
porter for Wo and its homologs HD7 and HD7L.
Histological staining showed that all three
HD-Zip IV genes were specifically expressed
in the upper region of the style, whereas Wo
expression was higher than that of HD7 and
HD7L (Fig. 3A). In support of their function
in anthers, the three HD-Zip IV genes were
highly expressed at the inter-anther junction,
with the expression of HD7 and HD7L higher than
that of Wo (Fig. 3B and fig. S3F). The spatio-
temporal expression pattern suggests that Wo
may function mainly in the style whereas HD7
and HD7L play a dominant role in anther cone
formation.
HD7, HD7L, and Wo coordinately regulate
reproductive development
To further test the function of HD7 and HD7L,
we generated loss-of-function mutants using
CRISPR/Cas9 (fig. S4, A and B). Single mutants
of Wo, HD7, or HD7L had no phenotype in the
anther cone or style development (fig. S4, C
to H). In line with our hypothesis, cr-hd7/hd7l
double mutants had almost no anther trichomes;
we also observed that the anther cones split
apart (fig. S5, A to D). By means of SEM we
found that the initiation of anther trichomes
in cr-hd7/hd7l was less than that in the WT at
10 days before anthesis (Fig. 3, C to D). Con-
sistently, the nuclear size in mature anther
trichomes was considerably reduced (Fig. 3,
E to F). cr-hd7/hd7l double mutants had no
obvious style shortening phenotype but style

Fig. 2. Dominant negative mutation of Wo inhibits interlocking trichomes
and style length. (A) Interlocking trichomes are moderately affected in the
woW106R mutant. (B) Mutation site and style phenotype of woW106R mutant.
(C) Phenotype of style epidermal cells and (D) quantification of cell size. Note
that cell size is smaller but cell number is considerably larger in woW106R. Data are
mean ± SD (n > 8) (****P < 0.0001). (E) Style phenotype in different backgrounds.
Note that Wo knockout results in no style shortening phenotype and knockout
of woW106R mutant allele partially rescues style phenotype. (F) woW106R interacts with
Wo, HD7, and HD7L proteins in Co-IP. (G) The woW106R protein inhibits the ability of
the interacting HD-Zip IV proteins to activate downstream gene MX1 in Luciferase
(LUC) reporter assay. Data are mean ± SD (n = 3 biological replicates) (**P < 0.01,
***P < 0.001). (H) Working model of HD-Zip IV and interference by woW106R. Wo
forms dimers with other HD-Zip IV proteins to bind to DNA through the HD domain
during the activation of downstream genes. The dimers formed with woW106R
protein disrupt this DNA binding ability, resulting in inactivation of downstream
genes. Unpaired t tests were used for all statistical analyses.

126 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

shortening was most evident in cr-hd7/wo
double mutants (Fig. 3G). Together with the
specific tissue expression patterns of HD7, Wo,
and HD7L, we propose that these three HD-
Zip IV proteins jointly regulate anther cone
formation and coordinated style growth in
tomato (Fig. 3, G to I), albeit with different
levels of involvement (Fig. 3, A and B).
In addition to the functional indispensabil-
ity of interlocking trichomes for anther cones,
we found that the misexpression of the domi-
nant allele of Wo (WoV) in the petal caused
dispersed trichomes to become interlocked,
resulting in a closed petal structure (fig. S5, E
to F). This result indicates that interlocking of
trichomes between adjacent separate organs
can generate organ fusion.
HD-Zip IV promotes endoreduplication in the style
Style development exhibits a spatially polar
distribution of cell division and cell expansion.
Cells in the upper region of the style (near the
stigma) were all smaller whereas the cells in
the middle and lower regions were larger (fig.
S6A). Using a histone (H2B-GFP) marker, we
found that cell division occurred specifically in
the upper region of the style. Away from the
division region, the style cells became progres-
sively endoreduplicated resulting in a gradual
increase in cell size in the middle and bottom
parts. The polar distribution of division, elon-
gation, and maturation zones in the style is
similar to that of plant root development (Fig.
4A and fig. S6B). In the mature zone of the style,
nuclei were highly polyploid with many reach-
ing 10C (C represents the DNA amount of the
unreplicated haploid genome) (Fig. 4B).
Among different combinations of double
mutants of Wo, HD7, and HD7L, cr-hd7/wo
had the smallest style cells (Fig. 4C). How-
ever, the number of style cells in this mutant
was increased compared with WT (Fig. 4, D to
E). Using the H2B marker, we found that the
transition from cell division to endoredupli-
cation was blocked in the cr-hd7/wo double
mutant, which showed active cell division
throughout the whole style at both early and
late stages of development (Fig. 4F). This ulti-
mately resulted in the style cells having smaller
nuclei in the cr-hd7/wo double mutant com-
pared with the WT (Fig. 4G). The three HD-Zip
IV genes are likely to jointly promote the tran-
sition from cell division to endoreduplication
(Fig. 4H) though the style phenotype varied in
their respective mutants. The reduction in endo-
reduplication was most severe in cr-wo/hd7
double mutants and was least severe in cr-hd7/
hd7l double mutants (Fig. 4, C to E). The distinct
involvement of the three HD-Zip IV genes in
style development may be due to their different
expression patterns in style tissue (Fig. 3A).
Based on our results, we propose that the
three HD-Zip IV genes function redundantly

Fig. 3. HD7, HD7L, and Wo coordinate the development of anther cone and
style. (A) Promoter-driven GUS assay shows that HD7, WO, and HD7L are
specifically expressed in the upper region of the style. −8d and −4d are 8 and
4 days before flower opening, respectively. (B) GUS staining of the cross sections
of -10d anthers. Note that HD7, WO, and HD7L are specifically expressed at
the junction between adjacent anthers. (C) Trichome phenotype in the -10d
anther of cr-hd7/wo, cr-wo/hd7l, and cr-hd7/hd7l. (D) Quantification of trichome
number. Data are mean ± SD (n > 5). (E) DAPI staining of the mature interlocking
trichomes in different backgrounds. (F) Quantification of nuclear size. Nuclear
DNA content is normalized to that in guard cells. Data are mean ± SD
(n > 10). (G) Style phenotype in different backgrounds and the quantification
of mature style length. Data are mean ± SD (n > 5). Unpaired t tests were used
for statistical analysis (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).
(H) Model depicting the roles of HD7, WO, and HD7L in the initiation of interlocking
trichomes. (I) Illustration showing that the neighboring anthers are united
together by the intertwining interlocked trichomes to form an anther cone.

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 127

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

Fig. 4. HD-Zip IV regulators promote the endoreduplication of style
giant cells. (A) (top) Tomato style exhibiting zone-specific features. (Middle)
SEM images showing that style cell size progressively increases from top to
bottom. The H2B-GFP marker indicates that division occurs only in the
upper region of the style. (B) Quantification of the nuclei size in mature style cells
expressing H2B-GFP. Nuclear DNA content is normalized to that in guard cells
(n > 10). (C) SEM images of style cells in cr-hd7/wo, cr-wo/hd7l, and cr-hd7/hd7l.
(D and E) Quantification of cell size (n > 10) and cell number of the style in the
mutants (n > 5). (F) Endoreduplication of style cells in cr-hd7/wo (marked by
H2B-GFP). The mitotic activity of cr-hd7/wo is substantially higher than that of
the WT at both the early (-10 d) and late stage (−2d) of style development.
(G) Quantification of nuclear size in mature style cells of the WT and cr-hd7/wo
(n > 10). (H) Model illustrating the spatial expression of HD-Zip IV genes and
polar distribution of mitosis and endoreduplication in tomato style development.
Data are mean ± SD for all quantifications. Unpaired t tests were used for all
statistical analyses (*P < 0.05, ****P < 0.0001).

in anther cone formation and style length reg-
ulation. However, we were unable to obtain
triple mutants likely because of embryonic
lethality. To overcome this we made a chimeric
fusion protein of Wo and HD7 proteins with
an SRDX transcriptional repressor (pWo:Wo-
SRDX and pHD7:HD7-SRDX). We found that
all transgenic lines expressing these chimeric
repressive proteins showed phenotypes of
shortened styles with smaller cells, similar to
cr-hd7/wo; they also showed disappearance of
interlocking trichomes on the anthers, result-
ing in split anther cones (fig. S7). These results
suggest that HD7, Wo, and HD7L are HD-Zip IV
regulators that jointly promote interlock-
ing trichome formation and style cell endo-
reduplication (Fig. 3, H and I, and Fig. 4H).
Cleistogamy was selected for during
tomato domestication
Although the modern tomato is a cleistoga-
mous species, wild tomato species reproduce
by crosspollination. To understand the domes-
tication process we examined the transcription
of Wo, HD7, and HD7L in wild tomato relatives
and wild tomato species, including Solanum
lycopersicoides, Solanum pennellii, and Solanum
habrochaites (29–32). Trichome-containing
anthers did not evolve in S. lycopersicoides
and the anther of S. pennellii only forms stunt
trichomes without interlocking capacity (Fig. 5A).
Consistent with these differences, our transcrip-
tional analyses indicated that the expression
of Wo, HD7, and HD7L in the anthers of S.
lycopersicoides and S. pennellii was 2 to 4 times
lower than that of the modern tomato (Fig. 5B).
It is possible that domestication enhanced
the expression of the HD-Zip IV regulators in
modern varieties, leading to the formation of
anther cones and an exserted style. To test this
possibility we observed the floral organs of the
WoP635R mutant, a functionally enhanced mu-
tation that substantially increases Wo pro-
tein stability (25). In this mutant, we observed
an increase in interlocking trichomes on the
anthers compared with the WT and stigma
exsertion resulting from an elongated style
(fig. S8, A to G).
The expression of Style 2.1, a previously iden-
tified key factor for stigma shortening, was 1.5
to 2.5 times lower in the modern tomato than
that of modern tomato ancestors including
S. lycopersicoide (Fig. 5C). In addition, we
found that the expression Style 2.1 was con-
siderably decreased in the styles of cr-hd7/wo
double mutants but considerably increased in
the styles of WoP635R mutants (Fig. 5D and fig.
S8H). In both wild tomato S. pennellii and the
modern tomato MT, the spatial expression
pattern of Style 2.1 in the style resembles that
of the HD Zip IV genes (Fig. 5E and Fig. 3A). In
addition, the expression of all Style 2.1 genes
from different tomato species and relatives
was repressed by woW106R (fig. S9, A and B).
Raising Style 2.1 expression level in the woW106R
background considerably mitigated the cell
division phenotype in the woW106R mutant (Fig.
5F and fig. S9, C and F). These results suggest
that Style 2.1 is downstream of the pathways

128 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

 RESE ARCH | R E S E A R C H A R T I C L E S

Fig. 5. Domestication of anther cone is putatively mediated by enhanced
expression of HD-Zip IV genes. (A) SEM images showing the incomplete
anther cone in tomato close relatives and wild tomato varieties. (B) Transcription
of HD7, HD7L, and Wo is considerably enhanced in the anther of MT.
(C) However, Style 2.1 expression in the style of MT is substantially reduced
compared with S. lycopersicoides, S. pennellii, and S. habrochaites (n = 3
biological replicates). (D) Expression of Style 2.1 in a cr-hd7/wo mutant (n = 3
biological replicates). (E) GUS reporter of the Style 2.1 promoters from wild
tomato (S. Pennellii) and cultivated tomato (MT). SP, S. Pennellii; MT, micro-Tom.
(F) Misexpression of Style 2.1 driven by the Wo promoter partially rescued the
reduced endoreduplication phenotype in the style of woW106R mutants.
(G) Predicted model of the transition from crosspollination to cleistogamy in wild
tomatoes. The expression of HD-Zip IV gene was increased first, resulting in a
cylinder-shaped anther and exposed stigma in wild tomatoes, followed by the
recessed stigma which appeared after removal of self-incompatibility. (H) Fruit
set and (I) quantification for the woW106R mutant. Data are mean ± SD for all
quantifications. Unpaired t tests were used for all statistical analyses (ns, not
significant) (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).

regulated by HD7 and Wo. Based on our results
we propose that the increased expression of
HD-Zip IV genes led to an exserted stigma and
another cone structure in wild tomato during
its evolution. Subsequently, the domestication
of stigma shortening occurred only after the loss
of self-incompatibility (Fig. 5G). These stepwise
processes may have eventually produced cleis-
togamy in the modern tomato.
Discussion
It has been previously reported that the con-
version from exserted to inserted stigma in
modern tomatoes is due to the mutations in
the Style 2.1 promoters (11). When the HD-Zip
regulators promote anther cone formation
they also promote style elongation, which ap-
pears to be detrimental to the formation of a
fully closed cleistogamous flower. However,
this mechanism is likely to be evolutionarily
advantageous because it may have facilitated
crosspollination in wild tomato species before
the self-incompatibility trait was lost. Simulta-
neous promotion of anther fusion and style
elongation seems to ensure the exserted style
for the pollination by insects or other pollina-
tors. Upon the loss of self-incompatibility, the
exserted stigma was not necessary, the style
evolved to be shorter, and the stigma was con-
verted to an inserted one, eventually leading
to cleistogamy in the modern tomato. In sup-
port of cleistogamy favoring fruit set, we found
that the dominant negative mutant woW106R
had a 15 to approximately 30% higher fruit set
rate than the WT (Fig. 5, H and I). This could
provide an advantage during evolution or be
deliberately selected in domestication. The
mechanism identified in tomato could also
be applied to many other plant species with
anther cones in nature (14, 33).
AtML1 and PDF2 are homologs of HD7, Wo,
and HD7L in Arabidopsis. They promote the
formation of giant cells on Arabidopsis sepals
through a dose-dependent manner (23, 24).
Additionally, style cells of different sizes were
formed in different combinations of cr-hd7/
wo, cr-wo/hd7l, etc., suggesting that the dosage
effect may also exist in the development of in-
terlocking trichomes and style giant cells.
However, the threshold of functional doses
seems to be tissue-specific. High levels of Wo
appear to promote cell division in trichomes
(34), while inhibiting cell division in the style.
A similar phenomenon has also been observed
with another HD Zip IV protein, GL2, which
has opposite effects on trichomes and root hairs
in the shoot and root, respectively (35, 36).
Dominant negative effects often occur during
regulation that requires the formation of homo-
or heterodimers (37, 38). The Arabidopsis
gene GL2, also a HD-Zip IV protein, binds DNA
through its HD domain whereas the START
domain of GL2 is required to form a dimer (39).

SCIENCE science.org 5 APRIL 2024 • VOL 384 ISSUE 6691 129

RES EARCH | R E S E A R C H A R T I C L E S
The dominant mutation or deletion mutation
of HD7 and HD7L in the HD domain could lead
to the formation of non-functional dimers sim-
ilar to woW106R with other HD-Zip IV proteins
(fig. S10, A to C). Because the HD and START
domains are highly conserved structural do-
mains in HD-Zip IV proteins (39), the dominant-
negative effect caused by mutations in the HD
domain is likely to be a common mechanism.
RE FE RENCES AND N OT ES
1. M. M. Strelin, P. K. Diggle, M. A. Aizen, Trends Plant Sci. 28,
1360–1369 (2023).
2. D. W. Schemske, Ecology 59, 596–613 (1978).
3. D. J. Schoen, D. G. Lloyd, Biol. J. Linn. 23, 303–322 (1984).
4. G. L. Stebbins, Am. Nat. 91, 337–354 (1957).
5. S. C. H. Barrett, Nat. Rev. Genet. 3, 274–284 (2002).
6. C. M. Rick, W. H. Dempsey, Bot. Gaz. 130, 180–186 (1969).
7. S. Takayama, A. Isogai, Annu. Rev. Plant Biol. 56, 467–489
(2005).
8. W. Li, R. T. Chetelat, Science 330, 1827–1830 (2010).
9. J. W. Busch, D. J. Schoen, Trends Plant Sci. 13, 128–136
(2008).
10. L. Shang et al., Plant Cell 33, 3293–3308 (2021).
11. K.-Y. Chen, B. Cong, R. Wing, J. Vrebalov, S. D. Tanksley,
Science 318, 643–645 (2007).
12. R. Subramanya, H. Y. Ozaki, Euphytica 33, 13–16 (1984).
13. G. Kowalska, Acta Sci. Pol. Hortorum Cultus 7, 63–76
(2008).
14. R. Hilgenhof et al., Taxon 72, 811–847 (2023).
15. B. J. Glover, S. Bunnewell, C. Martin, Gene 331, 1–7
(2004).
16. P. Lu, R. Porat, J. A. Nadeau, S. D. O’Neill, Plant Cell 8,
2155–2168 (1996).
17. A. Sessions, D. Weigel, M. F. Yanofsky, Plant J. 20, 259–263
(1999).
18. M. Abe, T. Takahashi, Y. Komeda, Plant J. 26, 487–494
(2001).
130 5 APRIL 2024 • VOL 384 ISSUE 6691
19. M. Abe, H. Katsumata, Y. Komeda, T. Takahashi, Development
130, 635–643 (2003).
20. E. Ogawa et al., Plant Cell Physiol. 56, 1183–1192 (2015).
21. B. Rombolá-Caldentey, P. Rueda-Romero, R. Iglesias-Fernández,
P. Carbonero, L. Oñate-Sánchez, Plant Cell 26, 2905–2919
(2014).
22. N. Kamata, H. Okada, Y. Komeda, T. Takahashi, Plant J. 75,
430–440 (2013).
23. A. H. K. Roeder, A. Cunha, C. K. Ohno, E. M. Meyerowitz,
Development 139, 4416–4427 (2012).
24. H. M. Meyer et al., eLife 6, e19131 (2017).
25. M. Wu et al., Dev. Cell 58, 278–288.e5 (2023).
26. B. Hua et al., Plant Biotechnol. J. 19, 375–393 (2021).
27. C. Yang et al., Proc. Natl. Acad. Sci. U.S.A. 108, 11836–11841
(2011).
28. I. Herskowitz, Nature 329, 219–222 (1987).
29. P. A. Bedinger et al., Sex. Plant Reprod. 24, 171–187 (2011).
30. M. A. Mutschler, B. E. Liedl, Genetic Control of Self-
Incompatibility and Reproductive Development in Flowering
Plants, E. G. Williams, A. E. Clarke, R. B. Knox, Eds. (Springer,
1994), pp. 164–188.
31. I. E. Peralta, D. M. Spooner, S. Knapp, Syst. Bot. Monogr. 84,
1–186 (2008).
32. N. Li et al., Nat. Genet. 55, 852–860 (2023).
33. N. Holstein, M. Gottschling, Flora 240, 7–15 (2018).
34. M. L. Wu et al., Plant Cell koae077 (2024).
35. W. G. Rerie, K. A. Feldmann, M. D. Marks, Genes Dev. 8,
1388–1399 (1994).
36. J. D. Masucci et al., Development 122, 1253–1260 (1996).
37. E. D. Shpak, M. B. Lakeman, K. U. Torii, Plant Cell 15,
1095–1110 (2003).
38. Q. Xin et al., Plant J. 103, 1723–1734 (2020).
39. T. Mukherjee et al., Plant Physiol. 190, 2315–2334
(2022).
ACKN OWLED GMEN TS
We thank Z. Yang and J. Chang for critical comments, A. Wang
for sharing S. habrochaites (LA1777), K. Shi for sharing
S. lycopersicoides, and C. Li for sharing VF36 and the introgression
lines. We thank B. Li and L. Xu for assistance with microCT
scanning. Wild tomato species were obtained from the Tomato
Genetics Resource Center (TGRC). Funding: This work is
supported by the grant from the National Natural Science
Foundation of China (32370354), and Outstanding Talent Fund
and Science Promotion Fund of Fujian Agriculture and Forestry
University (125-118992201A and KSYLX003, to S.W.). Author
contributions: M.L.W. and S.W. conceived and designed the
experiments; M.L.W., X.X.B., B.B.H., Y.D.D., S.R.H., J.Y.S., and
Y.L.W. performed most of the experiments and analyzed the data;
X.X.B., B.B.H., S.R.H., and J.Y.S. performed tomato stable
transformation; M.L.W., X.X.B., and S.W. wrote the article.
Competing interests: The authors declare no competing interests.
Data and materials availability: All data are available in the main
text or the supplementary materials. Tomato mutants and
constructs generated in this study are all available under a material
transfer agreement from the Fujian Agriculture and Forestry
University. Accession Numbers: Woolly, Solyc02 g080260; HD7,
Solyc10 g005330; HD7-LIKE, Solyc03 g031760; HD1, Solyc06
g050160; HD2, Solyc06 g035940; HD3, Solyc03 g026070; HD4,
Solyc03 g120620; HD5, Solyc07 g041850; HD6, Solyc12 g005830;
HD8, Solyc03 g098200; HD9, Solyc06 g072310; HD10, Solyc08
g076370; HD11, Solyc10 g018000; HD12, Solyc09 g066030; HD13,
Solyc11 g011940; CD2, Solyc01 g091630; Style 2.1, Solyc02
g087860; MX1, Solyc01 g010910; MTR1, Solyc10 g083140; ACTIN2,
Solyc11 g005330. License information: Copyright © 2024 the
authors, some rights reserved; exclusive licensee American
Association for the Advancement of Science. No claim to original
US government works. https://www.science.org/about/science-
licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adl1982
Figs. S1 to S10
Tables S1 and S2
References (40–47)
MDAR Reproducibility Checklist
Movie S1
Submitted 4 October 2023; accepted 4 March 2024
10.1126/science.adl1982
science.org SCIENCE
Produced by the Science/AAAS Custom Publishing Office LIFE SCIENCE TECHNOLOGIES
n e w p ro d u c t s : c e l l / t i s s u e c u l t u re
 %$$ $ &%'#($)*%%&$ ,*' $-'$$%%//%$'$
Advanced POCT Kit ,/**$ '$%$/%' $ /$ $./'$&$+'$0%$12($
The SBS Genetech RPA Multiplex Fluorescent 34%& $&&%$%0$'$21 $' $%$ $*4%%$%-"%$
Lyophilized Kit (Gen-4) advances nucleic **% $'$/%,$%$4'/%'$$*%/*$ ,,*'$%0$12$$%%//%$
acid research and point-of-care testing and surrounding regions, broadening access to non-invasive
(POCT). Its multiplex recombinase '  $$' '$' ($34%&$ $4*'$%$&'* '$' * $ /-*$&$
   "$*&  reduce the number of failures due to low fetal fraction, which might
!#  #  % otherwise occur following shipping delays.
2-4 targets within 10 to 20 minutes. This capability supports the Yourgene Health
'   %$$! (+    For info: +44-758-714-0199
and sensitivity, achieving a minimum detection limit of 400 www.yourgenehealth.com
-,.# (0 (/ % # ('1 " 1  
aerosol contamination system, engineered to eradicate aerosol Stable DRI Detectors
    %,   ((   $ */*$%' $$ '$%0$&''*$'0/ '$&'!$
between two contamination prevention systems tailored to meet 2#$&''/% $&'  '&$&$,% '$%$'/'$'$,'0%/'$
( '%)#  '" ( &/,
0 %0$*% $$+ $
3$%$
3$  '($'  $'$
leveraging advanced freeze-drying technology, the product is refractive index of an analyte relative to the solvent—a high-
 ' ( '%0 # &&   ( ' ,'0%/'$2$&''/%$ $&'*$/-%*'& '&$ $%'$%0$'$
thermostability. This allows for shipment at room temperature, % $"'!4*'$&''/% $0%$* &$/%% ,$ '($'$4 '*'$
thereby substantially reducing shipping costs while maintaining  *$ 4*$%0$$* &$/%% ,$&''/%$ $,/**$
product integrity. ,%$ $$ $$* $0/%$$,'-$&''%($''/%$
SBS Genetech stability gains even more importance where analysis time is long, as
For info: +86-10-82784290 is commonly encountered in high-resolution applications. Operating
www.sbsgenetech.com from room temperature up to 80°C with high thermal stability,
   $2$&''/%$ $4''$&'  '&$$,/*$'%$
Replication Competent Virus Kits %$'* $&0($ $$/% ' '/'$%0$ $/'&$4 '*'$
Bio-Rad Laboratories has announced the recent launch of the 4*$'$2$&''/%$&'* ' $'!/'**'$,'0%/'$0%$' '$
   !" !# " $"&()++ (- * ( /**'  $/%% ,/$ ',% ($'  '&$0/'&$
!&  ' !(.& (! !"( ,!  /(!  %"!" &$ *$' '&$$+'$  $ */*$2$&''/% $'$
of replication competent lentivirus (RCL) and replication competent available on competitive delivery times.
"!((! &( ++ ' ( !"  (, " .& TESTA Analytical
production of cell and gene therapies. Replication competent viruses For info: +49-30-864-24076
(RCVs) can infect cells and replicate to produce additional virions. www.testa-analytical.com
RCVs can be generated through recombination events between
the transfer plasmid and either the production plasmids or the Multi-Layer Microarray Platform
"  "( "!#- "#" !  (, !, #" Arrayjet has announced that its ArrayPlex discovery platform has
cell and gene therapies, RCV testing of samples must be performed been selected by Immunome, Inc. to expand its in-house high-
!!  !!" !((- !(   ( throughput screening tools. The proprietary multi-layer microarray
for detection of RCL and RCAAV provide a validated solution for   "    
  (  "   ("(& !" " %"!" !,  "     
viral replication genes, potentially indicating an RCV. The tests are and performing screening in-house. ArrayPlex was created by
(" ,! !( 
!     ,!#('     # $$  
  (#  ! ! " " !#  " (( "  ! " %&   
/  (#  ! - successful 7-year partnership with Immunome, Arrayjet performed
Bio-Rad screening services using the ArrayPlex platform, and the platform
For info: +1 510-356-7909 underwent further optimization, most recently to integrate internal
www.bio-rad.com assay controls and enhanced data normalization. Arrayjet is pleased
that Immunome has now chosen to purchase an instrument and
Expanded Prenatal Testing internalize the ArrayPlex assay as an in-house, high-throughput
 """!,!".$!%& %!("#2!#    (% "%
(#!")"+#*($!#($""!!"#( ,./'2-03-( therapies. In addition to screening between antibody and lysate
%%4!#")"/."#!" "")" ( (" (!) libraries, the platform is also compatible with a broad spectrum of
, !4(!)2  &"( (##"$(%"%!& !4(!)%!4!)" other sample types (e.g. small molecule libraries, protein panels,
(#"( #!)$()"" !""!#"(!4""##   %   %  $   " 
that reduce the need for invasive tests and the associated stress  %'
&
!)!1("&"1"%!!"#  ""#/."#"") Arrayjet
)""!" "#4& !4(!)(#!!)$!%")"!! For info: +44-131-440-5220
#%""( "#"#(!""(#0!!"#!&4"!"%")-( www.arrayjet.com
(#&,
-##&)"2(#& ,)-!)##&)"2

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.

SCIENCE https://www.science.org/content/page/new-products-section 5 APRIL 2024 • VOL 384 ISSUE 6691 131

ScienceCareers.org

Science Careers helps you advance

your career. Learn how !
 Register for a free online account on
ScienceCareers.org.
 Search hundreds of job postings and
find your perfect job.
 Sign up to receive e-mail alerts about job
postings that match your criteria.
 Upload your resume into our database and
connect with employers.
 Watch one of our many webinars on
different career topics such as job searching,
networking, and more.
 Download our career booklets, including
Career Basics, Careers Beyond the Bench,
and Developing Your Skills.
 Complete an interactive, personalized career
plan at “my IDP.”
 Visit our Employer Profiles to learn more
about prospective employers.
 Read relevant career advice articles from our
library of thousands.

Visit ScienceCareers.org
today — all resources are free

SCI E N CECA RE E RS.O RG


Tel.: +86-28-84078792/84078793
Contact: Mr./Ms. Li, Kang & Zhang
Email: hr@cdut.edu.cn
Address: No.1 East Third Road, Erxianqiao,
Chenghua District, Chengdu, Sichuan (610059)
Scan the QR Code to follow CDUT on WeChat
ScienceCareers.org
RECRUITMENT OF HIGH-LEVEL
TALENTS BY CDUT
Chengdu University of Technology (CDUT) is one of China's first and second batches of Double-First Class universities.
It is located in Chengdu, a famous historical and cultural city, national central city, and the powerhouse of the "Cheng-
du-Chongqing Economic Circle". CDUT was previously known as Chengdu College of Geological Exploration, which was
founded in 1956 as one of the three geological colleges in the early days of the founding of the People's Republic of China.
Currently, it has two state-key laboratories (State Key Laboratory of Geohazard Prevention and Geoenvironment Protection,
and National Key Laboratory of Oil and Gas Reservoir Geology and Exploitation), 33 provincial and ministerial science and
technology platforms under the Ministry of Education, the Ministry of Natural Resources, the Ministry of Ecology and
Environment, the Ministry of Emergency Management, and Sichuan Province. CDUT attaches great importance to interna-
tional exchanges and cooperation and has successively established exchange and cooperation relationships with over 180
universities and research institutions in the UK, Canada, Russia, the U.S., Australia, New Zealand, the Netherlands, Italy, and
other countries.
For more information about CDUT, please visit the official website at http://www.cdut.edu.cn.
A wealth of talents holds the key to the success of a great cause. While striving to construct "a world first-class universi-
ty with first-class disciplines" in the new era, CUDT is willing to provide a broad stage for all talents who are interested in
starting a new career and carrying out educational and scientific research and strives to create a new platform where "talents
from all over the world can gather together to display their capability".
CDUT is now globally recruiting well-known scholars with international influence, leading experts with innovation
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.
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.

Interviewing students for my first and share ideas, materials, and

batch of Ph.D. candidates didn’t links—and vice versa. Before they
bode well. Most of the applicants get going on any experiment, I
lived far from my university and ask them to write a detailed pro-
interviews were virtual. I have tocol as well. That has allowed
trouble lip-reading through 2D me to double-check and intervene
camera images, so I had to depend whenever I spot something that
on unreliable captions. Sometimes I must have missed during our
I had to guess what the students verbal discussions.
said. Luckily other colleagues were Initially, I feared my students
on the call and could fill me in on would feel that writing out all of
what I missed. But the experience their experimental design steps
was unsettling. and data analysis plans was an un-
I took a chance and offered po- necessary burden, something they
sitions to the candidates I thought would not have been subjected to
were best. Some declined, and I under a different supervisor. So, I
wondered whether my disability
factored into their decision. But
“Fostering multiple lines made an effort to ensure it would
benefit them as well: I taught them
others, including some who had dis-
abilities themselves, said yes, and
of communication can benefit about preregistration—the practice
of publishing a study plan in ad-
I’ve been able to slowly grow my lab. all mentors and trainees.” vance of conducting the research—
When my first student arrived, I and encouraged them to prepare
thought back to how I engaged with my supervisors earlier their written protocol with the intention of publishing it.
in my career. Worried that I’d missed key tidbits of informa- They’ve told me this step has been helpful because it allows
tion when they gave me instructions, I often visited their of- them to deepen their understanding of the project before
fices multiple times per day to ask for clarification. I realized starting experiments.
I couldn’t work that way with all my students. So I needed It’s still early in my faculty career. But I feel these strate-
to implement other safeguards against misunderstanding. gies are working well for me. I no longer worry that I’m
In addition to regular one-on-one meetings with my stu- missing critical information. And I’ve been able to guide
dents, I decided to have each student work with me for students in a way that would have been difficult for me
entire days in the lab when they first join my group. That to do through verbal communication alone. Looking back,
way, they can see what I’m doing and ask questions as we I wish I’d followed these strategies earlier in my career.
ILLUSTRATION: ROBERT NEUBECKER

go along. And time with the students allows me to get to
know them better and carefully explain the rationale be-
hind each experimental step.
I also established written communication as the most
important way to exchange information. Through instant
messaging services, I can quickly reach out to my students
Fostering multiple lines of communication can benefit all
mentors and trainees, preventing misunderstandings that
conversation alone can lead to. j
Denis Meuthen is a Freigeist fellow and a research group leader at Biele-
feld University. Send your career story to SciCareerEditor@aaas.org.

134 5 APRIL 2024 • VOL 384 ISSUE 6691 science.org SCIENCE

Publish your research in the Science family of journals
The Science family of journals (Science, Science Advances, Science Immunology, Science
Robotics, Science Signaling, and Science Translational Medicine) are among the most highly-
regarded journals in the world for quality and selectivity. Our peer-reviewed journals are
committed to publishing cutting-edge research, incisive scientific commentary, and insights
on what’s important to the scientific world at the highest standards.

Submit your research today!

Learn more at Science.org/journals
Call for Proposals:
Shape the Future of Science
Present your groundbreaking research, lead engaging workshops,
and connect with a diverse community of experts:
Submit proposals for the 2025 AAAS Annual Meeting.

WE ARE SEEKING PROPOSALS FOR:

Scientific Sessions: Discuss your research with fellow experts

in multi-disciplinary panels. Share your findings and engage in
stimulating discussions on topics that advance knowledge and
address societal needs.

Workshops: Share your expertise and guide attendees through

interactive learning experiences. Workshops are perfect for
developing skills, exploring new technologies, and advancing
careers in science and engineering.

Deadline for submissions: Monday, May 6, 2024.

Visit aaas.org/meetings to learn more and
submit your proposal today!