CONTENTS
3 0 J U N E 2 0 2 3 • VO LU M E 3 8 0
I S S U E 6 6 52
1313
A rat kidney goes into cold storage.
NEWS
IN BRIEF
1034 News at a glance
IN DEPTH
1306 Long-sought gravitational wave hum
finally detected
Subtle shifts in stellar signals reveal pervasive
waves from mergers of giant black holes
By A. Mann
PHOTOS: (FROM TOP) EVAN TAYLOR STUDIOS ASSISTED BY TOM OKINS; RITESH SHUKLA/GETTY IMAGES
1308 Scientists, U.S. counter claims
of pandemic patient zero at lab
Intelligence report offers little new on
SARS-CoV-2’s origin By J. Cohen
1309 Consciousness hunt yields results
but not clarity
Two rival theories of how it emerges
went head-to-head in brain-scanning
experiments By E. Finkel
1310 WHO’s new chief scientist wants
to look to ‘tomorrow’
Jeremy Farrar says more must be done
about COVID-19, but agency should aim to
anticipate its next major challenges
By K. Kupferschmidt
1311 U.S. military contract could boost
status of Black university
Defense research institute aims to turn
Howard into a top research university
By J. Mervis
SCIENCE science.org
FEATURES POLICY FORUM
1313 Frozen in time
Scientists are learning how to cryopreserve
1324 Emissions and energy impacts
of the Inflation Reduction Act
living tissues, organs, and even whole Economy-wide emissions drop 43 to 48%
organisms, then bring them back to life below 2005 levels by 2035 with accelerated
By W. Cornwall clean energy deployment By J. Bistline et al.
PODCAST
BOOKS ET AL.
1328 Water, now and then
INSIGHTS Reflecting on humanity’s intertwined history
with water, a scientist offers actionable
advice for meeting future needs By M. Aczel
PERSPECTIVES
1329 Thalidomide in America
1318 Galactic neutrinos in the Milky Way A journalist corrects the record on US
A source of neutrinos may lie within the exposure By J. Olszynko-Gryn
midplane of the Galaxy By L. A. Fusco
REEARCH ARTICLE p. 1338
1320 A new class of antiparasitic drugs
Cyanotriazole compounds “poison”
topoisomerase II of pathogenic
trypanosomatids
By R. Aphasizhev and I. Aphasizheva
REEARCH ARTICLE p. 1349
1321 Mapping RNA translation
A new method maps the location of
thousands of translating RNAs in cells and
tissues By R. Fan
REEARCH ARTICLE p. 1337
1322 Staving off cell death
A signaling pathway that senses energy
stress opposes necroptotic cell death
By D. G. Hardie
REEARCH ARTICLE p. 1372
1328
30 JUNE 2023 • VOL 380 ISSUE 6652 1299
 CONTE NTS

1318 & 1338

LETTERS 1344 Hydrology DEPARTMENTS

1330 Editorial Expression of Concern Agricultural expansion raises groundwater 1303 Editorial
By H. H. Thorp and increases flooding in the South American Not teaching evolution is an injustice
plains J. Houspanossian et al. By L. S. Shashidhara and A. Joshi
1330 Will Canada permit killer whale
extinction? By M. MacDuffee et al. 1349 Microbiology 1398 Working Life
Cyanotriazoles are selective topoisomerase What I owe my mentor
1330 Border conservation in Hindu II poisons that rapidly cure trypanosome By A. Joshi
Kush-Himalaya By X . Chen and V. Hull infections S. P. S. Rao et al.
PERSPECTIVE p. 1320

1331 Past as Prologue: A family’s pride

in educated daughters By Q. Tul Ain 1357 Chromatin biophysics
Stochastic motion and transcriptional
1398
1332 Technical Comment abstracts dynamics of pairs of distal DNA loci on a
compacted chromosome D. B. Brückner et al.

RESEARCH 1363 Martian geology

Gullies on Mars could have formed by melting
of water ice during periods of high obliquity
J. L. Dickson et al.
IN BRIEF
1367 2D materials
1333 From Science and other journals Observation of Rydberg moiré excitons
Q. Hu et al.
RESEARCH ARTICLES ON THE COVER
1336 Genome evolution 1372 Signal transduction Vapor from liquid nitrogen wafts over a rat
Virus-like transposons cross the species Metabolic orchestration of cell death by kidney awaiting a groundbreaking preserva-
barrier and drive the evolution of genetic AMPK-mediated phosphorylation of RIPK1 tion method at the University of Minnesota.
incompatibilities S. A. Widen et al. T. Zhang et al. Scientists there have learned how to cool the
CREDITS: (PHOTO) PHOTO: BRUNO GILLI/ESO/CC BY; (ILLUSTRATION) ROBERT NEUBECKER

PERSPECTIVE p. 1322 organ to –150°C and rewarm it while minimiz-

RESEARCH ARTICLE SUMMARY; FOR FULL TEXT:
DOI.ORG/10.1126/SCIENCE.ADE0705
ing freezing damage, enabling it to work after
1381 Quantum sensing being transplanted. This
1337 Spatial omics Improving metrology with quantum laboratory and others are
scrambling Z. Li et al. harnessing extreme cold
Spatially resolved single-cell translatomics at
to preserve everything
molecular resolution H. Zeng et al. from organs to endan-
RESEARCH ARTICLE SUMMARY; FOR FULL TEXT: 1384 Chiral materials gered coral to fresh
DOI.ORG/10.1126/SCIENCE.ADD3067 A magnetic assembly approach to chiral tomatoes. See page 1313.
PERSPECTIVE p. 1321 superstructures Z. Li et al. Photo: Evan Taylor Studios
assisted by Tom Okins
1338 Neutrino astrophysics 1390 Gamma-ray bursts
Observation of high-energy neutrinos from A tera–electron volt afterglow from a narrow
the Galactic plane IceCube Collaboration jet in an extremely bright gamma-ray burst Science Staff ............................................1302
PERSPECTIVE p. 1318 LHAASO Collaboration Science Careers ....................................... 1397

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 © 2023 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): $2411; 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 30 JUNE 2023 • VOL 380 ISSUE 6652 1301

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

1302 30 JUNE 2023 • VOL 380 ISSUE 6652 science.org SCIENCE

 ED ITORIAL
Not teaching evolution is an injustice
S
ince April, India has been roiled by controversy
around the excision of several topics, including
evolution and the periodic table, from school text-
books (up to grade 10) by the National Council for
Educational Research and Training (NCERT). This
was projected as an exercise in “rationalization” of
content aimed at reducing the study load on stu-
dents. The move was opposed by large numbers of aca-
demics and worried citizens. As the exclusion of specific
topics in history and contemporary politics appeared to
be in line with the ideology of the ruling party, many crit-
ics assumed that the removal of science topics was also
ideologically motivated. In turn, this spurred supporters
of NCERT and the government to dismiss all criticism as
being entirely political, rather than
academic. Both sides in this debate
have traded exaggerated accusations
of mala fide intent, leading to crucial
broader issues being obscured.
“…limiting
In 1947, upon attaining indepen-
dence from British rule, India was
the teaching
a poor country of about 340 million
people, with only 12% literacy. Yet,
of evolution in
its commitment to science educa-
tion and research, and the use of
India…is unfair
science and technology to solve soci-
etal problems, have enabled India to
to students
become the fifth-largest economy in
the world. With the help of expertise and a travesty…”
in basic and applied sciences across
all disciplines, Indian scientists have
contributed substantially to advances made in technol-
ogy-intensive areas such as food and nutrition, health
care, energy, information technology services, and space.
However, when it comes to teaching and research in
evolution, India has begun to show signs of irrationality
and extreme religiosity, reminiscent of what happened in
parts of the United States earlier and even now. Opposi-
tion to evolution in India is relatively recent, influenced
in part by notions of intelligent design. Traditionally, un-
like Abrahamic religions, Indic religions did not perceive
evolution or Darwinism as a threat to their beliefs. The
emergence of specifically anti-Darwinian stances, even
while often still accepting evolution as a general process,
may partly be motivated by the urge among some in India
to claim that all major scientific insights can be traced to
ancient India. This parochial nationalism, resulting in an
uncritical ascription of most scientific knowledge to an-
cient India, creates a particularly acute problem when it
leads to the denial of much of our modern understanding
of evolution. This emergence of anti-evolutionary ideas is
SCIENCE science.org
also particularly unfortunate as Indian evolutionary bi- L. S. Shashidhara
ologists have contributed substantially to the field as well is director of the
as to national and global problems of infectious diseases, Tata Institute
biodiversity, and conservation, and have helped charac- of Fundamental
terize India’s population diversity. Research–National
India should not forget that technology is science- Centre for
driven and, therefore, investment in technology will not Biological Sciences,
yield dividends if basic science is ignored. Evolution is Bengaluru, India.
widely recognized as important not only to academic bi- lsshashidhara@
ology, but also to understanding, managing, and solving
ncbs.res.in
many challenges that societies face worldwide. These
include multidrug resistance in microbes, aging, in-
Amitabh Joshi
creasing incidence of cancers, zoonotic pandemics, and
ecological problems stemming from a warming world is chair of the
and anthropogenic environmental Evolutionary
degradation. Evolution also directly and Organismal
addresses the manner in which hu- Biology Unit of the
mans conceive of themselves; their
origins, spread, identity, diversity,
Jawaharlal Nehru
Centre for Advanced
and health; and their relationships
with other species. It is also one of
Scientific Research,
Bengaluru, India.
the pillars of a rational worldview,
in contradistinction to a supersti-
ajoshi@jncasr.ac.in
tious or mythological view. In an as-
pirational and rapidly modernizing
country such as India, whose con-
stitution exhorts citizens to nurture
the scientific temper, limiting the
teaching of evolution in India to the
small subset of students who opt to
take biology in grades 11 and 12 will
result in the vast majority of students completing their
basic schooling with little or no exposure to evolution-
ary thought. This is unfair to students and a travesty of
a modern, liberal, rational, and evidence-based educa-
tion that is, ironically, envisioned in India’s New Educa-
tion Policy, 2020.
It is important that the entire scientific community,
and not just evolutionary biologists, respond to this issue.
In recent decades, biologists in India and worldwide have
emphasized reductionist molecular and cellular biology,
often ignoring evolution and ecology, thereby limiting ef-
forts to tackle important biological challenges faced by so-
cieties. Focusing only on applied and not basic research,
and fueling hypernationalism, are problems not just in
India, but in many other countries as well. Consequently,
if the strong, unified pushback against the exclusion of
fundamental concepts of evolution from grade 10 text-
books in India were to trigger a reversal of the decision,
it may also prove helpful beyond India.
–L. S. Shashidhara and Amitabh Joshi
10.1126/science.adj3557
30 JUNE 2023 • VOL 380 ISSUE 6652 1303
 1.3 billion
NEWS Adults expected to have diabetes globally by 2050—more than double
the current number—because of rising obesity and growing health inequality,
according to studies recently published in The Lancet journals.

IN BRIEF U.S. OKs sale of lab-grown chicken

Edited by Matthew Warren and Michael Price
FOOD SCIENCE | U.S. consumers will soon
get their first taste of lab-grown meat, after
the U.S. Department of Agriculture gave
two companies—Upside Foods and Good
Meat—permission to sell chicken produced
from cells grown in bioreactors. Last
week’s approval follows earlier decisions by
the Food and Drug Administration stating
that the chicken is safe to eat. The compa-
nies plan to initially sell the meat, which
will be labeled “cell cultivated chicken,”
only to select restaurants. The United
States is the second country to approve a
lab-grown meat product, after Singapore
gave the green light to Good Meat chicken
in 2020. The Good Food Institute, an
industry think tank, predicts other govern-
ments will soon follow suit.

Fighting racial bias in medicine

PHILANTHROPY | A new $10 million fund-
ing initiative will support five scientific
organizations as they investigate racial
biases in medical decisions. The grants
represent the largest single investment
in such research to date, says the Doris
Duke Foundation, which launched the
initiative this week. Physicians routinely
use algorithms that take into account
At a 2016 U.S. FDA advisory panel meeting, Jen McNary and her son, Austin Leclaire (left), spoke in
a patient’s race when assessing risk of
support of an earlier experimental muscular dystrophy treatment developed by Sarepta Therapeutics.
disease, making a diagnosis, or deciding
on a course of treatment, but the conse-
quences of including race as a variable
BIOMEDICINE are poorly understood. The newly funded
projects will explore how these algo-
Muscular dystrophy therapy approved rithms may exacerbate health disparities
and how medical organizations can
adjust their guidelines around the use

T
he U.S. Food and Drug Administration has approved the first
of algorithms to improve health equity.
gene therapy for Duchenne muscular dystrophy (DMD), a genetic Grant recipients include the American
disease that cripples boys and usually results in death by age 30. Academy of Pediatrics, the American
The treatment from Sarepta Therapeutics introduces a short ver- Heart Association, and the Coalition to
sion of the gene for dystrophin, a crucial muscle protein, which End Racism in Clinical Algorithms.
is mutated in patients with DMD. A one-time intravenous infu-
sion of a virus delivers the functioning “microdystrophin” gene into HIV prevention disparities persist
patients’ muscle cells. The 22 June approval is only for boys 4 to 5 years | The use of anti-HIV
P U B L I C H E A LT H

old, a group that appeared likely to benefit from the drug based on their drugs to prevent infection remains
PHOTO: JOHN BOAL

disproportionately low among Black and

elevated levels of the microdystrophin protein in a clinical trial. The Hispanic people in the United States,
approval could be revoked if an ongoing trial does not show improved according to a new analysis of pharmacy
muscle function. A single infusion will cost $3.2 million, Sarepta says. and consumer databases. Pre-exposure

1304 30 JUNE 2023 • VOL 380 ISSUE 6652 science.org SCIENCE

 The critically endangered
Saharan addax is hunted for
its skin, horns, and meat.

ECOLOGY

Scientists size up human predatory footprint

H
umans are the ultimate predators, trapping, hunting, or oth-
erwise exploiting 15,000 species of vertebrates—300 times
more species than jaguars and 113 times more than great
white sharks. These numbers, which represent about one-
third of all vertebrates, come from the first broad-scale
analysis of humans’ toll on these animals, published this week
in Communications Biology. The analysis, based on data from
the International Union for Conservation of Nature, also shows
that more than half of exploited terrestrial species—particularly
birds, reptiles, and amphibians—are caught live and sold as pets.
Because of other factors such as climate change and habitat
destruction, the researchers expect the proportion of targeted
species threatened with extinction—which currently sits at 39%—
will grow. Those most at risk, including large grazers such as the
addax (Addax nasomaculatus) in the Sahara Desert, fill unique
roles in ecosystems and are irreplaceable, says lead author Chris
Darimont, a conservation scientist at the University of Victoria
and the nonprofit Raincoast Conservation Foundation.

prophylaxis (PrEP), which protects
uninfected, at-risk people from HIV, has
become increasingly popular in the U.S.
since receiving regulatory approval in
2012. Data published last week on AIDSVu,
an interactive map created by Emory
University researchers, show that nation-
wide use increased 20% from 2021 to
2022. But only 14% of PrEP users are Black
and 17% are Hispanic, even though these
groups make up 42% and 27% of new HIV
diagnoses, respectively. AIDSVu released
the new data on the eve of National HIV
Testing Day, 27 June, a public health initia-
tive that encourages the wider use of PrEP.
Windpipe transplanter sentenced
| Surgeon Paolo
R E G E N E R AT I V E M E D I C I N E
Macchiarini last week was sentenced to
PHOTO: ROLAND SEITRE/MINDEN
Institute in the early 2010s. At the time, the
procedure was hailed as a breakthrough in
regenerative medicine, but all the patients
died after the implants failed and by 2016
Macchiarini had been fired amid allegations
of fraud and ethical violations. The verdict,
announced on 21 June, overturns a 2022 rul-
ing that found Macchiarini guilty of “bodily
harm” in two cases and gave him a sus-
pended sentence. Macchiarini’s lawyer has
said the defense team will appeal the ruling
to the Swedish Supreme Court.
Building a better TB vaccine
| A large-scale efficacy
P U B L I C H E A LT H
trial of a promising tuberculosis (TB)
vaccine will soon launch with substantial
backing from two major philanthropies.
The vaccine, dubbed M72, has so far
developed more than a century ago offers
some protection to young children, but
does little to prevent pulmonary disease
in adolescents and adults.) The new trial
will enroll 26,000 people at more than
50 sites in Africa and Asia. In 2020, a
research branch of the Bill & Melinda
Gates Foundation acquired the license
from the vaccine’s initial developer, GSK,
to further develop the candidate and then,
if it proves its worth in efficacy trials,
make it available to poor countries hit
hardest by TB. The Gates foundation will
foot $400 million of the estimated
$550 million needed to complete the
multiyear trial, with the Wellcome Trust
making up the balance.
Fauci to join Georgetown faculty

2 years and 6 months in prison by a Swedish
appeals court after it found him guilty
of gross assault against three patients into
whom he implanted artificial windpipes
seeded with stem cells. Macchiarini per-
formed the experimental surgeries on the
patients while working at the Karolinska
crawled through clinical trials. A placebo-
controlled study completed in 2015 found
that in adults who had latent infections
with the TB-causing mycobacterium,
those who received the vaccine had half
as many cases of the life-threatening pul-
monary form of the disease. (A TB vaccine
| Anthony Fauci, who in
ACA D E M I C S
December 2022 stepped down as director
of the National Institute of Allergy and
Infectious Diseases and as chief medical
adviser to President Joe Biden, announced
this week that he will be a professor at
Georgetown University’s School of Medicine.

SCIENCE science.org 30 JUNE 2023 • VOL 380 ISSUE 6652 1305

 IN DEP TH
ASTRONOMY
Long-sought gravitational
wave hum finally detected
Subtle shifts in stellar signals reveal pervasive
waves from mergers of giant black holes
By Adam Mann
B
y turning networks of dead stars into
galaxy-size gravitational wave detec-
tors, radio astronomers have tuned
into the slowly undulating swells
in spacetime thought to arise from
pairs of supermassive black holes
(SMBHs) that are about to collide.
In a simultaneous announcement on
28 June, five separate international teams
said that after nearly 20 years of effort they
had found evidence for these gravitational
waves. They are far longer than the waves
first captured by ground-based detectors
in 2015, which emanate from collisions
of star-size objects. The findings not only
open up a new window in gravitational
wave astronomy, but will also help re-
searchers answer questions about the ori-
gin and evolution of SMBHs, objects that
sit at the center of galaxies and weigh as
much as billions of Suns. The results could
even uncover hints of new physics.
“I’m so thrilled to live through the era
of not only one [gravitational wave] detec-
tion, but multiple detections,” says Vicky
Kalogera, an astrophysicist at Northwest-
ern University who wasn’t involved with
1306 30 JUNE 2023 • VOL 380 ISSUE 6652
the work. “This is really epic,” agrees
Jason Hessels of the University of Amster-
dam, who formerly worked with one of the
teams announcing the findings.
The new results rely on millisecond pul-
sars, highly magnetized stellar remnants
that emit beams of radiation as they spin
as fast as 1000 times per second. Their
lighthouse-like radio signals sweep past
our planet with a regularity that rivals an
atomic clock. Should a passing gravitational
wave intrude anywhere along the path be-
tween Earth and the pulsar, it will stretch
and squash the fabric of space—and cause
small variations in the timing of the flashes.
“Those ultraprecise signals will arrive a
little bit early, and then a little bit late,”
says Chiara Mingarelli, an astrophysicist at
Yale University and a member of the North
American Nanohertz Observatory for Gravi-
tational Waves. NANOGrav is one of the five
international pulsar timing arrays (PTAs),
which draw on data from a dozen radio tele-
scopes around the world to monitor dozens
of the beacons for evidence of the waves.
The challenge is immense because of
the many potential corrupting sources of
noise: radio interference from terrestrial
technology and satellites, scattering of the
Like buoys in an ocean, pulsars bob on undulating
gravitational waves from supermassive black holes.
pulsar’s radio beams by intervening clouds
of gas, and tiny random variations in the
spins of the pulsars themselves. “You have
to beat down all these confusing signals,”
Hessels says.
But NANOGrav investigators now feel
they have strong evidence for the waves, in
a 15-year data set of 67 pulsars located up
to 20,000 light-years away. They identified
anomalies of one part in a quadrillion—
comparable to measuring the distance
between Earth and the Moon to within ILLUSTRATION: AURORE SIMONNET/NANOGRAV COLLABORATION
1/1000th of 1 millimeter. The team is report-
ing the detection at a 3.5- to 4-sigma level,
indicating more than 99% confidence that
the signal is real. However, this falls short
of the 5-sigma threshold that physicists of-
ten want before claiming a discovery. “As a
group we’ve agreed to avoid the D-word,”
says team member Scott Ransom of the
National Radio Astronomy Observatory.
NANOGrav’s results were published in
five papers in The Astrophysical Journal
Letters, while simultaneous papers ap-
peared in other journals from the European
PTA, the Australian PTA, the Indian PTA,
science.org SCIENCE
 NE WS

and the Chinese PTA. The data sets are really going into new territory and set- from NASA’s JWST space telescope hinting
different: NANOGrav monitored the most ting bounds and constraints in ways which that galaxies grew larger and faster in the
pulsars, the European PTA had the longest have never been done before.” early universe than once thought.
running data set, and the Australian PTA If galaxy mergers really are the source of Each PTA has its own quirks and sources
was the only effort watching pulsars in the the background, the finding would settle a of noise, says Priyamvada Natarajan, a theo-
southern skies. But, “We’re confident about lingering question. Theoretical work in the retical astrophysicist at Yale. But because
the result because we’re all seeing the 1980s suggested pairs of orbiting SMBHs the five collaborations together monitor
same thing,” says Michael Kramer, a leader would slowly get closer as they lost energy more than 100 pulsars, “a combined analy-
of the European PTA at the Max Planck In- to gravitational waves. But they would even- sis would likely help us cross the [5-sigma]
stitute for Radio Astronomy. tually stall out just a few light-years apart, threshold to an actual detection.” That could
The signal is very different from what where they would hang on in their dance happen in the next year or two, she thinks.
ground-based detectors such as the Laser for longer than the age of the universe be- By mapping how the overall background
Interferometer Gravitational-wave Observa- fore approaching close enough to generate varies on the sky, the PTAs might eventu-
tory (LIGO) pick up. Those are sensitive to strong gravitational waves at frequencies ally get good enough to detect individual
the milliseconds-long blasts of waves pro- most detectable by the PTAs. The fact that SMBH pairs in the nearby universe, en-
duced by individual mergers of black holes the PTAs detected a signal at all suggests abling other telescopes to zoom in to study
and neutron stars with masses tens of times that friction from gas and dust—or other what happens in galaxies after they merge.
that of the Sun. The pulsar timing results in- stars and black holes—must be helping More pulsars would help. The Canadian
stead represent many waves, each years to the SMBHs shed energy and pass the stall- Hydrogen Intensity Mapping Experiment,
decades long, overlapping in a cumulative out threshold. a radio survey telescope, is on the hunt,
background hum. Each wave is generated The background signal is stronger than and China’s Five-hundred-meter Aperture
by a pair of SMBHs somewhere in the uni- expected, implying that this population of Spherical radio Telescope, the world’s big-
verse whose host galaxies have collided and imminently merging SMBHs numbers in gest, is finding pulsars too faint for other
merged. The black holes orbit one another the hundreds of thousands, perhaps even facilities. The Deep Synoptic Array, a pro-
for tens of millions of years, emitting gravi- millions. It could also mean the SMBHs posed collection of 2000 radio dishes in the
tational waves in the lead-up to an eventual are either bigger or merging more quickly Nevada desert, could add to the harvest.
merger. (Unlike LIGO, the PTAs are not sen- than astronomers previously predicted— The European Space Agency’s Laser
sitive to the mergers themselves—just the a possibility consistent with observations Interferometer Space Antenna (LISA), a
death spirals.) space-based detector, might
The background signal help fill out the spectrum of
could include other, more ex- gravitational waves. LISA,
otic sources. Primordial gravi- A gravitational wave spectrum which is scheduled to launch
The 100-meter-wide Green Bank Telescope in West Virginia was a critical tool
tational waves might have in 2037, would be able to see
in the pulsar timing arrays that have now detected long-period gravitational
been generated shortly after gravitational waves in a wave-
waves, likely from gargantuan black holes. Detectors on the ground and in
the big bang if the universe space are sensitive to shorter period waves from other sources.
length range between those
experienced convulsions from observed by LIGO and the
a still-hypothetical breakneck PTAs. These would come from
DETECTOR TYPE PERIOD SOURCES
expansion in size known as mergers of intermediate-size
inflation. Some theorists also Pulsar timing Years to Merging supermassive black holes, those between
posit that the violent emer- arrays decades black holes (SMBHs), big bang 100,000 and 10,000,000 times
gence of one of the funda- ripples, exotic new physics our Sun’s mass, such as the
mental forces at the dawn of one that lives in the center of
Space-based Hours to Midsize SMBHs, stellar-size black
time left defects in spacetime interferometers seconds hole mergers, white dwarf mergers the Milky Way. “The hope is
called cosmic strings, which that we can paint a picture of
would now stretch across the Ground-based Milliseconds Stellar black hole mergers, merging the entire gravitational wave
universe, vibrating like piano interferometers neutron stars, supernovae landscape, reaching from very
wires and emitting gravita- low frequencies all the way up
tional waves in the range that to very high frequencies,” says
PTAs are sensitive to. Kai Schmitz, a NANOGrav
Certain kinds of dark team member at the Univer-
matter—the mysterious sub- sity of Münster.
stance that makes up 80% of Having heard the cosmic
the matter in the universe— hum, the teams now want to
might also produce the long- listen more closely, digging
period oscillations. The PTA deeper into their data sets to
results already help rule out find out what else they might
some lightweight dark mat- learn. “We’re not even close to
ter particles, which would the end of the story,” Hessels
have produced a bigger says. “It’s a game of dedica-
PHOTO: JEE SEYMOUR

anomaly than was observed, tion and patience.” j

says Andrea Mitridate, a
NANOGrav team member With reporting by Daniel Clery.
at DESY, Germany’s particle Adam Mann is a journalist in
physics laboratory. “This is Oakland, California.

SCIENCE science.org 30 JUNE 2023 • VOL 380 ISSUE 6652 1307


COVID-19
Scientists, U.S. counter claims
of pandemic patient zero at lab
Intelligence report offers little new on SARS-CoV-2’s origin
By Jon Cohen
T
he fiery, politicized debate over
whether COVID-19 began with a labo-
ratory mishap flared anew last week. A
Wall Street Journal (WSJ) article said
anonymous sources had confirmed
an online newsletter’s report naming
three scientists at China’s Wuhan Institute of
Virology (WIV) as the pandemic’s first cases
in November 2019, supposedly infected dur-
ing a risky coronavirus research project. Ma-
jor newspapers, TV stations, and social media
spread the charges. But Ben Hu, alleged by
some outlets to be the first case at WIV, de-
nied that he was ill in late 2019 or that his
work had any link to the emergence of SARS-
CoV-2. And a declassified U.S. intelligence
report on COVID-19’s origin released on
23 June did not name him or substantiate
that any WIV scientists had the initial cases.
“The recent news about so-called ‘patient
zero’ in WIV are absolutely rumors and ridic-
ulous,” Hu emailed Science in his first public
response to the charges, which the newsletter
Public reported on 13 June, citing unnamed
U.S. government officials. A colleague of Hu’s
whom Public also named denies having had
one of COVID-19’s first cases.
Republicans in Congress may have thought
classified intelligence would tell a different
story about the long-rumored “sick” WIV
scientists. Earlier this year, they introduced
a bill—which Congress passed and President
Joe Biden signed into law on 20 March—
that ordered declassification and release of
1308 30 JUNE 2023 • VOL 380 ISSUE 6652
COVID-19 origin information within 90 days.
The law asked for the names and other de-
tails of any WIV researchers who were ill just
before COVID-19 surfaced publicly in Wuhan.
The report, issued by the Office of the Di-
rector of National Intelligence (ODNI), says
some at WIV were sick at that time with
“symptoms consistent with but not diagnos-
tic of COVID-19.” But it doesn’t identify the
three scientists and it further states, “We have
no indications that any of these researchers
were hospitalized because of the symptoms
consistent with COVID-19.”
Hu is an appealing suspect for those who
believe the pandemic virus leaked from WIV,
rather than coming from infected animals at
a Wuhan market, the other main theory. Hu
and the other WIV scientists named in media
reports, Yu Ping and Zhu Yan, worked in a
lab run by Shi Zhengli, a coronavirus scientist
long the target of allegations she has covered
up a lab leak. Hu was also a lead author on
a 2017 paper in PLOS Pathogens describing
an experiment that created chimeric viruses
by combining genes for proteins from bat
coronaviruses that would not grow in cul-
tures with the genome of one that did.
This paper has received intense scrutiny
because it was partially funded by the U.S.
National Institutes of Health (NIH) and, lab-
leak proponents insist, resulted in a virus
more dangerous to humans. NIH officials
have denied this and noted that the viruses
created were not close SARS-CoV-2 relatives.
Hu tells Science he never worked with
live viruses in that experiment or any others
Whether SARS-CoV-2 came from the Wuhan Institute
of Virology remains a divisive mystery.
done in Shi’s lab. “My work … was mainly
genome characterization and evolutionary
analysis of viruses,” Hu wrote. Yu also denied
being involved with live virus experiments.
“I like bioinformatics and I mainly engage
in gene sequencing and data analysis in the
laboratory,” she wrote. Both also insisted to
Science they did not get sick in late 2019,
let alone had COVID-19 or symptoms of
it. Yu emailed Science that the charges are
“fake news.” (Zhu did not reply to emails.)
Shi backs up their accounts. She wrote in
an email that Hu, Yu, and Zhu “worked on
genome sequencing based on extract RNA
and never worked on live virus. … All the al-
legations about the lab-associated accident
of COVID-19 are wrong.”
Lab-leak proponents have discounted Shi’s
protests before, saying she and China have a
clear reason to lie if work at WIV led to the
pandemic. They stress that WIV has refused
to allow outside investigators to conduct an
independent review of lab notebooks and the
like, and to make public a bat coronavirus
database it removed from the internet.
ODNI’s new report notes that agencies in
the U.S. intelligence community (IC) disagree
about the pandemic’s origin. The Depart-
ment of Energy and FBI think there was a
lab leak, it says, noting that they have differ-
ent rationales but leaving them unexplained.
Five other intelligence branches lean toward
a spillover from an animal, ODNI says. The
report also dismisses the allegation, circu-
lated by some lab-leak proponents, that the
virus was developed or enhanced in a labo-
ratory, perhaps for military purposes. “Most
agencies assess that SARS-CoV-2 was not
laboratory-adapted,” it says. “All IC agencies
assess that SARS-CoV-2 was not developed as
a biological weapon.”
The brevity of the report—which only has
five pages of original content—enraged Re-
publicans who had spearheaded the declas-
sification law. Senator Mike Braun (R–IN)
tweeted, “The COVID-19 Origin Act calls for a
full declassification, not Cliffs Notes to cover
for Dr. Fauci and protect China.”
Republicans in the House of Representa-
tives who run an intelligence subcommittee
on the pandemic have continued to push hard
on the lab-leak theory, recently interview-
ing Scripps Research evolutionary biologist
Kristian Andersen for more than 8 hours.
PHOTO: THOMAS PETER/REUTERS
He worked on a 2020 paper that dismissed
the idea that a lab engineered the virus. The
House members have also subpoenaed his
Slack messages connected to that work. “We
are following the breadcrumbs of a COVID
cover-up straight to the source!” tweeted the
subcommittee’s Twitter handle. j
science.org SCIENCE

NEUROSCIENCE
Consciousness hunt yields results but not clarity
Two rival theories of how it emerges went head-to-head in brain-scanning experiments
By Elizabeth Finkel, in New York City
A
mid  rock music,  a rap  about con-
sciousness, and the calling in of
a 25-year-old drunken bet, camps
backing two leading theories of how
consciousness arises from the brain
waited anxiously in a Greenwich Vil-
lage theater last week to hear who had won
the first round of an “adversarial collabo-
ration.” On balance, the initial results from
the experiments, which recorded brain ac-
tivity as volunteers watched screens and
pressed buttons, favor the idea that con-
sciousness arises in sensory areas at the
back of the brain. But the opposing
camp is far from ready to concede.
It still contends that consciousness
emerges within the brain’s “execu-
tive” center, the prefrontal cortex.
“The results ended up challeng-
ing both [groups], with key pre-
dictions of the two theories being
disconfirmed by the data,” says
Liad Mudrik, a cognitive neuro-
scientist at Tel Aviv University who
helped design the study and assess
its results.
The 23 June revelation at the
26th annual meeting of the As-
sociation for the Scientific Study
of Consciousness (ASSC) also
settled a wager placed in 1998 A study participant observes onscreen images while researchers record
at the second such conference.
Cognitive neuroscientist Christof
Koch bet philosopher David Chalmers
that the neural correlates of consciousness
would be nailed down in 25 years. Koch
conceded at the meeting that those corre-
lates remain unclear.
Philosophers and neuroscientists seek a
working theory of consciousness as the first
step toward measuring the phenomenon—
whether to make life and death decisions
about brain-damaged patients, ascribe
rights to animals, or determine whether ar-
tificial intelligence may have it. Two major
theories have emerged. Integrated informa-
PHOTO: LING LIU/PEKING UNIVERSITY
tion theory (IIT) posits a structure of sen-
sory neurons in a posterior “hot zone” as
the source of consciousness, whereas the
global neuronal workspace theory (GNWT)
likens networks of neurons in the front of
the brain to a clipboard where sensory sig-
nals, thoughts, and memories combine be-
fore being broadcast across the brain. The
SCIENCE science.org
collaboration, funded by the Templeton
World Charity Foundation (TWCF), set out
to pit the theories against each other.
Historically, GNWT has gained support
from studies in which participants report
when they become aware of a stimulus, such
as an image flashed on a screen. In those
tests—many led by GNWT’s  chief archi-
tect, Stanislas Dehaene of the French bio-
medical research agency INSERM’s Cog-
nitive Neuroimaging Unit—brain scans
showed the prefrontal cortex lit up at the
moment of perception.
But philosophers and experimentalists
questioned whether these studies captured
brain activity with magnetoencephalography.
the neural markers of conscious perception,
or simply processes required for reporting
it, such as memory and attention, which
involve the prefrontal cortex. “No-report”
studies, where participants passively view
images, seemed to offer a workaround.
When different images are shown to a per-
son’s left and right eye, so-called binocular
rivalry causes conscious perception to flip
between the images. The flips can be moni-
tored, independent of participants’ report,
by tracking eye movements. These experi-
ments found signals of conscious percep-
tion at the back of the brain, as IIT predicts.
The front-of-brain camp argued these
studies were themselves rife with confound-
ers. For example, participants might be so
bored by staring at onscreen images that
they let their minds wander to other tasks.
It was this cauldron of contested evi-
dence that fueled the adversarial collabora-
NE WS | I N D E P T H
tion. The project, launched in 2019, was the
brainchild of  Koch, then chief scientist  at
the Allen Institute and a proponent of IIT,
and Dawid Potgieter, director of Discovery
Science programs at TWCF, which com-
mitted $20 million to a series of grants for
studies testing theories of consciousness
(Science, 18 October 2019, p. 293).
Mudrik and the two other indepen-
dent project leaders, psychologists Lucia
Melloni at the Max Planck Institute for
Empirical Aesthetics and Michael Pitts at
Reed College, spent a year working closely
with the leading proponents of the two
theories:  Dehaene  and Giulio Tononi, a
psychiatrist and neuroscientist at
the University of Wisconsin (UW)-
Madison and a chief architect of IIT.
They agreed on the design of two
experiments for which the theories
made clearly distinct predictions.
Dehaene and Tononi had no
further role in the study. Instead,
six  theory-neutral labs scanned
the brains of 250 total partici-
pants using three  techniques:
functional magnetic resonance im-
aging,  magnetoencephalography,
and  electrocorticography, in which
electrodes are placed on the brain’s
surface prior to a surgery.
The first of the two planned ex-
periments  showed participants
images with and without an accom-
panying task—pressing a button in
response to target pictures. IIT predicts that
passive perception will activate the back of
the brain, while adding tasks will also spark
the front. GNWT predicts frontal activation
for both active and passive perception.
Key to the study were algorithms called
multivariate pattern decoders, which aim to
predict certain features of the onscreen im-
age (what category of object was displayed,
or whether a face pointed forward or to
the side) based solely on the viewer’s brain
signals. These decoders, which pick up pat-
terns of activity across the brain, increased
the sensitivity of the experiment compared
with methods of analysis that simply mea-
sure the strength of a signal in a specific
brain region. Researchers initially “trained”
the decoders by feeding them examples of a
participant’s brain activity data along with
the corresponding image.
GNWT posits that frontal networks sup-
30 JUNE 2023 • VOL 380 ISSUE 6652 1309
NEWS | I N D E P T H

porting both active and passive perception Jeremy Farrar, here at May’s World
should be similar enough to allow the de- Health Assembly, has landed a new job.
coder to cross train: If it’s only been trained
on signals from a person passively observ-
ing a face, it should still be able to decode
data from the button-pressing task. IIT pre-
dicts that cross-training will work best with
brain signals from the posterior regions, the
proposed site of conscious perception.
The results, also published in a
26 June preprint, were surprisingly mixed.
When it came to decoding categories of ob-
jects,  the  data  provided strong support
for  GNWT.  But for decoding the  orienta-
tion of faces, IIT was the better fit.
Other analyses revealed interactions be-
tween visual and frontal zones during con-
scious perception, Mudrik says—a finding
that aligns with GNWT.
But the duration of the signatures as-
sociated with conscious perception offered
stronger support for IIT. Activity persisted at
the back of the brain for as long as the image
was presented. In contrast, GNWT predicts
an initial activity spike—the “ignition” of the
frontal workspace—and another when the
stimulus disappears. That theory scored a
partial win: There was evidence for an initial
spike, but not the “off” spike.
Many IIT advocates are ready to declare
victory. “The results corroborate IIT’s overall
claim that posterior cortical areas are suf-
ficient for consciousness, and neither the PUBLIC HEALTH
involvement of [the prefrontal cortex] nor
global broadcasting are necessary,” Melanie
Boly, a neurologist and neuroscientist at UW
Madison and a leading proponent of IIT, said
WHO’s new chief scientist wants
at the ASSC presentation.
Although Dehaene agrees the new results
bolster back-of-the-brain theories, he says
to look to ‘tomorrow’
many theories beyond IIT make similar Jeremy Farrar says more must be done about COVID-19,
predictions. More importantly, he says, the
design of the study compromised the sensi-
but agency should aim to anticipate its next major challenges
tivity of decoding from the front of the brain
that would have supported GNWT. Tononi By Kai Kupferschmidt
was keen on that design, he says.

But the competition will go another round,
and Dehaene will get his preferred design.
The next TWCF-funded experiment, which
the research team hopes to present at next
year’s ASSC meeting, will compare brain sig-
nals when participants are aware of seeing
images and when they fail to consciously see
them because of a distraction.
Although Koch’s favored theory now has
a leg up on GNWT, he had to concede that
science hasn’t yet found the neural corre-
lates of consciousness. He paid off the bet
A
t the beginning of May, after almost 10 years at the helm of one of the world’s richest
biomedical foundations, British physician Jeremy Farrar traded funding clout for a
bigger international stage, moving to Geneva to become chief scientist at the World
Health Organization (WHO). Farrar had helped make the U.K.-based Wellcome Trust
a major player in global issues such as infectious diseases and the health effects of
climate change. He also wasn’t shy about criticizing WHO’s leadership, specifically its
slow response to the West African Ebola outbreak in 2014.
Only the second person in the chief scientist post, he succeeds Indian pediatrician Soumya
Swaminathan, who was promoted from another WHO position in March 2019 and left the
new post late last year. “I think it’s fair to say that the role of chief scientist is still to be fully
framed,” Farrar says. But in his eyes it offers an opportunity to provide countries with sci-
ence to help their political decision-making. “I wanted to do something closer to countries,
PHOTO: ANTOINE TARDY/WHO

to Chalmers with a case of fine wine. “I’ve
lost the battle,” he declared onstage, “but
won the war for science.” j
Elizabeth Finkel is a journalist based in
Melbourne, Australia.
because ultimately, that’s where health and science is going to be delivered.”
COVID-19 will remain a focus, Farrar said in an interview last week, because the world
needs to prepare for the unlikely but possible scenario “that we could go back in a bad direc-
tion.” But he also wants to look forward. “Part of what I would like to achieve is bringing a
sense of tomorrow into the World Health Organization.” Science’s exchange with Farrar has
been edited for brevity and clarity; a fuller version is at https://scim.ag/WHOFarrar.

1310 30 JUNE 2023 • VOL 380 ISSUE 6652 science.org SCIENCE

Q: Why did you decide to take on this job?
A: The world is facing some really big
challenges and I think almost all of
them are transnational. Climate change,
pandemics, antimicrobial resistance,
demographic shifts, inequality: These ulti-
mately require the world to find a way of
working together even if they can’t agree
on anything. And if you believe that, then
you can’t just sit on the sidelines and say:
“Multilateralism is great.” … If you can
make a contribution to [it], your career
will have had meaning.
Q: How do you hope to make a contribution?
At Wellcome you were running one of the
wealthiest, biggest science funders in the
world, now you are in a chronically under-
funded international organization.
A: It is very different and maybe that was
actually the attraction. I think science will
be central to [WHO] and it will be central
as a mechanism to feed into how govern-
ments make decisions. Money is critical,
but it’s not enough. If you’re not willing to
use the best evidence to inform the best
policies, then no matter how much money
you’ve got, you’re going to struggle.
Q: What exactly is the job of WHO’s chief
scientist?
A: I think governments, U.N. agencies,
WHO need to anticipate what is coming,
and then debate this in nations, in the
public, in the media: data, artificial intel-
ligence, genomics, and the ability to ma-
nipulate genomes in agriculture, human
and animals, climate change, and so on.
What is coming down the track? What do
we need to prepare for now, to ensure that
inequality and health disparities are not
exaggerated by those new technologies?
Q: You’ve said very often that in February
2020, we really knew everything we needed
to know about SARS-CoV-2. So how do you
make sure knowledge has impact?
A: There’s no doubt that countries around
the world made their own decisions
[about COVID-19] based on their own
drivers. No individual country got every
decision right. I believe that you are bet-
ter in policymaking if you have access to
the available evidence and science and
data. The decisions you then make are
largely political. But try and do that in the
absence of science and you have weaker
options. I think there is a much greater
appreciation in 2023 than there was in
2020 of the importance of having scien-
tific input in decision-making.
Q: That surprises me. Science is under
attack around the world and it seems like
SCIENCE science.org
dismissing science has become easier
because it is cast as part of the elite.
It’s part of the populist playbook to
denigrate scientists.
A: I agree with you that we are in that de-
bate. But there has always been a debate
about these issues, going back to Galileo,
and to Darwin and John Snow. If in Feb-
ruary 2020 you had told me we will have
a vaccine available before the end of the
year, and that within 2 years, probably
somewhere over 6.5 billion people will
have lined up in supermarkets, football
stadiums, etc. to be vaccinated by some-
body they’ve never met before and will
never meet again, and will have repeat-
edly done that once, twice, three times …
I would have been amazed.
Q: How much of a threat is COVID-19 now?
A: The most likely scenario is the one that
the world is sort of now in: that with a
combination of vaccination and natural im-
munity from infection, new variants are to
a large degree constrained. My concern has
always been that there are other scenarios.
… It would be unthinkable in 2026, 2027,
2028 for us to go back to March 2020. I
think that is unlikely, but we should be
prepared. We are not where we need to be.
We don’t have transmission-blocking vac-
cines. Frankly, we don’t have good enough
therapies yet. And the impact of what is
called Long Covid is going to be huge.
When you’ve had 6.5, 7 billion people
in the world infected, you only have to
increase the risk of diabetes, dementia,
cardiovascular disease by a tiny percent-
age, and you’ve got a huge long-term
problem. We’ve got to understand Long
Covid and find ways of preventing it, but
also treatments.
Q: What about your own future? What
would make you feel you made a difference
at the end of this job?
A: My contract covers 2 years for now
and I haven’t thought beyond that. I
think an individual rarely changes things
dramatically. [But] if you can be part of
a leadership team, help the organization
through [WHO Director-General] Tedros
[Adhanom Ghebreyesus] to move forward
in this peri-pandemic era, to learn the true
lessons of the last 3 years … I would look
back in 2 years’ time and say that the team
here pushed it forward a little bit. We’re all
tired at the moment. We’re all fed up with
COVID. We’ve all been affected personally,
professionally, but we need to lift our eyes
because I think the future is bright. But it
won’t just happen by chance, we have to
push our future. And if I can contribute to
that, that’s great. j
FUNDING
U.S. military
contract could
boost status of
Black university
Defense research institute
aims to turn Howard into
a top research university
By Jeffrey Mervis
T
he U.S. Air Force wants scientists
at Howard University to figure out
how its fighter pilots can team up
with robotic co-pilots to defeat
enemy combatants.
But aerial superiority isn’t the
only goal of Howard’s new Research Insti-
tute for Tactical Autonomy (RITA). Univer-
sity officials are also hoping the Air Force’s
initial 5-year, $90 million investment in
RITA—under a special arrangement that
makes additional funding all but certain—
will help Howard became the first histori-
cally Black institution to ascend to the top
ranks of the nation’s research universities.
“This should be a real game changer,”
predicts Danda Rawat, an electrical en-
gineer and associate dean for research at
Howard who will be RITA’s founding direc-
tor. “We’ve been doing a lot with a small
amount of resources,” says Rawat, who
now directs a $7.5 million Department of
Defense (DOD) center of excellence on ar-
tificial intelligence and machine learning.
But the new university-affiliated research
center (UARC) announced in January “is a
huge step.”
The Air Force’s expected long-term in-
vestment in Howard, one of the nation’s
104 historically Black colleges and univer-
sities (HBCUs), addresses a congressional
mandate that DOD begin to distribute its
massive research budget more equitably.
The UARC—the 15th such center funded
by DOD and the first for the Air Force—is a
potent mechanism for doing so.
Special rules that allow a UARC to re-
ceive a steady stream of DOD contracts
without external competition have fueled
the growth of such academic powerhouses
as Johns Hopkins University. In 1942, a
project at Hopkins to develop a more le-
thal artillery shell morphed into DOD’s first
30 JUNE 2023 • VOL 380 ISSUE 6652 1311
NEWS | I N D E P T H
UARC, the Applied Physics Laboratory. APL
is now a $2-billion-a-year juggernaut with
more than 8200 employees.
No HBCU has a research budget even
1/10th that size. “I’m told this is largest re-
search award ever received by an HBCU,”
says Victoria Coleman, chief scientist for
the Air Force. “The underresourcing of
HBCUs didn’t get created overnight, and
the problem isn’t going to get solved over-
night. And in my book, [this center] is still
not enough. But a vehicle like UARC, some-
Howard University President Wayne Frederick (left) and Air Force Secretary Frank Kendall celebrate a new center.
thing that’s going to be there for the long
haul, can be an important stepping stone.”
In announcing the Howard contract, Sec-
retary of Defense Lloyd Austin noted that
HBCUs train 30% of all Black scientists and
engineers yet receive only 0.05% of DOD’s
massive R&D budget, which is approaching
$100 billion. In 2020, Congress directed the
agency to do more to build the research ca-
pacity of HBCUs, and the next year it gave
DOD permission to use the UARC funding
mechanism to achieve those goals.
Last year, legislators were even more
explicit, telling DOD to target funding
to HBCUs striving to join the roughly
150 universities labeled R1 under a volun-
tary classification system for academic insti-
tutions based on their research activity. In
response, DOD took the unprecedented step
of restricting competition for the UARC to
1312 30 JUNE 2023 • VOL 380 ISSUE 6652
the 11 HBCUs that had announced the goal
of becoming R1s. Four submitted proposals,
and in January the Air Force chose Howard,
arguably the best known HBCU, as the lead
institute in a consortium of eight HBCUs
that will operate RITA.
The institute’s mission is to develop tech-
nologies that enable seamless and instanta-
neous joint reasoning and decision-making
between human and machine. That is a
much bigger challenge than the autonomy
needed for self-driving vehicles, Rawat says.
Those who follow defense research give
the Air Force’s decision a thumbs-up. “This
award makes perfect sense given the con-
gressional directive,” says Andrew Metrick,
a defense fellow at the Center for a New
American Security. “Tactical autonomy is a
critical need for the Air Force,” he notes,
and the cachet of a UARC will make it eas-
ier for Howard to win contracts from other
branches of the military for related work.
Coleman says some senior Air Force offi-
cials had reservations about forming a ma-
jor, long-term relationship with an HBCU
because it lacked the pedigree of the top-
tier research universities that operate most
UARCs. But Coleman says her boss, Secre-
tary of the Air Force Frank Kendall, backed
her up when she presented the idea of an
HBCU-only competition for a new UARC.
“‘What took you so long?’” Coleman recalls
his reaction. “That was a really good day.”
In holding the competition for its new
UARC, the Air Force decided it would only
accept proposals from consortia of institu-
tions, seeing that structure as a good way
to both build capacity across HBCUs and to
draw on their diverse populations. DOD first
used that approach 15 years ago in funding a
UARC on systems engineering, and its direc-
tor says there is strength in numbers.
“When you’re a large university, your
natural inclination is to think you can do it
all yourself,” says Dinesh Verma, a systems
engineer at the Stevens Institute of Techno-
logy who has led the Systems Engineering
Research Center since its formation. “There
were quite a few naysayers at the begin-
ning because all the previous UARCs were
located at large, well-established universi-
ties. But a smaller university is more open
to the idea of a network, which has worked
very well.”
Bindu Nair, head of DOD’s basic research
office, hopes the RITA will not only help
Howard attain R1 status, but also expand
the nation’s pool of STEM talent. “Being
an R1 is what higher ed uses to measure
research capacity,” says Nair, a materials
scientist who came up through the Army’s
research system. But a more important
goal, she says, “is creating more students
capable of doing high-quality research. And
if institutions are committed to doing that,
then we will support them in getting to
where they want to be.”
The heavy teaching load carried by HBCU
faculty is a major impediment to becom-
ing a research powerhouse, Nair says. “We
hear it from them all the time: ‘How am I
supposed to focus on my research if I have
these requirements in terms of teaching?’
But they don’t want to give up the teaching,
because that’s where a lot of the mentoring
takes place.”
The promise of continued funding that
comes with a UARC is designed to remove
that obstacle, Coleman says. “If you have
money to do research, you also have money
through overhead that allows you to hire
additional people,” she says. “And the UARC PHOTO: MATT MCCLAIN/THE WASHINGTON POST VIA GETTY IMAGES
will give them the confidence to add faculty,
equip their labs, and build up their pro-
grams by training more students.”
Now in temporary quarters on Howard’s
main campus in northwestern Washington,
D.C., RITA expects to move early next year
into a 10-story building being renovated at
an off-campus site a few kilometers away.
Bruce Jones, Howard’s vice president for re-
search, can’t wait to see it in action.
“When DOD commits to a UARC, it’s
not just to fund more research,” Jones
says. “It’s also a promise to mentor the
next generation.” j
science.org SCIENCE

FEATURES
A kidney recently removed from a rat normally decays quickly. But if preserved in extreme cold and artfully warmed, it can have a second life.
T
he rat kidney on the operating
table in front of Joseph Sushil Rao
PHOTO: EVAN TAYLOR STUDIOS
looked like it had been through hell.
Which it had—a very cold one.
ASSISTED BY TOM OKINS
Normally a deep pink, this
thumbnail-size organ was blanched
a corpselike gray. In the past
6 hours, it had been plucked from
the abdomen of a white lab rat,
SCIENCE science.org
NE WS
Scientists are learning how to
cryopreserve living tissues,
organs, and even whole organisms,
then bring them back to life
By Warren Cornwall small plastic box and gently laid it inside the
open abdomen of another white rat. Peer-
pumped full of a black fluid, stuck in a ing through a microscope, the transplant
freezer cooled to –150°C, and zapped by a surgeon–in–training deftly spliced the kid-
powerful magnet. ney’s artery and vein into the rat’s abdominal
Now, in a cramped, windowless room on blood vessels using a thread half the thick-
the 11th floor of the University of Minne- ness of a human hair.
sota’s (UMN’s) Malcolm Moos Health Sci- When he finally removed the tiny clips
ences Tower, Rao lifted the kidney from a pinching off the blood supply from the aorta,
30 JUNE 2023 • VOL 380 ISSUE 6652 1313
 Joseph Sushil Rao, a transplant surgeon–in–training at the University of Minnesota, prepares to remove a kidney from a laboratory rat.
the kidney blushed pink, a good first sign.
Then he waited. Forty-five minutes later, a
golden drop of urine emerged from the ure-
ter that would normally feed from the kidney
to the bladder.
Just before midnight, Rao snapped a
close-up photo with his iPhone, proof that
the kidney was working. He sent the photo
and an ecstatic email to his two bosses,
transplant surgeon Erik Finger and bio-
medical engineer John Bischof, titled
“First successful transplant of vitrified,
nanowarmed rat kidney.”
“I’m out of words,” he wrote. “This is a
proud moment for us all. It was not easy. But,
it paid off.”
That moment in April 2022 was one in a
series of recent breakthroughs in the quest
to effectively stop biological time. After de-
cades of frustration and halting progress,
scientists in the past 10 years have made
major advances using extreme cold to slow
or even halt the decay that is the usual fate
of all living things. They’ve developed new
ways to reduce the toxicity of chemical anti-
freeze treatments, minimize the formation of
destructive ice, and thaw objects rapidly and
evenly. Since 2018, labs have frozen and then
revived bits of coral, fruit fly larvae, zebrafish
embryos, and rat kidneys. They have also ap-
plied gentler techniques to cool everything
from tomatoes to entire pig livers to just be-
low freezing without ice formation, keeping
them virtually fresh for days or weeks.
Medical uses, particularly organ trans-
plants, are a key driver for today’s work.
Scientists hope to eventually create cryo-
preserved banks of tissues such as skin, en-
tire organs, or even limbs, easing shortages
1314 30 JUNE 2023 • VOL 380 ISSUE 6652
and giving doctors time to better prepare
recipients for transplants. But the advances
in preservation also extend to specks of hu-
man tissue used to screen pharmaceuticals,
species on the brink of extinction, fruit flies
studied by geneticists, produce bound for
grocery stores, and fish embryos stored for
aquaculture. Mehmet Toner, a bioengineer
at Massachusetts General Hospital (MGH)
and one of the leaders in the field of cryo-
preservation, likens the vision of stored liv-
ing tissue available on demand to a more
familiar cornucopia. “I call it,” he says, “the
Amazon of living things.”
SOMEONE RAISED on Hollywood movies
might think the technology to freeze and
revive entire organisms is right around the
corner. Star Wars’s Han Solo is trapped in
“carbonite” and resuscitated. Tom Cruise
gets turned into a human popsicle in a dys-
topian prison in Minority Report. Captain
America is entombed in Arctic ice in a Mar-
vel movie and rewarmed nearly 70 years later
for a sequel.
Reality is far less simple. The largest living
thing routinely stored at temperatures well
below zero and brought back to life is the size
of a grain of table salt: a human embryo. Try
that with an entire person using today’s tech-
nology and the result would be a lifeless body
filled with toxic chemicals, says cryobiologist
Greg Fahy. “You would be in sorry shape.”
Fahy was one of a pair of scientists whose
1985 Nature paper revealed a chemical pro-
cess that allowed mouse embryos to be stored
at nearly –200°C. Their technique addressed
the major barrier to freezing living tissue: ice.
When water freezes, it can wreak havoc
inside tissue. The water molecules go from a
tightly packed, amorphous fluid to a rigid lat-
tice. Ice crystals tear through cells like knives.
Salts in cell fluids get concentrated at toxic
levels in the tissue parts that freeze last. Any-
one who has frozen and thawed a strawberry
has seen the result: a mushy, discolored ver-
sion of what came off the plant.
Getting tissue below the freezing point
while minimizing ice is crucial. (That’s why
cryobiologists don’t like to say they “freeze”
tissue.) For the mouse embryo, Fahy and his
colleague at the American Red Cross, William
Rall, first soaked the little ball of cells in a
chemical cocktail that leached out much of
the water, replacing it with chemicals simi-
lar to the antifreeze in a car’s radiator. These
cryoprotectants, as they are known, dilute
the water molecules and form a viscous fluid
that discourages ice crystal formation.
Then they cooled the embryo, kept in a
slender plastic straw, using –196°C liquid
nitrogen. Between the rapid cooling and the
cryoprotectant, ice didn’t have time to form.
Rather than line up in a tidy crystalline pat-
tern, water molecules were stuck in a random
mass like a rigid liquid, a process known as
PHOTO: EVAN TAYLOR STUDIOS ASSISTED BY TOM OKINS
vitrification. The result was a hard, smooth,
glasslike substance without the problematic
properties of ice. To rewarm the embryo, Rall
stirred the straw in 0°C water.
The mouse embryo work paved the way for
banking similar-size human embryos, trans-
forming fertility treatment. But what works
for a tiny embryo of about 100 cells doesn’t
size up easily to whole organs. It’s hard to get
cryoprotectant to soak evenly into a bigger
piece of tissue. The center can take longer to
solidify, which fosters ice formation. Pump-
science.org SCIENCE
 NE WS | F E AT U R E S

Stopping the biological clock

Advances in using extreme cold to slow biological processes could touch everything from donated organs to fresh produce. A University of Minnesota team has
developed one approach, dubbed “nanowarming,” which thaws an organ evenly to avoid damage from ice.

Cryoprotectant and Nanoparticles Flushing fluid infused

iron nanoparticles radiate heat in renal artery
are infused

Vena cava
Aorta
through
renal
artery

Outflow
through Ureter Nitrogen Magnetic Cryoprotectant and nanoparticles
rtticl
iclees Urinne
Urine
renal vein freezer coil flow out through renal vein
1 Cryoprotectant 2 Big chill 3 Magnetic rewarming 4 Fluid flush 5 Ba
acck in
Back n action
aactio
on
A cocktail of chemicals that The treated kidney is cooled A current flowing through a Iron particles and toxic In five rrats
ats
t that rreceived
rece
re ceeive
eived
i d
resist ice formation is to –150°C. The extreme cold coil creates a magnetic field cryoprotectants are vitrified kidneys, the organs
infused into a rat kidney. causes the liquid in the kidney that flips orientation 360,000 flushed from the kidney were functioning 30 days later,
The liquid also contains tiny to vitrify, entering a glasslike times per second, warming before it is transplanted raising hopes that the same
iron oxide particles that state that is less damaging to the iron particles and thawing into a rat, where it resumes approach might someday work
spread through the organ. living cells than crystalline ice. the organ evenly. functioning. for donated organs for humans.

ing in more cryoprotectant to counter ice can
be damaging because the chemicals are toxic.
Rewarming poses its own problems. If an
object warms too slowly, ice crystals can ma-
terialize as the tissue approaches the melting
point. If it doesn’t warm uniformly, stresses
caused by uneven expansion or contraction
can crack the object like an ice cube dropped
in a glass of water.
In 2002, Fahy stepped up his work in
mouse embryos to rabbit kidneys. He got as
far as implanting a previously vitrified organ
into an animal. The rabbit survived nearly
7 weeks. But a necropsy revealed that al-
though the kidney was functional enough
to keep the animal alive, much of it
was damaged.
Fahy has been chipping away at the prob-
CREDITS: (GRAPHIC) C. BICKEL/SCIENCE; (PHOTO) EVAN TAYLOR STUDIOS ASSISTED BY TOM OKINS
irretrievably damaged. For a liver, the win-
dow is 8 to 12 hours. For a kidney it’s about
1 day.
The rush creates burdens for the medi-
cal system and for patients. Surgeons are
called to the hospital in the middle of the
night. Transplant recipients have a foreign
organ plugged into their body without
time for treatments that would help their
immune system acclimate. More than 60%
of donated hearts and lungs never make it
to a recipient in time. Fewer than 10% of
people who need organ transplants actu-
ally get them, the World Health Organiza-
tion estimates.
Cryopreservation holds out the possibil-
ity that organs could be stored for days,
weeks, or even years before they are im-
planted. That could save organs from get-
ting tossed after a few hours and would
enable doctors to find organs more easily
when needed or choose those that are a
closer immunologic match to recipients.
“It could touch so many aspects of bio-
medicine, truly change the way that we can
treat health,” says Sebastian Giwa, an econ-
omist and former hedge fund manager who
founded the nonprofit Organ Preservation
Alliance in 2012.
Giwa has helped launch several cryo-
preservation-related companies. One, Gaia-
Life, is experimenting with vitrifying ova-
ries. The goal is to remove the egg-bearing
organs from people before they undergo
ovary-damaging medical treatment such as
chemotherapy, then reimplant them after

lem ever since, testing different
chemical mixtures and cooling
and warming protocols. “It
turned out to be harder than I as-
sumed,” says Fahy, who is now ex-
ecutive director of 21st Century
Medicine, a private cryopreserva-
tion research company. “I think
all of this will pay off, but we’re
not quite there yet.”
THERE’S GOOD REASON to per-
sist. The rapid decay of organs
is one of the biggest problems
bedeviling organ transplants
for people. From the mo-
ment a human heart or lung
is disconnected from a donor,
doctors have 4 to 6 hours to
get it hooked up to a new pa-
tient’s blood supply before it is
the treatment is over. So far re-
searchers working with the com-
pany have reimplanted vitrified
ovaries into five sheep; in four
of the animals the ovaries pro-
duced progesterone, a sign they
were working, says Alison Ting, a
reproductive biologist and Gaia-
Life’s chief scientific officer. Ting
declined to describe the details
of the company’s methods but
says the progress “gives me the
optimism to say that the first
in human could be sooner than
5 years.”
BY VITRIFYING animal organs,
Fahy demonstrated a key first
step, Bischof says. “The problem
Postdoctoral researcher Zonghu Han connects a rat kidney to tubes that pump was he couldn’t rewarm them.”
antifreeze chemicals and tiny iron oxide particles into the organ before it is chilled. Finding ways to warm vitri-

SCIENCE science.org 30 JUNE 2023 • VOL 380 ISSUE 6652 1315

 In a rat kidney cooled to –150°C, water has been vitrified, turning smooth and glasslike rather than forming sharp ice crystals.
fied tissue quickly and evenly has been the
focus of Bischof ’s lab. In the past few years,
his team has tried everything from lasers
to heat-conducting mesh. With larger ob-
jects, such as rat kidneys, they have made
progress with a powerful magnetic field
coupled with iron nanoparticles.
On an unseasonably hot day in April,
Zonghu Han, a UMN mechanical engineer-
ing postdoctoral researcher, connected a
slender plastic tube to a rat kidney rest-
ing on a bed of gauze. He made a few key-
strokes on a computer and a black fluid
began to flow into the organ. The color
came from the iron nanoparticles sus-
pended in cryoprotectant. When the organ
turned a glossy ebony from the infusion,
Han slipped it into a small plastic bag,
and lowered it into a nearby
freezer cooled to –148°C.
The clock for the kidney’s
survival had been ticking
for more than 3 hours, since
Rao, the transplant surgeon,
had removed it from a rat in
a reenactment of that 2022
breakthrough surgery. Now,
as the kidney’s temperature
plummeted inside the freezer,
the biological processes grad-
ually destroying the organ
ground to a halt. “We have
stored [a rat kidney] up to
100 days before transplanta-
tion,” Han says. “It’s safe in
there indefinitely.”
In this case the kidney got
just 45 minutes. Han opened Zonghu Han removes a plastic bag containing a vitrified kidney from a device that
the lid in a billow of vapor and used a fast-changing magnetic field to heat the iron particles inside the organ.
1316 30 JUNE 2023 • VOL 380 ISSUE 6652
lifted out a tiny, rigid packet containing
the vitrified organ. He placed the packet
inside a small metal cup attached to a
cream-colored metal box. When he pressed
a button, the box generated a magnetic field
around the cup that flipped the positive and
negative poles 360,000 times every second.
That fluctuation heated the iron particles
and thawed the kidney in 90 seconds.
“That’s our secret sauce,” Bischof says of
the process as he watches.
In a Nature Communications paper in
early June, Bischof ’s team reported put-
ting five rat kidneys through this treat-
ment and reimplanting them. All of the
recipient animals lived a month before
they were killed to study their condition.
Now, the researchers have graduated to
pig kidneys, closer to the size of a human
kidney. Bischof declined to discuss details
of the unpublished pig work. “There’s no
physical reason that we’re aware of why
this [warming procedure] won’t work” in
larger organs, he says.
Although nanowarming, as Bischof calls
it, is his tool of choice for the kidneys, it re-
quires costly machinery and one-at-a-time
treatment. In May, Smithsonian Institution
marine biologist Mary Hagedorn was at a
laboratory outside Tampa, Florida, test-
ing a simpler approach developed by the
Bischof lab: a fine metal mesh engineered
to quickly transmit temperature—both
cold and heat. She is trying it on batches of
coral larvae. If it works and can be scaled
up, Hagedorn thinks this process could be a
critical piece of her campaign
to bank dozens of coral species
in the coming decade, before
increasingly hot and polluted
oceans spell their end.
The mesh has already proved
successful on fruit fly larvae in
Minnesota, and with two spe-
cies of mushroom coral in Ha-
PHOTO: EVAN TAYLOR STUDIOS ASSISTED BY TOM OKINS
waii and Australia. In Florida,
Hagedorn and colleagues were
trying it on Diploria labyrin-
thiformis, a kind of brain coral
whose larvae are more than
100 times bigger than those
of mushroom coral. In the
first few attempts, rewarmed
larvae were falling apart.
Each larval size, Hagedorn
was learning, needs its own
version of the treatment.
science.org SCIENCE

“We’re struggling a little bit to get this to
work,” she says.
WHILE SCIENTISTS such as Bischof and
Hagedorn wrestle with vitrification, oth-
ers are seeking an easier route by avoiding
ultralow temperatures that require large
infusions of cryoprotectant and make re-
warming so challenging.
At Harvard University and MGH, scien-
tists are taking cues from nature to push
tissues below freezing while holding back
the ice. The wood frog, Rana sylvatica, is
a champion of this realm. Found in much
of North America, including the frigid Ca-
nadian Arctic, it can spring to life after
spending months with as much as two-
thirds of its body frozen at temperatures
as low as –16°C.
As winter arrives, a cascade of physio-
logic changes prepares the frog to sur-
vive freezing, says Shannon Tessier,
an MGH biomedical re-
searcher who has stud-
ied it and other animals
that hibernate at near
freezing. The frog’s liver
churns out glucose that
acts as antifreeze inside
tissues. Antioxidant levels
there increase, protecting
against damage caused
by sudden changes in
the amount of oxygen in
cells. Special proteins in
the frog’s bloodstream
act as seeds for ice crys-
tals, steering their growth
to begin in the more du-
rable vasculature and not The wood frog (Rana sylvatica) can survive partially frozen for months. Strategies used by
in more delicate tissues. the frog to endure such extremes are inspiring efforts to better preserve human tissue and organs.
Such changes “are the the-
matic things that we want to pull forward
to human organs and tissues,” Tessier says.
In Boston, a team of scientists including
Tessier mimicked the frogs by flooding hu-
man livers with a synthetic sugar that, un-
like natural glucose, can’t be metabolized
into toxic byproducts. In 2019, they an-
nounced the approach had enabled them
to store human livers at –4°C for 27 hours,
more than double the standard life span of
donated livers.
More recently, the team combined this
synthetic sugar with an infusion of Sno-
PHOTO: J.M. STOREY/CARLETON UNIVERSITY
max, an ingredient normally used as a seed
for snow in snowmaking machines at ski
resorts. Like the proteins in the wood frog’s
blood, Snomax slowed ice formation and
concentrated it in the blood vessels of rat
livers, enabling them to be stored partially
frozen for up to 5 days at –15°C, then thawed
with limited damage. Five days is far short
of the virtually limitless storage time for a
SCIENCE science.org
vitrified organ, but still useful, says Toner,
who is working with Tessier. The extra time
could, for example, allow a patient wait-
ing for an organ to receive a bone marrow
transplant, a possible measure to coax the
immune system to accept the new organ.
On the other side of the country, Boris
Rubinsky had the idea that higher pres-
sures might help him supercool organs
without damage. In the early 2000s,
Rubinsky, a biomedical engineer at the
University of California, Berkeley, began
to cool objects inside sealed metal contain-
ers. As water inside approached freezing it
expanded, raising the pressure. The higher
pressure, he discovered, limited the forma-
tion of ice. “In all modesty, it’s a revolution-
ary approach to preservation,” he says.
In April, his team cooled a pig heart to
–4°C for 21 hours in one of its pressure
chambers, then warmed and implanted it
into another pig where it began to beat on
its own. In addition to avoiding the chal-
lenges that can come with full vitrification,
Rubinsky finds his strategy enables him to
use less cryoprotectant than other methods
require, reducing toxic side effects.
Because it is simple and relatively low-
tech, the approach could also be put to work
to preserve food without the ice damage
that can degrade today’s frozen vegetables
and meats. The metal storage chambers
fit in standard commercial freezers. And
because the method works at warmer tem-
peratures than those often used for storing
frozen food, Rubinsky and colleagues proj-
ect widespread use could cut global en-
ergy consumption by more than 6 billion
kilowatt-hours per year—equivalent to Lat-
via’s total annual electricity demand.
Cristina Bilbao-Sainz, a food technologist
at a U.S. Department of Agriculture (USDA)
lab in Albany, California, started to work on
the approach after Rubinsky gave a presen-
NE WS | F E AT U R E S
tation to the center in 2016. “It was a simple
idea, but so novel,” Bilbao-Sainz says. “We
totally saw that this could have a future.”
The first attempt was discouraging. Rasp-
berries, a delicate fruit, turned to mush
when thawed. Then Bilbao-Sainz tried
tomatoes. Twenty grape tomatoes sat for
1 month in a sealed container filled with
sugar water at –2.5°C. They emerged look-
ing like they had just been picked, she says.
Bilbao-Sainz has also had good results
with spinach, cherries, and potatoes. Now,
she is trying blueberries. In December
2022, the USDA lab started to work with
an unnamed company interested in using
the technology. “It could have a big im-
pact,” she says.
THE ADVANCES are coming swiftly enough
to make those Hollywood scripts seem less
outlandish. Some scientists raise the pos-
sibility of stockpiling vitrified human or-
gans grown in genetically
engineered pigs for fu-
ture transplants. Tessier
is working to keep whole
animals—tiny zebrafish
larvae—in suspended
animation. So far, she has
partially frozen them at
–10°C for 3 days. When
thawed, half of the fish
survived and kept grow-
ing. Substitute people for
fish and improve the suc-
cess rate, and a science-
fiction staple could even-
tually become a reality.
“You can think about even
long-term space travel,”
Tessier says. Hello, 2001:
A Space Odyssey.
But that’s a far cry from today. As if to il-
lustrate the gap between vision and reality,
the UMN demonstration goes awry at one
of the last steps. As the black nanoparticles
are flushed from the rat kidney, water con-
denses on the chilled equipment, causing a
device to malfunction. It pumps a solution
commonly used to preserve organs dur-
ing transplants into the kidney at out-of-
control pressures.
Han and Bischof stare dejectedly at
the damaged organ. Before any of this
technology makes its way into operating
rooms, it will not only have to scale up
to human size, but will also need to pass
muster with safety regulators such as the
U.S. Food and Drug Administration. “All
of this engineering that we’re doing in the
lab has to be made failsafe,” Bischof says as
he points toward the kidney. “This is just
an example of some of the things that can
go wrong.” j
30 JUNE 2023 • VOL 380 ISSUE 6652 1317
 INSIGHTS
PERSPECTIVES
ASTRONOMY
Galactic neutrinos in the Milky Way
A source of neutrinos may lie within the midplane of the Galaxy
By Luigi Antonio Fusco
V
isible light has long been the probe
for investigating the Universe. Over
the past decades, however, it has be-
come evident that there is much to
gain by using other wavelengths of
the light spectrum as well as using
cosmic messengers such as gravitational
waves (1) and subatomic particles called
neutrinos (2, 3). The plane of the Milky
Way—the galaxy where Earth resides—is the
most clear feature of the sky at all wave-
lengths of light. On page 1338 of this issue,
the IceCube Collaboration (4) reports sta-
tistically significant evidence for neutrinos
coming from the inner parts of the Milky
Way in the data collected by the IceCube
Neutrino Observatory (5). Existence of
this long-sought signal paves the way
for the future of astroparticle physics in
our Galaxy.
Dipartimento di Fisica “E.R. Caianiello”, Università degli Studi
di Salerno, 84084 Fisciano, Italy. Email: lfusco@unisa.it
1318 30 JUNE 2023 • VOL 380 ISSUE 6652
The Milky Way is filled with matter (gas
and dust) in the interstellar medium (ISM).
Its density is far from being homogeneous,
with much denser regions located along
the Galactic plane. This region, lying along
the Galactic equator, is the midplane of the
Milky Way’s spinning disk of stars. Cosmic
rays—protons and nuclei traveling through
space at speeds approaching that of light—
also fill the Milky Way. Accelerated in the
aftermath of cosmic explosions, they are
trapped by magnetic fields for thousands to
millions of years within the Galaxy. As they
travel around, they may occasionally col-
lide with an atomic constituent of the ISM,
breaking it into pieces. The released energy
becomes mass in the form of new particles.
Most of these particles are short lived and
can decay into gamma rays and neutrinos.
Although neutrinos can only be produced as
the consequence of these interactions, com-
peting reactions exist for the production of
gamma rays—namely, energy-loss processes
of cosmic electrons. Thus, detection of neu-
trinos informs about their parent cosmic
rays (highly energetic particles), whereas
gamma rays cannot provide such an unam-
biguous identification.
Such collisions in the ISM are more
probable in the denser inner Galactic
plane, so an intense neutrino flux is ex-
pected to originate from there. Neutrinos
have the special feature of interacting
with matter very weakly and can traverse
the cosmos unscathed. However, to detect
cosmic neutrinos, cubic-kilometer-scale
detectors must be built. Once a neutrino
interacts with the medium in which the
detector is situated (the Antarctic ice, for
example), it will produce relativistic, sub-
atomic, charged particles. If the medium is
transparent to light, the Cherenkov radia-
tion emitted in the process can be detected.
PHOTO: BRUNO GILLI/ESO/CC BY
The IceCube Collaboration has managed to
build such a cubic-kilometer detector that
has been operating in the Antarctic ice
shell for 15 years.
Even if a few cosmic neutrinos interact in
a cubic-kilometer detector, they likely will be
hidden in the plethora of signals produced
science.org SCIENCE
by the passage of particles in the atmosphere
produced by cosmic rays reaching Earth.
This foreground can be very hard to elimi-
nate. The best strategy is to look for neutri-
nos that have traversed Earth. Only neutri-
nos can cross the entirety of Earth, and this
path shields the detector from most particles
produced in the atmosphere. However, this
makes IceCube (located at the geographic
South Pole) blind to the southern sky, where
the Milky Way neutrino signal lies. Analysis
strategies based on artificial intelligence are
commonly used to overcome these issues
and have already helped provide evidence
for cosmic neutrinos (6, 7).
Cosmic rays have been measured for
more than a hundred years now, but much
remains unknown about their composi-
tion, sources, and accelerators and how
they behave during the time they travel
through the cosmos. The cosmic rays fur-
thest from Earth have been detected by us-
ing the Voyager probes, in the interstellar
space (8). Cosmic rays may be very far from
their sources when detected, or detection
may be influenced by other nearby sources
of cosmic rays. Because the Milky Way is
not homogeneous, the cosmic-ray popula-
tion detected on Earth cannot be extrapo-
lated to reflect those of the whole Galaxy.
Cosmic rays can only be directly observed
up to certain energies because of the cur-
rent limitations in the size of detectors
SCIENCE science.org
that are sent into space. Indirect measure-
ments carried out on Earth can only offer a
limited amount of information (9).
Detection of neutrinos (and gamma
rays) overcomes these issues by providing
information on what cosmic rays look like
when they first interact with the ISM (ei-
ther close to their sources, or as they travel
through the Galaxy). This window of ob-
servation has finally been opened by the
IceCube Collaboration.
A hint of a neutrino signal had been re-
cently reported by the undersea neutrino
telescope Antares in the Mediterranean
Sea (10). Although smaller in size, Antares
could perform well in searches for neutri-
nos from the Galaxy because of its location
in the Northern Hemisphere. Its position
allowed searching for neutrinos with less
contamination from events originating
from Earth’s atmosphere.
The IceCube discovery and the Antares
finding indicate that there is still much in-
formation to extract from neutrinos. Indeed,
a new generation of neutrino telescopes
is under construction, including KM3NeT
(Cubic Kilometre Neutrino Telescope) in the
Mediterranean Sea; GVD (Gigaton Volume
Detector) in Lake Baikal, Russia; as well
as proposals for IceCube Gen-2, P-One (off-
shore Canada), and TRIDENT (The Tropical
Deep-sea Neutrino Telescope; off-shore
China). These next-gen observatories will
The coordinated observation of disparate
“messenger” signals from the cosmos may include
neutrinos from the Milky Way.
be accompanied by upcoming gamma-ray
observatories (the Large High Altitude Air
Shower Observatory, Cherenkov Telescope
Array, and Southern Wide-field Gamma-ray
Observatory). The multi-messenger combi-
nation of photons and neutrinos will be cru-
cial in providing new insights on the proper-
ties of cosmic rays in the Milky Way.
The overdensity of gas in the ISM may
provide hotspots for neutrino searches and
for investigating the reciprocal influence
between the ISM and cosmic rays. As next-
generation observatories start unveiling the
individual sources of cosmic rays (11), we will
eventually answer the questions of their ori-
gins and what propelled them, and possibly
open some new windows on the Milky Way. j
RE FE RE N CES AN D N OT ES
1. B. P. Abbott et al., Astrophys. J. Lett. 848, L12 (2017).
2. K. Hirata et al., Phys. Rev. Lett. 58, 1490 (1987).
3. M. G. Aartsen et al., Science 361, eaat1378 (2018).
4. IceCube Collaboration, Science 380, 1338 (2023).
5. https://icecube.wisc.edu
6. R. Abbasi et al., Astrophys. J. 928, 50 (2022).
7. A. Albert et al., Astrophys. J. Lett. 853, L7 (2018).
8. E. C. Stone et al., Nat. Astron. 3, 1013 (2019).
9. C. Evoli, Zenodo (2020); https://doi.org/10.5281/
zenodo.4396125.
10. A. Albert et al., Phys. Lett. B 841, 137951 (2023).
11. S. Aiello et al., Astropart. Phys. 111, 100 (2019).
10.1126/science.adi6277
30 JUNE 2023 • VOL 380 ISSUE 6652 1319
I NS I GHTS | P E R S P E C T I V E S
MEDICINE
A new class of antiparasitic drugs
Cyanotriazole compounds “poison” topoisomerase II of pathogenic trypanosomatids
By Ruslan Aphasizhev and Inna Aphasizheva
T
ransmitted by insect vectors, hu-
man African trypanosomiasis (HAT),
Chagas disease, and leishmaniasis
cause millions of infections, which are
often fatal unless treated with toxic
drugs. Although there are promising
new therapeutics for HAT (1), development of
reliable, safe, and affordable drugs for Chagas
disease and leishmaniasis remains challeng-
ing. On page 1349 of this issue, Rao et al. (2)
introduce cyanotriazoles (CTs), a new class
of compounds with potent and broad try-
panocidal activity. These covalent inhibitors
of topoisomerase II, an enzyme involved in
nuclear DNA replication, show high efficacy
in clearing trypanosomes and leishmania
parasites from infected animals, efficient tis-
sue distribution, and low toxicity in the mam-
malian host. Acting on a different target than
those of existing interventions, these drug
candidates bring the possibility of eradica-
tion and effective management of many ne-
glected tropical diseases.
Trypanosomiasis, Chagas disease, and
leishmaniasis are all caused by protozoan
hemoflagellates belonging to the class
Kinetoplastea. They are typified by the “ki-
netoplast,” a condensed body of mitochon-
drial DNA, and include agents of many
tropical diseases. Historically called “sleeping
sickness,” West and East HAT are caused by
Trypanosoma brucei gambiense and T. brucei
rhodesiense subspecies, respectively. During
a tsetse fly blood meal, parasites enter the
circulatory system, where they differentiate
into proliferative form and divide rapidly.
Ultimately, the trypanosomes populate adi-
pose and other tissues (3) and penetrate the
blood-brain barrier, leading to neurological
damage and death.
Leishmania donovani and Leishmania
infantum are introduced into hosts through
sand fly bite, which triggers systemic life-
threatening visceral leishmaniasis. Both par-
asites infect and proliferate in macrophages,
with L. infantum also persisting extracellu-
larly. A combinatorial drug regimen is often
required to clear both intracellular and extra-
cellular forms. Also spread by sandflies, other
Leishmania species give rise to mucocutane-
ous leishmaniasis that inflicts disfigurements
Department of Molecular and Cell Biology, Boston University
Medical Campus, Boston, MA, USA. Email: ruslana@bu.edu
1320 30 JUNE 2023 • VOL 380 ISSUE 6652
or skin lesions. An infection by Trypanosoma
cruzi could be transmitted through the bite of
the triatomine bug, congenitally, or through
contaminated blood. After the initial acute
phase with nonspecific symptoms, the often
undiagnosed Chagas disease may remain as-
ymptomatic for years while progressing to
heart failure or digestive tract damage (4).
Rao et al. used an automated assay to
evaluate inhibitory effects of approximately
2 million compounds on proliferation of a
nonpathogenic T. brucei strain, Lister 427, in
axenic culture. This phenotypic screen iden-
tified a heterocyclic CT molecule as a potent
Shown is a colored scanning electron micrograph of
Leishmania donovani promastigotes. Transmitted to
humans by the bite of phlebotomine sand flies, these
parasites cause visceral leishmaniasis.
growth inhibitor that displayed similar ef-
fects on pathogenic T. brucei gambiense and
T. brucei rhodesiense. Remarkably, the lead
compound also demonstrated effectiveness
in killing T. cruzi and L. donovani parasites
in vitro and low cytotoxicity for human and
mouse cultured cells and in mouse mod-
els of infection. During lead optimization,
chemical modifications improved efficacy,
and pharmacological properties manifested
from a robust clearance of T. brucei and T.
cruzi systemic infections in mice. A single
dose of a variant molecule named CT3 dem-
onstrated fast-sterilizing activity by cur-
ing early and late stages of HAT in mouse
models more effectively than the current
standard-of-care drugs.
High selectivity and comprehensive ac-
tivity of CT compounds against multiple ki-
netoplastid protozoans hinted at a conserved
mechanism of action. Rao et al. gained initial
insights from observations of nuclear DNA
damage and cell cycle arrest after CT3 treat-
ment, whereas mitochondrial DNA remained
unaffected. With the understanding that CTs
damage nuclear but not organellar DNA, the
authors obtained a series of drug-resistant
T. brucei isolates, most of which carried
mutations in the topoisomerase II (Top2)
gene that is essential for parasite viability
(5). Further analyses of these drug-resistant
strains established CTs as a new class of cova-
lent topoisomerase II inhibitors.
An adenosine triphosphate (ATP)–de-
pendent molecular machine, topoisomer-
ase II modulates supercoiling and resolves
topological conflicts that arise during DNA
replication. By transiently breaking double-
stranded DNA (dsDNA) and forming a cova-
lent adduct with a nucleic acid, the enzyme
translocates the double helix through the
gap and repairs breaks. Because DNA breaks
may destabilize the genome, ligation is ki-
netically favored, and the topoisomerase II–
DNA cleavage complexes are short-lived and
readily reversible. It follows that a compound
that can stabilize the topoisomerase II–DNA
complex, a so-called topoisomerase II “poi-
son,” would trigger accumulation of dsDNA
breaks and eventual cell death (6). In vitro
activity assays demonstrated that CTs induce
accumulation of linearized DNA in a reac-
tion catalyzed by the recombinant T. cruzi
topoisomerase II, indicating that CTs indeed
stabilize the topoisomerase II–trypanosomal
DNA cleavage complex but spare the ho-
mologous human topoisomerase II enzyme.
These findings reveal a new class of cova-
lent topoisomerase II inhibitors but raise
the question of selectivity. Topoisomerase
II poisons are commonly used as antibiot-
ics (7) and in cancer chemotherapeutics (8)
and tend to exhibit general inhibitory activ-
ity against this family of essential cellular
PHOTO: DENNIS KUNKEL
enzymes.
To understand the specificity of CTs to ki-
netoplastids, Rao et al. used cryo–electron
microscopy (cryo-EM) to determine a high-
resolution structure of the T. cruzi topoisom-
science.org SCIENCE
 erase II bound to DNA and their lead com-
pound, CT1. The structure revealed interca-
lation of the aromatic trifluoromethylphenyl
moiety shared by all CT compounds between
dsDNA bases and spatial occupancy of the
active site, such as those observed with hu-
man and bacterial topoisomerase II poisons.
Conversely, formation of a covalent bond be-
tween cyanoazole heterocycle and a cysteine
residue distinctly conserved in topoisomerase
II among trypanosomatids provides a struc-
tural basis for the observed selectivity of CT
compounds toward trypanosomal enzymes.
Therapeutic potency of CTs—which ex-
ceeds that of currently available drugs—and
their broad trypanocidal activity, low cyto-
toxicity, and oral bioavailability approach-
ing 100% in a mouse model provide strong
encouragement for future clinical trials. As
a new class of trypanocides, these molecules
have great promise to aid the World Health
Organization’s efforts in eliminating West
African trypanosomiasis by 2030 and to im-
prove treatment of Chagas disease and leish-
maniasis. The potential for using CTs to treat
African animal trypanosomiasis, arguably
the most economically important livestock
disease of the continent (9), also seems ap-
parent. Rightfully called “neglected,” these
diseases are ultimately yielding to collective
efforts of international organizations and
synergistic research prowess.
These pathogens have likely played an
important role in early hominid evolution,
and kinetoplastids have been studied as a
model of early branching eukaryotes. This
work has yielded discoveries of antigenic
variation, trans-splicing, polycistronic tran-
scription, RNA editing and guide RNAs, and
other phenomena of general biological im-
portance (10). Although the threat of para-
sites to human life and economy is reduced
by advances in pharmacological interven-
tions, these organisms remain an endless
source of unconventional biology. j
REF ERENC ES AND NOTES
1. V. Kande Betu Ku Mesu et al., Lancet Glob. Health 9,
e999 (2021).
2. S. P. S. Rao et al., Science 380, 1349 (2023).
3. S. Trindade et al., Cell Host Microbe 19, 837 (2016).
4. M. De Rycker, S. Wyllie, D. Horn, K. D. Read, I. H. Gilbert,
Nat. Rev. Microbiol. 21, 35 (2023).
5. T. Kulikowicz, T. A. Shapiro, J. Biol. Chem. 281, 3048
(2006).
6. K. R. Vann, A. A. Oviatt, N. Osheroff, Biochemistry 60,
1630 (2021).
7. B. D. Bax et al., Nature 466, 935 (2010).
8. Y. Pommier, E. Leo, H. Zhang, C. Marchand, Chem. Biol.
17, 421 (2010).
9. S. Richards et al., Trends Parasitol. 37, 831 (2021).
GRAPHIC: K. HOLOSKI/SCIENCE
10. C. E. Clayton, Curr. Opin. Microbiol. 32, 46 (2016).
AC K N OW LEDGMENTS
R.A. and I.A. acknowledge support from National Institutes
of Health grants AI101057 and AI152408 (to R.A.) and
AI113157 (to I.A.).
10.1126/science.adi5925
SCIENCE science.org
MOLECULAR BIOLOGY
Mapping RNA translation
A new method maps the location of thousands of
translating RNAs in cells and tissues
By Rong Fan
A
lmost a decade ago, spatial transcrip-
tomics (1) emerged as a transforma-
tive technology for biological and
biomedical research but is largely
limited to detecting gene expression.
Yet the life of an RNA molecule is
rich and dynamic, starting from DNA tran-
scription to nascent RNA synthesis, splic-
ing, export, translation, and degradation.
Therefore, it is desirable to not only detect
the presence of a given RNA molecule but
A three-probe system for mapping translation
Barcoded “padlock” and primer probes bind to an mRNA molecule of interest. Amplification of the padlock
probe cannot occur unless the mRNA is adjacent to a ribosome, enabling a splint probe to circularize the
padlock probe. After amplification, the barcodes are detected with fluorescent probes to provide information
on the location of translating mRNAs.
Splint probe hybridized
to ribosomal RNA
Ribosome
Barcoded
padlock
probe
5' mRNA
also to have information on its stage of life.
On page 1337 of this issue, Zeng et al. (2)
describe a spatial transcriptomic approach
that is specific for mRNAs that are undergo-
ing translation, enabling the study of their
biological function in situ.
Previously, two major approaches were
undertaken to realize spatial transcrip-
tomics. Repeated imaging coupled with
a combinatorial coding method extended
the capability of single-molecule fluores-
cence in situ hybridization (smFISH) to
enable the detection of mRNA correspond-
ing to 10,000 genes, approaching whole-
transcriptome scale (3, 4). Next-generation
sequencing (NGS) of mRNA labeled with
location-specific DNA tags either captured
on a solid surface (5) or directly in tissue
(6) enabled genome-wide unbiased tran-
scriptome profiling at cellular resolution.
Department of Biomedical Engineering, Yale University,
New Haven, CT, USA. Email: rong.fan@yale.edu
The purpose of mRNA is to make pro-
teins through translation. However, mRNA
and protein levels are poorly correlated,
and one hypothesis is that the number of
actively translating mRNA molecules is
more closely associated with protein ex-
pression. Thus, single-cell or spatial profil-
ing of translating mRNAs might provide a
surrogate measure of protein expression,
as well as information on RNA processing
and function.
The method developed by Zeng et al.
(RIBO-map) is built upon a targeted in situ
Fluorescent probe
Amplified
padlock DNA
Barcode
Barcoded
primer
probe
AAA
hybridization approach called STARmap
(7) but uses three probes to selectively de-
tect and amplify ribosome-bound mRNAs
(see the figure). A “padlock” probe targets
mRNA molecules of interest and encodes a
gene-specific barcode. A second probe binds
to the padlock probe and serves as a primer
for rolling circle amplification (RCA), which
replicates a circular DNA template into a
long, single-stranded stretch of DNA (the
amplicon) containing  multiple copies of
the same DNA sequence. However, for RCA
to proceed, a third probe must be present;
this probe hybridizes to ribosomal RNAs
and acts as a splint to circularize the pad-
lock probe. After circularization and RCA,
the barcodes in the amplicon are identified
with in situ sequencing, which involves
fluorescently labeled nucleotides that are
imaged to provide information on the loca-
tion of the target mRNA. Without hybrid-
ization to the third probe, which requires
proximity of the mRNA to a ribosome, the
30 JUNE 2023 • VOL 380 ISSUE 6652 1321
I NS I GHTS | P E R S P E C T I V E S
padlock probe fails to be circularized for
amplification, thus allowing discrimination
between actively translating and nontrans-
lating mRNA molecules.
Zeng et al. used RIBO-map to profile
the translation of 981 genes in single HeLa
cells and identified translation of differ-
ent mRNAs at different stages of the cell
cycle. RIBO-map was also used to identify
translationally co-regulated gene modules,
which previous single-molecule and single-
gene imaging techniques were not able to
detect. In mouse brain slices, RIBO-map de-
tected 5413 genes across
119,173 single cells—an
“…spatial epigenomics
impressive throughput.
Interestingly, strong dis-
cordance was observed
and multiomics could
between the translatome
and transcriptome in
be integrated with
oligodendrocyte lineage RIBO-map to link
cells that may contribute
to distinct pathways of RNA translation to
oligodendrocyte matura-
tion and heterogeneity.
epigenetic control,
Furthermore, subcellular protein expression, and
differences in the regula-
tion of translation (for metabolic function.”
example, between the cell
body and periphery) were observed in the
hippocampus.
Although RIBO-map has demonstrated
spatial mapping of an impressive number
of genes in tissue, it is still limited to detect-
ing a finite set of known mRNA molecules.
Previously, genome-wide unbiased profil-
ing of translating mRNAs was achieved by
fixing and precipitating ribosome-bound
mRNAs for NGS (8) and was recently ex-
tended to single cells (9). However, this
approach has a low throughput and has
not yet been applied to spatial mapping.
Spatially profiling translating mRNAs base
by base in complex tissues would provide
the ability to detect sequence-specific RNA–
protein interactions and translational activ-
ity but remains a challenge.
Previous studies have explored other
stages of the life cycle of mRNA. For exam-
ple, transcriptional bursts were observed at
sites where nascent RNAs are synthesized
(transcription active sites, TAS) in individual
nuclei (10). A multiplexed smFISH technique
called seqFISH and probes designed to la-
bel the TAS of >10,000 genes were used to
capture the nascent transcriptome in single
cells (11), but the technique has not yet been
applied to complex tissues. Global run-on se-
quencing permits unbiased measurement of
nascent RNAs at genome scale (12) but has
not been successfully implemented in single-
cell or spatial profiling. To explore the matu-
ration stage (that is, splicing) of an RNA
molecule, NGS-based approaches have been
1322 30 JUNE 2023 • VOL 380 ISSUE 6652
used for single-cell isoform measurement in
mouse motor cortex and cerebellar cells, as
well as for spatial isoform mapping (13).
The dynamic and multistaged life cycle of
RNA is not a linear path from transcription
to translation and degradation but is actively
regulated by a large array of RNA bind-
ing proteins and other RNA species. Thus,
mapping the location of RNA–protein and
RNA–RNA interactions will be important for
understanding the regulation of gene expres-
sion. In addition, chemical modification of
RNAs alters stability and translation, repre-
senting another important
mechanism that would be
interesting to map spa-
tially and at genome scale.
The three-probe method
developed by Zeng et al.
has the potential to be
modified to investigate
other stages of the RNA
life cycle; for example,
mRNA degradation rate
might be studied by
detecting endonuclease-
bound mRNA molecules.
In the future, the latest
breakthroughs in spatial
epigenomics (14, 15) and multiomics could
be integrated with RIBO-map to link RNA
translation to epigenetic control, protein
expression, and metabolic function. Such an
approach would span the central dogma of
molecular biology, which defines the flow of
genetic information from DNA to RNA, to
protein and cellular function. j
R EFER ENCES AN D N OT ES
1. V. Marx, Nat. Methods 18, 9 (2021).
2. H. Zeng et al., Science 380, eadd3067 (2023).
3. E. Lubeck, A. F. Coskun, T. Zhiyentayev, M. Ahmad, L. Cai,
Nat. Methods 11, 360 (2014).
4. K. H. Chen, A. N. Boettiger, J. R. Moffitt, S. Wang, X.
Zhuang, Science 348, eaaa6090 (2015).
5. P. L. Ståhl et al., Science 353, 78 (2016).
6. Y. Liu et al., Cell 183, 1665 (2020).
7. X. Wang et al., Science 361, eaat5691 (2018).
8. N. T. Ingolia, S. Ghaemmaghami, J. R. Newman, J. S.
Weissman, Science 324, 218 (2009).
9. M. VanInsberghe, J. van den Berg, A. Andersson-Rolf, H.
Clevers, A. van Oudenaarden, Nature 597, 561 (2021).
10. A. M. Femino, F. S. Fay, K. Fogarty, R. H. Singer, Science
280, 585 (1998).
11. S. Shah et al., Cell 174, 363 (2018).
12. L. J. Core, J. J. Waterfall, J. T. Lis, Science 322, 1845
(2008).
13. A. Joglekar et al., Nat. Commun. 12, 463 (2021).
14. Y. Deng et al., Science 375, 681 (2022).
15. Y. Deng et al., Nature 609, 375 (2022).
A CKNOWLEDGMENTS
R.F. acknowledges funding from the National Institutes of
Health (U54AG076043, U54AG079759, UG3CA257393,
UH3CA257393, R01CA245313, RF1MH128876,
U54CA274509) to support his research in spatial omics. R.F.
is scientific cofounder of and adviser to IsoPlexis, Singelron
Biotechnologies, and AtlasXomics with financial interest.
10.1126/science.adi6844
CELL BIOLOGY
Staving off
cell death
A signaling pathway that
senses energy stress
opposes necroptotic cell death
By D. Grahame Hardie
M
aintenance of size and shape of or-
gans in multicellular organisms is
determined by the balance between
cell proliferation and cell death.
It was thought that apoptosis, a
mechanism driven by cleavage of in-
tracellular proteins by caspases, was the only
genetically programmed cell death pathway.
However, it became apparent that necrosis,
once regarded merely as an accidental form
of cell death, can be a regulated process, with
necroptosis being the major subtype (1). A
key player in both apoptosis and necropto-
sis is receptor-interacting protein kinase 1
(RIPK1). On page 1372 of this issue, Zhang et
al. (2) report that the cellular energy sensor
adenosine monophosphate (AMP)–activated
protein kinase (AMPK) phosphorylates and
inactivates RIPK1, opposing necroptosis and
thus promoting cell survival.
During apoptosis, cell fragments are re-
moved by phagocytic cells in an immunologi-
cally silent manner, whereas during necrop-
tosis, cells burst, triggeringnflammation.
The latter might seem deleterious—indeed,
necroptosis is implicated in several inflam-
matory disorders in humans (3). However,
it has been suggested that necroptosis may
have arisen as a backup pathway to kill vi-
rus-infected cells, in which viral proteins
sometimes inhibit apoptosis (4).
The classical cell-extrinsic trigger of apop-
tosis and necroptosis is tumor necrosis factor
(TNF). Its binding to TNF receptors (TNFRs)
recruits RIPK1 and TNFR type 1–associated
DEATH domain protein (TRADD), by associ-
ation of their death domains with that of the
receptor, to form a TNFR-bound signaling
complex (complex-I). Subsequent polyubiqui-
tylation of lysine residues on RIPK1 generates
scaffolds that recruit various components to
initiate nuclear factor kB (NF-kB) signaling
and promote cell survival [this scaffolding
role of RIPK1 does not require its kinase ac-
tivity (3)]. Alternatively, in response to TNFR
stimulation, RIPK1 forms two intracellular
School of Life Sciences, University of Dundee, Dundee,
Scotland, UK. Email: d.g.hardie@dundee.ac.uk
science.org SCIENCE
 complexes: complex-IIa, which Opposing necroptosis Why should AMPK oppose
also contains caspase-8 andFAS- Necroptosis is activated by TNFRs, which lead to formation of the necrosome that necroptosis but not apoptosis?
associated death domain protein contains RIPK1-RIPK3 amyloid-like filaments that activate MLKL. However, if cells One possibility is that, during
(FADD), and complex-IIb (also experience an insult that causes energy stress, AMPK is activated. This phospho- apoptosis, cellular ATP levels are
called the necrosome), which rylates Ser416 (S416) of RIPK1, which inhibits the activating S166 phosphorylation, maintained until the last mo-
contains the protein kinase thus repressing necroptosis. Amino acid numbers are for human proteins. ment, whereas ATP depletion is
RIPK3 and mixed lineage kinase associated with necrosis and can
domain-like (MLKL). MLKL has TNFR even be a trigger for it (8). Thus,
a pseudokinase domain that the AMPK pathway is more
binds adenosine triphospa1te likely to be activated during ne-
(ATP) but lacks two amino acids Necroptosis suppression crosis rather than apoptosis.
required for kinase activity (4). Energy stress causes increased ADP:ATP ratio, It is becoming apparent that
RIPK1 which is amplified by adenylate kinases into
When caspase-8 in complex- larger increases in AMP:ATP that activate AMPK.
a major function of AMPK is
IIa is activated, it initiates apop- Ubiquitin to maintain cellular mitochon-
tosis while also cleaving RIPK1 Energy ATP ADP drial networks, which are the
at Asp324 within its intermediate Necroptosis activation stress main source of ATP. When a
Upstream pathways (e.g.,
domain, thus inhibiting necrop- RIPK1 deubiquitylation) can
mitochondrial network becomes
tosis. Conversely, in the necro- trigger autophosphorylation Adenylate damaged, it must first be cleaved
some, RIPK1 and RIPK3 interact of RIPK1 at S166. kinase into segments small enough to
AMP
through their RIP homotypic in- AMPK be removed by mitophagy, which
teraction motifs (RHIMs) to form AMPK achieves by promoting
amyloid-like protein filaments. Inhibition of S166 fission and inhibiting fusion of
phosphorylation
Through a complex sequence P P represses necroptosis.
mitochondria through the phos-
of association and autophos- phorylation of mitochondrial
RIPK1 N S166 S416 C
phorylation events (5), RIPK3 fission factor (MFF) (9) and mi-
Autophos- Kinase Intermediate RHIM Death
is activated and phosphorylates phorylated RIPK1 tochondrial fission regulator 1
domain domain domain
MLKL, which goes on to execute and RIPK3 interact like (MTFR1L) (10), respectively.
P
necroptosis (4). through RHIM Next, AMPK activates mitoph-
AMPK is expressed in nearly all domains to form RIPK3 N S227 C agy by phosphorylating unc-51–
amyloid-like Kinase RHIM
eukaryotic cells and is switched filaments. RIPK3 then
like autophagy-activating kinase
on by cellular energy stress that domain (ULK1) (11). Finally, by phos-
phosphorylates MLKL.
P
it normally senses by detect- phorylating folliculin-interact-
ing increases in AMP relative to MLKL N S358 C ing protein 1 (FNIP1) (12), AMPK
ATP (see the figure), although MLKL executes Four-helix Pseudokinase stimulates both lysosomal and
glucose starvation can also be necroptosis via bundle domain mitochondrial biogenesis—the
its N-terminal
sensed by an AMP-independent
four-helix bundle. Necrosome former to ensure an adequate
mechanism (6). AMPK then supply of lysosomes to support
Necroptosis
phosphorylates numerous down- mitophagy, and the latter to re-
ADP, adenosine diphosphate; AMP, adenosine monophosphate; AMPK, AMP-activated protein kinase; ATP,
stream target proteins (7), adenosine triphosphate; MLKL, mixed lineage kinase domain-like; P, phosphorylation; RHIM, RIP homotypic place damaged mitochondrial
switching on catabolic pathways interaction motif; RIPK, receptor-interacting protein kinase; TNFR, tumor necrosis factor receptor. components that had been re-
that generate ATP from adenos- cycled by mitophagy. How does
ine diphosphate (ADP) while switching off created mice in which Ser415 of RIPK1 was this fit in with the findings of Zhang et al.?
ATP-consuming processes, thus preserving replaced by a nonphosphorylatable ala- In the longer term, maintenance of the mi-
cellular energy status and promoting sur- nine. These mice were viable and initially tochondrial network by AMPK would help
vival. Zhang et al. deprived mouse embryonic appeared normal, but increased cell death to ensure cellular energy homeostasis, thus
fibroblasts (MEFs) of glucose and noticed a and inflammation occurred when the liver making it less likely that necroptosis would
reduced mobility of RIPK1 during gel elec- was stressed by ischemia and reperfusion. be promoted by ATP depletion. However, if
trophoresis that was reversed by treatment Overall, these data indicate that AMPK op- necroptosis was triggered, the phosphoryla-
with a protein phosphatase, indicating that it poses necroptosis through direct phosphor- tion of RIPK1 by AMPK might delay the pro-
was a result of phosphorylation. The mobility ylation of Ser415 or Ser416 on RIPK1. Because cess to allow time for repair of the mitochon-
shift was abolished in AMPK-deficient MEFs. this appears to reduce the phosphorylation drial network and stave off cell death. j
Further analyses identified eight phosphory- of Ser166 within the activation loop, which
RE FE RE N CES AN D N OT ES
lation sites on RIPK1, of which one (Ser415 in is thought to be an autophosphorylation
416
1. P. Vandenabeele et al., Nat. Rev. Mol. Cell Biol. 11, 700
mice and Ser in humans) was increased by event, AMPK-mediated phosphorylation (2010).
glucose deprivation. The sequence around of RIPK1 may inhibit its kinase activity, al- 2. T. Zhang et al., Science 380, 1372 (2023).
3. W. Li, J. Yuan, Front. Immunol. 14, 1159743 (2023).
this site is a reasonable fit to the recogni- though this was not directly demonstrated. 4. P. E. Czabotar, J. M. Murphy, FEBS J. 282, 4268 (2015).
tion motif for AMPK, and in cell-free assays, There have been attempts to examine the 5. X.-N. Wu et al., Cell Death Differ. 21, 1709 (2014).
6. C.-S. Zhang et al., Nature 548, 112 (2017).
GRAPHIC: K. HOLOSKI/SCIENCE

AMPK phosphorylated Ser416 in RIPK1. effects of AMPK activation on apoptosis, but

To test the role of AMPK in the regula-
tion of necroptosis in vivo, Zhang et al. in-
duced deletion in mice of the genes encod-
ing both AMPK catalytic subunits (a1 and
a2) and noted increased activation of RIPKI
and cell death in several tissues. They also
these studies have often been marred by the
use of nonselective modulators of AMPK, and
no clear picture has emerged. The findings
of Zhang et al., demonstrating both a mecha-
nism and a molecular target by which AMPK
opposes necroptosis, are a clear step forward.
7. G. R. Steinberg, D. G. Hardie, Nat. Rev. Mol. Cell Biol. 24,
255 (2023).
8. Y. Tsujimoto, Cell Death Differ. 4, 429 (1997).
9. E. Q. Toyama et al., Science 351, 275 (2016).
10. L. Tilokani et al., Sci. Adv. 8, eabo7956 (2022).
11. D. F. Egan et al., Science 331, 456 (2011).
12. N. Malik et al., Science 380, eabj5559 (2023).
10.1126/science.adi6827

SCIENCE science.org 30 JUNE 2023 • VOL 380 ISSUE 6652 1323

I NS I GHTS

els underestimate emissions reductions,

P OLICY FORUM though many of IRA’s largest provisions are
represented and the degree of additional-
ENERGY AND CLIMATE POLICY ity of the remaining provisions is unclear
(materials and methods S7). Other sim-

Emissions and energy impacts plifications (e.g., limited representations

of frictions associated with infrastructure
deployment, supply chains, and non-cost

of the Inflation Reduction Act barriers) could result in higher emissions

relative to modeled outcomes. These uncer-
tainties imply that model results should not
Economy-wide emissions drop 43 to 48% below 2005 levels be interpreted as predictions (materials and
by 2035 with accelerated clean energy deployment methods S1).
To evaluate impacts on emissions and en-
ergy systems, IRA scenarios are compared
By John Bistline1, Geoffrey Blanford1, Maxwell Brown2, Dallas Burtraw3, Maya Domeshek3, to their counterfactual reference scenarios
Jamil Farbes4, Allen Fawcett5, Anne Hamilton2, Jesse Jenkins6, Ryan Jones4, Ben King7, Hannah without IRA (materials and methods S2).
Kolus7, John Larsen7, Amanda Levin8, Megan Mahajan9, Cara Marcy5, Erin Mayfield10, James IRA scenarios focus on central estimates of
McFarland5, Haewon McJeon11, Robbie Orvis9, Neha Patankar12, Kevin Rennert3, Christopher climate and energy provisions, which are
Roney1, Nicholas Roy3, Greg Schivley13, Daniel Steinberg2, Nadejda Victor14, Shelley Wenzel9, not harmonized across models.
John Weyant15, Ryan Wiser16, Mei Yuan17, Alicia Zhao11
EMISSIONS REDUCTION PATHWAYS

I
f goals set under the Paris Agreement
are met, the world may hold warming
well below 2°C (1); however, parties are
not on track to deliver these commit-
ments (2), increasing focus on policy im-
plementation to close the gap between
ambition and action. Recently, the US gov-
ernment passed its most prominent piece
of climate legislation to date—the Inflation
Reduction Act of 2022 (IRA)—designed to
invest in a wide range of programs that,
among other provisions, incentivize clean
energy and carbon management, encourage
electrification and efficiency measures, re-
duce methane emissions, promote domestic
supply chains, and address environmental
justice concerns (3). IRA’s scope and com-
plexity make modeling important to un-
derstand impacts on emissions and energy
systems. We leverage results from nine in-
dependent, state-of-the-art models to exam-
ine potential implications of key IRA pro-
visions, showing economy-wide emissions
reductions between 43 and 48% below 2005
levels by 2035.
This multimodel analysis provides a
range of decision-relevant information. For
example, international policy-makers and
negotiators need to track progress toward
Paris Agreement pledges, and assessing
IRA’s impacts is important to monitor US
efforts and to provide a template for mea-
suring the performance of other sectors and
jurisdictions. Federal and state policy-mak-
ers can use this IRA analysis to compare
updated baselines with policy targets—for
emissions, electric vehicle deployment, and
others—to understand the magnitude of ad-
ditional policies and private-sector actions
needed to narrow implementation gaps.
Electric companies need to know how long
IRA incentives will be available, because
these subsidies can continue until electric-
ity emissions are below 25% of their 2022
levels, which requires national models to
evaluate. Industry- and technology-specific
deployment can support investors, technol-
ogy developers, researchers, and companies
to quantify market opportunities.
MODELING IRA PROVISIONS
Some of the models used in our analysis in-
formed legislative debates preceding IRA’s
passage (4–7). We build on these prelimi-
nary analyses by updating IRA represen-
tations, increasing the number of models,
and providing a systematic comparison of
decision-relevant metrics. Multimodel stud-
ies highlight the robustness of insights and
potential uncertainties to alternate model
structures and input assumptions. Models
in this study vary in their coverage and im-
plementation of IRA provisions (table S1).
These differences are due to the models’
scopes (e.g., six models of the full US energy
system versus three that focus on the power
sector only) and resolutions (e.g., level of
technological and sectoral detail). Varia-
tions in IRA implementation across models
are also caused by the bill’s complexity and
pending guidance from government agen-
cies, which require subjective judgments
from modeling teams. The partial coverage
of IRA provisions could imply that mod-
Economy-wide emissions reductions from
IRA are 33 to 40% below 2005 levels in
2030 across multisector models with a 37%
average (see the first figure). This reflects a
range of IRA provisions modeled, input as-
sumptions, and model structures (materials
and methods S1). The 2030 range with IRA
of 33 to 40% is a substantial reduction from
the reference without IRA incentives, which
is 25 to 31% below 2005 (28% average).
Emissions reductions from IRA grow over
time and lead to 43 to 48% declines by 2035
from 2005 (compared with 27 to 35% in the
reference). IRA helps to narrow the imple-
mentation gap in achieving the US 2030
target to reduce net greenhouse gas (GHG)
emissions by at least 50% (8, 9). The emis-
sions gap is 1.0 to 1.6 gigatonnes of carbon
dioxide equivalent per year (Gt-CO2e/year)
without IRA, falling to 0.5 to 1.1 Gt-CO2e/
year with IRA [19 to 25 percentage points
(p.p.) to 10 to 17 p.p., fig. S6].
Emissions reductions are not evenly dis-
tributed across sectors (see the first figure).
Most IRA-induced mitigation comes from
electricity, representing 38 to 80% of 2030
reductions (64% average) from the refer-
ence in economy-wide models. There is
consistency across models that IRA will ac-
celerate power sector decarbonization (fig.
S7). In 2030, power sector emissions with
IRA are 47 to 83% (68% average) below
their 2005 levels compared to 41 to 60%
(51% average) in the reference. IRA-induced
reductions continue through 2035, and the
range narrows to 66 to 87% (77% average)
below 2005 levels with IRA compared to 39

1
Electric Power Research Institute, Palo Alto, CA, USA. 2National Renewable Energy Laboratory, Golden, CO, USA. 3Resources for the Future, Washington, DC, USA. 4Evolved Energy Research, San
Francisco, CA, USA. 5US Environmental Protection Agency, Washington, DC, USA. 6Princeton University, Princeton, NJ, USA. 7Rhodium Group, Washington, DC, USA. 8Natural Resources Defense
Council, Washington, DC, USA. 9Energy Innovation, San Francisco, CA, USA. 10Dartmouth College, Hanover, NH, USA. 11Center for Global Sustainability, University of Maryland, College Park, MD,
USA. 12Binghamton University, Binghamton, NY, USA. 13Carbon Impact Consulting, New Providence, NJ, USA. 14National Energy Technology Laboratory, Pittsburgh, PA, USA. 15Stanford University,
Stanford, CA, USA. 16Lawrence Berkeley National Laboratory, Berkeley, CA, USA. 17MIT Joint Program on the Science and Policy of Global Change, Cambridge, MA, USA. Email: jbistline@epri.com

1324 30 JUNE 2023 • VOL 380 ISSUE 6652 science.org SCIENCE

 to 68% (53% average) in the reference (fig. Cross-model comparison of US emissions reductions under the
S7B), short of the goal of totally “carbon
pollution-free electricity” by 2035 (8). The
Inflation Reduction Act (IRA)
Results reflect modeling of both IRA and reference scenarios, relative to the year 2005. The six models used
technology-neutral tax credits under IRA across both panels are EPS-EI, GCAM-CGS, MARKAL-NETL, NEMS-RHG, REGEN-EPRI, and RIO-REPEAT.
for zero-emitting resources continue after
2032 until electricity emissions are 25% of Historical Reference IRA
2022 levels. Three of nine models reach this
threshold by 2035.

Economy-wide emissions (percent below 2005 levels)

0

EFFECTS OF IRA ON THE POWER SECTOR

IRA includes many electricity-related pro- -10
visions (table S1), including investment
(30%) and production [$27.5 per megawatt-
-20
hour (MWh), 10 year] tax credits for clean
Reference *
electricity resources, tax credits for energy +
* 2030: Min. = −25%, Avg. = −28%, Max. = −31%
storage and carbon capture, and tax credits -30 +2035: Min. = −27%, Avg. = −31%, Max. = −35%
to maintain existing nuclear plants. Some IRA
provisions involve long-term extensions of § 2030: Min. = −33%, Avg. = −37%, Max. = −40%
pre-IRA tax credits (e.g., for wind and so- -40 × 2035: Min. = −43%, Avg. = −45%, Max. = −48%
§
lar), some involve increases in tax credit
levels (e.g., carbon capture credits, bonuses
for production and investment credits), and -50 ×
others are new (e.g., support for existing 2030 US target: 50–52%
nuclear and stand-alone storage).
-60
Models consistently show that IRA leads
2005 2010 2015 2020 2025 2030 2035
to large increases in wind and solar de-
ployment but with substantial variation in Historical and projected economy-wide greenhouse gas (GHG) emissions in CO2 equivalent terms. Historical
magnitudes (fig. S13A). Across all models, emissions and 100-year global warming potential values (for models representing non-CO2 GHGs) are based on
growth rates from 2021 to 2035 range from the US Environmental Protection Agency’s “Inventory of US Greenhouse Gas Emissions and Sinks.”
10 to 99 GW/year for solar and wind under
IRA (58 GW/year average), which is more
Net Electric Industry Buildings Transport Other CO2 Non-CO2 GHGs Land
than twice the average of 27 GW/year with-
out IRA and higher than the record 33 GW 200
installed in 2021. There is wide variation
in the expected increase in energy storage
across models, 1 to 18 GW/year (7 GW/year
0
average), compared with 0 to 8 GW/year in
the reference.
Results also exhibit reductions of un-
GHG change from reference (Mt-CO2e)

abated coal generation [i.e., without any −200

carbon capture and storage (CCS)], ranging
from 38 to 92% declines from 2021 levels
by 2030 with IRA (see the second figure) −400
versus 3 to 60% without IRA. Five models
show retrofits of some share of coal capac-
ity with CCS, driven by the high value of
−600
tax credits for stored CO2 (increasing from
$50/t-CO2 historically to $85/t-CO2) (fig.
S13). Although all models suggest that gas-
fired capacity will increase to provide firm −800
capacity as load grows and coal retires,
GRAPHIC: D. AN-PHAM/SCIENCE BASED ON J. BISTLINE AT AL.

most models also suggest that natural gas

generation will decline under IRA relative −1000
to today’s levels (fig. S13).
Overall, generation shares from low-emit-
ting technologies—including renewables,
−1200
nuclear, and CCS—in 2030 vary between 49
2030

2030

2030

2030

2030

2030

2030

2030

2030
2035

2035

2035

2035

2035

2035

2035

2035

2035
2025

2025

2025

2025

2025

2025

2025

2025

2025

and 82% (68% average) across models with

IRA (fig. S13B), up from about 40% today EPS-EI GCAM-CGS Haiku-RFF* IPM-NRDC* MARKAL- NEMS-RHG ReEDS- REGEN- RIO-
and from 46 to 65% without IRA (54% aver- NETL NREL* EPRI† REPEAT
age), an 11 to 33 p.p. increase. Power sector Emissions reductions by sector and model over time under IRA scenarios relative to reference levels. Models
generation and emissions outcomes under with * designate that electric sector IRA provisions only are represented, and † denotes energy CO2 IRA
IRA are more closely aligned by 2035 across provisions only. Additional information on participating models and study assumptions can be found in
models (figs. S7 and S13). supplementary materials and methods S1 and S2.

SCIENCE science.org 30 JUNE 2023 • VOL 380 ISSUE 6652 1325

I NS I GHTS | P O L I C Y F O RU M
IMPLICATIONS OF IRA ON
END-USE DEMAND
IRA reduces transport emissions by acceler-
ating electrification. Across models, electric
vehicles (EVs) are 32 to 52% of new light-
duty vehicle sales by 2030 with IRA (41%
average), compared with 22 to 43% (31%
average) in the reference (see the second
figure) (fig. S14), which is several times 2022
sales levels of about 7%. EV sales decrease
or flatten between 2030 and 2035 owing to
the expiration of IRA tax credits. The electri-
fied service demand share, stock share, and
emissions lag new sales, given the turnover
rate of the fleet (fig. S15). EV credits in IRA
also exist for commercial vehicles, includ-
ing vans, buses, and trucks. These incentives
help electrify these segments, representing
about one-third of transport electricity de-
mand by 2030 (fig. S17). Transport exhibits
the highest growth in electrification, though
the magnitude varies by model (2 to 6% of
final energy use in the sector by 2030 with
IRA, 1 to 4% in the reference).
IRA incentives also encourage building
efficiency and electrification, especially the
adoption of heat pumps for space and water
heating. Buildings currently have the highest
electrification share of end-use sectors (just
under 50% of final energy use in the sector),
which increases as end-uses electrify (fig.
S16). Fuel switching from fossil fuels to elec-
tricity in buildings, transport, and industrial
sectors leads electricity’s economy-wide share
of final energy to increase from about 21%
today to 23 to 26% by 2030 (see the second
figure) and 25 to 29% by 2035.
A driver charges an electric vehicle at a charging station in California in 2022. The Inflation Reduction Act is
expected to help drive electrification of the transportation sector.
1326 30 JUNE 2023 • VOL 380 ISSUE 6652
These projected changes in demand may
drive the first sustained period of declining
petroleum use in US history (fig. S18). By
2030, IRA scenarios show an 11 to 32% (22%
average) decrease in petroleum consumption
from 2005, but much of this happens in the
reference (11 to 31% with 20% average) owing
to the competitiveness of transport electrifi-
cation and continued efficiency improvement
in the remaining internal combustion engine
fleet. Natural gas consumption also decreases
with IRA relative to the reference scenario,
especially in the power sector, but declines
relative to its historical peak are not as high
as they are for petroleum and coal. IRA may
catalyze several new markets in industry and
fuels, including CCS, hydrogen, biofuels, and
sustainable aviation fuels. Tax credits for cap-
tured CO2 may accelerate emissions reduc-
tions across a variety of industries and lead to
CCS deployment in the industrial sector, fuel
production, and the power sector (fig. S22).
SOCIETAL IMPLICATIONS OF IRA
The climate-related benefits of IRA are es-
timated to be substantial (fig. S10), even
though there is uncertainty about the mag-
nitude of the social cost of CO2 (10) (fig. S9)
and model-specific emissions impacts of
IRA (see the second figure). Climate benefits
range from $44 billion to $220 billion annu-
ally by 2030 across models using central so-
cial cost of CO2 values with a 2% near-term
discount rate ($20 billion to $100 billion with
a 3% rate). Implied average abatement costs
of IRA incentives per unit of CO2 reduced
range from $27 to $102/t-CO2 with an aver-
age of $61/t-CO2 across all models (fig. S11).
Although these costs are higher for economy-
wide models (materials and methods S4),
abatement costs are much lower than many
updated social cost of CO2 estimates, even be-
fore accounting for other co-benefits.
Declining fossil fuel use lowers not only
GHG emissions but also conventional air
pollutants, which improves public health
outcomes. Figure S12 compares reductions in
SO2 and NOx in the power sector and across
the economy. Individual studies indicate that
monetized health benefits from power sector
air pollution reductions alone are $9 billion
to $22 billion annually by 2030 (11, 12), and
$53 billion annually by 2030 from particulate
matter reductions across the economy (8).
These declines in fossil fuel use gener-
ally mean that IRA lowers energy costs for
households and businesses (fig. S19), despite
increases in electricity spending. The magni-
tudes of these consumer cost changes vary
over time and across the four economy-wide
models reporting fuel expenditures, but net
spending declines $2 billion to $26 billion/
year by 2030 ($13 to $190 per household)
relative to the reference and $10 billion to
$52 billion per year by 2035 ($73 to $370 per
household). Electrification and accelerated
investments increase electricity expenditures
for most economy-wide models, even though
there is a reduction in total net energy ser-
vice costs. Incentives for the adoption of end-
use technologies such as EVs and heat pumps
also can lower capital and maintenance costs.
The magnitude and composition of tax
credit value under IRA vary by model (fig.
S21). Estimates from economy-wide mod-
els suggest that $330 billion to $870 billion
could be spent on tax credits through 2030
($510 billion average), which suggests that
uptake of IRA incentives could be greater
than initial Congressional Budget Office–
Joint Committee on Taxation estimates indi-
cated (3). Cumulative tax credit expenditures
through 2035 span $640 billion to $1300 bil-
lion ($910 billion average), indicating that
some IRA impacts may take longer to mate-
rialize. Individual studies perform analysis
to understand broader societal implications
of IRA, including improving distributional
outcomes (11), increasing jobs (5, 6, 12), and
bolstering domestic manufacturing (13).
PHOTO: FREDERIC J. BROWN/AFP/GETTY IMAGES
CONCLUSIONS AND KEY UNKNOWNS
Although IRA accelerates decarbonization,
including beyond 2030, no models indicate
that the 2030 US climate target would be met
with IRA alone. Overall, the analysis suggests
that IRA may have its largest effects in the
power sector, as its incentives amplify trends
already underway and lower decarbonization
costs. Modeling IRA impacts is challenging,
because net effects depend on clean energy
science.org SCIENCE

Key Inflaton Reduction Act (IRA) indicators across models
Indicators reflect values estimated for the year 2030. Clean generation and capacity shares include renewables,
nuclear, and carbon capture and storage–equipped generation. Electric vehicle sales include battery or plug-in
hybrid electric vehicles. Model-specific values for these metrics are provided in table S6.
Reference IRA Electric sector metrics
0 50
Electric sector CO2 reduction
(% from 2005)
Clean generation share (%)
Clean capacity share (%)
Unabated coal reduction
(% from 2021)
Economy-wide GHG reduction
(% from 2005)
Share of new vehicle sales
that are electric (%)
Electricity share of final energy (%)
Petroleum reduction (% from 2005)
adoption, producer choices, household pur-
chases, and actions by policy-makers. Several
key unknowns remain.
For example, lowering clean energy costs
may increase ambitions of other federal
agencies, state and local governments, and
companies, though there are many complex
drivers of such goals. These dynamic effects
of ratcheting ambition are not accounted
for in the modeled scenarios described here
but may be key to closing the 2030 imple-
mentation gap. In recent months, the US
Environmental Protection Agency has re-
leased proposed standards that target emis-
sions from cars, trucks, and power plants.
These complementary regulations are aimed
GRAPHIC: D. AN-PHAM/SCIENCE BASED ON J. BISTLINE ET AL.
at leveraging IRA incentives to further accel-
erate decarbonization trends and lower costs
of future regulations.
Also, expanded and new tax credits in
IRA—combined with grants, loans, fees,
domestic manufacturing incentives, and
Infrastructure Bill incentives—support
technologies that have experienced recent
growth (e.g., wind, solar, batteries) and cata-
lyze new markets for other decarbonization
options (e.g., CCS, new nuclear, hydrogen,
biofuels). Making clean technologies cheap
to accelerate adoption can, in turn, buy
SCIENCE science.org
Economy-wide metrics
100
Min. Max.
Avg.
down learning curves and encourage fur-
ther deployment. By making low-emitting
technologies cheaper domestically, IRA may
have international spillovers, bolstering the
political economy of those making their own
investments and policies. Emerging technol-
ogies have larger uncertainty about induced
technical change owing to their limited de-
ployment and nascent markets.
Models attempt to capture many economic
factors that could influence technology adop-
tion, but several implementation challenges
are difficult to model, including the scale-up
of supply chains and materials, siting and
permitting, infrastructure expansion, net-
work effects, non-cost barriers to consumer
uptake of incentives, and the economic in-
cidence of subsidies. These questions have
prompted debate about permitting reform to
streamline the infrastructure approval pro-
cess and increase the pace of clean energy de-
ployment, though legislative disagreements
have emerged about the types of projects
included and about balancing these reforms
with environmental safeguards and commu-
nity participation. Broader macroeconomic
trends of escalating interest rates, materials
costs, and labor costs can lower decarboniza-
tion rates, though IRA incentives can help to
offset these factors (14). Additional analysis
is important for understanding potential im-
pacts of partial coverage of IRA provisions
and IRA implementation uncertainties, as
well as uncertainties about external factors,
including inflationary trends, domestic mac-
roeconomic environment, and global drivers
(materials and methods S2). j
RE FE RE N CES AN D N OT ES
1. M. Meinshausen et al., Nature 604, 304 (2022).
2. United Nations Environment Programme, “Emissions
Gap Report 2022: The Closing Window” (2022); https://
www.unep.org/resources/emissions-gap-report-2022.
3. Congressional Budget Office, “Summary: Estimated
Budgetary Effects of Public Law 117-169, to Provide for
Reconciliation Pursuant to Title II of S. Con. Res. 14”
(2022); https://www.cbo.gov/system/files/2022-09/
PL117-169_9-7-22.pdf.
4. J. Larsen et al., “A Turning Point for US Climate Progress:
Assessing the Climate and Clean Energy Provisions in
the Inflation Reduction Act” (Rhodium Group, 2022).
5. J. Jenkins et al., “Preliminary Report: The Climate and
Energy Impacts of the Inflation Reduction Act of 2022”
(REPEAT Project, 2022).
6. M. Mahajan et al., “Modeling the Inflation Reduction Act
Using the Energy Policy Simulator” (Energy Innovation,
2022).
7. N. Roy et al., “Retail Electricity Rates Under the Inflation
Reduction Act of 2022” (Resources for the Future, 2022).
8. US Government, “Reducing Greenhouse Gases
in the United States: A 2030 Emissions Target”
(2021); https://www4.unfccc.int/sites/ndcstaging/
PublishedDocments/United%20States%20of%20
America%20First/United%20States%20NDC%20
April%2021%202021%20Final.pdf.
9. J. Bistline et al., Science 376, 922 (2022).
10. K. Rennert et al., Nature 610, 687 (2022).
11. N. Roy et al., “Beyond Clean Energy: The Financial
Incidence and Health Effects of the IRA” (Resources for
the Future, 2022).
12. A. Levin, J. Ennis, “Clean Electricity Tax Credits in the
Inflation Reduction Act Will Reduce Emissions, Grow
Jobs, and Lower Bills” (National Resources Defense
Council, 2022).
13. B. Jiang et al., “US Inflation Reduction Act: A Tipping
Point in Climate Action” (Credit Suisse, 2022).
14. J. Bistline et al., Brookings Papers on Economic Activity
(Brookings Institution, 2023).
15. J. Bistline et al., “Data for Bistline, et al. (2023)
‘Emissions and Energy Impacts of the Inflation
Reduction Act,’” Zenodo (2023); https://doi.
org/10.5281/zenodo.7879732.
ACK N OW LE D G M E N TS
The views and opinions expressed in this paper are those of
the authors alone and do not necessarily represent those of
their respective institutions, the US Department of Energy
(DOE), the US Government, or other funding agencies, and
no official endorsement should be inferred. The authors
thank J. Wingenroth for assistance with social cost of CO2
data. J.F., J.J., R.J., E.M., N.P., and G.S. were funded by the
William and Flora Hewlett Foundation. B.K., H.K., and J.L. were
funded by Bloomberg Philanthropies, the William and Flora
Hewlett Foundation, and the Heising-Simons Foundation.
H.M. and A.Z. were funded by Bloomberg Philanthropies.
R.W. was funded by Lawrence Berkeley National Laboratory
under Contract no. DE-AC02-05CH11231 with the US DOE.
M.B., A.H., and D.S. were funded by the DOE Office of Policy
through the National Renewable Energy Laboratory, operated
by Alliance for Sustainable Energy, LLC, for the DOE under
Contract no. DE-AC36-08GO28308. For disclosures of com-
peting interests, refer to the supplementary materials. The
American Association for the Advancement of Science rec-
ognizes the US Government’s nonexclusive rights to use the
Work for noncommercial, governmental purposes where such
rights are established in the contract. All data and materials
associated with the analysis are available at Zenodo (15).
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adg3781
10.1126/science.adg3781
30 JUNE 2023 • VOL 380 ISSUE 6652 1327
I NS I GHTS | B O O K S
B O OKS et al .
HYDROLOGY
Water, now and then
Reflecting on humanity’s intertwined history with water,
a scientist offers actionable advice for meeting future needs
By Miriam Aczel
P
eter Gleick’s The Three Ages of Water
is a fascinating and timely examina-
tion of how water—both its abun-
dance and its scarcity—has shaped
human life and history. It is also an
important call to protect existing wa-
ter sources before they run dry that demon-
strates what we can do to transition from an
era defined by rampant water use to a more
sustainable future.
The “First Age of Water,” as Gleick con-
ceives of it, began with the formation of Earth
and continued through the emergence of hu-
man civilization around 12,000 years ago.
During this period, humans were intimately
dependent on nature for fresh water. As pop-
ulations grew, however, they began to control
the water around them, building dams and
aqueducts that helped support early agricul-
tural efforts. This era also “brought violent
conflict directly associated with the control
of vital water resources.”
According to Gleick, the “Second Age of
Water” began as humans learned to manip-
ulate the natural hydrologic cycle for their
benefit, as they built large dams and aque-
ducts and treated wastewater, and as they ir-
rigated large areas for agricultural purposes.
We are presently in this age, which has also
included many of the intellectual and techno-
logical revolutions that define modernity and
The reviewer is at the California Institute for Energy and
Environment, University of California, Berkeley, CA 94720,
USA. Email: aczel@berkeley.edu
1328 30 JUNE 2023 • VOL 380 ISSUE 6652
A man crosses a dry reservoir in Maharashtra, India,
a region experiencing an unprecedented water crisis.
improvements could reduce the electric-
ity needs of California’s water system by up
to 19% between 2015 and 2035 (1). Because
emissions reductions will take time to imple-
ment, it will also be essential to plan for the
consequences of already unavoidable climate
changes by strengthening the resilience of
vulnerable water systems.
Gleick concludes on an optimistic note, ar-
guing that “a positive future in the Third Age
of Water is not only possible but inevitable.”
He describes how previous assumptions that
“exponential growth in population, consump-
tion, and production is not only necessary
but good” are giving way to an understanding
that healthy ecosystems and a stable climate
are indeed the “real foundation of a healthy,
resilient society.”
Gleick presents clear actions that might
be taken by governments, political leaders,
businesses, and communities to address wa-
ter sustainability. We need to invest in the
creation and deployment of more-efficient
infrastructure, such as dishwashers, wash-
an explosion in the human population. ing machines, and toilets, he argues. But a
The traditional approach to providing wa- key part of the change “must be social and
ter services, known as the “hard path,” relies institutional.” Rather than focusing exclu-
almost exclusively on physical infrastructure sively on increasing water supply, we must
and centralized water-treatment plants with- also focus on doing more with what we al-
out consideration for the needs of impover- ready have. “High-quality water produced
ished or marginalized communities. Gleick from wastewater,” for example, “is an asset,
demonstrates that this approach not a liability.” Wastewater, Gleick
has “delivered dying ecosystems, explains, which largely flows,
rivers that do not reach the sea, untreated, back into the environ-
and aquifers that are tapped out ment, should instead be collected,
and cannot be restored.” purified, and reused.
Gleick argues that as the sec- Governments, he argues, have a
ond age of water ends, we must responsibility to expand efforts fo-
embrace a sustainable “Third cused on meeting long-term water
Age” to overcome the challenges demands sustainably, rather than
of population growth, environ- The Three Ages short-term priorities. Individuals
of Water:
mental degradation, and cli- Prehistoric Past,
should educate themselves and
mate change. He advocates for Imperiled Present, and use water more efficiently and
a “soft path for water,” focusing a Hope for the Future should advocate for improved lo-
on meeting water-related needs. Peter Gleick cal water management institu-
This approach recognizes the hu- PublicAffairs, 2023. tions. Corporations have a social
368 pp.
man right to water as well as the responsibility to minimize unnec-
true value of water. It protects the health of essary water use and to protect local water
ecosystems, maximizes social and individ- resources and the communities in which they
ual well-being, and expands the available are situated.
sources of water. In the end, The Three Ages of Water pro-
Effective water management will also be vides a hopeful and practical vision for a
PHOTO: RITESH SHUKLA/GETTY IMAGES
key to tackling climate change, Gleick main- more sustainable water future, one that sup-
tains. To mitigate climate risks, we must ports human health, ecosystems, and eco-
reduce greenhouse gas emissions. Energy nomic development. j
and water systems are interconnected, and
RE FE RE N CES AN D N OT ES
improving water-use efficiency and produc-
1. J. Szinai, S. Abraham, H. Cooley, P. Gleick, “The future of
tivity is a simple and cheap way to reduce California’s water-energy-climate nexus” (Next 10 and
energy consumption. He cites a recent study the Pacific Institute, 2021); https://www.next10.org/
publications/water-energy.
conducted by the Pacific Institute, for ex-
ample, which found that low-cost, low-effort 10.1126/science.adg9222
science.org SCIENCE

PUBLIC HEALTH
Thalidomide in America
A journalist corrects the record on US exposure
By Jesse Olszynko-Gryn
T
he story of thalidomide—the notorious
West German drug that caused chil-
dren around the world to be born with
conspicuous limb anomalies in the
years around 1960—is worth revisit-
ing, not least because the survivors are
still among us, now approaching retirement
age. In Britain, Canada, and Germany, they
are an exceptionally well-documented co-
hort. After decades of invisibility, those born
in the United States deserve to be seen, too.
In Wonder Drug, journalist Jennifer Vander-
bes excavates an important corrective to the
mythology that has accreted around the US
thalidomide experience.
US Food and Drug Administration (FDA)
reviewer Frances Kelsey’s stubborn convic-
tion (or indecisiveness, as her critics saw it)
regarding the safety of thalidomide stalled
US approval of the drug, which was widely
marketed in the late 1950s and early 1960s
in other countries as a sedative and as a
treatment for nausea related to pregnancy.
In the meantime, concerning reports started
to come in associating the drug with nerve
damage in adults after long-term use and
with congenital limb differences in the chil-
dren of mothers who ingested the product in
the early months of pregnancy. Yet, despite
Kelsey’s efforts to prevent thalidomide from
coming to market in the US, more pregnant
American women (and fetuses) were ex-
posed to the drug than is generally realized.
Although it is an exaggeration to say
that Kelsey’s life has “never been written
about”—a biography was published 5 years
ago—Vanderbes pushes beyond heroizing
accounts of the lone woman who single-
handedly prevented a global drug disaster
from harming Americans. She provides the
first sustained account of thalidomide’s cir-
culation in the US, where it was distributed
primarily under the name “Kevadon” as part
of legal premarket testing by the Cincinnati-
based William S. Merrell Company and oth-
ers ahead of anticipated FDA approval.
PHOTO: MICHAEL A. MCCOY/REUTERS
A key claim of the book, anchored in ar-
chival research and interviews, is that dozens
of US children were exposed to thalidomide
The reviewer is at the School of Humanities, University
of Strathclyde, Glasgow G1 1XQ, UK, and is the
author of A Woman’s Right to Know: Pregnancy Testing
in Twentieth-Century Britain (MIT Press, 2023).
Email: jesse.olszynko-gryn@strath.ac.uk
SCIENCE science.org
in utero during the premarket trial phase
that saw the widespread distribution of free
unmarked samples. We will never know how
many children were affected, but Vanderbes
persuasively argues that it was considerably
more than the original FDA count of nine.
Some of her other claims are more con-
tentious and require additional evidence,
contextualization, and comparison. For in-
stance, it is difficult to determine whether
the human thalidomide trials conducted
were insufficient or whether more testing on
pregnant animals should have been done be-
fore human trials, as Vanderbes argues, with-
out a more complete picture of the standards
of the time. Her preferred analytic categories
of “greed,” “incompetence,” and “dishonesty”
are better at holding specific people and
companies to account than explaining the
structural factors at play.
To understand how such a devastating
worldwide event could transpire with ago-
nizing slowness, it helps to also consider how
information was shared in the early 1960s.
Decades before the internet or social media,
much effort was required to connect the dots
between local clusters of similarly affected
infants in different parts of the world. The
hard-won consensus that an unprecedented
catastrophe was not only in progress, but
was being propelled by an apparently safe
medication, helped end an era of optimism
in scientific progress and usher in a new,
less-trusting one.
SCIENCE, SEX, AND GENDER
Activists show support for incarcerated Black mothers in 2019.
I N S I G HT S
Wonder Drug:
The Secret History of
Thalidomide in America
and Its Hidden Victims
Jennifer Vanderbes
Random House, 2023. 432 pp.
Historians, who tend to view the past as
contingent, may object to the more deter-
ministic theses of Wonder Drug. A different
account might have placed less emphasis on
Germany’s Nazi past, for example, and more
on the genuine enthusiasm that surrounded
thalidomide—a drug that appeared to be a
safer alternative to barbiturates (which
themselves had been marketed as less
harmful than opioids) during a period
characterized by addiction and overdose.
Vanderbes herself provides much evidence
that many male doctors took the promising
new medicine and gave it to their pregnant
wives in good faith.
“The past is never dead,” wrote William
Faulkner in 1951 in Requiem for a Nun. “It’s
not even past.” As with so many disasters
of the 20th century, the effects of thalido-
mide are not history; survivors continue to
live with them every day, and Vanderbes
is finely attuned to the crucial role played
by journalists since the 1960s in provok-
ing outrage and keeping it alive to support
their campaigns for recognition, compen-
sation, and justice. Wonder Drug is both a
moving account of the uphill battles faced
by survivors and a striking example of
how investigative journalism can help. It
will have accomplished an important and
laudable mission if its publication leads to
greater awareness about US thalidomide
exposure. j
10.1126/science.adi5325
PODCAST
Killing the Black Body:
Race, Reproduction,
and the Meaning of Liberty
Dorothy Roberts
Pantheon, 1997. 373 pp.
More than two decades after
its initial publication, Dorothy
Roberts’s Killing the Black Body
remains an authoritative critique
of reproductive inequality in the
United States. This week on the
Science podcast, Roberts revisits
her scholarship on the fraught
history of Black motherhood
in America and discusses its
modern-day relevance.
https://bit.ly/3J0ZxKk
10.1126/science.adj1265
30 JUNE 2023 • VOL 380 ISSUE 6652 1329

LET TERS
A killer whale breaches at the site of the proposed shipping terminal expansion in Canada’s Fraser River estuary.
Editorial Expression
of Concern
On 30 September 2021, Science published
the Research Article “Light-induced mobile
factors from shoots regulate rhizobium-
triggered soybean root nodulation” by Tao
Wang et al. (1). The editors have been made
aware that data presented in Fig. 5 assessed
GmNSP1 expression rather than GmNIN
expression. We are alerting readers to these
concerns while the authors provide addi-
tional data to correct the manuscript.
H. Holden Thorp
Editor-in-Chief, Science
REF ERENC ES AND NOTES
1. T. Wang et al., Science 374, 65 (2021).
10.1126/science.adj3306
Will Canada permit killer
whale extinction?
The 2022 UN Biodiversity Conference
(COP15) culminated in a landmark agree-
ment to halt biodiversity loss (1), increas-
ing pressure on governments to confront
the conflict between mandates to protect
species and pressure to protect econo-
mies. Canada’s commitment to ecological
1330 30 JUNE 2023 • VOL 380 ISSUE 6652
priorities has been put to the test by the
CAN$3.5 billion Roberts Bank port expan-
sion in British Columbia’s Fraser River
estuary (2). Canada cannot ignore its biodi-
versity commitments to pursue growth at
the expense of ecosystems.   7. Fisheries and Oceans Canada, “Recovery strategy
The Canadian government approved
the port expansion in April, despite its
own conclusion (3) that the effects, which
would include destroying wildlife habitat
protected by the Species at Risk Act, would
be substantial and irreversible. If the proj-
ect proceeds, more than 100 at-risk species
(4) that are already grappling with climate
change and habitat loss will face additional
habitat destruction, as well as pollution and
noise caused by increased shipping traf-
fic. Further destruction of the Fraser River
estuary, one of the most important salmon-
bearing ecosystems on the continent (5, 6),
will inevitably have cascading consequences
for wildlife, food webs, and people.
The project will occur within the legally
protected critical habitat of the endangered
Southern Resident killer whale. This ceta-
cean relies on the port’s busy transboundary
area to feed and raise its offspring, activi-
ties that require individuals to hear and
be heard. Both American and Canadian
recovery plans for the species (7) identify
the need for quieter waters, reduced con-
taminant inputs, and increased access to
the whale’s primary prey: Chinook salmon.
Despite some threat reduction efforts, this
whale is at high risk of extinction (8–11)
under current—let alone worse—conditions.
Although the plans have been condition-
ally approved, construction cannot begin
until a permit is issued that allows the
destruction of critical habitat under the
Species at Risk Act (3). If granted, Canada
would betray commitments made at COP15
and demonstrate to the world that the
country is incapable of finding ways to bal-
ance economic prosperity with biodiversity
protection. Until governments can match
words with actions, the unraveling of spe-
cies and ecosystems will continue.
Misty MacDuffee*, Lance Barrett-Lennard, Auston
Chhor, Allison M. Dennert, Peter S. Ross, David C.
Scott, Valeria Vergara, Kristen Walters
Raincoast Conservation Foundation, Victoria, BC,
Canada.
*Corresponding author. Email: misty@raincoast.org
RE FE RE N CES AN D N OT ES
1. “Kunming-Montreal Global biodiversity framework,”
Conference of the Parties to the Convention on
Biological Diversity (2022); https://www.cbd.int/doc/c/
e6d3/cd1d/daf663719a03902a9b116c34/cop-15-l-
25-en.pdf.
2. D. Penner, “Port of Vancouver’s Roberts Bank Terminal 2:
6 things to know about the contentious $3.5B proposal,”
The Vancouver Sun (2023).
3. “Federal Review Panel report for the Roberts Bank
Terminal 2 project” (Registry Reference 80054, Impact
Assessment Agency of Canada, Ottawa, 2020); https://
iaac-aeic.gc.ca/050/documents/p80054/134506E.pdf.
4. L. J. Kehoe et al., Conserv. Sci. Pract. 3, e310 (2021).
5. C. D. Levings, in Fraser River Delta, British Columbia:
Issues of an Urban Estuary, B. J. Groulx, D. C Mosher, J. L.
Luternauer, D. E. Bilderback, Eds. (Geological Survey of
Canada, issue 567, 2004), pp. 213–236.
6. J. Gordon, M. Arbeider, D. Scott, S. M. Wilson, J. W. Moore,
Estuaries Coasts 38, 2337 (2015).
for the Northern and Southern Resident killer whales
(Orcinus orca) in Canada,” Species at Risk Act Recovery
Strategy Series (2018).
8. L. A. Vélez-Espino et al., Aquat. Conserv. Mar. Freshw.
Ecosyst. 25, 756 (2015).
9. M. Kardos et al., Nat. Ecol. Evol. 7, 675 (2023).
10. C. C. Murray et al., Biol. Conserv. 257, 109124 (2021).
11. R. C. Lacy et al., Sci. Rep. 7, 14119 (2017).
COM P E T IN G IN T E RESTS
D.C.S. served as a paid expert witness for Ecojustice. He pro-
vided an expert opinion on potential impacts of the proposed
project on Chinook salmon as part of the environmental
assessment process.
10.1126/science.adi5984
Border conservation in
Hindu Kush-Himalaya
The Hindu Kush-Himalaya region (HKH)
is one of the most biodiverse places in the
world. The area hosts four global biodi-
PHOTO: GARY SUTTON
versity hotspots (1) as well as 12 terrestrial
and 5 freshwater Global 200 Ecoregions
(2), yet multiple species are at risk of
imminent extinction (3). Both physical
barriers and policy differences across
science.org SCIENCE

international borders hinder conservation
efforts. Protecting the region’s biodiversity
will require multilateral cooperation.
Fences and walls are in place to
restrict illegal immigrants on the India-
Bangladesh and Myanmar-Bangladesh
borders, mitigate conflicts on the India-
Pakistan border, prevent crime on the
India-Myanmar border, and defend
against terrorism on the Pakistan-
Afghanistan border (4). During the pan-
demic, a new line of fences was erected on
the border between China and Myanmar
to impede illegal immigrants and com-
bat the spread of COVID-19 (5). These
physical barriers cannot necessarily stop
humans from crossing the borders (6), but
they increase the risk of wildlife extinc-
tions by increasing mortality; preventing
animals from accessing food, water, and
other resources; and dividing populations
and habitats (7).
International borders in HKH mark
differences in national priorities, socioeco-
nomic conditions, and conservation gover-
nance. However, they rarely align with the
edges of species’ ranges and often bisect
terrestrial ecoregions (2). In addition, the
ranges of terrestrial mammals and birds
in HKH are increasingly likely to shift
across international borders because of
climate change (8).
Maintaining HKH’s rich biodiversity
will require an integrated plan on cross-
boundary conservation and collaboration
among countries, as advocated by interna-
tional treaties such as the Convention on
Biological Diversity and the Convention
on the Conservation of Migratory Species
of Wild Animals (9). To restore pathways
for wildlife species, physical barriers along
international borders should be removed
or replaced with electronic monitor-
ing devices. If physical barriers must be
retained, they may not need to align with
political borders; instead, they should
be relocated to mitigate their impacts on
wildlife populations (10). Political borders
can be buffered by conservation areas
such as peace parks to enhance the con-
nectivity and coverage of biodiversity in
protected areas (11). Nature-based tourism
across international borders can enhance
the engagement of local communities
toward conservation (12). Sharing con-
servation costs among countries through
mechanisms such as payments for eco-
system services can leverage conserva-
tion to alleviate poverty and increase the
PHOTO: QURA TUL AIN
well-being of local communities. National
governments across HKH must involve
all relevant parties, including nongovern-
mental organizations, local stakeholders,
and academic institutions, in the planning
SCIENCE science.org
I N S I G HT S
PAST AS PROLOGUE
A family’s pride in educated daughters
The day I graduated from high school, I sat on the grass near my house contemplating my
future. I lived in Pakistan’s small city of Attock, where low education levels, mobility chal-
lenges, and gender norms limit women’s involvement in the formal workforce. Although
only 10% of the women in my community worked, and few pursued a university degree, I
aspired to increase the opportunities available to other women by continuing my educa-
tion. However, I hadn’t yet discussed my ambitions with my family. Without their support,
I would be forced to follow the traditional path: marriage, children, and domestic duties.
My father, who had spent time abroad, had shown that he valued my education by send-
ing me to tuition academies and ensuring that I had all the necessary materials. My mother,
a traditional housewife, also encouraged me. However, I was tasked with domestic chores,
even during exams, that were not expected of my brothers. If I objected, my brothers would
say that my education had made me rebellious.
My father held the final authority in our household. Under his direction, my brothers
would accompany me to school, ensuring that I could continue my education safely. But
would his support extend to the pursuit of a higher degree?
As I pondered, my father approached me. I shared that I had earned my degree. Thrilled,
he ran through the house celebrating my achievement. Then, to my relief, he announced
that he would take me to the local university to register for classes. But my brothers con-
fronted him. They argued that I had received ample education to fulfill my role as a wife.
They feared that further education would tarnish my obedience by enabling me to work
and challenge injustices.
My father refused to heed their warnings. Instead, he drove me to the college. I was wear-
ing a traditional dress, covered from head to toe, with even my face concealed, so my father
could not see my expression of joy and gratitude as he escorted me to the admission office.
I went on to graduate, move to China, earn a PhD in applied mathematics, and secure
a postdoc position. My achievements have brought pride to my parents and inspired my
younger sister. Even my brothers have grown accustomed to my path; they offered much
less resistance when my sister decided to pursue higher education. With every educated
woman, a new chapter begins—one that empowers future generations and shapes a more
inclusive, prosperous, and equitable society for all.
Qura Tul Ain
School of Mathematics, Guizhou University, Guiyang, China. Email: ainquratul@gzu.edu.cn
10.1126/science.adi2884
Call for Submissions Past as Prologue is an occasional feature highlighting the role of family history in the life of sci-
entists. What role did your family background play in your decision to pursue science, your field, or your career? Submit
your story to www.submit2science.org.
The author poses with her
parents on graduation day.
30 JUNE 2023 • VOL 380 ISSUE 6652 1331
 INSIGHTS | LET TERS

and practice of cross-boundary collabora-

tive conservation.
Xiaodong Chen1,2* and Vanessa Hull3
1
State Key Laboratory of Urban and Regional
Ecology, Research Center for Eco-Environmental
Sciences, Chinese Academy of Sciences, Beijing
100085, China. 2College of Resources and
Environment, University of Chinese Academy of
Sciences, Beijing 100049, China. 3Department of
Wildlife Ecology and Conservation, University of
Florida, Gainesville, FL 32611, USA.
*Corresponding author. Email: xdchen@rcees.ac.cn

RE FE RE N CES AN D N OT ES
1. N. Myers, R. A. Mittermeier, C. G. Mittermeier, G. A. da
Fonseca, J. Kent, Nature 403, 853 (2000).
2. D. M. Olson, E. Dinerstein, Ann. Missouri Bot. Garden 89,

Share
199 (2002).
3. T. H. Ricketts et al., Proc. Natl. Acad. Sci. U.S.A. 102,
18497 (2005).
4. J. D. C. Linnell et al., PLOS Biol. 14, e1002483 (2016).
5. “Iron wire on China-Myanmar border built during virus
spike in Sep 2020, prevents illegal border crossing amid

Your
COVID-19 resurgence,” Global Times (2021).
6. E. Flores, Harvard Int. Rev. 38, 10 (2017).
7. R. Peters et al., Bioscience 68, 740 (2018).
8. M. A. Titley, S. H. M. Butchart, V. R. Jones, M. J.
Whittingham, S. G. Willis, Proc. Natl. Acad. Sci. U.S.A. 118,
e2011204118 (2021).

Robotics 9. Convention on the Conservation of Migratory Species of

Wild Animals, “Ecological connectivity in the post-2020
global biodiversity framework” (2019).
10. C. R. Margules, R. L. Pressey, Nature 405, 243 (2000).
11. P. Visconti et al., Science 364, 239 (2019).
12. A. Balmford et al., PLOS Biol. 7, e1000144 (2009).

Research 10.1126/science.adi7958

with the
TECHNICAL COMMENT ABSTRACTS
Comment on “The lower Cambrian
lobopodian Cardiodictyon resolves the origin
of euarthropod brains”
Graham E. Budd, Georg Mayer, Ralf Janssen,

World. B. Joakim Eriksson

Shaping the future of robotics with high
impact research!
As a multidisciplinary online-only journal, Science Robotics publishes
original, peer-reviewed, research articles that advance the field of
robotics. The journal provides a central forum for communication of
new ideas, general principles, and original developments in research
and applications of robotics for all environments.
Submit your research today.
Learn more at: science.org/journal/scirobotics
Twitter: @SciRobotics
Facebook: @ScienceRobotics
Strausfeld et al. (Report, 24 Nov 2022,
p. 905) claim that Cambrian fossilized
nervous tissue supports the interpretation
that the ancestral panarthropod brain was
tripartite and unsegmented. We argue that
this conclusion is unsupported, and devel-
opmental data from living onychophorans
contradict it.
Full text: dx.doi.org/10.1126/science.adg1412
Response to Comment on “The lower Cambrian
lobopodian Cardiodictyon resolves the origin of
euarthropod brains”
Nicholas J. Strausfeld, Xianguang Hou,
Marcel E. Sayre, Frank Hirth
Budd et al. challenge the identity of neural
traces reported for the Cambrian lobo-
podian Cardiodictyon catenulum. Their
argumentation is unsupported, as are objec-
tions with reference to living Onychophora
that misinterpret established genomic,
genetic, developmental, and neuroanatomi-
cal evidence. Instead, phylogenetic data
corroborate the finding that the ancestral
panarthropod head and brain is unseg-
mented, as in C. catenulum.
Full text: dx.doi.org/10.1126/science.adg6051

science.org SCIENCE
RES EARCH

◥ erals in †C. catenulum, which they suggest were

T E C H N I C A L CO M M E N T derived from iron-yielding nervous tissue. How-
ever, muscle (for example) also contains iron
ARTHROPOD EVOLUTION (2), and we argue the evidence they present

Comment on “The lower Cambrian

does not support their conclusion. The faint
dorso-ventral lineation seen in the trunk could
just as well be related to the exterior annula-
lobopodian Cardiodictyon resolves the origin tion of the animal. In addition, the diffuse and
irregular anterior coloration they interpret as
of euarthropod brains” a tripartite nonsegmental brain shows no ob-
vious threefold division (e.g., their fig. S5F),
Graham E. Budd1*, Georg Mayer2, Ralf Janssen1, B. Joakim Eriksson3 and in part clearly follows the outline of the
gut (their fig. S5, A and B).
Strausfeld et al. (Report, 24 Nov 2022, p. 905) claim that Cambrian fossilized nervous tissue supports In the extant lobopodians (tardigrades and
the interpretation that the ancestral panarthropod brain was tripartite and unsegmented. We argue that onychophorans), the developing nervous sys-
this conclusion is unsupported, and developmental data from living onychophorans contradict it. tem and somites can be studied in detail, with
tissue-specific stains (3). In onychophorans,

T
he debate over the composition of the
panarthropod head has been ongoing
for well over a century without abating
[we use panarthropods to refer to the clade
consisting of onychophorans, tardigrades,
and euarthropods (mandibulates and cheli-
cerates)]. The proposal that the panarthropod
brain includes a pre-oral and non-segmental
“acron”, historically based on a supposed affi-
nity with annelids, has proved to be particularly
durable. Strausfeld et al. (1) argue that the
Cambrian lobopodian †Cardiodictyon catenulum
reveals the ancestral panarthropod brain to
consist of three non-segmental units: the proso-,
proto- and deutocerebrum (their ce1, ce2, and
ce3). In addition, they argue these three com-
ponents originated separately from the euar-
thropod ventral ganglionated nerve cord. They
thus revive the ideas of entomologists such as
Snodgrass (and earlier workers), who argued
that essentially everything in front of the in-
sect mouth was non-segmental in origin.
Strausfeld et al. base their conclusions on
the distribution of rust/magenta-colored min-
the origin and subsequent fate of the coelomic
cavities is clear: the somites associated with
the antenna and jaws form part of a series
with the others that originate posteriorly and
move anterior (Fig. 1, A to F), so the first (an-
tennal) somite comes to lie in front of the mouth
1
Department of Earth Sciences, Palaeobiology, Uppsala
University, 752 36, Uppsala, Sweden. 2Department of
Zoology, Institute of Biology, University of Kassel, 34132,
Kassel, Germany. 3Department für Neurowissenschaften und
Entwicklungsbiologie, Universität Wien, A-1030, Vienna,
Austria.
*Corresponding author. Email: graham.budd@pal.uu.se

Fig. 1. Segment formation during early onychophoran development. (A to F) DAPI staining of embryos of Euperipatoides kanangrensis at consecutive
developmental stages in ventral view. Anterior is to the left. Magenta arrowheads indicate anterior border of embryonic slit; yellow arrowheads point to anterior
borders of germ band (B) and first/antennal pair of somites (C) to (F). Note that like in subsequent trunk segments, the head somites (as, js, ss) appear in the
posterior segment addition zone (sz) and migrate anteriorly (dotted arrows). Abbreviations: as, somite of antennal segment; js, somite of jaw segment; ss, somite of
slime papilla segment; sz, segment addition zone. Scale bars: 200 μm [for (A) to (F)]. New images of sequence illustrated in figure 1 of (5).

Budd et al., Science 380, eadg1412 (2023) 30 June 2023 1 of 3

RES EARCH | T E C H N I C A L C OM M E N T

the classical argument that “segments” should

contain pairs of ganglia, because the evidence
from at least onychophorans is compatible
with the groundplan of the panarthropod (and
indeed ecdysozoan) central nervous system (CNS)
being largely unganglionated. The nervous
system of onychophorans nevertheless has a
distinct segmental appearance because of the
segmental spacing of nerves leading into the legs
and head appendages (3), which themselves
correspond to the embryonic somites: and the
most elaborated of these segmental features is
naturally the most anterior, which includes the
optic and major brain neuropils. It is tempting
to speculate that these aggregations were in-
deed the evolutionary origins of the euarthropod
segmental ganglia (3, 9) but if so, appendage-
associated head ganglia would have evolved in
the same way as those in the trunk.
Every anterior part of the onychophoran body
is thus segmental and underlain by a pair of
mesodermal somites with coeloms; and the
CNS, although probably at base non-segmental,
nevertheless has sets of neurons and swellings
along it that correspond to the segmental ap-
pendages including the antenna and jaws. In
addition, there is no part of the anterior ner-
vous system or other tissue that develops ante-
Fig. 2. Segmental features of the onychophoran head. Bisbenzimide staining (A and C), mRNA
rior to the anterior-most somite (Fig. 2A, C).
localization of segment-polarity gene engrailed (C), and anti-acetylated α-tubulin immunolabeling (D).
Finally, it can be noted that the segment-polarity
Confocal (A), (C), (D) and light (B) micrographs. Anterior is to the left in (A) and (B) and up in (C) and (D).
gene network established in the ectoderm re-
(A) Optical sagittal section of embryo of Principapillatus hitoyensis, illustrating the space-filling mesodermal
veals that this, too, can be considered segmen-
somites with their coeloms. Note that there is no additional tissue anterior to the first/antennal somite. The
tal (4) (Fig. 2B).
first somite corresponds to the prosocerebrum and protocerebrum of (2), and the second to their
Fossil evidence from †C. catenulum must be
deutocerebrum, all three of which are described as asegmental. (B) Segmentally repeated expression of
considered in this light, together with infer-
engrailed in the ectoderm of a segmenting embryo of Euperipatoides rowelli. Note that all three cephalic
ences that can be drawn from other fossils.
segments are demarcated by posterior engrailed stripes (arrowheads). (C) Anterior region of stage IV embryo
Cambrian lobopodians show a variety of an-
of E. rowelli in ventral view. Note that there is no additional tissue anterior to antennal segment, although
terior adaptations with differing numbers of
neuroectoderm (ne) corresponding to each segment is already present at this stage [cf. (4, 5)]. (D) Confocal
specialized appendages (10). However, the
micrograph illustrating the transitory nephridium (arrowhead) at the base of the antenna in the embryo of
widespread occurrence of a single differen-
Epiperipatus biolleyi, suggesting serial homology with the walking limbs (9). Abbreviations: an, antennal
tiated frontal appendage (and implied single
neuropil; as, antennal somite/segment; at, antenna; cn, developing central neuropil; co, coelomic remnant of
brain compartment), seen also in the more
antennal segment; ec, ectoderm; ey, developing eye; jn, jaw nerve; js, jaw somite/segment; jw, developing jaw;
euarthropod-ward taxa such as †Kerygmachela
le, developing first leg; ne, neuroectoderm of developing proto- and deutocerebrum, corresponding to ce1 to
kierkegaardi and radiodontids, suggests this is
ce3 of (2); sp, developing slime papilla; ss, slime papilla somite/segment; st, stomodaeum; sz, segment
the ancestral character state (10, 11). The char-
addition zone; ts, anteriormost somite/segment of trunk. Scale bars: 200 μm (A); 500 μm (B); 100 μm (C);
acter state of †C. catenulum is thus likely to
300 μm (D). (A) and (B) reproduced from figures 1L and 3A of (7); (C) from figure 3B of (13); and (D) from
be derived and not informative about the
figure 3D of (14), all reproduced under a Creative Commons license.
panarthropod groundplan state. Strausfeld et al.
argue that †K. kierkegaardi and lobopodians
(4). We therefore believe that there can be no not an “assumption” as per Strausfeld et al., but such as †Aysheaia pedunculata have lost their
doubt that there are mesodermal segmental a well-established empirical fact, as noted long ce2 and ce3 domains but they provide no evi-
components of the very anterior of the ony- ago by Manton (8). Manton’s arguments remain dence for this, and in addition, this character
chophoran head (despite being anterior to as valid today as then and additional support distribution in their own phylogeny would im-
Hox gene expression) (1). In addition to the comes from recognizing that panarthropods ply these domains would have to then re-evolve
coelom, each of these somites (including the are monophyletic (so that onychophorans are in the more crownward stem-euarthropods.
most anterior) is associated with a posterior directly relevant to the study of euarthropods) More simply, the anterior structures in ony-
engrailed stripe [(5, 6) contrary to figure 3 of and from molecular developmental data. chophorans and Cambrian lobopodians in-
Strausfeld et al.], an appendage (7), and an at Whether or not the ancestral panarthropod cluding †C. catenulum are likely to be modified
least transitory nephridium (7) (Fig. 2, A to D). ventral nervous system comprised a chain of segmental walking appendages, as they demon-
Developmentally then, all these components, ganglia, as in tardigrades and euarthropods, or strably are in onychophorans and it is unclear
including appendages, can be seen to be ex- a medullary pair of unsegmented nerve cords, why the evidence from †C. catenulum should
tremely similar to each other and to be clearly as in onychophorans, is unclear, although re- dissuade one from this view.
segmental. We argue the segmental composi- cent study has failed to find ganglionic struc- The onychophoran and euarthropod heads
tion of the panarthropod head at least, is thus ture in onychophorans (3). This complicates may well incorporate ancient features of the

Budd et al., Science 380, eadg1412 (2023) 30 June 2023 2 of 3

RES EARCH | T E C H N I C A L C OM M E N T
brain that were inherited from their deep bi-
laterian ancestors (12). However, if so, they are
all contained within the anterior appendage-
bearing segment in at least onychophorans,
and do not correspond to the overt anterior-
posterior divisions of the CNS, which are clearly
segmental in origin.
RE FE RENCES AND N OT ES
1. N. J. Strausfeld, X. Hou, M. E. Sayre, F. Hirth, Science 378,
905–909 (2022).
Budd et al., Science 380, eadg1412 (2023) 30 June 2023
2. F. Saleh, A. C. Daley, B. Lefebvre, B. Pittet, J. P. Perrillat,
BioEssays 42, e1900243 (2020).
3. G. Mayer, P. M. Whitington, Dev. Biol. 335, 263–275
(2009).
4. R. Janssen, G. E. Budd, Dev. Biol. 382, 224–234 (2013).
5. B. J. Eriksson, N. N. Tait, G. E. Budd, J. Morphol. 255, 1–23
(2003).
6. B. J. Eriksson, N. N. Tait, G. E. Budd, M. Akam, Dev. Genes Evol.
219, 249–264 (2009).
7. F. A. Franke, G. Mayer, PLOS ONE 9, e114383
(2014).
8. G. Mayer, M. Koch, Arthropod Struct. Dev. 34, 471–480
(2005).
9. S. M. Manton, Philos. Trans. R. Soc. B 233, 483–580
(1949).
10. G. E. Budd, Evol. Dev. 3, 332–342 (2001).
11. G. E. Budd, Arthropod Struct. Dev. 62, 101048 (2021).
12. M. R. Smith, J. Ortega-Hernández, Nature 514, 363–366
(2014).
13. Q. Ou, D. Shu, G. Mayer, Nat. Commun. 3, 1261 (2012).
14. G. Mayer, P. M. Whitington, P. Sunnucks, H.-J. Pflüger,
BMC Evol. Biol. 10, 255 (2010).
15. N. J. Strausfeld, F. Hirth, Science 340 (2013), 157–161.
Submitted 7 December 2022; accepted 26 April 2023
10.1126/science.adg1412
3 of 3
RES EARCH

◥ Budd et al. (2) dispute that neural collaterals

TECHNICAL RESPONSE extend from the C. catenulum ventral nerve
cord [figure 2, A to D, in (1)], claiming that
ARTHROPOD EVOLUTION they might instead represent body wall an-

Response to Comment on “The lower Cambrian

nulation or, failing that, then musculature.
The decay of neural tissue faster than muscle
appears to favor iron deposition and hence
lobopodian Cardiodictyon resolves the origin carbon (3), the latter demonstrated by chro-
matic filtering and energy dispersive x-ray
of euarthropod brains” spectroscopy (Fig. 1A). Micro-annulation is
absent for the C. catenulum trunk but is re-
Nicholas J. Strausfeld1*, Xianguang Hou2, Marcel E. Sayre3, Frank Hirth4* solved for trunk limbs (Fig. 1C). Sparse mus-
cles conjectured by Budd (4) for stilt-legged
Budd et al. challenge the identity of neural traces reported for the Cambrian lobopodian Cardiodictyon lobopodians would bear no correspondence
catenulum. Their argumentation is unsupported, as are objections with reference to living Onychophora to branched collaterals of the ventral nerve
that misinterpret established genomic, genetic, developmental, and neuroanatomical evidence. cords revealed by chromatic filtering [figure
Instead, phylogenetic data corroborate the finding that the ancestral panarthropod head and brain is S4 in (1)]. If in C. catenulum there existed
unsegmented, as in C. catenulum.

C
ontemporary genetics and genomics
can inform paleontological observations
for comparing traits across taxa. Using
this combined approach to compare ce-
rebral arrangements across panarthrop-
ods, including the lower Cambrian lobopodian
Cardiodictyon catenulum, resolves a common
ground pattern of organization (1). Without
new evidence, Budd et al. (2) resort to two
lines of argumentation to dispute this find-
ing. The first challenges our characteriza-
tion of the Cardiodictyon nerve cord and
brain. The second argues for a segmented
head and brain of Onychophora. Neither
argument counters evidence reported by
Strausfeld et al. (1).
1
Department of Neuroscience, University of Arizona, Tucson,
AZ 85721, USA. 2Yunnan Key Laboratory for Palaeobiology,
Institute of Palaeontology, Yunnan University, Kunming,
China. 3Lund University, Lund Vision Group, Department of
Biology, Lund, Sweden & Macquarie University, Department
of Biological Sciences, Sydney, Australia. 4Department
of Basic and Clinical Neuroscience, Institute of Psychiatry,
Psychology and Neuroscience, King’s College London,
London SE5 9RT, UK.
*Corresponding authors. Email: flybrain@arizona.edu (N.J.S.);
frank.hirth@kcl.ac.uk (F.H.)

Fig. 1. Chromatic filtering and energy dispersive

x-ray spectroscopy reveal fossil neural traces.
(A) Chromatic filtering reveals the retina, optic tract,
and neuropils of the radiodontan Lyrarapax unguispinus
(Cong et al. Nature 513:538–542). Upper panel: raw
image taken with white light; middle panel: suppression
of spectra other than blue showing additional evidence
of structures supporting energy dispersive x-ray
spectroscopy of carbon deposits (lower panel).
Scale bar = 1mm. (B-G) Trunk limbs of Cardiodictyon
catenulum are distinct from its cephalic appen-
dages. (B) Paired trunk limbs extend laterally from loci
immediately adjacent to the trunk’s midline (arrows).
(C) Trunk rotated counterclockwise by about 15°
around the anterior-posterior axis shows empty limb
socket (upper box) paired with socket (lower box)
of the intact contralateral limb. Whereas the trunk
lacks microannulation, the enlargement (inset) shows
microannulations (three indicated by dots) typical of
trunk limbs, which are the only appendage with a
terminal claw (inset, arrow). (D) Absent the epidermis,
the trunk reveals limbs and their articulation points
(spherical density, boxed) from which rod-like
elements (double arrows) extend to the limb’s claw.
(E) Longitudinally split fossil reveals the ce3 cephalic
appendage seemingly unattached (white arrow, boxed
area high contrast in E’), contrasting with a trunk
limb attached to its point of articulation (open arrow).
(F) Oblique top-down view of the head showing one
seemingly unattached limb extending from under the
trunk (white arrow), whereas three cephalic appen-
dages originate from points of articulation from
beneath the head’s margin (open arrows, box refers to
high contrast image in F’). (G). Summary diagram
comparing dispositions of segmental trunk limbs and
three unique pairs of appendages from the asegmental head. Scale bar for inset to C = 50μm. B, D = 200μm; E, F = 250μm.

Strausfeld et al., Science 380, eadg6051 (2023) 30 June 2023 1 of 3

RES EARCH | T E C H N I C A L R E S P ON S E
body wall muscles within segments and lon-
gitudinal muscles extending between seg-
mental sclerites, neural processes extending
ventrodorsally would be consistent with both
organizations.
Budd et al. misdirect in asserting we identify
a tripartite brain in C. catenulum. We empha-
size a continuum of trace deposits, interpreted
as neuropil resolved in an unsegmented head,
in which three domains align with three unique
(Fig. 1, E to G) appendage pairs, distinct from
trunk appendages (Fig. 1, B to D), and seriate
features of the foregut—the suggestion being
that these domains further evolved as neuro-
meres in crownward panarthropods (1). Budd
et al.’s objection to deposits partially outlin-
ing the foregut ignores the brains of extant
euarthropods, notably chelicerates, which are
perforated by the foregut.
Placing Cardiodictyon basal within an ab-
breviated ecdysozoan phylogeny but outside
the panarthropod crown group is justified [see
supplemental material in (1)], as is attributing
in Kerygmachela apical traces to a prosocere-
brum (ce1) distant from the endomesodermal
interface. Budd et al. object that loss of ce2 and
ce3 requires their re-evolution in more crown-
ward stem euarthropods. This is demonstrably
wrong since loss of ce2 and ce3 applies only to
the branch providing Kerygmachela as shown
in figure S8 of (1). The same applies to trunk
ganglia in Onychophora: evolved loss of gan-
glia but longitudinal expansion of synapse-rich
nerve cords is a trait mapped by us exclusively
onto the onychophoran trajectory (1).
Fig. 2. Corresponding genetics A Onychophora
and brain center organization
in Onychophora and Euarthro-
Six3
poda. (A) Alignment of onychoph-
Pax6
Hox (1-9)
oran and mandibulate with
reference to their endomesodermal
interface (emi) demonstrates
corresponding expression patterns
of homologous genes, the combi-
nation of which define the brain as
three cerebral domains. Note that
homeotic (Hox) gene expression
defining mesoderm-related segmen-
tation of the trunk does not extend
anterior to the emi (1). (B) Each
cerebral domain ce1-ce3 is further
characterized across panarthropods B
by corresponding arrangements
of circuits, the development of which
depend on specific combinatorial
expression patterns of homeobox
transcription factors (listed to the
right of the relevant domain).
For Euarthropoda, these domains
are referred to as the proso-, proto-,
and deutocerebrum (1). T1 indicates
the first segment of the trunk.
Onychophora
Strausfeld et al., Science 380, eadg6051 (2023)
Budd et al.’s second line of argument focuses
on living onychophorans, evoking supposed
ancestral traits to challenge interpretations
of C. catenulum. Onychophora is, however,
an extant taxon allowing alignment of its brain,
neuropil centers, and unique cephalic ap-
pendages with those of other panarthropods
(Fig. 2) (1). Budd et al. refers to what are de-
scribed as transitory coelomic cavities of “neg-
ligible volume” (5) and a possible nephridium
(6) to claim head mesoderm is segmental thus
indistinguishable from trunk segmentation
despite the absence of mesoderm-specific
homeotic (Hox) gene expression. In contrast,
neither early embryonic development nor ex-
pression patterns of developmental control
genes (Fig. 2A) support Budd et al.’s claim.
Budd et al. lean heavily on Manton’s 1949 sche-
matics of onychophoran development. These
included recognition of rostral pre-stomodeal
“optic rudiments” in Crustacea appearing
before the specification of trunk segments
[figure 6 in (7)]. A corresponding feature in
Onychophora would not have suited Manton’s
polyphyletic Arthropoda. Yet her sketches of
the developing onychophoran Peripatopsis
balfouri [figure 7, K to M, in (7)] suggest tissue
rostral to the stomodeum prior to the first ap-
pearance of segmentation. Manton reported
that mesoderm migrates rostrally from a post-
oral position thereby populating the head [see
plate 38, figs. 79 to 84, in (7). However, there is
no genetic evidence for mesoderm-related coe-
lomic cavities in the onychophoran head, con-
sistent with the absence of Hox gene expression
Mandibulata
vnd/Nk2.1
metanephirida
FoxQ2
NK2.2
FoxQ2
hbn
nephrocytes
exd
hbn
exd
Hox (1-9)
Twist
Twist
T1
T1
T1
Chelicerata Mandibulata
30 June 2023
(Fig. 2A). Manton (7) also described the post-
oral progression of growth caudally, accordant
with an onychophoran-specific role of Notch-
Delta in posterior growth but not in segmen-
tation (8). The gene Twist is an evolutionarily
conserved key regulator for mesoderm spec-
ification (9), the expression of which is re-
stricted to the trunk in Onychophora (10), as
it is in euarthropods (Fig. 2A) and in other
phyla (9). The genetic network for mesoderm-
related nephridia formation applies to the
onychophoran trunk (11), but not to its head
(Fig. 2A). Together these data establish that
head mesoderm in Onychophora is no indi-
cator for segmentation unless one would pos-
tulate non-Twist-non-Hox mesoderm, for which
there is no evidence.
Correspondingly, there is no genetic evidence
for brain segmentation in Onychophora. Early
transient, anterior engrailed expression is
consistent with its ancestral role in neural
fate specification, but not segmentation (12).
Likewise, the reported ‘segment-polarity’ gene
expression patterns suggest a role in organo-
genesis but not segmentation (13). This also
applies to homologs of the conserved pattern-
ing genes extradenticle (exd) and ventral ner-
vous system defective (vnd)/NK2.1/NK2.2 (14, 15),
with corresponding expression across euarthro-
pods (16–18) (Fig. 2A). In Onychophora, neuro-
genesis is characterized by “massive irregular
segregation of neural precursors” (19), in con-
trast to the regular, neuromere-related segrega-
tion typical for euarthropod trunk neurogenesis.
Expression of the patterning genes FoxQ2 and
Six3
Pax6
ce1 prosocerebral
ce2 protocerebral asegmental brain
ce3 deutocerebral
endomesodermal interface (emi)
segmented trunk nervous system
ocelli/principal eyes
ce1 prosocerebral
FoxQ2-Six3-hbn labral/neurosecretory
central body complex
ce2 protocerebral secondary/compound
Six3-Otx-Pax6 eyes, retino-neuropils
optic glomeruli
mushroom bodies
T1
ce3 deutocerebral chemoreceptive receptors
Emx-NK2-exd chemoreceptive centers.
segmental trunk sensory-motor
Hox genes coordination
2 of 3
RES EARCH | T E C H N I C A L R E S P ON S E
homeobrain (hbn) unambiguously identifies
the ce1 domain across total Panarthropoda
(18, 20, 21), consistent with Six3, Pax6 and
non-Hox expression territories that together
demarcate ce2 and ce3, which are genetically
distinct from the trunk (Fig. 2A).
Developmental genetics is consistent with
and regulates the formation and functions of
domain-specific brain centers that correspond
across panarthropods, including Onychophora
(Fig. 2B): ce1-specific Six3, FoxQ2 and hbn de-
fine prosocerebral rostral visual pathways and
the central body; ce2-specific Otx and Pax6
define protocerebral optic and memory pro-
cessing neuropils; ce3-specific Emx, Nk2 and
Exd determine deutocerebral chemosenso-
ry integration centers (1, 18, 21). Taken to-
gether, we conclude that arguments offered by
Strausfeld et al., Science 380, eadg6051 (2023)
Budd et al. comprehensively fail to refute any of 14. R. Janssen, B. J. Eriksson, G. E. Budd, M. Akam, N. M. Prpic,
the findings reported in Strausfeld et al. (1). Evol. Dev. 12, 363–372 (2010).
15. S. Treffkorn, L. Kahnke, L. Hering, G. Mayer, Evodevo 9, 17 (2018).
16. T. Nagao, K. Endo, H. Kawauchi, U. Walldorf,
RE FERENCES AND NOTES
K. Furukubo-Tokunaga, Dev. Genes Evol. 210, 289–299 (2000).
1. N. J. Strausfeld, X. Hou, M. E. Sayre, F. Hirth, Science 378,
17. S. R. Wheeler, M. L. Carrico, B. A. Wilson, J. B. Skeath, Dev. Biol.
905–909 (2022).
279, 491–500 (2005).
2. G. E. Budd, G. Mayer, R. Janssen, B. J. Eriksson. Comment on
18. V. S. Hunnekuhl, M. Akam, Dev. Biol. 396, 136–149 (2014).
“The lower Cambrian lobopodian Cardiodictyon resolves the
19. B. J. Eriksson, A. Stollewerk, Proc. Natl. Acad. Sci. U.S.A. 107,
origin of eurarthropod brains” (Science, 2023); 10.1126/
22576–22581 (2010).
science.adg1412
20. R. Janssen, Evodevo 8, 1 (2017).
3. F. Saleh, A. C. Daley, B. Lefebvre, B. Pittet, J. P. Perrillat,
21. P. Kitzmann, M. Weißkopf, M. I. Schacht, G. Bucher, Development
BioEssays 42, e1900243 (2020).
144, 2969–2981 (2017).
4. G. E. Budd, Evol. Dev. 3, 332–342 (2001).
5. B. J. Eriksson, N. N. Tait, G. E. Budd, J. Morphol. 255, 1–23 (2003).
AC KNOWLED GME NTS
6. G. Mayer, M. Koch, Arthropod Struct. Dev. 34, 471–480 (2005).
We are indebted to C. Strausfeld for refining the text and editing this work.
7. S. M. Manton, Philos. Trans. R. Soc. Lond. B Biol. Sci. 233,
Funding: This work was supported by the National Science Foundation
483–580 (1949).
8. R. Janssen, G. E. Budd, Dev. Genes Evol. 226, 69–77 (2016). under grant 1754798 awarded to N.J.S., and by the University of
9. I. Castanon, M. K. Baylies, Gene 287, 11–22 (2002). Arizona Regents Fund. F.H. acknowledges support from the UK
10. R. Janssen, Arthropod Struct. Dev. 46, 341–353 (2017). Biotechnology and Biological Sciences Research Council (BB/N001230/1).
11. L. Gąsiorowski et al., Curr. Biol. 31, 3629–3638.e2 (2021).
12. J. M. Gibert, Dev. Genes Evol. 212, 307–318 (2002). Submitted 9 February 2023; accepted 26 April 2023
13. F. A. Franke, G. Mayer, PLOS ONE 9, e114383 (2014). 10.1126/science.adg6051
30 June 2023 3 of 3
 RESEARCH
IN S CIENCE JOURNAL S
Edited by Michael Funk

HYDROLOGY

Farming and floods

A
griculture affects surface- and groundwater properties,
including widespread groundwater depletion in areas reli-
ant on crop irrigation. Houspanossian et al. documented
a large-scale occurrence of a very different phenomenon
across the South American plains: increasing flooding
associated with ongoing expansion of agriculture. Combining
remotely sensed data and groundwater-monitoring stations, the
authors showed that over about the past 20 years, agricultural
areas were more likely to start experiencing floods as ground-
water transitioned from deep to shallow states more responsive
to large precipitation events. Data compiled from several field
studies suggest that loss of deep-rooted native vegetation may
play a role in this hydrological transition, which can affect entire
regions. —BEL Science, add5462, this issue p. 1344

Changes in land use and vegetation have led to increased flooding in the South American plains, such as this site near Firmat, Santa Fe, Argentina.

CHROMATIN BIOPHYSICS
Enhancer-promoter
encounters
A crucial step in gene regulation
is the physical encounter of dis-
persed enhancer-promoter pairs
across the genome. However,
how distal DNA elements find
each other in the nuclear space
PHOTOS (TOP TO BOTTOM): © PATRICIO MURPHY VIA ZUMA WIRE; DICKSON ET AL.
MARTIAN GEOLOGY
Conditions for forming
Mars’ gullies
Some steep slopes on Mars
have gullies with morpholo-
gies suggesting that they were
formed by a fluid. However, the
planet’s current climate is not
conducive to the melting of
water ice at those locations,
and mechanisms involving car-
bon dioxide ice do not explain
the distribution of the gullies.
Dickson et al. simulated how
the climate of Mars differed
when its axis tilted by different
amounts over the past few
million years. At a tilt of 35
degrees, the ice caps partially
melted, raising the atmo-
spheric pressure, and there
were higher summer tempera-
tures. Under these conditions,
the atmospheric pressure at
the gullies would be above the
triple point of water, so it could

remains unclear. Brückner et al.

visualized the three-dimensional
motion of pairs of DNA loci of
varying separations along the
chromosome and their transcrip-
tional output in developing fly
embryos. They found an unex-
pected combination of dense
packing and rapid diffusion,
leading to encounter times with
a weak dependence on genomic
separation. These results imply
that transcriptional contacts are
possible across large genomic
distances, with crucial implica-
tions for gene regulation. —DJ
Science, adf5568, this issue p. 1357 Gullies on Mars may have been formed by liquid water that formed when Mars was more strongly tilted on its axis.

SCIENCE science.org 30 JUNE 2023 • VOL 380 ISSUE 6652 1333

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

melt to form a liquid. —KTS three-dimensional folds

Science, abk2464, this issue p. 1363
QUANTUM SENSING
Quantum enhanced
metrology
Taking advantage of the quan-
tum mechanical properties of
systems is expected to provide
advantages over classical-based
techniques. Li et al. showed that
the technique of quantum scram-
bling, in which local information
is rapidly distributed across
different degrees of freedom of
a quantum system, can provide
an advantage for sensing and
precision of metrological mea-
surements. These experiments
demonstrated the relationship
between quantum metrology and
quantum information scram-
bling, and the approach could
be applied to determine the
properties and dynamics of other
complex many-body quantum
systems. —ISO
Science, adg9500, this issue p. 1381
CHIRAL MATERIALS
upon engaging their tar-
gets. Using state-of-the-art,
solution-state nuclear
magnetic resonance tech-
niques, Madhurima et al.
have revealed the presence
of a sparsely populated state
on the amino terminus of
the transcriptional regulator
CytR. Surprisingly, it is this
conformationally excited
state rather than the natively
disordered ground state that
appears to be compatible
with DNA binding. Whether
similar recognition events
driven through the selection
of prefolded binding-compat-
ible states indeed represent
a general phenomenon in
disordered systems that was
missed by previous analyses
remains to be seen. —RG
Sci. Adv. (2023)
10.1126/sciadv.adh4591
Edited by Caroline Ash
IN OTHER JOURNALS and Jesse Smith
GENE THERAPY
Through the cochlear TROPHIC REWILDING
aqueduct
Although gene therapy can Release the tortoises!

Magnetically directing
chiral assembly
The assembly of particles into
chiral superstructures is usually
directed with chiral templates
such as DNA, which bind the par-
ticles, or by using lithographic
methods. Li et al. showed that
applying the chiral quadrupole
field generated by permanent
magnets to dispersed magnetic
nanoparticles creates long-
range chiral superstructures
that adopt the field chirality. The
tunability of the field enables
control over the superstructure
handedness, chirality, and pitch.
Hybrid particles composed of
gold nanorods and magnetite
displayed tunable circular
dichroism and optical rotary
dispersion signals. —PDS
treat genetic hearing loss
in neonatal mice, treating
adult animals is more difficult
because of the location of
the cochleae and the risk of
damage to inner ear struc-
tures. Mathiesen et al. showed
that the cochlear aqueduct
connects the cerebrospinal
fluid (CSF) and cochlear
fluid in adult mice. Tracers
injected into the CSF reached
the mouse inner ear, and
intracisternal injection of an
adeno-associated virus car-
rying the solute carrier family
17 member 8 gene (Slc17A8)
restored vesicular gluta-
mate transporter-3 protein
in the inner hair cells of the
cochlea and rescued hearing
in mice that lacked Slc17A8.
These findings suggest that
R
eestablishing large herbivores into their former habitats
may help to restore trophic relationships and ecosystem
functions. Galapagos giant tortoises were historically not
only key herbivores, but also trampled vegetation and
dispersed seeds. Tapia Aguilera and Gibbs documented
the effects of reintroducing giant tortoises to Española
Island over a 15-year monitoring period. Tortoise reintroduc-
tion decreased woody plants and increased grass cover at a
landscape scale and enlarged open habitats for other species,
such as the endemic Galapagos albatross, to use. The largest
effect was on the recovery of tree cactus, an important spe-
cies for many island animals. —BEL
Conserv. Lett. (2023) 10.1111/conl.12968
HOST DEFENSE 5-hydroxyindoleacetic acid
(5-HIAA), which is produced
Serotonin metabolite by platelets and mast cells, is
promotes fungus recognized by the eosinophil
Cryptococcus neoformans is a chemokine receptor GPR35,
fungal pathogen that can cause which facilitates eosinophil
fatal opportunistic infections of recruitment into the lung.
PHOTO: TUI DE ROY/MINDEN PICTURES

Science, adg2657, this issue p. 1384
MOLECULAR INTERACTIONS
Folding before binding
Intrinsically disordered
sequences have been
known to acquire defined
CSF administration of gene
therapy through the cochlear
aqueducts could eventually
be a treatment for genetic
deafness in adult patients.
—MN
Sci. Transl. Med. (2023)
10.1126/scitranslmed.abq3916
the lung and brain (cryptococ-
cosis). High levels of eosinophils,
which are a hallmark of type
2 responses that normally
target helminths, also occur
during cryptococcosis. De
Giovanni et al. found in mice
that a serotonin metabolite,
Knocking out GPR35 or
elements of the serotonin-pro-
cessing machinery resulted in
milder pulmonary eosinophilia,
increased type 1 helper T cells,
and less disease. —STS
Immunity (2023)
10.1016/j.immuni.2023.05.006

1334 30 JUNE 2023 • VOL 380 ISSUE 6652 science.org SCIENCE

 CANINE GENETICS
DNA details matter in
conservation
One large source of pressure
on wild populations is “genetic
swamping” caused by hybridiza-
tion with domesticated relatives.
Cairns et al. used genome-wide
genotyping arrays to assess
the genetic diversity of 391 wild
and captive dingoes across
Australia. Previous studies using
microsatellite assays estimated
that in some areas, hybrids
were the norm. In contrast to
these past studies, genome-
wide genotyping found that
dingoes are distributed among
five distinct populations across
Australia and show limited
admixture with domesticated
dogs. —CNS
Mol. Ecol. (2023)
10.1111/mec.16998
component, such as changes
in movements. Rodent models
of epilepsy are used to develop
effective therapies. Recent
developments allow recording
of brain activity at single-cell
resolution; however, behavioral
analysis still mostly relies on
subjective and often imprecise
analysis. Gschwind et al. devel-
oped a machine learning–based
video analysis that is able
to detect distinct behavioral
phenotypes in mice at sub-
second resolution (which they
called “syllables”) that are not
detectable by eye. The method
sensitively detected responses
to anti-epileptic drugs and
was able to uncover previously
unknown sex-specific behaviors
in genetic epilepsy. This type of
analysis could accelerate drug
development by allowing fast
and unbiased testing. —MMa
Neuron (2023)
10.1016/j.neuron.2023.02.003
OPTICAL TWEEZING
A merging route to
making molecules
The ability to trap and cool
atoms and molecules in single
optical trap, or a lattice of
traps, provides a powerful
platform for quantum sens-
ing technologies, as well as for
testing fundamental concepts
in physics and chemistry.
Although certain atoms have
been coaxed to pair up and
form molecules by applying a
magnetic field, not all atoms
have this magnetic property,
which limits the number of
molecular species that can be
formed. Ruttley et al. showed
that individual rubidium and
cesium atoms trapped in sepa-
rate optical traps could form
molecules as the traps were
merged. The authors were also
able to confirm the molecule
formation, establishing its
state and binding energy. This
so-called “mergoassociation”
approach should provide a gen-
eral route to forming various
molecular species. —ISO
Phys. Rev. Lett. (2023)
10.1103/PhysRevLett.130.223401

QUANTUM PHYSICS
Scaling up a paradox
One of the strangest conse-
quences of quantum mechanics
is that a measurement of one
part of an entangled system can
yield information about another,
physically distant part. This
so-called Einstein-Podolsky-
Rosen (EPR) paradox has been
experimentally observed in small
systems, but it is unclear to what
scale the paradox still holds.
Colciaghi et al. demonstrated the
EPR paradox on a Bose-Einstein
condensate of several hundreds
of rubidium atoms. The research-
ers first entangled the atoms,
then split the condensate into
two physically separated parts.
Measuring the collective spin
components of the two halves
revealed correlations between AIR POLLUTION
them that were indicative of the
EPR paradox. —JS Losing visibility

N
Phys. Rev. X (2023)
10.1103/PhysRevX.13.021031
PHOTO: PAWAN SHARMA/GETTY IMAGES
orthern India frequently experiences extreme smog in late autumn and winter, a situation
that has worsened in recent decades. Is this trend caused by more than just increasing
amounts of anthropogenic air pollution or is the problem more complex? Gautam et al.
present 40 years of satellite data showing that greater lower tropospheric stability and an

EPILEPSY
Decoding behaviors
in epilepsy
Epilepsy is a neurological
disorder that, in addition to
electrophysiological abnormali-
ties, can also have a behavioral
increase in relative humidity have driven a more than fivefold increase in low-visibility days
over that period. A better understanding of this important aerosol-radiation-meteorological feed-
back should aid in the development of better pollution mitigation strategies. —HJS
Geophys. Res. Lett. (2023) 10.1029/2023GL103105
Greater lower tropospheric stability and an increase in relative humidity have dramatically increased the frequency
of low visibility days in Northern India over the past four decades.

SCIENCE science.org 30 JUNE 2023 • VOL 380 ISSUE 6652 1335

R E S EARC H
ALSO IN SCIENCE JOURNALS
SPATIAL OMICS
RIBOmap charts
translation
A long-held goal of biological
research is to measure protein
translation with spatially
resolved, single-cell resolution.
Existing methods such as ribo-
some profiling measure protein
translation as an average of
many cells and lack spatial
resolution. Zeng et al. developed
a highly multiplexed, ribosome-
bound messenger RNA imaging
technique called RIBOmap and
applied it in single cells in situ to
profile translation events with
spatial coordinates (see the
Perspective by Fan). Thousands
of genes were simultaneously
mapped in intact cells and
tissues at molecular resolu-
tion, revealing the regulatory
principles that specify the loca-
tion and efficiency of protein
production for functionally
relevant gene programs across
different cell types and tissue
regions. —DJ
Science, add3067, this issue p. 1337;
see also adi6844, p. 1321
NEUTRINO ASTROPHYSICS
Neutrinos from within the
Milky Way
Observations of high-energy
astrophysical neutrinos
have shown that they mostly
originate from extragalactic
sources such as active galaxies.
However, gamma-ray observa-
tions show bright emission from
within the Milky Way Galaxy, and
astrophysical gamma rays and
neutrinos are expected to be
produced by the same physi-
cal processes. The IceCube
Collaboration searched for neu-
trino emission from within the
Milky Way (see the Perspective
by Fusco) and found evidence
of extra neutrinos emitted along
the plane of the Galaxy, which is
consistent with the distribution
of gamma-ray emission. These
results imply that high-energy
neutrinos can be generated by
1335-B 30 JUNE 2023 • VOL 380 ISSUE 6652
Edited by Michael Funk
nearby sources within the Milky
Way. —KTS
Science, adc9818, this issue p. 1338;
see also adi6277, p. 1318
MICROBIOLOGY
Tripping up trypanosomes
Treatments for diseases caused
by trypanosomatid parasites,
including Chagas disease and
African sleeping sickness,
remain a crucial unmet medi-
cal need for millions of people
globally. Rao et al. identified and
characterized a cyanotriazole
compound class that was able
to kill trypanosomes by selec-
tive inhibition of the parasite
topoisomerase II enzyme (see
the Perspective by Aphasizhev
and Aphasizheva). Cryo–elec-
tron microscopy revealed that
cyanotriazoles poison the
parasite topoisomerase, thereby
causing irreversible and lethal
DNA damage. This mechanism
led to rapid clearance of para-
sites both in vitro and in mouse
models of Chagas disease and
sleeping sickness. Cyanotriazole
treatment has the potential to
improve therapies for these
neglected diseases. —SMH
Science, adh0614 this issue p. 1349;
see also adi5925, p. 1320
GENOME EVOLUTION
Mavericks ignore species
boundaries
Horizontal gene transfer occurs
when DNA from one species is
nonsexually integrated into the
germline of another. In studying
the nematode Caenorhabditis
briggsae, Widen et al. identified
a toxin-antidote system with
an origin that can be traced
back to an ancient virus–like
transposon called Maverick. The
authors elucidated the muta-
tions responsible for the toxin’s
action and identified a compli-
cated pattern of horizontal gene
transfer between deeply diverged
nematode species. These trans-
fers have resulted in the exchange
of several genes, often in the
proximity of Mavericks, suggest-
ing that these transposons may
act as vectors for horizontal gene
transfer in animals more broadly.
—CNS
Science, ade0705, this issue p. 1336
2D MATERIALS
Generating and trapping
Rydberg excitons
Rydberg states are excited states
of atoms and molecules, the
“inflated size” of which provides
enhanced interactions that can
be used in quantum simulators
and sensor applications. Hu et
al. showed that optically excited
Rydberg excitons (excited
Coulomb-bound electron-hole
pairs) in monolayer tungsten
diselenide could be confined and
controlled using the narrow and
sharp potential wells of a moiré
lattice generated in an adjacent
small-angle twisted bilayer
graphene. The generation, trap-
ping, and eventual controlled
interactions between Rydberg
excitons held in a moiré lattice
would be useful for developing a
solid-state platform for quantum
simulation and sensing. — ISO
Science, adh1506, this issue p. 1367
GAMMA-RAY BURSTS
A bright GRB observed at
very high energy
Long gamma-ray bursts (GRBs)
are produced by the explosion
of a high-mass star, which pro-
duces a jet of material moving
close to the speed of light. The
LHAASO Collaboration observed
the extremely bright GRB
221009A in very-high-energy
(tera–electron volt) gamma
rays. The GRB serendipitously
occurred within the large field
of view of their detector, so
these data cover the rapid rise,
peak emission, and gradually
dimming afterglow. Modeling of
the observations showed that
the jet had an opening angle
of less than one degree, which
must have been pointed almost
exactly toward Earth, explaining
the unusual brightness of this
GRB. —KTS
Science, adg9328, this issue p. 1390
SIGNAL TRANSDUCTION
Kinases link metabolic
state to cell death
A critical phosphorylation event
links the metabolic state of a
cell with control of cell death
and inflammation. Adenosine
monophosphate–dependent
protein kinase (AMPK) is a sen-
sor of the nutrient status and
energy state of the cell. Zhang
et al. found that activated AMPK
phosphorylates and inhibits
receptor-interacting protein
kinase 1 (RIPK1) in nutrient-
deprived cells, thus inhibiting cell
death and inflammation (see the
Perspective by Hardie). Under
more prolonged nutrient stress,
such inhibition was lost, allow-
ing cell death to ensue. These
results also confirm RIPK1’s role
in cell death caused by ischemia,
thus implicating the AMPK-
RIPK1 interaction as a potential
therapeutic target. —LBR
Science, abn1725, this issue p. 1372;
see also adi6827, p. 1322
FUNGAL INFECTION
Activating immunity in
rewilded mice
Although mice are a valuable
model organism for studying the
immune system, laboratory mice
kept in specific pathogen–free
facilities lack the complex micro-
bial exposure that humans have.
Using “rewilded” laboratory
mice exposed to a seminatural
outdoor environment, Chen et
al. found that fungal coloniza-
tion induced a distinct program
of granulopoiesis that could
also be elicited by Candida
albicans intestinal colonization.
C. albicans–induced granu-
lopoiesis was dependent on
interleukin-6 receptor signaling
and the hypha-associated factor
candidalysin. Fungal colonization
science.org SCIENCE
 R E S E A RCH

led to sustained neutrophil

expansion and protected against
bloodborne bacterial infection
with Staphylococcus aureus,
highlighting the impact of
environmental fungi on myeloid
cell differentiation and innate
immunity. —CO
Sci. Immunol. (2023)
10.1126/sciimmunol.add6910

IMMUNOLOGY
T cells develop
survival skills
Mouse thymocyte survival
depends on the suppression of
cell death by the inhibitor of kB
kinase (IKK) complex, which
phosphorylates and inactivates
the cell death–promoting kinase
receptor-interacting protein
kinase 1 (RIPK1). Using lineage-
specific, conditional knockout
mice, Carty et al. investigated
the role of IKK in the survival of
mature naïve T cells after they
left the thymus. In addition to
inhibiting RIPK1-dependent cell
death, IKK also activated prosur-
vival signaling, which depended
on the transcription factor
nuclear factor-kB and one of its
targets, the cytokine receptor
interleukin-7R. —JFF
Sci. Signal. (2023)
10.1126/scisignal.abo4094

SCIENCE science.org 30 JUNE 2023 • VOL 380 ISSUE 6652 1335-C

RES EARCH
RESEARCH ARTICLE SUMMARY
GENOME EVOLUTION
Virus-like transposons cross the species barrier
and drive the evolution of genetic incompatibilities
Sonya A. Widen†, Israel Campo Bes†, Alevtina Koreshova, Pinelopi Pliota,
Daniel Krogull, Alejandro Burga*
INTRODUCTION: When we think of DNA, we
think of inheritance from parent to offspring.
However, DNA can also be transmitted be-
tween individuals that are not related and even
between individuals belonging to different
species in a process known as horizontal gene
transfer (HGT). Although best characterized
in bacteria, a growing body of evidence indi-
cates that HGT can also take place in a wide
range of eukaryotes, including fungi, plants,
and animals. However, one of the most fascinat-
ing aspects of HGT, the mechanism of transfer
in metazoans, remains largely uncharacterized.
RATIONALE: Many microorganisms can take
up DNA from their environment, and some
have even evolved specialized machinery to
transfer DNA between peers. In sharp con-
trast, transfer of DNA between multicellular
organisms poses a greater logistical challenge.
For HGT to take place, DNA must first find its
way out of the donor species, come in close con-
tact with the germ line of a second species (which
is physically and molecularly guarded), and, final-
ly, integrate into the genome of the new host.
Mavericks mediate A Viruses
HGT between nematodes.
(A) Mavericks share features
of both transposable
Mavericks
elements and viruses. In
nematodes, they acquired Host genome
a new protein, MFUS-1,
that structurally resembles
known viral fusogenes.
(B) We found that two
B Species A
new protein families
(cargo genes) have been
extensively transferred
between nematode species
on a global scale because of Millions of
years ago
their association with
Mavericks, even between
species belonging to differ-
ent genera.
Widen et al., Science 380, 1336 (2023) 30 June 2023
◥ occurred between species that last shared a
Viruses have long been hypothesized to medi-
ate HGT because of their known ability to take
up DNA sequences from their hosts, integrate
into genomes, and infect different species. How-
ever, evidence is sparse and mostly limited to
highly specialized viral symbionts that have co-
evolved in the genomes of parasitoid wasps.
RESULTS: While studying nematodes, we dis-
covered one of the long-sought vectors of HGT
in eukaryotes: Maverick elements. Mavericks,
also known as Polintons, are unique because
they share features of both transposons and
viruses. Like transposons, Mavericks are
flanked by terminal inverted repeats and can
readily jump and insert into genomes. Like
viruses, they code for a large number of pro-
teins, including a type-B DNA polymerase, a
retroviral-like integrase, as well as major and
minor capsid proteins. We discovered that two
new nematode gene families, wosp proteases
and krma kinases, are preferentially taken up
as cargo by Mavericks and have been exten-
sively transferred between different nematode
species on a global scale. Many of these transfers
Transposons
Nematode Maverick (Polinton)
Cargo Capsid
mfus-1
TIR Integrase Fusogen
(novel acquisition)
Putative Maverick
infective particles
>95% Horizontal
sequence gene transfer
identity
Cargo
Species B
common ancestor likely hundreds of millions
of years ago. We also found that nematode
Mavericks captured a new fusogen, MFUS-1 (for
Maverick fusogen), which is structurally most
similar to the glycoprotein B from Herpes
simplex virus 1. This event likely fueled the
spread of Mavericks through the formation
of enveloped infective particles, analogous to
the inception of retroviruses from genomic
retroelements. Finally, we show how the un-
ion between a horizontally transferred wosp
protease, msft-1, and a MULE (Mutator-like
elements) transposon gave birth to a new
class of selfish gene in the nematode Caeno-
rhabditis briggsae: a mobile toxin-antidote
element that causes genetic incompatibilities
in wild populations.
CONCLUSION: Our study illustrates how the
intertwined biology of viruses and transposons
can ultimately affect gene flow between popula-
tions. Indeed, HGT can result in regions of high
sequence similarity between distantly related
populations, masquerading as selection or intro-
gression. Although different vectors of HGT
likely exist in nature, we predict that Mavericks,
and analogous transposable elements with viral
properties, could mediate HGT in other animal
eukaryotic lineages, including vertebrates.
▪
The list of author affiliations is available in the full article online.
*Corresponding author. Email: alejandro.burga@imba.oeaw.ac.at
†These authors contributed equally to this work.
Cite this article as S. A. Widen et al., Science 380, eade0705
(2023). DOI: 10.1126/science.ade0705
READ THE FULL ARTICLE AT
https://doi.org/10.1126/science.ade0705
DNA polymerase
TIR
C. plicata Maverick Herpes Simplex
MFUS-1 Virus 1 gB
1 of 1
RES EARCH

◥ (27). We expected the genetic background of

RESEARCH ARTICLE the backcrossed population to be AF16 except
for the HK104 Chr. III region that contains the
GENOME EVOLUTION TA. Consistent with this prediction, we found
a single introgressed haplotype spanning ~10 Mb
Virus-like transposons cross the species barrier (Chr. III: 0 to 10 Mb; Fig. 1B). To narrow down
the candidate region, we derived 50 indepen-
and drive the evolution of genetic incompatibilities dent, near-isogenic lines (NILs) from the back-
crossed population (fig. S1A) and genotyped
Sonya A. Widen1†, Israel Campo Bes1†‡, Alevtina Koreshova1,2, Pinelopi Pliota1, them using established markers (28, 29). From
Daniel Krogull1,2, Alejandro Burga1* these lines, we selected two NILs carrying large
but partially overlapping introgressions (over-
Horizontal gene transfer, the movement of genetic material between species, has been reported lap region: 4.94 to 6.75 Mb; fig. S1B). Using
across all major eukaryotic lineages. However, the underlying mechanisms of transfer and their impact crosses, we determined that the TA was located
on genome evolution are still poorly understood. While studying the evolutionary origin of a selfish in the overlapping introgressed region (fig.
element in the nematode Caenorhabditis briggsae, we discovered that Mavericks, ancient virus-like S1C and table S1). Repeated attempts to narrow
transposons related to giant viruses and virophages, are one of the long-sought vectors of horizontal the introgressions through backcrossing were
gene transfer. We found that Mavericks gained a novel herpesvirus-like fusogen in nematodes, leading to unsuccessful, likely because of the low rate of
the widespread exchange of cargo genes between extremely divergent species, bypassing sexual recombination in the center of C. briggsae chro-
and genetic barriers spanning hundreds of millions of years. Our results show how the union between mosomes (24). To circumvent this, we took ad-
viruses and transposons causes horizontal gene transfer and ultimately genetic incompatibilities vantage of existing AF16 × HK104 recombinant
in natural populations. inbred lines (RILs) and selected three of them
carrying different breakpoints across the candi-

L
ike the early explorers who braved the
seas, genes can also embark on extraor-
dinary journeys. Horizontal gene trans-
fer (HGT), the nonsexual movement of
genetic material between species, is a
well-characterized and common phenomenon
in prokaryotes (1). By contrast, transfer of
genes between eukaryotes is thought to be
rare, especially in metazoans, because it requires
a chain of unlikely events: DNA must find its
way out of the donor species, come in close
contact with the germ line of a second species,
and finally integrate itself in the genome of
the new host. Nonetheless, from antiparasitic
toxins in butterflies to antifreeze proteins in
fish, a growing body of evidence indicates that
HGT between eukaryotes is far more common
than was previously thought and could be an
important evolutionary force (2–11). However,
one of the most fascinating aspects of HGT,
the mechanism of transfer, remains a mystery.
Viruses have been proposed to mediate HGT
because of their known ability to take up se-
quences from their hosts and integrate into
genomes (12–16). But evidence for this is sparse
and mostly limited to highly specialized viral
symbionts that have co-evolved in the ge-
nomes of parasitoid wasps for millions of years
(12, 17–20). Here, by studying the evolutionary
origin of a selfish toxin-antidote (TA) element
in the nematode Caenorhabditis briggsae, we
found that Mavericks, ancient eukaryotic virus-
1
Institute of Molecular Biotechnology of the Austrian
Academy of Sciences (IMBA), Vienna BioCenter (VBC), 1030
Vienna, Austria. 2Vienna BioCenter PhD Program, Doctoral
School of the University of Vienna and Medical University of
Vienna, A-1030 Vienna, Austria.
*Corresponding author. Email: alejandro.burga@imba.oeaw.ac.at
†These authors contributed equally to this work.
‡Present address: Centre for Genomic Regulation (CRG), Barcelona
Institute of Science and Technology, 08003 Barcelona, Spain.
like transposons, are widespread vectors of HGT
between nematodes. Further, we show how HGT
fueled the evolution of genetic incompatibilities
in natural populations.
Results
A maternal-effect selfish TA in C. briggsae
TAs are typically made up of two linked genes,
a toxin and its cognate antidote. In a maternal-
effect TA, the toxin is expressed in the germ line
of mothers and loaded into all eggs before
fertilization (21). Only embryos that inherit at
least one copy of the TA can express the anti-
dote to counteract the effect of the toxin and
survive (22). In this way, TAs selectively de-
crease the fitness of haplotypes lacking them
(hereafter referred to as “susceptible alleles”)
and rapidly spread in populations. We recently
discovered that a genetic incompatibility be-
tween the C. briggsae reference strain AF16
(Ahmedabad, India) and the wild isolate HK104
(Okayama, Japan) stems from a maternal-effect
TA (23–26). This TA causes developmental delay
in ~20% of their F2 progeny, and all delayed
progeny are homozygous for the susceptible
allele (AF16) (Fig. 1A) (23). When grown at 25°C,
C. briggsae typically take 2 days to develop from
embryos to sexually mature adults; however,
poisoned progeny are unable to counteract the
effect of the toxin and take 3 days or longer to
reach sexual maturity (23, 24). Although the
delay locus was previously mapped to HK104
chromosome (Chr.) III, the toxin- and antidote-
encoding genes were unknown.
To fine-map the underlying genes, we took
advantage of the inherent gene drive activity
of TAs (23). We backcrossed hybrid hermaph-
rodites to AF16 males for 16 generations and
sequenced the genome of the resulting popu-
lation (fig. S1A and materials and methods)
date region (24). By crossing each RIL to AF16
and testing for delay in the F2 population, we
narrowed the candidate region to 151 kb (Chr.
III: 4.94 to 5.09 Mb; Fig. 1C). For all subsequent
experiments, we used NIL1, which had the small-
est introgression (hereafter simply referred to as
NIL), and confirmed that it behaved similarly to
HK104 in both maternal and paternal back-
crosses to AF16 (fig. S1D and table S1) (23).
To identify candidate genes, we de novo
assembled the HK104 genome using a com-
bination of Oxford Nanopore long reads and
Illumina short reads. The final assembly con-
sisted of 18 scaffolds, with an N50 of 14,833,495 bp
(Fig. 1D). We then sequenced the HK104 tran-
scriptome and created comprehensive gene
annotations using both transcriptome-guided
and de novo gene calling. Our final annotation
included 26,658 genes (27,569 transcripts),
and analysis of conserved genes showed very
high completeness (96.7% of conserved nema-
tode genes were identified using BUSCO). Among
the 39 genes in the candidate region, we filtered
for genes in HK104 that were either absent,
mutated, or highly divergent in AF16 or lowly
expressed in AF16 and in tight genetic linkage
(usually close neighbors). Only one pair of genes
satisfied these criteria: ORF14767 and ORF14768
(Fig. 1E and table S2).
Because of its high expression level, we hy-
pothesized that ORF14767 is the toxin-encoding
gene. To test this, we first used CRISPR/Cas9
homology-directed repair in the NIL back-
ground to introduce a premature stop codon
in the second exon of ORF14767, leading to
early truncation of the predicted protein (p.S77X;
fig. S2A). During the generation of these worms,
ORF14767 was duplicated, but the second locus
coded for an identical mutant protein (fig. S2).
The resulting ORF14767(–) worms did not show
any abnormal phenotypes (n = 0/100; fig. S3A).

Widen et al., Science 380, eade0705 (2023) 30 June 2023 1 of 14

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

A Toxin Antidote
HK104
B Chr.
1
I II III IV V X

supporting HK104 allele

AF16

Fraction of reads
0.75
tox ant
HK104
P0 x 0.5 AF16

NIL AF16
0.25

0
F1 0 6 12 0 6 12 0 6 12 0 6 12 0 6 12 18 0 6 12 18
Position (Mb)

C AF16
4.88 4.98 5.03 5.19*
Genetic markers
5.52 6.73 6.92*
F2 delay
Chr. III (Mb): (cross to AF16)
NIL 18% (n=281)

PB1171 22% (n=78)

F2
PB1113 24% (n=94)
25% WT 50% WT 20% delayed
5% WT PB1151 3% (n=99)
candidate region

D E F
AF16 chromosomes HK104
ORF14767

Chr. III (Mb): 6.486 6.488 6.49 6.492 6.494 6.496

50
WT
6.487 6.489 6.491 6.493 6.495 6.497
Null mutant

Delayed F2 progeny (%)

40
ORF14768 ORF14769
ORF14767 ORF14770

n=281
AlignmentsTrack

150
RNA-seq

30
HK104

100

50

20
0

n=100
short-reads short-reads

10
HK104

0
tox(+) tox(-)
AF16

F1 genotype (selfing)
HK104 scaffolds

Fig. 1. A maternal-effect selfish TA in C. briggsae. (A) Mechanism of that was gathered from Ross et al. (24). (D) Synteny analysis between
action of the HK104 TA. In crosses between an NIL carrying the TA and the C. briggsae reference strain AF16 chromosomes and the HK104 de novo assembly
susceptible strain (AF16), all of the F2 progeny are poisoned, but only ~20% of scaffolds. The 20 HK104 scaffolds are numbered according to size (largest to
them are developmentally delayed because AF16 homozygotes cannot express smallest), and the 11 remaining much smaller scaffolds are not shown.
the antidote. The F2 delay phenotype is not fully penetrant (23). (B) Allele (E) Alignment of short reads and mixed embryo-stage RNA-seq to HK104 de
frequencies after 16 generations of maternal backcrosses to AF16 illustrating a novo assembly (y axis is coverage). Highlighted in blue is the strongest candidate
large introgression on Chr. III. (C) Genotype of NIL1 and the AF16 × HK104 gene, ORF14767. (F) A frameshift mutation in ORF14767 in the NIL background is
RILs PB1171, PB1113, and PB1151 at genetic markers across the Chr. III candidate sufficient to abolish the delay phenotype in F2 progeny (bottom). In all panels,
region. To determine whether each line still carries the HK104 TA, they were HK104 is shown in cyan and AF16 is shown in salmon. Error bars indicate
crossed to the susceptible strain AF16 and their F2 progeny were analyzed for 95% confidence intervals calculated with the Agresti-Coull method. tox, toxin;
developmental delay (right). Asterisk (*) indicates genetic marker information ant, antidote.

To determine whether the ORF14767 muta- sequences. Although our annotation pipeline between the TA and the MULE, but also a
tion disrupted toxicity of the TA, we crossed did not identify any clear open reading frame recent transposition event. In light of this,
ORF14767(–) NIL hermaphrodites to AF16 males. (ORF) in the ~3.7-kb region between the toxin we named the toxin msft-1 (for maternal serine-
We observed no delayed individuals among the and the downstream gene ORF14768, we de- protease toxin featuring transposon and
F2 progeny (n = 0/100; Fig. 1F) compared with tected low levels of transcription (Fig. 1E). By pronounced “misfit”).
17.8% in the ORF14767 wild-type cross (Fig. 1, C leveraging de novo gene predictions and evo- Because MULEs are poorly characterized in
and F), indicating that ORF14767 is the HK104 lutionary conservation, we identified a gene nematodes, we predicted the structure of the
toxin. An independently derived ORF14767-null with high homology to transposases belonging C. briggsae MULE transposase using Alpha-
allele (p.L104PfsX19; fig. S2C) using a second to the Mutator-like elements (MULE) family Fold2 (33) and identified the residues that
guide RNA also fully abrogated the toxicity of of transposable elements (Fig. 2A and fig. S5). make up its characteristic DDE motif catalytic
the TA (n = 0/99; fig. S3B). ORF14767 encodes an MULEs are “cut-and-paste” DNA transposons triad based on structural homology to ISCth4,
uncharacterized protein (325 amino acids long) that are known for their propensity to insert a prokaryotic transposase of the IS256 fam-
with a predicted serine-protease domain (fig. S4). into genic regions and acquire host genes (30–32). ily related to eukaryotic MULEs (fig. S6) (34).
The toxin and the MULE transposase were These three residues (Asp21, Asp83, and Glu198)
msft-1 is part of a novel mobile TA element flanked by terminal inverted repeats (TIRs) were perfectly conserved in MULE transposases
To study the evolutionary origins of the TA, 870 bp long and 96.7% identical, strongly from 49 different nematodes species, including
we focused our attention on the surrounding suggesting not only a functional association HK104, suggesting that the transposase linked

Widen et al., Science 380, eade0705 (2023) 30 June 2023 2 of 14

RES EARCH | R E S E A R C H A R T I C L E
A
HK104 TIR
MULE TIR
transposon
C Chr. III
HK104
AF16
JU439
VX34
QR24
QX1410
ED3036
0 5k 10k
Fig. 2. msft-1 is part of a mobile selfish TA. (A) The msft-1 TA is contained within a MULE transposon. A conserved MULE transposase is found in between
the toxin and ORF14768, and all three genes are flanked by TIRs. (B) Conservation of nematode MULE transposons. MULE transposase (green) and miso (light blue)
are flanked by TIRs (yellow). Alignment of representative MULE proteins across nematodes. Asterisks (*) show catalytic residues in the transposase. (C) Natural
genetic variation in C. briggsae populations supports the mobilization of the msft-1 TA in wild populations. Alignment of Chr. III and Chr. V regions that carry a copy of
the TA. (D) Sampling distribution of C. briggsae wild isolates. Color code shows the TA global localization.
to msft-1 is functional (Fig. 2B). In addition to
their conserved transposase, MULEs typically
contain additional ORFs (35). We then anno-
tated homologs of ORF14768 across a wide range
of divergent nematodes, including distantly
related parasitic species such as Haemonchus
placei and Ancylostoma ceylanicum (Fig. 2B).
Homologs of ORF14768 were found in close
proximity (±6 kb) to the transposase catalytic
domain in ~40% (67/161) of annotated MULE
transposons, suggesting a functional association
between the two genes. We therefore named
the novel protein family “mule internal second
ORF” (MISO) and the ORF14768, miso-1.
To determine whether the TA is indeed
mobile in nature, we assembled the genomes of
three additional C. briggsae wild isolates using
Nanopore long reads (fig. S7). Analysis of these,
in concert with two additional genomes (36),
revealed that the insertion site of msft-1 varied
between natural isolates. The msft-1 TA was
located on Chr. III in wild isolates from China
and Iceland, but was found on Chr. V in iso-
Widen et al., Science 380, eade0705 (2023)
msft-1 transposase
MULE transposon (8 kb)
Transposase catalytic domain
transposase
MULE internal second orf
Chr. V D
15k 20k 0 5k 10k 15k
lates from Canada, Saint Lucia, and Taiwan
(Fig. 2, C and D). The Chr. V TA was pseudo-
genized but largely complete. It lacked the
transposase and miso-1 sequences but did con-
tain most of the toxin and fragments of both
flanking MULE TIRs. Thus, standing variation
in wild C. briggsae populations supports the
view that the msft-1 TA was capable of excision
and integration into new genomic loci in recent
evolutionary history.
The MSFT-1 toxin belongs to a new family of
serine proteases
To better understand the mechanism underlying
MSFT-1 toxicity, we predicted its tertiary struc-
ture using AlphaFold2 (33). The resulting model
had a high accuracy (fig. S8) and revealed two
distinct domains that are connected by a flexible
linker (Fig. 3A). The N-terminal domain (amino
acids 1 to 83) displays a ubiquitin-like fold,
whereas the C-terminal one (amino acids 106
to 325) matches the serine protease domain
predicted by InterPro (fig. S4). An indepen-
30 June 2023
TIR
miso TIR
JU439
QR24
VX34 HK104
AF16 ED3036
QX1410
msft-1 Chr. III msft-1 Chr. V No msft-1
dent model generated using RoseTTAfold re-
sulted in a similar structure [root mean square
deviation (RMSD) = 1.3 Å and TM-align score =
0.85 for the protease domain; fig. S8]. We
then used the AlphaFold2 MSFT-1 model as
a template to search for proteins with a similar
structure in the Protein Data Bank (PDB)
using DALI (37). MSFT-1 was similar to the
high-temperature requirement A (HtrA) fami-
ly of serine proteases (Fig. 3B) (38–40). HtrA
proteases are widespread in bacteria and
eukaryotes and they typically contain one
C-terminal PDZ domain, which stabilizes
interactions with their substrates and modu-
lates the activity of the protease domain (41).
In bacteria, HtrA proteases, also known as
Deg, mediate protein quality control and can
also act as virulence factors (40, 42, 43). De-
spite sharing very low sequence similarity and
lacking a PDZ domain, DELTA-BLAST and
HHbit, two remote homology detection algo-
rithms, identified MSFT-1 as being homologous
to bacterial Deg proteases. We then aligned the
3 of 14
RES EARCH | R E S E A R C H A R T I C L E

MSFT-1
A 1 83 106 325 B C MSFT-1
AlgW
His148
MSFT-1 NTD CTD Ser256
Asp184

Ubiquitin-like fold Serine-protease

AlgW DegP
D
MSFT-1 WT
n=281
x AF16

256
MSFT-1 Ser Ala
n=136
(selfing)

256
MSFT-1 Ser Ala
n=120
x AF16
Linker
P. aeruginosa E. coli 0 10 20 30 40
Delayed F2 progeny (%)

E C. briggsae - C. plicata C. briggsae - C. auriculariae C. briggsae - C. parvicauda

msft-1 vs. Cparv-wosp-1

msft-1 vs. Cauri-wosp-5
Nucleotide alignment

msft-1 vs. Cpli-wosp-2

exon intron mismatch indel (1 bp) indel (>2 bp)

F
6

C. briggsae - C. plicata C. briggsae - C. auriculariae C. briggsae - C. parvicauda

6
5
substitution rate (dS)
5

5
4
Synonymous

4
3
3

3
2
2

2
1
1

msft-1 msft-1 msft-1

0
0

20 30 40 50 60 30 40 50 60 30 40 50 60
Effective number of codons (ENC)
G msft-1
C. briggsae Adintoc3
Adenain
Complete Maverick ATPase
Cpli-wosp-2 Cargo
C. plicata
Fusogen
Integrase
Cauri-wosp-5 Major capsid
C. auriculariae Minor capsid
pPolB
TIR - Maverick
Cparv-wosp-1 TIR - MULE
C. parvicauda
MULE transposase
MULE miso
0 5 kb 10 kb 15 kb 20 kb

Fig. 3. Horizontal gene transfer of msft-1 and related proteases. (A) Predicted C. parvicauda wosp-1. Alignments include intronic sequences. Mismatches, deletions,
domains and AlphaFold2 model of MSFT-1. (B) Comparison of the serine- and insertions are highlighted. (F) VHICA for each species pair. Number of single-
protease domain of MSFT-1 and two representative HtrA proteases, AlgW from copy orthologs analyzed for each species pair: C. briggsae HK104 and C. auriculariae
P. aeruginosa (PDB: 7CO5) and DegP from E. coli (PDB: 6JJK). a-chains are shown (1370); C. briggsae HK104 and C. plicata (1775); and C. briggsae HK104 and
in red, b-sheets in yellow, and linkers in green. (C) Structural alignment of the C. parvicauda (2172). msft-1/wosp gene pairs are shown in red. Solid black lines
protease domains of MSFT-1 (cyan) and AlgW (orange). Focus is on the three key represent the linear regression of ENC-dS, and dashed red lines correspond to
catalytic residues. (D) Mutation of MSFT-1 Ser256 to Ala abolishes its toxicity in the cutoff P value of 0.05. (G) Alignment of the C. briggsae TA and its closest
crosses (P < 0.0001, chi-square test). Error bars indicate 95% confidence intervals orthologous wosp proteases in basal Caenorhabditis species. Highlighted are
calculated with the Agresti-Coull method. (E) Nucleotide alignment of C. briggsae genes and TIRs of MULE and Maverick transposable elements. The Mavericks of
msft-1 to its closest orthologs, C. plicata wosp-2, C. auriculariae wosp-5, and C. auriculariae and C. parvicauda are pseudogenized.

Widen et al., Science 380, eade0705 (2023) 30 June 2023 4 of 14

RES EARCH | R E S E A R C H A R T I C L E
protease domain of MSFT-1 to Pseudomonas
aeruginosa AlgW, a representative HtrA pro-
tein (PDB: 7CO5), and identified the residues
that make up its catalytic triad based on struc-
tural homology: His148, Asp184, and Ser256
(RMSD = 1.9 Å; TM-align score = 0.71; Fig. 3C
and fig. S9). To determine whether the pro-
tease activity of MSFT-1 is necessary for its
toxicity, we mutated Ser256 to Ala in the endo-
genous msft-1 locus using CRISPR/Cas9. This
mutation abrogates the activity of serine pro-
teases because it renders them incapable of
performing the initial nucleophilic attack (44).
We then crossed msft-1 Ser256Ala NIL hermaph-
rodites to AF16 males and inspected their F2
progeny. In contrast to crosses involving the
wild-type TA, we observed no developmental
delay among F2 individuals (n = 0/120, P <
0.0001, chi-square test; Fig. 3D), and all F2
worms homozygous for the susceptible allele
(AF16) were phenotypically wild type. These
results were further confirmed in crosses with
an independently derived line carrying the same
point mutation (0.55% F2 delay, n = 1/180, P <
0.0001, chi-square test). Thus, our results in-
dicate that the proteolytic activity of MSFT-1 is
essential for its toxicity in vivo.
Horizontal gene transfer of msft-1 and
related proteases
To further investigate the evolutionary history
of MSFT-1, we retrieved homologous sequences
from 107 nematode genomes, curated gene
annotations, and built a comprehensive phylo-
genetic tree (fig. S10 and data S1). In total, we
identified 235 homologous proteins distrib-
uted across 52 species and 11 genera. To facil-
itate their study, we named this novel family of
proteases “worm HtrA-like serine protease”
(WOSP) (see data S1). Our analysis revealed
that WOSP proteases are very common in
nematodes, but they are not universal. For
instance, C. briggsae has two WOSP proteases,
CBG29787 and CBG26403, in addition to MSFT-1,
whereas C. elegans lacks WOSP orthologs alto-
gether. While inspecting the tree branch con-
taining the toxin in detail, we found that
MSFT-1 was almost identical to Cpli-WOSP-2, a
protease found in C. plicata, a rare Caenorhabditis
species associated with carrion beetles (45).
Their sequences were 95% identical at the
amino acid level (fig. S11). This is surprising
because C. plicata is one of the most basal
species of the Caenorhabditis genus (fig. S12)
(46). The coding and intronic sequences of
msft-1 and cpli-wosp-2 were 97.5% identical
at the nucleotide level, even though the ge-
nomes of C. briggsae and C. plicata are at least
as divergent as those of humans and zebrafish
(Fig. 3E and fig. S13) (47). In addition, we found
high nucleotide conservation between MSFT-1
and WOSP proteases found in C. parvicauda
(Cparv-WOSP-1) and C. auriculariae (Cauri-
WOSP-5) (Fig. 3E), two basal Caenorhabditis
Widen et al., Science 380, eade0705 (2023) 30 June 2023
species, and C. inopinata (Cino-WOSP-2), the
sister species of C. elegans (fig. S14) (46, 48, 49).
These observations raised a critical question:
How can MSFT-1 be almost identical to ortho-
logs in species that last shared a common ances-
tor tens of millions of years ago?
Introgressive hybridization, the exchange of
genetic material between different species
through hybridization and backcrossing, can
result in high levels of identity between ortho-
logs (50). However, this phenomenon is unlikely
to explain the extreme sequence identity of
msft-1 and its closest orthologs for two main
reasons. First, with the exception of C. nigoni,
there is complete reproductive isolation be-
tween C. briggsae and all other Caenorhabditis
species (51). Second, such high levels of sequence
identity would require a recent hybridization
event, resulting in extensive homology in the
vicinity of the msft-1 locus caused by linkage
disequilibrium (52). However, that was not the
case. Only the protease had a high sequence
identity, and the neighboring genomic regions
were not homologous (fig. S15). Because such
an isolated and extreme case of sequence
similarity was unlikely to arise by hybridiza-
tion, we hypothesized that msft-1 was horizon-
tally transferred.
To determine whether msft-1 has been
horizontally transferred, we studied its phylo-
geny in more detail. A hallmark of HGT is the
presence of discrepancies between species and
gene trees (53). In addition to the high sequence
similarity between C. briggsae msft-1 and or-
thologs found in C. plicata, C. parvicauda, C.
auriculariae, and C. inopinata, we found that
basal Caenorhabditis species and those belong-
ing to the Elegans and Japonica groups were
frequently intermixed in the protease tree (fig.
S10). Furthermore, Caenorhabditis species no
longer formed a monophyletic group (fig. S10).
For example, C. portoensis WOSP-9 is most sim-
ilar to Bursaphelenchus okinawaensis WOSP-3,
and C. virilis WOSP-1 is most similar to Meso-
rhabditis belari WOSP-8 (fig. S10).
Because gene loss and duplication can lead
to topological discrepancies between gene and
species trees (54), we also tested the HGT
hypothesis using a method that is independent
from phylogenetic reconstruction: Vertical and
Horizontal Inheritance Consistence Analysis
(VHICA) (55). VHICA is a statistical frame-
work that compares genome-wide rates of neu-
tral evolution in coding genes and identifies
horizontally transferred genes based on their
unusually low rate of synonymous substitu-
tions (dS). VHICA accounts for variation in
the number of synonymous sites under puri-
fying selection by using the effective number
of codons (ENC) as a proxy to distinguish
between codon selection and HGT. To imple-
ment VHICA, we first defined a set of high-
confidence single-copy orthologs between
C. briggsae HK104 and the three species
carrying proteases most similar to msft-1,
C. plicata, C. auriculariae, and C. parvicauda,
and estimated dS and ENC for each of them.
As expected, we observed an overall positive
correlation between dS and ENC on a genome-
wide scale (Fig. 3F). We then identified genes
that significantly deviated from this trend;
that is, they displayed both a low dS and high
ENC (low codon bias), a hallmark of HGT.
Following these criteria, msft-1 was the most
extreme outlier genome wide, strongly support-
ing an HGT event (Fig. 3F; P < 0.00001 for all
pairwise comparisons).
Mavericks transferred the precursor of
msft-1 into C. briggsae
Having established that msft-1 is horizontally
transferred in nematodes, we wondered by
what mechanism the transfers were accom-
plished. One key finding that has emerged
from various studies is that HGT usually involves
transposable elements (7, 8, 16). Because msft-1 is
embedded within a MULE in C. briggsae, we
wondered whether this DNA transposon was
responsible for the horizontal transfer of msft-1.
To test this idea, we reconstructed the evolution-
ary history of nematode MULEs and searched
for phylogenetic incongruencies that could in-
dicate horizontal transfer. MULEs are quite chal-
lenging to annotate because they are highly
divergent, oftentimes pseudogenized, and their
transposase is normally silenced or expressed
at very low levels. Thus, we focused our analysis
on the catalytic domain of the transposase,
which is by far the most conserved domain of
the enzyme (Fig. 2B). Our analysis revealed that,
unlike WOSP proteases, Caenorhabditis MULE
transposases form an almost perfect monophy-
letic group (fig. S16). Moreover, the C. briggsae
MULE transposase was most similar to the one
found in its sister species, C. nigoni (fig. S16).
Finally, we investigated whether the C. briggsae
MULE transposase or its TIRs were highly
conserved in any other species. Unlike msft-1,
we found no evidence for high sequence
conservation of the MULE in any other viral,
prokaryotic, or eukaryotic genome publicly
available to date. Taking all of these results
into consideration, we concluded that MULE
transposons are unlikely to be involved in
HGT of WOSP proteases.
Next, we inspected in closer detail the
genomic neighborhood of the Cpli-wosp-2 pro-
tease, the closest homolog of msft-1. Although
the protease was found in a very short scaffold,
a blastx search revealed that the region at the
closest edge of the scaffold and adjacent to the
protease shared homology to viral integrases
(fig. S17). We resequenced C. plicata using
Nanopore long reads and de novo assembled
its genome. The resulting assembly consisted
of 138 scaffolds and had an N50 of 3,276,966
bp. Most genes were complete in this assem-
bly despite the high levels of heterozygosity in
5 of 14
RES EARCH | R E S E A R C H A R T I C L E
the strain that hindered its assembly (BUSCO
score 90.0% complete and 2.6% partial genes;
fig. S17). The new long read–based assembly
revealed that Cpli-wosp-2 is found within a
complete Maverick transposon. In addition,
both Cauri-wosp-5 and Cparv-wosp-1 are found
within pseudogenized Mavericks. Mavericks,
which are also known as polintons, are a class
of ancient eukaryotic mobile elements that
share features of both transposons and viruses
(56, 57). Like transposons, Mavericks are
flanked by TIRs, but like viruses, they code for
a large number of proteins, including a type-B
DNA polymerase, a retroviral-like integrase,
an FtsK-like ATPase, an adenovirus-like pro-
tease, and major and minor capsid proteins
(58, 59). We found all of these intronless genes
in the Maverick carrying Cpli-wosp-2 (Fig. 3G).
The genetic link between Mavericks and giant
viruses, virophages, and adenoviruses has led to
the hypothesis that Mavericks could, under
some conditions, form virions (59–61). In sup-
port of this view, an AlphaFold2 model of the
C. plicata Maverick major capsid protein
(MCP) revealed extensive structural similarity
to the double jelly-roll capsids of numerous
double-stranded DNA (dsDNA) viruses, includ-
ing the virophage Sputnik, as well as the giant
viruses Faustovirus and Paramecium bursaria
chlorella virus (PBCV-1) (top Z score = 24.4;
Fig. 4, A and B). Viral MCP trimers, also known
as hexons, are the main building block of
capsids. The N-terminal tails of MCPs contribute
to the stability of the homotrimer by wrapping
around their nearest counterclockwise neighbor-
ing subunit (62). An AlphaFold2 model of the
Maverick MCP trimer revealed that their
N-terminal tails also show this conserved
structural feature (Fig. 4, A and B, and fig. S18),
suggesting that Mavericks can form virions.
The C. plicata Maverick additionally codes
for a putative viral-like fusogen, which we named
mfus-1 (for Maverick fusogen). Fusogens are
proteins used by enveloped viruses to facilitate
virus–host membrane fusion. C. plicata MFUS-1 is
866 amino acids long and, like most known
viral fusogens, it has a predicted N-terminal
signal peptide (amino acids 1 to 24) and a single
transmembrane domain (amino acids 775 to
811) (Fig. 4C). We used AlphaFold2 to predict
the structure of MFUS-1 and focused our analy-
sis on a portion of the ectodomain (amino acids
50 to 270) that showed the highest accuracy
(Fig. 4C and fig. S19). A structural homology–
based search revealed that MFUS-1 was most
similar to Herpes simplex virus 1 (HSV-1) gly-
coprotein B (gB) (top Z score = 11.1; RMSD =
3.88 Å; TM-align score = 0.52; Fig. 4C), and
also showed extensive similarity to the gB of
human cytomegalovirus and Epstein-Barr virus,
which are class III viral fusogens. Our approach
identified the highly conserved domain I of
class III proteins (DI), which forms an extended
b-sheet structure and contains two loops that
Widen et al., Science 380, eade0705 (2023) 30 June 2023
form the bipartite fusion peptide (Fig. 4C). Like
its viral counterparts, MFUS-1 has many hydro-
phobic residues in the putative loop regions of
DI that could interact with target membranes
(Fig. 4D). Additionally, we found strong support
for DII (Fig. 4C) and DIII, a large a-helix that
forms a trimeric coiled coil in viruses (fig. S19).
Evolutionary path from protease to toxin
Because Cpli-wosp-2 is embedded within a
Maverick that could form enveloped virions,
whereas MULE-associated msft-1 can only
directly mobilize within the C. briggsae genome,
we reasoned that Cpli-wosp-2 represents an
ancestral state from which the toxic activity of
msft-1 evolved after a horizontal transfer event.
To explore this scenario, we analyzed in detail
the coding differences between these two genes.
Of 16 nonsynonymous substitutions between
Cpli-wosp-2 and msft-1, only one of them af-
fects a highly conserved residue (Glu273; Fig. 4,
E and F, and fig. S11). This Glu is found in very
close proximity to the catalytic site of WOSP
proteases and is likely part of the substrate-
binding site (Fig. 4E). In MSFT-1, however, this
residue was replaced by a noncharged glycine
(Glu273Gly; Fig. 4, E and F). To test whether this
specific substitution is necessary for MSFT-1
toxicity, we first used CRISPR/Cas9 to revert
Gly273 to the inferred ancestral state (Glu) in
the endogenous msft-1 locus and then crossed
these mutants to the parental AF16 strain. We
observed almost no delayed individuals among
their F2 progeny (2.2%, n = 3/135, P = 0.0001, chi-
square test; Fig. 4G), indicating that this recent
substitution is necessary for MSFT-1 toxicity.
We confirmed these results in crosses with an
independently derived line carrying the same
point mutation (5.6% delayed F2 progeny, n =
9/159, P = 0.0018, chi-square test).
Overall, we conclude that C. briggsae msft-1
evolved from a Maverick-associated protease
that was horizontally transferred to C. briggsae,
likely from a basal Caenorhabditis species
where closely related homologs are found.
Mavericks carry the genetic toolkit necessary
for making enveloped virions, suggesting that
viral-like particles could directly mediate HGT
without the need of a vector, such as another
virus or pathogenic microorganism.
Widespread horizontal transfer of Maverick
cargo across genera
MSFT-1 and its close homologs are not the
only WOSP proteases that are horizontally
transferred; in fact, their patchy phylogeny
suggests this is a common occurrence in the wild
(Fig. 5A and fig. S10). Prompted by our discovery
in C. plicata, we next investigated whether
WOSP proteases are generally associated with
Mavericks. In 48 of 235 cases (20%), we iden-
tified at least one core component of Mavericks
in close proximity (±6 kb) to a WOSP protease,
even though Mavericks typically make up
<1% of nematode genomes (P < 0.00001; Fig.
5A and data S2). We think this number is an
underestimate, because the genome assem-
blies of most of these species are low quality
and the proteases are often found at the edge
of scaffolds. WOSP proteases were associated
with Mavericks not only in Caenorhabditis
species, but also in distantly related nematodes
such as Mesorhabditis belari, Oscheius tipulae,
and even Halicephalobus mephisto, which was
discovered inhabiting a South African mine
1.3 km below the Earth’s surface (63, 64).
Next, we investigated whether Mavericks
themselves were being horizontally transferred
regardless of their association with WOSP
proteases. To do so, we first annotated the
core set of Maverick genes across 107 available
nematode genomes and reconstructed their
phylogeny (fig. S20). For all Maverick genes
inspected, we found widespread incongruencies
between gene and species trees, support-
ing multiple independent Maverick transfer
events between species belonging to different
genera, including Caenorhabditis, Pristionchus,
Panagrellus, Heterorhabditis, Mesorhabditis,
Parapristionchus, and Micoletzkya (fig. S20).
This scenario is also supported by the extreme
nucleotide identity of Maverick genes found
in nematodes from different genera (fig. S21),
as well as their GC content, which is lower than
that of their host (fig. S22). Our analysis also
revealed that most nematode Mavericks are
either incomplete or pseudogenized, as is gen-
erally the case in eukaryotes (65). However, we
identified some noteworthy exceptions. For in-
stance, Mavericks found in C. wallacei, C. plicata,
C. bovis, M. belari, and H. bacteriophora were
largely complete (Fig. 5B). The Mavericks found
in C. wallacei, C. plicata, and C. bovis were
>95% identical to each other at the nucleotide
level, suggesting recent HGT events (Fig. 5B).
A new family of kinases is also associated
with Mavericks
A common feature across highly complete
Mavericks was the presence of an ~2- to 5-kb
region between the left TIR and the integrase
that could be distinguished because of its lack
of conservation and the presence of genes with
introns such as Cpli-wosp-2 (Fig. 5B). Some of
these Mavericks did not carry a wosp gene, but
rather a gene of unknown origin (Fig. 5B). To
characterize the novel cargo, we first focused
on the C. wallacei Maverick. An InterPro scan
of its predicted 430–amino acid–encoding car-
go revealed a ubiquitin-like fold at the N ter-
minus (amino acids 1 to 73) but no additional
known domains. We then predicted its struc-
ture using AlphaFold2 (figs. S23A and S24) and
searched for structurally similar proteins in the
PDB. The novel cargo was structurally homol-
ogous to numerous bacterial kinase effectors,
virulence factors, and toxins, including Legionella
pneumophila MavQ, Helicobacter pylori CtkA,
6 of 14
RES EARCH | R E S E A R C H A R T I C L E

A B C Ectodomain Cytoplasmic
Sputnik C. plicata Maverick
MFUS-1 SP DII DI D II TM
Major capsid trimer Major capsid trimer 24 50 68 210 270 774 811 866
Sputnik b b
virophage Ectodomain Trimer Herpes Simplex Virus 1 C. plicata Maverick
gB MFUS-1

DIV

DIII DII DII

a c a c DII

DV DI
DI DI

* AA:140-500 * AA:50-270
top bottom top bottom

D E Cpli-WOSP-2 (Glu273)

WOSP
protease
domain
gB
fusion loop Electrostatic surface

Catalytic site MSFT-1 (Gly273)

His148
Asp184 Ser256

MFUS-1
Glu273 *
Hydrophobic residue (F/Y/W) Conserved Variable Positive Negative

F 256 273 G MSFT-1 WT

MSFT-1 n=281
x AF16
Cpli-WOSP-2
Cbr-WOSP-2
MSFT-1 Gly273Glu n=134
Cino-WOSP-5 (selfing)
Ccas-WOSP-1
Cbov-WOSP-3
MSFT-1 Gly273Glu n=135
Hmeph-WOSP-4
x AF16
Boki-WOSP-1
Mbel-WOSP-8
0 10 20 30 40
Otip-WOSP-1
Delayed F2 progeny (%)
* *
Fig. 4. Mavericks transferred the precursor of msft-1 into C. briggsae. represent the evolutionary conservation of each residue. Highlighted in white are
(A) Structure of the Sputnik virophage capsid and major capsid protein trimer the three catalytic residues and the highly conserved Glu273 (left). Electrostatic
(hexon) (PDB: 3J26). (B) Predicted AlphaFold2 structure of C. plicata Maverick surface potential was calculated for AlphaFold2 models of Cpli-WOSP-2
major capsid protein trimer. (C) Domain architecture of MFUS-1. SP, Signal (top right) and MSFT-1 (bottom right). The Glu273Gly substitution found in
peptide; TM, transmembrane domain (top). Structure of the gB ectodomain MSFT-1 alters the negative charge of the region next to the catalytic site (*).
from HSV-1 (PDB: 2GUM). For simplicity, only one subunit of the homotrimer is (F) Partial protein alignment of the representative MSFT-1 orthologs used to
shown in green. Key domains (DI-V) are highlighted (bottom left and generate the surface plot in (E). Highlighted are residues Ser256 (catalytic serine)
middle). AlphaFold2 MFUS-1 model (bottom right). (D) Magnification of the and Glu273 as shown in (E). Highly conserved residues in magenta. (G) Reversal
HSV-1 fusion loops in domain I of gB (top) and the homologous loops in the of MSFT-1 Gly273 near the catalytic site to the ancestral Glu residue
predicted MFUS-1 structure (bottom). Highlighted in blue are hydrophobic (Gly273Glu) abolishes its toxicity in crosses (P = 0.0001, chi-square test).
residues in the loops that are thought to interact with target membranes. Error bars indicate 95% confidence intervals calculated with the
(E) AlphaFold2 model of the serine-protease domain of Cpli-WOSP-2. Colors Agresti-Coull method.

and Shewanella oneidensis HipA (top Z score = dues (RMSD = 4.8 Å; TM-align score = 0.59; family associated with Mavericks). To de-
11.9; fig. S24). Structural alignment of the novel fig. S24), suggesting that the Maverick cargo is termine whether KRMA kinases were being
cargo to H. pylori serine/threonine kinase CtkA a kinase (66). In light of this result, we named horizontally transferred, similar to WOSP pro-
revealed conservation of its key catalytic resi- the novel cargo Cwal-krma-1 (for kinase-related teases, we comprehensively annotated KRMA

Widen et al., Science 380, eade0705 (2023) 30 June 2023 7 of 14

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

A WOSP protease phylogeny

# of species

Elegans (12)

Japonica (6)

Guadeloupensis (2)

Drosophilae (7)

Angaria (6)

Basal § (5)
*
Non-Caenorhabditis (10)
nematodes

Associated with a Maverick

* MSFT-1
Cparv-WOS
P-1
Tree scale: 1 Cpli-WOS
P-2
Cauri-W
OSP-5
Cino-W
OSP-2

B Hba-orf15111
Mavericks
H. bacteriophora
77% 88.3%
Cwal-krma-1
C. wallacei Adintoc3
93% 99.5% Adenain
Cpli-wosp-2
Recent C. plicata
HGT ATPase
Cbov-krma-2
96%
Cargo
C. bovis (*)
Fusogen
Castr-g1865(**)
73.5% 91.2%
C. astrocarya Integrase
74.7% 75.3% 66% Major capsid
C. briggsae Minor capsid
75% 79% pPolB
M. belari
TIR
79% 70%
C. virilis

0 kb 5 kb 10 kb 15 kb 20 kb

C. plicata
C 45
Maverick D p<2.22e-16
p<2.22e-16
E p<2.22e-16
p<2.22e-16
80 p<1.8e-9 p=0.75
Synonymous codon usage

35 1.00
GC content of CDS (%)
GC content (%)

25
order (SCUO)

0 10 20 30 40 50 60 70 80 90 100 60 0.75
Scaffold 1 (2.9-3.0 Mb)
Cargo gene
55 C. wallacei Maverick Maverick core gene region
0.50
40
45
0.25
35

25
20 0.00
0 10 20 30 40 50 60 70 80 90 100 Host wosp/krma Maverick Host wosp/krma Maverick
Scaffold 2 (3.3-3.4 Mb)

Fig. 5. Widespread horizontal transfer of Maverick cargo across genera. (B) Alignment of complete and largely complete Mavericks from nematode
(A) Protein phylogenetic tree of WOSP proteases (only the protease domain was genomes. Highlighted in gray are homologous sequences with high nucleotide
used). The tree contains 182 of 235 WOSPs that have a protease domain that is sequence identity. *The C. bovis scaffold is circular. **This gene is likely a
at least 100 amino acids long. The Caenorhabditis group to which each krma pseudogene. (C) Two representative examples of the GC content of
ortholog belongs is represented by the color of the inner ring. These groups are Mavericks and their host. Dotted line is the average GC content of the genome.
monophyletic except for the basal Caenorhabditis group (§) and the group Cargo gene is shown in magenta, and Maverick core genes are shown in blue.
containing non-Caenorhabditis species. Black outer ring represents those WOSP Joint comparison of the GC content [CDS (D)] and codon usage bias [SCUO (E)]
proteases that are associated with Mavericks in the genome. Orange dots denote of host (n = 74,655), cargo (n = 67), and core Maverick (n = 66) genes for
branches with a bootstrap value >85. Zoom-in shows the branch containing the the genomes of C. plicata, C. parvicauda, C. auriculariae, and C. briggsae HK104.
MSFT-1 toxin. The closest orthologs in other species are colored light blue. P values in (C) and (D) were calculated using a Wilcoxon test.

Widen et al., Science 380, eade0705 (2023) 30 June 2023 8 of 14

RES EARCH | R E S E A R C H A R T I C L E
kinases across 107 nematodes genomes and
found a total of 531 KRMA kinases (data S2).
Similar to WOSP proteases, we found wide-
spread incongruencies in the gene tree com-
pared with the species phylogeny and extreme
DNA conservation in both coding and intronic
regions between KRMA kinase homologs in
highly divergent species (fig. S23, B and C).
Furthermore, 121 of 531 (22.7%) of KRMA ki-
nases were associated with either complete or
partial Mavericks (P < 0.00001; fig. S23C and
data S3). We also found that WOSP and KRMA
proteins associated with Mavericks had shorter
terminal branches (fig. S25).
Overall, our results indicate that two novel
nematode protein families, WOSP proteases
and KRMA kinases, are preferentially associ-
ated with Mavericks. These genes likely corre-
spond to cargo sequences that were originally
“captured” by Mavericks from nematode ge-
nomes during transposition and, as a result,
broadly exchanged between species that have
been reproductively isolated for tens or even
hundreds of millions of years. In support of
this view, we found that the GC content of
wosp and krma genes largely matched that
of host genes but differed significantly from
that of Maverick core genes (Fig. 5, C and D).
Similarly, the codon usage of wosp and krma
genes is in alignment with that of their host
rather than with Maverick genes (Fig. 5E).
These results were robust and consistent across
species and independent of the association of
wosp and krma genes with Mavericks (fig.
S22). This, together with the exclusive pres-
ence of introns in cargo genes and the fact that
hundreds of wosp and krma genes are not
associated with Mavericks but are otherwise
virtually indistinguishable from nematode
genes, strongly supports the capture model.
Although we cannot formally rule out the possi-
bility that these genes are integral components
of Mavericks that evolved to resemble nema-
tode genes, we believe that a nematode origin
is the most parsimonious explanation.
Modularity of ubiquitin-like domains in
cargo proteins
In addition to being associated with Mavericks,
WOSP proteases and KRMA kinases share
another feature: Both protein families have
ubiquitin-like domains at their N terminus
despite sharing low levels of sequence iden-
tity with ubiquitin (Fig. 3A and fig. S23A).
We found that these ubiquitin-like domains
are highly modular within each protein family.
For example, the ubiquitin-like domains of
MSFT-1 and Cpar-WOSP-2 were 92% identical
at the protein level, whereas their protease
domains were only 21% identical (Fig. 6A).
By contrast, the ubiquitin-like domains of
MSFT-1 and Mbel-WOSP-8 were 27% identical,
whereas their protease domains were 72%
identical (Fig. 6A). This pattern was also evident
Widen et al., Science 380, eade0705 (2023) 30 June 2023
between MSFT-1 and Cpar-WOSP-1 (Fig. 3E).
KRMA kinases were also modular (Fig. 6A).
This unusual pattern of genetic variation is con-
sistent with recombination between different
Mavericks, analogous to viral recombination
during infection and replication cycles (67).
Discussion
In this study, we provide evidence for wide-
spread Maverick-mediated horizontal gene
transfer across extremely divergent nematode
species (Fig. 6B) that drove the evolution of a
novel MULE-associated selfish TA element (Fig.
6C). The TA found in C. briggsae is capable of
both spreading in nature by poisoning individ-
uals homozygous for the susceptible allele
and changing its position in the genome through
transposition. The existence of elements such
as the msft-1 TA could explain why numerous
transposable elements that are present in only
one or two copies per genome can persist in
the face of genetic drift (68). Because of their
association with TAs, some transposons could
rapidly spread in populations while incurring
no fitness cost. Furthermore, elements such
as the msft-1 TA provide a cautionary tale to
evolutionary biologists, because they could
be responsible for “ghost” signatures of se-
lection in genomes. Quick fixation of the TA
element followed by transposition could be
erroneously interpreted as positive selection
acting on neutral variants that were originally
linked to the TA. The msft-1 TA is reminiscent
of the P element in Drosophila (2, 11) and the
Spok block, a recently reported selfish ele-
ment in the fungus Podospora anserina (69).
The Spok block is made up of numerous Spok
TAs capable of meiotic drive, which are em-
bedded within a putative giant DNA trans-
poson, the Enterprise. Like the msft-1 TA, the
Spok block is a hyperparasite capable of both
gene drive and transposition. We show in un-
precedented detail how cooperation between
parasitic genetic elements can lead to the evo-
lution of new biological function and ultimately
affect gene flow between populations. However,
several intriguing aspects of the TA evolution
are still unknown. For instance, it is not clear
whether msft-1 became a toxin before or after
its capture by the MULE transposon and if
msft-1 has nonselfish roles that could also
have contributed to the evolution of the TA.
Our work reveals Mavericks as one of the
long sought-after vectors of HGT. Because of
their unique biology, which shares features
of both transposons and viruses, Mavericks
are responsible for the widespread transfer
of genes across extremely divergent species.
Analogous to endogenous retroelements that
gained env genes, Mavericks captured a novel
fusogen, mfus-1, which is likely instrumental
for HGT in nematodes (70, 71). However, we
could not detect transcription of the fusogen
or any other Maverick gene. We hypothesize
that Mavericks, like E. coli l phage and human
herpesvirus, integrate into the genome of their
host and passively replicate until an environ-
mental factor triggers the formation of infec-
tive particles (Fig. 6B). In support of this model,
maviruses, virophages evolutionary related to
Mavericks, are not constitutively expressed
when integrated in protozoan genomes, but
they are specifically transcribed after super-
infection by the giant virus CrV (72, 73). Future
studies focusing on the reconstitution of the
Maverick life cycle will allow us to study HGT
with unprecedented mechanistic detail in vivo.
Moreover, this research could also lead to the
development of payload delivery systems in
nematodes, for which DNA viruses are unknown
(74). Of special interest are the wosp and krma
genes. Given their specific association to
Mavericks, we speculate that their ubiquitin-
like domains could facilitate infection or inte-
gration. The structural similarity of WOSP and
KRMA to bacterial proteins may hint at an
ancient transfer from eukaryotes to nematodes;
however, we cannot exclude that their resem-
blance was caused by convergent evolution.
Mavericks are widespread in eukaryotes—and
pseudogenized copies are even present in
humans—and their gene content is highly con-
served (58, 65, 75). Although we acknowledge
that there are likely many different vectors of
HGT in nature, Mavericks could also play a role
beyond nematodes. For instance, teleosts such
as ray-finned fishes have experienced extreme
levels of HGT and have a high number of
Mavericks in their genomes (7, 65). Therefore,
we predict that Mavericks, and analogous
selfish genes with viral properties, could medi-
ate HGT in other eukaryotic lineages.
Materials and methods
Maintenance of worm strains
All worms were maintained at 25°C on modified
nematode growth medium (NGM) plates unless
otherwise indicated. The modified NGM con-
tains 1% agar/0.7% agarose as C. briggsae and
other species can burrow into normal NGM plates.
Some of the strains were provided by the CGC,
which is funded by the NIH Office of Research
Infrastructure Programs (P40 OD010440). All
strains used in this study are listed in table S3.
Data source and code
The data and code necessary to replicate the
main and supplementary figures of this manu-
script are available at Zenodo (76).
Generation of C. briggsae near-isogenic and
mutant lines
All near-isogenic lines were generated as
previously described (23), and an overview is
shown in fig. S1A. For CRISPR/Cas9 genome
editing, we adapted previously described pro-
tocols (77). In brief, 250 ng/ml Cas9 or Cas12a
(IDT) protein was incubated with 200 ng/ml
9 of 14
RES EARCH | R E S E A R C H A R T I C L E

A MSFT-1 Cwal-KRMA-1
Ubiquitin-like Ubiquitin-like

Protein Identity (%)

domain domain
Protease Kinase
domain domain
27% 72% 92% 21% 21% 87% 99% 26%

Mbel-WOSP-8 Cpar-WOSP-2 C51-KRMA-1 Cpli-KRMA-2

72% 24% 42% 91% 83% 26% 25% 70%

Cdro-WOSP-4 C44-WOSP-2 Pgibl-KRMA-3 Cvir-KRMA-9

B Integrated Mavericks Maverick

viral-like particles
external
trigger ?
HGT
...

cargo

Maverick [pPolB, capsid, fusogen, integrase, etc]

C
MULE transposition HGT basal
Caenorhabditis
Maverick HGT
MULE wosp Maverick
1 MULE transposition

C. briggsae
wosp capture MULE transposition
wosp capture
wosp
2 MULE transposition

MULE TIR Glu273Gly

MULE transposase
msft-1
MULE miso Evolution of msft-1
3 TA element
Maverick TIR toxin

Fig. 6. Modularity of Maverick cargo and model for Maverick-mediated HGT
and TA evolution. (A) Examples of modularity between the N-terminal ubiquitin-
like domain and C-terminal catalytic domain of WOSP proteases (left) and between
the N-terminal ubiquitin-like domain and C-terminal catalytic domain of KRMA
kinases (right). (B) Model for Maverick-mediated HGT of cargo genes in nematodes.
We hypothesize that under some environmental perturbation, Mavericks are
transcribed and form enveloped viral-like particles. These particles can infect
individuals from other species and integrate into their germ line. (C) Model for the
origin of the C. briggsae TA. (1) A Maverick from a basal Caenorhabditis species
infected C. briggsae and integrated into the genome introducing its cargo wosp
protease (right). A C. briggsae MULE then transposed into this Maverick (left) and (2)
captured the protease during subsequent transposition. (3) The wosp protease
evolved into the msft-1 toxin. Alternatively, the toxic activity of msft-1 could have
evolved before the capture by the MULE. The Glu273Gly mutation close to the
catalytic site was critical for evolution of toxicity and may have changed the
substrate specificity of the MSFT-1 protease domain.

crRNA (IDT) and 333 ng/ml tracrRNA (IDT)
at 37°C for 10 min before adding 2.5 ng/ml
co-injection marker plasmid (pCFJ90-mScarlet-I).
For homology-directed repair, donor oligos
(IDT) or biotinylated and melted polymerase
chain reaction (PCR) products were added at
a final concentration of 200 or 100 ng/ml, re-
spectively. After injections into young adult
hermaphrodites, mScarlet-positive F1 progeny
were singled and their offspring screened by
PCR and Sanger sequencing to detect successful
editing events. All guide RNAs and homology-
directed repair templates are listed in table S4.
Phenotyping, genotyping, and statistical analysis
of crosses
All crosses were performed at 25°C. Crosses
were set up in a single well of a 12-well plate
with six L4-stage hermaphrodites and at least
25 adult males. Worms were then allowed to
mate and lay for 48 hours. For F1-selfing experi-
ments, L4-stage F1 s were then singled onto
3-cm plates and allowed to self-fertilize over-
night. To ensure that the F2s used for pheno-
typing were approximately the same age, the
F1 mothers were transferred to a new plate in
the morning, allowed to lay for 2 hours, and
then fresh F2 eggs were singled onto 3-cm
plates for phenotyping. All F2s were then staged
every 24 hours to track their development.
Eggs that did not hatch were scored as embryo-
nic lethal, and larvae that hatched but ar-
rested before sexual maturity were scored as
larval lethal. At 48 hours, wild-type worms
have reached sexual maturity and are laying
eggs; worms were considered to be develop-
mentally delayed if they did eventually reach
sexual maturity but had not yet done so at
48 hours (in most cases, delayed worms were
at the L4 stage or just molted to the adult
stage). In all cases, F2s were monitored until
they either died or reached sexual maturity. All
crosses were performed at least twice, with 10
to 20 F2 progeny analyzed from five to 10 dif-
ferent F1 mothers in each experiment. Each
n value given indicates the total number of F2
progeny that were counted. Where applicable,
P values were calculated using the chi-square
method. The expected number of delayed F2
progeny in each cross was calculated based
on the proportion observed in the control cross
(NIL × AF16). For maternal backcrosses, crosses
were performed as above except that L4-stage
F1s were transferred to a new 3-cm plate and

Widen et al., Science 380, eade0705 (2023) 30 June 2023 10 of 14

RES EARCH | R E S E A R C H A R T I C L E
mated with at least 50 males of the appropri-
ate parental strain overnight, and then singled
in the morning to lay fresh eggs for analysis.
Paternal backcrosses were performed by trans-
ferring at least 50 F1 young males to a new 3-cm
plate, left overnight to ensure they reached
sexual maturity, and then at least 10 L4-stage
hermaphrodites of the appropriate parental
strain were added. Worms were allowed to mate
overnight, and then the hermaphrodites were
singled in the morning and fresh eggs were
picked for analysis. Genotyping PCRs were
performed with proteinase K (ProK)–digested
worm lysates as template using in-house hot-
start Taq polymerase (IMBA’s Molecular Bi-
ology Service Core Facility) with an annealing
temperature of 56°C. ProK digestion was per-
formed for 30 min at 37°C with 200 ug/ml
enzyme diluted in Taq buffer, and then 1 ml
was used directly in genotyping PCRs. In all
crosses where we observed F2 delay, all F2
progeny were genotyped regardless of their
phenotype by PCR using template from ProK
digestion of either the F2 worm itself or at
least 10 pooled F3s. The genotypes for all rele-
vant crosses can be found in table S1. In crosses
where we observed no delay, all F1 mothers and
at least 40 F2 s were genotyped from at least
two different confirmed heterozygous F1 moth-
ers to ensure that we observed the expected
Mendelian ratios of all genotypes. Where ap-
plicable, the genotyping of all crosses and
lines used in this study were performed with
either amplified fragment length polymor-
phism (AFLP) assays or Sanger sequencing. A
list of all primers and genotyping assays can be
found in table S5.
Short-read Illumina whole-genome sequencing
We extracted genomic DNA (gDNA) from at
least two freshly starved 9-cm plates of the
appropriate worm strain using the Master-
Pure Complete DNA and RNA Purification kit
(Lucigen; catalog no. MC85200) following man-
ufacturer protocols, including the sample
freeze-thaw step to facilitate tissue lysis. We
then prepared Illumina sequencing libraries
using the Nextera DNA Flex Library Prep kit
(Illumina; catalog no. 20018705) with 100 ng
of input gDNA for each library. gDNA and
prepared libraries were quantified using the
DeNovix dsDNA High Sensitivity kit (Biozym;
catalog no. 31DSDNA-HI1). Libraries were
checked on a fragment analyzer and then
sequenced on an Illumina NextSeq or MiSeq
instrument. All sequencing runs used in this
study are listed in table S6.
RNA extraction and RNA sequencing
For AF16 and HK104 mixed-stage embryo
RNA sequencing (RNA-seq), three full, non-
starved 9-cm plates of the appropriate worm
strain were first bleached to isolate only the
embryos, and then total RNA was extracted
Widen et al., Science 380, eade0705 (2023) 30 June 2023
from them using a TRIzol extraction method
based on a previously published protocol (78)
with the following modifications. Chloroform
was used instead of bromochloropropane, and
phase-lock tubes were used during TRIzol/
chloroform steps. DNA was removed from the
samples with a DNase I digestion (30 min at
37°C), and then the enzyme was inactivated
with 2.5 ml of 25 mM EDTA and a 10-min in-
cubation at 65°C. Next, 100 ng of purified RNA
was used as input for the TruSeq Stranded
mRNA kit (Illumina; catalog no. 20020595), and
libraries were prepared according to manu-
facturer instructions. Libraries were then checked
on a fragment analyzer and sequenced on a 2 × 75
paired-end NextSeq550 lane. All samples were
run in biological triplicate.
Single-molecule nanopore sequencing
High-molecular-weight DNA was extracted
using either the MasterPure Complete DNA
and RNA Purification kit (Lucigen; catalog
no. MC85200) or by standard phenol/chloroform
extraction. Worms were washed off of freshly
starved plates and rinsed with M9 at least
three times before the pellet was frozen at –20
to facilitate sample lysis before the extraction
protocol was started. Either 10-kb (C. briggsae
strains) or unfragmented (C. plicata SB355)
sequencing libraries were prepared using the
1D Ligation Sequencing Kit from Oxford
Nanopore Technologies (ONT; catalog no.
SQK-LSK109) and sequenced on a PromethION
device (ONT). Libraries for C. briggsae HK104
and C. plicata SB355 were run individually,
whereas C. briggsae ED3036, JU439, and
QR24 were barcoded using the Native Barcod-
ing kit (ONT, catalog no. EXP-NBD104) and
sequenced together. Read calling and demul-
tiplexing (where applicable) was performed
with the GPU version of Guppy Basecalling
software v5.0.11+2b6dbffa5 and the base-calling
model dna_r9.4.1_450bps_sup_prom.cfg.
Genome assembly and annotation
Raw nanopore reads were error corrected
using NECAT v0.0.1 with an estimated genome
size of 105 Mb. The longest corrected reads
were taken so that the estimated coverage of
the whole genome would be 50×. These reads
were then trimmed by Canu v1.8 and assembled
using Flye v2.7.1 with default parameters
(79, 80). The resulting assembly was polished
by two to four rounds of Racon v1.4.2. Scaffolds
corresponding to residual E. coli contamination
were identified and removed using BLAST
v2.2.26. The assemblies of C. briggsae strains
QR24, ED3036, and JU439 consist of 37,
42, and 72 scaffolds; have N50s of 7,629,715,
6,649,622, and 8,765,825; and have BUSCO
completeness of 98.1%, 98.2%, and 98.7%,
respectively. Like HK104 (Japan), the strains
JU439 (Iceland) and VX34 (China) contain the
msft-1 TA in Chr. III. We have not crossed
these strains to AF16, but it is likely that their
respective TAs are active. JU439 MSFT-1 is
identical to HK104 MSFT-1 at the protein level.
VX34 MSFT-1 has five substitutions compared
with HK104 MSFT-1, but these changes do not
affect the protease catalytic residues nor the
putative substrate-binding site Glu273 residue.
For the C. plicata assembly, Flye was run in
the metagenomic mode because of extensive
bacterial contamination of the library. Circular
contigs with coverage more than twice higher
than coverage of true C. plicata contigs were
removed. True contigs were defined as nearly
chromosome-sized contigs showing sequence
homology to the existing C. plicata short-read
assembly. Genome annotation was performed
using funannotate v1.7.4, with a transcriptome
assembled by Trinity v2.10.0 (81, 82). For
funannotate, Augustus was run with a pre-
trained C. elegans model, and BUSCO was run
with–busco_db nematoda–busco_seed_species
Caenorhabditis options. GeneMark-ET was run
with default parameters.
Fine mapping of the HK104 toxin
We began mapping the HK104 TA by per-
forming 16 generations of maternal backcrosses
to the susceptible strain AF16, taking advantage
of the ability of TAs to select for themselves in
each generation. For each backcross, five to 10
L4-stage hermaphrodites were crossed with at
least 25 AF16 males in a 3-cm plate. The 16th
generation was Illumina short-read sequenced
(see “Short-read Illumina whole-genome se-
quencing” section and table S6) and we per-
formed single-nucleotide polymorphism (SNP)–
calling analysis to calculate allele frequencies
across the genome (see “SNP-calling analysis”
section). This allowed us to identify a single
large introgression from HK104 on Chr. III (~0
to 10 Mb; Fig. 1B) in an otherwise AF16 back-
ground. We then isolated 50 independent
NILs from the 16× backcrossed generation by
picking individual L4-stage hermaphrodites
and allowing them to self-fertilize. The intro-
gressions carried by each NIL were roughly
mapped using a combination of previously
published (29) and newly identified marker
indels (table S5) across Chr. III that had a large
enough size difference in the AF16 and HK104
genomes to be visualized by PCR and gel
electrophoresis. We identified two particularly
useful NILs (NIL1 and NIL2) that had partially
overlapping introgressions, but both carried
the HK104 TA (fig. S1, A to C). The exact mar-
gins of the overlap between NIL1 and NIL2
were determined by Illumina short-read se-
quencing, alignment of each dataset to the
AF16 genome, and manual inspection of SNPs
using Integrative Genomics Viewer (Chr. III:
4.94 to 6.75 Mb in AF16 coordinates). NIL1 was
used for all further experiments and carries a
3.24-Mb introgression (Chr. III: 3.51 to 6.75 Mb;
fig. S1B). To further narrow the candidate region,
11 of 14
RES EARCH | R E S E A R C H A R T I C L E
we analyzed existing AF16 × HK104 RILs that
had different recombination breakpoints across
the introgression (Fig. 1C) by genotyping them
for the same Chr. III markers and also crossing
them to AF16 and analyzing their F2 progeny
for developmental delay to determine whether
they carried the HK104 TA. We then used our
HK104 de novo assembled genome (see “Genome
assembly and annotation” section) to establish
a list of predicted genes within the overlap
region, and narrowed the list of candidates
first based on their gene expression during
embryonic development using our RNA-seq
data from both HK104 and AF16 (see “RNA
extraction and RNA-seq” section). After filter-
ing for genes that were expressed in HK104
and not expressed in AF16 (or, if expressed,
highly divergent), we then searched for pairs
of candidates that were immediate neighbors
or in tight linkage to each other (table S2). Only
one pair of genes, ORF14767 and ORF14768,
satisfied all criteria. Because of its high expres-
sion level, ORF14767 was subsequently targeted
with CRISPRs to generate knockout alleles
(see “Generation of C. briggsae near-isogenic
and mutant lines” section). ORF14767 mutants
were tested in genetic crosses as described
previously (23) to confirm that ORF14767 is the
toxin-encoding gene. All mutant lines are de-
scribed in detail in table S3.
SNP-calling analysis
The reads were aligned to the reference
genome using SAMtools v1.10. Duplicates
were removed using sambamba v0.6.6. Var-
iants were called using GATK v4.0.1.2. Var-
iants were quality filtered with a threshold of
QUAL = 200. Fractions of reads supporting
one or another allele were estimated using
custom R script. All scripts are available upon
request.
Annotation of MULE transposases and
phylogenetic reconstruction
For the annotation of MULE transposase
catalytic domains, we used OrthoFinder v2.5.4
with default settings (83). As an input, we used
a combination of genomes generated in this
study and those available in two public data-
bases: Caenorhabditis Genomes Project (V1)
and Wormbase Parasite (WBPS16). After ob-
taining the MULE transposase orthogroup, we
exclusively kept the MULE catalytic domains,
and sequences below 100 amino acids were
removed. After filtering, we obtained 161 MULE
catalytic transposase sequences. Then, the
sequences were aligned with MAFFT v.7.427
with default options (84). With the purpose of
removing spurious sequences, we used trimAL
(85) with the following options: -resoverlap
0.75 -seqoverlap 80. To build the phylogeny,
we used IQ-TREE v.1.6.12 (86) with the follow-
ing parameters: -st AA -m TESTNEW -bb 1000
-alrt 1000. The phylogenetic tree was colored
Widen et al., Science 380, eade0705 (2023) 30 June 2023
using the interactive Tree Of Life (iTOL soft-
ware v6.5.7) (87).
De novo protein structure prediction
and analyses
We predicted protein structures using Alpha-
Fold2 (33) and RoseTTAfold. AlphaFold2 was
run using the ColabFold AlphaFold2_advanced
notebook implementation in Google Colabor-
atory (88). For multiple sequence alignment,
we selected the mmseqs2 option. Models were
ranked based on pLDDT, and only the best
ranked out of five models was selected for fur-
ther analysis. We searched for structurally ho-
mologous proteins in PDB using the DALI server
(37). Graphics were generated using PyMol
(The PyMOL Molecular Graphics System v2.5,
Schrödinger) Protein alignment statistics for
the selected pairs were generated with TM-
align (89) and PyMol using the super algorithm
for protein pairs with high sequence identity
and cealign for protein pairs with low sequence
identity. We estimated and plotted the evolu-
tionary conservation of MSFT-1 residues using
the ConSurf server (90).
Horizontal gene transfer analysis of msft-1
We first obtained genome-wide single-copy
orthologs from C. briggsae HK104, C. plicata,
C. parvicauda, and C. auriculariae using
OrthoFinder v2.5.4 (83). We selected these
three additional species because they carried
proteases that were most closely related to
msft-1. The sequences for each pair of ortho-
logs were codon aligned using MACSE v2.04
software (91). To discriminate between ver-
tical and horizontal gene transfer, we then
used the vertical and horizontal inheritance
consistence analysis (VHICA) method, which
is implemented in the vhica R package (55).
Annotation of WOSP proteases and KRMA
kinases and phylogenetic reconstruction
For the annotation of WOSP proteases and
KRMA kinases protein families, we used
OrthoFinder v2.5.4 with default settings
(83). As an input, we used a combination of
genomes generated in this study and those
available in Caenorhabditis Genomes Project
V1 and V2 and WBPS16. Proteins identified
were randomly assigned a WOSP and KRMA
number in each species irrespective of orthology
between species (n = 235 total wosp genes, n =
531 total krma genes). Given the modularity of
the N- and C-terminal domains of WOSP and
KRMA proteins, the N-terminal ubiquitin-like
domains were removed from sequences and
phylogenetic reconstruction was performed on
only the remaining protease or kinase domain-
containing peptides, respectively (n = 182
WOSP proteins, n = 451 KRMA proteins).
Sequences with fewer than 100 amino acids
were removed. Orthogroups for each family
were then aligned with MAFFT v.7.427 with
default options (84). To increase the phyloge-
netic signal from our protein alignment, we
used trimAl v.1.4.1 to remove spurious sequen-
ces from the input alignment (85). trimAL was
run with the following options for the WOSP
family: -resoverlap 0.75 -seqoverlap 80. We then
used IQ-TREE v.1.6.12 to build the protein
phylogenetic tree with the following param-
eters: -st AA -m TESTNEW -bb 1000 -alrt
1000 (86, 92, 93). The phylogenies were
colored using the iTOL v6.5.7 (87). Two outlier
sequences (Cdol-WOSP-9 and Coiwi-WOSP-1)
with particularly long branch lengths were
manually removed from the WOSP protease
tree in Fig. 5A to better visualize the rest of
the phylogeny.
Annotation of Mavericks and
phylogenetic reconstruction
We used tblastn (blast+ v.2.8.1) to identify
highly conserved Maverick genes in the genome
of all nematode species used in this study (94).
We used as a query the following genes: ATPase,
fusogen, integrase, major capsid, minor capsid,
and PolB. We filtered out tblastn hits with an
e-value above 0.0001 and those genes that were
shorter (<60%) than the consensus Maverick
gene [except for protein-primed type B DNA
polymerase (pPolB) Maverick genes, sequen-
ces below 500 bp length were filtered out].
We aligned sequences with MUSCLE v.3.8.31
(default options), and IQ-TREE v.1.6.12 was
used to build the protein phylogenetic tree
(86, 92, 93). Unlike Mavericks that are inte-
grated in the genome of their host (Fig. 5B),
the C. bovis Maverick was found in a 17,366-bp
scaffold that corresponds to a circular DNA
sequence. This Maverick has only one TIR,
which connects the 3′ end of the pPolb gene
with the 3′ end of its cargo gene. This circular
DNA could represent the genome of a virus-
like particle after excision and recombination
of the TIRs that flank the Maverick or it could
be an intermediate state of transposition.
Genomic association of Mavericks with wosp
and krma gene families
We extracted the krma and wosp genomic
coordinates for every sequence found in each
of the two orthogroups. We did the same for
the Maverick genes annotated in the previous
step. wosp and krma genes considered asso-
ciated with a Maverick when they were <6 kb
upstream or downstream from a Maverick
gene. To estimate the significance of this as-
sociation, we built a null hypothesis by sim-
ulating random coordinates across nematode
genomes. For each iteration (n = 10,000), we
randomly generated genomic coordinates
(1500 bp in length) for each nematode species
carrying at least one wosp or krma gene. The
number of genomic coordinates picked for
each nematode species corresponds to the
number of wosp or krma genes found in that
12 of 14
RES EARCH | R E S E A R C H A R T I C L E
species. We used bedtools v.2.27.1 (95) to ob-
tain random coordinates (bedtools random -l
1500 -n ${number_interval} -g ${genome}).
Then, we tested whether these random regions
were associated with Mavericks, as we did for
wosp and krma gene families. P values were
estimated by comparing observed frequency
and expected frequency based on the 10,000
simulations. To compare terminal branch length
between wosp and krma genes that were as-
sociated or not with Mavericks, we obtained
these values from the estimated protein phy-
logenies using the R package ape (96).
Comparison of GC content and codon usage
We concatenated the coding sequences of all
nematode host genes, wosp and krma genes,
and ORFs of Maverick genes of the following
species: C. briggsae (HK104), C. plicata,
C. parvicauda, and C. auriculariae. The se-
quences were loaded in R with the package
Biostrings (97). For the codon usage bias
analysis, we used the synonymous codon
usage orderliness (SCUO) statistic implemented
in the R package coRdon (98), and for the GC
content analysis, a homemade script was used.
To show the different GC content of Maverick
genes and Maverick cargo genes with respect
to the rest of the genome, we estimated the GC
content in a sliding window manner (window
size of 600 bp) in two nematodes, C. plicata
and C. wallacei, in a genomic region (~100 kb)
in which a complete Maverick is located (Fig. 5C).
For C. auriculariae and O. tipulae, we extracted
a genomic region of ~500 kb (fig. S22).
Genomic alignments
Sequence alignments of genomic regions
containing MULEs and Mavericks were
generated using Minimap2 (99). We obtained
pairwise sequence alignments using the fol-
lowing parameters: -X -N 50 -p 0.1 -c. Final
plots were generated using the gggenomes R
package (https://github.com/thackl/gggenomes).
RE FE RENCES AND N OT ES
1. B. J. Arnold, I.-T. Huang, W. P. Hanage, Horizontal gene
transfer and adaptive evolution in bacteria. Nat. Rev. Microbiol.
20, 206–218 (2022). doi: 10.1038/s41579-021-00650-4;
pmid: 34773098
2. J. B. Clark, W. P. Maddison, M. G. Kidwell, Phylogenetic
analysis supports horizontal transfer of P transposable
elements. Mol. Biol. Evol. 11, 40–50 (1994). pmid: 8121285
3. E. A. Gladyshev, M. Meselson, I. R. Arkhipova, Massive
horizontal gene transfer in bdelloid rotifers. Science 320,
1210–1213 (2008). doi: 10.1126/science.1156407; pmid: 18511688
4. L. A. Graham, P. L. Davies, Horizontal gene transfer in
vertebrates: A fishy tale. Trends Genet. 37, 501–503 (2021).
doi: 10.1016/j.tig.2021.02.006; pmid: 33714557
5. J. Van Etten, D. Bhattacharya, Horizontal gene transfer in
eukaryotes: Not if, but how much? Trends Genet. 36, 915–925
(2020). doi: 10.1016/j.tig.2020.08.006; pmid: 33012528
6. D. A. Fitzpatrick, Horizontal gene transfer in fungi. FEMS
Microbiol. Lett. 329, 1–8 (2012). doi: 10.1111/j.1574-
6968.2011.02465.x; pmid: 22112233
7. H.-H. Zhang, J. Peccoud, M.-R.-X. Xu, X.-G. Zhang, C. Gilbert,
Horizontal transfer and evolution of transposable elements in
vertebrates. Nat. Commun. 11, 1362 (2020). doi: 10.1038/
s41467-020-15149-4; pmid: 32170101
Widen et al., Science 380, eade0705 (2023) 30 June 2023
8. A. Suh et al., Ancient horizontal transfers of retrotransposons
between birds and ancestors of human pathogenic nematodes.
Nat. Commun. 7, 11396 (2016). doi: 10.1038/ncomms11396;
pmid: 27097561
9. L. Gasmi et al., Horizontally transmitted parasitoid killing factor
shapes insect defense to parasitoids. Science 373, 535–541
(2021). doi: 10.1126/science.abb6396; pmid: 34326235
10. S. M. Soucy, J. Huang, J. P. Gogarten, Horizontal gene transfer:
Building the web of life. Nat. Rev. Genet. 16, 472–482 (2015).
doi: 10.1038/nrg3962; pmid: 26184597
11. S. B. Daniels, K. R. Peterson, L. D. Strausbaugh, M. G. Kidwell,
A. Chovnick, Evidence for horizontal transmission of the P
transposable element between Drosophila species. Genetics
124, 339–355 (1990). doi: 10.1093/genetics/124.2.339;
pmid: 2155157
12. C. Gilbert, R. Cordaux, Viruses as vectors of horizontal transfer
of genetic material in eukaryotes. Curr. Opin. Virol. 25, 16–22
(2017). doi: 10.1016/j.coviro.2017.06.005; pmid: 28672159
13. I. R. Arkhipova, I. A. Yushenova, Giant transposons in
eukaryotes: Is bigger better? Genome Biol. Evol. 11, 906–918
(2019). doi: 10.1093/gbe/evz041; pmid: 30796812
14. R. E. Benveniste, G. J. Todaro, Evolution of C-type viral genes:
Inheritance of exogenously acquired viral genes. Nature 252,
456–459 (1974). doi: 10.1038/252456a0; pmid: 4372546
15. N. A. T. Irwin, A. A. Pittis, T. A. Richards, P. J. Keeling,
Systematic evaluation of horizontal gene transfer between
eukaryotes and viruses. Nat. Microbiol. 7, 327–336 (2022).
doi: 10.1038/s41564-021-01026-3; pmid: 34972821
16. S. Schaack, C. Gilbert, C. Feschotte, Promiscuous DNA:
Horizontal transfer of transposable elements and why it
matters for eukaryotic evolution. Trends Ecol. Evol. 25,
537–546 (2010). doi: 10.1016/j.tree.2010.06.001;
pmid: 20591532
17. S. E. Schneider, J. H. Thomas, Accidental genetic engineers:
Horizontal sequence transfer from parasitoid wasps to their
Lepidopteran hosts. PLOS ONE 9, e109446 (2014). doi:
10.1371/journal.pone.0109446; pmid: 25296163
18. L. Gasmi et al., Recurrent Domestication by Lepidoptera of
Genes from Their Parasites Mediated by Bracoviruses. PLOS
Genet. 11, e1005470 (2015). doi: 10.1371/journal.
pgen.1005470; pmid: 26379286
19. C. Gilbert et al., Population genomics supports baculoviruses
as vectors of horizontal transfer of insect transposons. Nat.
Commun. 5, 3348 (2014). doi: 10.1038/ncomms4348;
pmid: 24556639
20. O. Piskurek, N. Okada, Poxviruses as possible vectors for
horizontal transfer of retroposons from reptiles to mammals.
Proc. Natl. Acad. Sci. U.S.A. 104, 12046–12051 (2007).
doi: 10.1073/pnas.0700531104; pmid: 17623783
21. E. Ben-David, A. Burga, L. Kruglyak, A maternal-effect selfish
genetic element in Caenorhabditis elegans. Science 356,
1051–1055 (2017). doi: 10.1126/science.aan0621;
pmid: 28495877
22. A. Burga, E. Ben-David, L. Kruglyak, Toxin-antidote elements
across the tree of life. Annu. Rev. Genet. 54, 387–415 (2020).
doi: 10.1146/annurev-genet-112618-043659; pmid: 32886546
23. E. Ben-David et al., Ubiquitous selfish toxin-antidote elements
in Caenorhabditis species. Curr. Biol. 31, 990–1001.e5 (2021).
doi: 10.1016/j.cub.2020.12.013; pmid: 33417886
24. J. A. Ross et al., Caenorhabditis briggsae recombinant inbred
line genotypes reveal inter-strain incompatibility and the
evolution of recombination. PLOS Genet. 7, e1002174 (2011).
doi: 10.1371/journal.pgen.1002174; pmid: 21779179
25. S. E. Baird, R. Stonesifer, Reproductive isolation in
Caenorhabditis briggsae: Dysgenic interactions between
maternal- and zygotic-effect loci result in a delayed
development phenotype. Worm 1, 189–195 (2012).
doi: 10.4161/worm.23535; pmid: 24058847
26. L. Velazco-Cruz, J. A. Ross, Genetic architecture and temporal
analysis of Caenorhabditis briggsae hybrid developmental
delay. PLOS ONE 17, e0272843 (2022). doi: 10.1371/
journal.pone.0272843; pmid: 35951524
27. Materials and methods are available as supplementary materials.
28. L. W. Hillier et al., Comparison of C. elegans and C. briggsae
genome sequences reveals extensive conservation of
chromosome organization and synteny. PLOS Biol. 5, e167
(2007). doi: 10.1371/journal.pbio.0050167; pmid: 17608563
29. D. C. Koboldt et al., A toolkit for rapid gene mapping in the
nematode Caenorhabditis briggsae. BMC Genomics 11, 236
(2010). doi: 10.1186/1471-2164-11-236; pmid: 20385026
30. A. D. Cresse, S. H. Hulbert, W. E. Brown, J. R. Lucas,
J. L. Bennetzen, Mu1-related transposable elements of maize
preferentially insert into low copy number DNA. Genetics 140,
315–324 (1995). doi: 10.1093/genetics/140.1.315;
pmid: 7635296
31. Q. H. Le, S. Wright, Z. Yu, T. Bureau, Transposon diversity in
Arabidopsis thaliana. Proc. Natl. Acad. Sci. U.S.A. 97,
7376–7381 (2000). doi: 10.1073/pnas.97.13.7376;
pmid: 10861007
32. D. Lisch, Mutator and MULE Transposons. Microbiol. Spectr. 3,
MDNA3-0032-2014 (2015). doi: 10.1128/microbiolspec.MDNA3-
0032-2014; pmid: 26104710
33. J. Jumper et al., Highly accurate protein structure prediction
with AlphaFold. Nature 596, 583–589 (2021). doi: 10.1038/
s41586-021-03819-2; pmid: 34265844
34. D. Kosek, A. B. Hickman, R. Ghirlando, S. He, F. Dyda,
Structures of ISCth4 transpososomes reveal the role of
asymmetry in copy-out/paste-in DNA transposition. EMBO J.
40, e105666 (2021). doi: 10.15252/embj.2020105666;
pmid: 33006208
35. M. Dupeyron, K. S. Singh, C. Bass, A. Hayward, Evolution of
Mutator transposable elements across eukaryotic diversity.
Mob. DNA 10, 12 (2019). doi: 10.1186/s13100-019-0153-8;
pmid: 30988700
36. L. Stevens et al., Chromosome-level reference genomes
for two strains of Caenorhabditis briggsae: An improved
platform for comparative genomics. Genome Biol.
Evol. 14, evac042 (2022). doi: 10.1093/gbe/evac042;
pmid: 35348662
37. L. Holm, P. Rosenström, Dali server: Conservation mapping in
3D. Nucleic Acids Res. 38, W545–W549 (2010). doi: 10.1093/
nar/gkq366; pmid: 20457744
38. T. Li et al., Molecular basis of the versatile regulatory
mechanism of HtrA-type protease AlgW from Pseudomonas
aeruginosa. mBio 12, e03299–e20 (2021). pmid: 33622718
39. H. Cho et al., Over-activation of a nonessential bacterial
protease DegP as an antibiotic strategy. Commun. Biol. 3, 547
(2020). doi: 10.1038/s42003-020-01266-9; pmid: 33005001
40. T. Krojer, M. Garrido-Franco, R. Huber, M. Ehrmann, T. Clausen,
Crystal structure of DegP (HtrA) reveals a new protease-
chaperone machine. Nature 416, 455–459 (2002).
doi: 10.1038/416455a; pmid: 11919638
41. T. Clausen, M. Kaiser, R. Huber, M. Ehrmann, HTRA proteases:
Regulated proteolysis in protein quality control. Nat. Rev. Mol.
Cell Biol. 12, 152–162 (2011). doi: 10.1038/nrm3065;
pmid: 21326199
42. S. Backert, S. Bernegger, J. Skórko-Glonek, S. Wessler,
Extracellular HtrA serine proteases: An emerging new strategy
in bacterial pathogenesis. Cell. Microbiol. 20, e12845 (2018).
doi: 10.1111/cmi.12845; pmid: 29582532
43. T. Clausen, C. Southan, M. Ehrmann, The HtrA family of
proteases: Implications for protein composition and cell fate.
Mol. Cell 10, 443–455 (2002). doi: 10.1016/S1097-2765(02)
00658-5; pmid: 12408815
44. P. Carter, J. A. Wells, Dissecting the catalytic triad of a serine
protease. Nature 332, 564–568 (1988). doi: 10.1038/
332564a0; pmid: 3282170
45. J. Völk, Zoologische Jahrbücher Abteilung für Systematik,
Geographie und Biologie der Tiere (G. Fischer, 1950).
46. L. Stevens et al., Comparative genomics of 10 new
Caenorhabditis species. Evol. Lett. 3, 217–236 (2019). doi:
10.1002/evl3.110; pmid: 31007946
47. K. Kiontke et al., Caenorhabditis phylogeny predicts
convergence of hermaphroditism and extensive intron loss.
Proc. Natl. Acad. Sci. U.S.A. 101, 9003–9008 (2004).
doi: 10.1073/pnas.0403094101; pmid: 15184656
48. M. Dayi et al., Additional description and genome analyses of
Caenorhabditis auriculariae representing the basal lineage of
genus Caenorhabditis. Sci. Rep. 11, 6720 (2021). doi: 10.1038/
s41598-021-85967-z; pmid: 33762598
49. N. Kanzaki et al., Biology and genome of a newly discovered
sibling species of Caenorhabditis elegans. Nat. Commun. 9, 3216
(2018). doi: 10.1038/s41467-018-05712-5; pmid: 30097582
50. E. Anderson, Introgressive hybridization. Biol. Rev. Camb.
Philos. Soc. 28, 280–307 (1953). doi: 10.1111/j.1469-185X.1953.
tb01379.x
51. K. C. Kiontke et al., A phylogeny and molecular barcodes for
Caenorhabditis, with numerous new species from rotting fruits.
BMC Evol. Biol. 11, 339 (2011). doi: 10.1186/1471-2148-11-339;
pmid: 22103856
52. R. A. Folk, P. S. Soltis, D. E. Soltis, R. Guralnick, New prospects
in the detection and comparative analysis of hybridization in
the tree of life. Am. J. Bot. 105, 364–375 (2018). doi: 10.1002/
ajb2.1018; pmid: 29683488
53. M. Ravenhall, N. Škunca, F. Lassalle, C. Dessimoz,
Inferring horizontal gene transfer. PLOS Comput. Biol. 11,
13 of 14
RES EARCH | R E S E A R C H A R T I C L E
e1004095 (2015). doi: 10.1371/journal.pcbi.1004095;
pmid: 26020646
54. M. Goodman, J. Czelusniak, G. W. Moore,
A. E. Romero-Herrera, G. Matsuda, Fitting the gene lineage into
its species lineage, a parsimony strategy illustrated by
cladograms constructed from globin sequences. Syst. Zool. 28,
132 (1979). doi: 10.2307/2412519
55. G. L. Wallau, P. Capy, E. Loreto, A. Le Rouzic, A. Hua-Van,
VHICA, a new method to discriminate between vertical and
horizontal transposon transfer: application to the Mariner
family within Drosophila. Mol. Biol. Evol. 33, 1094–1109 (2016).
doi: 10.1093/molbev/msv341; pmid: 26685176
56. E. J. Pritham, T. Putliwala, C. Feschotte, Mavericks, a novel
class of giant transposable elements widespread in eukaryotes
and related to DNA viruses. Gene 390, 3–17 (2007).
doi: 10.1016/j.gene.2006.08.008; pmid: 17034960
57. V. V. Kapitonov, J. Jurka, Self-synthesizing DNA transposons in
eukaryotes. Proc. Natl. Acad. Sci. U.S.A. 103, 4540–4545
(2006). doi: 10.1073/pnas.0600833103; pmid: 16537396
58. M. Krupovic, E. V. Koonin, Polintons: A hotbed of eukaryotic virus,
transposon and plasmid evolution. Nat. Rev. Microbiol. 13,
105–115 (2015). doi: 10.1038/nrmicro3389; pmid: 25534808
59. M. Krupovic, D. H. Bamford, E. V. Koonin, Conservation of major
and minor jelly-roll capsid proteins in Polinton (Maverick)
transposons suggests that they are bona fide viruses. Biol. Direct
9, 6 (2014). doi: 10.1186/1745-6150-9-6; pmid: 24773695
60. C. M. Bellas, R. Sommaruga, Polinton-like viruses are abundant
in aquatic ecosystems. Microbiome 9, 13 (2021). doi: 10.1186/
s40168-020-00956-0; pmid: 33436089
61. G. J. Starrett et al., Adintoviruses: A proposed animal-tropic
family of midsize eukaryotic linear dsDNA (MELD) viruses.
Virus Evol. 7, veaa055 (2021). doi: 10.1093/ve/veaa055;
pmid: 34646575
62. T. Klose et al., Structure of faustovirus, a large dsDNA virus.
Proc. Natl. Acad. Sci. U.S.A. 113, 6206–6211 (2016).
doi: 10.1073/pnas.1523999113; pmid: 27185929
63. G. Borgonie et al., Nematoda from the terrestrial deep
subsurface of South Africa. Nature 474, 79–82 (2011).
doi: 10.1038/nature09974; pmid: 21637257
64. D. J. Weinstein et al., The genome of a subterrestrial nematode
reveals adaptations to heat. Nat. Commun. 10, 5268 (2019).
doi: 10.1038/s41467-019-13245-8; pmid: 31754114
65. J. G. N. Barreat, A. Katzourakis, Phylogenomics of the Maverick
virus-like mobile genetic elements of vertebrates. Mol. Biol.
Evol. 38, 1731–1743 (2021). doi: 10.1093/molbev/msaa291;
pmid: 33481003
66. D. J. Kim et al., Helicobacter pylori proinflammatory protein up-
regulates NF-kB as a cell-translocating Ser/Thr kinase. Proc.
Natl. Acad. Sci. U.S.A. 107, 21418–21423 (2010). doi: 10.1073/
pnas.1010153107; pmid: 21098302
67. M. Pérez-Losada, M. Arenas, J. C. Galán, F. Palero,
F. González-Candelas, Recombination in viruses: Mechanisms,
methods of study, and evolutionary consequences. Infect.
Genet. Evol. 30, 296–307 (2015). doi: 10.1016/
j.meegid.2014.12.022; pmid: 25541518
68. E. W. Ganko, K. T. Fielman, J. F. McDonald, Evolutionary history
of Cer elements and their impact on the C. elegans genome.
Genome Res. 11, 2066–2074 (2001). doi: 10.1101/gr.196201;
pmid: 11731497
69. A. A. Vogan et al., The Enterprise, a massive transposon
carrying Spok meiotic drive genes. Genome Res. 31, 789–798
(2021). doi: 10.1101/gr.267609.120; pmid: 33875482
70. H. S. Malik, S. Henikoff, T. H. Eickbush, Poised for contagion:
Evolutionary origins of the infectious abilities of invertebrate
retroviruses. Genome Res. 10, 1307–1318 (2000). doi: 10.1101/
gr.145000; pmid: 10984449
71. M. Merchant et al., A bioactive phlebovirus-like envelope
protein in a hookworm endogenous virus. Sci. Adv. 8, eabj6894
(2022). doi: 10.1126/sciadv.abj6894; pmid: 35544562
72. M. G. Fischer, T. Hackl, Host genome integration and giant virus-
induced reactivation of the virophage mavirus. Nature 540,
288–291 (2016). doi: 10.1038/nature20593; pmid: 27929021
Widen et al., Science 380, eade0705 (2023) 30 June 2023
73. M. G. Fischer, C. A. Suttle, A virophage at the origin of large
DNA transposons. Science 332, 231–234 (2011). doi: 10.1126/
science.1199412; pmid: 21385722
74. M.-A. Félix, D. Wang, Natural viruses of Caenorhabditis
nematodes. Annu. Rev. Genet. 53, 313–326 (2019).
doi: 10.1146/annurev-genet-112618-043756; pmid: 31424970
75. J. G. N. Barreat, A. Katzourakis, A billion years arms-race
between viruses, virophages and eukaryotes. bioRxiv 440574
[Preprint] (2023); https://doi.org/10.1101/2021.04.20.440574.
76. Data and code from: S. A. Widen, I. Campo Bes, A. Koreshova,
P. Pliota, D. Krogull, A. Burga, Virus-like transposons cross
the species barrier and drive the evolution of genetic
incompatibilities, Zenodo (2023); https://doi.org/10.5281/
zenodo.7888571.
77. K. S. Ghanta, C. C. Mello, Melting dsDNA donor molecules
greatly improves precision genome editing in Caenorhabditis
elegans. Genetics 216, 643–650 (2020). doi: 10.1534/
genetics.120.303564; pmid: 32963112
78. M. R. Green, J. Sambrook, Molecular Cloning: A Laboratory
Manual (Cold Spring Harbor Laboratory, 2012).
79. S. Koren et al., Canu: Scalable and accurate long-read
assembly via adaptive k-mer weighting and repeat separation.
Genome Res. 27, 722–736 (2017). doi: 10.1101/gr.215087.116;
pmid: 28298431
80. M. Kolmogorov, J. Yuan, Y. Lin, P. A. Pevzner, Assembly of long,
error-prone reads using repeat graphs. Nat. Biotechnol. 37, 540–546
(2019). doi: 10.1038/s41587-019-0072-8; pmid: 30936562
81. B. J. Haas et al., De novo transcript sequence reconstruction
from RNA-seq using the Trinity platform for reference
generation and analysis. Nat. Protoc. 8, 1494–1512 (2013).
doi: 10.1038/nprot.2013.084; pmid: 23845962
82. J. M. Palmer, J. Stajich, Funannotate v1.8.1: Eukaryotic genome
annotation, Zenodo (2020); https://doi.org/10.5281/
zenodo.4054262.
83. D. M. Emms, S. Kelly, OrthoFinder: Phylogenetic orthology
inference for comparative genomics. Genome Biol. 20, 238
(2019). doi: 10.1186/s13059-019-1832-y; pmid: 31727128
84. K. Katoh, D. M. Standley, MAFFT multiple sequence alignment
software version 7: Improvements in performance and
usability. Mol. Biol. Evol. 30, 772–780 (2013). doi: 10.1093/
molbev/mst010; pmid: 23329690
85. S. Capella-Gutiérrez, J. M. Silla-Martínez, T. Gabaldón, trimAl:
A tool for automated alignment trimming in large-scale
phylogenetic analyses. Bioinformatics 25, 1972–1973 (2009).
doi: 10.1093/bioinformatics/btp348; pmid: 19505945
86. L.-T. Nguyen, H. A. Schmidt, A. von Haeseler, B. Q. Minh,
IQ-TREE: A fast and effective stochastic algorithm for
estimating maximum-likelihood phylogenies. Mol. Biol. Evol. 32,
268–274 (2015). doi: 10.1093/molbev/msu300;
pmid: 25371430
87. I. Letunic, P. Bork, Interactive Tree Of Life (iTOL) v5: An online
tool for phylogenetic tree display and annotation. Nucleic Acids
Res. 49, W293–W296 (2021). doi: 10.1093/nar/gkab301;
pmid: 33885785
88. M. Mirdita, K. Schütze, Y. Moriwaki, L. Heo, S. Ovchinnikov,
M. Steinegger, ColabFold - Making protein folding accessible to
all. bioRxiv 456425 [Preprint] (2022); https://doi.org/10.1101/
2021.08.15.456425.
89. Y. Zhang, J. Skolnick, TM-align: A protein structure alignment
algorithm based on the TM-score. Nucleic Acids Res. 33,
2302–2309 (2005). doi: 10.1093/nar/gki524;
pmid: 15849316
90. H. Ashkenazy et al., ConSurf 2016: An improved methodology to
estimate and visualize evolutionary conservation in
macromolecules. Nucleic Acids Res. 44, W344–W350 (2016).
doi: 10.1093/nar/gkw408; pmid: 27166375
91. V. Ranwez, S. Harispe, F. Delsuc, E. J. P. Douzery, MACSE:
Multiple Alignment of Coding SEquences accounting for
frameshifts and stop codons. PLOS ONE 6, e22594 (2011).
doi: 10.1371/journal.pone.0022594; pmid: 21949676
92. D. T. Hoang, O. Chernomor, A. von Haeseler, B. Q. Minh,
L. S. Vinh, UFBoot2: Improving the ultrafast bootstrap
approximation. Mol. Biol. Evol. 35, 518–522 (2018).
doi: 10.1093/molbev/msx281; pmid: 29077904
93. S. Kalyaanamoorthy, B. Q. Minh, T. K. F. Wong, A. von Haeseler,
L. S. Jermiin, ModelFinder: Fast model selection for accurate
phylogenetic estimates. Nat. Methods 14, 587–589 (2017).
doi: 10.1038/nmeth.4285; pmid: 28481363
94. C. Camacho et al., BLAST+: Architecture and applications. BMC
Bioinformatics 10, 421 (2009). doi: 10.1186/1471-2105-10-421;
pmid: 20003500
95. A. R. Quinlan, I. M. Hall, BEDTools: A flexible suite of utilities for
comparing genomic features. Bioinformatics 26, 841–842
(2010). doi: 10.1093/bioinformatics/btq033; pmid: 20110278
96. E. Paradis, K. Schliep, ape 5.0: An environment for modern
phylogenetics and evolutionary analyses in R. Bioinformatics
35, 526–528 (2019). doi: 10.1093/bioinformatics/bty633;
pmid: 30016406
97. H. Pagès, P. Aboyoun, R. Gentleman, S. DebRoy, “Biostrings”
(Bioconductor, 2022); https://bioconductor.org/packages/
Biostrings.
98. H. Elek, M. Kuzman, K. Vlahovicek, “BioinfoHR/coRdon”
(Github, 2022); https://github.com/BioinfoHR/coRdon.
99. H. Li, Minimap2: Pairwise alignment for nucleotide sequences.
Bioinformatics 34, 3094–3100 (2018). doi: 10.1093/
bioinformatics/bty191; pmid: 29750242
AC KNOWLED GME NTS
We thank E. Haag, S. Baird, and C. Braendle for strains; J. Ross
and K. Senti for discussions; and P. Duchek, B. Érdi, and
J. Gokcezade from the IMBA Fly & Worm facility for the design
and generation of transgenic worms. Library preparation and
sequencing were performed at the VBCF NGS Unit. We also thank
Life Science Editors for editing services. Funding: Research in the
Burga laboratory is supported by the Austrian Academy of
Sciences, the city of Vienna, a European Research Council (ERC)
starting grant under the European Union’s Horizon 2020 program
(grant ERC-2019-StG-851470), and the Austrian Science Fund
(FWF). S.A.W. is supported by the European Union’s Framework
Programme for Research and Innovation Horizon 2020
(2014‐2020) under Marie Curie Skłodowska grant 847548. I.C.B. is
supported by “la Caixa” Foundation (grant LCF/BQ/EU19/
11710050). A.K. is supported by a Boehringer Ingelheim Fonds
(BIF) PhD Fellowship. Author contributions: A.B. conceived and
supervised the project. S.A.W., I.C.B., and A.B. designed and
developed the project. S.A.W., P.P., and D.K. performed worm
genetics. S.A.W. and A.K. performed fine-mapping of TA. I.C.B.
performed HGT analyses, genome annotations, phylogenetics, and
TE characterization. A.K. performed Nanopore genome assembly
and annotations. A.B. annotated MULEs and Mavericks and
performed structural predictions. S.A.W. and A.B. wrote the
manuscript with input from I.C.B. Competing interests: The
authors declare no competing interests. Data and materials
availability: All materials generated for this study (worm strains,
plasmids, etc.) are available upon request from A.B. Code is
available at Zenodo (76). All raw sequencing data and genome
assemblies are available under National Center for Biotechnology
Information (NCBI) project no. PRJNA857754. License
information: Copyright © 2023 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.ade0705
Figs. S1 to S25
Table S1 to S6
Data S1 to S3
MDAR Reproducibility Checklist
View/request a protocol for this paper from Bio-protocol.
Submitted 22 July 2022; resubmitted 25 January 2023
Accepted 17 May 2023
10.1126/science.ade0705
14 of 14
RES EARCH
RESEARCH ARTICLE SUMMARY
SPATIAL OMICS
Spatially resolved single-cell translatomics
at molecular resolution
Hu Zeng†, Jiahao Huang†, Jingyi Ren†, Connie Kangni Wang, Zefang Tang, Haowen Zhou,
Yiming Zhou, Hailing Shi, Abhishek Aditham, Xin Sui, Hongyu Chen, Jennifer A. Lo, Xiao Wang*
INTRODUCTION: Cellular proteins are the final
products of gene expression that execute cel-
lular functions. To understand translational
gene regulation, it is crucial to measure protein
synthesis at single-cell and spatial resolution.
Current ribosome profiling methods achieve
transcriptome-wide translational analysis but
lack spatial context. By contrast, imaging-based
methods preserve spatial information but are
not multiplexed. A scalable method for single-
cell and spatially resolved profiling of protein
synthesis is still needed.
RATIONALE: We introduce ribosome-bound
mRNA mapping (RIBOmap), a highly multi-
plexed method for spatial charting of protein
synthesis at single-cell and subcellular resolu-
tion. RIBOmap employs a targeted-sequencing
strategy that selectively detects ribosome-bound
mRNAs through a specific design of tri-probes.
The tri-probe set consists of a splint DNA
probe hybridized to ribosomal RNA and a pair
RIBOmap for spatial trans-
latomics imaging. RIBOmap
is a three-dimensional (3D)
in situ profiling technology
designed to selectively mea-
sure ribosome-bound mRNAs,
providing spatially resolved
single-cell translatome analysis
at molecular resolution. In cell
culture, RIBOmap uncovers
cell-cycle-dependent and sub-
cellular localized translation.
When applied to intact mouse
brain tissue, RIBOmap gener-
ates spatial tissue atlases
and reveals cell-type–specific
and tissue region–specific
translational regulation.
Zeng et al., Science 380, 1337 (2023) 30 June 2023
◥ spatial resolutions of RIBOmap, we generated
of padlock and primer probes hybridized to
mRNAs, which together produce DNA ampli-
cons with gene-unique barcodes through in
situ amplification. These DNA amplicons are
then embedded in a polyacrylamide hydrogel
matrix and the gene-unique barcodes are read
out through in situ sequencing. We validated
the specificity of RIBOmap through the use of
noncoding RNAs, translation inhibitor, in vitro
transcribed mRNAs, and comparison with
established ribosome profiling technology.
RESULTS: We applied RIBOmap to HeLa cells
for a 981-gene multiplexed experiment and
designed a multimodal RIBOmap imaging
experiment to capture the cell-cycle phase and
subcellular organelles. The results revealed cell-
cycle–dependent translation and subcellular
localized mRNA translation in HeLa cells. We
further applied RIBOmap to intact mouse brain
tissue, measuring the translation of 5413 genes
simultaneously. Leveraging the single-cell and
a single-cell spatial map of the mouse brain. By
comparing the spatial translatomics generated
by RIBOmap with spatial transcriptomics, we
uncovered cell type–specific and brain region–
specific translational regulation.
CONCLUSION: RIBOmap presents a new spa-
tially resolved single-cell translatomics technol-
ogy, accelerating our understanding of protein
synthesis in the context of subcellular architec-
ture, cell types, and tissue anatomy. The pair-
wise spatial translatomic and transcriptomic
mapping enabled us to systematically identify
cell type- and tissue-region–specific translational
regulation, paving the way for uncovering novel
posttranscriptional gene regulation princi-
ples and mechanisms that shape the proteome
for cellular and tissue functions. RIBOmap by-
passes complicated polysome isolation steps
and genetic manipulation, making it promis-
ing for studies in post hoc human tissue and
disease samples. We anticipate that integrating
RIBOmap with other imaging-based measure-
ments will enable spatial multiomics mapping
for a comprehensive understanding of biolog-
ical systems.
▪
The list of author affiliations is available in the full article.
*Corresponding author. Email: xwangx@mit.edu
†These authors contributed equally to this work.
Cite this article as H. Zeng et al., Science 380, eadd3067
(2023). DOI: 10.1126/science.add3067
READ THE FULL ARTICLE AT
https://doi.org/10.1126/science.add3067
1 of 1
RES EARCH

◥ four experiments. First, RIBOmap and pre-

RESEARCH ARTICLE viously reported STARmap RNA imaging (20)
were used to detect protein-encoding ACTB
SPATIAL OMICS mRNA and two negative controls of noncod-
ing RNAs (Fig. 2A and table S1). As expected,
Spatially resolved single-cell translatomics RIBOmap only detected cytoplasmic ACTB
mRNAs whereas STARmap detected both
at molecular resolution ACTB mRNA and the noncoding RNAs (Fig. 2,
B and C and fig. S2). Second, RIBOmap spe-
Hu Zeng1,2†, Jiahao Huang1,2†, Jingyi Ren1,2†, Connie Kangni Wang2, Zefang Tang1,2, Haowen Zhou2, cificity was validated using Harringtonine,
Yiming Zhou1,2,3, Hailing Shi1,2, Abhishek Aditham2,4, Xin Sui1,2, Hongyu Chen1,2, a translation inhibitor that traps translation-
Jennifer A. Lo2, Xiao Wang1,2* initiating ribosomes at the start codon (21)
(Fig. 2D). After Harringtonine treatment, there
The precise control of messenger RNA (mRNA) translation is a crucial step in posttranscriptional was a significant decrease in RIBOmap signal
gene regulation of cellular physiology. However, it remains a challenge to systematically study mRNA intensity using probes targeting 115 or 405 nu-
translation at the transcriptomic scale with spatial and single-cell resolution. Here, we report the celotides (nt) downstream of the start codon,
development of ribosome-bound mRNA mapping (RIBOmap), a highly multiplexed three-dimensional in whereas there was minimal decrease in the
situ profiling method to detect cellular translatome. RIBOmap profiling of 981 genes in HeLa signal generated by the probe targeting the
cells revealed cell cycle–dependent translational control and colocalized translation of functional −16 nt from the start codon (Fig. 2, E to G).
gene modules. We mapped 5413 genes in mouse brain tissues, yielding spatially resolved single-cell Third, we used synthetic in vitro transcribed
translatomic profiles for 119,173 cells and revealing cell type–specific and brain region–specific (IVT) mRNAs and lipid-mediated transfec-
translational regulation, including translation remodeling during oligodendrocyte maturation. tion (Fig. 2H). Our data demonstrated that
Our method detected widespread patterns of localized translation in neuronal and glial cells in STARmap could detect both small puncta cor-
intact brain tissue networks. responding to free cytosolic mRNAs and larger
intracellular granules corresponding to lipid

M
easuring genome-wide protein synthe-
sis patterns with single-cell and spa-
tial resolution can help us understand
translational regulation in heteroge-
neous cell types and states. Although
large-scale single-cell and spatially resolved
proteome profiling remains challenging (1),
past research has focused on mapping mRNA
levels to infer the corresponding protein abun-
dances in single cells. However, numerous studies
have revealed poor correlation between mRNA
and protein levels (2–5), as gene regulation is
achieved at both transcriptional and transla-
tional levels. Moreover, many genes undergo
signal-dependent and subcellular-localized
translation (6). Therefore, we need scalable
single-cell and spatially resolved profiling of
protein synthesis for a comprehensive under-
standing of gene expression and translational
regulation.
Existing bulk and single-cell ribosome pro-
filing methods have enabled us to analyze pro-
tein translation at the transcriptome scale (7–12)
but cannot preserve spatial information. Cur-
rent imaging-based methods that can trace
mRNA translation with their physical coordi-
nates (13–19) are limited to low gene through-
put. Therefore, it remains challenging to achieve
highly multiplexed spatial ribosome profiling
in single cells with subcellular resolution. To
1
Department of Chemistry, Massachusetts Institute of
Technology, Cambridge, MA 02139, USA. 2Broad Institute of
MIT and Harvard, Cambridge, MA 02142, USA. 3Center for
Psychiatric Research, Broad Institute of MIT and Harvard,
Cambridge, MA 02142, USA. 4Department of Biological
Engineering, Massachusetts Institute of Technology,
Cambridge, MA 02139, USA.
*Corresponding author. Email: xwangx@mit.edu
†These authors contributed equally to this work.
fill this gap, we developed a three-dimensional
(3D) in situ ribosome-bound mRNA mapping
method (RIBOmap) for highly multiplexed char-
acterization of protein synthesis with single-
cell and subcellular resolutions.
RIBOmap design and validation
RIBOmap is built upon a targeted-sequencing
strategy in which a specific design of tri-probes
selectively detects and amplifies ribosome-
bound mRNAs (Fig. 1A and fig. S1A). The
RIBOmap tri-probe set includes: (i) a splint
DNA probe that hybridizes to ribosomal RNAs
(rRNAs) and serves as the splint to circularize
the adjacent padlock probe; (ii) a padlock
probe that targets specific mRNA species of
interest and encodes a gene-unique barcode;
(iii) a primer probe that targets the mRNA
site adjacent to the one targeted by the pad-
lock probe and serves as the primer for rolling
circle amplification resulting in a DNA nano-
ball (amplicon). DNA amplicon signals are only
produced when all three probes are present
in proximity (Fig. 1, B and C). The gene-unique
barcodes in the DNA amplicons are then de-
coded through in situ sequencing with error-
reduction by dynamic annealing and ligation
(Fig. 1A) (20). We also tested an alternative
RIBOmap workflow that uses primary anti-
bodies of ribosomal proteins (RPS3, RPL4)
and splint-conjugated protein A/G to target
ribosomes, which demonstrated a much lower
signal-to-noise ratio (SNR) than that of the
rRNA targeting strategy (fig. S1, B to G). Thus,
we proceeded with the rRNA targeting strat-
egy in the following studies.
To confirm that RIBOmap specifically tar-
gets ribosome-bound mRNAs, we designed
transfection vesicles containing many copies of
nontranslating mRNAs, whereas RIBOmap
signals were depleted in lipid transfection vesi-
cles (Fig. 2, I and J), indicating that RIBOmap
does not detect nontranslating mRNAs. Fi-
nally, we validated RIBOmap by comparing
it with RiboLace, an established ribosome pro-
filing technology (22). RIBOmap successfully
detected differentially translated mRNAs in
starved versus control human MCF7 cells,
which is consistent with the reported RiboLace
and proteomics measurements (22) (fig. S3).
Together, these results demonstrate the spe-
cificity of RIBOmap in detecting ribosome-
bound mRNAs.
Multiplexed RIBOmap in cultured cells
After benchmarking the SNR and specificity
of RIBOmap, we performed a highly multi-
plexed 981-gene RIBOmap experiment in HeLa
cells (Fig. 3A). The 981 genes represent a curated
list composed of cell-cycle gene markers, genes
with diverse subcellular RNA patterns, and
genes of varying RNA stabilities (23–26) (table
S2). We designed a multimodal RIBOmap im-
aging experiment that further incorporated
the information on cell-cycle stages and sub-
cellular organelles: (i) the cell-cycle phase
of each cell was captured by the fluorescent
ubiquitination-based cell cycle indicators (FUCCI)
(27, 28); (ii) ribosome-bound mRNAs (981 genes)
were in situ sequenced; (iii) finally, nuclei, en-
doplasmic reticulum, and cell shapes were
stained and imaged (Fig. 3B). We also con-
ducted a paired STARmap experiment for
comparison (Fig. 3A). In total, RIBOmap and
STARmap sequenced 1813 and 1757 cells, re-
spectively (fig. S4A). The median distance be-
tween each read and its nearest neighbor in

Zeng et al., Science 380, eadd3067 (2023) 30 June 2023 1 of 10

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

A
Ribosome-bound mRNA Non-translating RNA Tri-probes Lacking splint probe
Splint probe
Ribosome Hybridization hybridized
3' 5'P 3' to rRNA 3' 5'P 3'
5' 5'
(A)n (A)n (A)n (A)n

Ligation

DNA amplicon No DNA amplicon Circularized padlock probe Linear padlock probe
NH2 Rolling circle
3' amplification
3' 3' 5'P 3'
NH2 3' 5'P
NH2 5' 5'
5' (A)n (A)n
(A)n
NH2

Hydrogel-tissue chemistry
and SEDAL sequencing
Signal readout 5' Cap
AAGTTAC Translating gene 1
NNNAA
NN rcode TTGCACT Translating gene 2 18S rRNA
Ba Decode Annotate In situ single-cell
TTGCACT Translating gene 3 translatome with 3' Blocked splint probe
A C GT
… … subcellular localization
Barcoded padlock probe
A CGT

GTACTTA Translating gene n

Barcoded primer probe

B Tri-probes Without primer probe Without splint probe Incomplete splint probe, without C
1.5 **
rRNA hybridization part

(normalized to DAPI)
**
**

Signal intensity
1.0
(A)n (A)n (A)n (A)n

0.5

0.0

be

be
be

ob

ro

ro
ro

pr

tp

tp
i-p

er

in

lin
Tr

rim

pl

sp
ts
tp

e
ou

et
ou

ith

pl
ith

m
W
W

co
10 µm

In
DNA amplicon DAPI

Fig. 1. RIBOmap for in situ profiling of mRNA translation at subcellular resolu-
tion. (A) Schematic of RIBOmap. After the sample is prepared, a paired barcoded
padlock probe and primer probe are hybridized to a targeted intracellular RNA,
and the splint probe is hybridized to 18S rRNA of ribosomes. The splint probe is used
as a template for the proximity ligation to circularize the padlock probe. The
intact padlock probe can then be amplified to generate amine-modified DNA
amplicons in situ. Next, these DNA amplicons are copolymerized into hydrogel
through tissue-hydrogel chemistry for in situ sequencing. The gene-unique barcode
sequence (red) in the cDNA amplicons can then be read out through cyclic in situ
sequencing. (B) Tri-probe strategy. (Left) Fluorescent images of tri-probe condition
show the RIBOmap signal of ACTB mRNA in HeLa cells. (Middle) Fluorescent
images of negative control samples without the primer or splint probe show minimum
DNA amplicon signal. (Right) Fluorescent images of the control sample using a splint
probe without the rRNA hybridization sequence show minimal DNA amplicon signal.
(C) Quantification of the DNA amplicon signal intensity shown in (B). Error bars,
standard deviation. n = 3 images per condition. Student's t-test, **P < 0.01.

3D was 0.69 mm for RIBOmap and 0.61 mm for
STARmap (fig. S4, B and C), indicating that
both techniques have high spatial resolution.
To assess the reliability of RIBOmap in multi-
plexed experiments, we calculated the corre-
lation between the RIBOmap results for 981
genes with published HeLa proteome and
ribosome profiling datasets (26, 29, 30). The
results showed that the correlation between
RIBOmap and the proteome dataset is com-
parable to that between the ribosome profiling
dataset and the proteome dataset (fig. S4D).
Additionally, the correlation between RIBO-
map and the two ribosome profiling datasets
is comparable to the correlation between the
two ribosome profiling datasets (fig. S4E).
Next, we evaluated whether RIBOmap can
decipher cell cycle–dependent mRNA transla-
tion. The results showed that the G1, G1/S, and
G2/M cell cycle phases identified from both
RIBOmap and STARmap datasets agreed with
the expected cell cycle–dependent patterns of
FUCCI protein fluorescence (Fig. 3C and fig.
S5), demonstrating the accuracy of RIBOmap
in delineating cell states. We then identified
the differentially expressed genes (DEGs) across
cell cycle phases using RIBOmap and STARmap
data (fig. S6, A and B, and table S3). Most of the
RIBOmap DEGs overlapped with known cell
cycle–dependent genes (31) (fig. S6C). We also
identified DEGs detected in RIBOmap but not
in STARmap (fig. S6D). One example is NOL6
(nucleolar protein 6), whose translation was
significantly decreased from the G1 phase to
the G2/M phase as detected by both RIBOmap
and ribosome profiling (32), but whose overall
RNA level showed little change across cell cycle
phases as detected by STARmap and RNA-seq
(fig. S6, E to G).
Next, to leverage the single-cell resolution in
our dataset, we performed gene-expression co-
variation analysis and identified five coregulated
translation modules (RTMs) with substantial
intramodule correlation, each with enrichment
of distinct functional pathways (RTMs 1 to 5,
Fig. 3D, fig. S7, A to E, and table S4). RTM 3
and RTM 5 are enriched for G2/M and G1/S
cell cycle marker genes, respectively (Fig. 3D,

Zeng et al., Science 380, eadd3067 (2023) 30 June 2023 2 of 10

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

A B ACTB MALAT1 vtRNA1-1 C Normalized RIBOmap signal

mRNA (ACTB)
0.4 ****
(A)n

(normalized to STARmap signal)

****
STARmap 0.3

Signal intensity
Non-coding RNA (MALAT1 and vtRNA1-1)
0.2

STARmap RIBOmap
0.1

RIBOmap
RNA Ribosome-
bound RNA 0.0

TB

-1
AT

A1
AC
10 µm

AL

N
R
M

vt
DNA amplicon DAPI
D F Probe 1 Probe 2 Probe 3 G Normalized RIBOmap signal
(A)n 1.0 **
**

(normalized to no drug treatment)

Start
codon No drug 0.8
Harringtonine treatment treatment
(A)n

Signal intensity
0.6

0.4
E Probe 1 Probe 2 Probe 3
Harringtonine 0.2
(A)n
treatment
0.0
Start 10 µm

3
codon

e
ob

ob

ob
Pr

Pr

Pr
DNA amplicon DAPI
H I
Firefly mRNA STARmap Renilla mRNA STARmap DAPI Merge
Lipofectamine transfection

Ribosome-bound
transfected mRNA

10 µm
Non-translating
transfected mRNA
J
Lipofectamine Firefly mRNA RIBOmap Renilla mRNA STARmap DAPI Merge
vesicle

STARmap RIBOmap

Signal in No signal in
lipofectamine vesicles lipofectamine vesicles 10 µm

Fig. 2. RIBOmap specificity validation. (A) Schematic of RIBOmap signal
verification by targeting mRNA (ACTB) and noncoding RNA (MALAT 1 and
vtRNA1-1). (B) Fluorescent images show RNA detection results by STARmap and
translating RNA detection results by RIBOmap. (C) Quantification of the DNA
amplicon signal intensity shown in (B). Error bars, standard deviation. n = 5
images per condition. Student's t-test, ****P < 0.0001. (D) Schematic of
translational regulation by Harringtonine. (E) Three pairs of padlock and primer
probes targeting different sites of ACTB mRNA. (F) Fluorescent images show the
RIBOmap signal of 3 sets of probes targeted to different regions of ACTB mRNA
in HeLa cells before and after Harringtonine treatment. (G) Quantification of the
RIBOmap signal intensity shown in (F). Error bars, standard deviation. n = 3
images per condition. Student's t-test, **P < 0.01. (H) Schematic representation
of RIBOmap signal verification using transfected IVT mRNAs. (I) Fluorescent
images show RNA detection of transfected Firefly mRNA and Renilla mRNA by
STARmap. The lipofectamine vesicles are labeled by red arrows. (J) Fluorescent
images show RNA detection results of transfected Renilla mRNA by STARmap
and translating RNA detection of transfected Firefly mRNA by RIBOmap. The
lipofectamine vesicles are labeled as in (I).

right), and are negatively correlated (fig. S7F).
Moreover, RTM 2 contains genes encoding
protein translation machineries (fig. S7D);
these genes show a positive correlation with
RTM 5 (G1/S marker genes) and a negative
correlation with RTM 3 (G2/M marker genes)
(fig. S7G). This observation suggests that the
protein translation machinery is up-regulated
in the G1/S phase to support the demand for
protein products as cells enlarge and subcel-
lular organelles are replicated.
Subcellular localized translation has impli-
cations for signal transduction and proteome
organization (6). Using RIBOmap results, we
identified five colocalized translation modules
(LTMs) with highly correlated subcellular spa-
tial organizations and distinct functional en-
richment (LTMs 1 to 5, Fig. 3, E to G, fig. S8, A
to C, and table S4). Among them, LTMs 2, 4,
and 5 are enriched with a membrane protein
and secretion pathway (Fig. 3, F and G and

Zeng et al., Science 380, eadd3067 (2023) 30 June 2023 3 of 10

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

A B FUCCI In situ sequencing Organelle

RIBOmap of 981 genes
imaging Cycle 1 (MAX) Cycle1 (MAX) Cycle 1 (single frame) Cycle 2–6 (single frame) staining

Ribosome-bound RNA
Ribosome-bound 1 2 3 4 5 6
RNA mapping

Organelle
staining

1,813 HeLa/FUCCI cells Localized translation 50 µm 50 µm 10 µm

STARmap of 981 genes

1 2 3 4 5 6

RNA
RNA mapping

Organelle
staining
50 µm 50 µm 10 µm
X
1,757 HeLa/FUCCI cells RNA localization
mKO2 mAG DAPI XY view XY view Channel 1 2 3 4 DAPI Flamingo ConA
Y
Z DAPI
C Single-cell transcriptome Single-cell translatome D Single-cell translatome covariation RTM 3
(G2/M marker genes enriched)
G1 UBE2C -
G1/S - BIRC5
CDCA2 -
- CDK1
G2/M CENPF -
- CKAP5
CKS1B -
DC1

- CKS2
DNAJB1 -
- FAM83D
1 HJURP -
- HMGB2
Exp. KIF23 -
- MKI67
level NCAPD2 -
- NDC80
High NUF2 -
- NUSAP1
TACC3 -
- TOP2A

2 RTM 5
3 Low (G1/S marker genes enriched)
4 CDC6 - - DTL
DC1

Intensity r value GINS2 -

mKO2 0.2 - HNRNPF
MAZ - - MCM2
0.1
5 MCM5 - - MCM6
0.0
MSH6 - - PCNA
–0.1
mAG TMEM97 - - UHRF1
–0.2
DC2 DC2

E Ribosome-bound RNA subcellular colocalization RNA subcellular colocalization

F LTM 1 LTM 2 LTM 3 LTM 4 LTM 5
Cell division
Mitotic spindle
Chromosome, centromeric region
Integral component of plasma membrane
1 1 Plasma membrane
2 2 Cellular response to mechanical stimulus
Clusters

3 3 Polysomal ribosome
Cytosol
Membrane
p value Extracellular exosome
0.0
0.2 Endoplasmic reticulum lumen
0.4
4 4 0.6 0 2 4 0 2 4 0 2 4 0 5 10 15 0 2 4
0.8
5 5 1.0 –log10(FDR)

G LTM 3 LTM 4
H RIBOmap
I J Comparison with APEX-RIP dataset
TXN - AKAP1 - **** 100
- AP2B1 - ATP1A1
AP2S1 - ATP2B4 - ****
ER localized percentage (%)

- BTN3A2 - BFAR
CBX3 - CANX - - CHST14
- COX6C
Overlap percentage (%)

CRIP1 - CLU - - DYNC1H1 80 80

- CSTB ESYT1 -
CTNNA1 - - DDX3X - FAT1
DHX9 - FSTL1 - - GALNT2
- EEF2 GLG1 -
EIF3B - - GLT8D1 60 60
- GMNN HEXA -
GSTO1 - - HNRNPA2B1
- IGSF3
IQGAP1 -
MCM2 - - NCAPG LAMB1 -
- KTN1
NDUFS3 - - LAMC1 40
- PCBP1 LGALS3BP - - MET 40
PPM1G - - PRPF8 MOGS -
PSMD2 - - NOTCH2
- RPL36A PCYOX1 - - PLXNB2
RPS28 - - RPS3 PODXL - 20
20
- PRSS23
SLC16A5 - - TRAPPC1 PSEN1 -
- PXDN
TRMT112 - - YBX3 QSOX2 - - SERPINH1
SLC16A1 - 0
p value TFRC -
- SLC7A5 10 µm 0
- TIMP2
TOR1B -
LTM 3 LTM 4 ER
1.0 0.8 0.6 0.4 0.2 0.0
Nucleus Cell boundary

Fig. 3. RIBOmap simultaneously measures the subcellular translation of measurements along with corresponding protein fluorescence profiles of the
981 genes in human HeLa cells. (A) Schematic of RIBOmap detection in FUCCI cell-cycle markers (STARmap, bottom left; RIBOmap, bottom right).
HeLa FUCCI cells to measure localized mRNA translation. (B) Representative (D) Single-cell translatome covariation matrix showing the pairwise Pearson’s
images showing the sequential mapping of FUCCI fluorescence signal, cDNA correlation coefficients of the cell-to-cell variation, shown together with their
amplicons, and organelle staining in the same HeLa cell sample. (Left) the FUCCI averaged expression levels. Five strongly correlating blocks of coregulated
cell fluorescence imaging results for RIBOmap (upper) and STARmap (bottom). translation modules (RTMs) are indicated by the gray boxes in the matrix, with
(Middle) Representative images showing the maximum intensity projections RTMs 3 and 5 enlarged on the right. Cell cycle markers are highlighted in
(MAX) of the first sequencing cycle with zoom-in views of a representative cell the enlarged matrix. (E) Matrix of the pairwise colocalization P-values describing
and single-frame views of the representative cell across six sequencing cycles. the degree to which the reads of two genes tend to coexist in a sphere of a
(Right) magnified view of organelle staining of a representative cell. (C) Diffusion 3-μm radius in the same cell in RIBOmap results (left) and STARmap results
map embeddings of cell-cycle stage clusters determined using the single-cell (right), shown together with hierarchical clustering of these genes. The
expression profile for STARmap (upper left) and RIBOmap (upper right) STARmap matrix uses the same order of genes as the RIBOmap matrix. Five

Zeng et al., Science 380, eadd3067 (2023) 30 June 2023 4 of 10

RES EARCH | R E S E A R C H A R T I C L E
strongly correlating blocks of colocalized translation modules (LTMs) are
indicated by the gray boxes in the matrix. (F) Bar plots visualizing the most
significantly enriched gene ontology (GO) terms (maximum 3) in each of
the five LTMs. (G) Enlarged gene matrix of LTMs 3 and 4. (H) A representative
cell image showing the spatial distribution of RIBOmap signals of LTMs 3
fig. S8 A and B) and physically colocalize with
ER staining (Fig. 3, H and I and fig. S8, D to F),
suggesting they are ER-translated genes. In-
deed, 62.2 and 94.6% of LTM4 genes over-
lapped with the proximity ribosome profiling
dataset (8) and APEX-RIP dataset (33), respec-
tively (Fig. 3J, fig. S8D, and table S4), suggest-
ing the accuracy of RIBOmap for subcellular
spatial analysis. By contrast, LTMs 1 and 3
are enriched for genes encoding large protein
complexes of the mitotic spindle and trans-
lation machinery, respectively (Fig. 3, F and G,
and fig. S8, A and C). This observation sug-
gests that the subunits of large protein com-
plexes may be synthesized in spatial proximity
for efficient assembly. Finally, we explored
whether coregulated genes tend to be colo-
calized for translation and found that the five
RTMs also show strong subcellular colocaliza-
tion (fig. S8G). These results may imply that
functionally related gene groups can be coreg-
ulated through subcellular colocalized trans-
lation, possibly through shared regulatory RNA
elements or protein-protein interactions be-
tween nascent proteins.
RIBOmap in intact mouse brain tissue
We then applied RIBOmap to mouse brain
slices to reveal single-cell translatome profiles
in intact tissues. We mapped a targeted gene
list of 5413 genes (table S5) curated from pre-
viously published single-cell sequencing studies
of mouse cell atlas (34–40). We imaged two
biological replicates of the mouse hemisphere
coronal sections (60,481 cells for Rep1 and
58,692 cells for Rep2) containing multiple brain
regions (Fig. 4, A and B). We benchmarked
tissue RIBOmap data by comparing the spatial
translation pattern of well-known cell-type
marker genes and neurotransmitter genes with
corresponding in situ hybridization (ISH) im-
ages from the Allen Brain database (41) (Fig.
4C and fig. S9), where the comparison showed
consistent spatial patterns. The data quality of
RIBOmap is comparable to previous STARmap
PLUS data (42) (fig. S10A). Notably, we observed
a high correlation (Pearson r = 0.95) between
the two RIBOmap replicates (fig. S10B), validat-
ing the reproducibility of RIBOmap.
Encouraged by the benchmarking results,
we then adopted a hierarchical clustering strat-
egy (20, 36) to identify cell types (fig. S10, C to
E), resulting in 11 major cell types and 38 sub-
types (Fig. 4, D and E, fig. S10, C to E, and fig.
S11). The cross-reference analyses revealed
good correspondence between RIBOmap cell
clusters and published regional scRNA-seq
Zeng et al., Science 380, eadd3067 (2023) 30 June 2023
and 4 genes, overlaid on the ER and cell boundary. (I) Quantification of the
ER-localized percentage of genes of LTMs 3 and 4 versus all the detected
genes. Wilcoxon signed-rank test, ****P < 0.0001. (J) The overlap percentage
of LTM3 genes, LTM4 genes, and all 981 genes with ER-proximal RNAs
identified by APEX-RIP.
results of cell types (35) (fig. S10D). Based on
these cell typing results, we generated spa-
tial cell maps from the imaged hemibrain
region (Fig. 4, F and G and fig. S12). This spa-
tial cell-type map is consistent with previous
reports (36), demonstrating the potential of
RIBOmap single-cell translatomics to compre-
hensively identify diverse brain cell types and
regions.
Cell type–specific and brain
region–specific translational regulation
in mouse brain
To compare the RIBOmap results with spa-
tial transcriptomics, we conducted pairwise
STARmap on a brain slice adjacent to RIBOmap
sample followed by data integration for joint
analyses (Fig. 5A). We observed consistent
cell typing results between the two methods
with respect to gene expression, cell-type com-
position, and spatial distribution of cell types
(Fig. 5A and fig. S13). We also performed im-
munostaining of NeuN and GFAP proteins
during RIBOmap and STARmap sample prep-
aration and observed clear enrichment of NeuN
protein signal in neurons and GFAP protein
signal in astrocytes, respectively, in both sam-
ples (fig. S14), further supporting the accuracy
of the cell typing results.
Leveraging the single-cell and spatial reso-
lution of paired RIBOmap and STARmap data,
we analyzed the heterogeneity of translational
regulation across cell types and brain regions.
We determined the reads correlation of 5413
genes between the two datasets for individual
cell types or tissue regions (Fig. 5, B and C).
Lower correlation scores indicate a greater
degree of translational regulation, resulting in
disparities between the translatome and the
transcriptome. We found that non-neuronal
cell types, especially oligodendrocytes, have
lower correlation scores between translatome
and transcriptome than neuronal cell types
(Fig. 5B). Correspondingly, fiber tracts, the
brain region where oligodendrocytes are en-
riched, had the lowest correlation (Pearson r =
0.58) between the translatome and transcrip-
tome among the eight brain regions (Fig. 5C).
This suggests a previously undercharacterized
level of translational regulation in glial cells,
particularly in oligodendrocytes.
To further pinpoint translationally regulated
genes across various cell types, we performed
gene clustering using major cell-type–resolved
RIBOmap and STARmap profiles and identi-
fied 15 gene modules with distinct gene func-
tions and expression patterns (fig. S15, A and
B, and table S6). Given that the oligodendro-
cyte lineage showed the lowest correlation
between translatome and transcriptome among
major cell types (Fig. 5C), we focused on gene
modules that were highly expressed in oligo-
dendrocyte lineage cells (240 genes; fig. S15A
and table S6). Pseudotime trajectory analysis
revealed the differentiation path of the oligo-
dendrocyte lineage in our datasets (40), from
oligodendrocyte precursor cells (OPC) to oligo-
dendrocyte subtype 1 (OLG1), then to oligo-
dendrocyte subtype 2 (OLG2) (Fig. 5, D and E).
This is supported by the increased expression
gradients of genes involved in myelination
(Plp1 and Mbp) along oligodendrocyte matu-
ration (Fig. 5F) (43), and by the spatial enrich-
ment of the more mature OLG2 in fiber tracts
for myelination-based axon ensheathment (Fig.
5, G and H). To correlate translational regulation
with oligodendrocyte subtypes and maturation,
we further clustered oligodendrocyte-related
gene modules using their subtype-resolved
translatome and transcriptome profiles in the
oligodendrocyte lineage (Fig. 5I, and table S6).
We detected three gene modules: Module 1 and
Module 2 have consistent patterns between
RIBOmap and STARmap with the highest signals
in OPC and OLG2 for most genes, respectively,
whereas Module 3 has the highest RIBOmap
signal in OLG2 but the highest STARmap signal
in OLG1 for most genes (Fig. 5I). This obser-
vation suggested strong translational regulation
for genes in Module 3. We further calculated
the relative translation efficiency (RTE) for
each gene using their ratio of RIBOmap and
STARmap signals and found that genes in
Module 3 had higher RTE in OLG2 than OLG1
and OPC (Fig. 5I). GO analysis revealed that
Module 3 genes are involved in myelin sheath
(i.e., Cldn11 and Tspan2) (Fig. 5, J and K). Spa-
tial analysis of Module 3 RTE patterns and
representative genes across brain regions also
showed the highest RTE value in the fiber
tracts, where OLG2 is enriched (Fig. 5, L to O,
and fig. S16A). We experimentally verified one
example gene Cldn11 and found that Claudin 11
(the protein product of Cldn11) immunostaining
correlated better with Cldn11 RIBOmap signal
than STARmap signal, validating the accuracy
of the spatial translatomic pattern we observed
(fig. S16, B to E). Overall, through the integra-
tive analysis of RIBOmap and STARmap results
in the mouse brain, we identified gene modules
that are translationally regulated during the
maturation of oligodendrocyte lineage cells
across different brain regions. The enhanced
translation efficiency of Module 3 genes in OLG2
5 of 10
RES EARCH | R E S E A R C H A R T I C L E

Fig. 4. Spatially single-cell A B Cycle 1 (MAX) YZ view Cycle 1 (single frame) Cycle 2–9 (single frame)
translatomic profiling 2 3 4 5 6 7 8 9
1
of 5413 genes in the mouse
RIBOmap
brain. (A) Diagram of the
5,413 genes
imaged mouse coronal hemi-
brain region (red box) for Rep1: 60,481 cells 100 µm 5 µm 5 µm 5 µm
Mouse brain tissue Rep2: 58,692 cells X
RIBOmap. (B) Representative Y XY view XZ view XY view Channel 1 2 3 4 DAPI

C
Z
images showing the measure- D E
ments of localized translation of Slc17a7 Gad2 Pcp4
Major clusters Subclusters
32
31
5413 genes by RIBOmap in a VAS
OLG 26
mouse coronal hemibrain slice. 28
30 29
25 24
23
AC
(Left) Maximum intensity
RIBOmap

27
MLG OPC 37
38
PVM 36
CHOR, EPEN 22 34
(MAX) projection of the first DE/MEN
INH 21 15 16 18
20
35
33
14
CHO, PEP
sequencing round, showing all 10
19 17

6 7 9 8
five channels simultaneously.

UMAP 2
3

UMAP 2
5 1
1 mm TEPN 11
4 2 12

Red square, zoom region. 13

UMAP 1 UMAP 1
(Middle) Magnified view of Telencephalon projecting neurons (TEPN) TEPN INH AC CHOR, EPEN
Inhibitory interneurons (INH) 1 TEGLU COA 14 INH Pvalb1 23 AC1 33 CHOR
three cells showing the MAX
Allen Brain ISH

2 TEGLU CA1 15 INH Pvalb2 24 AC2 34 EPEN

Cholinergic, monoaminergic and peptidergic neurons (CHO, PEP)
3 TEGLU CA2 16 INH Sst 25 AC3
Di- and mesencephalon neurons (DE/MEN) PVM
4 TEGLU CA3 26 AC4
projection view of the first Astrocytes (AC)
Oligodendrocytes (OLG)
5
6
TEGLU L2/3
TEGLU L4/5
CHO, PEP
17 DECHO OLG
35 PVM1
36 PVM2
Oligodendrocytes precursor cell (OPC) 18 PEP1 27 OLG1
sequencing round. (Right) Microglia (MLG)
Vascular cells (VAS)
7
8
TEGLU L5
TEGLU L6 19 PEP2
20 PEP3
28
VAS
OLG2 37
38
MLG
OPC
9 TEGLU L6a
Magnified view of three cells Choroid epithelial cells and ependymal cells (CHOR, EPEN)
Perivascular macrophages (PVM)
10 TEGLU Mix
11 TEGLU PIR
DE/MEN
21 DEGLU1
29
30
Peri/VEC1
Peri/VEC2
Mix

Mix 12 DGGRC 31 VLMC

22 DEGLU2
showing the spatial arrangement 32 VSMC

F G
13 MSN

of amplicons in a single z-frame Isocortex TEPN AC

across nine sequencing rounds. Hippocampal region (HIP)
Fiber tracts
(C) RIBOmap and Allen Brain Thalamus (TH)
Cerebral nuclei (CNU)
ISH images (41) showing the Hypothalamus (HY)

expression patterns of the three

cell-type marker genes in Cortical subplate
(CTXsp)
comparable coronal hemibrain Olfactory areas
(OLF)

sections. (D and E) Uniform

manifold approximation and
projection (UMAP) plot visual-
ization of translational profiles of DE/MEN OLG
119,173 cells collected from
mouse coronal hemibrains.
11 major cell types (D) and
38 subtypes (E) were identified
using Leiden clustering. Cells
identified as mixed cells
numbered 7559 (see Methods)
and were excluded from
the downstream analysis.
(F) Representative spatial cell
CHO, PEP VAS
type atlas in the imaged coronal
hemibrain region using the
same color code as in (E).
(G) The respective spatial cell
maps of six major cell types in
the imaged coronal hemibrain
region using the same color
code as in (E).
250 µm

1 mm

suggests substantial translation remodeling. we performed gene clustering using region- (extracted from the HPA database) and less
This may functionally serve to augment the resolved RIBOmap and STARmap profiles and correlated with the STARmap signal (fig. S17,
protein production for axon ensheathment identified gene modules with distinct spatial D to I). By contrast, the RNA ISH signals of
and may be mechanistically relevant to the patterns between RIBOmap and STARmap Gng2 from the Allen Brain database (41) cor-
recent report on altered tRNA modification across different brain regions (e.g., Module 7, related better with the STARmap signal than
during oligodendrocyte maturation (44). fig. S17, A to C, and table S6). For example, G RIBOmap (fig. S17, J to M). These results were
To systematically uncover genes with dif- protein subunit gamma 2 (Gng2) in Module 7 consistent with previous studies in which the
ferential spatial patterns between transla- has a low RIBOmap signal in the thalamus, translatome is better correlated with the pro-
tome and transcriptome across brain regions, which is consistent with the protein signal teome than the transcriptome (7, 11, 12).

Zeng et al., Science 380, eadd3067 (2023) 30 June 2023 6 of 10

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

A B
STARmap Translatome and transcriptome correlation across cell types
RIBOmap

Adjacent mouse 0.60 0.62 0.63 0.68 0.68 0.68 0.69 0.70 0.72 0.72 0.78
brain slices Isocortex Isocortex

LG

PC

N
AC

EP
H

PN

EM
G

VA

PE

IN
PV
O

,P
O

TE

M
,E
HIP

E/
O
HIP

D
H
O

C
H
C
Non-neuron Neuron
TEPN CHO, PEP AC Translatome and Correlation
TEGLU COA DECHO AC1 Transcriptome in
PEP1 AC2 TH
C
TEGLU CA1
TH each brain region 0.60 0.65 0.70 0.75

Fib
TEGLU CA2 PEP2 AC3

Fib
PEP3 and cell type
TEGLU CA3 AC4

er
Translatome and transcriptome correlation across brain regions

er
TEGLU L2/3 DE/MEN CNU CNU

tra
VAS

tra
TEGLU L4/5 DEGLU1

cts
Peri/VEC1
TEGLU L5 DEGLU2

cts
Peri/VEC2 0.58 0.64 0.66 0.66 0.67 0.68 0.68 0.77
TEGLU L6 CHOR, EPEN VLMC
TEGLU L6a VSMC
TEGLU Mix CHOR
HY HY

CT
EPEN

CT
TEGLU PIR MLG

cts

TH

HY

F
rte
Xs

OL
HI
CN
OL
PVM

Xs
OPC OL

tra
Xs
DGGRC

co
CT
PVM1

er

Iso
Mix
F

p
MSN

Fib
INH PVM2
Correlation
INH Pvalb1 OLG 1 mm
INH Pvalb2 OLG1 Fiber tracts Fiber tracts
INH Sst OLG2 0.60 0.65 0.70 0.75
Cell typing and brain region segmentation

D F Plp1
G H RIBOmap STARmap
I Oligo lineage gene clustering
RIBOmap OPC RIBOmap STARmap RTE
100
Cell percentage across brain regions (%)

OLG1
80
OLG2

Relative translational efficiency (RTE)

60 1
40
20

Gene modules
OPC OLG1 OLG2 l (expression)
0

E 0
Mbp
1 2 100
80
STARmap 2 z score
1.0
0.0
60
−1.0
40
RTE
20 3 1.4
0 1.0
0.6
s
U

TH

sp

IP

x
LF
ct

te
H
N

l (expression) 1 mm
H
TX

O
ra

or
C

pseudotime
rt

oc
C

O C
O 1
2
O C
O 1
2

O C
O 1
2
be

LG
LG

LG
LG

LG
LG
Is

P
OPC OLG1 OLG2

O
0 2 4 6 0 1 2 3
Fi

J L N
GO enrichment of Module 3 genes RTE of Module 3 genes RTE Module 3 genes
Myelin sheath 1.25 RIBOmap STARmap RTE
Adult locomotory behavior
CN s
U
TH

HY

p
P

x
F
ct

rte
Xs

OL
HI

Main axon
tra

co
CT

Cell junction organization 1.00 Isocortex

er

Iso
Fib

Regulation of cytoskeleton organization

0 1 2 3 4
-log10(p-value)
M HIP Signal
RTE of example genes in Module 3
K Example genes of Module 3
Efnb3
Cldn11
intensity
High
RIBOmap STARmap RTE Tspan2 RTE
z score

Fib
Efnb3 1.0 Mal 2.2 TH

er
Cldn11 0.0 Rnf13 1.8
Tspan2 CNU

tra
−1.0 Slain1 1.4 Low
Mal

cts
Rnf13 RTE Lap3 1.0 RTE
Slain1 1.4 Sgk1 0.6 1.25

CT F
Lap3 1.0
HY

OL
Xs
CN s
U
TH

HY

p
P

x
F

0.6
ct

rte

Sgk1
Xs

OL
HI

p
ra

co
CT
rt

1 mm
Iso
PC

O 1
2
PC

O 1
2
PC

O 1
2

1.00
LG
LG

LG
LG

LG
LG

Fib

Fiber tracts
O

O
O

O
Cldn11 Rnf13
RIBOmap STARmap RTE RIBOmap STARmap RTE

Isocortex Isocortex

HIP HIP
Signal
intensity
High
Fib

Fib

TH TH
er

er

CNU CNU
tra

tra

Low
cts

cts

RTE
CT F

CT F

2.2
HY HY
OL

OL
Xs

Xs

1.8
p
p

1.4
1 mm
Fiber tracts Fiber tracts 1.0
0.6

Fig. 5. Comparison of spatial translatome and transcriptome in the showing the gene clustering using RIBOmap and STARmap results in the
mouse brain. (A) Two adjacent mouse brain coronal sections were used for three oligodendrocyte lineage cell types (left) and the relative translational
RIBOmap and STARmap, respectively, for generating the spatial cell-type map. efficiency (RTE) of these genes in each oligodendrocyte lineage cell type (right).
(B and C) The correlation of the translatome and transcriptome of the 5413 (J) The top 5 significantly enriched GO terms for Module 3 genes. (K) Heatmap
genes in each cell type (B) and brain region (C). (D) Diffusion map visualization showing the RIBOmap results (left), STARmap results (middle), and RTE
of oligodendrocyte and OPC in RIBOmap samples. (E) Diffusion map with (right) of example genes in Module 3. (L and M) Heatmap showing the average
pseudotime trajectory visualization of oligodendrocyte and OPC in RIBOmap RTE values of all Module 3 genes (L) and the RTE value of Module 3 example
sample generated by Monocle 3. (F) Diffusion map showing the normalized genes (M) in each brain region. (N and O) RIBOmp (left) and STARmap (middle)
expression level of oligodendrocyte lineage marker genes Plp1 and Mbp images show the translation and transcription levels of all Modules 3 genes
in oligodendrocytes and OPCs. (G) Cell percentage of oligodendrocyte and (N) and two Module 3 example genes (O), respectively. Each dot in the images
OPC population in each brain region of RIBOmap (top) and STARmap (bottom) represents a cell and the dot color represents the expression level. (Right)
samples. (H) Cell-resolved spatial map for the oligodendrocyte and OPC Spatial map showing the average RTE of all Module 3 genes (N) and the RTE of
population of RIBOmap (left) and STARmap (right) samples. (I) Heatmap two Module 3 example genes (O) in each brain region.

Zeng et al., Science 380, eadd3067 (2023) 30 June 2023 7 of 10

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

A B
TEPN
TEGLU COA
TEGLU CA1
TEGLU CA2
TEGLU CA3 Neurons
TEGLU L2/3
TEGLU L4/5
TEGLU L5 Oligodendrocytes
TEGLU L6
TEGLU L6a
TEGLU Mix
TEGLU PIR Astrocytes
DGGRC
MSN
INH
INH Pvalb1
INH Pvalb2
INH Sst
200 µm CHOR, EPEN
CHOR
EPEN Somata Neuronal and glial processes
OLG C
OLG1
OLG2
AC
AC1
AC2
AC3
AC4
VAS
Peri/VEC1
Peri/VEC2
VLMC
VSMC
MLG
OPC
Mix

50 µm 50 µm
X
XY view Somata reads Neuronal and glial processes reads
Y
Z
D G Shank1 Eef2
I Gfap (AC)
Shank1
Mbp Processes-enriched-
Gfap translation genes
Processes reads (%)

30 Kif5a
Eef2
Calm1
20

App
10 Atp1a1
Somata-enriched-
Mal
translation genes 200 µm 200 µm 200 µm
Genes
Kif5a Calm1 Mbp (OLG)
E Processes-enriched-translation genes
Cytosolic ribosome
Cytoplasmic translation
Translation
Ribosome
Cytosolic large ribosomal subunit
Cytosol
Cytosolic small ribosomal subunit
Postsynaptic density
Small ribosomal subunit
Cytoplasm
0 20 40 60
80 200 µm 200 µm 200 µm
–log10(FDR)
F H App Atp1a1
J Mal (OLG)
Somata-enriched-translation genes
Integral component of plasma membrane
Plasma membrane
Extracellular space
Extracellular region
Extracellular exosome
Cell adhesion
Cell surface
Integral component of membrane
Extracellular matrix
Angiogenesis
0 10 20 200 µm 200 µm 200 µm
–log10(FDR)

Fig. 6. Localized translation in the somata and processes of neuronal
and glial cells in the mouse brain. (A) Magnified sections in the hippocampus
and CA1 region showing the different cell types and the interspace.
(B) Schematic of a hippocampal slice showing the somata and processes of
hippocampal neurons, oligodendrocytes, and astrocytes. (C) Section in the CA1
region showing somata reads (blue) and neuronal and glial processes reads
(red). (D) Processes read percentages of individual genes in the 5413-gene
RIBOmap measurements, with genes rank-ordered based on their processes
reads percentage. Nine example genes were labeled inset. (E) The top 10
significantly enriched GO terms for processes-enriched-translation genes. (F) The
top 10 significantly enriched GO terms for somata-enriched translation genes.
GO was analyzed by DAVID (see Methods). (G and H) The spatial translation map
of representative processes-enriched translation genes (G) and somata-enriched
translation genes (H) in the hippocampus, showing somata reads (blue) and
processes reads (red). (I and J) The spatial translation map of glial cell marker
genes, including examples of processes-enriched translation genes (I) and
somata-enriched translation genes (J) in the hippocampus, showing somata
reads (blue) and processes reads (red).

Zeng et al., Science 380, eadd3067 (2023) 30 June 2023 8 of 10

RES EARCH | R E S E A R C H A R T I C L E
Subcellular localized mRNA translation
in mouse brain
In addition to brain-wide anatomical analysis,
we investigated potential translational con-
trol at the subcellular level. In the brain tissue,
subcellular localized translation in the cell
body (soma) and periphery branches (processes)
serves as a critical mechanism to assemble and
adjust the network of neuronal and glial cells
(45–50) in response to physiological signals
during neurodevelopment and memory. To
dissect localized translation in the somata and
processes of neuronal and glial cells, we di-
vided the RIBOmap reads into somata reads
(i.e., inside the cell body area identified by
ClusterMap) (51) and processes reads (i.e., the
rest of the reads) (Fig. 6, A to C). Next, we defined
the top 10% of genes with the highest and lowest
processes-to-somata ratios as processes- and
somata-enriched translation genes, respec-
tively (Fig. 6D and table S7). GO analysis showed
that processes-enriched translation genes are
associated with translation machinery and
postsynaptic density (Fig. 6E and fig. S18A).
By contrast, somata-enriched translation genes
are associated with the plasma membrane and
extracellular matrix (Fig. 6F and fig. S18B),
corresponding to localized translation at the
ER in the somata. Notably, we observed abun-
dant processes-enriched translation signals in
hippocampal neuropils (Fig. 6G and fig. S18C).
By contrast, the somata-enriched translation
genes showed sparse RIBOmap signals in hip-
pocampal neuropils (Fig. 6H and fig. S18D),
which may be associated with ER translation.
Beyond neuronal genes, we also observed lo-
calized translation in glial cells (Fig. 6, I and
J, and fig. S18, E and F). Overall, we demon-
strated that RIBOmap can be used to study
subcellular-localized translation in processes
of both neuronal and glial cells of the mouse
brain tissue.
In summary, RIBOmap is a spatial trans-
latomics method with single-cell and molec-
ular resolution, which can be used to study
mRNA regulation and protein synthesis in in-
tact cellular and tissue networks. In contrast
to existing approaches, RIBOmap bypasses
complicated polysome isolation steps and
genetic manipulation, thus holding promise
for spatial and single-cell resolved studies in
post hoc human tissue and disease samples.
The detection efficiency and spatial resolu-
tion of RIBOmap can be further increased by
combining with super-resolution imaging,
expansion microscopy, and sparse labeling
(52) to further illustrate translational events
in fine subcellular structures (e.g., dissecting
dendritic and axonic mRNA translation in
neurons). It is noteworthy that the accuracy
of RTE analysis depends on the precise joint
cell typing of RIBOmap and STARmap data.
Therefore, further computational develop-
ment to improve the cross-modality integra-
Zeng et al., Science 380, eadd3067 (2023) 30 June 2023
tion of various spatial omics modalities will
significantly enhance the analysis and inter-
pretation of RIBOmap. Whereas RIBOmap
in this manuscript focuses on mapping RNA
translation, such proximity-based tri-probe
design can be readily adapted to study RNA-
RNA interactions, RNA-protein interactions,
and RNA modifications. In the future, we en-
vision that RIBOmap can be combined with
other imaging-based measurements to enable
spatial multiomics mapping of epigenome,
transcriptome, and translatome in the same
samples for an integrative understanding of
biological systems.
RE FERENCES AND NOTES
1. J. A. Alfaro et al., The emerging landscape of single-molecule
protein sequencing technologies. Nat. Methods 18, 604–617
(2021). doi: 10.1038/s41592-021-01143-1; pmid: 34099939
2. T. Maier, M. Güell, L. Serrano, Correlation of mRNA and protein
in complex biological samples. FEBS Lett. 583, 3966–3973
(2009). doi: 10.1016/j.febslet.2009.10.036; pmid: 19850042
3. R. de Sousa Abreu, L. O. Penalva, E. M. Marcotte, C. Vogel,
Global signatures of protein and mRNA expression levels.
Mol. Biosyst. 5, 1512–1526 (2009). doi: 10.1039/b908315d;
pmid: 20023718
4. C. Vogel, E. M. Marcotte, Insights into the regulation of
protein abundance from proteomic and transcriptomic
analyses. Nat. Rev. Genet. 13, 227–232 (2012). doi: 10.1038/
nrg3185; pmid: 22411467
5. B. Schwanhäusser et al., Global quantification of mammalian
gene expression control. Nature 473, 337–342 (2011).
doi: 10.1038/nature10098; pmid: 21593866
6. S. Das, M. Vera, V. Gandin, R. H. Singer, E. Tutucci, Intracellular
mRNA transport and localized translation. Nat. Rev. Mol. Cell
Biol. 22, 483–504 (2021). doi: 10.1038/s41580-021-00356-8;
pmid: 33837370
7. N. T. Ingolia, S. Ghaemmaghami, J. R. S. Newman,
J. S. Weissman, Genome-wide analysis in vivo of translation
with nucleotide resolution using ribosome profiling.
Science 324, 218–223 (2009). doi: 10.1126/science.1168978;
pmid: 19213877
8. C. H. Jan, C. C. Williams, J. S. Weissman, Principles of ER
cotranslational translocation revealed by proximity-specific
ribosome profiling. Science 346, 1257521 (2014). doi: 10.1126/
science.1257521; pmid: 25378630
9. C. C. Williams, C. H. Jan, J. S. Weissman, Targeting and
plasticity of mitochondrial proteins revealed by proximity-
specific ribosome profiling. Science 346, 748–751 (2014).
doi: 10.1126/science.1257522; pmid: 25378625
10. M. VanInsberghe, J. van den Berg, A. Andersson-Rolf,
H. Clevers, A. van Oudenaarden, Single-cell Ribo-seq reveals
cell cycle-dependent translational pausing. Nature 597,
561–565 (2021). doi: 10.1038/s41586-021-03887-4;
pmid: 34497418
11. N. T. Ingolia, Ribosome Footprint Profiling of Translation
throughout the Genome. Cell 165, 22–33 (2016). doi: 10.1016/
j.cell.2016.02.066; pmid: 27015305
12. Y. Liu, A. Beyer, R. Aebersold, On the Dependency of Cellular
Protein Levels on mRNA Abundance. Cell 165, 535–550
(2016). doi: 10.1016/j.cell.2016.03.014; pmid: 27104977
13. J. M. Halstead et al., An RNA biosensor for imaging the
first round of translation from single cells to living animals.
Science 347, 1367–1671 (2015). doi: 10.1126/science.aaa3380;
pmid: 25792328
14. B. Wu, C. Eliscovich, Y. J. Yoon, R. H. Singer, Translation
dynamics of single mRNAs in live cells and neurons. Science
352, 1430–1435 (2016). doi: 10.1126/science.aaf1084;
pmid: 27313041
15. T. Morisaki et al., Real-time quantification of single RNA
translation dynamics in living cells. Science 352, 1425–1429
(2016). doi: 10.1126/science.aaf0899; pmid: 27313040
16. X. Yan, T. A. Hoek, R. D. Vale, M. E. Tanenbaum, Dynamics of
Translation of Single mRNA Molecules In Vivo. Cell 165,
976–989 (2016). doi: 10.1016/j.cell.2016.04.034;
pmid: 27153498
17. S. Boersma et al., Multi-Color Single-Molecule Imaging Uncovers
Extensive Heterogeneity in mRNA Decoding. Cell 178, 458–472.
e19 (2019). doi: 10.1016/j.cell.2019.05.001; pmid: 31178119
18. Z. B. Katz et al., Mapping translation ‘hot-spots’ in live cells by
tracking single molecules of mRNA and ribosomes. eLife 5,
e10415 (2016). doi: 10.7554/eLife.10415; pmid: 26760529
19. S. tom Dieck et al., Direct visualization of newly synthesized
target proteins in situ. Nat. Methods 12, 411–414 (2015).
doi: 10.1038/nmeth.3319; pmid: 25775042
20. X. Wang et al., Three-dimensional intact-tissue sequencing of
single-cell transcriptional states. Science 361, eaat5691 (2018).
doi: 10.1126/science.aat5691; pmid: 29930089
21. N. T. Ingolia, L. F. Lareau, J. S. Weissman, Ribosome profiling
of mouse embryonic stem cells reveals the complexity and
dynamics of mammalian proteomes. Cell 147, 789–802 (2011).
doi: 10.1016/j.cell.2011.10.002; pmid: 22056041
22. M. Clamer et al., Active Ribosome Profiling with RiboLace. Cell
Rep. 25, 1097–1108.e5 (2018). doi: 10.1016/j.celrep.2018.09.084;
pmid: 30355487
23. I. Tirosh et al., Dissecting the multicellular ecosystem of
metastatic melanoma by single-cell RNA-seq. Science 352,
189–196 (2016). doi: 10.1126/science.aad0501;
pmid: 27124452
24. F. M. Fazal et al., Atlas of Subcellular RNA Localization
Revealed by APEX-Seq. Cell 178, 473–490.e26 (2019).
doi: 10.1016/j.cell.2019.05.027; pmid: 31230715
25. X. Wang et al., N6-methyladenosine-dependent regulation of
messenger RNA stability. Nature 505, 117–120 (2014).
doi: 10.1038/nature12730; pmid: 24284625
26. H. Shi et al., YTHDF3 facilitates translation and decay of
N6-methyladenosine-modified RNA. Cell Res. 27, 315–328
(2017). doi: 10.1038/cr.2017.15; pmid: 28106072
27. A. Sakaue-Sawano et al., Visualizing spatiotemporal dynamics
of multicellular cell-cycle progression. Cell 132, 487–498
(2008). doi: 10.1016/j.cell.2007.12.033; pmid: 18267078
28. S.-B. Koh et al., Quantitative FastFUCCI assay defines cell cycle
dynamics at single-cell level. J. Cell Sci.jcs.195164 (2016).
doi: 10.1242/jcs.195164; pmid: 27888217
29. T. Robin, A. Bairoch, M. Müller, F. Lisacek, L. Lane, Large-Scale
Reanalysis of Publicly Available HeLa Cell Proteomics Data in
the Context of the Human Proteome Project. J. Proteome Res.
17, 4160–4170 (2018). doi: 10.1021/acs.jproteome.8b00392;
pmid: 30175587
30. Y. Lin et al., eIF3 Associates with 80S Ribosomes to Promote
Translation Elongation, Mitochondrial Homeostasis, and Muscle
Health. Mol. Cell 79, 575–587.e7 (2020). doi: 10.1016/
j.molcel.2020.06.003; pmid: 32589965
31. D. Mahdessian et al., Spatiotemporal dissection of the cell
cycle with single-cell proteogenomics. Nature 590, 649–654
(2021). doi: 10.1038/s41586-021-03232-9; pmid: 33627808
32. M. E. Tanenbaum, N. Stern-Ginossar, J. S. Weissman, R. D. Vale,
Regulation of mRNA translation during mitosis. eLife 4, e07957
(2015). doi: 10.7554/eLife.07957; pmid: 26305499
33. P. Kaewsapsak, D. M. Shechner, W. Mallard, J. L. Rinn,
A. Y. Ting, Live-cell mapping of organelle-associated RNAs via
proximity biotinylation combined with protein-RNA
crosslinking. eLife 6, e29224 (2017). doi: 10.7554/eLife.29224;
pmid: 29239719
34. A. Saunders et al., Molecular Diversity and Specializations
among the Cells of the Adult Mouse Brain. Cell 174,
1015–1030.e16 (2018). doi: 10.1016/j.cell.2018.07.028;
pmid: 30096299
35. Z. Yao et al., A taxonomy of transcriptomic cell types across
the isocortex and hippocampal formation. Cell 184,
3222–3241.e26 (2021). doi: 10.1016/j.cell.2021.04.021;
pmid: 34004146
36. A. Zeisel et al., Molecular Architecture of the Mouse Nervous
System. Cell 174, 999–1014.e22 (2018). doi: 10.1016/
j.cell.2018.06.021; pmid: 30096314
37. Tabula Muris Consortium, Overall coordination, Logistical
coordination, Organ collection and processing, Library
preparation and sequencing, Computational data analysis, Cell
type annotation, Writing group, Supplemental text writing
group,Principal investigators, Single-cell transcriptomics
of 20 mouse organs creates a Tabula Muris. Nature 562,
367–372 (2018). doi: 10.1038/s41586-018-0590-4
38. N. Almanzar et al., A single-cell transcriptomic atlas
characterizes ageing tissues in the mouse. Nature 583,
590–595 (2020). doi: 10.1038/s41586-020-2496-1;
pmid: 32669714
39. X. Han et al., Mapping the Mouse Cell Atlas by Microwell-Seq.
Cell 173, 1307 (2018). doi: 10.1016/j.cell.2018.05.012;
pmid: 29775597
40. J. Cao et al., The single-cell transcriptional landscape of
mammalian organogenesis. Nature 566, 496–502 (2019).
doi: 10.1038/s41586-019-0969-x; pmid: 30787437
9 of 10
RES EARCH | R E S E A R C H A R T I C L E
41. E. S. Lein et al., Genome-wide atlas of gene expression in the
adult mouse brain. Nature 445, 168–176 (2007). doi: 10.1038/
nature05453; pmid: 17151600
42. H. Zeng et al., Integrative in situ mapping of single-cell
transcriptional states and tissue histopathology in a
mouse model of Alzheimer’s disease. Nat. Neurosci. 26,
430–446 (2023). doi: 10.1038/s41593-022-01251-x;
pmid: 36732642
43. S. Marques et al., Oligodendrocyte heterogeneity in the mouse
juvenile and adult central nervous system. Science 352,
1326–1329 (2016). doi: 10.1126/science.aaf6463;
pmid: 27284195
44. S. Martin et al., Oligodendrocyte differentiation alters tRNA
modifications and codon optimality-mediated mRNA
decay. Nat. Commun. 13, 5003 (2022). doi: 10.1038/
s41467-022-32766-3; pmid: 36008413
45. M. S. Fernandopulle, J. Lippincott-Schwartz, M. E. Ward, RNA
transport and local translation in neurodevelopmental and
neurodegenerative disease. Nat. Neurosci. 24, 622–632
(2021). doi: 10.1038/s41593-020-00785-2; pmid: 33510479
46. A. Biever et al., Monosomes actively translate synaptic mRNAs
in neuronal processes. Science 367, eaay4991 (2020).
doi: 10.1126/science.aay4991; pmid: 32001627
47. C. Glock et al., The translatome of neuronal cell bodies,
dendrites, and axons. Proc. Natl. Acad. Sci. U.S.A. 118,
e2113929118 (2021). doi: 10.1073/pnas.2113929118;
pmid: 34670838
48. L. M. Meservey, V. V. Topkar, M.-M. Fu, mRNA Transport and
Local Translation in Glia. Trends Cell Biol. 31, 419–423 (2021).
doi: 10.1016/j.tcb.2021.03.006; pmid: 33840591
49. K. Sakers et al., Astrocytes locally translate transcripts in their
peripheral processes. Proc. Natl. Acad. Sci. U.S.A. 114,
E3830–E3838 (2017). doi: 10.1073/pnas.1617782114;
pmid: 28439016
50. C. Müller, N. M. Bauer, I. Schäfer, R. White, Making myelin basic
protein -from mRNA transport to localized translation.
Front. Cell. Neurosci. 7, (2013). doi: 10.3389/fncel.2013.00169
Zeng et al., Science 380, eadd3067 (2023) 30 June 2023
51. Y. He et al., ClusterMap for multi-scale clustering analysis of
spatial gene expression. Nat. Commun. 12, 5909 (2021).
doi: 10.1038/s41467-021-26044-x; pmid: 34625546
52. S. Alon et al., Expansion sequencing: Spatially precise in situ
transcriptomics in intact biological systems. Science 371,
eaax2656 (2021). doi: 10.1126/science.aax2656; pmid: 33509999
53. H. Zeng et al., Preprocessed dataset of the Spatially Resolved
Single-cell translatomics at Molecular Resolution, Version 1.0.1,
Zenodo (2023); https://zenodo.org/record/8041323.
54. H. Zeng et al., Gene expression dataset of the Spatially Resolved
Single-cell Translatomics at Molecular Resolution, Version 1.0.0,
Zenodo (2023); https://zenodo.org/record/8041114.
55. J. Huang, K. Wang, Z. Tang, J. Ren, H. Zeng, X. Wang, wanglab-
broad/RIBOmap-analysis: v1.0.2, Version 1.0.2, Zenodo (2023);
https://zenodo.org/record/8041365.
ACKN OWLED GMEN TS
We thank S. Zhu (Broad Institute), M. Pan (Broad Institute) for
technical assistance, J. Tian (Broad Institute), J. Liu (Harvard
University), and L. Gaffney (Broad Institute) for helpful discussions
and thoughtful comments on the manuscript. We thank F. Zhang
(Broad Institute and MIT) for kindly providing the psPAX2 and
VSVG plasmids. Funding: This work was funded by the
following: The Searle Scholars Program (to X.W.); Thomas D.
and Virginia W. Cabot Professorship (to X.W.); Edward Scolnick
Professorship (to X.W.); Ono Pharma Breakthrough Science
Initiative Award (to X.W.); Merkin Institute Fellowship (to X.W.);
National Institutes of Health DP2 New Innovator Award
1DP2GM146245-01 (to X.W.); Helen Hay Whitney Foundation
Postdoctoral Fellowship (to H.S.); and the National Institutes of
Health Under Award T32AR007098 (to J.A.L.) X.W. acknowledges
the support from the Searle Scholars Program, Thomas D. and
Virginia W. Cabot Professorship, Edward Scolnick Professorship,
Ono Pharma Breakthrough Science Initiative Award, Merkin
Institute Fellowship, and NIH DP2 New Innovator Award. H.S. is
supported by the Helen Hay Whitney Foundation Postdoctoral
Fellowship. J.A.L. is supported by the NIH Under Award.
Author contributions: H.Ze., J.R., and X.W. optimized RIBOmap
probe design and developed the method. H.Ze. performed RIBOmap
experiments with the help of J.R., Y.Z., A.A., X.S., and H.C. J.H.,
C.K.W., Z.T., and H.Zh. performed computational analyses. H.Ze.
prepared the FUCCI cell samples and Y.Z. prepared animal
samples. H.Ze., J.H., C.K.W., and X.W. analyzed and interpreted the
data with the inputs from Z.T., H.S., and J.A.L. All authors
contributed to writing and revising the manuscript and approved
the final version. X.W. conceptualized and supervised the project.
Competing interests: X.W., H.Z., and J.R. are inventors on pending
patent applications related to RIBOmap. X.W. is a scientific cofounder
of Stellaromics. All methods, protocols, and sequences are freely
available to nonprofit institutions and investigators. Data and
materials availability: The RIBOmap preprocessed dataset and gene
expression dataset are available on Zenodo (53, 54) and Single Cell
Portal (SCP) (https://singlecell.broadinstitute.org/single_cell/study/
SCP1835). All code and analysis are available on Zenodo (55).
Additional information is available at the Wang lab website (https://
www.wangxiaolab.org/). License information: Copyright © 2023 the
authors, some rights reserved; exclusive licensee American
Association for the Advancement of Science. No claim to original US
government works. https://www.sciencemag.org/about/science-
licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.add3067
Materials and Methods
Figs. S1 to S18
Tables S1 to S7
References (56–69)
MDAR Reproducibility Checklist
View/request a protocol for this paper from Bio-protocol.
Submitted 2 June 2022; resubmitted 29 January 2023
Accepted 7 May 2023
10.1126/science.add3067
10 of 10
RES EARCH

◥ neutrino (vt) with nuclei, as well as scattering

RESEARCH ARTICLE interactions of all three neutrino flavors [ve,
muon neutrino (vm), and vt] on nuclei. Be-
NEUTRINO ASTROPHYSICS cause the charged particles in cascade events
travel only a few meters, these energy deposi-
Observation of high-energy neutrinos from the tions appear almost point-like to IceCube’s
125-m (horizontal) and 7- to 17-m (vertical)
Galactic plane instrument spacing. This results in larger di-
rectional uncertainties than tracks. Tracks are
IceCube Collaboration*† elongated energy depositions (often several
kilometers long), which arise predominantly
The origin of high-energy cosmic rays, atomic nuclei that continuously impact Earth’s atmosphere, is from muons generated in cosmic-ray particle
unknown. Because of deflection by interstellar magnetic fields, cosmic rays produced within the Milky interactions in the atmosphere or muons pro-
Way arrive at Earth from random directions. However, cosmic rays interact with matter near their duced by interactions of vm with nuclei. The
sources and during propagation, which produces high-energy neutrinos. We searched for neutrino energy deposited by cascades is often con-
emission using machine learning techniques applied to 10 years of data from the IceCube Neutrino tained within the instrumented volume (un-
Observatory. By comparing diffuse emission models to a background-only hypothesis, we identified like tracks), which provides a more complete
neutrino emission from the Galactic plane at the 4.5s level of significance. The signal is consistent with measure of the neutrino energy (19).
diffuse emission of neutrinos from the Milky Way but could also arise from a population of unresolved Searches for astrophysical neutrino sources
point sources. are affected by an overwhelming background
of muons and neutrinos produced by cosmic-

T
he Milky Way emits radiation across the
electromagnetic spectrum, from radio
waves to gamma rays. Observations at
different wavelengths provide insight into
the structure of the Galaxy and have iden-
tified sources of the highest-energy photons.
For gamma rays with energies above 1 GeV,
the plane of the Milky Way is the most prom-
inent feature visible on the sky (Fig. 1B). Most
of this observed gamma-ray flux consists of
photons generated by the decay of neutral
pions (p0), themselves produced by cosmic
rays (high-energy charged particles) collid-
ing with the interstellar medium within the
Milky Way Galaxy (1).
Photons can also be produced from inter-
actions of energetic electrons with interstellar
photon fields or be absorbed by ambient in-
terstellar matter, so other messengers are
needed to determine the cosmic-ray inter-
actions and acceleration sites in the Galaxy.
In addition to neutral pions, cosmic-ray inter-
actions also produce charged pions. Charged
pions produce neutrinos when they decay.
Unlike photons, neutrinos rarely interact
on their way to Earth, so they directly trace
the location and energetics of the cosmic-
ray interactions.
Because both gamma rays and neutrinos
arise from the decay of pions, a diffuse neu-
trino flux concentrated along the Galactic
plane has been predicted (2–5). The expected
tera–electron volt energy neutrino flux is shown
in Fig. 1D, calculated from the giga–electron
volt gamma-ray observation (1). In addition to
the predicted diffuse emission, the Milky Way
is densely populated with numerous high-
*Corresponding author: analysis@icecube.wisc.edu; F. Halzen
(francis.halzen@icecube.wisc.edu)
†IceCube Collaboration authors and affiliations are listed in the
supplementary materials.
energy gamma-ray point sources (also visible
in Fig. 1B), several classes of which are po-
tential cosmic-ray accelerators and therefore
possible neutrino sources (6–10). This makes
the Galactic plane an expected location of
neutrino emission.
Previous searches for this signal by using
neutrino detectors (11–14) have not found a
statistically significant signal (P values ≥0.02).
The development of deep learning techniques
in data science has produced tools (15–17) that
can identify a larger number of neutrino inter-
actions in detector data, with improved angu-
lar resolution over earlier methods. We applied
these deep learning tools to data from the
IceCube Neutrino Observatory to search for
evidence of neutrino emission from the Galac-
tic plane.
Cascade events in IceCube
The IceCube Neutrino Observatory (18), located
at the South Pole, is designed to detect high-
energy (≳1 TeV) astrophysical neutrinos and
identify their sources. The detector construc-
tion, which deployed instruments within a
cubic kilometer of clear ice, was completed
in 2011; 5160 digital optical modules (DOMs)
were placed at depths between 1.5 and 2.5 km
below the surface of the Antarctic glacial ice
sheet. Neutrinos are detected through Cheren-
kov radiation, emitted by charged secondary
particles that are produced by neutrino inter-
actions with nuclei in the ice or bedrock. Be-
cause of the large momentum transfer from
the incoming neutrino, the directions of sec-
ondary particles are closely aligned with the
incoming neutrino direction, enabling the iden-
tification of the neutrino’s origin.
The two main detection channels are cas-
cade and track events. Cascades are short-
ranged particle showers, predominantly from
interactions of electron neutrino (ve) and tau
ray interactions with Earth’s atmosphere. At-
mospheric muons dominate this background;
IceCube records about 100 million muons for
every observed astrophysical neutrino. Whereas
muons from the Southern Hemisphere (above
IceCube) can penetrate several kilometers deep
into the ice, muons from the Northern Hemi-
sphere (below IceCube) are absorbed during
passage through Earth. Because of this shield-
ing effect, and the superior angular resolution
of tracks over cascades (≲1° compared with
≲10°, respectively; both above 10 TeV), searches
for neutrino sources that use IceCube typically
rely on track selections, making them most
sensitive to astrophysical sources in the North-
ern sky (20).
The Galactic Center, as well as the bulk
of the neutrino emission expected from the
Galactic plane, is located in the Southern sky
(Fig. 1, C and D). To overcome the muon back-
ground in the Southern sky, analyses of IceCube
data often use events in which the neutrino
interaction is observed within the detector
(21, 22). The selection of these events greatly
reduces the background rate of cosmic-ray
muons that enter the instrumented volume
from the Southern sky. Unlike these atmo-
spheric muons, atmospheric neutrinos (23)
generally cannot be distinguished (by their in-
teractions in the detector) from astrophysical
neutrinos. Nevertheless, with increasing ener-
gy, an increasing fraction of the atmospheric
neutrinos from the Southern sky (above IceCube)
can be excluded by eliminating events with
simultaneous muons that originate from the
same cosmic-ray air-shower that produced the
atmospheric neutrino (24, 25). At the tera–
electron volt energies relevant for searches of
Galactic neutrino emission, the majority of
these atmospheric neutrinos remain as a sub-
stantial background in searches for astrophys-
ical neutrinos. This background is dominated
by muon neutrinos, which are largely detected

IceCube Collaboration, Science 380, 1338–1343 (2023) 30 June 2023 1 of 6

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

Fig. 1. The plane of the Milky Way Galaxy in photons and neutrinos. (A) to
(E) are in Galactic coordinates, with the origin being at the Galactic Center,
extending ±15° in latitude and ±180° in longitude. (A) Optical color image (39),
which is partly obscured by clouds of gas and dust that absorb optical photons.
[Credit: A. Mellinger, used with permission.] (B) The integrated flux in gamma
rays from the Fermi Large Area Telescope (Fermi-LAT) 12-year survey (40)
at energies greater than 1 GeV, obtained from the Fermi Science Support Center
and processed with the Fermi-LAT ScienceTools. (C) The emission template
calculated for the expected neutrino flux, derived from the p0 template that
matches the Fermi-LAT observations of the diffuse gamma-ray emission (1).
(D) The emission template from (C), after including the detector sensitivity to
cascade-like neutrino events and the angular uncertainty of a typical signal event
(7°, indicated by the dotted white circle). Contours indicate the central regions
that contain 20 and 50% of the predicted diffuse neutrino emission signal.
(E) The pretrial significance of the IceCube neutrino observations, calculated
from the all-sky scan for point-like sources by using the cascade neutrino event
sample. Contours are the same as in (D). Gray lines in (C) to (E) indicate the
northern-southern sky horizon at the IceCube detector.

as tracks in IceCube. The selection of cascade
events instead of track events therefore reduces
the contamination of atmospheric neutrinos—
by about an order of magnitude at tera–electron
volt energies—and permits the energy thresh-
old of the analysis to be lowered to about 1 TeV.
In the Southern sky, the lower background,
better energy resolution, and lower energy
threshold of cascade events compensate for
their inferior angular resolution, compared
with those of tracks. This is particularly true for
searches for emission from extended objects,
such as the Galactic plane, for which the size
of the emitting region is larger than (or similar
to) the angular resolution. Compared with
track-based searches, cascade-based analyses
are more reliant on the signal purity and less
on the angular resolution of individual events.
We therefore expect analyses based on cascades
to have substantially better sensitivity to ex-
tended neutrino emission in the tera–electron
volt energy range from the Southern sky.
Application of deep learning to cascade events
To identify and reconstruct cascade events in
IceCube, we used tools based on deep learn-
ing. These tools are designed to reject the
overwhelming background from atmospheric
muon events, then to identify the energies and
directions of the neutrinos that generated the
cascade events. IceCube observes events at a
rate of about about 2.7 kHz (18), arising mostly
from background events (atmospheric muons
and atmospheric neutrinos) that outnumber
signal events (astrophysical neutrinos) at a
ratio of roughly 108:1. To search for neutrino
sources, event selection was required to im-
prove the signal purity by orders of magnitude.
Previously used event selections for cascade
events (22, 26, 27) relied on high-level observ-
ables, such as the event location within the
IceCube volume and total measured light levels,
to reduce the initial data rate. In subsequent
selection steps, more computing-intensive se-
lection strategies were performed, such as the
definition of veto regions within the detector,
to further reject events identified as incoming
muons. We adopted a different approach,
using tools based on convolutional neural net-
works (CNNs) (15, 28) to perform event selec-
tions. The high inference speed of the neural
networks (milliseconds per event) allowed us
to use a more complex filtering strategy at
earlier stages of the event selection pipeline.
This retains more low-energy astrophysical
neutrino events (Fig. 2) and includes cascade
events that are difficult to reconstruct and dis-
tinguish from background because of their lo-
cation at the boundaries of the instrumented
volume or in regions of the ice with degraded
optical clarity (from higher concentrations of
impurities in the ice).
After the selection of events, we refined
event properties, such as the direction of the
incoming neutrino and deposited energy, using
the patterns of deposited light in the detector.
The likelihood of the observed light pattern
under a given event hypothesis was maximized
to determine the event properties that best
describe the data. For this purpose, we used
a hybrid reconstruction method (16, 17) that
combines a maximum likelihood estimation
with deep learning. In this approach, we used
a neural network (NN) to parameterize the
relationship between the event hypothesis
and expected light yield in the detector. This
smoothly approximates a (more computation-
ally expensive) Monte Carlo simulation while
avoiding the simplifications that limit other
reconstruction methods (19, 29). Starting with
an event hypothesis, the NN models the photon

IceCube Collaboration, Science 380, 1338–1343 (2023) 30 June 2023 2 of 6

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

with the interstellar medium in the Galactic

10 2 A B
101 plane, tests that use catalogs of known Galac-
tic sources of tera–electron volt gamma rays,

< -5°)
10 1
and a test for neutrino emission from the
< -5°)

10 0 Fermi Bubbles (large areas of diffuse gamma-

NAstro [year− 1] (-90° <

100 ray emission observed above and below the
AEff [m2 ] (-90°<

10− 1 Galactic Center) (32). We also performed an all-

sky point-like source search and a test for emis-
10− 2 sion from a catalog of known giga–electron volt
10− 1 (mostly extragalactic) gamma-ray emitters (sup-
10− 3
This Work
plementary text). The results for each test (30)
10− 4 Cascades (12) are summarized in Table 1.
Tracks (20)
Galactic plane neutrino searches
10− 5 10− 2
103 104 105 106 103 104 105 106 We tested three models of Galactic diffuse
E [GeV] E [GeV]
neutrino emission, extrapolated from the ob-
Fig. 2. Neutrino effective area and event selection comparison. (A) The all-flavor southern sky effective area servations in gamma rays (Fig. 1B). These mod-
(AEff) of the IceCube dataset, averaged over a solid angle in the declination (d) range between –90° and –5° els are referred to as p0, KRA5g , and KRA 50
g (33)
as a function of Ev, the true neutrino energy. Results are shown for the deep learning event selection used in this and are each derived from the same under-
work (dark blue), a previous cascade event selection (light blue) (12), and a previous track event selection (gray) lying gamma-ray observations (1). The model
(20) applied to the IceCube data. (B) The number of expected signal events (NAstro) in the Southern sky per predictions depend on the distribution and
energy bin per year for each event selection, assuming an isotropic astrophysical flux (22). Calculations are based emission spectrum of cosmic-ray sources in
on equal contributions of each neutrino flavor at Earth because of neutrino oscillations. the Galaxy, the properties of cosmic-ray diffu-
sion in the interstellar medium, and the spa-
tial distribution of target gas. Each neutrino
yield at each DOM. Symmetries (such as rota- the observed flux (22). The remaining 87% of emission model was converted to a spatial tem-
tion, translation, and time invariance of the the events are atmospheric neutrinos. These plate, then convolved with the detector ac-
neutrino interaction) and detector-specific do- fractions are not used in the analysis directly; ceptance and the event’s estimated angular
main knowledge are exploited by directly in- instead, we used the entire sample to derive a uncertainty, to produce an event-specific spatial
cluding them in the network architecture, which data-driven background estimate. probability density function (shown for a typical
is analogous to how a Monte Carlo simulation event angular uncertainty of 7° in Fig. 1D).
would exploit this information. This differs Searches for Galactic neutrino emission The p0 model assumes that the mega–electron
from previous CNN-based methods used in We used this event selection to perform volt–to–giga–electron volt p0 component, infer-
neutrino telescopes (15), which inferred the searches based on several neutrino emission red from the gamma-ray emission, follows a
event properties directly from the observed hypotheses (30). For each hypothesis, we used power law in photon energy (E) of E–2.7 and
data. However, the observed IceCube data a previously described maximum likelihood– can be extrapolated to tera–electron volt en-
are already convolved with detector effects, based method (31), modified to account for ergies with the same spatial emission profile.
making it difficult to exploit the underlying signal contamination in the data-derived back- The KRAg models include a variable spectrum
symmetries. Our hybrid method is intended ground model (11, 12). These techniques, decided in different spatial regions, use a harder (on
to provide a more complete use of available a priori and blind to the reconstructed event average) neutrino spectrum than that of the p0
information. A description of the hybrid meth- directions, infer the background from the data model, and include a spectral cutoff at the
od has been published previously (16), and itself, avoiding the uncertainties introduced by highest energies (33). In this analysis, the KRAg
we discuss its application to our dataset (30). background modeling. We calculated P values models are tested with a template that uses a
We found that this deep learning event se- by comparing the experimental results with constant, model-averaged spectrum over the
lection retains more than 20 times as many mock experiments performed on randomized sky, roughly corresponding to an E–2.5 power
events as that retained with the selection meth- experimental data. The backgrounds for these law, with either a 5 or 50 PeV cosmic-ray en-
od used in the previous cascade-based Galactic searches—consisting of atmospheric muons, ergy cutoff for the KRA5g and KRA50 g models,
plane analysis of IceCube data (Fig. 2) (12). It atmospheric neutrinos, and the flux of ex- respectively. The KRAg models predict more
also provides improved angular resolution, by tragalactic astrophysical neutrinos—are each concentrated neutrino emission from the Ga-
up to a factor of 2 at tera–electron volt energies largely isotropic. The rotation of Earth ensures lactic Center region, whereas the p0 model
(fig. S5) (16). The increased event rate is mostly that for a detector located at the South Pole, predicts events more evenly distributed along
due to the reduced energy threshold and the the detector sensitivity to neutrinos at differ- the Galactic plane. The corresponding neutrino
inclusion of events near the boundaries of ent right ascensions is fairly uniform in each spectrum predicted by each of these models has
the instrumented volume (fig. S3). We analy- declination band. Therefore, we estimated back- a cutoff at about 10 times lower energies.
zed 10 years of IceCube data, collected be- grounds by scrambling the right ascension We performed Galactic template searches
tween May 2011 and May 2021. A total of value of each event, preserving all detector- with the same methods as those of previous
59,592 events were selected over the entire specific artifacts in the data. Any systematic Galactic diffuse emission searches (11, 12).
sky in the energy range of 500 GeV to several differences between the modeling of signal Because of the uncertainties in the expected
peta–electron volts, compared with 1980 events hypotheses and the true signal could reduce distribution of sources, and their emission spec-
from 7 years in the previous selection (12). We the sensitivity of our search but would not trum and cosmic-ray diffusion, we make no
estimate that the remaining sample has an invalidate the resulting P values. assumption about the absolute model nor-
atmospheric muon contamination of about 6% The source hypothesis tests were defined a malization. Instead, the analyses include an
(30), whereas the astrophysical neutrino con- priori. They include tests for the diffuse emis- unconstrained free parameter for the number
tribution is estimated to about 7%, assuming sion expected from cosmic rays interacting of signal events (ns) in the entire sky, which

IceCube Collaboration, Science 380, 1338–1343 (2023) 30 June 2023 3 of 6

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

Observed data Background scramble

A D

B E

C F

Fig. 3. Galactic plane test-statistic contributions. The contribution to the test-statistic t is shown in galactic coordinates (longitude l and latitude b) for each of
the three tested Galactic plane models. The overall test-statistic value was obtained through integration over the whole sky. (A to C) The contributions of each
model for the observed data. (D to F) The contributions of each model for a single randomly selected mock experiment by using scrambled data. In (A) to (F),
contours enclose 20% (white) and 50% (gray) of the predicted model flux; for the p0 model, these are the same as in Fig. 1, D and E. The 50% contours contain about
1.64, 0.70, and 0.65 sr for the p0, KRA5g , and KRA50
g models, respectively.

Fig. 4. All-sky point source search. A map of

the best-fitting pretrial significance for the all-sky
search, shown on an Aitoff projection of the celestial
sphere, in equatorial coordinates (J2000 equinox).
The Galactic plane is indicated with a grey curve,
and the Galactic Center is indicated with a gray
dot. Although some locations appear to have
significant emission, the trial factor for the number
of points searched means that these points are
all individually statistically consistent with
background fluctuations.

provides flux normalization, while keeping the
spectrum fixed to the model. Results for each
model are summarized in Table 1. We rejected the
background-only hypothesis with significances
of 4.71s, 4.37s, and 3.96s for the p0, KRA5g , and
KRA50
g models, respectively. Although these
three hypotheses are correlated, we applied a
conservative trial factor of 3 to the most signif-
icant of these values, leading to a trial-corrected
P value equivalent to a significance of 4.48s.
The best-fitting fluxes are also listed in Table 1.
The flux normalization of the p0 model is
quoted at 100 TeV, assuming a single power
law; however, the KRAg models have a more
complex spectral prediction and are therefore
quoted as multiples of the predicted model
flux. These fluxes correspond to best-fitting
values of 748, 276, and 211 signal events (ns)
in the IceCube dataset for the p0, KRA5g , and
KRA 50g models, respectively. A visualization
of the template results is shown in Fig. 3, A
to C, in the form of a map of the per-steradian
contribution to the results in the sky region
near the Galactic Center for each of the Ga-
lactic plane models. Similar maps for a ran-
domly selected mock experiment are also
shown for comparison (Fig. 3, D to F).
An all-sky point source search was also per-
formed (supplementary text), in which the sky
was divided into a grid of equal solid-angle
bins, spaced 0.45° apart, and each point was
tested as a neutrino point source. The resulting
significances are shown in Fig. 4. Some loca-
tions have excess emission over the background
expectations, including some in spatial coinci-
dence with known gamma-ray emitters, such as
the Crab Nebula, 3C 454.3, and the Cygnus X
region. However, after accounting for trial fac-
tors, no single point in the map is statistically
significant (Table 1). This also implies that the
emission that is present in the Galactic template
analyses is not due to a single point source.
Searches using catalogs of Galactic sources
The total gamma-ray signal from the Galactic
plane includes a contribution from several
strong gamma-ray point sources (1). We there-
fore searched for correlated neutrino emission
from three distinct catalogs of Galactic sources.
Previous work had classified each source as a
supernova remnant (SNR), pulsar wind nebula
(PWN), or other unidentified (UNID) Galactic
source, based on observations in tera–electron
volt gamma rays (34, 35). Under the assump-
tion that multiple sources in each class emit
neutrinos, stacking these sources in a single
analysis provides higher sensitivity compared

IceCube Collaboration, Science 380, 1338–1343 (2023) 30 June 2023 4 of 6

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

detection because the objects in these Galactic

Fig. 5. Energy spectra for KRA 5 Model KRA 5 Best-Fit Flux
source catalogs overlap spatially with regions
each of the Galactic plane KRA 50 Model KRA 50 Best-Fit Flux
that predict the largest neutrino fluxes in the
models. Energy-scaled, sky- 0
Model 0
Best-Fit Flux
Galactic plane diffuse emission searches.
integrated, per-flavor neutrino IceCube All-Sky Flux (22)
flux is shown as a function of 10−6 Implications of Galactic neutrinos
neutrino energy (Ev) for each of
The neutrino flux we observed from the Galac-

[GeV s−1 cm− 2]

the Galactic plane models.
Dotted lines are the predicted
values for the p0 (dark blue),
KRA5g (orange), and KRA50 g (light
blue) models. Solid lines are our
best-fitting flux normalizations
tic plane could arise from several different
emission mechanisms. The predicted energy
spectra integrated over the entire sky is shown
in Fig. 5 for each of the Galactic plane models
10−7 and their best-fitting flux normalization. Model-
dN
dE

to-model flux comparisons depend on the

from the IceCube data. Shaded
E2

regions of the sky considered. The KRAg best-

regions indicate the 1s uncer- fitting flux normalizations are lower than pre-
tainties; they extend over the dicted, which could indicate a spectral cutoff
energy range that contributes that is inconsistent with the 5 and 50 PeV
to 90% of the significance. values assumed. The simpler extrapolation of
These results are based on the 10−8 the p0 model from giga–electron volt energies
all-sky (4p sr) template and are to 100 TeV predicts a neutrino flux that is a
presented as an all-sky flux. For factor of ~5 below our best-fitting flux. How-
103 104 105 106 107
comparison, the gray hatching ever, the best-fitting flux for the p0 model ap-
E [GeV]
shows the IceCube total neu- pear to be consistent with recent observations
trino flux (22), scaled to an all-sky flux by multiplying by 4p, with its 1s uncertainty. of 100-TeV gamma rays by the Tibet Air Shower
Array (fig. S8) (37). The p0 model mismatch
could arise from propagation or spectral differ-
Table 1. Summarized results of the neutrino emission searches. The flux sensitivity and best-fitting ences for cosmic rays in the Galactic Center
flux normalization (F) are given in units of model flux (MF) for the KRAg templates and for the p0 analyses region, or from contributions from unresolved
–12
as E2 dN
dE at 100 TeV, in units of 10 TeV cm–2 s–1 (where dNdE is the differential number of neutrinos per neutrino sources.
flavor, N, and neutrino energy, E). P values and significances are calculated with respect to the We used model injection tests to quantify
background-only hypothesis. Pretrial P values for each individual result are listed for the three diffuse the ambiguity between different source hy-
Galactic plane analyses and three stacking analyses, and posttrial P values are given for the other analyses potheses. In these tests, the best-fitting neu-
(supplementary text). Because of the spatial overlap of the stacking catalogs with the diffuse Galactic trino signal from one source search was
plane templates, strong correlations between these searches are expected. More detailed results for each simulated, then the expected results in all
search are provided in tables S1 to S5. other analyses were examined. Injecting a
signal from the p0 model analysis, with a flux
Flux sensitivity F P value Best-fitting flux F normalization equal to the best-fitting value
Diffuse Galactic plane analysis from the observations, produces a median sig-
.....................................................................................................................................................................................................................
nificance that is consistent with the best-fitting
p0 5.98 1.26 × 10–6 (4.71s) 21:8þ5:3
4:9
..................................................................................................................................................................................................................... values for all other tested hypotheses (within
KRA5g 0.16 × MF 6.13 × 10–6 (4.37s) 0:55þ0:18
0:15  MF the expected statistical fluctuations). This in-
.....................................................................................................................................................................................................................
50 –5 þ0:13 cludes the 3s excess observed in Galactic source
KRA g 0.11 × MF 3.72 × 10 (3.96s) 0:37 0:11  MF
..................................................................................................................................................................................................................... catalog searches. Individually injecting the
Catalog stacking analysis
..................................................................................................................................................................................................................... best-fitting flux of any one of the tested Ga-
SNR 5.90 × 104 (3.24s)*
..................................................................................................................................................................................................................... lactic source catalogs, at the flux level observed,
PWN 5.93 × 104 (3.24s)*
..................................................................................................................................................................................................................... did not recover the observed p0 or KRAg model
UNID 3.39 × 104 (3.40s)*
..................................................................................................................................................................................................................... results. However, the angular resolution of the
Other analyses
..................................................................................................................................................................................................................... sample and the small number of equally
Fermi bubbles 0.06 (1.52s)
..................................................................................................................................................................................................................... weighted sources included in these catalogs
Source list 0.22 (0.77s)
..................................................................................................................................................................................................................... does not constrain emissions from these broad
Hotspot (north) 0.28 (0.58s)
..................................................................................................................................................................................................................... source populations. It is plausible that many
Hotspot (south) 0.46 (0.10s) independently contributing sources from the
.....................................................................................................................................................................................................................

*Significance values that are consistent with the diffuse Galactic plane template search results. Galactic plane could show a similar result to
diffuse emission from interactions in the inter-
stellar medium. These tests favor a neutrino
with individual source searches, because the them under the hypothesis that each contrib- signal from Galactic plane diffuse emission,
neutrino fluxes add together, whereas random utes equally to the flux (supplementary text). but we do not have sufficient statistical power
background adds incoherently (36). The ob- The total number of signal events and the to differentiate between the tested emission
jects in each catalog were selected according spectral index are left as free parameters for models or identify embedded point sources.
to the observed gamma-ray emission above each catalog search. The resulting P value for The neutrinos observed from the Galactic
100 GeV and the detector sensitivity, following each catalog search is shown in Table 1. Each plane contribute to the all-sky astrophysical
previously described methods (20). We chose result rejects the background-only hypothesis diffuse flux previously observed by IceCube
the 12 sources from each category with the at the 3s level or above. However, we do not (Fig. 5) (21, 22, 38). The fluxes we infer for each
strongest expected neutrino flux and weighted interpret these neutrino event excesses as a of the Galactic template models contribute

IceCube Collaboration, Science 380, 1338–1343 (2023) 30 June 2023 5 of 6

RES EARCH | R E S E A R C H A R T I C L E
~6 to 13% of the astrophysical flux at 30 TeV.
These comparisons are complicated because
of different spectral assumptions and tested
energy ranges used in each analysis. Addition-
ally, the observed Galactic flux is integrated
over the entire sky, so local emission along the
central region of the Galactic Plane contrib-
utes a higher fraction to the total flux.
The observed excess of neutrinos from the
Galactic plane provides strong evidence that
the Milky Way is a source of high-energy neu-
trinos. This evidence is consistent with the
expected interactions of cosmic rays within
the Galaxy, as established with gamma-ray mea-
surements. It complements IceCube’s measure-
ments of the diffuse extragalactic flux to provide
a more complete picture of the neutrino sky.
RE FE RENCES AND N OT ES
1. M. Ackermann et al., Astrophys. J. 750, 3 (2012).
2. F. W. Stecker, Astrophys. J. 228, 919 (1979).
3. V. Berezinsky, T. Gaisser, F. Halzen, T. Stanev, Astropart. Phys.
1, 281–287 (1993).
4. G. Ingelman, M. Thunman, arXiv:hep-ph/9604286 (1996).
5. C. Evoli, D. Grasso, L. Maccione, J. Cosmol. Astropart. Phys.
2007, 003 (2007).
6. M. Ackermann et al., Science 339, 807–811 (2013).
7. D. Guetta, E. Amato, Astropart. Phys. 19, 403–407 (2003).
8. W. Bednarek, Astron. Astrophys. 407, 1–6 (2003).
9. M. C. Gonzalez-Garcia, F. Halzen, S. Mohapatra, Astropart. Phys. 31,
437–444 (2009).
10. F. Halzen, A. Kheirandish, V. Niro, Astropart. Phys. 86, 46–56
(2017).
11. M. G. Aartsen et al., Astrophys. J. 849, 67 (2017).
12. M. G. Aartsen et al., Astrophys. J. 886, 12 (2019).
13. A. Albert et al., Astrophys. J. Lett. 868, L20 (2018).
14. A. Albert et al., Phys. Rev. D 96, 062001 (2017).
15. R. Abbasi et al., J. Instrum. 16, P07041 (2021).
16. The IceCube Collaboration, Combining maximum-likelihood with
deep learning for event reconstruction in IceCube, presented
at 2021 ICRC The Astroparticle Physics Conference, 12 to
23 Junly 2021, Berlin, Germany, 10.22323/1.395.1065 (2021).
17. M. Huennefeld, icecube/event-generator: Version 1.0.2. Zenodo
(2022); https://doi.org/10.5281/zenodo.7412035.
18. M. Aartsen et al., J. Instrum. 12, P03012 (2017).
19. M. Aartsen et al., J. Instrum. 9, P03009 (2014).
20. M. G. Aartsen et al., Phys. Rev. Lett. 124, 051103 (2020).
21. R. Abbasi et al., Phys. Rev. D 104, 022002 (2021).
22. M. G. Aartsen et al., Phys. Rev. Lett. 125, 121104 (2020).
23. T. K. Gaisser, M. Honda, Annu. Rev. Nucl. Part. Sci. 52, 153–199
(2002).
24. S. Schönert, T. K. Gaisser, E. Resconi, O. Schulz, Phys. Rev. D
Part. Fields Gravit. Cosmol. 79, 043009 (2009).
25. T. K. Gaisser, K. Jero, A. Karle, J. van Santen, Phys. Rev. D Part.
Fields Gravit. Cosmol. 90, 023009 (2014).
26. M. G. Aartsen et al., Phys. Rev. D Part. Fields Gravit. Cosmol.
91, 022001 (2015).
27. IceCube Collaboration, Science 342, 1242856 (2013).
28. Y. LeCun et al., Advances in Neural Information Processing
Systems 2 (Morgan-Kaufmann, 1990), pp. 396–404.
29. R. Abbasi et al., J. Instrum. 16, P08034 (2021).
30. Materials and methods are available as supplementary materials.
31. J. Braun et al., Astropart. Phys. 29, 299–305 (2008).
32. M. Su, T. R. Slatyer, D. P. Finkbeiner, Astrophys. J. 724,
1044–1082 (2010).
33. D. Gaggero, D. Grasso, A. Marinelli, A. Urbano, M. Valli,
Astrophys. J. Lett. 815, L25 (2015).
34. S. Wakely, D. Horan, TeVCat, http://tevcat2.uchicago.edu
(2022).
35. A. Voruganti, C. Deil, A. Donath, J. King in 6th International
Symposium on High Energy Gamma-Ray Astronomy (AIP Conference
Proceedings, 2017) 1792, p. 070005; http://gamma-sky.net/.
36. R. Abbasi et al., Astrophys. J. 732, 18 (2011).
37. M. Amenomori et al., Phys. Rev. Lett. 126, 141101 (2021).
38. R. Abbasi et al., Astrophys. J. 928, 50 (2022).
39. A. Mellinger, Publ. Astron. Soc. Pac. 121, 1180–1187 (2009).
40. Fermi’s 12-year view of the gamma-ray sky, NASA Goddard
Space Flight Center (2022); https://svs.gsfc.nasa.gov/14090.
IceCube Collaboration, Science 380, 1338–1343 (2023)
ACKN OWLED GMEN TS
The IceCube Collaboration thanks D. Gaggero, D. Grasso,
A. Marinelli, A. Urbano, and M. Valli for providing the templates
for the KRA-g models and for fruitful discussions. Funding: The
authors gratefully acknowledge support from the following
agencies and institutions. USA: US National Science Foundation–
Office of Polar Programs, US National Science Foundation–Physics
Division, US National Science Foundation–EPSCoR, Wisconsin
Alumni Research Foundation, Center for High Throughput
Computing (CHTC) at the University of Wisconsin–Madison, Open
Science Grid (OSG), Extreme Science and Engineering Discovery
Environment (XSEDE), Frontera computing project at the Texas
Advanced Computing Center, US Department of Energy–National
Energy Research Scientific Computing Center, Particle
Astrophysics Research Computing center at the University of
Maryland, Institute for Cyber-Enabled Research at Michigan State
University, and Astroparticle physics computational facility at
Marquette University. Belgium: Funds for Scientific Research
(FRS-FNRS and FWO), FWO Odysseus and Big Science
programmes, and Belgian Federal Science Policy Office (Belspo).
Germany: Bundesministerium für Bildung und Forschung (BMBF),
Deutsche Forschungsgemeinschaft (DFG), Helmholtz Alliance
for Astroparticle Physics (HAP), Initiative and Networking Fund of
the Helmholtz Association, Deutsches Elektronen Synchrotron
(DESY), and High Performance Computing cluster of the RWTH
Aachen. Sweden: Swedish Research Council, Swedish Polar Research
Secretariat, Swedish National Infrastructure for Computing (SNIC), and
Knut and Alice Wallenberg Foundation. Australia: Australian Research
Council. Canada: Natural Sciences and Engineering Research Council of
Canada, Calcul Québec, Compute Ontario, Canada Foundation for
Innovation, WestGrid, and Compute Canada. Denmark: Villum Fonden
and Carlsberg Foundation. New Zealand: Marsden Fund. Japan:
Japan Society for Promotion of Science (JSPS) and Institute for Global
Prominent Research (IGPR) of Chiba University. Korea: National
Research Foundation of Korea (NRF). Switzerland: Swiss National
Science Foundation (SNSF). United Kingdom: Department of Physics,
University of Oxford. Author contributions: Major contributions to
this manuscript were made by M. Hünnefeld and S. Sclafani. The
IceCube Collaboration designed, constructed, and now operates the
IceCube Neutrino Observatory. Data processing and calibration, Monte
Carlo simulations of the detector and of theoretical models, and
data analyses were performed by a large number of collaboration
members, who also discussed and approved the scientific results. The
manuscript was reviewed by the entire collaboration before
publication, and all authors approved the final version. Contributions
are in the areas of (a) Conceptualization, Formal Analysis, and
Methodology; (b) Data Curation and Resources; (c) Funding
Acquisition and Project Administration; (d) Investigation; (e) Software;
(f) Supervision; (g) Validation; (h) Writing – original draft and
visualization; and (i) Writing – review and editing. M. Ackermann
(b, c, d), J. Adams (d, e, g), M. Ahlers (c, d, i), M. Ahrens (d),
J. M. Alameddine (d), A. A. Alves Jr. (e), N. M. Amin (d), K. Andeen
(b, c, d), T. Anderson (d), G. Anton (c, d), C. Argüelles (b, d),
Y. Ashida (d), S. Athanasiadou (d), S. Axani (e), X. Bai (a, c, d),
A. Balagopal V. (d), V. Basu (d), R. Bay (b, d), J. J. Beatty (d),
K.-H. Becker (d), J. Becker Tjus (c, d), J. Beise (d), C. Bellenghi (b, e),
S. Benda (d), D. Berley (c), E. Bernardini (c, d), D. Bindig (d),
E. Blaufuss (a, b, c, d, e, g, h, i), S. Blot (c, d), F. Bontempo (d),
J. Borowka (g, i), S. Böser (c, d), O. Botner (a, b, c, d, i), J. Böttcher
(b, e, g, i), E. Bourbeau (d), F. Bradascio (d), J. Braun (b, d, e),
B. Brinson (d), J. Brostean-Kaiser (d), R. T. Burley (d), R. S. Busse (d),
M. A. Campana (d, e), E. G. Carnie-Bronca (d), C. Chen (d), D. Chirkin
(b, d, e), B. A. Clark (d), L. Classen (d), A. Coleman (b, d, e),
G. H. Collin (e), J. M. Conrad (d), P. Coppin (d), P. Correa (d),
D. F. Cowen (c, d), C. Dappen (b, g, i), P. Dave (d), C. De Clercq (c, d),
J. J. DeLaunay (b, g, i), D. Delgado López (d), H. Dembinski (b, d, e),
K. Deoskar (d), A. Desai (d), P. Desiati (b, c, d, e), K. D. de Vries
(d), G. de Wasseige (d), T. DeYoung (b, c, d), A. Diaz (e),
J. C. Díaz-Vélez (b, d, e), M. Dittmer (d), H. Dujmovic (d), M. Dunkman
(d), M. A. DuVernois (b, d), T. Ehrhardt (d), P. Eller (b, e), R. Engel
(f), H. Erpenbeck (b, g, i), J. Evans (b, d), P. A. Evenson (b, d),
K. L. Fan (b, d), A. Fedynitch (d), N. Feigl (d), S. Fiedlschuster (d),
A. T. Fienberg (d), C. Finley (a, c, d, g, h), L. Fischer (d), D. Fox (d),
A. Franckowiak (c, d), E. Friedman (b, d, e), A. Fritz (d), P. Fürst
(a, b, e, g, i), T. K. Gaisser (a, b, c, d), J. Gallagher (b, d), E. Ganster
(b, e, g, i), A. Garcia (d), S. Garrappa (d), L. Gerhardt (b, d),
A. Ghadimi (b), C. Glaser (d), T. Glauch (b, e), T. Glüsenkamp (d),
N. Goehlke (d), A. Goldschmidt (a, c, d), J. G. Gonzalez (b, c, d, e),
S. Goswami (b), D. Grant (c, d), T. Grégoire (d), C. Günther (g, i),
P. Gutjahr (d), C. Haack (b, d, g), A. Hallgren (a, b, c, d), R. Halliday
(b, d), L. Halve (g, i), F. Halzen (a, b, c, d, h, i), M. Ha Minh (e),
K. Hanson (b, c, d, e), J. Hardin (b, d, e), A. A. Harnisch (d, e),
A. Haungs (c, f), K. Helbing (d), F. Henningsen (e), S. Hickford (d),
30 June 2023
J. Hignight (d, e), C. Hill (b), G. C. Hill (a, b, c, d), K. D. Hoffman (b, c,
d), K. Hoshina (b, d, e), W. Hou (d), F. Huang (d), M. Huber (b, e),
T. Huber (d), K. Hultqvist (c, d), M. Hünnefeld (a, d, e, h, i), R. Hussain
(b, d, e), K. Hymon (d), S. In (d), A. Ishihara (b, c, d, e), M. Jansson
(d, e), G. S. Japaridze (d), M. Jeong (d), D. Kang (d), W. Kang (d),
X. Kang (d, e), A. Kappes (c, d), D. Kappesser (d), L. Kardum (d),
T. Karg (b, c, d), M. Karl (b, e), A. Karle (a, b, c, d, i), U. Katz (c, d),
M. Kauer (b, d, e), M. Kellermann (g, i), J. L. Kelley (b, c, d, e),
A. Kheirandish (d, i), K. Kin (b), S. R. Klein (a, c, d), A. Kochocki (d),
R. Koirala (d), H. Kolanoski (c, d), T. Kontrimas (b, e), L. Köpke (c, d),
C. Kopper (b, c, d, e, g, i), D. J. Koskinen (c, d), P. Koundal (d),
M. Kovacevich (d, e), M. Kowalski (b, c, d), T. Kozynets (d),
E. Krupczak (d), E. Kun (d), N. Kurahashi (a, b, c, d, e, f, g, h, i),
N. Lad (d), C. Lagunas Gualda (d), J. L. Lanfranchi (d), M. J. Larson
(b, c, d, e), F. Lauber (d), J. P. Lazar (b, d), K. Leonard (b, d, e),
A. Leszczyńska (b, c, d, e), Y. Li (d), M. Lincetto (b, d), Q. R. Liu
(b, d, e), M. Liubarska (d), E. Lohfink (d), C. J. Lozano Mariscal (d),
L. Lu (b, d, e), F. Lucarelli (a, d, e), A. Ludwig (d), W. Luszczak (b, d,
e), Y. Lyu (d), W. Y. Ma (d), J. Madsen (c, d), K. B. M. Mahn (d),
Y. Makino (b, d), S. Mancina (b, d, e), I. Martinez-Soler (d),
R. Maruyama (c, d), S. McCarthy (d), T. McElroy (d), F. McNally (d),
J. V. Mead (d), K. Meagher (b, d, e), S. Mechbal (d), A. Medina (d, e),
M. Meier (e), S. Meighen-Berger (e), Y. Merckx (d), J. Micallef (d),
T. Montaruli (a, b, d, e), R. W. Moore (d), R. Morse (a, b, c, d),
M. Moulai (b, d, e), T. Mukherjee (d), R. Naab (d), R. Nagai (b),
R. Nahnhauer (b, d), U. Naumann (d), J. Necker (d), H. Niederhausen
(b, d, g, i), M. U. Nisa (d), S. C. Nowicki (d), A. Obertacke Pollmann
(d), M. Oehler (d), B. Oeyen (d), A. Olivas (b, c, d, e), E. O’Sullivan
(c, d), H. Pandya (d), D. V. Pankova (d), N. Park (a, b, c, i),
E. N. Paudel (d), L. Paul (b, d), C. Pérez de los Heros (a, b, c, d),
L. Peters (b, g, i), J. Peterson (d), S. Philippen (b, e, g, i), S. Pieper
(d), A. Pizzuto (b, d, e), M. Plum (b, e), Y. Popovych (d), A. Porcelli
(d), M. Prado Rodriguez (b, d, e), B. Pries (d), J. Rack-Helleis (d),
K. Rawlins (b, c, d, e), I. C. Rea (e), Z. Rechav (d), A. Rehman (d),
P. Reichherzer (d), R. Reimann (b, e, g, i), E. Resconi (a, b, f, i),
S. Reusch (d), W. Rhode (b, c, d, f), M. Richman (a, b, d, e, g),
B. Riedel (b, c, d, e), E. J. Roberts (d), G. Roellinghoff (d), M. Rongen
(d), C. Rott (b, c, d), T. Ruhe (b, c, d), D. Ryckbosch (c, d),
D. Rysewyk Cantu (d), I. Safa (b, d, e), J. Saffer (d), D. Salazar-Gallegos
(d), P. Sampathkumar (d), S. E. Sanchez Herrera (d), A. Sandrock
(d), M. Santander (b, c, i), S. Sarkar (d), K. Satalecka (d), M. Schaufel
(g, i), H. Schieler (d), S. Schindler (d), T. Schmidt (b, d, e),
A. Schneider (b, d, e), J. Schneider (d), F. G. Schröder (b, c, d),
L. Schumacher (a, b, e), G. Schwefer (a, b, d, e, g, i), S. Sclafani (a, b,
d, e, h, i), D. Seckel (a, b, c, d), A. Sharma (d), S. Shefali (d),
N. Shimizu (b), M. Silva (b, d, e), B. Skrzypek (d), R. Snihur (b, d, e),
J. Soedingrekso (d), D. Soldin (b, c, d, e), C. Spannfellner (e),
G. M. Spiczak (d), C. Spiering (b, c, d), M. Stamatikos (d), T. Stanev
(a, d), R. Stein (d), J. Stettner (b, e, g, i), B. Stokstad (a, c, d),
T. Stürwald (d), T. Stuttard (c, d), G. W. Sullivan (b, c, d), I. Taboada
(c, d, i), J. Thwaites (d), S. Tilav (a, b, c, d, e, g), F. Tischbein (g, i),
K. Tollefson (c, d), C. Tönnis (d), D. Tosi (b, d, e), A. Trettin (d),
M. Tselengidou (d), C. F. Tung (d), A. Turcati (b, e), R. Turcotte (d),
C. F. Turley (d), J. P. Twagirayezu (d), B. Ty (d), M. A. Unland Elorrieta
(d), N. Valtonen-Mattila (b, d), J. Vandenbroucke (b, c, d, e, h, i),
N. van Eijndhoven (c, d), D. Vannerom (e), J. van Santen (b, c, d, e),
J. Veitch-Michaelis (b, d), S. Verpoest (d), C. Walck (d), W. Wang
(d), C. Weaver (d, e), P. Weigel (e), A. Weindl (d), M. J. Weiss (d),
J. Weldert (d), C. Wendt (b, c, d, e), J. Werthebach (d), M. Weyrauch
(d), N. Whitehorn (c, d, e, i), C. H. Wiebusch (a, c, d, f, g, i), N. Willey
(d), D. R. Williams (c, d, e, i), M. Wolf (b, d, e), G. Wrede (d),
J. P. Yanez (d, e), S. Yu (b, d, e, g, i), S. Yoshida (b, c, d), E. Yildizci
(d), and T. Yuan (b, c, d, e). Competing interests: There are no
competing interests to declare. Data and materials availability: The
data and analysis software used in this paper are available from
the IceCube data archive: https://icecube.wisc.edu/science/data/
neutrino-emission-galactic-plane. The event selection code is archived
at Zenodo (17). License information: Copyright © 2023 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.adc9818
Materials and Methods
Supplementary Text
Figs. S1 to S12
Tables S1 to S5
IceCube Collaboration Authors
References (41–62)
Submitted 12 May 2022; accepted 4 May 2023
10.1126/science.adc9818
6 of 6
RES EARCH

HYDROLOGY flooded land since 1977 (pixels that display

water cover for the first time) at a rate of
Agricultural expansion raises groundwater and ~700 km2 year−1 in the central plains that is
unseen in the rest of the continent (Fig. 1A,
increases flooding in the South American plains black line rectangle). Analysis at a coarser
spatial resolution (0.5° cells for which the pro-
Javier Houspanossian1,2*†, Raul Giménez1,2, Juan I. Whitworth-Hulse1, Marcelo D. Nosetto1,3, portion of flooded areas incorporated every
Wlodek Tych4, Peter M. Atkinson4, Mariana C. Rufino4,5, Esteban G. Jobbágy1*† year was calculated) (table S1 and fig. S1)
shows that the areas where floods expanded
Regional effects of farming on hydrology are associated mostly with irrigation. In this work, we show are zones of extreme flatness [i.e., height
how rainfed agriculture can also leave large-scale imprints. The extent and speed of farming above nearest drainage (19) ≤7 m; darker area
expansion across the South American plains over the past four decades provide an unprecedented case in Fig. 1A] with a relatively high proportion
of the effects of rainfed farming on hydrology. Remote sensing analysis shows that as annual crops of rainfed cultivation (i.e., agricultural frac-
replaced native vegetation and pastures, floods gradually doubled their coverage, increasing their tion >25%). Notably, these flooded areas con-
sensitivity to precipitation. Groundwater shifted from deep (12 to 6 meters) to shallow (4 to 0 meters) tinued increasing in size after 2000 (time at
states, reducing drawdown levels. Field studies and simulations suggest that declining rooting which the expanding trend of agriculture
depths and evapotranspiration in croplands are the causes of this hydrological transformation. These across the plains consolidated), whereas the
findings show the escalating flooding risks associated with rainfed agriculture expansion at rest of the continent exhibited stable flooding
subcontinental and decadal scales. conditions throughout 1985 to 2019. Before
2000, the plains of South America accounted

G
lobal demand for grain is leading to
the replacement of large areas of South
America’s native grasslands and forests
as well as cultivated pastures by agri-
cultural crops, mainly soybean (1–3),
providing sustainability and environmental
challenges. Currently, the area covered by an-
nual crops throughout the continent expands
at a rate of 2.1 Mha year−1 (1), particularly in
the sedimentary plains of the Pampas tem-
perate grasslands and the Gran Chaco sub-
tropical forests (4). Flanked by the Andes to
the west and the Brazilian shield to the east,
this region represents one of the flattest and
most hydrologically stagnant plains on the
planet (5, 6). Such flat sedimentary settings
host some of the best farming soils on Earth,
and their hydrology is particularly sensitive
to water balance shifts introduced by land
and water use changes (7, 8). Thus, the ex-
pansion of agriculture on these plains at such
a rate and scale provides both a sustainability
challenge and an outstanding experimental
setting to explore vegetation’s role in mod-
ifying hydrology. This includes the effects of
shifts in the magnitude and timing of water
demand by cultivated canopies as well as the
changing depths of soil explored by rooting
systems (9).
Crop irrigation has introduced some of the
most prominent, large-scale hydrological trans-
formations on Earth, including substantial
water table level rises when surface water is
1
Grupo de Estudios Ambientales, CONICET, San Luis,
Argentina. 2Departamento de Geología, National University of
San Luis, San Luis, Argentina. 3Cátedra de Climatología
Agrícola, Universidad Nacional de Entre Ríos, Entre Ríos,
Argentina. 4Lancaster Environment Centre, Lancaster
University, Lancaster, UK. 5Livestock Systems, TUM School
of Life Sciences, Technical University Munich, Freising,
Germany.
*Corresponding author. Email: jhouspa@gmail.com (J.H.);
jobbagy@gmail.com (E.G.J.)
†These authors contributed equally to this work.
the major source and groundwater deple-
tion when its use gains predominance (10, 11),
as detected at continental scales with remote
sensing tools (12, 13). By contrast, subtle changes
in rainfed cultivation can also cause water
excess and water table level rises (7, 8) ob-
servable at local scales and associated with
regional land degradation processes by water-
logging and salinization of both croplands
and natural vegetation relicts in western Aus-
tralia (14), the Sahel (15, 16), and, more re-
cently, the South American Pampas (17) and
the Chaco (18). In this study, we present evi-
dence of such a process taking place at a sub-
continental scale in the central plains of South
America, revealing the critical role that vege-
tation plays in hydrology there and, likely,
over the water storage and fluxes of other
major sedimentary plains across the globe.
Hydrological shifts and their link to land use
change are characterized by a continental-
scale remote sensing analysis of flooding trends
and exploration of their spatial drivers, a re-
gional description of groundwater level regime
shifts based on long-term public records, and
a statistical synthesis of previous local studies
that compares the effects of contrasting vege-
tation types on groundwater dynamics. These
analyses together with simulations using an
ecohydrological model identify the likely mech-
anisms that link rainfed cropland expansion
with rising groundwater levels and increasing
flooding.
Shifting hydrological regimes
We found that the large-scale replacement of
native vegetation and pastures by rainfed crop-
lands that has been taking place over the past
four decades in the major grain-producing
area of South America was accompanied by
increasing flooding in time and space. Fine-
resolution remote sensing imagery shows a
progressive appearance of clusters of newly
for 40% (10.44 Mha) of the total flooded area,
with most of it (72%, or 7.52 Mha) located in
wetlands and other less-cultivated land (table
S1). By contrast, more than three-quarters of
the newly flooded areas (i.e., those flooded
for the first time after 2000) are in the plains,
and half of these are highly cultivated land
(1.26 Mha). Boosted regression modeling (BRT)
applied to gridded data (water-covered frac-
tion in 0.5° cells) of the whole continent, con-
sidering the proportion of flooded territory
before 2000 (pre-2000) and that flooded for
the first time after that year (post-2000), shows
that wetland area is the strongest explanatory
covariate with a positive association with flood-
ing in both periods (33.6 and 32.1% of the
variance in pre- and post-2000, respectively).
The fraction of woody vegetation cover shows
the next-strongest association with flooding,
being negative in both periods (14.6 and 25.2%
of the variance in pre- and post-2000, re-
spectively; figs. S2 to S4). Conversely, although
the agricultural fraction is not substantially
influential pre-2000, it becomes the third
variable in importance—explaining 9.9% of
the variance—post-2000, with a positive asso-
ciation with flooded area (figs. S4 and S5). The
importance of agricultural fraction becomes
stronger when the BRT is restricted to the
plains of South America, being the second
variable in importance explaining 19.8% of the
variance of post-2000 flooding (fig. S5). Accord-
ing to the BRT analysis, newly flooded areas
appear only in grid cells with >50% cultivated
area (figs. S4 and S5).
Analysis of the evolution of floods within the
cultivated areas of the central plains reveals
a gradual incorporation of areas newly cov-
ered by water following a north-west direc-
tion. The newly flooded areas expanded away
from the historically flooded environments,
known as the Flooding Pampas (boundary
shown by the black dotted line in Fig. 1C),
into the drier aeolian landscapes of the western

Houspanossian et al., Science 380, 1344–1348 (2023) 30 June 2023 1 of 5

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

Fig. 1. Changes in the surface

flooded in South America during
the past four decades. (A) Initial
flooded area before 2000 (bubble
size; 1985 to 1999 time period) and
newly flooded area added after
2000 (bubble color; 2000 to 2019
time period) across 0.5° grid cells
in South America based on the
Landsat Surface Water Cover
product with 30-m spatial resolu-
tion. Shades of gray represent the
terrain topography using height
above the nearest drainage
(HAND). (B) Proportion of the area
occupied by agriculture in 2019
based on the Copernicus Land
Cover product, with 100-m spatial
resolution applied to 0.5° grid cells.
Inset area of the central plains
of South America is displayed (black
rectangle). (C) Detail of the intensive
agricultural area of the central
plains of South America in the
Argentine grain belt and the timing
of floods (pentad of first flooding
episode since 1977 based on Land-
sat imagery). The locations
[stations (ST) 1 to 8] of long-term
public water table level monitoring
sites used in this study as well as
aridity index isolines are shown.
Dotted lines indicate the boundary
of the subregion that has been
considered flood prone historically
(the Flooding Pampas). Dashed lines
indicate Mar Chiquita Lake and the
Parana river delta. (D to H) Land-
scape-level detail displaying the pro-
gression of flooding from 1977 to
2019 in five sample areas.

Pampas between the late 1980s and 1990s
and even to the drier and warmer fluvial and
aeolian landscapes of the southern Chaco after
2000 and at a faster rate after 2015 (Fig. 1C).
Within each of these regions, fine-resolution
analysis of flood progression reveals concen-
tric patterns of expanding temporary water
bodies within the cultivated landscape (Fig. 1,
D to H). In the highly cultivated areas of the
plains (grid cells with >25% agriculture cover;
fig. S1) newly flooded areas were annexed in
the four regional flooding cycles observed
between 1977 and 2019 (Fig. 2A), reaching a
maximum area of 3.8 Mha in 2018—almost
doubling the maximum surface flooded in
the first decade of that period. Although all
long-lasting flooding episodes coincided with
periods of relatively high precipitation (Fig.
2B), their magnitude of flooding increased
with time, particularly in the last episode,
although no significant concurrent increase

Houspanossian et al., Science 380, 1344–1348 (2023) 30 June 2023 2 of 5

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

Fig. 2. Increase in the flooded surface of

the central South American plains in
response to precipitation and agricultural
fraction increases. (A) Annual flooded area in
megahectares of the grain belt of the central
plains of South America (Fig. 1C; Mar Chiquita
Lake and Parana river delta areas were not
considered in the curve) with four steps
of newly flooded area since 1977 colored by the
pentad of first detection (colored bars). Modeled
flooded area subtracting the output of the
annual precipitation-driven dynamic transfer
function (DTF) model is shown as the output
(solid blue line; DTF model fit Rt2 = 0.9)
with its standard error (dashed blue lines).
(B) Mean annual precipitation (black circles)
and integrated random walk (IRW) smooth
trend estimate for annual precipitation (black
solid line) shown with its approximate 95%
confidence interval (black dashed lines).
(C) DTF time-varying gain parameters (blue
circles) and smoothed trend (blue line)
and their approximate 95% confidence intervals
(blue dotted lines), showing effective time-
varying sensitivity of flooding to precipitation.
Smoothed agricultural land use percentage of
Argentine grain belt (red solid and dotted
lines) is also shown.

in precipitation was observed (fig. S6). The
analysis of last-century precipitation, based
on both aggregated and individual station
data (fig. S7, A and B), shows the previously
documented rise between the 1970s and the
1990s (20) but no sustained trend over the
past 30 years. In this period, there are no
significant trends in the characteristics of
rainfall events, including the frequency and
magnitude of the most extreme events (21).
Although associated with positive precipita-
tion anomalies, floods in the central plains
of South America display an increasing sen-
sitivity to rainfall through time, as shown by
the analysis using a dynamic transfer func-
tion model. The dynamic transfer function
produced a good model fit, explaining just
>90% of the variance of the flooded area (Fig.
2A). The temporal analysis reveals a growing
sensitivity of flooding to positive precipitation
anomalies, indicated by the increase in the
gain parameter of the dynamic transfer func-
tion model linking flooded area to mean an-
nual precipitation (Fig. 2C).
Expanding floods in highly cultivated and
flat landscapes were accompanied by rising
water table levels, as indicated by the long-
term records from eight public monitoring
stations distributed across the study region.
The time series analysis shows a generalized
rise in levels that reaches, with different tim-
ings, a similarly stable shallow groundwater
position (−4 to 0 m) at all sites (Fig. 3 and fig.
S8). Hidden Markov models (22) show that
at all sites, station-level variability is best ex-
plained by two or three states (lower Akaike
information criterion), which suggests shift-
ing regimes (figs. S9 to S16), with deep, transi-
tional, and shallow groundwater level states
(Fig. 3 and fig. S8). Notably, shallow states
did not transition to deeper states even during
years with low precipitation.
The hydrological effects of rainfed cultivation
All available field observations provide inde-
pendent and convergent evidence about the
imprint of vegetation and land use changes
on groundwater levels. The analysis of all pub-
lished studies (seven reports for 19 sites) of
groundwater levels and discharge-recharge
fluxes across paired stands comparing crop-
lands with native or planted forests and pas-
tures of the region shows the consistent effects
of croplands on the hydrology of the plains
(table S2). Pooled together, these observa-
tions indicate higher water table levels (mean
effect of +1.54 m), and, individually, they evi-
dence increased groundwater recharge and/or
decreased discharge in croplands compared
with the rest of the vegetation types at all sites
(Fig. 4). In the subset of studies conducted in
drier sites with water table levels >10 m deep,
salt profiles in the unsaturated zone indicate
a full inhibition of deep drainage in native
forests that lasted millennia until the present.
This no-recharge condition was interrupted
by the onset of deep drainage under neigh-
boring cropland (table S2). The remaining
studies, conducted where water tables are
shallower than 10 m, offer evidence of ground-
water consumption (in most cases through
diurnal water table level fluctuations) down

Houspanossian et al., Science 380, 1344–1348 (2023) 30 June 2023 3 of 5

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

Fig. 3. Hidden Markov models applied to time series Hydrological state Agriculture (%)
of water table depth at three sites in the central Deep Shallow Transitional
plains of South America in the Argentine grain belt, 25 50 75
indicating three alternative hydrological states: 0
deep, transitional, and shallow. Agricultural fraction
per year and per site are shown, drawn from depart-
A
-3
mental agricultural statistics corresponding to the

station 1
counties where the stations are located. (A) For station -6
1 (located at Laboulaye, Argentine; ST1 in Fig. 1C), water
table levels rose in the mid-1970s, reaching surface -9
levels in the 1990s. (B and C) For station 2 (B) (located
in Anguil, Argentine; ST2 in Fig. 1C) and for station 3

Water table depth (m)

(C) (located in Marcos Juarez, Argentine; ST3 in Fig. 1C), 0
the water table rose in the mid-1970s gradually to
B
surface levels in the 2000s. Similar situations are -3

station 2
displayed in the rest of the stations (fig. S8). The
vertical bars in (A), (B), and (C) denote the 5th and 95th -6
percentile estimates of water table depth observed.
-9

C
-3

station 3
-6

-9

1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 2020
Date

Fig. 4. Infield evidence

of water table depth
change under land cover
changes in the South
American plains. (A) Mean
water table depth in paired
plots of croplands and native
dry forests or eucalypt
plantations and of croplands
and natural grasslands or
alfalfa pastures across the
central plains of South
America. Numbers in
parentheses indicate the
studies referenced in table
S2 as follows: (1) Giménez
et al. (2016) (18); (2)
Jobbágy et al. (2021) (33);
(3) Nosetto et al. (2013)
(34); (4) Amdan et al.
(2013) (35); (5)
Nosetto et al. (2015) (36);
and (6) Pal et al. (2021) (37).
(B) Boxplots of water table
depth in forests and/or plantations, grasslands and/or pastures, and croplands. The crossbar within the box indicates the median, the length of the box reflects the interquartile
range, the whiskers show the 95% confidence intervals, and the dots represent the outliers. (C) Comparison between the water table depths in croplands and their adjacent forests
and plantations (n = 4 pairwise comparisons) and grasslands and pastures (n = 3). Small squares and triangles refer to the mean effect size (Hedges’s d) for individual pairwise
comparison, and large squares and triangles represent the overall mean effect size estimate for comparisons between forests and/or plantations and croplands and between
grasslands and/or pastures and croplands, respectively. The horizontal bars around the means refer to the 95% confidence intervals. The absolute values of the means and
confidence intervals (in brackets) are shown on the right. Positive mean effect sizes indicate that the water table depth is shallower in croplands than in forests or plantations and
grasslands or pastures. A mean effect size is significantly different from zero when its 95% confidence interval does not include zero (significance level, ***P < 0.001).

Houspanossian et al., Science 380, 1344–1348 (2023) 30 June 2023 4 of 5

RES EARCH | R E S E A R C H A R T I C L E
to 5- to 9-m depth under native forests, tree
plantations, and perennial pastures but only
down to 3-m depth or less under annual crop-
lands (table S2). These contrasts explain in the
first place how cultivation may trigger water
table level rises during wet periods and limit
water table drawdowns during subsequent
dry periods, reducing the threshold of cumu-
lative water excess needed to cause water-
logging and flooding.
Rising flooding risk is widely assumed to
be a phenomenon associated with unusually
high precipitation events, reduced infiltration
rates, and limited run-off caused by land cover
changes (23, 24), whereby the recovery after
floods is a function of the duration of subse-
quent dry periods (22). Although this is the
case for sloped watersheds, extremely flat sedi-
mentary regions such as the South American
plains with stagnant hydrological systems
favor shallow water table level equilibria under
a broad range of climatic conditions (6). Under
these circumstances, vegetation changes play
a major role in modulating flooding, mostly
through the capacity of plants to draw down
unsaturated and saturated water reserves dur-
ing dry periods (figs. S17 to S19). The two most
critical and interconnected features of vegeta-
tion water use that dictate the speed at and
extent to which water reserves are depleted
are its evaporative capacity and rooting depth
(25). These factors control the depth of the
drawdowns and how much of the water stor-
age capacity can be emptied. Therefore, when
rooting depths and groundwater drawdowns
are reduced, there is a higher likelihood of
saturating the soil profile and bringing water
table levels to the surface during subsequent
wet periods, as suggested by one-dimensional,
long-term hydrological simulations using the
Hydrus model (26) calibrated for the west
Pampas (fig. S19).
Discussion
The replacement of forests and pastures with
rainfed annual crops may have been enough
to create a new dynamic hydrological equi-
librium that, even under low-precipitation
conditions, may not recover the initial deep
groundwater levels, making the whole region
more flood prone. Although ecohydrological
recovery is likely under a scenario of farming
abandonment and native vegetation recovery
and/or cultivated pastures reestablishment,
the effects of the observed hydrological shifts
on these agricultural systems do not seem to
have discouraged annual crop expansion until
the present (1). In the specific case of agricul-
ture and in the short term, the positive effects
of a shallower water table buffering crop pro-
duction during droughts appear to compen-
sate for the negative effects of reduced sowing
area (27). However, as flooding continues to
expand into drier areas with more saline and
Houspanossian et al., Science 380, 1344–1348 (2023)
erodible soils, this compensatory effect may
disappear (14, 17). Additionally, important
trade-offs with other sectors of the rural econ-
omy have already been reported, including
damages to the infrastructure of dairy farms
and disruptions to the logistics of small agri-
cultural towns. Climate may also be affected
at global-to-local scales by these hydrological
shifts. Increased intermittent waterlogging is
likely to increase methane and water vapor
fluxes to the atmosphere, limiting soil organic
matter sequestration and contributing to glo-
bal warming (28). At a regional level, floods
have been shown to cause more-frequent fog
events, reduce thermal amplitude, and extend
frost-free periods, affecting both cultivated
and natural ecosystems (29). The historical con-
text of low investment in agricultural infra-
structure and taxation (rather than subsidies)
on grain production (30) in Argentina sug-
gests that drainage would not be a viable fu-
ture response to the increase in flooded area,
and neither would nature-based solutions, such
as hydrology-smart rotations and landscape
designs, become feasible (31). Any one of these
approaches needs to consider the critical role
of rooting depths in the stagnant context of
sedimentary plains and the possibilities to
adjust them through crop and cultivar choice,
native vegetation conservation, and hydrolog-
ically informed decision-making. The insights
provided in this work on the connection be-
tween extensive land use change and the grow-
ing flood sensitivity in the South American
plains will help improve our understanding
of hydrological changes in other regions of
the world with similar characteristics, includ-
ing dry and stagnant hydrological settings and
expanding rainfed grain production systems,
such as those of Ukraine or Canada (32). The
findings presented here are critical for future
land use policies that support farming, water
management, and rural towns in smarter and
more-integrated ways.
RE FERENCES AND NOTES
1. X.-P. Song et al., Nat. Sustain. 2021, 784–792 (2021).
2. V. Zalles et al., Sci. Adv. 7, eabg1620 (2021).
3. D. M. Olson et al., Bioscience 51, 933 (2001).
4. P. Potapov et al., Nat. Food 3, 19–28 (2022).
5. K. E. Giller et al., Food Secur. 13, 1073–1099 (2021).
6. Y. Fan, H. Li, G. Miguez-Macho, Science 339, 940–943
(2013).
7. S. Kuppel, J. Houspanossian, M. D. Nosetto, E. G. Jobbágy,
Water Resour. Res. 51, 2937–2950 (2015).
8. N. Ramankutty et al., Annu. Rev. Plant Biol. 69, 789–815
(2018).
9. E. Hauser, P. L. Sullivan, A. N. Flores, D. Hirmas, S. A. Billings,
Earths Futur. 10, e2022EF002897 (2022).
10. B. R. Scanlon, I. Jolly, M. Sophocleous, L. Zhang, Water Resour. Res.
43, W03437 (2007).
11. B. R. Scanlon et al., Proc. Natl. Acad. Sci. U.S.A. 109,
9320–9325 (2012).
12. M. Rodell, I. Velicogna, J. S. Famiglietti, Nature 460, 999–1002
(2009).
13. D. J. MacAllister, G. Krishan, M. Basharat, D. Cuba,
A. M. MacDonald, Nat. Geosci. 15, 390–396 (2022).
14. C. J. Clarke, R. J. George, R. W. Bell, T. J. Hatton, Soil Res. 40,
93 (2002).
30 June 2023
15. W. M. Edmunds, C. B. Gaye, J. Hydrol. 156, 47–59
(1994).
16. G. Favreau et al., Water Resour. Res. 45, W00A16
(2009).
17. M. D. Nosetto, E. G. Jobbágy, A. B. Brizuela, R. B. Jackson,
Agric. Ecosyst. Environ. 154, 2–11 (2012).
18. R. Giménez, J. Mercau, M. Nosetto, R. Páez, E. Jobbágy,
Hydrol. Processes 30, 2603–2616 (2016).
19. A. D. Nobre et al., J. Hydrol. 404, 13–29 (2011).
20. R. A. Maenza, E. A. Agosta, M. L. Bettolli, Int. J. Climatol. 37,
445–463 (2017).
21. M. A. Lovino, O. V. Müller, E. H. Berbery, G. V. Müller, Global
Planet. Change 168, 78–97 (2018).
22. T. J. Peterson, M. Saft, M. C. Peel, A. John, Science 372,
745–749 (2021).
23. M. Rogger et al., Water Resour. Res. 53, 5209–5219
(2017).
24. A. Alaoui, M. Rogger, S. Peth, G. Blöschl, J. Hydrol. 557,
631–642 (2018).
25. Y. Fan, G. Miguez-Macho, E. G. Jobbágy, R. B. Jackson,
C. Otero-Casal, Proc. Natl. Acad. Sci. U.S.A. 114, 10572–10577
(2017).
26. J. Šimůnek, M. Šejna, H. Saito, M. Sakai, M. T. van Genuchten,
“The HYDRUS-1D Software Package for Simulating the One-
Dimensional Movement of Water, Heat, and Multiple Solutes in
Variably-Saturated Media” (Department of Environmental
Sciences, University of California–Riverside Research Reports,
2005).
27. J. I. Whitworth-Hulse et al., Agric. Ecosyst. Environ. 349,
108440 (2023).
28. Z. P. Wang, C. W. Lindau, R. D. Delaune, W. H. Patrick Jr., Biol.
Fertil. Soils 16, 163–168 (1993).
29. J. Houspanossian et al., Theor. Appl. Climatol. 131, 111–120
(2016).
30. P. Tittonell et al., Front. Sustain. Food Syst. 4, 584605
(2020).
31. L. A. Garibaldi et al., Conserv. Lett. 14, e12773 (2021).
32. S. Wang, J. Li, Can. J. Rem. Sens. 42, 190–202 (2016).
33. E. G. Jobbágy et al., Hydrol. Processes 35, e14188
(2021).
34. M. D. Nosetto et al., Agric. Water Manage. 129, 120–129
(2013).
35. M. L. Amdan, R. Aragón, E. G. Jobbágy, J. N. Volante,
J. M. Paruelo, Water Resour. Res. 49, 6601–6612 (2013).
36. M. D. Nosetto, R. A. Paez, S. I. Ballesteros, E. G. Jobbágy,
Agric. Ecosyst. Environ. 206, 60–70 (2015).
37. S. Pal et al., Water Resour. Res. 57, e2021WR029704
(2021).
38. J. Houspanossian et al., Supplementary Materials for:
Agricultural expansion raises groundwater and increases
flooding in the South American plains, data set, Zenodo
(2022); https://doi.org/10.5281/zenodo.7962455.
AC KNOWLED GME NTS
Funding: We thank the UUKi Rutherford Fund Strategic Partner
Grants; the UK Department for Business, Energy & Industrial
Strategy (BEIS); CONICET (PIP 363/2020); and the ANPCyT
(PICT 504/2020) for funding this work. We also acknowledge a
grant from the Inter-American Institute for Global Change
Research (IAI) SGP-HW 056 (GovernAgua Project). Author
contributions: Conceptualization: J.H. and E.G.J. Methodology:
J.H., R.G., W.T., J.I.W.-H., M.D.N., P.M.A., and M.C.R. Investigation:
J.H., R.G., and E.G.J. Visualization: J.H., W.T., J.I.W.-H., and E.G.J.
Funding acquisition: M.C.R. and E.G.J. Project administration:
J.H., M.C.R., and E.G.J. Supervision: M.C.R. and E.G.J. Writing –
original draft: J.H., W.T., M.C.R., and E.G.J. Writing – review &
editing: J.H., P.M.A., M.C.R., and E.G.J. Competing interests: The
authors declare that they have no competing interests. Data
and materials availability: All data and code are deposited at
Zenodo (38). License information: Copyright © 2023 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.add5462
Materials and Methods
Figs. S1 to S19
Tables S1 and S2
References (39–66)
Submitted 25 July 2022; accepted 24 May 2023
10.1126/science.add5462
5 of 5
RES EARCH

MICROBIOLOGY antileishmanial lead compounds that are ad-

vancing to human clinical trials are the 20S
Cyanotriazoles are selective topoisomerase II poisons proteasome (8, 10, 11), cdc2 related kinase 12
(CRK12) (12), and cleavage and polyadenylation
that rapidly cure trypanosome infections specificity factor 3 (CPSF3) (9); these may offer
potential for use against the human trypano-
Srinivasa P. S. Rao1,2,3*, Matthew K. Gould4, Jonas Noeske2, Manuel Saldivia1,2, Rajiv S. Jumani1,2, somiases, too.
Pearly S. Ng3, Olivier René1,2, Yen-Liang Chen1,2,3, Marcel Kaiser5,6, Ryan Ritchie4, To identify growth inhibitors for trypano-
Amanda Fortes Francisco7, Nila Johnson1, Debjani Patra1,2, Harry Cheung1,2, Colin Deniston8, somatids, we performed whole-cell high-
Andreas D. Schenk9, Wilian A. Cortopassi2, Remo S. Schmidt5,6, Natalie Wiedemar5,6, throughput screening of the Novartis compound
Bryanna Thomas1,2, Rima Palkar1, Nahdiyah A. Ghafar3, Vanessa Manoharan3, Catherine Luu2, collection. Hit compounds belonging to the
Jonathan E. Gable1,2, Kah Fei Wan3, Elmarie Myburgh10, Jeremy C. Mottram11, Whitney Barnes8, cyanotriazole (CT) class were further optimized
John Walker8, Charles Wartchow2, Natasha Aziz1,2, Colin Osborne1,2, Juergen Wagner3,9, through medicinal chemistry to improve phar-
Christopher Sarko1,2, John M. Kelly7, Ujjini H. Manjunatha1,2,3, Pascal Mäser5,6, Jan Jiricek1,3, macological properties and allow the evalua-
Suresh B. Lakshminarayana1,2,3, Michael P. Barrett4*, Thierry T. Diagana1,2,3* tion of the lead molecules in in vitro and in vivo
models of HAT and Chagas disease. We iden-
Millions who live in Latin America and sub-Saharan Africa are at risk of trypanosomatid infections, tified the CT target in trypanosomatids using
which cause Chagas disease and human African trypanosomiasis (HAT). Improved HAT treatments are phenotypic, genomic, and metabolomic ap-
available, but Chagas disease therapies rely on two nitroheterocycles, which suffer from lengthy proaches and further validated the target
drug regimens and safety concerns that cause frequent treatment discontinuation. We performed using genetic, biochemical, and biophysical
phenotypic screening against trypanosomes and identified a class of cyanotriazoles (CTs) with potent approaches. We ultimately used cryo–electron
trypanocidal activity both in vitro and in mouse models of Chagas disease and HAT. Cryo–electron microscopy microscopy (cryo-EM) to resolve the atomic
approaches confirmed that CT compounds acted through selective, irreversible inhibition of trypanosomal structure of CT binding to its cognate trypano-
topoisomerase II by stabilizing double-stranded DNA:enzyme cleavage complexes. These findings somal target to provide structural insight into
suggest a potential approach toward successful therapeutics for the treatment of Chagas disease. the compound class’s potency and selectivity.

CTs are pan-kinetoplastid inhibitors with rapid

M
ore than 1 billion people who live in trop-
ical and subtropical regions worldwide
are at risk of infection from kinetoplas-
tid parasites. For example, Trypanosoma
cruzi is the etiological agent of Chagas
disease (1); Trypanosoma brucei subspecies
cause human African trypanosomiasis (HAT),
or sleeping sickness (2); Leishmania donovani
causes visceral leishmaniasis (3); and several
other Leishmania species underlie cutaneous
leishmaniasis (4). The current treatments are
decades old and suffer from toxicity and lim-
1
Novartis Institute for Tropical Diseases, Emeryville, CA, USA.
2
Novartis Institutes for BioMedical Research, Emeryville, CA,
USA. 3Novartis Institute for Tropical Diseases, Singapore.
4
College of Medical Veterinary and Life Sciences, University
of Glasgow, Glasgow, UK. 5Swiss Tropical and Public Health
Institute, Allschwil, Switzerland. 6Faculty of Science,
University of Basel, Basel, Switzerland. 7London School of
Hygiene and Tropical Medicine, London, UK. 8Novartis
Institutes for BioMedical Research, San Diego, CA, USA.
9
Novartis Institutes for BioMedical Research, Basel,
Switzerland. 10York Biomedical Research Institute, Hull York
Medical School, University of York, York, UK. 11York
Biomedical Research Institute, Department of Biology,
University of York, York, UK.
*Corresponding author. Email: srinivasa.rao@novartis.com
(S.P.S.R.); michael.barrett@glasgow.ac.uk (M.P.B.);
thierry.diagana@novartis.com (T.T.D.)
ited efficacy. A fledgling drug pipeline, powered
by public-private partnerships, has addressed
some unmet medical needs for HAT (5–7) and,
to some level, visceral leishmaniasis (8–12). How-
ever, progress against Chagas disease has been
scant (13, 14). The two standard Chagas dis-
ease therapies, benznidazole and nifurtimox,
are poorly tolerated, and the treatment regi-
mens are lengthy and often discontinued be-
cause of adverse effects (1). This leaves patients
vulnerable to life-threatening cardiac and gastro-
intestinal complications associated with chronic
T. cruzi infection (1). More effective, safer, and
shorter-duration therapeutics for Chagas dis-
ease are urgently needed.
Assay miniaturization and automation have
enabled large chemical library screens against
causative agents of neglected tropical diseases.
Over the past two decades, such phenotypic
screening approaches have yielded notable
advances against malaria (15), tuberculosis
(16), leishmaniasis, and the African trypano-
somiases (8–12, 17). This approach has enabled
lead chemical entities to enter clinical trials
and the discovery of mechanisms of action
and drug targets. Among the targets for such
sterilizing activity
Identification of CTs
We performed a whole cell–based growth in-
hibition screen of the Novartis compound li-
brary against bloodstream form T. brucei brucei,
which identified CT0 as a potent growth inhib-
itor that was also active against Trypanosoma
brucei gambiense and Trypanosoma brucei
rhodesiense, pathogens that cause HAT (Table 1
and Fig. 1). CT0 belongs to the CT compound
class that also proved to be potent growth in-
hibitors of a panel of kinetoplastid parasites,
which includes T. cruzi and L. donovani (Table 1
and fig. S1, A and B). CTs were noncytotoxic
(Table 1; Fig. 1; and fig. S1, C and D) and not
active against a range of nontrypanosomatid
microbial pathogens (table S1). In a prelimi-
nary structure-activity analysis, analogs of CT0
were amenable for optimization and retained
cellular potency against multiple kinetoplastid
parasites, which indirectly suggests that CTs
may have a conserved mechanism of action
across all kinetoplastids (fig. S1, A and B).
Despite its high potency, CT0 displayed lim-
ited solubility and unfavorable physicochem-
ical properties, which resulted in poor oral

Fig. 1. CTs for trypanosomatids. Chemical structures.

Rao et al., Science 380, 1349–1356 (2023) 30 June 2023 1 of 8

RES EARCH | R E S E A R C H A R T I C L E
bioavailability and no measurable brain ex-
posure, which is necessary for curing stage II
HAT (Table 1 and Fig. 1). To obtain a suitable
lead molecule for evaluation in rodent dis-
ease models, medicinal chemistry efforts opti-
mized properties of the CT scaffold for oral
dosing and brain penetration. Removing the
hydrogen-bond donor of the aminomethyl-
furan core of CT0 through cyclization led to an
aza-isoindoline scaffold. Further diminishing
the cyanophenyl substituent’s polarity resulted
in molecule CT1 with improved solubility, oral
bioavailability, and moderate free brain expo-
sure (Table 1 and Fig. 1). Subsequently, re-
placement of the aza-isoindoline core with an
isoindoline moiety and restoring polarity by in-
troducing a 5-fluoro-2-(trifluoromethyl)pyridine
substituent on the left-hand side of the mol-
ecule led to molecule CT3. This nitrogen shift
strategy allowed CT3 to preserve solubility
while displaying excellent oral bioavailability
and free brain exposure (Table 1 and Fig. 1),
which yielded pharmacokinetic properties
suitable for profiling in vivo in various ani-
mal models of human trypanosomatid dis-
eases (table S2).
In vitro and in vivo potency of CTs
Rapid onset of action and a short dosing regi-
men are desirable properties for anti-infective
therapies. These properties typically require
drugs with fast-sterilizing activity that is time
and concentration dependent. Rate of kill
and wash-off experiments revealed time- and
concentration-dependent sterilizing activity of
CT3 that compared favorably with the standard-
of-care drugs for Chagas disease and HAT,
which indicates a potential to act fast and offer
shorter drug regimens than those of the cur-
rent therapies (fig. S2, A to D).
In vitro to in vivo translatability of CT3 ster-
ilizing activity was evaluated in mouse models
of trypanosomatid diseases. Chagas disease
requires complete parasite clearance in vivo to
prevent relapse from residual parasites, as
seen with posaconazole in mouse models of
infection (18) and during human clinical trials
(19). We thus tested CT3 in a bioluminescent
mouse model of chronic Chagas disease, with
immunosuppression to confirm sterile cure
(20). In this model, dosing at 30 mg · kg−1 once
daily for 5 days reduced parasitemia beyond
the limit of detection. Parasites remained un-
detectable in all treated mice even after three
rounds of immunosuppression, which sug-
gests that a complete and sterile cure had
been achieved. Ex vivo imaging of selected
organs and tissues confirmed the absence of
detectable parasites (Fig. 2A). Pharmacoki-
netic analysis of treated mice showed that
the free blood concentration of CT3 was above
the median effective concentration (EC50) for
~17 hours (fig. S3A). CT3 was also evaluated
in the more stringent acute model of infec-
Rao et al., Science 380, 1349–1356 (2023) 30 June 2023
Table 1. CTs inhibit growth of trypanosomatid parasites in vitro. Antiparasitic activity, cytotoxicity,
physicochemical properties, and pharmacokinetics of CT compounds. HepG2, human hepatocellular
carcinoma cells; NIH 3T3, fibroblast cells; Kp, brain to plasma ratio, total; Kpuu, brain to plasma ratio,
unbound (corrected by mouse plasma protein binding and rat brain tissue binding); EC50 and CC50
represent the half-maximum growth inhibition concentration against bloodstream form of T. brucei,
intracellular T. cruzi amastigotes and extracellular L. donovani amastigotes, and mammalian cell lines.
Growth inhibition measurements were generated with two or more independent biological replicates
and with two technical replicates. Values shown are geometric mean ± SD. Physicochemical properties
such as polar surface area and hydrogen-bond donors were calculated by using Medchem Focus software.
In vivo pharmacokinetic properties for CT compounds were determined by using mice (n = 3). Oral
bioavailability was determined by using dose-normalized exposures after oral and intravenous dosing of
compounds, and Kp was measured 1 hour after intravenous dosing in mice.
CT0 CT1 CT3
In vitro antiparasitic activity EC50 (nM)
.....................................................................................................................................................................................................................
T. b. brucei 66 ± 26 160 ± 35 180 ± 58
.....................................................................................................................................................................................................................
T. b. gambiense 94 ± 25 110 ± 11 31 ± 15
.....................................................................................................................................................................................................................
T. b. rhodesiense 16 ± 3 22 ± 5 57 ± 14
.....................................................................................................................................................................................................................
T. cruzi 69 ± 15 87 ± 27 78 ± 19
.....................................................................................................................................................................................................................
L. donovani 497 ± 70 1350 ± 1200 134 ± 52
.....................................................................................................................................................................................................................
Cytotoxicity CC50 (nM)
.....................................................................................................................................................................................................................
HepG2 14,600 24,000 >50,000
.....................................................................................................................................................................................................................
NIH 3T3 >20,000 >20,000 >20,000
.....................................................................................................................................................................................................................
Physicochemical properties
.....................................................................................................................................................................................................................
Polar surface area 121 88 88
.....................................................................................................................................................................................................................
Hydrogen-bond donors 1 0 0
.....................................................................................................................................................................................................................
Solubility (pH 6.8, mM) <2 230 34
.....................................................................................................................................................................................................................
Mouse in vivo pharmacokinetics
.....................................................................................................................................................................................................................
Oral bioavailability (%) 5 80.8 100
.....................................................................................................................................................................................................................
Kp/Kpuu <0.05/<0.05 0.7/0.5 0.7/1.4
.....................................................................................................................................................................................................................
tion with higher parasite burden, in which velopment for HAT, in which its long half-life
benznidazole, the standard of care, fails to has enabled a cure, even of stage II disease,
achieve cure with a dose of 100 mg · kg−1 once with a single dose (6).
daily for 5 days. However, CT3 showed com-
plete sterile cure when dosed at 30 mg · kg−1 CT-treated parasites showed defects in
once daily for 5 days (fig. S4A). nuclear DNA replication, exhibited
We also assessed CT3 in mouse models of up-regulated DNA damage response
HAT (17, 21), which also require parasite erad- Understanding the mechanism of action and
ication to achieve complete cure. Treatment identifying drug target(s) facilitates medicinal
with 10 mg · kg−1 CT3 once daily for 4 days chemistry through structure-based optimiza-
cured the stage I (hemolymphatic) infection tion that helps design molecules with better
(fig. S4B), whereas 7 days’ treatment with potency and selectivity. We used a multipronged
10 mg · kg−1 once daily was required to cure strategy—which included phenotypic, metab-
the stage II (meningoencephalitis) infection olomic, genetic, and biochemical approaches—
(fig. S4C). Given the rapid in vitro sterilizing to deconvolute the CTs’ target.
activity (6 hours) of CT3 treatment during wash- First, the effect of 6 hours of treatment with
off assays (fig. S2D), we also evaluated the abil- five times EC50 of CT1 and CT3 on cell cycle
ity to achieve single-dose cure in both mouse progression was examined in bloodstream
models described above. CT3 achieved relapse- form T. b. brucei. DNA staining that used
free sterile cure after single-dose oral admin- 4′,6-diamidino-2-phenylindole (DAPI) showed
istration of 40 mg · kg−1 in stage I (fig. S4B) and arrest in S-phase, denoted by an increased 2K1N
100 mg · kg−1 in stage II HAT mouse models population, compared with untreated con-
(Fig. 2B). Ex vivo analysis of brains after treat- trol (fig. S5A), which suggests that CTs af-
ment showed complete clearance of parasites. fected nuclear, but not kinetoplast, DNA
Pharmacokinetic analysis of mice treated with replication. We used ethynyl deoxyuridine
a single dose of 100 mg · kg−1 CT3 showed that (EdU) (a thymidine analog that incorporates
the free brain concentration was above EC50 for into newly synthesized DNA) staining to
>24 hours (fig. S3B). Acoziborole, an oxaborole probe the impact of CTs on DNA synthesis.
molecule, has completed phase 2/3 clinical de- Two hours treatment of T. cruzi with 50× EC50
2 of 8
RES EARCH | R E S E A R C H A R T I C L E

Figure 2

Rao et al., Science 380, 1349–1356 (2023) 30 June 2023 3 of 8

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

Fig. 2. CTs are efficacious against trypanosomatid parasites in vivo.
(A) In vivo assessment of nontreated, benznidazole (BZ)–treated, and CT3-
treated mice during chronic-stage T. cruzi infections as seen in the outline
of experiment (top left). BALB/c mice were given vehicle (n = 3), BZ
(100 mg ∙ kg−1) (n = 3), and CT3 (30 mg ∙ kg−1) (n = 6) starting on day 102
(orally, 5 days, once daily). Ventral images of each mouse are shown at different
time points after infection. All BZ- and CT3-treated mice were immunosup-
pressed on days 116, 120, and 124 by using cyclophosphamide (200 mg ∙ kg−1
intraperitoneally). All treated mice were bioluminescence negative at the
experimental endpoint (day 141) and assessed further through ex vivo imaging
(bottom left panel and central ex vivo organs and tissue display). At this
point, as an example, organs from nontreated mouse 3 were removed and
assessed for bioluminescence. There was infection in the lungs, mesentery,
and cecum (black arrows). All treated mice were cured. Heatmaps are
on log10 scales and indicate the intensity of bioluminescence from low (blue)
to high (red). DPI, days post-infection. (B) In vivo efficacy of CT3 in
meningoencephalitis mouse model of HAT. After 21 days of infection with a
bioluminescent T. b. brucei GVR35 strain, three mice were treated intra-
peritoneally with a single dose of 40 mg ∙ kg−1 of diminazene (negative
control), and six mice were treated orally with one dose of 100 mg ∙ kg−1 of
CT3. Parasitemia levels were monitored by imaging mice after administration
of D-luciferin (150 mg ∙ kg−1, subcutaneously). Diminazene-treated mice
relapsed by day 42 and were humanely euthanized, whereas CT3-treated
mice remained parasitemia free for >92 days; no brain signal was observed
during ex vivo imaging on day 160 (mouse 3 died because of natural
causes in day 144; no parasitemia was detected). In both animal models, the
presence of parasites is represented by bioluminescence radiance as
indicated with the color scales.

of CT3 prevented EdU incorporation into nu-
clear DNA, which confirmed that CTs hinder
nuclear but not kinetoplastid DNA replication
(Fig. 3A).
We also performed untargeted metabo-
lomics, which has helped assign the drug
mode of action in kinetoplastids (22, 23). Af-
ter exposing T. b. brucei to CTs (CT1 and CT3),
more than 600 metabolites were detected
with mass spectrometry, and profound in-
creases in deoxynucleotide abundance were
observed (fig. S6 and data S2), a profile sim-
ilar to that of Staphylococcus aureus after ex-
posure to fluoroquinolone inhibitors of DNA
gyrase (24).
Because DNA replication appeared to be af-
fected, we next evaluated whether CTs cause
DNA damage in trypanosomatids. Immuno-
blotting with antibodies against phosphoryl-
ated histone (gH2A), a marker of double-strand
DNA (dsDNA) breaks (25), revealed increases
in gH2A in T. b. brucei (fig. S5B). In situ fluo-
rescence microscopy also showed increased
gH2A in both T. b. brucei and T. cruzi exposed
to CTs, whereas GNF6702, a known kineto-
plastid proteasome inhibitor, did not induce
increased gH2A (Fig. 3, B and C). Moreover, T. b.
brucei mutants, which lacked either one of
the dsDNA break repair enzymes BRCA2 or
RAD51, were hypersensitive to CT compounds,
which suggests that they either interfered
with the cellular DNA strand-breakage repair
machinery or caused DNA damage that is en-
hanced when the repair machinery is disrupted
(table S3).
Drug resistance to CTs involved mutations
in the topoisomerase IIa gene of T. b. brucei
Drug-resistant mutants were independently
selected against two different CTs (CT1 and CT4)
in T. b. brucei in vitro (table S4). Resistant clones
were isolated by using limiting dilution, and
each showed substantial resistance to CT mol-
ecules while retaining sensitivity to other known
trypanocides (Fig. 4A and fig. S7). Whole-
genome sequencing of mutants revealed multi-
ple single-nucleotide polymorphisms (SNPs)
in each individual mutant, but three of the
four mutant strains (CT1.1, CT1.2, and CT4.1)
carried an F540L SNP in the topoisomerase II
alpha (topoIIa) gene (data S2). All other SNPs
were specific to individual parasite mutant
lines. Sanger sequencing of the topo II region
of the mutants confirmed a homozygous F540L
mutation in the resistant lines CT1.1, CT1.2, and
CT4.1 (fig. S8A). Sanger sequencing of CT4.2
parasites revealed, in the N-terminal encoding
region of topoIIa, multiple heterozygous muta-
tions that code for the matching amino acids
found in topoIIb (fig. S8B). Independent CT-
resistant mutants were also generated against
a third CT compound (CT5) (table S4), and
Sanger sequencing of the topoIIa gene region
identified homozygous SNPs that encoded the
L491F and S492N mutations in CT5.1 and the
G598S mutation in CT5.2 (fig. S8A). (Single-
letter abbreviations for the amino acid residues
are as follows: C, Cys; F, Phe; G, Gly; I, Ile; L, Leu;
R, Arg; and S, Ser. In the mutants, other amino
acids were substituted at certain locations; for
example, F540L indicates that phenylalinine
at position 540 was replaced by leucine.)
T. b. brucei has two topoisomerase II en-
zymes, TopoIIa and TopoIIb, and RNA inter-
ference (RNAi) experiments have shown that
only topoIIa is essential for T. b. brucei blood-
stream form survival (26). The depletion of
topoIIa by RNAi affects nuclear but not kineto-
plast DNA replication, which is similar to our
observations with CT-treated parasites. Sequence
alignment of TopoIIa and TopoIIb shows that
the N-terminal region is highly conserved, where-
as the C-terminal region after amino acid 1067
is more divergent (fig. S9). F540 is present in
both TopoII isoforms, whereas L491, S492,
and G598 in TopoIIa are F491, N492, and S598,
respectively, in TopoIIb. The SNPs identified
in topoIIa in the resistant lines CT5.1 and CT5.2
corresponded to amino acids found in wild-
type TopoIIb.
The biologically active form of TopoIIa is a
homodimer. All TopoIIa mutations found in
the drug-resistant mutants clustered in the pro-
tein’s Toprim domain (Fig. 4B), which is essen-
tial for DNA cleavage and religation activity
(27, 28). Residue F540 is conserved in kineto-
plastids, yeast, and humans (fig. S10) pointing
into the hydrophobic core of the Toprim domain,
whereas G598 is located at the topoisomerase
dimer interface (Fig. 4B) (29, 30). Furthermore,
comparison with structures of the full-length
yeast TopoII reveals that residues equivalent
to L491 and S492 lie proximal to residues in
the adenosine triphosphatase (ATPase) trans-
ducer domain structures (29), whereas in the
pre-open state structure of the full-length hu-
man topoisomerase IIa (hTOP2A), the equiv-
alent residues are proximal to the K-loop of
the transducer domain (30). The K-loop is
important in coupling DNA binding with
adenosine 5′-triphosphate (ATP) turnover and
strand passage (29). The presence of TopoIIa
mutations in multiple T. b. brucei CT-resistant
mutants points to TopoIIa as a potential tar-
get, which is consistent with genetic, pheno-
typic, and metabolomic observations.
We assessed the impact of reduced expression
of TopoIIa by means of doxycycline-inducible
RNAi. Because topoIIa is essential, we achieved
partial rather than full suppression with tran-
sient induction of RNAi by pulsing a range of
doxycycline concentrations to allow gene ex-
pression to recover before the enzyme’s ab-
sence led to cell death. The higher the dose of
doxycycline in pulsed induction of RNAi, the
less sensitive cells became to CTs (table S5).
This is consistent with observations in mam-
malian cells exposed to TopoII poisons, for
which reduced enzyme activity diminishes
dsDNA break formation (31).
To further confirm TopoIIa as the target for
CTs, we introduced one of the drug-resistant
allele mutations into T. b. brucei using the
CRISPR-Cas9 system (fig. S11A). The resulting
CRISPR-edited parasites were selected for CT
drug resistance at five times its EC50 against
the wild type (fig. S11B). Parasites that were
transfected with the mutating DNA repair tem-
plate and induced to express the Cas9 protein
recovered within days, whereas the control
transfection with no Cas9 expression did not

Rao et al., Science 380, 1349–1356 (2023) 30 June 2023 4 of 8

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

Fig. 3. CTs induce a DNA damage response in trypanosomatid parasites. (A) Evaluation of
active DNA replication by using EdU incorporation in T. cruzi parasites. Vero cells were infected
with T. cruzi CL-Brener tdTomato strain for 48 hours, followed by 2 hours of treatment with
DMSO (no drug), 50× EC50 (4 mM) CT3, or benznidazole (BZN, 40 mM). Representative images of
intracellular T. cruzi expressing tdTomato (red), Hoechst-stained nuclei (blue), EdU-labeling
(green) are shown. There was a reduction in EdU incorporation into T. cruzi parasites incubated
with CT3 compared with the no-drug control and BZN-treated cultures. Images are representative
of at least three independent experiments. Scale bar, 5 mm. (B) Quantification of gH2A induction
in T. b. brucei treated with CT3 and other compounds. Representative microscopic images of
T. b. brucei, DAPI-stained DNA (red), T. b. brucei gH2A antibody staining (green), and merged image
suggesting localization of gH2A induction with nuclear DNA (yellow) are shown. Parasites were
treated with DMSO (no drug control), 11× EC50 (2 mM) of CT3, 20 mM phleomycin (DNA damaging
agent), or 285× EC50 (20 mM) of GNF6702 (kinetoplastid proteasome inhibitor) for 4 hours.
Data show mean and SD of at least four independent biological replicates. Scale bar, 5 mm. DIC,
differential interference contrast; Phleo., phleomycin. (C) Quantification of gH2A induction in
T. cruzi treated with CT3 and other compounds. Representative images of gH2A induction (green)
in T. cruzi intracellular amastigotes that express tdTomato (red) are shown. Vero cells were
infected with T. cruzi for 48 hours, followed by 24 hours of DMSO (no drug control), 10× EC50
(740 nM) of CT3, 25× EC50 (20 mM) of BZN, or 200× EC50 (20 mM) of GNF6702. Data show mean
and SD of at least four independent experiments. Scale bar, 5 mm. Statistical analysis was
carried out by using Student’s t test. ns, nonsignificant; **P < 0.01; ****P < 0.0001.

grow under selective pressure. EC50 values
with a range of CT compounds confirmed that
the CRISPR-edited parasites were highly resist-
ant to CTs (Fig. 4C). Whole-genome sequencing
of the CRISPR-edited parasites identified no
mutations outside the intended topoII gene
cluster. Sanger sequencing revealed that both
topoIIa and topoIIb genes carried the F540L
mutation (expected because the guide-RNA
target sequence and homology flanks of the
transfected repair template are conserved in
both genes). The TopoIIa sequence also re-
vealed other changes in the vicinity of F540L,
including L491F and S492N, which were ob-
served in the TopoIIa sequence of other CT-
resistant mutants and also the wild-type sequence
of TopoIIb (Fig. 4C). Thus, in selecting CRISPR-
edited parasites for CT resistance, a domain
swap may have introduced a section of the
topoIIb gene into the topoIIa gene in addition
to the F540L mutation.
CTs selectively inhibited the DNA relaxation
activity of trypanosome TopoII by stabilizing
the dsDNA:enzyme complex
TopoII inhibitors are divided into two broad
classes: catalytic inhibitors and TopoII poi-
sons. The latter stabilize the covalent DNA:
enzyme complex, which leads to dsDNA breaks
(32). To characterize the inhibitory activity of
the CTs, we expressed and purified T. cruzi
TopoII (Tc TopoII) using a baculovirus expres-
sion system. Tc TopoII shares 82% identity
with T. b. brucei over the first 1181 amino acids
(in contrast to the mammalian ortholog, which
diverges substantially in this region) (fig. S10).
Residues F540 and G598, which were detected
as mutations in the T. b. brucei drug-resistant
mutants, are conserved in all kinetoplastids,
whereas TbTopoIIa L491 and S492 residues are
isoleucine (I491) and arginine (R492), respec-
tively, in T. cruzi (fig. S10).
We measured the effect of CTs on Tc TopoII
using a Topo2 cleavage assay, which mea-
sures the stabilization of dsDNA breaks. The

Rao et al., Science 380, 1349–1356 (2023) 30 June 2023 5 of 8

RES EARCH | R E S E A R C H A R T I C L E
Fig. 4. CT-resistant mutants show mutation in topoisomerase IIa.
(A) Concentration-response curves for T. b. brucei wild-type (WT) and CT-resistant
bloodstream forms in the presence of increasing concentrations of CTs. (Left)
Mutants showed a significant EC50 shift (represented in fold-over WT) when treated
with CT compounds, compared with proteasome inhibitor (GNF6702), CLK1 inhibitor
Rao et al., Science 380, 1349–1356 (2023) 30 June 2023
(AB1), and suramin, a known antitrypanosomal agent. (Right) Representative
EC50 curves are shown for CT1, CT3, and CT5. EC50 values of two biological
replicates were calculated from concentration-response curves performed in
duplicate after nonlinear fitting with GraphPad Prism. Statistical analysis was
carried out by using Student’s t test. ns, nonsignificant; *P < 0.05; **P < 0.01;
6 of 8
RES EARCH | R E S E A R C H A R T I C L E

***P < 0.001; ****P < 0.0001. (B) Schematic representation of the topoisomerase II
domain structure. Triangles highlight positions of homozygous mutations. These
mutations were mapped onto the yeast full-length topoisomerase II structure (PDB
ID 4GFH) in magenta and labeled with their type and residue number from the T. b.
brucei sequence. One monomer of topoisomerase II is colored in cyan, and the
second monomer is colored in green. GHKL (gyrase, Hsp90, histidine kinase, MutL);
WHD (winged helix domain). (C) Antiparasitic activity of CT compounds against
parental and CRISPR T. b. brucei strains. Data presented were generated with
a minimum of four independent biological replicates. Statistical analysis was carried
out by using Student’s t test (***P < 0.001). Sanger sequencing of the parental and
CRISPR strains of T. b. brucei is shown. Sequencing of the topoIIa and topoIIb
gene fragments around the mutated residue showed that both the a and b variants
had introduced the mutation, with the a variant appearing to have also converted
to the b sequence upstream. This suggests a probable recombination event with an
initial introduction of the mutation into the beta variant, followed by recombination to
introduce the mutation plus surrounding residues to the a locus under drug
selection. Triangles are homozygous mutations, and circles are heterozygous
mutations seen with CT-resistant mutants. Cpd, compound.

Fig. 5. Biochemical and struc-

tural validation of TopoII as
a target of CTs. (A) Biochemical
characterization by using TopoII
DNA cleavage assay in the
absence of ATP. Full-length
T. cruzi TopoII was incubated
with supercoiled DNA (scDNA) in
the presence of varying concen-
trations (micromolar) of CT
compounds, and the ability to
induce dsDNA breaks was
measured by running products
on a 1.2% agarose gel. The
experiment was conducted with
two biological replicates; one of
the representative images is
shown. CT1 caused increasing
levels of linear dsDNA (asterix)
to appear as the concentration
increased. (B) Representative
SPR sensorgram of SPR experi-
ment showing the association of
truncated T. cruzi topoIIDC 2
to 1182 with immobilized dsDNA
in the presence of CT1 (green) or
DMSO control (blue). The 1:1
model of the response is shown
(black). Rmax, CT1-2263; DMSO-
2200; RU, response units. (C) CTs
bind to the DNA cleavage site,
as revealed with cryo-EM. View of
the binding site of the TcTopoII-
DNA complex bound to CT1
(shown in cyan) is given. CT1
intercalates between the DNA
base pairs at the DNA cleavage
site and forms a covalent bond
with the reactive cysteine C477
(shown in magenta), which is
conserved in trypanosomatid
TopoII. C, cysteine. (D) Etoposide
(from PDB ID 5GWK, shown in
yellow) superimposed on the CT1
cyro-EM structure reveals that CT1
binds to the same binding site
as etoposide, which is a known
human TOP2A inhibitor.
(E) Mutations found in T. b.
brucei–resistant mutants (CT1.1,
CT1.2, CT4.1, CT5.1, and CT5.2)
are highlighted as magenta
spheres on the cryo-EM structure of the Tc TopoII DNA binding domain (amino acids 401 to 1178) bound to DNA and CT1. Tc TopoII monomers are colored in cyan and green, and
the DNA backbone is colored in orange. (F to I) Close up views of the different resistant mutation sites with the mutated amino acids shown in magenta in stick representation.

Rao et al., Science 380, 1349–1356 (2023) 30 June 2023 7 of 8

RES EARCH | R E S E A R C H A R T I C L E
addition of CTs (CT1 and CT3) to Tc TopoII pro-
moted selective accumulation of linearized
DNA, which is indicative of CT stabilizing the
topoisomerase dsDNA cleavage complex (Fig.
5A and figs. S12 and S13). By contrast, a phar-
macologically inactive, closely related analog
(CT2) did not result in linearized DNA accu-
mulation by Tc TopoII (Fig. 5A and table S4).
CTs (CT1 and CT3) did not affect human
TOP2A cleavage activity (fig. S14). The effect
of CT1 on the binding kinetics of Tc TopoII to
DNA was determined by using surface plas-
mon resonance (SPR). The equilibrium disso-
ciation constant (Kd) of Tc TopoII binding to
immobilized 36–base pair dsDNA decreased by
~1000-fold from 1.33 mM to 1 nM in the presence
of CT1, which suggests a stabilization of the
topoisomerase DNA cleavage complex as ob-
served in the DNA cleavage assay. The disso-
ciation rate constant (kd) of the protein-DNA
complex decreased by >130-fold in the presence
of CT1 compared with the dimethyl sulfoxide
(DMSO) control, which indicates that the in-
crease in binding affinity for this complex is
primarily driven by a slower kd (Fig. 5B).
CTs interact covalently with C477
and bind at the DNA cleavage site
Mechanistically, CTs belong to the class of
topoisomerase poisons characterized by their
stabilizing effect on the topoisomerase cleavage
complex. These include clinically used drugs
for cancer (such as etoposide) and bacterial
infections (such as fluoroquinolones) (33). To
provide a structural rationale for the mode of
action of the CTs, we solved the cryo-EM struc-
ture of the Tc TopoII DNA binding domain
(amino acids 401 to 1178) bound to dsDNA and
CT1 at ~2.9-Å resolution (figs. S15 and S16).
EM density for CT1 was observed at the DNA
cleavage site that extends to C477 (Fig. 5C and
fig. S16, A and D). The trifluoromethylphenyl
portion of CT1 intercalated between the DNA
base pairs at the cleavage site and occupied a
space similar to etoposide bound to hTOP2A
[Protein Data Bank (PDB) ID 5GWK] (Fig. 5D)
(34) and fluoroquinolones bound to bacterial
gyrase (fig. S16B) (35). The CT portion of CT1 is
positioned near C477, which allows the forma-
tion of a covalent bond with the nucleophilic
thiol of the cysteine, represented by continu-
ous EM density between CT1 and C477 (fig. S16,
A and D). Cyanoazole heterocycles are known to
form reversible covalent interactions with cys-
teines (36). C477, which is conserved in trypano-
somatid TopoII, provides the basis for CT
compound selectivity (the equivalent residue
is asparagine in human TOP2A). The muta-
tions found in T. b. brucei strains selected for
CT resistance are highlighted in the cryo-EM
structure of Tc TopoII (Fig. 5, E to I), and res-
idues L491 and S492, which mutate in selected
T. b. brucei CT-resistant mutants, were only
~9 Å from the reactive cysteine.
Rao et al., Science 380, 1349–1356 (2023) 30 June 2023
Topoisomerase II inhibitors as treatment
of trypanosomatid diseases
Our data demonstrated that CT compounds are
potent growth inhibitors of multiple kineto-
plastid parasites. The CTs acted through selec-
tive kinetoplastid topoisomerase II inhibition,
which enabled rapid and complete steriliz-
ing activity in mouse models of Chagas disease
and HAT. Given their therapeutic efficacy,
we are undertaking advanced preclinical pro-
filing of further optimized CT analogs to iden-
tify clinical candidates with a suitable safety
profile. CT3 was relatively safe in preliminary
preclinical safety profiling (table S7). Fur-
thermore, the identification of trypanosomal
TopoII as the molecular target of CTs offers
the opportunity to discover alternative chem-
ical classes through high-throughput screening
and structure-based drug design. We antic-
ipate that these findings will underpin the de-
velopment of potent, safe, and shorter-duration
therapies that will improve the treatment and
management of Chagas disease and other
kinetoplastid parasitic infections.
RE FERENCES AND NOTES
1. J. A. Pérez-Molina, I. Molina, Lancet 391, 82–94
(2018).
2. P. Büscher, G. Cecchi, V. Jamonneau, G. Priotto, Lancet 390,
2397–2409 (2017).
3. S. Burza, S. L. Croft, M. Boelaert, Lancet 392, 951–970
(2018).
4. R. Reithinger et al., Lancet Infect. Dis. 7, 581–596
(2007).
5. E. A. Dickie et al., Trop. Med. Infect. Dis. 5, 29 (2020).
6. V. K. Betu Kumeso et al., Lancet Infect. Dis. 23, 463–470
(2022).
7. V. K. B. K. Mesu et al., Lancet 391, 144–154 (2018).
8. S. Khare et al., Nature 537, 229–233 (2016).
9. C. E. Mowbray et al., J. Med. Chem. 64, 16159–16176
(2021).
10. A. Nagle et al., J. Med. Chem. 63, 10773–10781
(2020).
11. S. Wyllie et al., Proc. Natl. Acad. Sci. U.S.A. 116, 9318–9323
(2019).
12. S. Wyllie et al., Nature 560, 192–197 (2018).
13. S. P. S. Rao et al., ACS Infect. Dis. 5, 152–157 (2019).
14. A. M. Padilla et al., Nat. Microbiol. 7, 1536–1546
(2022).
15. Medicines for Malaria Venture, Pipeline of antimalarial drugs;
https://www.mmv.org/mmv-pipeline-antimalarial-drugs.
16. Stop TB Partnership, Working Group on TB drugs, Clinical
pipeline; https://www.newtbdrugs.org/pipeline/clinical.
17. S. P. S. Rao et al., Trop. Med. Infect. Dis. 5, 28
(2020).
18. A. F. Francisco et al., Antimicrob. Agents Chemother. 59,
4653–4661 (2015).
19. I. Molina et al., N. Engl. J. Med. 370, 1899–1908
(2014).
20. M. D. Lewis et al., Cell. Microbiol. 16, 1285–1300 (2014).
21. E. Myburgh et al., PLOS Negl. Trop. Dis. 7, e2384
(2013).
22. P. C. Steketee et al., PLOS Negl. Trop. Dis. 12, e0006450
(2018).
23. A. Trochine, D. J. Creek, P. Faral-Tello, M. P. Barrett, C. Robello,
PLOS Negl. Trop. Dis. 8, e2844 (2014).
24. K. Dörries, R. Schlueter, M. Lalk, Antimicrob. Agents
Chemother. 58, 7151–7163 (2014).
25. L. Glover, D. Horn, Mol. Biochem. Parasitol. 183, 78–83
(2012).
26. T. Kulikowicz, T. A. Shapiro, J. Biol. Chem. 281, 3048–3056
(2006).
27. L. Aravind, D. D. Leipe, E. V. Koonin, Nucleic Acids Res. 26,
4205–4213 (1998).
28. B. H. Schmidt, A. B. Burgin, J. E. Deweese, N. Osheroff,
J. M. Berger, Nature 465, 641–644 (2010).
29. B. H. Schmidt, N. Osheroff, J. M. Berger, Nat. Struct. Mol. Biol.
19, 1147–1154 (2012).
30. A. Vanden Broeck et al., Nat. Commun. 12, 2962
(2021).
31. A. C. Porter, Methods Mol. Biol. 582, 245–263 (2009).
32. K. R. Vann, A. A. Oviatt, N. Osheroff, Biochemistry 60,
1630–1641 (2021).
33. J. L. Delgado, C. M. Hsieh, N. L. Chan, H. Hiasa, Biochem. J.
475, 373–398 (2018).
34. Y. R. Wang et al., Nucleic Acids Res. 45, 10861–10871
(2017).
35. D. A. Veselkov et al., Acta Crystallogr. D Struct. Biol. 72,
488–496 (2016).
36. A. Keeley, P. Ábrányi-Balogh, G. M. Keserű, MedChemComm
10, 263–267 (2018).
AC KNOWLED GME NTS
The authors acknowledge G. Baxley, H. Roughley, L. C. Bok,
Y.-B. Chen, S. Williams, and J. Roquigny for the technical
assistance. The authors also thank D. Beer and F. Blasco for
insightful discussions. The authors are grateful to R. McCulloch
(Glasgow) for providing T. brucei RAD51 and BRCA2 mutants and
D. Horn (Dundee) for the T. brucei inducible Cas9 cell line.
The authors are grateful to the project management, alliance
management and partnering, and legal and finance team (G. Feng,
T. Krucker, M. Hopkins, and J. C. Poilevey). Funding: This research
was supported by the Wellcome Trust (grants 103024/Z/13Z,
104976/Z/14/Z, and 108517/Z/15/Z, 219639/Z/19/Z). M.P.B. and
M.K.G. were also funded by an MRC Newton grant, “Bridging
epigenetics, metabolism and cell cycle in pathogenic
trypanosomatids,” MR/S019650/1 and the Wellcome Trust Wellcome
Centre for Integrative Parasitology (grant no. 104111/Z/14/Z). Author
contributions: Conceptualization: S.P.S.R., M.K.G., U.H.M., J.J.,
M.P.B., and T.T.D.; Methodology: S.P.S.R., M.K.G., J.N., M.S., R.S.J.,
D.P., P.S.N., M.K., R.R., A.F.F., H.C., C.D., A.D.S., W.A.C., C.L., E.M.,
C.W., C.O., J.M.K., U.H.M., and J.J.; Investigation: D.P., B.T., P.S.N.,
J.N., M.S., R.S.J., R.R., O.R., Y.-L.C., A.F.F., N.J., D.P., H.C., C.D., A.D.S.,
W.A.C., R.S.S., N.W., B.T., R.P., N.A.G., V.M., C.L., C.W., K.F.W., E.M.,
and W.B.; Formal analysis: S.P.S.R., M.K.G., J.N., M.S., R.S.J., P.S.N.,
O.R., Y.-L.C., M.K., R.R., A.F.F., N.J., H.C., C.D., A.D.S., R.S.S., N.W.,
C.L., J.E.G., K.F.W., E.M., W.B., J.Wal., C.W., C.O., J.M.K., U.H.M., J.J.,
and S.B.L.; Visualization: W.B., J.Wal., J.N., C.D., A.D.S., W.A.C., and
M.S.; Funding acquisition: S.P.S.R., M.P.B., J.C.M., P.M., and T.T.D.;
Project administration: S.P.S.R., M.P.B., N.A., and T.T.D.; Supervision:
S.P.S.R., M.P.B., C.O., C.S., M.K., P.M., J. Walker, J.J., J.E.G., J.C.M.,
E.M., C.W., J.Wag, J.M.K., U.H.M., and T.T.D.; Writing – original draft:
S.P.S.R., M.K.G., J.N., O.R., M.P.B., and T.T.D.; Writing – review
and editing: S.P.S.R., J.N., M.S., R.S.J., O.R., Y.-L.C., S.B.L., E.M., C.S.,
E.M., U.H.M., P.M., M.P.B., and T.T.D. Competing interests: S.P.S.R.,
J.N., M.S., R.J., O.R., Y.-L.C., D.P., H.C., C.D., A.D.S., W.A.C., B.T.,
C.L., J.E.G., W.B., J.Wal., C.W., N.A., C.O., J.Wag, C.S., U.H.M., S.B.L.,
and T.T.D. are Novartis employees, and some own shares in Novartis.
CT compounds have been patented by Novartis with J.J., P.S.N.,
and S.P.S.R. listed as authors (US patent application no. 17/253,737,
2022). All other authors declare that they have no competing
interests. Data and materials availability: The cryo-EM map is
available from the Electron Microscopy Data Bank with accession
code EMD-29930. The atomic model has been deposited in the PDB
with the accession code PDB ID 8GCC. Metabolomics data has
been deposited into metabolomics database with accession number
MTBLS6845 and MTBLS7666. The whole-genome sequencing of the
CRISPR mutant is deposited into European Nucleotide Archive
with accession no. PRJEB52675. All compounds and reagents can be
obtained through a materials transfer agreement from Novartis by
contacting T. Krucker at thomas.krucker@novartis.com. License
information: Copyright © 2023 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.adh0614
Materials and Methods
Figs. S1 to S16
Tables S1 to S7
References (37–50)
Data S1 and S2
MDAR Reproducibility Checklist
View/request a protocol for this paper from Bio-protocol.
Submitted 22 February 2023; accepted 24 May 2023
10.1126/science.adh0614
8 of 8
RES EARCH

CHROMATIN BIOPHYSICS we require an approach to simultaneously

monitor the movement of DNA locus pairs
Stochastic motion and transcriptional dynamics of and transcription across a series of genomic
separations in vivo.
pairs of distal DNA loci on a compacted chromosome Here, we addressed this problem by live im-
aging the joint dynamics of two cis-regulatory
David B. Brückner1†, Hongtao Chen2,3†, Lev Barinov3,4, Benjamin Zoller5,6, Thomas Gregor3,5,6* DNA elements, an enhancer and a promoter,
while monitoring the transcriptional output
Chromosomes in the eukaryotic nucleus are highly compacted. However, for many functional resulting from their functional dynamic en-
processes, including transcription initiation, the pairwise motion of distal chromosomal elements counters in developing fly embryos. We sys-
such as enhancers and promoters is essential and necessitates dynamic fluidity. Here, we used tematically varied the genomic separation
a live-imaging assay to simultaneously measure the positions of pairs of enhancers and promoters and between these loci spanning many TADs.
their transcriptional output while systematically varying the genomic separation between these two Stochastic real-time trajectories of the 3D
DNA loci. Our analysis reveals the coexistence of a compact globular organization and fast subdiffusive motion of the two loci showed a dynamic
dynamics. These combined features cause an anomalous scaling of polymer relaxation times with search process, with physical proximity re-
genomic separation leading to long-ranged correlations. Thus, encounter times of DNA loci are much quired for successful transcription and a
less dependent on genomic distance than predicted by existing polymer models, with potential power-law scaling of transcription probability
consequences for eukaryotic gene expression. with genomic separation. Although typical 3D
distances between the locus pair follow a com-

L
iving systems are built based on informa-
tion encoded in chromosomes confined
in each cell’s nucleus. These meter-long
DNA polymers must be highly compacted
to fit into the micrometer-sized structure
(1, 2). At the same time, for cells to function,
chromosome organization must allow the in-
formation content to be accessed and read out
through transcription (3, 4). Often, transcrip-
tion can only occur through the spatial inter-
action of DNA loci such as enhancers and
promoters, which find each other dynamically
and remain in physical proximity (5–8). Al-
though the distances over which many en-
hancers function in higher eukaryotes can
be up to mega–base pairs in genomic separa-
tion (9–12), it is unknown how these elements
come into proximity, what their typical dis-
tance is in three-dimensional (3D) space, and
how they explore this space dynamically in the
process. Specifically, it remains unclear how
the real-time physical motion of such coupled
pairs of DNA loci determines transcriptional
encounters and how this depends on their
genomic separation.
Over the past decade, the advent of chromo-
some capture and imaging methods (13) has
given key insights into the 3D spatial orga-
nization of chromosomes, with the discovery
of structural features such as topologically
associating domains (TADs) (14–17), phase-
separated nuclear condensates (18–20), and
larger-scale compartments (21, 22). These or-
1
Institute of Science and Technology, Am Campus 1,
Klosterneuburg, Austria. 2School of Life Science and
Technology, ShanghaiTech University, Shanghai, China.
3
Lewis-Sigler Institute for Integrative Genomics, Princeton
University, Princeton, NJ, USA. 4Memorial Sloan Kettering
Cancer Center, New York, NY, USA. 5Joseph Henry
Laboratories of Physics, Princeton University, Princeton, NJ,
USA. 6Department of Developmental and Stem Cell Biology,
CNRS UMR3738 Paris Cité, Institut Pasteur, Paris, France.
*Corresponding author. Email: tg2@princeton.edu
†These authors contributed equally to this work.
ganizing structures have key implications for
transcriptional regulation (23), but they are
not static. Rather, they have been revealed to
be heterogeneous across cells (24, 25) and dy-
namic and short lived in time (26, 27). The role
of the real-time dynamics of pairs of loci is
only beginning to be understood and remains
elusive for focal contacts that are key to es-
tablishing enhancer–promoter interactions in
many systems (28).
Similarly, from a polymer physics perspec-
tive, there is a gap in our understanding of the
static and dynamic properties of chromosomes.
At large scales, across tens to hundreds of
TADs, chromosome organization has been
shown to be highly compacted in a space-filling
configuration (22, 29, 30). A useful null model
for this configuration is the crumpled chain
(also referred to as fractal globule) with fractal
dimension three (22, 31–33). However, the
real-time dynamics of DNA loci revealed by
live-imaging experiments exhibit subdiffu-
sion with exponents in the range of 0.5 to 0.6
(26, 27, 34–36), close to the predictions of the
simple Rouse polymer model, which predicts a
loosely packed ideal chain polymer configura-
tion with fractal dimension two that is in
contrast to the compacted architecture of the
crumpled chain model. A promising technique
to address this gap are scaling approaches that
combine fractal organization and subdiffusive
dynamics (37–39), but these have never been
tested experimentally.
Thus far, experimental datasets have given
insight into static organization (14–17, 22, 30, 40),
dynamic properties of chromosomes (26, 27,
34, 35, 41), or transcription (8, 36, 42–44), but
rarely all at the same time. For instance, pre-
vious live measurements of locus pairs occurred
at fixed genomic separation in transcription-
ally silent loci (26, 27). To investigate how 3D
spatial organization and dynamic locus mo-
tion control the encounter times of functional
DNA loci and thus transcriptional activation,
pact packing consistent with the crumpled
chain model, the dynamic properties exhibit
fast diffusion, albeit with a diffusion coeffi-
cient that increases with genomic separation.
These features give rise to an anomalous
scaling of polymer relaxation times and long-
range correlations in the relative motion of the
two loci. This suggests that the enhancer-
promoter search process is much less depen-
dent on genomic separation than expected
based on existing polymer models.
Results
Live imaging of chromosome dynamics
and transcription
To simultaneously monitor the coupled mo-
tion of enhancer-promoter pairs and trans-
cription across multiple genomic separations,
we generated fly lines in which a reporter gene
was introduced at various genomic locations
from the well-studied Drosophila even-skipped
(eve) locus (8). The locations of both the endo-
genous eve enhancers and the promoter of the
reporter gene, as well as the transcriptional
activity of the reporter gene, were measured
together using a three-color imaging system
(see the materials and methods, section 1.2,
and Fig. 1A) (8). To facilitate transcription, the
reporter cassette contained the insulator ele-
ment homie, which allowed stable loop forma-
tion with the endogenous homie element in
the eve locus (Fig. 1B).
We built seven such reporter constructs,
with genomic separations s varying over close
to two orders of magnitude from 58 kb to
3.3 Mb, comparable to the distances over
which many enhancers function in higher
eukaryotes (see the materials and methods,
section 1.1) (9–12). These genomic length
scales span across multiple TADs in the Dro-
sophila genome, with typical median TAD
sizes of ~90 kb (45) (here, 18 kb for the eve
locus).
Imaging took place for ~30 min during
the second half of nuclear cycle 14 (NC14) of

Brückner et al., Science 380, 1357–1362 (2023) 30 June 2023 1 of 6

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

Fig. 1. Simultaneous tracking of DNA loci and transcriptional activity in
living embryos. (A) Typical surface view of a representative fly embryo
displaying fluorescent foci for MS2, parS, and PP7 in the corresponding blue
(top), green (center), and red (bottom) channels. Top inset shows schematic
with image location in the embryo; bottom inset shows a close-up. (B) Top:
schematic of the gene cassettes used for three-color imaging. The endogenous
eve locus (left) is tagged with MS2 stem loops that are labeled with blue
fluorescence. A reporter with an eve promoter driving PP7 transcription (labeled
with red fluorescence) is integrated at a genomic separation s from the eve
locus on the second chromosome in the Drosophila genome. It includes a homie
insulator sequence allowing loop formation through homie-homie pairing and
a parS sequence that is permanently labeled with green fluorescence. Seven such
constructs were generated with varying genomic separation s (triangles).
Bottom: sample interlocus distance trajectories R(t) for six genomic separations,
with standardized y-axis limits (0, 2 mm) and x-axis limits (0, 30 min) obtained
after nucleus and locus segmentation, tracking, chromatic aberration, and
motion correction (see the materials and methods, section 1). The sampling time
interval is 28 s. (C) Trajectories of interlocus distance R and transcriptional
activity, with inferred topological states shown by the colored top bar (blue,
Ooff; cyan, Poff; red, Pon; see the materials and methods, section 2). Inset:
schematic of the three topological states. (D) 200 examples of state trajectories
sampled from a total set of N = 579 trajectories acquired in n = 30 embryos
(genomic separation s = 149 kb). Colors are as in (C). Gray trajectory parts
correspond to untrackable time points.

embryo development (Fig. 1C), well after the
completion of DNA replication. Sister chro-
matids are tightly coupled together at inter-
vals <10 kb (46). Therefore, our two tagged
DNA loci are connected by a single chromatin
polymer composed of two coupled chromatids
that were not resolved by our microscopy.
Accordingly, our measurements are associ-
ated with increased localization uncertainty
and reflect both intra- and interchromosomal
interactions, which may not be fully represen-
tative of pure intrachromosomal interactions.
Interlocus distance scaling suggests a
space-filling organization
In previous work using a single fixed genomic
distance (s = 149 kb) (8), this system was
shown to exhibit three topological states (Fig.
1C): an open configuration, Ooff, in which the
homie elements are not bound to each other,
and two paired configurations, Poff and Pon,
in which a loop is formed with either inac-
tive or active transcription, respectively. As-
suming that these configurations apply to all
genomic distances, we determined the in-
stantaneous topological and transcriptional
states of the system. To this end, we used an
inference approach with a hidden Markov
model that is based on the time series of inter-
locus distances and transcriptional activity
(see the materials and methods, section 2).
We assigned one of these states to each mea-
sured configuration, including the hidden Poff
state (Fig. 1D).
A key question is how the interlocus dis-
tances R in the open configuration Ooff vary
with the linear genomic separation s. These
distances exhibit broad distributions, which
shift systematically with larger separation
(Fig. 2A). From a polymer physics perspec-
tive, the mean distance hRi is expected to
scale as s1/d, where d is the fractal dimension.
Whereas an ideal chain polymer, as predicted
by the simple Rouse model, has fractal dimen-
sion d = 2, the compact crumpled chain or-
ganization has dimension d = 3 (33, 47). Our
experiments show a scaling exponent of 1/d =
0.31 ± 0.07 for genomic separations up to s =
190 kb, consistent with the crumpled chain
model (Fig. 2B). The smaller-than-expected
average distances observed for the largest sep-
arations (s = 595 kb, 3.3 Mb) are most likely
affected by the average folding of the chromo-
some (48).
The distances of the paired configurations
were independent of genomic separation, as
anticipated, and exhibited typical distances of
350 to 400 nm (Fig. 2B), consistent with pre-
vious measurements of distances within the
eve locus (8, 49). Together, these results re-
veal a compact crumpled chain architecture
of chromosome configurations in a range of
genomic separations consistent with Hi-C ex-
periments in Drosophila (17).
Transcriptional activity scales with
genomic separation
From the latent state trajectories revealed by
our inference approach, we estimated the sur-
vival curves of the transcriptionally active
state (see the materials and methods, section
2.4, and Fig. 2C). We found a median tran-
scriptional lifetime independent of genomic
separation of 10 ± 5 min (SD across separa-
tions; Fig. 2D). This corresponds to about three
to five independent rounds of transcription on
average, given the typical promoter switch-
ing correlation time of the system (50). Sim-
ilarly, the relative proportion of transcriptionally
active states within the paired subpopulation
is insensitive to genomic separation (Fig. 2E).

Brückner et al., Science 380, 1357–1362 (2023) 30 June 2023 2 of 6

RES EARCH | R E S E A R C H A R T I C L E 1 1

Fig. 2. Scaling of interlocus distances and transcriptional activity across
genomic separations. (A) Probability distributions of the interlocus distances
R. Distributions are separated by state, with paired states pooled across genomic
separations, and individual distributions are shown for the open state.
(B) Average interlocus distances hRi for each of the three transcriptional states.
Blue dashed line indicates a linear best fit to the Ooff data for the range of
genomic separations 58 to 190 kb, with exponent 1/d = 0.31 ± 0.07. Dashed cyan
and red lines are average values of the interlocus distances of the Poff and
Pon states, respectively, with shaded areas indicating SEM. Solid dark green and
red lines indicate predictions for ideal and crumpled polymers, respectively.
(C) Survival curves S(t) of the transcriptionally active state Pon, giving the
probability that transcription remains active after time t. Orange curve: data
for no-homie constructs (s = 58 kb). Curves were estimated using the
Kaplan-Meier estimator, which accounts for censoring that occurs if the
trajectory begins or ends in the transcriptionally active state (81). Shaded
areas show 95% confidence intervals (see the materials and methods,
section 2.4). (D) Median lifetime of the transcriptionally active state Pon as a
function of genomic separation using the Kaplan-Meier estimator (dots) and a
maximum-likelihood estimator assuming exponential decay of the survival
curves (triangles) (see the materials and methods, section 2.4). (E) Probability
of the paired on and off states conditioned on the system being in one of
these two paired configurations. (F) Overall probability of the paired
configurations Poff and Pon as a function of genomic separation. Gray line is
the best fit with exponent 0.9 ± 0.2. Green and dark red lines indicate predicted
exponents for the contact probabilities of the ideal and crumpled chain
polymer models, respectively.

By contrast, the overall probability of observ-
ing either of the paired configurations decreases
with genomic separation and exhibits a power-
law scaling P(s) ~ s–f, with f = 0.9 ± 0.2 (Fig.
2F). Because transcriptional lifetimes are in-
dependent of distance, the scaling of P(s) is
likely dominated by the search of the two loci
to come into contact. Different polymer mod-
els make distinct predictions of the scaling
of contact probabilities (22, 33, 51). For ideal
chains, f = 3/2, whereas crumpled chains ex-
hibit f ≈ 1.15 (52), which is close to the scaling
that we observed.
To determine how these results depend on
the nature of the homie insulator–mediated
focal contacts in our system, we used a re-
porter construct in which the homie sequence
was replaced by a l DNA sequence of the same
length. At a 58-kb separation, transcriptional
encounters still occur, albeit with a shorter
median lifetime of 4.9 ± 1.2 min (Fig. 2C and
fig. S13). Furthermore, the probability of observ-
ing a transcriptional state was reduced from
(30 ± 5)% for the homie version to (8.5 ± 0.8)%
in the no-homie version. By contrast, very few
such encounters were found for a 149-kb no-
homie separation (8), where contact probabil-
ity decreases from (6 ± 1)% to >1% when the
homie sequence was replaced by a l DNA.
Together, these results demonstrate quanti-
tatively how both genomic sequence and geno-
mic separation control the rate of transcriptional
encounters. The scaling of transcription prob-
abilities with separation suggests that the
transition from the open to the paired config-
uration is a key limiting step in transcriptional
activation of distal DNA loci, which is limited
by the time taken to diffuse into proximity.
Characterizing the subdiffusive locus
search process
To understand these diffusive timescales, we
considered the real-time dynamics of the blue-
and green-labeled DNA loci. We found that
the majority of single-cell trajectories sampled
the whole range of physical distances in each
topological state, because they showed a simi-
lar spread as the ensemble-averaged distri-
bution (Fig. 3, A to C, and fig. S8). Thus, rather
than existing in constrained configurations
as observed in other genomic contexts (41),
this observation supports the picture of a dy-
namic search process exploring a broad range
of distances.
We quantified how this search process is
reflected in the motion of individual DNA loci
by computing the single-locus MSD M1 ðt Þ ¼
hðri ðt0 þ t Þ  ri ðt0 ÞÞ2 it0 ¼ Gt b , where ri(t) is
the 3D position of the locus, G is diffusivity,
and b is the dynamic exponent. This expo-
nent quantifies how locus diffusion scales with
time and can be related theoretically to the
packing of the chromosome through the frac-
tal dimension d: b = 2/(2 + d) (37, 39, 53).
Although the ideal chain model predicts b =
1/2 (54), we expected b = 2/5 for a crumpled
polymer (37). Our system showed a scaling
exponent of b = 0.52 ± 0.04 across genomic
separations (error bar: SD calculated from

Brückner et al., Science 380, 1357–1362 (2023) 30 June 2023 3 of 6

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

total variance across separations) for both the

endogenous eve locus (blue) and the ectopic
reporter (green), which is close to the predic-
tion of the ideal chain model and consistent
with previous works (26, 27, 35) (see the mate-
rials and methods, section 3, and Fig. 3, D and
E). Our data further indicate that the single-
locus dynamics are not affected by transcrip-
tional activity, unlike previous accounts (43),
because they were consistent across the three
topological states (Fig. 3F).
To further understand how the locus dynamics
are determined by the interplay of chromosome
organization and single-locus dynamics, we an-
alyzed the joint dynamics of the two coupled
chromosomal loci. From the statistics of the 3D
distance vector R(t), we computed the two-
locus MSD M2 ðt Þ ¼ hðR ðt0 þ t Þ  R ðt0 ÞÞ2 it0
(26), which quantifies the crossover between
two intuitive regimes. Whereas at small time
lags, the MSD is determined by the indepen-
dent diffusion of the two loci [M2(t) = 2Gtb],
it exhibits a crossover to a plateau at large
Fig. 3. Dynamics of DNA locus search and single-locus fluctuations. (A) Single-cell interlocus
times, given by the average squared interlo-
distance trajectories for the three topological states (s = 149 kb). For each state, 80 trajectories are
cus distance [M2(t) = 2hR2i] (see the ma-
shown, with one sample trajectory highlighted in bold. (B) Distance distributions (bar histogram) of
terials and methods, section 5, and Fig. 4, A
the highlighted trajectory in (C) compared with the ensemble distribution obtained by averaging over all
and B). Consistent with the observed single-
cells (line). (C) Single-cell interlocus distance distributions (thin lines) of all trajectories for the three
locus dynamics, we found that the subdiffu-
states compared with ensemble distributions in bold (s = 149 kb). Distributions are smoothed using
sive regime of the experimental two-locus
Gaussian kernel density estimation with a width of 100 nm. Only trajectories with at least 10 time points
MSD exhibited an exponent close to 1/2 for
are included to ensure sufficient statistics for comparison. (D) Single-locus MSDs for all genomic
those datasets in which this regime was sam-
separations (color code corresponds to Fig. 2A). Single-locus MSDs were calculated by estimating 3D
pled (Fig. 4A and fig. S16). Similarly, for large
MSDs from motion-corrected trajectories in the x-y plane of the system (see the materials and methods,
time lags, the two-locus autocorrelation re-
section 3). Open data points correspond to a shorter imaging time interval Dt = 5.4 s (s = 149 kb).
vealed agreement with the ideal chain scal-
(E) Single-locus MSDs comparing enhancer (blue) and promoter (green) fluctuations (s = 149 kb).
ing (Fig. 4, C and D). Thus, the full time
(F) Single-locus MSDs comparing fluctuations in the three states (s = 149 kb).
dependence of the MSD is well described by
the ideal chain predictions, both for single
and coupled loci.

Interlocus relaxation times exhibit an anomalous
scaling with genomic separation
Having established the static and dynamic
properties of the system, we next investigated
the consequences of these features for the time-
scales of the two-locus search process. This
process is determined by the interplay of chro-
mosome dynamics and organization and can
be characterized by a relaxation time t, which
corresponds to the timescale of the crossover
of the two regimes of the two-locus MSD (Fig.
2A). Specifically, t is the time taken by the two
loci to diffuse (dynamics) over their typical dis-
tance of separation (organization): Gtb ~ s2/d.
This relationship predicts a scaling of relax-
ation times with genomic separation t ~ sg.
For ideal chains with fractal dimension d = 2
and a diffusion exponent b = 1/2, this yields
the classical result g = 2. By contrast, for crum-
pled chains, b = 2/5 and d = 3, yielding g = 5/3
(see the materials and methods, section 5,
and table S7).
To infer the relaxation time in our data as a
function of genomic separation, we performed
a Bayesian fitting of the two-locus MSD with
the ideal chain expression (26) (see the mate-
rials and methods, section 4.1). We found that
the fitted two-locus diffusion coefficient in-
creased with genomic separation up to 595 kb,
with an approximate scaling G(s) ~ s0.27 ± 0.03
(Fig. 4E). This scaling appeared to plateau for
the largest genomic separation (3.3 Mb) at a
value close to the single-locus diffusion, which
remained approximately constant across separa-
tions (Fig. 4E). The absolute value of the diffu-
sivity at the plateau was almost 20-fold larger
than previous measurements in mammalian
stem cells with similar genomic separation (26),
suggesting comparatively fast chromosome dy-
namics (fig. S23).
The relaxation time was determined by com-
bining our estimate of the two-locus diffusivity
with the average interlocus distances. The com-
bination of static and dynamic exponents in
our system, as well as the scale-dependent dif-
fusivity, results in an anomalous scaling of
relaxation times with genomic separation with
an exponent g = 0.7 ± 0.2 (Fig. 4F). This ex-
ponent corresponds to a much shallower scal-
ing with separation than predicted by either
the ideal or crumpled chain theory. This result
was further confirmed by a data collapse of the
two-locus autocorrelation functions (Fig. 4D
and fig. S20). Although these results are de-
rived from the trajectories in the Ooff state,
they are insensitive to the details of the state
inference (fig. S11). In sum, the key result here
is that the relaxation time, which sets the time-
scale of two-locus encounters, is much less de-
pendent on genomic separation than predicted
by existing polymer models.
Anomalous relaxation time scaling induces
long-ranged velocity correlations
The anomalous relaxation time scaling makes
a key prediction for the correlations of the ab-
solute motion of DNA loci, quantified by the
ðdÞ ðdÞ
velocity cross-correlation Cvv ðt Þ ¼ hv i ðt0 Þ
ðdÞ
v j ðt0 þ t Þit0 . These correlations are deter-
mined by the relaxation time through the
dimensionless ratio d/t, where d is the exper-
imental observation timescale (Fig. 4G) (55).
Having determined the relaxation times t,
one can therefore make a parameter-free
prediction of the correlations, which decay

Brückner et al., Science 380, 1357–1362 (2023) 30 June 2023 4 of 6

RES EARCH | R E S E A R C H A R T I C L E 1 1

Fig. 4. Joint dynamics of DNA locus pairs. (A) Ideal

chain Rouse prediction of the two-locus MSD M2(t) =
2Gt1/2(1 – e–t/pt) + 2J erfc[(t/pt)1/2] (26) (gray line) using
best-fit values G, J, b = 1/2, and t = (J/G)2, compared
with experiment (s = 595 kb). Green and red lines
give expected scaling tb for t ≪ t for the generalized
Rouse model for ideal and crumpled chains, respectively
(see the materials and methods, section 5). (B) All
experimental two-locus MSDs with relaxation times
(dashed lines) and expected asymptotes 2hR2i
(solid lines; color code corresponds to Fig. 2A).
(C) Scaling of the diffusion coefficients G from two-locus
MSD fits (black dots) compared with single-locus
diffusion coefficients obtained from single-locus MSDs
(Fig. 3, F to H). Dashed line is the best fit to two-locus
diffusivity with exponent 0.27 ± 0.03 (s = 58 to
595 kb); solid lines are the average value of single-
locus diffusivities; shaded area shows SE calculated from
total variance across separations. (D) Two-locus auto-
correlation function (ACF) C2 ðtÞ ¼ hRðt0 Þ Rðt0 þ tÞit0 ¼
hR2 i  M2 ðtÞ=2 (gray) compared with data (s = 149 kb).
Green and red curves indicate the power-law exponent
l = 2(1 – d)/(2 + d) of the correlation function C2(t) ~ tl
for ideal and crumpled chains for t ≫ t, respectively
(39). (E) Collapsed correlations C2 ~ C2(ts–g)/hR2i with
g = 0.7. Inset: raw correlations C2(t) for varying
genomic separation. Open data points correspond to
data obtained with a higher sampling rate. (F) Scaling
of inferred relaxation times compared with predicted
ideal and crumpled chain exponents. Gray line is
the best fit with exponent g = 0.7 ± 0.2. (G) Predicted
ðdÞ
velocity cross-correlation functions Cvv ðtÞ ¼
ðdÞ ðdÞ
hvi ðt0 Þ  vj ðt0 þ tÞit0 for increasing values of the
dimensionless ratio d/t (55). Velocities are calculated
on a time interval d as v(d)(t) = [x(t + d) – x(t)]/d.
(H) Scaling of the zero-time velocity cross-correlation
intercept normalized by the zero-time autocorrelation,
ðdÞ ðdÞ
Cvv ð0Þ=Cv ð0Þ, for the Ooff (blue) and Pon (red) states;
d = 300 s. Green line is the prediction based on ideal
chain Rouse scaling of the relaxation times (g = 2) with an
intercept determined based on the 58-kb data point; gray
line is the parameter-free prediction using the inferred anomalous relaxation time scaling (g ≈ 0.7) (see the materials and methods, section 4.3); dashed red line is the
average correlation in the Pon state.

substantially more slowly than for the ideal
Rouse model (see the materials and methods,
section 4.3, and Fig. 4H, green and gray lines).
We found that the experimental correlations
were quantitatively captured by this parameter-
free prediction (Fig. 4H), including the full
time dependence of the correlations (fig. S22).
This demonstrates that the anomalous relaxa-
tion time scaling indeed leads to long-range
velocity cross-correlations of chromosomal
loci, pointing toward potential long-range
interactions.
Discussion
We developed an experimental approach to
perform in vivo imaging of the joint dynam-
ics of enhancer-promoter pairs with varying
genomic separation and simultaneous moni-
toring of their transcriptional output. Observ-
ing the dynamics of pairs of DNA loci has only
become possible recently and has been done
for tagged DNA loci at a single fixed genomic
separation (8, 26, 27, 36). Here, we show how
imaging across genomic separations gives
insight into the relative motion, dynamic
encounters, and transcriptional activation
of such loci.
Many features of the two-locus dynamics,
including the subdiffusive exponent close to
0.5, are very well conserved with measure-
ments of CTCF sites at TAD boundaries in
mammalian systems (26, 27), despite CTCF
not being essential for Drosophila embryo-
genesis (56). In absolute numbers, however,
our measurements revealed large diffusion
coefficients of DNA loci that are ~20-fold
larger than in mammalian cells (26) (fig. S23).
Early fly development follows a tight sched-
ule, suggesting that the chromosome dynam-
ics may have evolved to operate on much
faster timescales than mammalian systems.
By contrast, the median lifetime of focal con-
tacts in our system of 12 ± 5 min is well within
the range of typical CTCF loop lifetimes of
10 to 30 min in mammalian cells (26, 27). These
timescales facilitate transcriptional lifetimes
of 10 ± 5 min in our system, which in the ab-
sence of the homie insulator are reduced to
4.9 ± 1.2 min (Fig. 2C and fig. S13), highlighting
the importance of focal elements for contact
formation in Drosophila.

Brückner et al., Science 380, 1357–1362 (2023) 30 June 2023 5 of 6

RES EARCH | R E S E A R C H A R T I C L E
To initiate such transcriptional encoun-
ters, the two loci perform a search process to
reach physical proximity. The timescale of this
search process is given by the lifetimes of the
unpaired Ooff state, which depends on multi-
ple factors. These factors typically include the
landscape of the search process (57, 58), the
biochemical binding properties of the focal
elements when proximity has been reached
(59), and the correlation time of the system
(i.e., the relaxation time). Indeed, we found
that lifetimes increased with the relaxation
times and were ~10 times larger on average
(fig. S19).
We have demonstrated how key features of
our system, tight crumpled chain packing, sub-
diffusion with exponent 0.5, and a separation-
dependent two-locus diffusivity, lead to relaxation
times that are much less dependent on genomic
separation than predicted by existing polymer
models. Indeed, for an ideal Rouse polymer,
the relaxation time for our largest genomic
separation (3.3 Mb) would be ~3000 times
longer than for the shortest 58-kb separation.
Our measurements, however, revealed that it
only takes ~20 times longer, corresponding
to a >100-fold reduction. This reduced depen-
dence on distance implies that transcriptional
encounters are possible across large genomic
distances, allowing enhancers dispersed across
the chromosome to find their target promoter
efficiently. This might be one of the reasons
that evolution can act on distal sequences
from a given target promoter. Overall, our find-
ings have crucial implications for the spatio-
temporal organization of the cell nucleus,
including the dynamics of long-range focal con-
tacts (28) and mammalian enhancer-promoter
interactions (9–12, 44).
From a polymer physics perspective, our
measured exponents suggest that the rela-
tionship between static and dynamic prop-
erties in the generalized Rouse framework,
which relies on the assumption of local fric-
tion, does not apply to chromosomes. This
implies that long-range interactions such as
hydrodynamics or active motor-mediated
interactions (60, 61) could play a role. Indeed,
the simplest polymer model that relaxes the
Rouse assumption and includes long-range
hydrodynamic interactions, the Zimm mod-
el (54), predicts a scaling relationship of re-
laxation times with genomic separations with
an exponent of g = 1 (table S7), which is close
to our measured value of g ≈ 0.7. Furthermore,
the observed separation-dependent diffusivity
points to additional interactions or heteroge-
neities along the polymer. Such heterogeneities
could be caused by a number of processes,
such as cross-linking (41), out-of-equilibrium
activity (61), entanglements (62), or the pres-
ence of condensates (18–20). Together, these
processes may orchestrate the anomalous
scaling of relaxation times with genomic sep-
Brückner et al., Science 380, 1357–1362 (2023)
aration. In future work, the mechanistic un-
derpinnings of our findings should be tested
using polymer simulations (40, 41, 51, 63–69) to
generate hypotheses for new sets of experiments.
RE FERENCES AND NOTES
1. W. Flemming. Zellsubstanz, Kern Und Zelltheilung (Vogel,
1882).
2. R. Milo, R. Phillips, Cell Biology by the Numbers (CRC Press,
2015).
3. R. Stadhouders, G. J. Filion, T. Graf, Nature 569, 345–354
(2019).
4. B. van Steensel, E. E. M. Furlong, Nat. Rev. Mol. Cell Biol. 20,
327–337 (2019).
5. M. Levine, Curr. Biol. 20, R754–R763 (2010).
6. T. K. Kim, R. Shiekhattar, Cell 162, 948–959 (2015).
7. D. Vernimmen, W. A. Bickmore, Trends Genet. 31, 696–708
(2015).
8. H. Chen et al., Nat. Genet. 50, 1296–1303 (2018).
9. B. Tolhuis, R. J. Palstra, E. Splinter, F. Grosveld, W. de Laat,
Mol. Cell 10, 1453–1465 (2002).
10. V. V. Uslu et al., Nat. Genet. 46, 753–758 (2014).
11. Y. Zhang et al., Nature 504, 306–310 (2013).
12. A. Sanyal, B. R. Lajoie, G. Jain, J. Dekker, Nature 489, 109–113
(2012).
13. Y. Huang, R. Neijts, W. de Laat, FEBS Lett. 594, 3439–3449
(2020).
14. J. R. Dixon et al., Nature 485, 376–380 (2012).
15. C. Hou, L. Li, Z. S. Qin, V. G. Corces, Mol. Cell 48, 471–484
(2012).
16. E. P. Nora et al., Nature 485, 381–385 (2012).
17. T. Sexton et al., Cell 148, 458–472 (2012).
18. A. Boija et al., Cell 175, 1842–1855.e16 (2018).
19. W.-K. Cho et al., Science 361, 412–415 (2018).
20. M. Conte et al., Nat. Commun. 13, 4070 (2022).
21. C. S. Osborne et al., Nat. Genet. 36, 1065–1071
(2004).
22. E. Lieberman-Aiden et al., Science 326, 289–293
(2009).
23. O. Symmons et al., Genome Res. 24, 390–400 (2014).
24. T. J. Stevens et al., Nature 544, 59–64 (2017).
25. M. V. Arrastia et al., Nat. Biotechnol. 40, 64–73
(2022).
26. M. Gabriele et al., Science 376, 496–501 (2022).
27. P. Mach et al., Nat. Genet. 54, 1907–1918 (2022).
28. M. Levo et al., Nature 605, 754–760 (2022).
29. S. Wang et al., Science 353, 598–602 (2016).
30. B. Bintu et al., Science 362, eaau1783 (2018).
31. A. Grosberg, Y. Rabin, S. Havlin, A. Neer, Europhys. Lett. 23,
373–378 (1993).
32. A. Yu. Grosberg, S. K. Nechaev, E. I. Shakhnovich, J. Phys.
(Paris) 49, 2095–2100 (1988).
33. L. A. Mirny, Chromosome Res. 19, 37–51 (2011).
34. S. C. Weber, A. J. Spakowitz, J. A. Theriot, Phys. Rev. Lett. 104,
238102 (2010).
35. J. S. Lucas, Y. Zhang, O. K. Dudko, C. Murre, Cell 158, 339–352
(2014).
36. J. M. Alexander et al., eLife 8, e41769 (2019).
37. M. V. Tamm, L. I. Nazarov, A. A. Gavrilov, A. V. Chertovich,
Phys. Rev. Lett. 114, 178102 (2015).
38. S. Shinkai, T. Nozaki, K. Maeshima, Y. Togashi, PLOS Comput.
Biol. 12, e1005136 (2016).
39. K. E. Polovnikov, M. Gherardi, M. Cosentino-Lagomarsino,
M. V. Tamm, Phys. Rev. Lett. 120, 088101 (2018).
40. M. Nicodemi, A. Pombo, Curr. Opin. Cell Biol. 28, 90–95
(2014).
41. N. Khanna, Y. Zhang, J. S. Lucas, O. K. Dudko, C. Murre,
Nat. Commun. 10, 2771 (2019).
42. T. Germier et al., Biophys. J. 113, 1383–1394 (2017).
43. B. Gu et al., Science 359, 1050–1055 (2018).
44. J. Zuin et al., Nature 604, 571–577 (2022).
45. S. V. Ulianov et al., Nat. Commun. 12, 41 (2021).
46. B. L. Hua, T. L. Orr-Weaver, Genetics 207, 29–47
(2017).
47. B. Avşaroğlu et al., PLOS ONE 9, e102474 (2014).
48. J. Dekker, M. A. Marti-Renom, L. A. Mirny, Nat. Rev. Genet. 14,
390–403 (2013).
49. L. Barinov, S. Ryabichko, W. Bialek, T. Gregor, Transcription-
dependent spatial organization of a gene locus. arXiv:2012.
15819 [q-bio.MN] (2020).
30 June 2023
50. B. Zoller, S. C. Little, T. Gregor, Cell 175, 835–847.e25
(2018).
51. A. Rosa, N. B. Becker, R. Everaers, Biophys. J. 98, 2410–2419
(2010).
52. J. D. Halverson, W. B. Lee, G. S. Grest, A. Y. Grosberg,
K. Kremer, J. Chem. Phys. 134, 204904 (2011).
53. P. G. De Gennes, Macromolecules 9, 587–593 (1976).
54. M. Rubinstein, R. H. Colby, Polymer Physics (Oxford Univ.
Press, 2003).
55. T. J. Lampo, A. S. Kennard, A. J. Spakowitz, Biophys. J. 110,
338–347 (2016).
56. A. Kaushal et al., Nat. Commun. 12, 1011 (2021).
57. G. Wilemski, M. Fixman, J. Chem. Phys. 60, 866–877 (1974).
58. A. Szabo, K. Schulten, Z. Schulten, J. Chem. Phys. 72,
4350–4357 (1980).
59. M. Fujioka, H. Mistry, P. Schedl, J. B. Jaynes, PLOS Genet. 12,
e1005889 (2016).
60. R. Bruinsma, A. Y. Grosberg, Y. Rabin, A. Zidovska, Biophys. J.
106, 1871–1881 (2014).
61. M. M. Tortora, H. Salari, D. Jost, Curr. Opin. Genet. Dev. 61,
37–43 (2020).
62. R. Kawamura et al., J. Cell Biol. 188, 653–663 (2010).
63. A. Rosa, R. Everaers, PLOS Comput. Biol. 4, e1000153
(2008).
64. M. Barbieri et al., Proc. Natl. Acad. Sci. U.S.A. 109, 16173–16178
(2012).
65. L. Giorgetti et al., Cell 157, 950–963 (2014).
66. L. Liu, G. Shi, D. Thirumalai, C. Hyeon, PLOS Comput. Biol. 14,
e1006617 (2018).
67. A. Buckle, C. A. Brackley, S. Boyle, D. Marenduzzo, N. Gilbert,
Mol. Cell 72, 786–797.e11 (2018).
68. C. A. Brackey, D. Marenduzzo, N. Gilbert, Nat. Methods 17,
767–775 (2020).
69. H. Salari, M. Di Stefano, D. Jost, Genome Res. 32, 28–43
(2022).
70. D. B. Brückner, B. Zoller, dbrueckner/DynamicChromosome:
initial release, Zenodo (2023); https://zenodo.org/record/
7926778.
71. D. B. Brückner, H. Chen, L. Barinov, B. Zoller, T. Gregor,
Trajectory data for: Stochastic motion and transcriptional
dynamics of pairs of distal DNA loci on a compacted
chromosome, Zenodo (2023); https://zenodo.org/record/
7616203.
AC KNOWLED GME NTS
We thank R. Evearers, L. Giorgetti, A. Grosberg, S. Grosse-Holz,
M. Levo, L. Mirny, A. Rosa, V. Scolari, A. Spakowitz, and G. Tkacik
for helpful comments and discussion; K. Bystricky for introducing
us to the ParB/parS system; and F. Payre and P. Valenti for
sharing the ParB-eGFP plasmid and the parS sequence. Funding:
This work was supported in part by the U.S. National Science
Foundation, the Center for the Physics of Biological Function (grant
PHY-1734030), and the National Institutes of Health (grants
R01GM097275, U01DA047730, and U01DK127429). D.B.B. was
supported by the NOMIS Foundation as a fellow and by an EMBO
postdoctoral fellowship (ALTF 343-2022). H.C. was supported
by a Charles H. Revson Biomedical Science Fellowship. Author
contributions: H.C. and T.G. designed the experiments. H.C. and
L.B. performed the experiments and analyzed the images. D.B.B.
and B.Z. analyzed the data. D.B.B. and T.G. interpreted the
data. D.B.B., B.Z., and T.G. wrote the manuscript. Competing
interests: The authors declare no competing interests. Data
and materials availability: The code and software used in this
study (70) and the raw trajectory data (71) are freely available
through Zenodo. License information: Copyright © 2023 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.adf5568
Materials and Methods
Figs. S1 to S23
Tables S1 to S8
References (72–86)
Movies S1 to S7
MDAR Reproducibility Checklist
View/request a protocol for this paper from Bio-protocol.
Submitted 29 October 2022; accepted 31 May 2023
10.1126/science.adf5568
6 of 6
RES EARCH

MARTIAN GEOLOGY pole-facing slopes (11–15), locations that now

contain gullies (14, 15). Meanwhile, large slabs
Gullies on Mars could have formed by melting of of CO2 ice (currently sequestered in the south-
ern polar ice cap) sublimated (16), producing
water ice during periods of high obliquity an atmosphere with double its current pres-
sure (17). The extent of terrain that exper-
J. L. Dickson*1,2, A. M. Palumbo2, J. W. Head2, L. Kerber3, C. I. Fassett4†, M. A. Kreslavsky5 ienced conditions capable of hosting liquid
H2O would have been greater at that time
Gullies on Mars resemble water-carved channels on Earth, but they are mostly at elevations where than in the present.
liquid water is not expected under current climate conditions. It has been suggested that sublimation
General circulation models
of carbon dioxide ice alone could have formed Martian gullies. We used a general circulation model to show
that the highest-elevation Martian gullies coincide with the boundary of terrain that experienced pressures We tested the hypothesis that liquid H2O was
above the triple point of water when Mars’ rotational axis tilt reached 35°. Those conditions have occurred involved in the formation of gullies by in-
repeatedly over the past several million years, most recently ~630,000 years ago. Surface water ice, if vestigating whether the mapped locations
present at these locations, could have melted when temperatures rose >273 kelvin. We propose a dual gully of gullies (3) correlate with the locations of
formation scenario that is driven by melting of water ice followed by carbon dioxide ice sublimation. terrain that could host liquid H2O when Mars
experienced 35° obliquity. We used a three-

G
dimensional general circulation model (GCM)
ullies on Mars resemble H2O-carved uid H2O. Previous studies have shown that at of Mars (13, 18) to simulate climatic condi-
channels on Earth (1). Their concentra- 35° obliquity, H2O ice accumulated on mid- tions at three obliquities and associated pres-
tion at Mars’ midlatitudes, where near- latitude (30° to 45° in each hemisphere), sures (16, 17) that occurred in the past million
surface ice is stable (1–3), is consistent
with an H2O-melting model for their
formation. However, the observed distribution
includes elevations where present-day atmo-
spheric pressure is always below the triple
point of H2O (2, 4), so solid H2O ice is expected
to sublimate to form vapor rather than melt
to form liquid. The seasonal timing of con-
temporary mobilization of surface material
within gullies is consistent with sublimation
of solid CO2 (5, 6), so it is possible that gullies
formed from CO2-mediated processes alone
(7). However, the mechanism of such a CO2-
only process is uncertain and lacks an Earth
analog. Repeat orbital imaging shows that
present-day erosion of gullies is rare. Although
examples have been documented (5, 6), ero-
sion of gully channels is minor (Fig. 1A) and
infrequent. At high latitudes (>70°), where
solid CO2 ice is emplaced and removed every
year, ~98.3% of gullies show no activity and
none experience channel erosion (8).
An alternative possibility is that gullies were
incised by small amounts of liquid H2O during
earlier climate conditions that were more con-
ducive to melting of H2O ice. This scenario
would be consistent with the stratigraphy of
gully fans (9, 10), which indicates cycles of em-
placement of fan sediment, fracturing (Fig. 1B),
and incision. Mars’ axial tilt (obliquity) is known
to vary over hundreds of thousands of years
(11), so earlier periods of higher obliquity could
have provided more-favorable climates for liq-

1
Division of Geological and Planetary Sciences, Caltech,
Pasadena, CA, USA. 2Department of Earth, Environmental
and Planetary Sciences, Brown University, Providence,
RI, USA. 3Jet Propulsion Laboratory, Caltech, Pasadena,
CA, USA. 4NASA Marshall Space Flight Center, Huntsville,
AL, USA. 5Earth and Planetary Sciences, University of
California, Santa Cruz, CA, USA.
*Corresponding author. Email: jdickson@caltech.edu
†Present address: Applied Physics Laboratory, Johns Hopkins
University, Laurel, MD, USA.
Fig. 1. Active gullies and gully stratigraphy on Mars. (A) Orbital image of Terra Sirenum (37.45°S, –137.05°E)
showing a gully channel–forming event (5). The image was taken with the High Resolution Imaging Science Experiment
(HiRISE) instrument in the red channel, 570 to 830 nm (image ESP_032011_1425). The white line (within white box)
indicates a channel that formed between 14 May 2009 and 25 May 2013. This illustrates the scale of current activity
compared with entire gully systems. (B) HiRISE image ESP_013115_1420 taken before the activity showing buried
channel (white arrows) and fan material (black arrows) (5). (C) Enlargement of the area within the white box in (A)
showing exhumed channel and fan material. (D) HiRISE image PSP_007592_1425 of another gully system at 176.52°E,
37.40°S. Fractures cross-cut the gully fan material, including gully channels incised into large gully fans (10).

Dickson et al., Science 380, 1363–1367 (2023) 30 June 2023 1 of 5

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

years (fig. S1 and table S1). These three sce-
narios represent conditions that occurred re-
peatedly through the Amazonian period (19),
the most recent ~3 billion years of Mars’ geo-
logic history. For all three scenarios, simu-
lations were performed twice, with perihelion
(closest approach to the Sun) occurring at
summer solstice in the northern hemisphere
and southern solstice in the southern hemi-
sphere, for a total of six model runs. For each
run, we mapped the locations where the at-
mospheric pressure predicted by the GCM is
>612 Pa and regional soil temperatures are
>273 K, corresponding to the triple point of
H2O (20).
At 25° obliquity, similar to present day
conditions, locations that meet those thresh-
olds (Fig. 2B and movie S2) correlate poorly
with the global distribution of gullies (3). The
atmospheric pressure does exceed 612 Pa at
low elevations (<–2500 m) in the low and
midlatitudes during the spring and summer,
but only for brief periods of the year when
regional soil temperatures exceed 273 K (<2%
of GCM time steps).
At 30° obliquity, the locations of northern
hemisphere gullies experience conditions
above the triple point of H2O, whereas most
southern hemisphere gullies remain <612 Pa
(Fig. 2, C and D). Close to perihelion, gullies
near 30°N experience potential H2O melt-
ing conditions for ~13% of the year (Fig. 2C
and movie S3). Gullies in the southern high-
lands typically do not surpass 612 Pa (Fig. 2,
C and D).
At 35° obliquity, pressures rise >612 Pa at
nearly all gully sites during seasons and times
of day when H2O ice is most likely to reach
273 K (Fig. 2, E and F, and movie S5). Gullies
across Mars are only observed on steep slopes
(2, 21, 22), so they do not form on the smooth
terrain that also exceeds the triple point of
H2O in our model (see the supplementary
text and fig. S2). Within 25° of the equator,
most terrain in the model reaches pressures
>612 Pa, but there was probably no H2O ice
to melt in these regions. Most gullies on Mars
(78.4%) occur between 25°S and 50°S (3); in
the model, these locations all exceed 612 Pa
during seasons and times of day when re-
gional temperatures are >273 K (Fig. 2F and
movie S6). This latitude band hosts the highest-
elevation gullies observed on Mars, the loca-
tions of which all surpass 612 Pa (Fig. 3). The
model predicts a sharp boundary between
lowlands, which are frequently >612 Pa during
the warmest time of the season and day, and
highlands, which rarely exceed 612 Pa. This
boundary occurs at an elevation of ~4500 m
and extends for ~1700 km. It matches the
upper elevation limit of gullies, which is 4500 m
(3) (Fig. 4). South of this boundary, gullies
are deeply incised with large bedrock alcoves

Fig. 2. Percentage of time above the

triple point of H2O in each GCM
simulation. In all panels, the
background image is a shaded
relief map from the Mars Orbiter
Laser Altimeter (MOLA) instrument
(31) in Robinson projection. Black dots
indicate the locations of gullies (3).
Colored points indicate the percentage
of time steps in the GCM simulations
that were above the triple point of
H2O. (A, C, and E) Results for simula-
tions in which perihelion occurs at
summer solstice in the northern
hemisphere. (B, D, and F) Results
for simulations in which perihelion
occurs at summer solstice in the
southern hemisphere. An orbital
schematic is shown below each
panel. Obliquities are 25° [(A)
and (B)], 30° [(C) and (D)], and
35° [(E) and (F)]. Increasing
obliquity leads to corresponding
increases in mean pressure. Each
simulation ran for 1 Mars year.
Pressure >612 Pa at all gully
locations is only reached in the
simulation at 35° obliquity, with
most gully locations reaching those
conditions during peak southern
summer. The white box in (F)
indicates the region shown
in Fig. 4.

Dickson et al., Science 380, 1363–1367 (2023) 30 June 2023 2 of 5

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

shading and surface composition. The GCM

that we used does not have sufficient resolu-
tion to fully characterize these temperature
variations. Therefore, we considered whether
the pressure at ice-rich gully sites could have
surpassed the triple point of H2O during sea-
sons and times of day when surface temper-
atures were most likely to surpass 273 K.
Temperatures output by the GCM are mean
values for a cell 3.75° in longitude by 5.625° in
latitude, assuming that the surface is dry rego-
lith on approximately flat terrain (20). The
temperature of H2O ice on steep slopes in
shaded alcoves could be very different. We
expect that the thermal inertia of H2O ice,
which is higher than that of dry regolith,
would dampen diurnal or seasonal temper-
ature rises, so they would surpass 273 K more
slowly or over more limited areas than pre-
dicted by the model.
We therefore compared our global results
with previous studies of local regions that
used higher-resolution models (25–28) and
investigated whether they could predict ice
>273 K during the same seasons and times of
day as our models. Models of specific gullies
predict H2O ice surface temperatures exceed-
ing 273 K under present conditions (25–27).
Additionally, sudden heating of H2O frost can
generate temperatures to within 10°C of melt-
ing (28), with melting conditions enhanced
during past climates with increased atmo-
spheric pressure, as in our high-obliquity mod-
Fig. 3. Elevation distributions of gullies and simulated locations that reach the triple point of H2O.
els. H2O ice surface temperatures exceeding
The black histogram shows the elevations of gullies on Mars (3), and the gray histogram shows the elevations
273 K at high obliquity are predicted by pre-
of terrain that experience potential melting conditions in the GCM simulations at 25° obliquity and
vious high-resolution models (26, 27, 29). We
6.0 mbar (A), 30° obliquity and 8.0 mbar (B), and 35° obliquity and 12.0 mbar (C). For each obliquity,
therefore regard H2O ice surface temperatures
the northern and southern summer periods are combined. Minimum conditions for melting are achieved at
that exceed 273 K as plausible during seasons
all gully elevations in the 35° obliquity simulation. The dashed gray line indicates the distribution of
and times of day when our models predict
topography in latitude bands that contain gullies; it was extracted from an equal-area gridded global MOLA
pressures that exceed 612 Pa at all gully loca-
topography map (32). The dashed vertical line indicates the maximum gully elevation. At 35° obliquity,
tions on Mars.
H2O ice at the surface in the northern lowlands is common, holding the surface temperature at 273 K
(movie S5) because of the latent heat of melting, so there are less-frequent conditions above the triple point Implications for gully formation
of H2O at low elevations than at lower obliquity.
Our results show that at 35° obliquity, the lo-
cations of gullies reached pressures >612 Pa,
and ice surface temperatures likely surpassed
273 K. These locations have abundant H2O ice
(Fig. 4, A, D, and E). In the highlands to the into an ice-cemented substrate, referred to as near the surface today (23) and are thought to
north of the contact, gullies are absent (Fig. 4) pasted-on terrain (15) (see the supplementary have had tens of meters more H2O ice within
despite the presence of steep pole-facing slopes text). Ground ice that was emplaced during the past 1 million years (10, 11).
draped with smooth, coherent regolith (figs. periods of high obliquity (11) is present on If gullies were formed by small amounts of
S3 and S4) that is rich in H2O ice (11) (Fig. 4B midlatitude, pole-facing slopes today (23), liquid H2O during periods of high obliquity,
and fig. S4A). where gullies are most abundant (2, 3, 21, 22). then this would explain the maximum eleva-
This is consistent with model predictions that tion of known gullies at ~4500 m (Figs. 3 and 4),
Spatial distribution of H2O melting locations incorporate local topography (24). We there- as well as why stratigraphically lower gully
Our model shows that at 35° obliquity, all fore infer that abundant H2O ice was present channels and fans (Fig. 1B) have not been
gullies in the southern hemisphere exceeded at nearly all gully sites in the ice-rich pasted- erased by mobilization of loose material with-
612 Pa for >15% of a Martian year (Fig. 2) on terrain. This same requirement applies to in gullies by CO2-mediated processes (6, 7)
during seasons and times of day when re- alternative explanations involving erosion by (Fig. 1A). We propose the following scenar-
gional soil temperatures surpass 273 K. Melt- CO2, in which vertical deepening of gully io for gully formation and evolution that
ing of H2O in the gullies also requires the channels is achieved through sublimation incorporates both H2O and CO2 on the same
presence of H2O ice on or near the surface of an ice-rich layer (7). slopes.
and surface temperatures >273 K. Gully chan- Surface temperature, unlike pressure, fluc- First, gullies formed during periods of high
nels are observed to be most commonly incised tuates both diurnally and due to topographic obliquity throughout the Amazonian period.

Dickson et al., Science 380, 1363–1367 (2023) 30 June 2023 3 of 5

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

Fig. 4. Uppermost vertical

extent of gullies on Mars.
(A) Mars Reconnaissance Orbiter
Context Camera (CTX) global
panchromatic mosaic (33) overlain
by MOLA topography (31) of
the Thaumasia highlands. Yellow
stars represent mapped gullies
(3). Grayscale dots show the per-
centage of time spent above the
triple point of H2O in the GCM
simulation. There is a large
difference between lowlands and
highlands. Gridlines are drawn
at GCM cell boundaries. White
letters show locations for (B) to
(E). (B) CTX panchromatic image
D10_031205_1382_XN_41S091W
of a steep, pole-facing slope
overlain with pasted-on terrain;
no gullies are visible at this
resolution (5 m per pixel).
Figure S4 shows a wider area.
(C) CTX panchromatic image
F09_039196_1414_XN_38S084W
of the highest-elevation gully
on Mars [see context in (A)]. The
gully has formed in pasted-on
terrain that fills previously
eroded bedrock alcoves.
(D) CTX panchromatic image
B12_014445_1332_XN_46S086W
of deeply incised gullies in
Aonia Terra. (E) Additional
CTX panchromatic image
P15_006916_1370_XN_43S083W
of deeply incised gullies in
Aonia Terra.

Those conditions occurred several times in the
past few million years, most recently ~630,000
years ago (19) (fig. S1). Gully formation oc-
curred at locations where conditions for liquid
H2O were reached, which includes elevations
up to ~4500 m.
Second, Mars returned to lower obliquity
(25°). Melting of H2O ice ceased across Mars
and sublimation dominated, leaving loose mate-
rial that collected in gully channels and could
be easily mobilized.
Finally, alcoves and channels now serve as
cold traps for both H2O and CO2 seasonal frost
and ice. Both CO2 (6) and H2O (30) undergo
phase changes through the course of Mars’
year as temperatures vary, producing occasional
(8) and local (6) modifications to gullies, as
observed today (Fig. 1A).
This scenario is consistent with the observed
spatial distribution of gullies (1–3), the mor-
phology of gully channels (1), gully fan strat-
igraphy (9, 10), the observed mobilization of
regolith within gullies (5, 6), and the lack of for-
mation of new gully channels on Mars today (8).

Dickson et al., Science 380, 1363–1367 (2023) 30 June 2023 4 of 5

RES EARCH | R E S E A R C H A R T I C L E
In our proposed scenario, slopes with gullies
have hosted both CO2 and H2O ices, and both
ices were involved in the formation and mod-
ification of gullies on Mars.
RE FE RENCES AND N OT ES
1. M. C. Malin, K. S. Edgett, Science 288, 2330–2335 (2000).
2. J. L. Heldmann, M. T. Mellon, Icarus 168, 285–304 (2004).
3. T. N. Harrison, G. R. Osinski, L. L. Tornabene, E. Jones,
Icarus 252, 236–254 (2015).
4. R. M. Haberle et al., J. Geophys. Res. 106 (E10), 23317–23326
(2001).
5. C. M. Dundas, S. Diniega, A. S. McEwen, Icarus 251, 244–263
(2015).
6. C. M. Dundas, A. S. McEwen, S. Diniega, C. J. Hansen, S. Byrne,
J. N. McElwaine, “The formation of gullies on Mars today,”
in Geological Society, London, Special Publications 467
(The Geological Society, 2019), pp. 67–94.
7. C. Pilorget, F. Forget, Nat. Geosci. 9, 65–69 (2016).
8. J. Raack et al., Icarus 350, 113899 (2020).
9. J. L. Dickson, J. W. Head, Icarus 204, 63–86 (2009).
10. J. L. Dickson, J. W. Head, T. A. Goudge, L. Barbieri, Icarus 252,
83–94 (2015).
11. J. W. Head, J. F. Mustard, M. A. Kreslavsky, R. E. Milliken,
D. R. Marchant, Nature 426, 797–802 (2003).
12. M. A. Mischna, M. I. Richardson, R. J. Wilson, D. J. McCleese,
J. Geophys. Res. 108 (E6), 5062 (2003).
13. J.-B. Madeleine et al., Geophys. Res. Lett. 41, 4873–4879
(2014).
14. R. E. Milliken, J. F. Mustard, D. L. Goldsby, J. Geophys. Res. 108
(E6), 5057 (2003).
15. P. R. Christensen, Nature 422, 45–48 (2003).
16. R. J. Phillips et al., Science 332, 838–841 (2011).
17. C. J. Bierson et al., Geophys. Res. Lett. 43, 4172–4179
(2016).
18. F. Forget et al., J. Geophys. Res. 104, 24155–24175 (1999).
19. J. Laskar, B. Levrard, J. F. Mustard, Nature 419, 375–377
(2002).
20. Materials and methods are available as supplementary materials.
Dickson et al., Science 380, 1363–1367 (2023) 30 June 2023
21. J. L. Dickson, J. W. Head, M. Kreslavsky, Icarus 188, 315–323
(2007).
22. S. J. Conway, T. N. Harrison, R. J. Soare, A. W. Britton,
L. J. Steele, “New slope-normalized global gully density and
orientation maps for Mars,” in Geological Society, London,
Special Publications 467 (The Geological Society, 2019),
187–197.
23. M. Vincendon et al., Geophys. Res. Lett. 37, n/a (2010).
24. O. Aharonson, N. Schorghofer, J. Geophys. Res. 111 (E11),
E11007 (2006).
25. M. Hecht, Icarus 156, 373–386 (2002).
26. K. E. Williams, O. B. Toon, J. L. Heldmann, C. McKay,
M. T. Mellon, Icarus 196, 565–577 (2008).
27. K. E. Williams, O. B. Toon, J. L. Heldmann, M. T. Mellon,
Icarus 200, 418–425 (2009).
28. N. Schorghofer, Astrophys. J. 890, 49 (2020).
29. F. Costard, F. Forget, N. Mangold, J. P. Peulvast, Science 295,
110–113 (2002).
30. M. Vincendon, J. Geophys. Res. Planets 120, 1859–1879
(2015).
31. D. E. Smith et al., J. Geophys. Res. 106, 23689–23722
(2001).
32. C. S. Edwards et al., J. Geophys. Res. Planets 116,
2010JE003755 (2011).
33. J. Dickson, B. L. Ehlmann, L. Kerber, C. I. Fassett, paper
presented at the 54th Lunar and Planetary Science Conference,
The Woodlands, TX, 13–17 March 2023.
34. J. Dickson, A. M. Palumbo, J. W. Head, L. Kerber, C. I. Fassett,
M. A. Kreslavsky, Data and software for: “Gullies on Mars could
have formed by melting of water ice during periods of high
obliquity,” Zenodo (2023); https://doi.org/10.5281/zenodo.
7946988.
ACKN OWLED GMEN TS
We thank three reviewers and the editor for the thoughtful and
detailed comments; F. Forget for helpful feedback and for making
the Laboratoire de Météorologie Dynamique General Circulation
Model available to the public; and J-B. Madeleine and C. Dundas for
helpful discussions. Funding: J.L.D. received support from a
donation to the Caltech Murray Lab by Foster and Coco Stanback.
Author contributions: Conceptualization: J.L.D., C.I.F., J.W.H.;
Investigation: J.L.D., A.M.P., L.K., J.W.H., C.I.F., M.A.K.; Methodology:
J.L.D., A.M.P., L.K.; Project administration: J.L.D., J.W.H.; Visualization:
J.L.D.; Writing: J.L.D., A.M.P., L.K., J.W.H., C.I.F., M.A.K. Competing
interests: The authors declare no competing interests. Data and
materials availability: The data we used and the analysis code we
wrote are archived at Zenodo (34). This includes the output files
of the GCM simulations (in NetCDF format), Python code for
extraction and visualization of the GCM data, and higher-resolution
versions of movies S1 to S6. We obtained the Laboratoire de
Météorologie Dynamique General Circulation Model from https://web.
lmd.jussieu.fr/~lmdz/planets/mars/. The HiRISE data were retrieved
from https://hirise-pds.lpl.arizona.edu/PDS/RDR/ (we used the
files listed in the figure captions). The MOLA global DEM was acquired
from https://pds-geosciences.wustl.edu/mgs/mgs-m-mola-5-
megdr-l3-v1/mgsl_300x/ and MOLA point data from https://pds-
geosciences.wustl.edu/mgs/mgs-m-mola-3-pedr-l1a-v1/mgsl_21xx/
data/. The global CTX mosaic was retrieved from https://murray-lab.
caltech.edu/CTX/ [we used the V01 release (33)]. Raw CTX
data used in the construction of the DEM for fig. S4 are available at
https://pds-imaging.jpl.nasa.gov/volumes/mro.html (we used
the files listed in the fig. S4 caption). The THEMIS global mosaics
(32) were retrieved from https://astrogeology.usgs.gov/search (we
used version 12). License information: Copyright © 2023 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.abk2464
Materials and Methods
Supplementary Text
Figs. S1 to S5
Tables S1 and S2
References (35–38)
Movies S1 to S6
Submitted 30 June 2021; accepted 19 May 2023
10.1126/science.abk2464
5 of 5
RES EARCH

2D MATERIALS standing-wave laser beam or arrays of opti-

cal tweezers for Rydberg atom trapping (1, 2, 4).
Observation of Rydberg moiré excitons Given the nanoscale size (≪rB) of the confin-
ing potential well, we would further elaborate
Qianying Hu1,2,3†, Zhen Zhan4,5†, Huiying Cui1,2, Yalei Zhang4, Feng Jin1, Xuan Zhao1,2, that the rigid quantum impurity approxima-
Mingjie Zhang1,2, Zhichuan Wang1,2, Qingming Zhang1,6, Kenji Watanabe7, Takashi Taniguchi8, tion (29–36) is no longer valid for Rydberg
Xuewei Cao3, Wu-Ming Liu1,2, Fengcheng Wu4,9, Shengjun Yuan4,9*, Yang Xu1,2* excitons. The XRM realize electron–hole sep-
aration and exhibit the character of long-lived
Rydberg excitons, the solid-state counterparts of Rydberg atoms, have sparked considerable charge-transfer excitons.
interest with regard to the harnessing of their quantum application potentials, but realizing
Rydberg sensing
their spatial confinement and manipulation poses a major challenge. Lately, the rise of two-
dimensional moiré superlattices with highly tunable periodic potentials provides a possible Figure 2A illustrates the typical device sche-
pathway. Here, we experimentally demonstrate this capability through the spectroscopic matic. The directly contacted TBG and mono-
evidence of Rydberg moiré excitons (X RM ), which are moiré-trapped Rydberg excitons in layer WSe2 are encapsulated by hexagonal
monolayer semiconductor tungsten diselenide adjacent to twisted bilayer graphene. In the boron nitride (hBN) dielectrics and graphite
strong coupling regime, the X RM manifest as multiple energy splittings, pronounced red shift, electrodes where the gate voltages are applied.
and narrowed linewidth in the reflectance spectra, highlighting their charge-transfer character Because of the band misalignment between
wherein electron–hole separation is enforced by strongly asymmetric interlayer Coulomb TBG and WSe2, charge carriers are only doped
interactions. Our findings establish the excitonic Rydberg states as candidates for exploitation into TBG, whereas WSe2 remains charge neu-
in quantum technologies. tral within the experimentally accessible gat-
ing range. We used a broadband light source

T
he Rydberg states of matter are ubiquitous-
ly encountered in various physical plat-
forms, ranging from atoms to molecules
to solids (1–3). They share common fea-
tures as exemplified by Bohr’s descrip-
tion of the highly excited hydrogen atoms. The
large spatial extent of the Rydberg-state wave
function promotes large dipole moments with
substantially enhanced sensitivity to weak ex-
ternal fields. Over the past two decades, Rydberg
atoms have drawn much more attention owing
to the experimental developments in trapping
and manipulation of cold atoms, facilitating
the study of quantum many-body physics and
quantum information processing (4–8). Sim-
ilarly, as the high-order Coulomb bound states
of electron–hole pairs emerged in semicon-
ductors, Rydberg excitons have also been pro-
posed to host various potential applications
such as simulating the topological Haldane
phase and implementing quantum optimiza-
tion algorithms (3, 9, 10). Their solid-state
nature allows for compatibility with modern
semiconductor technologies. However, the
requisite controllability on spatial trapping
for Rydberg excitons can be difficult to achieve
1
Beijing National Laboratory for Condensed Matter
Physics, Institute of Physics, Chinese Academy of
Sciences, Beijing 100190, China. 2School of Physical
Sciences, University of Chinese Academy of Sciences,
Beijing 100049, China. 3School of Physics, Nankai
University, Tianjin 300071, China. 4School of Physics and
Technology, Wuhan University, Wuhan 430072, China.
5
Imdea Nanoscience, 28015 Madrid, Spain. 6School of
Physical Science and Technology, Lanzhou University,
Lanzhou 730000, China. 7Research Center for Functional
Materials, National Institute for Materials Science,
Tsukuba 305-0044, Japan. 8International Center for
Materials Nanoarchitectonics, National Institute for
Materials Science, Tsukuba 305-0044, Japan. 9Wuhan
Institute of Quantum Technology, Wuhan 430206, China.
*Corresponding author. Email: s.yuan@whu.edu.cn (S.Y.);
yang.xu@iphy.ac.cn (Y.X.)
†These authors contributed equally to this work.
in bulk materials. In this work, we instead used
two-dimensional (2D) semiconductor mono-
layers (specifically WSe2), which have strong
light–matter interaction and support excitonic
Rydberg states to high orders (11–16).
In recent years, the Rydberg sensing tech-
nique has been applied to the detection of near-
by exotic electronic states and phase transitions
by the environmentally sensitive Rydberg ex-
citons in atomically thin semiconductors (17–19).
In this experiment, we placed 2D moiré su-
perlattices [specifically twisted bilayer graph-
ene (TBG); lower layer in Fig. 1] beneath the
monolayer WSe2 (upper layer in Fig. 1) to
provide spatially periodic modulations. When
the wavelength l of the potential landscape
created by TBG is smaller than (or just com-
parable to) the exciton size rB [~7 nm for the
2s states (14)], the Wannier-type exciton’s
wave packet is spread over a few moiré unit
cells and does not lose its mobile character,
as illustrated on the left side of Fig. 1. The
optical response of the system is dominated
by the Rydberg sensing scheme (17, 20).
To realize efficient trapping, the moiré po-
tential must have a spatial profile (lº q1 at
small twist angles q) that is properly larger
than the exciton size (illustrated on the right
side of Fig. 1), as has been shown for the
ground-state excitons in transition metal
dichalcogenide (TMD) heterobilayers (21–28).
In our system, the accumulated charges in the
TBG AA sites could strongly attract the op-
positely charged electron or hole of the loosely
bound 2s exciton in WSe2, thus achieving spa-
tial confinements of Rydberg moiré excitons
(XRM) with in situ controllable interaction
strength up to 75 meV tuned by the charge
density. The potential wells generated by

the periodic charge distribution  in the
of TBG
strong coupling regime rlB > ∼2:4 render
an analog of the optical lattices created by a
to excite the electron–hole pairs in monolayer
WSe2 and detect their resonance energies
through reflectance contrast (DR/R0) spec-
troscopy. A HeNe laser at 632.8 nm was used
for the photoluminescence measurements. The
energies of excitons carry information about
dielectric screenings and interactions with
the charges in TBG. More details on device
fabrication and optical measurements can
be found in (37).
We first examined the device with relatively
large twist angles, meaning small l/rB. The
doping-dependent reflectance contrast (DR/R0)
spectrum of device D1 with q = 10° TBG (l/rB ≈
0.2 for the 2s excitons) is shown in Fig. 2B. In
contrast to the barely changed 1s exciton (the
excitonic ground state of WSe2 near 1.71 eV),
the 2s exciton (resonance near 1.78 to 1.8 eV)
red shifts and merges into the renormalized
band edge with increasing carrier densities
(n). This observation is similar to that seen in
the monolayer graphene/WSe2 system (20, 37).
For TBG with a twist angle as large as 10°, the
low-energy band structure maintains the linear
dispersion of isolated graphene. In such small
l/rB limits, the neighboring 2D electron gas
(10° TBG here) provides a uniform dielectric
background to screen the Coulomb interac-
tions in WSe2. The Rydberg exciton (2s state
here) energy can be safely expressed as the
subtraction of the binding energy from the
quasiparticle bandgap, both of which get re-
normalized by increasing the density of states
in TBG. The exciton Rydberg states therefore
become delicate dielectric sensors to probe the
dielectric function and electronic compress-
ibility of the neighboring TBG. See (20, 38) for
more details.
The Rydberg sensing scheme also works for
the near-magic-angle TBG (device D2, q = 1.14°,
l/rB ≈ 1.9) as shown in Fig. 2D. The doping-
dependent 2s state (near 1.75 to 1.78 eV) ex-
hibits a symmetric sawtooth pattern around

Hu et al., Science 380, 1367–1372 (2023) 30 June 2023 1 of 6

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

tions to the formation of Rydberg moiré ex-

citons (XRM).
The photoluminescence measurements are
presented in Fig. 3B, where only the lowest-
energy branch of the XRM shows bright emis-
sion and follows an almost identical trend to
that in Fig. 3A. Such a prominent doping de-
pendence is, at first sight, notable. The lowest-
energy XRM approaches the 1s exciton resonance,
by only ~10 meV larger at nm. The energy
shift magnitude |Eshift| from the charge neu-
trality point is extracted in Fig. 3C, where the
carrier density is normalized by its full filling
Fig. 1. Schematic illustration of the interplay between a Rydberg exciton (size rB) and TBG density ns = 8.37 × 1011 cm−2. It shows a nearly
moiré superlattices with small and large periodicity (wavelength l). In the small l/rB limit (left linear dependence on the density for |n/ns|
panel), the moiré system provides a nearly uniform dielectric environment, and the optical response < ~4, as guided by the dashed lines in Fig. 3C.
of the exciton is dominated by the Rydberg sensing features. The exciton maintains its mobile character. The |Eshift| reaches its maximum energy shift
In the large l/rB limit (right panel), the Rydberg exciton can be confined by the moiré potential well Em = 73 meV on the electron-doped side,
generated by the accumulated charges in the AA site of TBG. The rB is typically an order of magnitude which is much larger than the internal electron–
larger than the interlayer spacing d, rendering much stronger interlayer interaction than the intralayer hole binding energy of the 2s state (~9 meV
one in the formed Rydberg moiré exciton (XRM) complex. for the WSe2 in proximity to undoped TBG)
(37). A simple dielectric screening cannot ac-
count for such large energy shifts.
the zero density, with periodic intensity en- bandgap, as shown by the schematic in the The interaction between an exciton and a
hancements on both the electron and hole- left panel of Fig. 2E (20). The mechanism is uniform Fermi sea has been extensively studied
doped sides, labeled as n = −4, −3, …, 3, 4, similar to the emergence of additional bright (29–36). The conventional wisdom of dealing
respectively. Among them, the n = ±4 states resonances due to the Bragg-Umklapp scatterings with the exciton is treating it as a rigid and
correspond to the gap openings when the in the exciton dispersion picture (46, 47). The mobile impurity in a degenerate Fermi system.
first moiré subbands of small-angle TBG are experimentally observed energy separation be- However, such an assumption is no longer valid
empty or fully filled, as schematically illus- tween the 2s state (or band edge) and its replica here owing to the large spatial extension of
trated in the right panel of Fig. 2E. The fea- D is ~32 meV, slightly larger than the value h2/ the Rydberg excitons and the presence of
tures at full filling densities (denoted as ns) 6mrl2 = 22 meV (where h is Planck’s constant, long-wavelength moiré potentials. Because
allow us to determine the accurate twist an- and the reduced exciton mass mr is ~0.15 free the change in intralayer binding energy is
gles in a wide range from ~0.73° to 1.6° in our electron mass) expected from the weak per- much smaller compared with the interlayer
devices (37). turbation limit. interaction, the energy shift of XRM reflects
On the other hand, the n = ±3 (±2, ±1, and 0) the Coulomb interaction energy between the
states, corresponding to 3 (2, 1, and 0) charges Rydberg moiré excitons at smaller twist angles 2s exciton and free charges in TBG. A plau-
per moiré site, are beyond the scope of single- When the twist angle q is further reduced, sible explanation is related to the spatially
particle band theory. In the right panel of we observe a pronounced enhancement of the confined charge distributions in TBG that
Fig. 2D, these features show little temper- interlayer Rydberg exciton-charge interactions, can help promote unequal interlayer inter-
ature dependence in the range of 1.6 to 10 K, indicating access to the strong coupling regime. actions for the constituent electron and hole
well above the onset of correlated insulating Similar reflectance spectrum measurements of the exciton.
states observed by transport measurements in device D3 with q ≈ 0.6° TBG (l ≈ 23.5 nm, We hence carried out numerical simulations
(39, 40). They are consistent with the ob- l/rB ≈ 3.6) is shown in Fig. 3A. Upon doping to extract the real-space charge distribution in
servations by scanning tunneling microscopy the TBG with either positively or negatively TBG at different doping levels, as shown in
and local electronic compressibility measure- charged carriers, the 1s resonance only slightly Fig. 3E [see (37) and figs. S6 to S8 for further
ments (41, 42), both of which reveal a cas- shifts, whereas the 2s state near 1.783 eV splits information]. Starting from the charge neutral
cade of fourfold (spin and valley degrees of into multiple branches and shows nonmono- point, the local charge density rises drastical-
freedom) symmetry-broken states. We have tonic doping dependencies. We focus on the 2s ly at the AA-stacked regions first, while it is
observed these states with a twist angle rang- state in the following paragraphs, and the merely changed at the AB/BA-stacked regions.
ing from 1.06° to 1.15°, demonstrating the detailed discussion on other excitonic states is This is closely related to the much larger local
prominent role of electronic correlations in presented in (37). The resonance energies red density of states (LDOS) of the AA site at small
the near-magic-angle TBG with band flatten- shift first and then blue shift, reaching their densities. While the AA-stacked region has
ing (39–45). minima at densities nm ≈ 3.8 × 1012 cm−2 for a radius of ~2.6 nm (estimated from the full
In addition to the Rydberg sensing features the electron-doped side and −4.7 × 1012 cm−2 for width at half maximum of the spatially en-
discussed above, the spatially periodic dielec- the hole-doped side. Some of the main branches hanced charge accumulation peak) that is
tric screening environment provided by TBG could survive at relatively high temperatures up much smaller than rB, the areas of the AB/BA
modulates the spectrum as well. The emer- to 140 K, as shown in fig. S5. The splitting be- stacked region are greatly enlarged owing
gence of the replica at higher energies (differ havior reminds us of the multiple peaks of the to lattice reconstruction (with the schematic
by D) is a manifestation of the formation of ground-state moiré excitons observed in the superlattices shown in the lowest map in Fig.
moiré bands in WSe2 (20). additional optical TMD heterobilayers, suggesting that the ex- 3E) (48). Taking the electron-doped side as
transitions between states at high-symmetry citon wave function resides at different moiré an example, the accumulated charges cen-
points of the mini-Brillouin zone boundary stacking sites and experiences inequivalent tered at the AA-stacked region create deep
become allowed above the fundamental potentials (20–27). We attribute our observa- and narrow potential wells for trapping the

Hu et al., Science 380, 1367–1372 (2023) 30 June 2023 2 of 6

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

A B C
-0.67 0.45
WSe2
10° TBG
R/R0

Vtg
TBG
EF
Eg
1s 2s
WSe2
Vbg
19
10° TBG

D E
-0.4 1.6 1.6 K
WSe2 1.14° TBG
R/R0

4
5K
3
2
1 E g+ Eg
1s 2s 0
-1
10 K
-2
-3
17 -4 K’ K
1.14° TBG

d( R/R0)/dE

Fig. 2. Rydberg sensing of WSe2 adjacent to 10° and 1.14° TBG. (A) Schematic
structure of a typical device with electrically grounded TBG and monolayer WSe2
embedded in hBN/graphite dual gates. (B) Doping-dependent reflectance contrast
(DR/R0) spectrum of device D1 with 10° TBG (l/rB = 0.2). The WSe2 serves as a
dielectric sensor, whose 2s exciton energy reflects the dielectric screening of
the neighboring TBG. (C) Schematic band alignment between monolayer WSe2 and
10° TBG. (D) Doping-dependent DR/R0 spectrum of device D2 with 1.14° TBG
(l/rB = 1.9). Sawtooth features appear at filling factors −4 to 4 (highlighted by the
red rectangle), and a replica is observed at higher energies of D = 32 meV. The right
three panels show the temperature-insensitive sawtooth features at 1.6, 5, and
10 K, respectively. (E) Schematic band structure of near-magic-angle TBG
and the adjacent monolayer WSe2. The TBG features gap openings and flat
bands, which result in the band insulating states (n = ±4) and the cascade of
symmetry-breaking phase transitions (n = 0, ±1, ±2, ±3) observed in (D),
respectively. Meanwhile, the spatially periodic screening of TBG folds the band
of WSe2 into the mini-Brillouin zone, generating a new optically allowed transition
at Eg + D. The DR/R0 spectra above 1.74 eV in (B) and (D) are multiplied by a
factor of 19 and 17, respectively, for better illustration.

hole of the exciton, while the trend to mini-
mize the repulsion energy will push the elec-
tron of the exciton toward the AB/BA-stacked
regions, as schematically shown in the inset
of Fig. 3D. The role of electron and hole is re-
versed on the hole-doped side. This process
renders a spatial separation of the electron–
hole pair, supporting a charge-transfer–type
exciton that typically only forms in molecular
crystals with electron or hole occupying adja-
cent molecules or across an interface between
two kinds of materials (49).
In this scenario, with one charge residing
at the AA site and the other charge mostly on
the AB/BA site, the electron and hole of the
Rydberg exciton have highly asymmetric elec-
trical potential energies. The Eshift can then be
approximately estimated from the difference
in attraction on the AA site and repulsion on
the AB/BA site as Eshift ≈ (eUAA − eUAB/BA) º
(nAA − nAB/BA), where UAA (UAB/BA) and nAA
(nAB/BA) are the electric potential and carrier
density at the AA (AB/BA) site, respectively.
Quantitatively, the calculated nAA − nAB/BA
as a function of n/ns is presented in Fig. 3D,
where a nonmonotonic trend similar to that
of Eshift is observed, with the critical densi-
ties (nm) almost identical to our experimental
observations on both the electron and hole-
doped sides. As the doping reaches nm, where
LDOS of TBG nearly equalizes at the two sites,
the charge density at the AB/BA site begins
to grow faster than that in the AA site (see
detailed charge-filling maps in fig. S7), and the
lattice becomes more evenly filled at higher
doping densities. Such a process explains the
blue shift for XRM at |n| > |nm| and the con-
vergence of the multiple branches in the large
density limit.
Meanwhile, considering the size of the 2s
exciton (rB ~ 7 nm, which could be underesti-
mated here owing to the enhanced screening
effect) that is about an order of magnitude
larger than the interlayer spacing d (≈0.5 nm),
the interlayer Coulomb interaction with TBG
charges can be much stronger than the intra-
layer one. It is thus possible to see the pro-
nounced energy shift of the lowest-energy
branch that is much larger than its original
binding energy after the TBG is doped. We
reproduced these results in another ~0.6° de-
vice and performed control experiments by in-
serting additional hBN spacers to increase
the interlayer distance, as shown in fig. S9. The

Hu et al., Science 380, 1367–1372 (2023) 30 June 2023 3 of 6

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

Fig. 3. Rydberg moiré exciton formation in WSe2 adjacent to 0.6° TBG.
(A) Reflectance contrast spectrum of device D3 with 0.6° TBG (l/rB = 3.6). The
WSe2 2s exciton resonance splits into multiple branches and exhibits
nonmonotonic dependences when the TBG is doped. The DR/R0 spectra
above 1.71 eV are multiplied by 30. (B) Photoluminescence spectrum of
device D3 measured at the same location. (C) Extracted energy shift of the
lowest-energy branch in (A) as a function of n/n s (n s denoting the full filling
density of the first moiré band). (D) The calculated local carrier density
difference between the AA and AB/BA sites with varying n/n s . Insets are
the schematic exemplification of the lowest-energy X RM confinement on
the electron-doped side. The moiré potential landscape facilitates the
charge-transfer–type exciton configuration with the hole (blue sphere)
residing on the AA site and the electron (red sphere) on the AB/BA site.
The n AA – n AB/BA versus n/n s approximately reproduces the energy shift in
(C) as E shift ≈ (eUAA – eUAB/BA ) º (n AA – n AB/BA ). a.u., arbitrary units.
(E) Calculated spatial charge distribution (in logarithmic scales) of 0.6°
TBG at representative doping densities. The lowest map is a schematic of
relaxed TBG moiré superlattices with AA, AB, and BA sites marked.

energy shift is then strongly suppressed, and
the system exhibits characteristics with weak
interlayer interactions (37).
Crossover between the weak and strong
coupling regimes
To better reveal the evolution to the strong cou-
pling regime, more devices were fabricated and
measured with twist angles ranging from ~0.6°
to 1.23°, as presented in Fig. 4A. Reflection con-
trast and photoluminescence of additional de-
vices are given in fig. S10. As the twist angle
becomes smaller, the maximum red shift (de-
noted as Em, at the critical density nm) of the
2s state or the lowest-energy XRM keeps grow-
ing, accompanied by the emergence of more
replica states. Notably, as shown in the bottom-
right panel of Fig. 4A, the energy separation
between 1s and 2s states at nm nearly vanishes
in the smallest-twist-angle device where q is
slightly smaller than 0.6°, suggesting a strong
interlayer Coulomb interaction of the XRM that
is even comparable to the binding energy of the
1s state and the possibility of forming new mo-
lecular states (50).
We summarize the twist-angle dependences
of Em and nm on the electron-doped side in
Fig. 4, B and C, respectively. The Em is a direct
evaluation of the maximum interlayer inter-
action, and the experimentally obtained nm at
three twist angles is in good agreement with
that obtained from the theoretical calculations
given in fig. S8. The nm/ns and Em share a
similar trend upon varying the angle q, likely
stemming from the positive correlation be-
tween charge accumulation at each moiré site
and the interlayer Coulomb interaction en-
ergy. Meanwhile, an obvious reduction of XRM
linewidth (extracted for the lowest-energy
branch) from ~8 meV to ~1.5 meV is observed
with decreasing the twist angle into the strong
coupling regime (Fig. 4D). The reduced line-
width is consistent with our interpretation of
XRM where the spatially separated electron–
hole configuration is likely to support longer
coherent lifetimes.
In this study, we developed and experimen-
tally demonstrated a method for spatially con-
fining and manipulating Rydberg excitons
using long-wavelength moiré potentials. The
strongly bound XRM complex can be domi-
nated by interlayer interactions and approach
the energy of ground-state excitons. The sys-
tem provides easy access to control the po-
tential well depth by electrostatic doping, to
tune the moiré wavelength by the twist angle,
and to achieve longer lifetimes guaranteed
by the electron–hole separation. All these fea-
tures would be helpful for further realizing
excitonic Rydberg-Rydberg interactions and
coherent controls. Our study could open up un-
precedented opportunities for the implemen-
tation of quantum information processing

Hu et al., Science 380, 1367–1372 (2023) 30 June 2023 4 of 6

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

A 10
-0.4 0.5
B 80
n ( 1012 cm-2)

5 60

E (meV)
nm

40

m
0
Em
10 20 20
1.23° 0.8°
-5
C 5
Calculation
4 Experiment

s
n /n
m
3

20 30 2
1.06° ӗ0.6°

D 8

Linewidth (meV)
6

2
20 30
0.88° < 0.6°
0
0.6 0.8 1 1.2
(°)

Fig. 4. Twist-angle dependences and crossover to the strong coupling
regime. (A) Doping-dependent reflectance contrast spectra of devices with
1.23°, 1.06°, 0.88°, 0.8°, ~0.6°, and <0.6° TBG. The nearly parallel 2s resonance
and replica evolve into XRM with increasing l/rB (decreasing twist angle).
(B to D) Twist-angle dependence of the maximum energy shift Em (B), the
normalized critical density nm/ns (C), and the estimated linewidth (D) extracted
from the 2s resonance or the lowest-energy branch of XRM on the electron-
doped side. The critical density expected from the calculated nAA – nAB/BA is
marked in blue (C), in good agreement with the experimental results. The
strongly reduced optical linewidth for q < ~0.9° is an indication of longer
coherent lifetimes of the XRM, in accordance with its charge-transfer character
in the strong coupling regime.

and quantum simulation based on the ver-
satile Rydberg states in solid-state systems
(9, 10).
RE FE RENCES AND N OT ES
1. T. F. Gallagher, Rep. Prog. Phys. 51, 143–188 (1988).
2. N. Šibalić, C. S. Adams, Rydberg Physics (IOP Publishing,
2018).
3. M. Aßmann, M. Bayer, Adv. Quantum Technol. 3, 1900134
(2020).
4. M. Saffman, T. G. Walker, K. Mølmer, Rev. Mod. Phys. 82,
2313–2363 (2010).
5. E. Urban et al., Nat. Phys. 5, 110–114 (2009).
6. H. Weimer, M. Müller, I. Lesanovsky, P. Zoller, H. P. Büchler,
Nat. Phys. 6, 382–388 (2010).
7. H. Labuhn et al., Nature 534, 667–670 (2016).
8. D. Bluvstein et al., Science 371, 1355–1359 (2021).
9. A. N. Poddubny, M. M. Glazov, Phys. Rev. Lett. 123, 126801
(2019).
10. J. Taylor et al., Quantum Sci. Technol. 7, 035016
(2022).
11. K. He et al., Phys. Rev. Lett. 113, 026803 (2014).
12. A. Chernikov et al., Phys. Rev. Lett. 113, 076802
(2014).
13. A. Raja et al., Nat. Commun. 8, 15251 (2017).
14. A. V. Stier et al., Phys. Rev. Lett. 120, 057405
(2018).
15. K. F. Mak, D. Xiao, J. Shan, Nat. Photonics 12, 451–460
(2018).
16. G. Wang et al., Rev. Mod. Phys. 90, 021001
(2018).
17. Y. Xu et al., Nature 587, 214–218 (2020).
18. Z. Zhang et al., Nat. Phys. 18, 1214–1220 (2022). 43. R. Bistritzer, A. H. MacDonald, Proc. Natl. Acad. Sci. U.S.A. 108,
19. A. Popert et al., Nano Lett. 22, 7363–7369 (2022). 12233–12237 (2011).
20. Y. Xu et al., Nat. Mater. 20, 645–649 (2021). 44. L. Balents, C. R. Dean, D. K. Efetov, A. F. Young, Nat. Phys. 16,
725–733 (2020).
21. K. L. Seyler et al., Nature 567, 66–70 (2019).
22. K. Tran et al., Nature 567, 71–75 (2019). 45. E. Y. Andrei, A. H. MacDonald, Nat. Mater. 19, 1265–1275
23. C. Jin et al., Nature 567, 76–80 (2019). (2020).
46. T. Smoleński et al., Nature 595, 53–57 (2021).
24. E. M. Alexeev et al., Nature 567, 81–86 (2019).
25. N. P. Wilson, W. Yao, J. Shan, X. Xu, Nature 599, 383–392 47. Y. Shimazaki et al., Phys. Rev. X 11, 021027 (2021).
(2021). 48. H. Yoo et al., Nat. Mater. 18, 448–453 (2019).
26. D. Huang, J. Choi, C. K. Shih, X. Li, Nat. Nanotechnol. 17, 49. X. Y. Zhu, Q. Yang, M. Muntwiler, Acc. Chem. Res. 42,
227–238 (2022). 1779–1787 (2009).
27. E. C. Regan et al., Nat. Rev. Mater. 7, 778–795 50. I. Schwartz et al., Science 374, 336–340 (2021).
(2022). 51. Q. Hu et al., Data repository for “Observation of Rydberg moiré
28. M. H. Naik et al., Nature 609, 52–57 (2022). excitons,” version 1, Zenodo (2023); https://doi.org/10.5281/
29. R. A. Suris, Optical Properties of 2D Systems with Interacting zenodo.7944742.
Electrons (Springer Netherlands, 2003).
30. A. V. Koudinov et al., Phys. Rev. Lett. 112, 147402 AC KNOWLED GME NTS
(2014). We thank X. Cui for helpful discussions. Funding: This work was
31. Y. C. Chang, S. Y. Shiau, M. Combescot, Phys. Rev. B 98, supported by the National Key R&D Program of China (grant nos.
235203 (2018). 2021YFA1401300, 2021YFA1400900, 2021YFA0718300,
32. M. Sidler et al., Nat. Phys. 13, 255–261 (2017). 2021YFA1402100, and 2022YFA1402704), the National Natural
33. D. K. Efimkin, A. H. MacDonald, Phys. Rev. B 95, 035417 Science Foundation of China (grant nos. 12174439, 11974263,
(2017). 12174291, 61835013, 12174461, 12234012, 12104491, U1932215,
34. C. Fey, P. Schmelcher, A. Imamoglu, R. Schmidt, Phys. Rev. B and 12274186), the Strategic Priority Research Program of the
101, 195417 (2020). Chinese Academy of Sciences (grant no. XDB33010100), the
35. K. Wagner et al., Phys. Rev. Lett. 125, 267401 (2020). Innovation Program for Quantum Science and Technology (grant
36. E. Liu et al., Nat. Commun. 12, 6131 (2021). no. 2021ZD0302400), Supercomputing Center of Wuhan
37. See supplementary materials. University, and the Synergetic Extreme Condition User Facility
38. A. C. Riis-Jensen, J. Lu, K. S. Thygesen, Phys. Rev. B 101, (SECUF). Z.Z. was partially supported by the European Union’s
121110 (2020). Horizon 2020 research and innovation programme under the Marie
39. Y. Cao et al., Nature 556, 80–84 (2018). Skłodowska-Curie grant no. 101034131 and the “Severo Ochoa”
40. X. Lu et al., Nature 574, 653–657 (2019). Programme for Centres of Excellence in R&D (CEX2020-001039-
41. D. Wong et al., Nature 582, 198–202 (2020). S/AEI/10.13039/501100011033). The growth of hBN crystals
42. U. Zondiner et al., Nature 582, 203–208 (2020). was supported by the Elemental Strategy Initiative of MEXT, Japan,

Hu et al., Science 380, 1367–1372 (2023) 30 June 2023 5 of 6

RES EARCH | R E S E A R C H A R T I C L E
and CREST (JPMJCR15F3), JST. Author contributions: Y.X.
and S.Y. conceived of and supervised the project. Q.H., H.C., F.J.,
X.Z., M.Z., and Z.W. built the experimental setup under the
supervision of Q.Z., X.C., and Y.X. Q.H. fabricated the devices,
performed the measurements, and analyzed the data. K.W.
and T.T. grew the bulk hBN crystals. Z.Z., Y.Z., and S.Y. performed
the theoretical calculations, with substantial input from F.W.
and W.-M.L. Q.H. and Y.X. designed the scientific objectives.
Q.H., Z.Z., S.Y., and Y.X. co-wrote the manuscript. All authors
Hu et al., Science 380, 1367–1372 (2023) 30 June 2023
discussed the results and commented on the manuscript. SUPPLEMENTARY MATERIALS
Competing interests: The authors declare no competing interests. science.org/doi/10.1126/science.adh1506
Data and materials availability: All data are available in this Materials and Methods
paper or the supplementary materials or are deposited at Supplementary Text
Zendo (51). License information: Copyright © 2023 the authors, Figs. S1 to S10
some rights reserved; exclusive licensee American Association References (52–63)
for the Advancement of Science. No claim to original US
government works. https://www.science.org/about/science- Submitted 15 February 2023; accepted 24 May 2023
licenses-journal-article-reuse 10.1126/science.adh1506
6 of 6
RES EARCH

SIGNAL TRANSDUCTION including colon cancer HT-29 cells, mouse

microglia BV2 cells, human T lymphocyte
Metabolic orchestration of cell death by Jurkat cells, and human U2OS osteosarcoma
cells (fig. S1, D to I). Increased insolubility is
AMPK-mediated phosphorylation of RIPK1 another hallmark of activated RIPK1, RIPK3,
and MLKL (22). In glucose-starved MEFs, the
Tao Zhang1†, Daichao Xu2,3†*, Elijah Trefts4†, Mingming Lv1†, Hiroyuki Inuzuka1, Guobin Song5, amounts of RIPK1, RIPK3, and MLKL were
Min Liu6, Jianlin Lu7, Jianping Liu2, Chen Chu8,9, Min Wang10, Huibing Wang3, Huyan Meng11, Hui Liu1, decreased in a mild-detergent (NP-40)–soluble
Yuan Zhuang12, Xingxing Xie2, Fabin Dang1, Dongxian Guan5, Yuqin Men5, Shuwen Jiang13,5, Cong Jiang1, fraction but increased in an NP-40–insoluble
Xiaoming Dai1, Jing Liu1, Zhen Wang1, Peiqiang Yan1, Jingchao Wang1, Zhenbo Tu1, Mrigya Babuta12, and 6M urea–soluble fraction, which was large-
Emily Erickson1, Alissandra L. Hillis1, Christian C. Dibble1, John M. Asara14, Gyongy Szabo12, ly blocked by the Ripk1D138N/D138N mutation
Piotr Sicinski8,9, Ji Miao5, Yu-Ru Lee15, Lifeng Pan16*, Reuben J. Shaw4*, Junying Yuan2,3*, Wenyi Wei1* (Fig. 1C and fig. S1, C, F, and G). We also ob-
served interaction of RIPK1 and RIPK3 in MEFs
Adenosine monophosphate–activated protein kinase (AMPK) activity is stimulated to promote metabolic after glucose starvation (fig. S1J). Glucose dep-
adaptation upon energy stress. However, sustained metabolic stress may cause cell death. The mechanisms by rivation induced apoptosis, as marked by the
which AMPK dictates cell death are not fully understood. We report that metabolic stress promoted receptor- cleavage of RIPK1 and caspase-3 (CC3) (Fig. 1B
interacting protein kinase 1 (RIPK1) activation mediated by TRAIL receptors, whereas AMPK inhibited RIPK1 and fig. S1K). We exposed cells to various con-
by phosphorylation at Ser415 to suppress energy stress–induced cell death. Inhibiting pS415-RIPK1 by centrations of glucose and found that only
Ampk deficiency or RIPK1 S415A mutation promoted RIPK1 activation. Furthermore, genetic inactivation of severe energy stress induced necroptosis and
RIPK1 protected against ischemic injury in myeloid Ampka1-deficient mice. Our studies reveal that AMPK apoptosis (Fig. 1D). Prolonged treatment of
phosphorylation of RIPK1 represents a crucial metabolic checkpoint, which dictates cell fate response to cells with the glucose analog 2-deoxyglucose
metabolic stress, and highlight a previously unappreciated role for the AMPK-RIPK1 axis in integrating (2DG), which blocks cellular glucose utilization
metabolism, cell death, and inflammation. by indirectly inhibiting hexokinase, also acti-
vated RIPK1 to promote necroptosis (fig. S1L).

A
denosine monophosphate (AMP)–activated
protein kinase (AMPK) is an evolution-
arily conserved sensor of cellular nutrient
status and regulator of energy homeosta-
sis in eukaryotes (1). In response to in-
creases in intracellular AMP that always
accompany decreases in adenosine triphos-
phate (ATP), AMPK is activated and serves
as a metabolic checkpoint (1–4). However,
1
Department of Pathology, Beth Israel Deaconess Medical
Center, Harvard Medical School, Boston, MA 02215, USA.
2
Interdisciplinary Research Center on Biology and Chemistry,
Shanghai Institute of Organic Chemistry, Chinese Academy
of Sciences, 201203 Shanghai, China. 3Department of Cell
Biology, Harvard Medical School, Boston, MA 02115, USA.
4
The Salk Institute for Biological Studies, La Jolla, CA 92037,
USA. 5Division of Endocrinology, Boston Children’s Hospital,
Harvard Medical School, Boston, MA 02115, USA.
6 to TNFR (13–18). Notably, inhibition of RIPK1
Transfusion Medicine, Boston Children’s Hospital, Harvard
Medical School, Boston, MA 02115, USA. 7Department of
Immunology, St. Jude Children’s Research Hospital,
Memphis, TN 38105, USA. 8Department of Cancer Biology,
Dana-Farber Cancer Institute, Boston, MA 02215, USA.
9
Department of Genetics, Blavatnik Institute, Harvard
Medical School, Boston, MA 02115, USA. 10Department of
Biliary-Pancreatic Surgery, Affiliated Tongji Hospital, Tongji
Medical College, Huazhong University of Science and
Technology, 430030 Wuhan, Hubei, China. 11F.M. Kirby
Neurobiology Center, Boston Children’s Hospital, Boston,
MA 02115, USA. 12Department of Medicine, Beth Israel
Deaconess Medical Center, Harvard Medical School, Boston,
MA 02215, USA. 13Department of Metabolic and Bariatric
Surgery, The First Affiliated Hospital of Jinan University,
510632 Guangzhou, China. 14Division of Signal Transduction,
Department of Medicine, Beth Israel Deaconess Medical
Center, Harvard Medical School, Boston, MA 02215, USA.
15
Institute of Biomedical Sciences, Academia Sinica, Taipei
115201, Taiwan. 16State Key Laboratory of Bioorganic and
Natural Products Chemistry, Center for Excellence in
Molecular Synthesis, Shanghai Institute of Organic
Chemistry, University of Chinese Academy of Sciences,
Chinese Academy of Sciences, 200032 Shanghai, China.
*Corresponding author. Email: xudaichao@sioc.ac.cn (D.X.);
panlf@sioc.ac.cn (L.P.); shaw@salk.edu (R.J.S.);
junying_yuan@sioc.ac.cn (J.Y.); wwei2@bidmc.harvard.edu (W.W.)
†These authors contributed equally to this work.
when extensive metabolic stress overrides
AMPK-mediated adaptation, it activates cell
death (5, 6). The mechanisms by which AMPK
modulates cell survival under metabolic stress
are not fully understood. Receptor-interacting
protein kinase 1 (RIPK1) is a key mediator of
cell death and inflammation (7, 8). Activated
RIPK1 may mediate receptor-interacting protein
kinase 3 (RIPK3) and mixed lineage kinase
domain-like (MLKL)–dependent necroptosis or
caspase-8–dependent apoptosis upon stimulation
of tumor necrosis factor receptor 1 (TNFR1) by
tumor necrosis factor–a (TNFa) (9–12). RIPK1
is also activated downstream of the death re-
ceptors of Fas (also called CD95), as well as
TRAIL receptor 1 (TRAIL-R1, also called DR4)
and TRAIL-R2 (also called DR5) in addition
activation is highly effective in protecting against
ischemic damage (19, 20). However, it remains
unknown whether and how RIPK1 may be ac-
tivated under ischemic conditions.
Metabolic stress promotes the activation
of RIPK1
To this end, we incubated cells with glucose-
free medium for various lengths of time. Wild-
type (WT) mouse embryonic fibroblasts (MEFs)
died after 48 hours of glucose deprivation,
whereas most MEFs expressing catalytically in-
active RIPK1 (Ripk1D138N/D138N) survived (Fig.
1A and fig. S1A). Moreover, glucose deprivation
increased phosphorylation of RIPK1 (S166), a
biomarker of RIPK1 activation (21, 22), as well
as that of RIPK3 (T231/S232) and MLKL (S345),
hallmarks of necroptosis (11), which were all
blocked in Ripk1D138N/D138N MEFs (Fig. 1, B
and C, and fig. S1, B and C). Glucose depriva-
tion activates RIPK1 in multiple cell lines,
RIPK1 promotes cell death through its
kinase activity, whereas it inhibits cell death
through its scaffold functions (21, 23–25). Un-
like Ripk1D138N/D138N MEFs, which showed re-
sistance to glucose deprivation–induced cell
death, Ripk1 knockout (KO) MEFs were more
sensitive to cell death induced by glucose star-
vation than were WT MEFs (fig. S2A). However,
in Jurkat cells, RIPK1 deficiency decreased cell
death and the activation of RIPK3 and MLKL
induced by glucose starvation (fig. S2, B and
C), indicating that the effects of RIPK1 de-
ficiency on cell death in response to glucose
starvation depend on the cellular context. To
investigate the role of necroptosis in glucose
deprivation–induced cell death, we deprived
WT, Ripk3, and Mlkl KO cells of glucose for
various lengths of time. Deletion of either
Ripk3 or Mlkl partially prevented cell death
induced by glucose deprivation (fig. S2, D to
F). Notably, the loss in viability and the degree
of RIPK1 activation were comparable between
WT and Tnfr1/2 double knockout (DKO) cells
(fig. S2, G and H), which indicates that glucose
deprivation–induced activation of RIPK1 and
cell death is independent of the TNFa signal-
ing pathway.
Next we subjected WT mice, Ripk1D138N/D138N
mice, and Tnfr1/2 DKO mice to hepatic ische-
mia, a condition that causes mouse livers to
become pale, enlarged, and damaged (fig. S3, A
and B). Histological analysis showed severe
inflammatory infiltration and cell death in
ischemic WT and Tnfr1/2 DKO mice but not
in Ripk1D138N/D138N mice (Fig. 1E). We detected
activation of RIPK1 as determined by p-RIPK1
(S166) immunostaining and cell death as deter-
mined by terminal deoxynucleotidyl transferase–
mediated deoxyuridine triphosphate nick end
labeling (TUNEL) assay in the livers of WT

Zhang et al., Science 380, 1372–1380 (2023) 30 June 2023 1 of 9

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

Zhang et al., Science 380, 1372–1380 (2023) 30 June 2023 2 of 9

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

Fig. 1. Metabolic stress promotes cell death and inflammation in a RIPK1-
dependent manner. (A) WT or Ripk1D138N/D138N MEFs were cultured in glucose-
free medium for the indicated times (hours), followed by cell viability analyses
using propidium iodide (PI) uptake assay. Data are mean ± SD of n = 3 biological
independent samples. Two-way analysis of variance (ANOVA); ***P < 0.001.
(B) MEFs were cultured in glucose-free medium for the indicated times. The
levels of proteins were determined by immunoblotting. n = 3 independent
biological repeats. (C) WT or Ripk1D138N/D138N MEFs were cultured in glucose-free
medium for the indicated times. The levels of proteins in mild-detergent
(NP-40)–soluble fraction and 6 M urea–soluble fraction were determined by
immunoblotting. n = 3 independent biological repeats. (D) MEFs were cultured in
the medium with different glucose concentrations for 36 hours. The levels of
proteins were determined by immunoblotting. n = 3 independent biological
repeats. (E) WT, Tnfr1/2 DKO, and Rpik1D138N/D138N mice were subjected to a
protocol of liver ischemia for 18 hours. Histological analysis and immunostaining
for p-RIPK1(S166) were performed on liver sections (n = 4 mice in each group).
DAPI (4′,6-diamidino-2-phenylindole) for nuclei. Representative images are
shown. Quantification of p-RIPK1(S166)–positive cells is shown at the bottom.
Data are mean ± SEM, one-way ANOVA, post hoc Dunnett’s test; ***P < 0.001;
n.s., not significant. Scale bars, 100 mm. (F) Heatmap of genes differentially
expressed in the whole livers derived of WT, Tnfr1/2 DKO, and Rpik1D138N/D138N
mice subjected to liver ischemia for 18 hours. (G) Gene Ontology analysis of
genes that are up-regulated in the livers of mice subjected to liver ischemia for
18 hours in a RIPK1-dependent manner. (H) ELISA analyses of the levels of
indicated cytokines and chemokines in serum from WT, Tnfr1/2 DKO, and
Rpik1D138N/D138N mice subjected to liver ischemia for 18 hours. Data are
presented as mean ± SEM, and each dot represents one mouse. One-way
ANOVA, post hoc Dunnett’s test; *P < 0.05, ***P < 0.001.

and Tnfr1/2 DKO mice but not in those of
Ripk1D138N/D138N mice (Fig. 1E and fig. S3C).
We analyzed the gene expression profiles of
whole livers by RNA sequencing (RNA-seq).
Compared to WT sham control mice, WT and
Tnfr1/2 DKO mice showed increased expres-
sion of genes associated with an inflammatory
response, including interferon alpha (Ifn-a),
transforming growth factor beta (TGF-b), C-X-C
motif chemokine ligand 10 (Cxcl10), C-X-C mo-
tif chemokine ligand 2 (Cxcl2), and C-C motif
chemokine ligand 2 (Ccl2) (Fig. 1, F and G, and
fig. S3, D and E); and down-regulation of genes
involved in oxidation-reduction processes and
the transport signaling pathway (fig. S3, F and
G). These gene expression abnormalities were
largely rescued by genetic inhibition of RIPK1
in Ripk1D138N/D138N mice (Fig. 1, F and G, and
fig. S3, F and G). We confirmed the increased
production of Ccl2, Cxcl2, interleukin 1 alpha
(Il1a), interleukin 1 beta (Il1b), and interleu-
kin 6 (Il6) by enzyme-linked immunosorbent
assay (ELISA), all of which were reduced in the
blood of Ripk1D138N mice but not in that of
Tnfr1/2 DKO mice (Fig. 1H). Thus, these re-
sults indicate that ischemia promotes RIPK1-
dependent cell death and inflammation in a
TNFa-independent manner.
We next explored the possible roles for other
death receptors including Fas, DR4, and DR5
in the regulation of RIPK1 activity upon glu-
cose starvation. Deletion of DR4 or DR5 but
not FAS reduced RIPK1 activation and cell
death (fig. S4, A to F). Glucose deprivation
activates TRAIL receptors through an activat-
ing transcription factor 4 (ATF4) and C/EBP
homologous protein (CHOP)–dependent mecha-
nism, which promotes cell death in a ligand-
independent manner (26). Consistently, we
observed increased up-regulation of TRAIL
receptors after glucose starvation (fig. S4, D
and F). We used the liver ischemia injury mouse
model to examine the role of Dr5, which is the
single TRAIL receptor expressed in mice (27).
Dr5 deficiency suppressed RIPK1 activation,
cell death, and inflammation (fig. S4, G and
H). Thus, our data demonstrate that TRAIL
receptors function as upstream drivers for
RIPK1 activation and cell death under glucose
deprivation conditions.
AMPK phosphorylates RIPK1 in response to
metabolic stress
AMPK is a highly conserved sensor of cellular
energy status, which is activated under condi-
tions of low cellular energy such as glucose
deprivation (28). RIPK1 derived from cells ex-
posed to glucose deprivation showed a pro-
found RIPK1 mobility shift on standard SDS–
polyacrylamide gel electrophoresis (SDS-PAGE),
which was diminished by the treatment of
the protein with lambda-phosphatase. Thus,
glucose deprivation appears to induce RIPK1
phosphorylation (Fig. 2, A and B, and fig. S5,
A and B). Moreover, this glucose deprivation–
induced RIPK1 mobility shift was not reversed
by RIPK1-specific inhibitor necrostatin-1s
(Nec-1s) (19) (fig. S5, C to E). These data in-
dicate that glucose deprivation–induced RIPK1
phosphorylation is not entirely dependent on
RIPK1 autophosphorylation. In keeping with
this notion, cells expressing RIPK1 catalyt-
ically inactive mutant (K45M) (29) showed a
strong upshifted band, which was diminished
in AMPK DKO cells, indicating that AMPK
might target RIPK1 for phosphorylation (fig.
S5F). We conducted mass spectrometry analy-
sis of RIPK1 phosphorylation under glucose
starvation conditions and found that Ser416
of RIPK1 is likely a specific phosphorylation
site in response to glucose deprivation (fig. S6,
A and B). Further sequence analysis showed
that the RIPK1 Ser416 residue conforms to the
AMPK substrate motif validated in some pre-
viously identified AMPK substrates (30) (fig.
S6C) and represents an evolutionarily con-
served residue in human, rat, and mouse
(fig. S6C).
To examine endogenous RIPK1 phosphoryl-
ation, we developed a phosphospecific anti-
body against p-S416 of human RIPK1, which
corresponds to p-S415 of mouse RIPK1 (fig. S6,
D and E). p-S416 of RIPK1 was increased in
response to a constitutively active AMPKa1
allele (31) (fig. S7A). We observed increased
endogenous p-S415 of murine RIPK1 or p-S416
of human RIPK1 in WT but not in Ampk DKO
MEFs and HT29 cells, respectively, after glu-
cose starvation (Fig. 2, C and D). We tested
whether direct pharmacological activation of
AMPK was sufficient to induce phosphoryla-
tion of RIPK1. Phosphorylation of RIPK1 at
Ser415 was induced by the treatment with AMP-
mimetic aminoimidazole carboxamide ribo-
side (AICAR) in an AMPK-dependent manner
(Fig. 2E and fig. S7B). Similarly, treatment
with compound 991, a small molecule that di-
rectly binds to and activates AMPK (32), was
sufficient to induce RIPK1 phosphorylation at
Ser415 (Fig. 2F). Consistently, phosphorylation
of RIPK1 was also increased in response to
treatment with metformin, a widely prescribed
drug for type 2 diabetes that is known to ac-
tivate AMPK both in vitro and in vivo (33) (Fig.
2G and fig. S7, C and D).
Liver kinase B1 (LKB1) is the major kinase
that activates AMPK under conditions of en-
ergy stress (34). Treatment with metformin
or glucose starvation resulted in the phos-
phorylation of RIPK1 at Ser415 , which was
abolished in Lkb1 KO MEFs (fig. S7, E and F).
The phosphorylation of RIPK1S415 in these
cells paralleled the phosphorylation of a well-
established AMPK substrate, acetyl–coenzyme
A carboxylase (ACC) (Fig. 2, C to G, and fig.
S7, C and D) (35). Phosphorylation quickly
returned to basal levels when cells were re-
stimualted with 25 mM glucose after glucose
deprivation for 4 hours (Fig. 2H and fig.
S7G). We therefore tested whether AMPK
might directly phosphorylate RIPK1. We de-
tected a physical interaction between overex-
pressed Flag-tagged RIPK1 and hemagglutinin
(HA)–tagged AMPK (fig. S7H). Recombinant
AMPK induced phosphorylation of Flag-tagged
RIPK1 at Ser416 or a myelin basic protein fu-
sion with a RIPK1 fragment (390–436) in an
in vitro kinase assay (Fig. 2I and fig. S7I).
Furthermore, we used 32P-labeled ATP in an
in vitro kinase assay (36). The inactive Flag-
tagged RIPK1 K45M mutant or glutathione

Zhang et al., Science 380, 1372–1380 (2023) 30 June 2023 3 of 9

RES EARCH | R E S E A R C H A R T I C L E
Fig. 2. AMPK phosphorylates RIPK1 at S416 in response to metabolic
stress. (A) WT or Ripk1D138N/D138N MEFs were cultured in glucose-free
medium for 12 hours. Cell lysates were then subjected to immunoblotting
using anti-RIPK1 antibody. n = 3 independent biological repeats. (B) U2OS
were cultured in glucose-free medium for 4 hours. Cell lysates were treated
with lambda-phosphatase (lPPase), as indicated, for 30 min and then
subjected to immunoblotting using anti-RIPK1 antibody. (C) WT and Ampk
DKO MEFs were cultured in glucose-free medium for the indicated times.
The levels of proteins were determined by immunoblotting. n = 3 independent
biological repeats. (D) WT and AMPK DKO HT-29 cells were cultured in
glucose-free medium for 12 hours. The levels of proteins were determined
by immunoblotting. n = 3 independent biological repeats. (E) WT and Ampk
DKO MEFs were treated with 2 mM AICAR in the presence of glucose for
the indicated times. The levels of proteins were determined by immuno-
blotting. n = 3 independent biological repeats. (F) MEFs were treated with
50 mM 991 in the presence of glucose for the indicated times. The levels of
Zhang et al., Science 380, 1372–1380 (2023) 30 June 2023
proteins were determined by immunoblotting. n = 3 independent biological
repeats. (G) WT and Ampk DKO MEFs were treated with 2 mM metformin in
the presence of glucose for various times. The levels of proteins were
determined by immunoblotting. n = 3 independent biological repeats.
(H) U2OS were starved of glucose (−Glucose) for 4 hours, and then the
culture was switched to glucose-containing (25 mM) medium for indicated
times (Re-Glucose), and samples were harvested. n = 3 independent
biological repeats. (I) Flag-RIPK1 purified from human embryonic kidney 293T
cells expressing Flag-tagged WT or S416A mutant of RIPK1 for 24 hours
was combined with recombinant (Recomb.) active AMPK as indicated in an
in vitro kinase reaction. The amounts of p-RIPK1 (S416) were determined by
immunoblotting. n = 3 independent biological repeats. (J) Western blot
analysis of p-RIPK1 (S415) in livers from fed and fasted (16 hours) animals
(n = 5). Quantification of p-RIPK1(S415) is shown on the right. Data are
presented as mean ± SEM, and each dot represents one mouse. Unpaired
two-tailed t test; ***P < 0.001.
4 of 9
RES EARCH | R E S E A R C H A R T I C L E

Fig. 3. AMPK deficiency promotes RIPK1-driven cell death and inflammation Two-way ANOVA; ***P < 0.001. (B) WT or Ampk DKO MEFs were cultured in
in vitro and in vivo. (A) WT or Ampk DKO MEFs were cultured in glucose- glucose-free medium for 30 hours. The levels of proteins in mild-detergent
free medium for the indicated times followed by cell viability analyses using PI (NP-40)–soluble fraction and 6 M urea–soluble fraction were determined by
uptake assay. Data are mean ± SD of n = 3 biologically independent samples. immunoblotting. n = 3 independent biological repeats. (C) WT or Ampk DKO

Zhang et al., Science 380, 1372–1380 (2023) 30 June 2023 5 of 9

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

MEFs were cultured in glucose-free medium for 24 hours. Afterward, cell lysates
were immunoprecipitated with anti-RIPK1 antibody, and the immunocomplexes
were analyzed by immunoblotting using anti-RIPK3 antibody. n = 3 independent
biological repeats. Quantified values for Western blot images are shown on
the right. **P < 0.01. (D) Ampka1/a2f/f;Ubc-CreER mice were treated with
or without tamoxifen (4 mg per day, for 5 days) to induce Ampk deletion.
Spleens were harvested 8 weeks after tamoxifen treatment, and then
immunostaining for p-RIPK1(S166) was performed on spleen sections (n = 5
mice in each group). DAPI for nuclei. Representative images are shown. Data are
presented as mean ± SEM, and each dot represents one mouse. Unpaired
two-tailed t test; **P < 0.01. Scale bars, 100 mm. (E) MEFs were cultured
in glucose-free medium for the indicated times. The levels of proteins were
determined by immunoblotting. n = 3 independent biological repeats. Quantified
values for Western blot images are shown at the bottom. (F) Ripk1−/− MEFs
reconstituted with WT or S415A mutant of RIPK1 were cultured in glucose-free
medium for the indicated times followed by cell viability analyses using PI uptake
assay. Data are mean ± SD of n = 3 biological independent samples. Two-
way ANOVA; ***P < 0.001. The expression of RIPK1 was determined by
immunoblotting and is shown at the bottom. (G) Ripk1S415A/S415A and control
WT littermate mice were subjected to liver IR injury. Histological analysis on liver
sections was performed (n = 4 mice in each group). (H) Ripk1S415A/S415A and
control WT littermate mice were subjected to liver IR injury. Immunostaining for
p-RIPK1(S166) on liver sections was performed (n = 4 mice in each group).
DAPI for nuclei. Representative images are shown. Microscopic quantification
of p-RIPK1(S166)–positive cells is shown at the bottom. Data are presented as
mean ± SEM, and each dot represents one mouse. One-way ANOVA, post hoc
Dunnett’s test; ***P < 0.001. (I) TUNEL assays were performed on liver sections
from Ripk1S415A/S415A and control WT littermate mice subjected to liver IR
injury (n = 4 mice in each group). DAPI for nuclei. Representative images
are shown. Microscopic quantification of TUNEL-positive cells was shown on the
right. Data are presented as mean ± SEM, and each dot represents one mouse.
One-way ANOVA, post hoc Dunnett’s test; ***P < 0.001. (J) ELISA analyses
of the concentrations of indicated cytokines and chemokines in serum from
Ripk1S415A/S415A and control WT littermate mice subjected to liver IR injury (n = 4
mice each group). Data are presented as mean ± SEM, and each dot represents one
mouse. One-way ANOVA, post hoc Dunnett’s test; **P < 0.01, ***P < 0.001.

S-transferase (GST) fusion of RIPK1 fragment
(390–436) was phosphorylated by AMPK (fig.
S7, J and K). The stoichiometry of the phos-
phorylation of RIPK1 at Ser416 by AMPK was
estimated to be 0.51 mol phosphate per mol of
RIPK1 fragment (390–436). AMPK-mediated
incorporation of 32P into RIPK1 was not ob-
served for the RIPK1 fragment containing the
S416A mutation (fig. S7K), indicating that
AMPK predominantly phosphorylates RIPK1
at Ser416. Furthermore, we detected RIPK1
p-S415 in mouse tissues including lung and
brain (fig. S7L). Moreover, in mice subjected
to fasting, which causes low glucose in vivo
(fig. S7M) (37), RIPK1 p-S415 was increased in
liver and pancreas, two organs that are sen-
sitive to fasting, a condition in which AMPK
is activated (Fig. 2J and fig. S7N).
Ampk deficiency promotes RIPK1-driven cell
death and inflammation
We tested whether phosphorylation of RIPK1
by AMPK represents a survival mechanism in
response to glucose deprivation. Ampk defi-
ciency sensitized cells to glucose deprivation–
induced cell death (Fig. 3A and fig. S8A).
Moreover, genetic deletion of Ampk in MEFs
promoted activation of RIPK1 (Fig. 3B). Ac-
cordingly, glucose deprivation induced greater
association of RIPK3 with immunoprecipi-
tated RIPK1 in Ampk DKO MEFs than in WT
MEFs, indicating that Ampk deficiency pro-
motes RIPK1-driven necroptosis (Fig. 3C).
Similar results were obtained in AMPK DKO
HT-29 cells (fig. S8, B to E). Because the ac-
tivation of RIPK1 by glucose deprivation in-
duces both necroptosis and apoptosis, and
RIPK1-induced apoptosis is caspase-8 depen-
dent (38), we explored the contribution of
necroptosis and apoptosis to the glucose
deprivation–induced cell death in Ampk DKO
MEFs. Cell death induced by glucose depri-
vation in Ampk DKO MEFs was reduced by
10 mM pan-caspase inhibitor quinoline-Val-
Asp-difluorophenoxymethylketone (QVD-oph)
or 3 mM RIPK3 inhibitor GSK’872 (fig. S8, F
and G).
To determine the role of AMPK in the reg-
ulation of RIPK1 activity in vivo, we measured
RIPK1 activation by p-S166 RIPK1 immuno-
staining in a diverse set of tissues from
Ampka1a2 f/f ;Ubc-Cre ERT2 mice that were
treated with tamoxifen to induce the deletion
of both Ampka1 and Ampka2. Loss of Ampka1
and Ampka2 induced RIPK1 activation and
increased cell death in multiple tissues, in-
cluding spleen, kidney, and intestine (Fig. 3D
and fig. S9, A to G). We did not observe in-
creased RIPK1 activity in the livers of Ampka1
and Ampka2–deficient mice (fig. S9, H and I).
Thus, there may be tissue-specific activation of
RIPK1 in mice lacking Ampka1 and Ampka2.
Notably, prolonged glucose deprivation or
AICAR activation of AMPK overcame the
adaptive response and triggered RIPK1 activa-
tion and cell death (Fig. 3E and fig. S10A). We
also treated MEFs with 2-DG for various lengths
of time. Short-term treatment of cells with 2-DG
activated AMPK (39) and increased phospho-
rylation of RIPK1 at Ser415 (fig. S10B). However,
phosphorylation of RIPK1 at Ser415 decreased
after prolonged treatment of cells with 2-DG,
which caused RIPK1 activation, as determined
by p-RIPK1 (S166) (fig. S10B). Thus, modified
AMPK-mediated phosphorylation of RIPK1 may
contribute to the switch from adaptive homeo-
stasis to cellular demise in cells exposed to
prolonged metabolic stress.
Consistent with the notion that AMPK-
mediated phosphorylation of RIPK1 directly
inhibits its activity, overexpression of RIPK1
S416A mutant led to higher amounts of p-S166
RIPK1 (fig. S11A). We reintroduced WT and
S415A mutant of RIPK1 into Ripk1 KO MEFs.
Ripk1 KO MEFs reconstituted with the RIPK1
S415A mutant were more sensitive to cell
death induced by glucose deprivation than
were the cells reconstituted with WT RIPK1
(Fig. 3F and fig. S11B). Treatment of cells with
either caspase inhibitor QVD-oph or RIPK3
inhibitor GSK’872 reduced cell death (fig. S11,
C and D). To further investigate the physio-
logical importance of AMPK-mediated phos-
phorylation of RIPK1 in vivo, we generated
Ripk1S415A/S415A knock-in mice by the CRISPR-Cas9
technology (fig. S11, E and F). Ripk1S415A/S415A
mutant mice were born in normal Mendelian
ratios, and their growth appeared normal (fig.
S11, G and H). However, the RIPK1 S415A mu-
tation increased cell death and inflammation
(Fig. 3, G to J, and fig. S11I). These results in-
dicate that RIPK1 S415 phosphorylation inhib-
its RIPK1 activity and blocks cell death in
response to energy stress.
We also investigated whether AMPK has a
role in restraining RIPK1 activity in the con-
text of TNFa-induced cell death. Ampk defi-
ciency had no effect on RIPK1 activation and
inflammation induced by TNFa alone or TNFa-
SM-164 (TS) or TNFa-SM-164-zVAD.fmk (TSZ)
(fig. S12). Moreover, the RIPK1 S415A mutation
had no effect on RIPK1 activation, cell death,
or inflammation induced by TNFa in combi-
nation with SM-164 or zVAD.fmk (fig. S13).
Thus AMPK-mediated RIPK1 phosphorylation
appears not to affect TNFa-induced cell death
and inflammation.
Inhibition of RIPK1 protects against liver
ischemia-reperfusion injury in myeloid
Ampk KO mice
AMPK exerts anti-inflammatory effects, espe-
cially in macrophages, as metabolic activity
and inflammatory status are directly linked
in these cells (40). We conditionally deleted
Ampka1, which is the only catalytic subunit

Zhang et al., Science 380, 1372–1380 (2023) 30 June 2023 6 of 9

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

Zhang et al., Science 380, 1372–1380 (2023) 30 June 2023 7 of 9

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

Fig. 4. Inhibition of RIPK1 activity protects against liver IR injury in
myeloid Ampk KO mice. (A) Schematic of liver IR injury involving 60 min of
global ischemia followed by an 18-hour reperfusion period. (B) Serum ALT and
AST levels were measured from control (n = 9), Ampka1f/f;LysM Cre (n = 9), and
Ampka1f/f;LysMCre;Ripk1D138N/D138N (n = 9) mice subjected to liver IR injury.
Data are presented as mean ± SEM, and each dot represents one mouse. One-
way ANOVA, post hoc Dunnett’s test; ***P < 0.001. (C) Histological analyses
were performed on liver sections of Ampka1f/f;LysM Cre, Ampka1f/f;LysM
Cre;Ripk1D138N/D138N, and control littermate mice subjected to liver IR injury (n = 4).
Representative images are shown. Scale bars, 100 mm. (D) Immunostaining
for p-RIPK1(S166) was performed on liver sections of Ampka1f/f;LysM Cre,
Ampka1f/f;LysM Cre;Ripk1D138N/D138N, and control littermate mice subjected to
liver IR injury (n = 4). DAPI for nuclei. Representative images are shown. Scale
bars, 100 mm. (E) Quantification of p-RIPK1(S166)–positive cells on liver
sections from (D). Data are presented as mean ± SEM, and each dot represents
one mouse. One-way ANOVA, post hoc Dunnett’s test; ***P < 0.001.
(F) TUNEL assays were performed on liver sections of Ampka1f/f;LysM Cre,
Ampka1f/f;LysM Cre;Ripk1D138N/D138N, and control littermate mice subjected to liver
IR injury (n = 4). DAPI for nuclei. Representative images were shown. Scale bars,
100 mm. (G) Quantification of TUNEL-positive cells on liver sections from (F). Data
are presented as mean ± SEM, and each dot represents one mouse. One-way
ANOVA, post hoc Dunnett’s test; ***P < 0.001. (H) Quantitative reverse
transcription polymerase chain reaction analysis of the mRNA expression of
cytokines and chemokines in livers from control (n = 4), Ampka1f/f;LysM Cre
(n = 4), and Ampka1f/f;LysM Cre;Ripk1D138N/D138N (n = 4) mice subjected to liver IR
injury. Data are presented as mean ± SEM, and each dot represents one mouse. One-
way ANOVA, post hoc Dunnett’s test; **P < 0.01, ***P < 0.001. (I) A schematic model
to illustrate a delicate and temporal cellular response of RIPK1 to metabolic stress:
Cells activate AMPK to suppress RIPK1 activation, allowing survival under energy stress
in the short term, whereas over longer time periods, as the AMPK phosphorylation of
RIPK1 is lost, the inhibition is relieved, promoting a switch to activated RIPK1-mediated
cell death and inflammation. Moreover, long-term glucose starvation causes the
ATF4/CHOP-dependent up-regulation and activation of TRAIL receptors DR4 and
DR5, which promotes RIPK1 activation, cell death, and inflammation.

isoform expressed in macrophages (41), in mye-
loid cells using Ampka1 fl/fl mice expressing
LysM-Cre and then crossed these mice with
Ripk1D138N/D138N mice to generate Ampka1f/f;
LysM-Cre;Ripk1 D138N/D138N mice. Ampka1
expression was blocked in bone marrow–
derived macrophages (BMDMs) isolated from
Ampka1fl/fl;LysM-Cre and Ampka1f/f;LysM-
Cre;Ripk1D138N/D138N mice (fig. S14A). Ampka1fl/fl;
LysM-Cre mice showed normal liver and body
weight, normal amounts of serum alanine
aminotransferase (ALT) and aspartate amino-
transferase (AST), and normal liver morphology
(fig. S14, B to E). Similar to MEFs, Ampk-
deficient BMDMs showed increased RIPK1
activation when deprived of glucose (fig. S14,
F to H).
Next we subjected these mice to liver
ischemia-reperfusion (IR) injury (Fig. 4A).
Ampka1fl/fl;LysM-Cre mice showed increased
abnormal ALT and AST in serum compared
with WT mice (Fig. 4B). Moreover, Ampka1fl/fl;
LysM-Cre mice showed increased infiltration
of inflammatory cells, which was reduced in
Ampka1 fl/fl;LysM-Cre;Ripk1D138N/D138N mice
(Fig. 4C and fig. S15, A and B). We also de-
tected increased numbers of p-S166 RIPK1+
and TUNEL+ cells as well as increased ex-
pression of proinflammatory cytokines in the
livers of Ampka1fl/fl;LysM-Cre mice compared
with those of WT mice (Fig. 4, D to H). We
immunostained liver tissues after IR injury
with a phosphospecific RIPK3 antibody vali-
dated in previous studies (42). Although we
detected a substantial amount of TUNEL+
cells in livers of Ampka1 fl/fl;LysM-Cre mice
after hepatic IR injury, none of them showed
phosphorylation of T231/S232 of RIPK3, in-
dicating the absence of necroptosis in IR-
injured livers (fig. S15C). These data indicate
that RIPK1-dependent apoptosis, more so
than necroptosis, mediated cell death and
inflammation in response to ischemic stress
(fig. S15).
Next we transplanted bone marrow from
mice in which Ampk was deleted in myeloid
cells with or without Ripk1D138N/D138N into
WT, Ripk1D138N/D138N, and Ripk1S415A/S415A mice
(fig. S16A) (43). Lethally irradiated WT recipient
mice reconstituted with Ampka1f/f;LysM-Cre
bone marrow cells showed increased con-
centrations of ALT and AST in serum com-
pared with WT recipient mice reconstituted
with Ampka1f/f bone marrow cells in response
to liver IR injury (fig. S16B). Thus, Ampka1
deficiency in myeloid cells appears to pro-
mote liver damage after hepatic IR injury in
a non–cell-autonomous manner. Consistently,
Ripk1D138N/D138N recipient mice reconstituted
with Ampka1fl/fl;LysM-Cre bone marrow cells
showed decreased concentrations of ALT
and AST in serum compared with WT reci-
pient mice reconstituted with Ampka1 fl/fl;
LysM-Cre bone marrow cells (fig. S16B). Thus,
Ampka1 deficiency in myeloid cells facilitates
RIPK1 activation by IR injury, which promotes
liver proinflammatory immune activation and
contributes to the development of hepatocel-
lular damage.
To further link the effects of AMPK on cell death
and inflammation to RIPK1 phosphorylation
at Ser415, we crossed the Ampka1 fl/fl;LysM-
Cre mice into the Ripk1S415A/S415A background.
Ampka1 fl/fl;LysM-Cre;Ripk1S415A/S415A mice
did not show additional increase in IR injury–
induced RIPK1 activation, cell death, or inflam-
mation as compared with that of Ripk1S415A/S415A
mice (fig. S16, C to E). Thus, the effects of AMPK
on cell death and inflammation appear to be
predominantly through AMPK-mediated RIPK1
Ser415 phosphorylation in the liver IR injury
model we used.
Discussion
Our results indicate that phosphorylation-
regulated control of RIPK1 activity enables
cellular metabolic control of cell death and
inflammation in response to metabolic stress
(Fig. 4I). In parallel with other substrates of
AMPK, AMPK directly inhibits RIPK1 by phos-
phorylation, which in turn suppresses energy
stress–induced cell death and inflammation
(fig. S16F). We revealed a delicate and tempo-
ral cellular response of RIPK1 to metabolic
stress: Cells activate AMPK to suppress RIPK1
activation, allowing survival in the short term
under energy stress, whereas over the longer
term, as AMPK phosphorylation of RIPK1 is
lost, the inhibition is relieved, promoting a
switch to activated RIPK1–mediated cell death
and inflammation. The work presented here
expands our understanding of the interplay
between metabolism and cell death regulation,
which may help inform the development of
therapeutic drugs aimed at preventing ische-
mia-induced cell death and tissue damage.
REFERENCES AND NOTES
1. D. G. Hardie, Nat. Rev. Mol. Cell Biol. 8, 774–785 (2007).
2. D. G. Hardie, F. A. Ross, S. A. Hawley, Nat. Rev. Mol. Cell Biol.
13, 251–262 (2012).
3. A. González, M. N. Hall, S. C. Lin, D. G. Hardie, Cell Metab. 31,
472–492 (2020).
4. S. C. Lin, D. G. Hardie, Cell Metab. 27, 299–313 (2018).
5. D. R. Green, L. Galluzzi, G. Kroemer, Science 345, 1250256 (2014).
6. R. J. Shaw et al., Cancer Cell 6, 91–99 (2004).
7. B. Tummers et al., Immunity 52, 994–1006.e8 (2020).
8. J. Yuan, P. Amin, D. Ofengeim, Nat. Rev. Neurosci. 20, 19–33
(2019).
9. D. Xu et al., Cell 174, 1477–1491.e19 (2018).
10. L. Sun et al., Cell 148, 213–227 (2012).
11. H. Wang et al., Mol. Cell 54, 133–146 (2014).
12. D. R. Green, Cell 177, 1094–1107 (2019).
13. G. Pan et al., Science 276, 111–113 (1997).
14. G. Pan et al., Science 277, 815–818 (1997).
15. S. M. Kavuri et al., J. Biol. Chem. 286, 16631–16646 (2011).
16. S. P. Cullen et al., Mol. Cell 49, 1034–1048 (2013).
17. T. Hartwig et al., Mol. Cell 65, 730–742.e5 (2017).
18. C. M. Henry, S. J. Martin, Mol. Cell 65, 715–729.e5 (2017).
19. A. Degterev et al., Nat. Chem. Biol. 1, 112–119 (2005).
20. A. Linkermann et al., Proc. Natl. Acad. Sci. U.S.A. 110,
12024–12029 (2013).
21. A. Degterev et al., Nat. Chem. Biol. 4, 313–321 (2008).
22. D. Ofengeim et al., Cell Rep. 10, 1836–1849 (2015).
23. C. P. Dillon et al., Cell 157, 1189–1202 (2014).
24. D. R. Green, A. Oberst, C. P. Dillon, R. Weinlich, G. S. Salvesen,
Mol. Cell 44, 9–16 (2011).
25. R. Weinlich, D. R. Green, Mol. Cell 56, 469–480 (2014).
26. R. Iurlaro et al., Mol. Cell. Biol. 37, e00479-16 (2017).

Zhang et al., Science 380, 1372–1380 (2023) 30 June 2023 8 of 9

RES EARCH | R E S E A R C H A R T I C L E
27. P. M. Chaudhary et al., Immunity 7, 821–830 (1997).
28. J. Kim, M. Kundu, B. Viollet, K.-L. Guan, Nat. Cell Biol. 13,
132–141 (2011).
29. H. Meng et al., Proc. Natl. Acad. Sci. U.S.A. 115, E2001–E2009
(2018).
30. D. G. Hardie, B. E. Schaffer, A. Brunet, Trends Cell Biol. 26,
190–201 (2016).
31. B. E. Crute, K. Seefeld, J. Gamble, B. E. Kemp, L. A. Witters,
J. Biol. Chem. 273, 35347–35354 (1998).
32. B. D. Stein et al., Cell Rep. 29, 3331–3348.e7 (2019).
33. J. J. Howell et al., Cell Metab. 25, 463–471 (2017).
34. R. J. Shaw et al., Proc. Natl. Acad. Sci. U.S.A. 101, 3329–3335
(2004).
35. G. J. Gowans, S. A. Hawley, F. A. Ross, D. G. Hardie, Cell Metab.
18, 556–566 (2013).
36. S. A. Hawley et al., Biochem. J. 459, 275–287 (2014).
37. C. S. Zhang et al., Nature 548, 112–116 (2017).
38. K. Newton et al., Nature 574, 428–431 (2019).
39. S. A. Hawley et al., Cell Metab. 11, 554–565 (2010).
40. B. Kelly, L. A. O’Neill, Cell Res. 25, 771–784 (2015).
41. R. Mounier et al., Cell Metab. 18, 251–264 (2013).
42. K. Heger et al., Nature 559, 120–124 (2018).
43. A. Dey et al., Science 337, 1541–1546 (2012).
ACKN OW LEDG MEN TS
We thank M. Kelliher (University of Massachusetts) and
M. Pasparakis (University of Cologne, Germany) for providing
Ripk1D138N/D138N mice and V. Dixit (Genentech) for RIPK3 KO mice
and the antibodies recognizing phosphorylated RIPK3 Thr231/
Ser232 (GEN135-35-9). Funding: This work was supported in
Zhang et al., Science 380, 1372–1380 (2023) 30 June 2023
part by NIH grants [CA177910 and R35CA253027 (W.W.); R00
CA194314 (C.C.D.); RO1 AA017729m and R01 AA011576 (G.Sz.);
CA236226, CA239660, CA202634, P01 CA250959, and P01
CA120964 (P.S.); and RO1DK124328 and RO1DK133331 (J.M.)],
the National Key R&D Program of China (2022YFC2303102 to L.P.;
2022ZD0213200 to D.X.), and the National Natural Science
Foundation of China (92253301 to L.P.; 32070737 to D.X.). J.Y.
was supported in part by the China National Natural Science
Foundation (82188101, 21837004, 91849204, and 92049303).
D.X. and J.Y. were supported in part by the Shanghai Municipal
Science and Technology Major Project (grant no. 2019SHZDZX02),
the Strategic Priority Research Program of the Chinese Academy
of Sciences (XDB39030200 to J.Y.; XDB39030600 to D.X.).
Y.-R.L. was supported in part by a Career Development Award,
Academia Sinica Taiwan (AS-CDA-110-L07), and the Ministry of
Science and Technology Taiwan (110-2320-B-001-029-MY2).
Author contributions: T.Z. was the key contributor in designing
and conducting most of the experiments. T.Z., D.X., L.P., R.J.S.,
J.Y., and W.W. conceived of and directed the project. T.Z., D.X.,
J.Y., and W.W. wrote the manuscript. D.X., E.T., and M.Lv.
conducted key experiments. H.I., G.So., M.Li., J.Lu, Jia.Liu, C.C.,
M.W., H.W., H.M., H.L., Y.Z., X.X, F.D., D.G., Y.M., S.J., C.J., X.D.,
Jin.Liu, Z.W., P.Y., J.W., Z.T., M.B., E.E., A.L.H., and J.M.A.
conducted some of the experiments. C.C.D., G.Sz., P.S., J.M., and
Y.-R.L. commented on and edited the manuscript. Competing
interests: W.W. is a cofounder and consultant for ReKindle
Therapeutics. P.S. has been a consultant at Novartis, Genovis,
Guidepoint, The Planning Shop, ORIC Pharmaceuticals, Cedilla
Therapeutics, Syros Pharmaceuticals, Exo Therapeutics, Curie Bio
Operations, Exscientia, Ligature Therapeutics, and Redesign
Science; and his laboratory receives research funding from
Novartis. G.Sz. is a paid consultant for Cyta Therapeutics, Durect
Co, Evive, Merck, Pfizer, Surrozen, Terra Firma, Pandion
Therapeutics, LabCorp, Glympse Bio, Satellite Bio, and Zomagen.
G.Sz. has additional financial interests in Glympse Bio, Satellite Bio,
and Zomagen. The other authors declare that they have no
competing financial interests. Data and materials availability:
RNA-seq data used to support the present study have been
deposited in the Gene Expression Omnibus with an access number
of GSE173938. The biological material is available upon request to
W.W. All data are available in the main text or the supplementary
materials. License information: Copyright © 2023 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.abn1725
Materials and Methods
Figs. S1 to S16
Table S1
References (44–50)
MDAR Reproducibility Checklist
View/request a protocol for this paper from Bio-protocol.
Submitted 8 November 2021; resubmitted 9 December 2022
Accepted 4 May 2023
10.1126/science.abn1725
9 of 9
RES EARCH

QUANTUM SENSING adding a global rotation term S^ x to the one-

axis-twisting (OAT) (35) Hamiltonian S^z2
Improving metrology with quantum scrambling ^ ¼ cS^ 2 þ WS^ x
H ð1Þ
z
1 1 1,2 1,3 4
Zeyang Li ( ) , Simone Colombo , Chi Shu , Gustavo Velez , Saúl Pilatowsky-Cameo ,

Roman Schmied5, Soonwon Choi4, Mikhail Lukin2, Edwin Pedrozo-Peñafiel1, Vladan Vuletić1* Here S ¼ S^ x ; S^ y ; S^ z represents the total
spin of the system consisting of N ¼ 2S spin-21
Quantum scrambling describes the spreading of information into many degrees of freedom in particles, c is the shearing parameter for OAT,
quantum systems, such that the information is no longer accessible locally but becomes distributed and W is the transverse rotation frequency.
throughout the system. This idea can explain how quantum systems become classical and acquire a Although the time evolution is not chaotic
finite temperature, or how in black holes the information about the matter falling in is seemingly erased. because of the conservation of S ^ 2 , the LMG
We probe the exponential scrambling of a multiparticle system near a bistable point in phase space Hamiltonian nevertheless features a quantum
and utilize it for entanglement-enhanced metrology. A time-reversal protocol is used to observe Lyapunov exponent for 0 < W=ðScÞ < 2 due
a simultaneous exponential growth of both the metrological gain and the out-of-time-order correlator, to an unstable (bifurcating) trajectory in the
thereby experimentally verifying the relation between quantum metrology and quantum information system phase space (Fig. 1A) (32, 36, 37).
scrambling. Our results show that rapid scrambling dynamics capable of exponentially fast Our experiments operate with N ¼ 200 171 Yb
entanglement generation are useful for practical metrology, resulting in a 6.8(4)-decibel gain beyond atoms whose magnetic sublevels j↑; ↓i in the
the standard quantum limit. electronic ground state represent a spin- 21 sys-
tem. One of the two spin states (j↑i) couples to
an electronically excited state jei via sþ-polarized

E
ven though all unitary dynamics of quan- tum metrology, this enables a family of powerful light that circulates inside the optical cavity
tum systems are in principle reversible, quantum amplification protocols (17–24) such (Fig. 1B). The coupling between a single atom
it is extremely challenging in practice to as signal amplification through time-reversed and the cavity is characterized by the single-
reverse the arrow of time in generic in- interaction (SATIN) (23). Such protocols can atom cooperativity h ¼ 8:8ð2Þ (38). We imple-
teracting many-body systems. This is be- be robust against many limitations that usual- ment the LMG Hamiltonian in the rotating
cause any small perturbations or imperfections ly affect entanglement-enhanced atomic sen- frame by adding an oscillating transverse mag-
in the time-reversed dynamics can lead to sors, including imperfect measurements. In the netic field to the OAT Hamiltonian (34) [Fig.
highly complicated, nonlocal changes in quan- presence of exponential scrambling dynamics 1B and supplementary materials (SM) (39)].
tum wave functions, similar to the butterfly (Fig. 1A), the SATIN signal is also amplified The experiments start by initializing the
effect in chaos theory. Dubbed information exponentially over time. system in a coherent spin state (CSS) pointing
scrambling (1–3), this quantum-mechanical along the x axis by means of optical pumping
effect gives rise to a variety of unusual pheno- Experimental setup followed by a p=2 spin rotation. Analytical
mena and applications ranging from models The Lipkin-Meshkov-Glick (LMG) Hamiltonian solutions using the Holstein-Primakoff approx-
of quantum gravity (4, 5) to quantum metrol- (24, 25–34) is of particular interest in this con- imation (40) show that for W=ðScÞ < 0 or
ogy (6). The speed of information scrambling text, as it exhibits exponential evolution in W=ðScÞ > 2, the system evolution is periodic
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
can be quantified by out-of-time-ordered cor- phase space while it can be experimentally with a frequency w ¼ W2
2ScW (41). How-
relators (OTOCs) (7, 8), which constitute a realized in a cavity quantum electrodynamics ever, for 0 < W=ðScÞ < 2 the frequency w
measure of how fast the noncommutativity (cQED) system. The latter is accomplished by becomes imaginary, corresponding to an
between two different quantum operations
is established (9). In certain systems, the
OTOC grows exponentially fast over time
elQ t , where lQ > 0 defines the generalized
quantum Lyapunov exponent (8). OTOCs have
been measured (10) and used as probes for
various many-body phenomena, such as ther-
malization (11), quantum phase transitions
(12), many-body entanglement growth (13), and
quantum scrambling (14–16). However, the
observation of exponential scrambling has
remained elusive.
One approach to effective time reversal in-
volves changing the sign of the Hamiltonian
H^→
H ^ during the evolution of highly engi-
neered quantum systems. In the field of quan-
Fig. 1. Time-reversal–based exponential growth of sensitivity in a system with an unstable fixed point.
(A) Classically, for a trajectory with a positive Lyapunov exponent l1 > 0, an initial signal (displacement) dð0Þ
1
increases exponentially over time. For quantum dynamics, however, an initial overlap between two states is
Department of Physics, MIT-Harvard Center for Ultracold
Atoms, Research Laboratory of Electronics, Massachusetts
preserved under unitary evolution. To amplify the signal similarly to the classical case, one needs to evolve the
Institute of Technology, Cambridge, MA 02139, USA. state under the nonlinear H,^ resulting in decreased quantum fluctuations along a direction with negative Lyapunov
2
Department of Physics, Harvard University, Cambridge, MA ^
02138, USA. 3Department of Electrical Engineering and coefficient l2 < 0. A displacement along this direction followed by application of the negative Hamiltonian
H
Computer Science, Massachusetts Institute of Technology, (such that l1;2 →
l1;2 ) is then used to amplify the signal. The plots represent the evolution on the Bloch
Cambridge, MA 02139, USA. 4Department of Physics, sphere, where the Sy and Sz axes are labeled as in the left subplot of the middle row. (B) Experimental setup. The
Massachusetts Institute of Technology, Cambridge, MA
02139, USA. 5Viewpointsystem GmbH, 1010 Wien, Austria. LMG Hamiltonian is generated by the interaction of the collective atomic spin with light inside a cavity on the
*Corresponding author. Email: vuletic@mit.edu transition j↑i→jei, while a radiofrequency magnetic field is applied to rotate the atomic spin.

Li et al., Science 380, 1381–1384 (2023) 30 June 2023 1 of 4

RES EARCH | R E S E A R C H A R T I C L E
unstable-fixed-point exponential evolution with
a Lyapunov exponent lQ ¼ jwj. For a fixed
Sc, choosing W ¼ Sc results in a maximum
Lyapunov exponent lQ ¼ jScj.
Metrological gain
We first measure the antisqueezing {largest
variance x2þ ≡ maxa ½varðSa Þ=ðS=2Þ} of the col-
lective spin S after an evolution under H
a function of the ratio W=ðScÞ . The anti-
squeezing xþ constitutes an upper bound on
the quantum Fisher information (QFI) with
respect to spin rotations (42). As shown in
Fig. 2B, the experimental data for x2þ agree
with the numerical simulation of the model
(solid red line) and show a peak at W ¼ Sc,
as expected.
We then measure in Fig. 2C how x2þ grows
with time for the two cases W ¼ 0 (OAT Hamil-
tonian) and W ¼ Sc (critically tuned LMG
Hamiltonian). The OAT data (gray) exhibit
quadratic growth of x2þ , as expected. The
LMG data (red) show exponential growth of
x2þ ¼ e2lQ t , with lQ ¼ Sc for times t ≲ ðScÞ
1.
For larger times, the growths slows because
of finite particle number and light-induced
decoherence (34, 39). The finite total spin fur-
ther causes the states to turn non-Gaussian,
which we characterize via the Binder cumu-
lant (43) (Fig. 2D).
The time evolution under the critically tuned
(W ¼ Sc) LMG Hamiltonian H ^ quickly pre-
pares an entangled collective quantum state.
To implement quantum metrology with the
SATIN protocol, we then apply a small rotation
U^ df ¼ e
iS^ adf, where S^ a ≡ S^ y cosa þ S^ z sina
represents a collective spin operator in the yz
plane. This encodes a signal phase df along
the a direction, with a ¼ p=4 chosen to max-
imize the metrological gain [Fig. 1A and SM
(39)]. To implement
H, ^ we switch to another
set of laser frequencies incident on the cavity
and flip the sign of the transverse field W [SM
(39)]. This generates an effective backward
evolution in time that amplifies the applied
signal df. The shifted state then undergoes a
bifurcated trajectory for df ≶ 0 (see Fig. 2C),
resulting in an exponentially amplified devia-
tion Gdf from the original position. We mea-
sure the mean spin value hS^ a i ¼ SsinðGdfÞ to
infer the signal amplification G. As shown in
Fig. 3, the squared signal amplification G 2
(orange) increases exponentially with the same
exponent 2lQ as the antisqueezing x2þ up to
times t ≈ ðScÞ
1 . The measured quantum noise
N 2 , i.e., the variance of spin projection noise
along the amplification direction S^ a normalized
to the standard quantum limit (SQL) (blue),
remains unity until t ≈ 0:8ðScÞ
1 . The subse-
quent increase of the noise N 2 results from
the residual light-atom entanglement (34) and
can be improved in the future by optimizing
the light detuning (39). The improvement of
the metrological gain over the SQL is 6.8(4) dB.
Li et al., Science 380, 1381–1384 (2023) 30 June 2023
^ as
Fig. 2. Collective-spin evolution in the cQED system. (A) Numerical calculation of the normalized variance x2þ
of the antisqueezed direction as a function of Sct and W=Sc with linecuts representing the measurements in
(B) and (C). (B) Experimentally measured antisqueezing (blue symbols) and theoretical expectation (red line) for
Sct ¼ 1:8 as a function of the rotation strength W. The shaded region indicates exponential growth, whereas
in the other regions the time evolution is either quasi-periodic or growing polynomially. (C) Comparison of
antisqueezing x2þ between the fastest exponential growth for a critical rotation strength W ¼ Sc, and the
polynomial growth of pure OAT (W ¼ 0). The two Bloch spheres represent the lines of classical evolution in
both situations. The dashed and dash-dotted red lines represent exponential growth based on the theoretically
predicted Lyapunov exponent and the full numerical result, respectively, with no free parameters. The gray dashed
line is calculated for W ¼ 0. (Inset) Logarithmic plot for W ¼ Sc showing the exponential growth of x2þ .
(D) The Binder cumulant B (characterizing the deviation from a Gaussian distribution for B > 0) for the
antisqueezed direction for the critical LMG condition W ¼ Sc versus time t. (Insets) Measured spin distributions
for Sct ¼ 0 (blue) and Sct ¼ 2 (purple), with the latter being strongly non-Gaussian.
Fig. 3. Metrological gain with
exponential LMG time-reversal
protocol. The squared signal
amplification G2 (pink open circles)
and system noise N2 (blue solid
squares) versus time t. The pink
dashed line represents the exponen-
tial growth of the antisqueezing shown
in Fig. 2, and constitutes an upper
bound to the QFI. The blue dash-dotted
line is the calculated noise (with no
free parameters) due to residual light-
atom entanglement. The maximum
metrological gain is 6:8ð4Þ dB.
The deviation of G 2 from an exponential for surements (39, 46, 47). The FOTOC F ðt Þ can
t ≳ ðScÞ
1 is due to the nonuniform coupling be expressed as the trace between the den-
between atoms and the cavity light (44), as sity matrix rð0Þ of the original state and that
well as the residual light-atom entanglement, of the state displaced by df evolved back-
both of which can be improved in the future ward in time, r′ t ð0Þ :¼ U ^ t^
r ð0ÞU^ †, where
t
^
iS^ adf
iHt
^ t :¼ eiHt ^
(34, 45). U e e
D E
′ 
F ðt Þ ≡ U ^ t^ ^ †^
r ð0ÞU t r ð0Þ ¼ Tr r t ð0Þrð0Þ
Quantum metrology and quantum information
To investigate the quantum information sci- ð2Þ
ence aspect of the time-reversal protocol, we
measure the fidelity OTOC (FOTOC) with quan- At fixed forward evolution time t, the FOTOC
tum state tomography using randomized mea- F depends on the small displacement df and
2 of 4
RES EARCH | R E S E A R C H A R T I C L E
Fig. 4. FOTOC and OTOC extracted from quantum state tomography. (A) Experimental Wigner functions
obtained from quantum state tomography after applying the LMG SATIN protocol with different signal
displacements df [for W ¼ Sc and t ¼ 0:57ðScÞ
1 ]. The dashed circle indicates the original CSS state. (B) The
solid blue line is a quadratic fit to the data (open circles), used to extract the OTOC I (see text and Eq. 3).
Fig. 5. Comparison between quantum
information and quantum metrology
parameters for the LMG model. The
purple open circles, solid green squares,
and solid blue diamonds represent the
antisqueezing, metrological gain, and
OTOC, respectively. All quantities increase
initially exponentially with time with a
fitted Lyapunov exponent lQ ¼ 1:01ð3ÞSc
that agrees well with the theoretical
prediction lQ ¼ Sc. The OTOC error bars
are obtained by using the bootstrapping
method (39, 50). For longer times
t ≳ ðScÞ
1 , the metrological gain and
OTOC decrease owing to decoherence
caused by light-atom entanglement, as is
well captured by the theoretical model
(solid blue line) without free parameters.
is related to the OTOC I ðt Þ by its second de-
rivative (12)
I ðt Þ ≡


1 @ 2 F ðt Þ
2 ð@dfÞ2 df¼0 j
¼ S^ a ðt Þ^r ð0ÞS^ a ðt Þ^
r ð0 Þ
with the Hermitian operator S^ a ðt Þ :¼ eiHt
^
S^ a e
iHt .
^
Choosing four different evolution times
(such that Sct1 ∈ f0:38; 0:57; 0:77; 0:96g), we
displace the entangled state for each t1 by
several different small angles df. We then
perform the tomographic reconstruction after
a reversed time evolution with
H ^ to obtain
F ðt1 Þ, as shown in Fig. 4A. The OTOC I ðt1 Þ is
then extracted from the data by fitting a quad-
ratic function in the displacement df to the
Li et al., Science 380, 1381–1384 (2023)
FOTOC (Fig. 4B). We notice that the fitted
quadratic curve is slightly shifted from df ¼ 0
and has slightly reduced peak fidelity. The

shift is likely due to a small difference between
the assumed and actual Larmor frequencies
ð3Þ between the spin states, whereas the reduc-
tion from unit peak fidelity is due to the im-
perfect time reversal associated with residual
light-atom entanglement. The small imperfec-
tions do not reduce the metrological gain ap-
preciably (39).
Figure 5 summarizes our findings regard-
ing the close relation between quantum scram-
bling and time-reversal quantum metrology:
The antisqueezing x2þ , metrological gain G :¼
G 2 =N 2 , and OTOC I all agree with each other
and scale exponentially with application time
t of the LMG Hamiltonian for t ≲ 0:8ðScÞ
1 .
30 June 2023
The exponential fit yields a Lyapunov exponent
lQ =ðScÞ ¼ 1:01 T 0:03, in excellent agreement
with the theoretical expectation lQ =ðScÞ ¼ 1.
Concluding remarks
Our experiments demonstrate a cQED real-
ization of the critically tuned (W ¼ Sc) LMG
model with an exponential evolution in phase
space. We also point out and experimentally
verify that time-reversal protocols represent a
powerful experimental tool giving access not
only to metrological gain beyond the SQL
(22, 23, 48), but also enabling the measure-
ment of quantum information scrambling in
many-body systems. This generalizes the pos-
sibilities of using quantum information science
to enhance quantum metrology (49). Further-
more, we observe exponential growth of both
the OTOC and the metrological gain for the
LMG model, thereby experimentally verify-
ing the intrinsic relation between these two
concepts from different subfields of quantum
science. The demonstrated methods to re-
verse time evolution may enable the exper-
imental investigation of complex many-body
quantum systems in which the information
spreads exponentially fast within many de-
grees of freedom.
REFERENCES AND NOTES
1. Y. Sekino, L. Susskind, J. High Energy Phys. 2008, 065
(2008).
2. B. Swingle, G. Bentsen, M. Schleier-Smith, P. Hayden,
Phys. Rev. A 94, 040302 (2016).
3. R. J. Lewis-Swan, A. Safavi-Naini, J. J. Bollinger, A. M. Rey,
Nat. Commun. 10, 1581 (2019).
4. T. Schuster et al., Phys. Rev. X 12, 031013 (2022).
5. S. Nezami et al., PRX Quantum 4, 010321 (2023).
6. J. Appel et al., Proc. Natl. Acad. Sci. U.S.A. 106, 10960–10965
(2009).
7. A. Larkin, Y. N. Ovchinnikov, Sov. Phys. JETP 28, 1200
(1969).
8. A. Kitaev, A simple model of quantum holography [video]
(2015). https://online.kitp.ucsb.edu/online/entangled15/
kitaev/. See also J. Maldacena, D. Stanford, Phys. Rev. D 94,
106002 (2016).
9. I. García-Mata, R. A. Jalabert, D. A. Wisniacki, Scholarpedia 18,
55237 (2023).
10. J. Li et al., Phys. Rev. X 7, 031011 (2017).
11. A. M. Green et al., Phys. Rev. Lett. 128, 140601
(2022).
12. K. X. Wei et al., Phys. Rev. Lett. 123, 090605
(2019).
13. M. Gärttner et al., Nat. Phys. 13, 781–786 (2017).
14. K. A. Landsman et al., Nature 567, 61–65 (2019).
15. J. Braumüller et al., Nat. Phys. 18, 172–178 (2022).
16. X. Mi et al., Science 374, 1479–1483 (2021).
17. E. Davis, G. Bentsen, M. Schleier-Smith, Phys. Rev. Lett. 116,
053601 (2016).
18. F. Fröwis, P. Sekatski, W. Dür, Phys. Rev. Lett. 116, 090801
(2016).
19. O. Hosten, R. Krishnakumar, N. J. Engelsen, M. A. Kasevich,
Science 352, 1552–1555 (2016).
20. S. P. Nolan, S. S. Szigeti, S. A. Haine, Phys. Rev. Lett. 119,
193601 (2017).
21. T. Macrì, A. Smerzi, L. Pezzè, Phys. Rev. A 94, 010102
(2016).
22. K. A. Gilmore et al., Science 373, 673–678 (2021).
23. S. Colombo et al., Nat. Phys. 18, 925–930 (2022).
24. T.-W. Mao et al., Quantum enhanced sensing by echoing
spin-nematic squeezing in atomic Bose-Einstein condensate
(2022). arXiv:2212.09124.
25. H. J. Lipkin, N. Meshkov, A. Glick, Nucl. Phys. 62, 188–198
(1965).
3 of 4
RES EARCH | R E S E A R C H A R T I C L E
26. L.-M. Duan, A. Sørensen, J. I. Cirac, P. Zoller, Phys. Rev. Lett.
85, 3991–3994 (2000).
27. T. Zibold, E. Nicklas, C. Gross, M. K. Oberthaler, Phys. Rev. Lett.
105, 204101 (2010).
28. C. D. Hamley, C. Gerving, T. Hoang, E. Bookjans,
M. S. Chapman, Nat. Phys. 8, 305–308 (2012).
29. H. Strobel et al., Science 345, 424–427 (2014).
30. W. Muessel et al., Phys. Rev. A 92, 023603 (2015).
31. J. Peise et al., Nat. Commun. 6, 8984 (2015).
32. S. Pilatowsky-Cameo et al., Phys. Rev. E 101, 010202
(2020).
33. J. A. Muniz et al., Nature 580, 602–607 (2020).
34. Z. Li et al., PRX Quantum 3, 020308 (2022).
35. M. Kitagawa, M. Ueda, Phys. Rev. A 47, 5138–5143
(1993).
36. S. Pappalardi et al., Phys. Rev. B 98, 134303 (2018).
37. E. B. Rozenbaum, L. A. Bunimovich, V. Galitski, Phys. Rev. Lett.
125, 014101 (2020).
38. H. Tanji-Suzuki et al., in Advances in Atomic, Molecular, and
Optical Physics, E. Arimondo, P. R. Bermna, C. C. Lin, Eds.
(Academic Press, 2011), vol. 60, pp. 201–237.
39. See supplementary materials for details.
40. T. Holstein, H. Primakoff, Phys. Rev. 58, 1098–1113
(1940).
41. C. K. Law, H. T. Ng, P. T. Leung, Phys. Rev. A 63, 055601
(2001).
42. L. Pezzè, A. Smerzi, M. K. Oberthaler, R. Schmied, P. Treutlein,
Rev. Mod. Phys. 90, 035005 (2018).
Li et al., Science 380, 1381–1384 (2023) 30 June 2023
43. K. Binder, Z. Phys. B Condens. Matter Quanta 43, 119–140
(1981).
44. J. Hu, W. Chen, Z. Vendeiro, H. Zhang, V. Vuletić, Phys. Rev. A
92, 063816 (2015).
45. B. Wu, G. P. Greve, C. Luo, J. K. Thompson, Site-dependent
selection of atoms for homogeneous atom-cavity coupling.
arXiv:2104.01201 [quant-ph] (2021).
46. R. Schmied, P. Treutlein, New J. Phys. 13, 065019
(2011).
47. A. Elben et al., Nat. Rev. Phys. 5, 9–24 (2023).
48. D. Barberena, S. R. Muleady, J. J. Bollinger, R. J. Lewis-Swan,
A. M. Rey, Fast generation of spin squeezing via resonant spin-
boson coupling. arXiv:2201.10622 [quant-ph] (2022).
49. C. D. Marciniak et al., Nature 603, 604–609 (2022).
50. W. Hardle, J. S. Marron, Ann. Stat. 19, 778 (1991).
51. Z. Li et al., Improving metrology with quantum scrambling, Dryad,
dataset (2023). https://doi.org/10.5061/dryad.zkh1893fj.
ACKN OWLED GMEN TS
We thank J. Thompson, M. Schleier-Smith, B. Braverman, and
A. Adiyatullin for inspiring discussions. We also thank M. Liu for
helping resolve the subtleties of using bootstrap method to
extract statistical uncertainties. Funding: We acknowledge
financial support from ONR (grant N00014-20-1-2428), the
National Science Foundation through the Center for Ultracold
Atoms (grant PHY-1734011) and NSF QLCI (Award OMA -
2016244), DOE QSA (through grant DE-SC0021013 and LBNL
QSA Center), and ARO (grant W911NF1910517). Author
contributions: Z.L., S. Colombo, C.S., G.V., and E.P. contributed
to building the experimental setup and performed the
measurements. Z.L. and S.Colombo analyzed the data, numerically
simulated the experiments, and contributed to theoretical
interpretation of the results. R.S. contributed to developing the
quantum state tomography method. S.P.-C. and S. Choi
contributed to the theoretical interpretation of the results. All
work was supervised by V.V. All authors discussed the results
and contributed to the manuscript. Competing interests:
The authors declare no competing interests. Data and materials
availability: All data needed to evaluate the conclusions in the
paper are present in the figures of the paper and/or the
supplementary materials. The data can be accessed at Dryad (51).
License information: Copyright © 2023 the authors, some
rights reserved; exclusive licensee American Association for the
Advancement of Science. No claim to original US government
works. https://www.sciencemag.org/about/science-licenses-
journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adg9500
Materials and Methods
Figs. S1 and S2
Tables S1 and S2
References (52–58)
Submitted 1 February 2023; accepted 19 May 2023
10.1126/science.adg9500
4 of 4
RES EARCH

CHIRAL MATERIALS We developed an analytical model to under-

stand the assembly of magnetic nanorods in
A magnetic assembly approach to such a chiral field (supplementary materials)
that could predict magnetic nanorod align-
chiral superstructures ment along the local field to form chiral
superstructures (Fig. 1C and fig. S4). For small
Zhiwei Li†, Qingsong Fan, Zuyang Ye, Chaolumen Wu, Zhongxiang Wang, Yadong Yin* magnetic nanorods (107.6 ± 5.2 nm in length
and 13.0 ± 1.7 nm in diameter), there were no
Colloidal assembly into chiral superstructures is usually accomplished with templating or lithographic obvious CD signals in rod dispersion without a
patterning methods that are only applicable to materials with specific compositions and morphologies magnetic field or in a magnetic field along
over narrow size ranges. Here, chiral superstructures can be rapidly formed by magnetically assembling the x axis (Bx), but changing the field direc-
materials of any chemical compositions at all scales, from molecules to nano- and microstructures. We tion from the x axis to the y and z axes cre-
show that a quadrupole field chirality is generated by permanent magnets caused by consistent field ated CD responses (Fig. 1, D and E). We
rotation in space. Applying the chiral field to magnetic nanoparticles produces long-range chiral observed positive and negative CD peaks of
superstructures controlled by field strength at the samples and orientation of the magnets. Transferring similar intensity for y (By) and z fields (Bz),
the chirality to any achiral molecules is enabled by incorporating guest molecules such as metals, respectively, which suggested chiral super-
polymers, oxides, semiconductors, dyes, and fluorophores into the magnetic nanostructures. structures with opposite handedness. The CD
responses were then measured in an aqueous

S
uperstructures of colloidal particles can
assemble with chiral symmetry (1–5).
The driving force for chiral assembly is
that the particles or structures are intrin-
sically chiral or rendered so with ad-
sorbed surface molecules and templates that
create chirality (usually helical structures) or
by lithographic methods (6–8). Chiral super-
structures, especially those made from plas-
monic nanoparticles, have distinctive optical
properties such as circular dichroism (CD)
under circularly polarized light excitation
(9–14) and have the potential for developing
electric and optical sensors and devices (15–20).
Templated assembly and lithography have
been used to create chiral superstructures
for sensing external stimuli through changes
in CD spectra (21, 22). For example, DNA-
templated assembly can transfer the helical
configuration of DNA templates to many nano-
structures (23–26) and be used to monitor
changes in temperature and chemical bind-
ing (27–31).
Controlling the collective orientation of
these chiral structures in either solution or
solid matrices remains challenging for optimiz-
ing chiroptical performance. Additionally, the
existing strategies such as templated assem-
blies and chiral superlattice formation only
work for materials of narrow length scales and
specific chemical compositions or shapes
(32–34). Designing miniature chiroptical de-
vices and understanding light-matter interac-
tions that involve distinct physical principles
would benefit from a general approach for
assembling achiral materials of diverse sizes,
shapes, and chemical compositions into chiral
superstructures with actively tunable chiroptical
responses (35–37).
Department of Chemistry, University of California, Riverside,
CA 92521, USA.
*Corresponding author. Email: yadong.yin@ucr.edu
†Present address: Department of Chemistry and International
Institute for Nanotechnology, Northwestern University, Evanston, IL
60208, USA.
Here, we report the rapid and reversible
assembly of materials of varying compositions
and length scales, from small molecules to
nano- and microstructures, into chiral struc-
tures, and the active tuning of the structural
handedness, collective orientation, and chirop-
tical properties by using the magnetic field of
a permanent magnet. Our analytical model
demonstrates the presence of a quadrupole
field chirality in the gradient magnetic field
of the magnet. Assembling nanorods in such
a magnetic field led to the formation of chiral
superstructures, with their handedness and
chiroptical properties being determined by
magnet position and orientation. The struc-
tural chirality could be transferred to organic
molecules and inorganic compounds by dop-
ing into or coating onto the host magnetic
building blocks, demonstrating a general
approach to chiral superstructures.
The quadrupole field chirality of
permanent magnets
We consider the magnetic field and field dis-
tribution of a cube-shaped permanent mag-
net (fig. S1), with the north pole of the magnet
pointing upward in fig. S1A. The calculated
azimuth of the local magnetic field (mapped in
Fig. 1A) undergoes gradual changes in the field
direction in each quadrant. A differential
magnetic field was calculated by subtracting
magnetic field vectors in two chosen yz cross
sections (fig. S2). In Fig. 1B, the resulting local-
field rotation is indicated by the black arrows,
and the magnitude is color mapped, with the
rotation angle (Dw) being defined as the dif-
ferences between the azimuth of the magnetic
fields in the two cross sections. The differential
field forms a quadrupole, with two left-handed
(positive Dw in red domains) and right-handed
(negative Dw in blue domains) magnetic field
domains. Figure S3 further delineates the
rotation of local fields along a chosen path-
way, demonstrating the helical magnetic field
distribution.
solution of glycerol (n = 1.475) with an in-
creasing volume ratio from 0 to 100%. The
increase in effective refractive index (n) of
the solution induced consistent CD-intensity
decrease under the same magnetic fields
(fig. S5). The nanorods’ CD signal weakened
but did not disappear as the solution refractive
index approached that of the silicon dioxide
(SiO2) layers (n = 1.475 ± 0.005) (38). This
refractive index–matching experiment demon-
strates that both the scattering and absorption
of the nanorods contribute to the overall CD
responses. We calculated an optical anisotropic
factor (g factor) of ~0.01 at 400 nm (fig. S6). To
verify the formation of chiral superstructures,
cyanine 3–doped Fe3O4@SiO2 core-shell nano-
rods (322.2 ± 16.5 nm in length, 70.2 ± 4.7 nm in
diameter, and 50.3 ± 1.5 nm in silica thickness)
were used as magnetic building blocks (39, 40)
and fixed in a polymer by photocuring under a
uniform magnetic field. Linear chains were
formed because of magnetic dipole-dipole
interactions and were parallel to the uniform
field with a standard deviation of 0.36° to
minimize the demagnetizing fields (figs. S7
and S8) (41). If a gradient field of a perma-
nent magnet (cube shape, 12 mm in edge
length) was used (fig. S9), the chain alignment
within one yz plane and the chain rotation
between different yz planes were determined
by the local magnetic fields and field rotation,
respectively (figs. S10 and S11). Thus, the chiral
superstructures made of large nanorods show
similar CD responses to magnetic fields (fig.
S12). Additionally, we characterized the chain
alignment in 10 sequential yz layers from x = 1 to
10 mm and in five layers along the y axis using
optical and electron microscopy, confirming the
chain rotation into chiral superstructures as
driven by the quadrupole chiral field (figs. S13
to S17).
Magnetic assembly and active tuning of
plasmonic chiral superstructures
To systematically study the CD dependence
on magnetic fields over a wide spectral range,

Li et al., Science 380, 1384–1390 (2023) 30 June 2023 1 of 7

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

Fig. 1. The quadrupole field chirality of a A 0 B 0 (deg)

90 6.10
permanent magnet. (A) Normalized magnetic
field (white arrows) and field-vector azimuth -1
(color maps) of a permanent magnet (high-

Z (cm)
Z (cm)
lighted in the middle square) with a cubic -2 −2
shape and edge length of 2 cm. (B) The field
rotating vectors (black arrows) and field angle -3
changes (color map) of the magnetic field -90 −6.10
along the x axis. Positive rotation angles (Dw) -4 −4
represent clockwise left-handed rotation of the -3 -2 -1 0 1 2 3 −3 −2 −1 0 1 2 3
magnetic field, and negative rotation angles Y (cm) Y (cm)
represent counterclockwise right-handed field
C D

Extinction
rotation. deg, degree. (C) Schematic illustra- 3

(a.u.)
tion of the magnetic assembly during CD z 2
measurement and simulated helical super- 1
0
structures assembled from magnetic nanorods x 300 450 600 750 900
under such a chiral magnetic field. (D) Extinction 0o 180o
Wavelength (nm)
spectrum and (E) CD spectra of Fe3O4@SiO2
kx E

CD (mdeg)
nanorods under different magnetic fields. 900
y
Arbitrary units, a.u.; mdeg, millidegree.
0
z z no magnet Bx
(0,1,-0.75)
By Bz
-900
x y 300 450 600 750 900
Wavelength (nm)

we introduced Fe3O4/Au hybrid nanorods as
building blocks by taking advantage of the
localized surface plasmon resonance (LSPR)
of Au nanorods and the magnetic responses
of Fe3O4 nanorods. The Au nanorods were
synthesized with a space-confined growth
method and had a length of 156.6 ± 15.2 nm
and a diameter of 48.9 ± 4.7 nm (42, 43). As
shown in fig. S18A, Fe3O4@SiO2 nanorods
were introduced (107.6 ± 5.2 nm in length,
13.0 ± 1.7 nm in diameter, and 5.0 ± 0.5 nm in
silica thickness) as initial templates, followed
by Au seed attachment (~2.0 nm in diameter).
During polymer coating, the SiO2 shells were
etched away, and defined gaps were formed
between the Fe3O4 nanorods and polymer
shells. Seeded growth was performed inside
the gaps to prepare the hybrid nanorods, which
comprise one Au nanorod and one Fe3O4
nanorod in a parallel configuration, as shown
in the transmission electron microscopy (TEM)
images (Fig. 2A) and elemental mapping (fig.
S19). Because of the confinement of the poly-
mer shells, radial growth was limited, and
preferable longitudinal growth produced the
Au nanorods. This growth mode allows easy
tuning of the nanorod length (fig. S18, B to E)
and shifts the LSPR of the Au nanorods from
560 to 880 nm (fig. S18F). The parallel align-
ment of the hybrid nanorods allowed the Au
nanorods to assemble into chiral superstruc-
tures under a chiral magnetic field and pro-
duce CD responses (fig. S20, A and B). Switching
the magnet dipole did not alter the CD profile,
but changing the magnet position to the
opposite side of the sample produced a CD
spectrum with an opposite sign (fig. S20C).
Applying two identical magnets in their at-
traction configuration generated a parallel
magnetic field with a reduced field gradient
between the two magnets and reduced the CD
signals (fig. S20, D and E). The disappearance
of CD signals confirmed that linear super-
structures assembled in a uniform magnetic
field could not produce CD signals in our ex-
perimental conditions. We also fixed the hybrid
nanorods under the absence and presence of
uniform and chiral magnetic fields using a
photocuring polymerization method, which
produced random, linear, and chiral struc-
tures, respectively. Although films containing
random and linear structures were not op-
tically active, the film with chiral superstructures
had CD signals. None of the three structures
showed evident responses to a magnet once
fixed in polymer films (fig. S21). These exper-
iments suggested that the observed CD re-
sponses in a single permanent magnet were
not induced by extrinsic chirality, a CD pro-
perty of achiral superstructures (44), or by mag-
netic circular dichroism (45).
We measured the CD spectra of Au nano-
rods while changing the position of the mag-
net vertically (Fig. 2B and fig. S22, A and B).
The CD signal declined gradually after the
magnet was moved from −1.25 to 0 cm along
the z axis (vertical direction) (fig. S22, C and
D), and the spectra changed the sign across
0 cm (Fig. 2C) in response to the field chirality
changes shown in fig. S22, E and F. The
dependence of CD signals on magnet posi-
tion was similar when the magnet dipole was
along the z axis (fig. S23). Changing the samples’
position relative to the magnet was equivalent
but induced a decrease in CD intensity when
the sample-magnet separation increased up to
3.0 cm along the x axis (fig. S24). The CD
spectra of Au nanorods were then measured
by decreasing field strength from 31.2 to 25.1,
18.3, 12.9, and 5.5 mT, which corresponded to
the magnet being 2.3, 2.5, 3, 4, and 5 cm away
from the nanorod dispersion. We observed a
consistent decrease in CD intensity for Au
nanorods with different LSPR positions and
for left- and right-handed chiral superstructures
(Fig. 2, D and E, and fig. S25), which cor-
responded to the decrease in field rotation
when the magnet departed the sample (Fig. 2F).
Detailed field analysis (Fig. 2, G to J) indicated
that the Dw decreased consistently in the four
chiral domains for increasing magnet-sample
separation. The sample in the first quadrant
was in a left-handed field in the By field (Fig. 2,
G and H), which caused the negative signals in
CD spectra. In the Bz field, the chiral field in the
first quadrant produced positive CD signals
(Fig. 2, I and J). The peak intensities of ex-
tinction and CD of the hybrid nanorods were
linearly proportional to Au concentration at
fixed magnet and sample positions (fig. S26),
which was consistent with Beer’s law and
molar ellipticity in CD. To verify the chirop-
tical properties, we modeled the chiral super-
structures and calculated their CD spectra using
a finite element method. The simulated spectra
in fig. S27 demonstrated CD responses of Au
nanorods with large separations. Considering
the lack of nanorod positional order in exper-
iments, complex models were further developed
to resolve the structures in figs. S15 and S17.
These models contain nanorods with random

Li et al., Science 380, 1384–1390 (2023) 30 June 2023 2 of 7

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

A B z C 400
1.25 cm
Fe3O4/Au y -1.25 cm
200

CD (mdeg)
(0,2.5,1.25)
0
x
(0,2.5,0)
-200
k
50 nm (0,2.5,-1.25) -400
Polymer 200 300 400 500 600 700 800 900
Wavelength (nm)
D E -455.0 F

7 6 5 4 3 2 1 1 2 3 4 5 6 7
400
31.2 mT 2.3 cm
300 25.1 mT

Perpen
18.3 mT

By
200 12.9 mT

CD (mdeg)
CD (mdeg)

100 5.5 mT By

0
3 cm

Vertical
-100 5.5 mT Bz

Bz
12.9 mT
-200
18.3 mT
-300 25.1 mT
5 cm
31.2 mT
-400
200 300 400 500 600 700 800 900 455.0
5.5 12.9 18.3 25.1 31.2
Wavelength (nm)
Field Strength (mT)

G 3 I 3

(deg)
(deg)

0 0

−3 −3
0 0 6.1
H J
Z (cm)

Z (cm)

−2 −2

−4 −4 −6.1
−6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6
Y (cm) Y (cm)

Fig. 2. Magnetic assembly of Fe3O4/Au hybrid nanorods into chiral super-
structures. (A) TEM image of the Fe3O4/Au hybrid nanorods wrapped within polymer
shells. (Inset) Schematic structure of the Fe3O4/Au hybrid nanorods. (B) Schematic
illustration of the magnet position during the CD measurement. (C) CD spectra of
the hybrid nanorods measured by changing the magnet position. (D) CD spectra of the
hybrid nanorods under magnetic fields with consistent direction and decreasing
strength. (E) Dependence of CD intensity on the field strength and the rod aspect
ratios. The magnetic fields are defined as By and Bz when the magnet dipole is parallel
to y and z axis, respectively. (F) Simulated field distributions of the cubic
magnet at given distances to the magnet surface. (G and I) Rotation angles
of magnetic fields between two yz cross sections, (–0.2, y, z) and (–0.8, y, z). The
orientation of the magnet is shown in the inset in each plot. (H and J) The
corresponding field rotation angles (color map) and the local magnetic field at
the cross section (–0.2, y, z). The magnet has a horizontal magnetic dipole in (G) and
(H) and a vertical magnetic dipole in (I) and (J). The field rotation is calculated
by Dw(y, z) = w(–0.8, y, z) − w(–0.2, y, z).

positions but constant rotation in different
planes (fig. S28), resembling the nanorod align-
ment in different cross sections in the SEM im-
ages. The simulated CD spectra show rotation
angle–dependent CD responses, which explains
the decrease of CD intensity with the rotation
angles of magnetic fields. We also observed
similar dependence in nanorods of increas-
ing aspect ratios, and the continuous redshift
of CD peaks in fig. S29 is qualitatively consistent
with the measured CD spectra in fig. S25.
Changing the directions of the magnetic
field produced more complex CD responses.
We considered the rotation of the magnet
within the xy and yz planes to explain the
involved mechanisms (Fig. 3A). The alignment
of the magnetic dipole of the magnet relative
to the axes was defined by the azimuth angle
q. The CD spectra showed intensity changes
when q increased to 180° in the xy plane (Fig.
3B), which increased to a saturated value and rods was determined by and could be predicted
decreased again (fig. S30). Given that the from the nanorod alignment, light incidence,
magnet was rotated in the xy plane (fig. S31), and light polarization (46). Because of the
the local magnetic fields and the field distri- changes in magnetic field direction and dis-
bution were analyzed for different q values. tribution shown in fig. S32, there was an
The schemes in fig. S31 illustrate the magnet’s associated change in the LSPR excitation of the
constant position and varying orientation. Au nanorods when q increased to 90°. At 0°, the
Because the incident light was along the field was nearly parallel to the x axis, and
x axis, the magnetic fields within the mea- the resulting parallel alignment of nanorods
sured domains in the yz and xz planes were to the incident light suppressed the longitu-
further plotted in Fig. 3, C and D, respectively. dinal mode. At 90°, the magnetic field being
We observed only slight changes in field di- nearly parallel to the z axis led to the exci-
rection and field distribution within the yz tation of the longitudinal mode. We could
plane for different magnet orientations (Fig. predict the longitudinal extinction of the
3C). In the xy planes, however, the field dis- Au nanorods through the equation sin2(ϕ),
tribution exhibited dramatic changes when q where ϕ is the angle between the local-field
increased to 90°, which led to excitation of direction and the light incident direction.
longitudinal mode with gradually increased Figure 3E shows a symmetric trend with maxi-
strength (Fig. 3D). For incident light along mum extinction between 60 and 120°. Compar-
the x axis, the plasmonic excitation of Au nano- ison of the predicted longitudinal extinction

Li et al., Science 380, 1384–1390 (2023) 30 June 2023 3 of 7

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

A B
y z 180 48.00
0o 0o
z 150

(deg)
y y 120
in x-y in y-z 90

Angle
x 60
y z
k 30

x (0,-2,-0.5) y 0 -147.0
200 300 400 500 600 700 800 900
Wavelength (nm)

C 0.5
-35
E
Z (cm)

0.0

-90
-0.5
-1.0 -0.5 0.0 -1.0 -0.5 0.0 -1.0 -0.5 0.0 -1.0 -0.5 0.0
D 0.5
Y (cm) Y (cm) Y (cm) Y (cm) F 50

CD (mdeg)
1 0
690 nm
Z (cm)

-50
0.0 535 nm
-100
-150
0
-0.5 0 30 60 90 120 150 180
-0.5 0.0 0.5 -0.5 0.0 0.5 -0.5 0.0 0.5 -0.5 0.0 0.5
X (cm) X (cm) X (cm) X (cm) Angle (deg)

Fig. 3. Tunable CD spectra of the assembled superstructures. (A) Schematic magnetic field of the cubic magnet in (C) yz and (D) xz planes. The azimuth
illustration of the azimuth changes of a cubic permanent magnet within the xy of the magnet is 0° to 30°, 60°, and 90° from left to right. (E) Predicted
and yz plane during CD measurements. (B) CD spectra of the hybrid nanorods longitudinal extinction of the nanorods based on the analytical solution.
measured by changing the magnet azimuth in the xy plane. (C and D) The (F) Dependence of the CD intensity on magnetic field q.

with the CD intensity measured at different q
(Fig. 3F) suggested that CD intensity would
depend on the magnet q and that the CD
responses of the assembled superstructures
would be determined by the longitudinal ex-
tinction changes when the magnet azimuth
increased within the xy plane.
Changing the magnet azimuth in the yz
plane altered the field chirality and led to a
different CD response. The measured CD spec-
tra are shown in fig. S33. The CD peaks at 545
and 698 nm simultaneously switch their signs
at about 20° and 110° during the measurement
(Fig. 4A, color map), which suggested changes
in handedness of the assembled superstruc-
tures. To verify this hypothesis, we used the
analytical model to directly map the local-field
rotation direction, chirality, and handedness
(fig. S2). At q = 0°, the sample was 2 cm away
from the magnet center with an upward off-
set of 0.5 cm, which led to an azimuth of
~14° relative to the magnet center (fig. S34).
Understanding the CD spectrum at q = 0° re-
quired access to field properties at this spe-
cific sample location in the yz plane.
The CD changes in response to magnet ro-
tation were studied by analyzing the local-
field properties in different q. In Fig. 4B, the
plots of field distribution and field chirality
within the yz plane showed a similar quadru-
pole field chirality. The normalized field vectors
at x = –0.2 and –0.8 cm were superimposed
in Fig. 4B to illustrate local-field rotation. The
field rotation angles along a trajectory high-
lighted in fig. S34C are shown in Fig. 4C,
which correspond to the field changes during
experimental measurement. The Dw was ini-
tially positive but changed its sign at q = 15°
and 105°, which is consistent with the experi-
mental chirality-transition angles plotted in
Fig. 4D. The Pearson correlation coefficients
between the field rotation angles and CD
intensity at 698 and 545 nm are –0.989 and
0.987, respectively (fig. S35). The strong negative
and positive correlation indicates that the field
chirality changes could explain the dependence
of the superstructure handedness and CD spec-
tra on magnet rotation in the yz plane.
Optical rotatory dispersion
We also studied the optical rotatory disper-
sion (ORD), which measures the polarization
rotation of a linearly polarized light. Left- and
right-handed circularly polarized light interacts
differently with chiral structures and travels
at different speeds inside of them. Because
linearly polarized light comprises two circu-
larly polarized light beams with the same
magnitude but opposite handedness, these
two highly correlated beams will develop a
phase difference, leading to the polarization
rotation of the incident beam (Fig. 4E). The
ORD effect was tested by applying an analyzer
to the incident beam and observing the color
changes. Only light of a specific wavelength
can pass through the analyzer at a polarization
angle (a) when the material is optically active.
Experimentally, a is defined as the angle
between the analyzer polarization direction
and the horizontal baseline, and the polariza-
tion of the polarizer is fixed along the vertical
direction.
Digital images of a nanorod dispersion are
shown in Fig. 4F, before (left) and after (right)
application of a z-directional magnetic field at
a = 3°. The original dispersion appeared dark
without noticeable colors because only mini-
mal light could transmit through the analyzer.
Under a z-directional magnetic field, the top
domain turned yellow, and the bottom turned
red, demonstrating the formation of chiral
superstructures with opposite handedness
in these two domains, which is consistent
with the predicted field chirality in the first

Li et al., Science 380, 1384–1390 (2023) 30 June 2023 4 of 7

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

1
A 180
B 6.1
C
180.0
0
150
(deg)

120 -1

Z (cm)
90
-2
D
Angle

200
60

CD (mdeg)
100
-3
30 0
-100
0 -255.0 -4 698 nm
-200
200 300 400 500 600 700 800 900 −6.1 545 nm
-300
Wavelength (nm) -5 0 30 60 90 120 150 180
-3 -2 -1 0 1 2 3
Y (cm) Angle (deg)

E Polarizer z Analyzer G -30o -27o-24o-21o-18o-15o -12o -9o -6o -3o 0o 3o 6o 9o 12o 15o 18o 21o 24o 27o 30o
α By
x

y Bz

F
(Bz) Bx

No
w/o mag w/ mag (Bz)

Fig. 4. Tunable CD spectra and optical rotatory dispersion of the assembled
superstructures. (A) CD spectra of the hybrid nanorods measured by changing the
magnet azimuth in the yz plane. (B) The magnetic field rotation (color map) and
the normalized field vectors (arrows) of the cubic magnet. The magnetic field
rotation is calculated by Dw(y, z) = w(–0.8, y, z) – w(–0.2, y, z). The normalized
magnetic field vectors in (–0.8, y, z) are indicated with black arrows, and the field
vectors in (–0.2, y, z) are indicated with orange arrows. (C) Dependence of field
rotation on the magnet field azimuth. (D) Dependence of the CD peak intensity
on the magnet field azimuth. (E) Schematic illustration of the ORD measurement.
The a is introduced as the angle between the polarization direction of the analyzer
and the horizontal direction. (F) Digital images of a hybrid nanorod dispersion in
a cuvette without (w/o mag) and with (w/ mag) a magnetic field. The a is 3°.
(G) Digital pictures of the hybrid nanorod dispersion under different magnetic field
conditions. The analyzer’s a was switched from –30 to 30° during the measurement.

and fourth quadrants of the magnet. This
ORD effect was determined by superstruc-
ture handedness and the angle a. Under a
z-directional magnetic field, we observed sim-
ilar two-color domains when a increased from
–30 to 30° (Fig. 4G), with red-orange-yellow-
green changes in the top domain and oppo-
site color changes in the bottom domain.
Changing the magnetic field to the y axis led
to color switching between the two domains,
corresponding to a field chirality transition.
Nanorod dispersion under the absence and
presence of the x-directional magnetic field
only exhibited contrast changes at different a
because of the negligible CD responses at these
two conditions.
Generalizing the all-scale chiral assembly
The chirality formed by nanoscale magnetic
assembly could be transferred to organic
molecules, polymers, oxides, and semiconduc-
tors. These guest materials were introduced to
the magnetic nanorods through surface-coating
and doping methods, which have the advan-
tages of wide material accessibility and easy
further processing. Starting with Fe3O4@SiO2
nanorods with a length of 107.6 ± 5.2 nm, a
diameter of 13.0 ± 1.7 nm, and silica thickness
of 5.0 ± 0.5 nm, we coated their surface with
Cu2O nanoparticles and resorcinol-formaldehyde
(RF), with the latter being converted into MnO2
by reacting with KMnO4 (47–49). The resulting
Fe3O4@SiO2@MnO2 nanorods (Fig. 5A) under-
went oxidation-induced volume expansion and
created nanogaps between porous MnO2 nano-
shells and Fe3O4@SiO2 cores. The extinction
spectra of these samples in Fig. 5B showed
broad extinction of Fe3O4@SiO2 and Fe3O4@
SiO2@Cu2O nanorods, and extinction peaks of
Fe3O4@SiO2@RF and Fe3O4@SiO2@MnO2
nanorods. These four samples showed evident
CD activities under magnetic fields that can
also be tuned by changing the field directions
(Fig. 5C). The CD peak positions being near
their extinction peak positions indicates that
the chirality in the superstructures transferred
from the host nanorods to the guest mole-
cules. In addition, changing the field direction
from the y axis to the z axis changed the chiral
superstructures from left-handed to right-handed
symmetry.
Transfer of chirality to small molecules was
demonstrated by doping organic dyes into RF
polymeric shells through electrostatic inter-
actions. Figure S36A shows a typical TEM
image of the Fe3O4@SiO2@RF nanorods with
100-nm RF shells that carry negative charges
after a condensation reaction. We chose three
dyes—methylene blue, methylene green, and
neutral red—and their molecular structures
are given in fig. S36B. These dyes develop
positive charges after dissociation of chloride
anions in water and can be doped into porous
RF shells by mixing with the nanorods at room
temperature. The dyed solutions appeared
blue, green, and red after removing the excess
molecules, with extinction spectra exhibiting
distinct peaks after successful doping (Fig. 5D).
Using the green dye–doped nanorods as
an example, we confirmed their CD responses
under different field directions (fig. S36C),
with negligible CD signals under no and
x-directional magnets and opposite CD peaks
in y- and z-directional magnetic fields. The
CD spectra of nanorods doped with all three
dyes (Fig. 5E) showed peaks at their charac-
teristic wavelengths. The observation is con-
sistent with that of plasmonic nanorods and
demonstrated the formation of chiral super-
structures and the successful transfer of chi-
rality from the nanoscale to molecular levels.
The g factor calculated in Fig. 5F has a max-
imum of ~0.003 for methylene blue–doped
nanorods, a value comparable to that of classic
chiral molecules and magnetically assembled
chiral superstructures of inorganic molecules,
polymers, and semiconductors (figs. S37 and

Li et al., Science 380, 1384–1390 (2023) 30 June 2023 5 of 7

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

A B C

Normalized absorbance (a.u.)

1.0 Fe3O4@SiO2 Fe3O4@SiO2/By
1

Normalized CD (mdeg)
Fe3O4@SiO2/Bz
Fe3O4@SiO2@CuO2
Fe3O4@SiO2@Cu2O/By
0.8 Fe3O4@SiO2@RF Fe3O4@SiO2@Cu2O/Bz
Fe3O4@SiO2@MnO2
0.6
0
0.4
Fe3O4@SiO2@RF/By
Fe3O4@SiO2@RF/Bz
0.2
Fe3O4@SiO2@MnO2/By
-1 Fe3O4@SiO2@MnO2/Bz

200 nm 0.0
300 400 500 600 700 800 900 300 400 500 600 700 800 900
Wavelength (nm) Wavelength (nm)
D E F
2.5 400 7
Fe3O4@RF undoped/By
Methylene blue
300 undoped/Bz 6 Methylene blue
Absorbance (a.u.)

2.0 Methylene green blue/By Methylene green

200

g-factor (10-3)
Neutral red blue/Bz 5 Neutral red
CD (mdeg)

1.5 100
4
0
1.0 3
-100
green/By 2
-200 green/Bz
0.5
-300 red/By 1
red/Bz
0.0 -400 0
300 400 500 600 700 800 900 1000 300 400 500 600 700 800 900 300 400 500 600 700 800 900
Wavelength (nm) Wavelength (nm) Wavelength (nm)

Fig. 5. All-scale magnetic assembly of achiral molecules into chiral
superstructures. (A) TEM image of the Fe3O4@SiO2@MnO2 core-shell
nanorods. (B) Extinction spectra of core-shell Fe3O4@SiO2 and Fe3O4@SiO2@RF
nanorods, Fe3O4@MnO2 yolk-shell nanorods, and Fe3O4@SiO2@Cu2O
core-satellite nanorods. (C) The corresponding CD spectra measured under
y- and z-directional magnetic fields. (D) Extinction spectra of the
Fe3O4@SiO2@RF nanorods and nanorods doped with the three organic
dyes. (Insets, left to right) Digital pictures of the nanorods before and after
doping with methylene blue, methylene green, and neutral red. (E) CD
spectra of the doped and undoped nanorods under y- and z-directional
magnetic fields. (F) The g factor of the doped nanorods under a z-directional
magnetic field.

S38). We further demonstrated that the mag-
netic assembly strategy could be extended
to the assembly of fluorophores for generat-
ing circularly polarized luminescence. When
europium-doped NaYF4 nanorods were decor-
ated with Fe3O4 nanoparticles, the fluorescent
nanorods could be assembled into chiral
superstructures and exhibited circularly polar-
ized luminescence (fig. S39).
Discussion
The consistent rotation of the local-field vec-
tors of the cubic permanent magnet forms
the quadrupole field chirality with alternat-
ing left-handed and right-handed magnetic
fields in the four quadrants. Such chiral mag-
netic fields induce the assembly of magnetic
nanorods into chiral superstructures, with
handedness and chirality determined by the
local features of the magnetic fields. This
strategy allows remote, reversible, and instan-
taneous assembly of chiral superstructures
from nanostructures of various chemical com-
pounds (plasmonic materials, polymers, oxides,
metals, semiconductors, fluorescent nanostruc-
tures, and molecular moieties) and active
tuning of their CD responses in a broad range
of spectra and circularly polarized lumines-
cence, as long as they can be properly bound to 16. C. Hao, L. Xu, H. Kuang, C. Xu, Adv. Mater. 32, e1802075
the magnetic nanorods. Fixing the chirality of (2020).
17. F. Neubrech, M. Hentschel, N. Liu, Adv. Mater. 32, e1905640
these chemical compounds at all scales is (2020).
possible by embedding the formed super- 18. Y. Y. Lee, R. M. Kim, S. W. Im, M. Balamurugan, K. T. Nam,
structures in polymer substrates, which could Nanoscale 12, 58–66 (2020).
19. Z. Li, Q. Fan, Y. Yin, Chem. Rev. 122, 4976–5067 (2022).
be realized by applying an external magnetic 20. N. J. Greenfield, Nat. Protoc. 1, 2876–2890 (2006).
field during polymerization. This simple 21. B. Shen et al., Sci. Adv. 4, eaap8978 (2018).
strategy makes the chiral superstructures 22. B. Frank et al., ACS Nano 7, 6321–6329 (2013).
nonvolatile without external magnetic fields 23. A. Kuzyk et al., Nature 483, 311–314 (2012).
24. D. Winogradoff et al., Adv. Sci. (Weinh.) 8, 2003113
and highly accessible for portable chiroptical (2021).
devices. 25. K. Martens et al., Nat. Commun. 12, 2025 (2021).
26. Y. Chen et al., Nat. Rev. Phys. 4, 113–124 (2022).
27. P. Zhan et al., Sci. Adv. 5, eaax6023 (2019).
RE FERENCES AND NOTES 28. L. Xin, C. Zhou, X. Duan, N. Liu, Nat. Commun. 10, 5394
1. G. Singh et al., Science 345, 1149–1153 (2014). (2019).
2. X. Lan et al., J. Am. Chem. Soc. 137, 457–462 (2015). 29. S. Li et al., J. Am. Chem. Soc. 138, 306–312 (2016).
3. W. Ma et al., Chem. Rev. 117, 8041–8093 (2017). 30. A. Cecconello, L. V. Besteiro, A. O. Govorov, I. Willner, Nat. Rev.
4. J. Lv et al., Nat. Rev. Chem. 6, 125–145 (2022). Mater. 2, 1–19 (2017).
5. M. Hentschel, M. Schäferling, X. Duan, H. Giessen, N. Liu, Sci. 31. A. Kuzyk, M. J. Urban, A. Idili, F. Ricci, N. Liu, Sci. Adv. 3,
Adv. 3, e1602735 (2017). e1602803 (2017).
6. C. Zhou, X. Duan, N. Liu, Acc. Chem. Res. 50, 2906–2914 32. S. Zhou et al., Nature 612, 259–265 (2022).
(2017). 33. Z. Cheng, M. R. Jones, Nat. Commun. 13, 4207 (2022).
7. R. Schreiber et al., Nat. Commun. 4, 2948 (2013). 34. J. Lu et al., Science 371, 1368–1374 (2021).
8. D. Zerrouki, J. Baudry, D. Pine, P. Chaikin, J. Bibette, Nature 35. G. M. Whitesides, B. Grzybowski, Science 295, 2418–2421
455, 380–382 (2008). (2002).
9. L. Xu et al., Nature 601, 366–373 (2022). 36. X. Lan, Q. Wang, Adv. Mater. 28, 10499–10507 (2016).
37. C. A. Batista, R. G. Larson, N. A. Kotov, Science 350, 1242477
10. N. A. Kotov, L. M. Liz-Marzan, P. S. Weiss, ACS Nano 15,
12457–12460 (2021). (2015).
11. K.-J. Jeong et al., ACS Nano 14, 7152–7160 (2020). 38. B. N. Khlebtsov, V. A. Khanadeev, N. G. Khlebtsov, Langmuir
12. H.-E. Lee et al., Nature 556, 360–365 (2018). 24, 8964–8970 (2008).
39. W. Xu et al., Nano Lett. 17, 2713–2718 (2017).
13. W. Jiang et al., Science 368, 642–648 (2020).
14. G. González-Rubio et al., Science 368, 1472–1477 (2020). 40. Z. Li, C. Qian, W. Xu, C. Zhu, Y. Yin, Sci. Adv. 7, eabh1289
15. L. Ohnoutek et al., Nat. Photonics 16, 126–133 (2022). (2021).

Li et al., Science 380, 1384–1390 (2023) 30 June 2023 6 of 7

RES EARCH | R E S E A R C H A R T I C L E
41. M. Wang, Y. Yin, J. Am. Chem. Soc. 138, 6315–6323
(2016).
42. Z. Li et al., J. Am. Chem. Soc. 143, 15791–15799 (2021).
43. Z. Li, J. Jin, F. Yang, N. Song, Y. Yin, Nat. Commun. 11, 2883
(2020).
44. E. Plum, V. Fedotov, N. Zheludev, Appl. Phys. Lett. 93, 191911
(2008).
45. B. Han, X. Gao, J. Lv, Z. Tang, Adv. Mater. 32, e1801491
(2020).
46. M. Wang et al., J. Am. Chem. Soc. 135, 15302–15305 (2013).
47. C. Liu, H. Dong, N. Wu, Y. Cao, X. Zhang, ACS Appl. Mater.
Interfaces 10, 6991–7002 (2018).
48. S. Zhou et al., Nanoscale 12, 15460–15465 (2020).
49. Y. Bai, X. Yao, X. Wang, Y. Yin, ChemNanoMat 6, 1186–1190
(2020).
50. Z. Li, Programming code for: A magnetic assembly approach to
chiral superstructures, version 1, Zenodo (2023); https://doi.
org/10.5281/zenodo.7686903
Li et al., Science 380, 1384–1390 (2023) 30 June 2023
ACKN OWLED GMEN TS The programming code to model magnetic fields and datasets
We thank the Central Facility for Advanced Microscopy and used to analyze the field chirality are available online at Zenodo
Microanalysis at University of California Riverside (UCR) for (50). License information: Copyright © 2023 the authors, some
help with TEM analysis and G. Ung’s group at University rights reserved; exclusive licensee American Association for the
of Connecticut for assistance with measuring the circularly Advancement of Science. No claim to original US government
polarized luminescence. Funding: US National Science Foundation works. https://www.science.org/about/science-licenses-journal-
CHE- 2203972 (Y.Y.). Authors contributions: Z.L. and Y.Y. article-reuse
conceived and planned this study. Z.L. and Q.F. developed the
methodology for modeling and simulation. Z.L. developed the SUPPLEMENTARY MATERIALS
synthesis and characterization of nanoparticles. Z.L., Z.Y., and C.W.
science.org/doi/10.1126/science.adg2657
performed the CD measurements. Z.W. contributed to the
Materials and Methods
circularly polarized luminescence studies. All authors discussed
Supplementary Text
and analyzed the results and contributed to the writing of this
Figs. S1 to S39
paper. Competing interests: Y.Y. and Z.L. have filed a patent
Table S1
application related to this work filed by UCR. The anthors declare
References (51–57)
that they have no other completing interests. Data and materials
availability: All data needed to evaluate the conclusions in the Submitted 12 December 2022; accepted 22 May 2023
paper are present in the paper or the supplementary materials. 10.1126/science.adg2657
7 of 7
RES EARCH

GAMMA-RAY BURSTS (XRT) observations began 143 s after the BAT

trigger (15).
A tera–electron volt afterglow from a narrow jet in The exceptionally large fluence of this event
saturated almost all gamma-ray detectors dur-
an extremely bright gamma-ray burst ing the main burst. The event fluence of GRB
221009A was measured by the Konus-Wind
LHAASO Collaboration*† spacecraft from T0 + 175 s to T0 + 1458 s as
≳5  102 erg cm2 in the energy range of 10 to
Some gamma-ray bursts (GRBs) have a tera–electron volt (TeV) afterglow, but the early onset of this has 1000 keV (16), higher than any previously ob-
not been observed. We report observations with the Large High Altitude Air Shower Observatory served GRB. Later optical observations mea-
(LHAASO) of the bright GRB 221009A, which serendipitously occurred within the instrument’s field of sured the redshift (z) of the afterglow, finding
view. More than 64,000 photons >0.2 TeV were detected within the first 3000 seconds. The TeV flux z = 0.151 (17, 18). Assuming standard cosmol-
began several minutes after the GRB trigger and then rose to a peak ~10 seconds later. This was ogy, the burst isotropic-equivalent energy re-
followed by a decay phase, which became more rapid ~650 seconds after the peak. We interpret the lease Eg,iso is at least 3 × 1054 erg, among the
emission using a model of a relativistic jet with half-opening angle of ~0.8°. This is consistent with the highest measured.
core of a structured jet and could explain the high isotropic energy of this GRB. We observed GRB 221009A with the Large
High Altitude Air Shower Observatory (LHAASO).

G
amma-ray bursts (GRBs) are explosions
observed in distant galaxies, character-
ized by a rapid flash of gamma rays last-
ing from a fraction of a second up to
several hundreds of seconds. The pro-
genitors of long-duration ð≳2 sÞ GRBs are thought
to be collapsing massive stars, whereas short-
duration ð≲2 sÞ GRBs are produced by the merger
of two compact objects (1). The emission of a
GRB consists of two stages—the prompt emis-
sion and the afterglow, which can partially
overlap in time. Prompt emission has irregular
variability on timescales as rapid as millisec-
onds, which is thought to result from internal
shocks or other dissipation mechanisms that
occur within the source. The afterglow emis-
sion is smoother and lasts much longer, with
the flux decay usually following a power-law in
time. The long-lasting afterglow is thought to
result from external shocks caused by the in-
teraction between relativistic jets (produced
by the GRB) with the ambient medium at large
distances from the source. The afterglow emis-
sion spans a wide range of the electromag-
LHAASO is a VHE gamma ray extensive air
and rising afterglow phases. Extensive air shower detector (19), consisting of three inter-
shower detectors have a larger instantane- connected detectors. At the time of the GBM
ous field of view, have an observing duty cycle trigger, the location of GRB 221009A was within
close to 100%, and do not need to be pointed, the field of view of LHAASO at a zenith angle of
so they could potentially observe the prompt 28.1° (fig. S2). We analyzed the observations of
GRB and afterglow onset phases. However, at- GRB 221009A with LHAASO’s Water Cherenkov
tempts to observe GRBs with extensive air Detector Array (WCDA) (fig. S1) (19), which
shower detectors have not resulted in detec- detected the burst at coordinates of right as-
tions (10–12). cension (RA) 288.295 ± 0.005 (stat) ± 0.05 (sys)
degrees and declination 19.772 ± 0.005 (stat) ±
Observations of GRB 221009A 0.05 (sys) degrees, with a statistical signifi-
On 9 October 2022, at 13:16:59.99 universal cance >250 standard deviations (fig. S3). With-
time (UT) (hereafter T0), the Gamma-ray Burst in the first ~3000 s after the trigger, WCDA
Monitor (GBM) on the Fermi spacecraft de- detected >64,000 photons with energies be-
tected and located the burst GRB 221009A tween ~200 GeV and ~7 TeV associated with the
(13). The Large Area Telescope (LAT) on Fermi GRB. It was observed by LHAASO for ~6000 s
also detected high-energy emission from the before moving out of the field of view. At these
burst (14). The Burst Alert Telescope (BAT) on energies, another detector reported no emis-
the Neil Gehrels Swift Observatory (Swift) space- sion ~8 hours after T0 (20).
craft detected this burst when it came into Figure 1 displays the WCDA light curve (count
view 53 min later, and Swift’s X-Ray Telescope rate as a function of time) of GRB 221009A.
900
Count rate [ count s-1 ]

netic spectrum. The radio to sub-GeV emission 800

800
is generally interpreted as synchrotron radia-
Count rate [ count s-1 ]

700
tion, generated by acceleration of relativistic 600
electrons by the external shock (1). The same 600 500
electrons can up-scatter the synchrotron pho-
400
tons through the inverse Compton (IC) mech-
300
anism, producing a synchrotron self-Compton
400 200
(SSC) emission component that extends to
100
very high energy [(VHE) >100 GeV] gamma
rays (2–5). 220 230 240 250 260 270 280 290 300 310 320

VHE emission has been detected from a few 200 T* = T0 + 226 s Time since GBM trigger [ s ]
GRBs during the afterglow decay phase (6–9).
However, those observations used pointed
instruments that had slew to the GRB position 0
(after it was detected by other instruments), so 0 500 1000 1500 2000 2500 3000 3500 4000
they did not capture the prompt emissions Time since GBM trigger [ s ]

*Corresponding authors: X. Y. Wang (xywang@nju.edu.cn);
Z. G. Yao (yaozg@ihep.ac.cn); Z. G. Dai (daizg@ustc.edu.cn);
M. Zha (zham@ihep.ac.cn); Y. Huang (huangyong96@ihep.ac.cn);
J. H. Zheng (mg21260020@smail.nju.edu.cn)
†LHAASO Collaboration authors and affiliations are listed in the
supplementary materials.
Fig. 1. Count rate light curve of GRB 221009A observed by LHAASO-WCDA. The energy range of
photons observed is ~0.2 to 7 TeV. The inset panel shows a zoomed-in view of the light curve during 220
to 320 s (yellow shaded zone) after the GBM trigger (T0), with the arrow indicating the reference time
T* = T0 + 226 s for our light curve analysis (see text). Blue histograms are the data, and black histograms are
the estimated background.

LHAASO Collaboration, Science 380, 1390–1396 (2023) 30 June 2023 1 of 6

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

The VHE emission exhibits a smooth tempo-

ral profile, with a rapid rise to a peak followed
5
A
by a more gradual decay that persists for at 10
least 3000 s after the peak. By contrast, the

E2 Flux [ erg cm-2 s-1 ]

MeV gamma-ray light curves, measured by GBM 10 6

(13) and other gamma-ray detectors (16, 21),

are highly variable. The GBM emission includes 10 7

an initial precursor pulse lasting ~10 s, which

sets T0, followed by an extended, much brighter, 8
10
and multipulsed emission episode (fig. S10).
The contrast between the TeV and MeV light 9
10
curves suggests that the TeV emission has a dif-
ferent origin than the prompt MeV emission. T*+[5, 14] s ( 10) T*+[14, 22] s T*+[22, 100] s ( 0.25) T*+[100, 674] s ( 0.2)
10
10 T*+[674, 1774] s ( 0.3) T*+[-6, 4] s ( 0.05)
Analysis of the gamma-ray data
Gamma rays emitted from distant astronom- 10 1 2 10 1
3 10 1
1 2 3 4 5 6 7 8 9
ical sources are attenuated by photon interac- 5
B Energy [ TeV ]
10
tions with the extragalactic background light
E2 Flux [ erg cm-2 s-1 ]

(EBL). The gamma-ray spectrum that would 6

10
be observed if the EBL were absent, referred to
as the intrinsic spectrum, can be inferred from
7
the observed events by correcting for EBL at- 10
tenuation. We perform this correction by as-
8
suming an EBL model (22). Figure 2 presents 10
both the observed and the EBL-corrected in-
trinsic flux spectra in the energy range from 10 9

~200 GeV to ~7 TeV, for five time intervals dur-

ing which the GRB was detected by LHAASO-
WCDA. The time intervals were chosen to cover
the rising phase, the peak, and three periods in
the decay phase. We fitted the intrinsic spectra
with a power-law model (23), which is con-
sistent with the data up to at least 5 TeV with
no evidence for a spectral break or cut-off. The
best-fitting values of the spectral indices g of
the power-law function are given in table S2.
These show a mild spectral hardening in time,
with g increasing by ~0.2 between the first and
last time intervals. The uncertainty in the EBL
model (22) is equivalent to a similar change in
the spectral index (table S2). During the initial
main burst phase ðT0 þ ½220; 230 sÞ , no TeV
emission is detected (significance <2.3s), with
the 95% upper limits on the flux shown in
Fig. 2.
Figure 3 shows the light curve of GRB 221009A,
converted to energy flux and integrated in the
energy range of 0.3 to 5 TeV. The peak ob-
served flux is ∼1:2  105 erg cm2 s1, which
corresponds to an apparent isotropic-equivalent
luminosity of Liso;TeV ∼ 7:3  1050 erg s1 in
the range of 0.3 to 5 TeV, which is higher than
other known sources at these energies. Be-
cause the Fermi GBM triggered on the pre-
cursor emission of this GRB, which was very
weak compared with the main burst that started
after a quiescent period of about 200 s (fig.
S10), we set the reference time (hereafter T *)
of the afterglow light curves at the onset of
the main component (24). Numerical studies
have shown that this is a better approximation
than using T0 when investigating the after-
glow emission (24, 25). The first main pulse of
GRB 221009A lasted from 225 to 228 s after
T*+[5, 14] s ( 10) T*+[14, 22] s T*+[22, 100] s ( 0.25)
10 10
T*+[100, 674] s ( 0.2) T*+[674, 1774] s ( 0.3)
10 1 2 10 1
3 10 1
1 2 3 4 5 6 7 8 9 10
Energy [ TeV ]
Fig. 2. Intrinsic and observed flux spectra for five time intervals. (A) The intrinsic spectra corrected
for EBL attenuation, assuming power-law functions in the calculation (23). (B) The observed spectra obtained
by reapplying the EBL attenuation to the corresponding intrinsic spectra. The shaded bands indicate
1s ranges of statistical uncertainties (inner bands) and systematic uncertainties (outer bands). The
systematic uncertainties are further divided into detector-related (middle band in light blue) and EBL-related
(outer band in light cyan) components. The 95% upper limits on the intrinsic flux during the initial main
burst phase ðT0 þ ½220; 230 sÞ are indicated by dark orange triangles in (A). The EBL model (22) and its
uncertainties are used to calculate the spectra and estimate the systematic uncertainties. The wave-like
features in the observed spectrum are caused by the EBL attenuation at different wavelengths. An alternative
calculation, independent on the EBL model, is shown in fig. S4. Goodness-of-fit tests are shown in fig. S5,
and the results are listed in table S1.
the GBM trigger (16, 21), indicating that T * is the slope of the steep decay after the break is
þ0:30
in the range of 225 to 228 s. We measured T* a3 ¼ 2:210:83 . The transition time from the
by fitting the LHAASO light curve with a rapid rise to the slow rise phase is Tb;0 ¼
model consisting of multisegment power-laws T ∗ þ 4:85þ0:15 ∗
0:10 s, the peak time is Tpeak ¼ T þ
(23), finding T ∗ ¼ 225:7 þ2:2
3:2 s (fig. S6). In the 18:0þ1:2
1:2 s, and the break time to the steep de-
following analysis, we adopt T * = 226 s. cay is Tb;2 ¼ T ∗ þ 670þ230110 s (table S3).
Figure 3 has a four-segment shape, consist- Although the rapid rise is evident in the light
ing of a rapid initial rise, a slower rise up to the curve (fig. S8), the observations have insufficient
peak, a slow decay after the peak, and then a temporal resolution to constrain the functional
steep decay after a break, each of which is con- form of such a rapid rise. We assume a power-
sistent with a power-law function of time. We law because it is consistent with the other three
fitted the data with a semismoothed quadruple- segments of the light curve.
power-law model [(SSQPL) in which a smoothed We verified that the steep decay is required
triple-power-law is connected to an initial rapid by comparing the four-segment model with a
power-law rise] (23). We find that the slope of three-segment model with only one segment
the rapid rise is a0 ¼ 14:9þ5:73:9 , the slope of the for the whole decay phase (Fig. 3). We find that
slow rise is a1 ¼ 1:82þ0:21
0:18, the slope of the slow the four-segment model improves the fit over
decay after the peak is a2 ¼ 1:115þ0:012 0:012 , and the three-segment model with a significance of

LHAASO Collaboration, Science 380, 1390–1396 (2023) 30 June 2023 2 of 6

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

sion in GRB 221009A suggests that it mainly

a b c d e
A results from an external shock. Synchrotron
emission of relativistic electrons has a maximum
Energy flux [ erg cm-2 s-1 ]

5
10 energy of ≪1 TeV if the electrons are accelerated
and radiate in the same zone, therefore the TeV
emission of GRB 221009A is probably produced
10 6 by SSC of relativistic electrons in the external
shock, as has been proposed for previous TeV
afterglows (7, 27, 28).
In the external shock model, the rise phase
7
10 before the peak corresponds to the afterglow
onset, where the forward-moving external shock
sweeps up an increasing amount of ambient
10 8 matter before being substantially decelerated
(known as the coasting stage). The density n of
1 a 10 b c 102 d 10e3
B the ambient matter is described by n(R) º R−k,
2.6 Time since T* [ s ]
where R is the radius of the external shock, k = 0
corresponds to a homogeneous medium, and
k = 2 is expected for a stellar wind from the
Spectral Index

2.4 GRB progenitor. In the homogeneous medium

case, the flux increases with time as t2 during
the coasting stage, where t is the time since T *
2.2
in the observer frame, provided that the ob-
served frequency is above the peak frequency
of the SSC spectrum (i.e., in the spectral re-
gime of Fn º n−p/2, where Fn is the flux density
2 at frequency n, and p is the power-law index of
the electron energy distribution) (23). In the
1 10 102 103 stellar wind case, the light curve of the TeV
Time since T* [ s ] afterglow is expected to be flat or possibly de-
cline with time in the same spectral regime
Fig. 3. Energy flux light curve and spectral evolution in the VHE band for GRB 221009A. (A) The light (23, 29). To produce a rising flux in the wind
curve converted to energy flux, integrated over the energy range of 0.3 to 5 TeV. Blue points indicate the medium pcase, ffi Lorentz factor G (defined as
ffiffiffiffiffiffiffiffiffiffiffiffithe
observations, with error bars indicating the ~1s statistical uncertainty [the systematic uncertainty of ~10.7% G ≡ 1= 1  b2 , where b is the velocity of the
(23) is not included]. The solid orange curve shows the fitted model, consisting of four joint power-laws that shock divided by the speed of light c) of the
describe the four-segment features in the light curve: rapid rise, slow rise, slow decay, and steep decay. The forward shock must increase with time, per-
dark red dashed line shows the three-segment model, which has only one segment for the entire decay haps as a result of energy injection. The rising
phase. The best-fitting parameter values for both models are listed in table S3. (B) The temporal evolution slope that we observe, a1 ¼ 1:82þ0:21 0:18 , is con-
of the power-law spectral index (blue circles) of photons, determined from the time-resolved intrinsic sistent with homogeneous medium without
spectra. The orange line is the function g(t) = a log(t) + b fitted to the data points. This model is flat before energy injection, so we infer k = 0. The initial
T0 + 230 s. The time intervals for the spectrum fitting in Fig. 2 are labeled a to e in gray in both panels. rapid rise phase has a high slope, albeit with a
The average spectral indices of these five intervals are indicated by gray hollow squares in (B). large uncertainty. Fn º t4 might apply at this
stage because the spectrum is expected to be
harder (Fn º n−(p−1)/2) at such an early time (23).
9.2s (23), indicating that a separate steep de- does not affect the identification of the steep This phase overlaps in time with the strongest
cay phase is justified by the data. decay. pulses of the prompt main burst emission
We also applied the four-segment model to A potential systematic uncertainty in the (16, 21), so energy could be transferred to the
light curves of sliced data samples with different flux arises from the adopted EBL model. To external shock by the inner ejecta, producing a
energy bands (Fig. 4 and fig. S7). The best-fitting test whether the uncertainty in the EBL affects rapid flux increase (23).
parameters are listed in table S3. The peak times the light curve, we recalculated the light curve After the peak, the expected decay of the SSC
for different energy bands agree with each other, using the EBL intensities at the lower and emission is t−(9p−10)/8 in the spectral regime
within the uncertainties, without any spectral upper boundaries of the EBL model (22). We of Fn º n−p/2, when the IC scattering is in the
evolution, which indicates that the peak corre- then fitted the light curve with the same pro- Thomson regime (28). Depending on the phys-
sponds to the onset of the afterglow phase. cedure, and the results are given in table S3. We ical parameters, the scatterings could enter
The break in the decay phase of the light curve find that the slopes and break times in the light into the Klein-Nishina (KN) regime, where the
is observed in all energy bands. We find no curves remain almost unchanged, although photon energy (in the rest frame of the scat-
systematic shift of the break time, which is the intrinsic fluxes change systematically. tering electron) becomes similar to the elec-
also constant within the uncertainties. tron rest-mass energy. In the KN regime, the
decay is t ð2 4 Þ in the spectral regime of Fn º
3 5p
We identify a small flare at T * þ ½320; 550 s Interpreting the TeV emission
(fig. S9). To check whether this flare affects TeV gamma-ray emission could be produced n−(p−1) (23). Both interpretations are consistent
þ0:012
the break at tb,2, we masked the data during by relativistic electrons accelerated by internal with our observed decay slope,a2 ¼ 1:1150:012 ,
the flare period when performing the fitting. shocks during the prompt emission (26) or by for p ~ 2.1, although the observed spectrum
This analysis shows that the break’s behav- external shocks during the afterglow phase (2–4). in this period is slightly softer than the mod-
ior remained the same, indicating that the flare The smooth temporal profile of the TeV emis- el prediction.

LHAASO Collaboration, Science 380, 1390–1396 (2023) 30 June 2023 3 of 6

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

A Emedian = 0.35 TeV B Emedian = 0.41 TeV

Energy flux [ erg cm-2 s -1 ]

Energy flux [ erg cm-2 s-1 ]

10
5 PSF: (1.07, 2.05)
10 5 PSF: (1.04, 2.89)

6 6
10 10

7 7
10 10

10 8 10 8

1 10 102 103 1 10 102 103

C Time since T* [ s ] Emedian = 0.54 TeV D Time since T* [ s ] Emedian = 0.81 TeV
Energy flux [ erg cm -2 s -1 ]

Energy flux [ erg cm-2 s-1 ]

10 5 PSF: (0.85, 2.26)
10 5 PSF: (0.62, 1.19)

10 6 10 6

10 7 10 7

10 8 10 8

1 10 102 103 1 10 102 103

E Time since T* [ s ]
Emedian = 1.62 TeV F Time since T* [ s ]
Emedian = 3.72 TeV
Energy flux [ erg cm-2 s -1 ]

PSF: (0.32, 0.61)

Energy flux [ erg cm-2 s-1 ]

5 PSF: (0.48, 1.20) 5

10 10

6 6
10 10

10 7 10 7

8 8
10 10
1 10 10 2
10 3
1 10 10 2 10 3
Time since T* [ s ] Time since T* [ s ]

Fig. 4. Energy flux light curve in the VHE band for six Nhit segments.
(A to F) The six segments are: [30, 33) (A), [33, 40) (B), [40, 63) (C), [63, 100)
(D), [100, 250) (E), and [250, +∞) (F). The median energy (Emedian) and
point spread function (PSF) are labeled in each panel; the PSF is given as 68
and 99% containment in degrees. The orange solid lines in (A) to (E) are
four-segment models fitted to the data. The overall fit in Fig. 3 is shown as
the dashed line in (F) for comparison. During fitting, two parameters—the
transition time from the rapid rise to the slow rise phase and the sharpness of
the transition from the slow decay to the steep decay phase—were fixed to the
values obtained from Fig. 3.

The light curve steepening at tb;2 ≃ 670 s medium, and we adopt the convention that
after T * cannot be a result of the KN scattering subscript numbers x indicate normalization
effect because the spectrum after the break does by a factor of 10x in centimeter-gram-second
not soften. The steepening that we observe re- (cgs) units. This reduces the required energy in
 to Eg;j ≡ Eg;iso q0 =2 ∼ 5:5  10
2 50
sembles a jet break, which occurs when the gamma rays
q0 2
Lorentz factor of a GRB jet drops to 1/q0, where Eg;iso;55 0:6° erg for GRB 221009A. This is con-
q0 is the initial half-opening angle of the jet. At sistent with the typical energy reservoir of GRB
this time, the jet edge becomes visible to the ob- jets (35). It has been suggested that GRBs could
server, causing a steepening in the light curve by have a quasi-universal beaming configuration—
t−3/4 for a homogeneous medium (30, 31). If the a structured jet with high anisotropy in its
lateral expansion of the jet is fast enough (32–34), angular distribution of the fireball energy about
a steeper decay is expected after the jet break the symmetry axis (36, 37). Under this assump-
Our identification of the TeV afterglow on-
set time can be used to estimate the initial
bulk Lorentz factor G0 of the jet. The peak
time (tpeak ~ 18 s after T *) of the light curve
corresponds to the deceleration time, when
most of the outflow energy is transferred to
the shocked external medium. The initial bulk
Lorentz factor is then
!1=8
3ð1 þ z Þ3 Ek
G0 ¼
32pnmp c5 tpeak
3

for the VHE emission (23). The early jet break tion of a universal jet structure for GRBs, a small  3=8
1=8 1=8 tpeak
of GRB 221009A implies a small q0, given by opening angle of GRB 221009A could imply ¼ 440 Ek;55 n0 ð2Þ
that the brightest core of a structured jet was 18 s
 
1=8 1=8 tb;2 3=8 visible from Earth before the break, explain-
q0 ∼ 0:6°Ek;55 n0 ð1Þ ing the high isotropic-equivalent energy of this where mp is the proton mass, and c is the
670 s
GRB. Combined with the low redshift of the speed of light. G0 is almost insensitive to Ek and
where Ek is the isotropic kinetic energy of the source, the small opening angle also explains n. The initial Lorentz factor of GRB 221009A
ejecta, n is number density of the circum-burst the high fluence (brightness) of this GRB. is consistent with the upper range of values

LHAASO Collaboration, Science 380, 1390–1396 (2023) 30 June 2023 4 of 6

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

10-4 A 10-3 B T +[5, 14]s ( 10) T +[100, 674]s ( 0.2)

T +[14, 22]s T +[674, 1774]s ( 0.3)
T +[22, 100]s ( 0.25)
10-5 10-4
Energy flux [erg cm -2 s-1]

Energy flux [erg cm -2 s-1]

10-6 10-5

10-7 10-6

10-8 10-7

0.3 - 5 TeV( 10)

10-9 10-8
0.3 - 10 keV
0.1 - 10 GeV( 0.1)
10-10 10-9
100 101 102 103 104 10-1 100 101
Time since T [s] Energy [TeV]

Fig. 5. Multiwavelength modeling of GRB 221009A. Data points indicate the
observations, and curves are the output of the model, which assumes afterglow
emission arising from external forward shocks, emitting synchrotron and SSC radiation.
The adopted parameters are: Ek = 1.5 × 1055 erg, G0 = 560, ee = 0.025, eB = 6 × 10−4,
p = 2.2, n = 0.4 cm−3, and q0 = 0.8°, although these are not a unique solution
(see text). (A) Light curves at keV (0.3 to 10 keV; black), GeV (0.1 to 10 GeV; red), and
TeV (0.3 to 5 TeV; blue) energies for the first ~104 s after the burst. (B) LHAASO-
WCDA spectra at five time intervals. The shaded regions indicate 1s ranges of
statistical uncertainties (inner band) and systematic uncertainties (outer bands).
The five colored lines indicate the output of the model for the five time intervals.

for previous GRBs with measured G0 inferred
from afterglow deceleration (38). This im-
plies that more energetic GRBs (in isotropic
energy) have a larger initial Lorentz factor for
the outflow.
Multiwavelength modeling
We performed multiwavelength modeling (23)
of the Swift-XRT, Fermi-LAT (39), and LHAASO
data assuming synchrotron plus SSC radiation,
within the framework of our interpretation of
the afterglow emission as arising from exter-
nal forward shocks. We use the full KN cross
section for the IC scattering and incorporate
two-photon annihilation (gg) absorption with-
in the source (23). For the jet break, we con-
sider only the geometric effect when the jet
edge is seen by the observer. Because the inner
core of the structured jet could be responsible
for the early-time afterglow emission, we con-
sider only the data for the first ~104 s after the
burst. The late-time afterglow emission could,
in principle, include additional contributions
from the outer wider components of the struc-
tured jet (40, 41).
We find a model that is consistent with
the broadband light curves and the LHAASO
spectra at various time intervals (Fig. 5) un-
der the following conditions (23): The ini-
tial isotropic-equivalent kinetic energy of
the forward shock is Ek ∼ 1:5  1055 erg, and
the initial bulk Lorentz factor is G0 ~ 560. The
electrons and magnetic field behind the shock
carry fractions ee ~ 0.025 and eB ~ 6 × 10−4,
respectively, of the dissipated energy of the
shock. The power-law index of the electron
distribution is p ~ 2.2, and the density of the
circum-burst medium is n ~ 0.4 cm−3. Because
there is degeneracy in the parameter space,
these parameters are not the only possible
choice.
In this model, the x-ray afterglow is produced
by synchrotron emission, and the TeV after-
glow is produced by SSC emission, whereas
the 0.1 to 10 GeV afterglow measured by
Fermi-LAT has contributions from both syn-
chrotron and SSC emission. We find that ee >
eB for the external shock, which is a neces-
sary condition for efficient SSC radiation. The
internal gg absorption is not strong, with op-
tical depth ≤1 for gamma rays of 5 TeV (23).
The half-opening angle in the model is q0 ~ 0.8°,
consistent with the analytical estimate. The
resulting model SSC spectrum is harder than
the observed spectrum at low energy during the
first two time intervals (Fig. 5B), similar to
the previously studied TeV spectrum of GRB
190829A (9). This discrepancy between the
model and observation could be a result of
additional contributions to the flux measured
by LHAASO-WCDA by some other emission
processes, such as external IC emission.
Limit on prompt TeV emission
In the optically thin synchrotron scenario for
prompt emission in GRBs (42, 43), the SSC
emission produced by the same population
of relativistic electrons generates GeV to TeV
gamma rays (26). Previous observations ob-
tained only loose upper limits on the TeV flux
during the prompt emission phase (44). GRB
221009A was observed by LHAASO during the
main burst phase, yielding a differential flux
limit of ~6 × 10−8 erg cm−2 s−1 at ~1 TeV from
T0 + 220 s to T0 + 230 s (Fig. 2). Compared with
the averaged MeV flux during the same period,
which is 3 × 10−3 erg cm−2 s−1 [in the 20 keV to
15 MeV range; we regard this as a conserva-
tive estimate because of the saturation effect
on the gamma-ray detectors (16)], the flux
ratio between TeV and MeV emission is R  ≡
FTeV =FMeV ≤ 2  105 . This is a stronger con-
straint than previous observations.
The internal gg absorption suppresses the
TeV flux during the prompt emission because
the radius of the internal shock or dissipa-
tion is much smaller than that of the external
shock. The internal dissipation radius can be
estimated if we know the variability timescale
of the prompt emission. Before the saturation
of the GBM data, the shortest variability time-
scale of GRB 221009A was tn ~ 0.082 s (39),
implying that internal dissipation occurs at
a distance from the source of Rin ∼ 2G20 ctv ¼
1015 cmðG0 =440Þ2 ðtn =0:082 sÞ. We estimate the
optical depth for TeV emission to be tgg ∼
1
190ðRin =1015 cmÞ (23), producing strong at-
tenuation of TeV photons, which could ex-
plain the very low flux ratio between TeV and
MeV emission.
Summary and conclusions
LHAASO observed the bright GRB 221009A at
the epochs covering both the prompt emission
phase and the early afterglow in the TeV band,
revealing the onset of afterglow emission in
the TeV band. We identify a jet break in the
light curve of GRB 221009A, indicating that
the opening angle of GRB 221009A is ~0.8°. Un-
der the assumption of a universal jet structure
for GRBs, this implies that the orientation of

LHAASO Collaboration, Science 380, 1390–1396 (2023) 30 June 2023 5 of 6

RES EARCH | R E S E A R C H A R T I C L E
this GRB was such that the brightest core of a
structured jet was visible from Earth, ex-
plaining the brightness of this GRB.
RE FE RENCES AND N OT ES
1. P. Kumar, B. Zhang, Phys. Rep. 561, 1–109 (2015).
2. R. Sari, A. A. Esin, Astrophys. J. 548, 787–799 (2001).
3. X. Y. Wang, Z. G. Dai, T. Lu, Astrophys. J. 556, 1010–1016
(2001).
4. B. Zhang, P. Mészáros, Astrophys. J. 559, 110–122 (2001).
5. Y. C. Zou, Y. Z. Fan, T. Piran, Mon. Not. R. Astron. Soc. 396,
1163–1170 (2009).
6. MAGIC Collaboration, Nature 575, 455–458 (2019).
7. P. Veres et al., Nature 575, 459–463 (2019).
8. H. Abdalla et al., Nature 575, 464–467 (2019).
9. H.E.S.S. Collaboration, Science 372, 1081–1085 (2021).
10. R. Alfaro et al., Astrophys. J. 843, 88 (2017).
11. B. Bartoli et al., Astrophys. J. 842, 31 (2017).
12. A. Albert et al., Astrophys. J. 936, 126 (2022).
13. P. Veres, E. Burns, E. Bissaldi, S. Lesage, O. Roberts, “GRB
221009A: Fermi GBM detection of an extraordinarily bright
GRB,” GCN Circular 32636 (NASA, 2022).
14. E. Bissaldi, N. Omodei, M. Kerr, “GRB 221009A or Swift J1913.1+1946:
Fermi-LAT detection,” GCN Circular 32637 (NASA, 2022).
15. S. Dichiara et al., “Swift J1913.1+1946 a new bright hard
X-ray and optical transient,” GCN Circular 32632 (NASA,
2022).
16. D. Frederiks et al., “Konus-Wind detection of GRB 221009A,”
GCN Circular 32668 (NASA, 2022).
17. A. de Ugarte Postigo et al., “GRB 221009A: Redshift from
X-shooter/VLT,” GCN Circular 32648 (NASA, 2022).
18. A. J. Castro-Tirado et al., “GRB 221009A: 10.4m GTC
spectroscopic redshift confirmation,” GCN Circular 32686
(NASA, 2022).
19. X.-H. Ma et al., Chinese Phys. C 46, 030001 (2022).
20. H. Ayala, “GRB 221009A: Upper limits from HAWC 8 hours
after trigger,” GCN Circular 32683 (NASA, 2022).
21. J. C. Liu et al., “GRB 221009A: HEBS detection,” GCN Circular
32751 (NASA, 2022).
22. A. Saldana-Lopez et al., Mon. Not. R. Astron. Soc. 507,
5144–5160 (2021).
23. Materials and methods are available as supplementary
materials.
24. S. Kobayashi, B. Zhang, Astrophys. J. 655, 973–979 (2007).
25. D. Lazzati, M. Begelman, Astrophys. J. 641, 972–977
(2006).
26. Ž. Bošnjak, F. Daigne, G. Dubus, Astron. Astrophys. 498,
677–703 (2009).
27. E. Derishev, T. Piran, Astrophys. J. Lett. 880, L27 (2019).
LHAASO Collaboration, Science 380, 1390–1396 (2023)
28. X. Y. Wang, R. Y. Liu, H. M. Zhang, S. Q. Xi, B. Zhang,
Astrophys. J. 884, 117 (2019).
29. Y.-Z. Fan, T. Piran, R. Narayan, D.-M. Wei, Mon. Not. R. Astron.
Soc. 384, 1483–1501 (2008).
30. P. Mészáros, M. J. Rees, Mon. Not. R. Astron. Soc. 306,
L39–L43 (1999).
31. Z. G. Dai, L. J. Gou, Astrophys. J. 552, 72–80 (2001).
32. J. E. Rhoads, Astrophys. J. 525, 737–749 (1999).
33. R. Sari, T. Piran, J. P. Halpern, Astrophys. J. 519, L17–L20
(1999).
34. Z. G. Dai, K. S. Cheng, Astrophys. J. 558, L109–L112 (2001).
35. D. A. Frail et al., Astrophys. J. 562, L55–L58 (2001).
36. E. Rossi, D. Lazzati, M. J. Rees, Mon. Not. R. Astron. Soc. 332,
945–950 (2002).
37. B. Zhang, P. Mészáros, Astrophys. J. 571, 876–879 (2002).
38. E.-W. Liang et al., Astrophys. J. 725, 2209 (2010).
39. R.-Y. Liu, H.-M. Zhang, X.-Y. Wang, Astrophys. J. Lett. 943, L2
(2023).
40. F. Peng, A. Königl, J. Granot, Astrophys. J. 626, 966–977
(2005).
41. Y. Sato, K. Obayashi, R. Yamazaki, K. Murase, Y. Ohira,
Mon. Not. R. Astron. Soc. 504, 5647–5655 (2021).
42. P. Mészáros, M. J. Rees, H. Papathanassiou, Astrophys. J. 432,
181 (1994).
43. Z. L. Uhm, B. Zhang, Nat. Phys. 10, 351–356 (2014).
44. J. Wood, in Proceedings of 35th International Cosmic Ray
Conference (ICRC2017), Y.-S. Kwak, H. S. Lee, S. Oh, I. H. Park,
Eds. (Proceedings of Science, 2017), p. 619.
ACKN OWLED GMEN TS
The LHAASO observatory, including its detector system, was
designed and constructed by the LHAASO project team. Since its
completion, it has been maintained by the LHAASO operating
team. We extend our gratitude to all members of these two teams,
especially those who work year-round at the LHAASO site, located
more than 4400 m above sea level. Their tireless efforts ensure
that the detector and all its components as well as the electricity
power supply operate smoothly. Funding: This work was supported
in China by the National Key R&D program of China under
grants 2018YFA0404201, 2018YFA0404202, 2018YFA0404203,
and 2018YFA0404204 and by NSFC under grants U1831208,
11833003, 12022502, 12121003, U2031105, 12005246, 12173039,
1221101008, and 12203022. Support was provided by the National
SKA Program of China under grant 2020SKA0120300, the Chinese
Academy of Sciences under grant YSBR-061, the Department
of Science and Technology of Sichuan Province under grant
2021YFSY0030, the Natural Science Foundation of Jiangsu
Province under grant BK20220757, and the Chengdu Management
Committee of Tianfu New Area for research with LHAASO data.
In Thailand, support was provided by the National Science and
30 June 2023
Technology Development Agency (NSTDA) and the National
Research Council of Thailand (NRCT) under the High-Potential
Research Team Grant Program (N42A650868). Author
contributions: Z. G. Yao and X. Y. Wang led the data analysis and
interpretation, respectively. S. C. Hu (supervised by Z. G. Yao)
performed spectrum analysis. Y. Huang (supervised by C. Liu and
Z. G. Yao) calculated the light curve. H. C. Li, C. Liu, and Z. G. Yao
performed light curve fitting. M. Zha coordinated the entire
data analysis, provided reconstruction and simulation data, and
participated in spectrum analysis. H. Zhou and Zhen Wang
provided cross checks. Z. G. Dai and B. Zhang contributed to
theoretical interpretation and manuscript structure. H. M. Zhang
performed multiwavelength data analysis and contributed to
interpretation. J. H. Zheng (supervised by X. Y. Wang) modeled the
multiwavelength data, and R. Y. Liu contributed to the modeling
of multiwavelength data. Zhen Cao is the spokesperson of the
LHAASO Collaboration and the principal investigator of the
LHAASO project and coordinated the working group for this study
along with the corresponding authors. Internal reviews were
provided by the Physics Coordination Committee of the LHAASO
Collaboration led by S. Z. Chen and the Publication Committee of
the LHAASO Collaboration led by S. M. Liu and D. della Volpe.
All other authors participated in data analysis, including detector
calibration, data processing, event reconstruction, data quality
check, and simulations, and provided comments on the
manuscript. Competing interests: There are no competing
interests to declare. Data and materials availability: Data and
software to reproduce the results are available at https://www.
nhepsdc.cn/resource/astro/lhaaso/paper.Science2023.adg9328/.
This includes the observed data and detector acceptance
parameters used as input for our analysis, the resulting light curve
data points (shown in Figs. 1, 3, 4, and 5), spectrum data points
(shown in Fig. 2 and fig. S4), the code we used for the light curve
fitting, and machine-readable versions of tables S1 to S3. License
information: Copyright © 2023 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.adg9328
LHAASO Collaboration Authors
Materials and Methods
Figs. S1 to S10
Tables S1 to S3
References (45–58)
Submitted 31 January 2023; accepted 25 May 2023
Published online 8 June 2023
10.1126/science.adg9328
6 of 6
 online @sciencecareers.org
Find your next job at ScienceCareers.org
 WORKING LIFE
By Zoe Weiss

What I owe my mentor

S
oon after I took my first steps into a research lab as a bright-eyed, first-year college student,
things went awry. I left all the enzymes at room temperature overnight, not knowing they
needed to be stored frozen. This setback could have discouraged me from performing solo ex-
periments. But it didn’t, thanks to the kindness of my mentor. Saurja, who was a postdoc, didn’t
respond with anger or disappointment after I ruined all the newly ordered enzyme stocks. In-
stead, he patiently guided me through the principles of enzymatic activity and how they vary
with temperature, turning my mistake into a teachable moment.

I went on to work with Saurja IT’S OK TO MAKE MISTAKES. From

for four more years, helping him
with his experiments and also
carrying out my own. After grad-
uating with plans to pursue re-
search, I reflected on all the ways
in which he nurtured my abili-
ties and set me up for success in
science. Here are the lessons he
taught me.
UNDERSTAND THE BIG PICTURE. In
my first few weeks in the lab, I
shadowed Saurja, following him
around from the gel running rigs
to chromatography machines to
group meetings—always with a
notebook and pen in hand. He “I wish all undergraduate
struck me as an energetic post-
doc whose brain seemed to spin
researchers could be so lucky.”
at twice the usual speed, gener-
ating outside-the-box questions. He explained how RNA
could act like an enzyme if it folds in specific ways, much
like proteins do. I had never considered such an idea; it
seemed to break the rules I learned in biology classes. I
found his enthusiasm and curiosity infectious, and I ap-
preciated that he took the time to explain to me what we
were doing and why. By understanding the overarching
question, I was better able to plow through the day-to-day
disappointments of failed experiments and troubleshoot
problems on my own.
ASK QUESTIONS. Before starting any experiment, Saurja
would draw out a schematic depicting what we’d be doing
and what our expected outcomes were, pausing after to
the start, Saurja showed a level
of trust in my experimental skills
that I did not always think I de-
served. He handed me the pipette
with total freedom to mess up
even the most important steps in
an experiment. Even when I used
10,000 times more of an expensive
reagent than I should have, Saurja
didn’t roll his eyes or get angry. In-
stead, he told me stories of his own
mistakes—for instance, how on his
first day as a postdoc in our lab, he
broke a gel plate and ruined an ex-
periment. I learned to reframe fail-
ure as motivation to learn from the
situation and do better next time.
SHARE YOUR SKILLS. After a few
months in the lab, Saurja and I
started a project together that required us to write com-
puter code. That was something I had experience in, and
it felt great to have the opportunity to teach him. Over the
years, he’s treated me like a colleague and has always been
open to a two-way exchange of information, which has led
us to develop many successful projects. We even published
a paper together—just the two of us—which has given me
a tremendous amount of confidence as a young scientist.
As I look back on my time working with Saurja, I’m
thankful I had a mentor who instilled in me an unwavering
desire to learn and who gave me the freedom to be creative
and contribute my own skills and ideas. I wish all under-
graduate researchers could be so lucky. I encourage other
mentors to tap into their own inner excitement for research
ILLUSTRATION: ROBERT NEUBECKER

see whether I had any questions. At first, I was afraid to
sound stupid. But he was always patient and didn’t make
me feel I should have mastered the concept already. That
gave me confidence to ask more questions and get help
tackling concepts that seemed alien and dense at first.
and pass it on to others. Because undergraduate researchers
aren’t just an extra set of hands in the lab—we’re also fledg-
ling scientists who need nurturing and encouragement. j
Zoe Weiss recently completed a bachelor’s degree at Harvard University.

1398 30 JUNE 2023 • VOL 380 ISSUE 6652 science.org SCIENCE

 Post date 30 June 2023

Editorial Expression of Concern

On 30 September 2021, Science published the Research Article “Light-induced mobile factors from shoots
regulate rhizobium-triggered soybean root nodulation” by Tao Wang et al. (1). The editors have been made
aware that data presented in Fig. 5 assessed GmNSP1 expression rather than GmNIN expression. We are
alerting readers to these concerns while the authors provide additional data to correct the manuscript.

H. Holden Thorp
Editor-in-Chief

R E F EREN C ES AND NOTES

1. T. Wang et al., Science 374, 65 (2021).

10.1126/science.adj3306

SCIENCE science.org EDITORIAL EXPRESSION OF CONCERN POST DATE • 30 JUNE 2023 1