Science - 4 11 2022

Berlin shows how biodiversity Neutrinos from an active Eppendorf & Science Prize for neural
can flourish in cities p. 466 galaxy pp. 474 & 538 underpinnings of aggression p. 484
$15
4 NOVEMBER 2022
SPECIAL ISSUE
science.org
BRAIN
CONNECTIVITY
The 2023 AAAS Annual Meeting will convene in-person
in Washington, D.C. for the first time in two years.
Virtual access to the meeting will be available as well.
Register today and join us March 2-5, 2023!
aaas.org/meetings | #AAASmtg
BE A PART OF THE CONVERSATION
Diversity, Equity and Inclusion Research Funding and Infrastructure
Science Communication Artificial Intelligence
STEM Education Information Technology and Security
International S&T Collaboration Science and Values
Social and Behavioral Sciences Global and Public Health
Science Policy Climate Change
and more...
science.org/journal/sciimmunol
READY TO
PUT THE
SPOTLIGHT
ON YOUR
RESEARCH?
Submit your research:
cts.ScienceMag.org
Twitter: @SciImmunology
Facebook: @ScienceImmunology
Advertorial
(Left) Junsuk Rho, (Far right) Soojin Park
The transformation of theory into practice

For researchers working in the applied sciences, the Pohang University of Science Out of the lab, into the market
and Technology in South Korea, commonly called POSTECH, acts as an academic Currently, Rho is focused on bringing the potential of metamaterials to the mass
springboard for bringing ideas to life. After opening its doors in 1986, the university market.
quickly established a reputation for excellence and was rated third in the Times POSTECH was founded and funded by the Pohang Steel Company (POSCO), one
Higher Education’s World’s Best Small University Rankings from 2019 to 2021. of the world’s leading steel-making companies, with a view to supporting cutting-
edge scientific and technological research. In 2021, POSCO awarded Rho’s lab USD
Junsuk Rho completed his Ph.D. in the United States at the University of California, 10 million to support a 10-year project to pilot a line of metamaterials that have the
Berkeley, in 2013. Now chair professor of mechanical and chemical engineering potential for mass production.
at POSTECH, he explains why he decided to continue his career researching “This is the first time in Korea that a company has invested such a large sum of
metamaterials in South Korea. money in a single lab,” says Rho.
“Everything at POSTECH, from the school’s policies to the layout of the campus, At the same time, the Research Institute of Industrial Science and Technology
is designed to improve the research atmosphere,” he says. (RIST), also located on the POSTECH campus, has invested USD 2 million in a high-
Attracted by the university’s highly collaborative and interdisciplinary research speed electron-beam lithography system that will further support the translation
ethic, which includes partnerships with Germany’s Max Planck Society for of Rho’s work.
the Advancement of Science (1), Rho leads a world-class team that has made “We’re competing with other leading groups from places such as Harvard
considerable advances in metamaterials, from developing optical materials and Stanford, but with our high levels of funding, strong research environment,
for ultracompact, high-resolution 3D holography and ultrathin optical lenses and excellent students, we are confident that we are one of the main hubs for
and extended reality (XR) devices, to pushing the boundaries of fabrication metamaterials research,” he says.
techniques. Current metamaterials that work at visible wavelengths are very expensive
For example, in 2018, Rho’s team created optically active, chiral gold to make using conventional nanofabrication techniques. One of Rho’s biggest PHOTOS: PROVIDED BY POSTECH
nanoparticles using amino acids and peptides. The nanoparticles change color challenges is to develop an economically viable way to mass produce
according to the degree of light polarization (2). metamaterials on a large scale. His recent success has come in the form of new
“The potential applications include active color displays, holography, techniques for fabricating nanostructures, such as a controllable lithographic
biochemical sensors, and all-angle negative refractive index materials,” says Rho. method for making single-nanometer sharp bowtie nanoantennas inspired by the
Produced by the Science
toppling of dominos (3), as well as single-step production of metasurfaces using a Charged with challenges
nanoparticle-embedded resin (4). “The former allows for the extreme confinement For battery materials to be commercialized, numerous scientific obstacles must
of light, while the latter provides a method of repeatedly printing metasurfaces at be overcome, says Park.
will with a single master stamp,” he explains. “We’re wrestling with fundamental questions: Whether low-cost, mass-
“This opens up the possibility for the mass production of metasurfaces, production is possible. Whether we can develop batteries with a longer charged
especially if such techniques can be combined with other printing methods such life. And whether we can genuinely make them completely safe,” he says.
as roll-to-roll printing,” he says. Silicon is a promising anode material for high-capacity batteries that can
realize 10 times the capacity per weight compared to graphite, he explains.
Battery-powered future However, its capacity degrades during the charging and discharging process.
Chair professor Soojin Park leads the Energy Materials Laboratory at POSTECH “In order for the characteristics of materials being developed at the laboratory
(5). His research focuses on battery materials and systems. For Park, the level to match the level desired by battery cell manufacturers, many problems
opportunities afforded by developing safer, longer-life batteries are both must be addressed,” he says. “Active joint research is necessary to narrow the
scientifically exciting and socially vital for a sustainable future. Park believes that distance between the laboratory level and the battery makers.”
battery research is key to fulfilling the energy gap that will necessarily need to be To bridge the gap between research and manufacturing, POSTECH has built
filled as the world weans itself off fossil fuels. He is also keen to explore the role a dry room to manufacture and evaluate large battery cells. Additionally, the
battery materials can play in revolutionizing health care, as we develop devices university has made the development of next-generation battery materials and
that can monitor our physical states in real time. systems a research priority and is in the process of building up its talent base by
Currently, rechargeable lithium-ion batteries are widely used in our daily lives, hiring experts in anode and cathode materials, battery systems, electrolytes, and
for example, in powering electric cars. However, these battery systems have advanced analysis.
drawbacks, such as being inherently flammable and having slow charging times.
These batteries are not currently powerful, safe, or flexible enough to enable Hub for real-life breakthroughs
society to switch to battery power, he explains. POSTECH’s campus offers researchers the opportunity to share their passion for
“We need to develop battery materials that can last a long time on a single futuristic technologies with a wide community of like minds and to benefit from
charge for electric vehicles; safe and flexible power sources for electronic devices a highly ambitious research infrastructure. “Consistently strong levels of funding,
that can be attached to the skin; and light, powerful batteries that can power the development of centers for joint research with manufacturers, and a focus on
drones and small airplanes,” he says. growing our research talent base are laying the foundation for the university to
become a key hub for next-generation research,” says Park.
Battery breakthroughs
Right now, Park’s research group is conducting joint research with South References
Korean battery material makers and battery cell manufacturers, such as 1. “Max Planck to Set Up Two Centres in South Korea” (2010); available at https://
sciencebusiness.net/news/68020/Max-Planck-to-set-up-two-centres-in-South-Korea.
LG Energy Solution, Samsung SDI, the Hyundai Motor Company, and POSCO.
2. H.-E. Lee et al., Nature 556, 360–365 (2018), https://www.nature.com/articles/s41586-
His team is jointly developing battery anode materials, functional separator 018-0034-1.
membranes, and functional polymeric binders for electric vehicles. Each of 3. I. Kim et al., Mater. Today 39, 89–97 (2020), https://doi.org/10.1016/j.mattod.2020.06.002.
4. G. Yoon, K. Kim, D. Huh, H. Lee, J. Rho, Nat. Commun. 11, 2268 (2020), https://www.nature.
these materials is intended to meet the needs of future batteries, with an com/articles/s41467-020-16136-5.
emphasis on fast charging capabilities, high output, high energy density, and 5. “Spark Lab. Polymer Based Energy Materials Laboratory” (2019), https://www.spark-
high stability. postech.com.
6. C. Hwang et al., Adv. Mater. 30, 1705445 (2018), https://onlinelibrary.wiley.com/doi.org/
Park also expects his work to expand into developing energy-dense, flexible abs/10.1002/adma.201705445.
batteries that can be combined with flexible devices. Recently, his team 7. W.-J. Song et al., Adv. Energy Mater. 8, 1702478 (2018), https://onlinelibrary.wiley.com/doi/
successfully developed high-capacity silicon anode materials that realize 10 times abs/10.1002/aenm.201702478.
more capacity than conventional graphite, which is the current commercialized Sponsored by
anode material for batteries. His research group has also developed a flexible
battery with high energy density that can be crumpled like paper (6). Furthermore,
Park’s group has developed a stretchable battery that can be operated while
being stretched by more than 50% (7).
CONTENTS
4 N O V E M B E R 2 0 2 2 • VO LU M E 3 7 8 • I S S U E 6 6 1 9
Wild boars have adapted

to living in Berlin. 466
NEWS
FEATURES 475 Molecular diversity of astrocytes
466 Urban oasis The tissue environment influences
In Berlin, pioneering research into astrocyte form and function in health
urban ecology has found surprising and disease
IN BRIEF biodiversity in the city’s green spaces By K. T. Baldwin
456 News at a glance By G. Popkin BRAIN CONNECTIVITY SECTION p. 486;
RESEARCH ARTICLE p. 514
PODCAST
IN DEPTH
POLICY FORUM
458 Should Webb telescope’s data be
open to all?
NASA plans to end policy of giving some
INSIGHTS 477 Protect global values of the
Southern Ocean ecosystem
observers 1 year’s exclusive access to Climate change and fishing present
data By D. Clery PERSPECTIVES dual threats By C. M. Brooks et al.
459 Teeth record pneumonia—and racism 470 Arctic wildfires at a warming

BOOKS ET AL.
Research raises standards for working with threshold
anthropological collections By R. Pérez Ortega Bigger wildfires in the Siberian Arctic signal 480 Where telemedicine always
release of more carbon to the atmosphere falls short
461 A cleaner way to mine clean-energy By E. Post and M. C. Mack Efforts to innovate our way to better health
mainstays RESEARCH ARTICLE p. 532 consistently fail to deliver on equity promises
An electric field can reduce the need for By V. Rampton
polluting chemicals to extract rare earth 471 Gene therapy for epilepsy
elements By D. Normile On-demand inhibition of neuronal activity 481 Reifying racism in medicine
reduced spontaneous seizures in mice False beliefs about racial differences
462 Prominent astronomer barred after By K. Staley have a long history in physician training
complaints RESEARCH ARTICLE p. 523 programs By S. Seth
PHOTO: GREGOR FISCHER/PICTURE ALLIANCE/GETTY IMAGES

Tim de Zeeuw had a pattern of “extremely
unacceptable” behavior, Leiden University 473 An optical interface for quantum LETTERS
says By C. O’Grady networks 482 Pakistan’s floods flow from
A silicon atom embedded in diamond can be climate injustice
463 NSF to end cost-sharing requirement
for some grants entangled with a photon By D. Gangloff By M. A. Waqas
REPORT p. 557
Instrumentation and teacher training
programs will no longer ask institutions to 482 Restore Pakistan’s forest
provide matching funds By J. Mervis 474 Neutrinos unveil hidden galactic to prevent floods
activities By A. Ali
464 After Lula’s win, ‘a huge relief!’ An obscured supermassive black hole
Many in Brazil’s scientific community may be producing high-energy cosmic 483 Predicting Pakistan’s
abhorred Jair Bolsonaro. But his successor neutrinos By K. Murase next flood
faces major challenges By S. Moutinho RESEARCH ARTICLE p. 538 By Y. Yao and Z. A. Khan
450 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE

PRIZE ESSAY 538 Neutrino astrophysics
484 Boiling over Evidence for neutrino emission from
An emergent encoding of aggressive the nearby active galaxy NGC 1068
motivation in neurons of the hypothalamus IceCube Collaboration
PERSPECTIVE p. 474
By A. Kennedy
543 Plant science

RESEARCH The control of carpel determinacy
pathway leads to sex determination in
cucurbits S. Zhang et al.
IN BRIEF REPORTS
511 From Science and other journals 549 Structural biology SPECIAL SECTION
CREDITS (LEFT TO RIGHT): IMAGE: NASA/JPL-CALTECH; IMAGE: MARKUS AXER AND KATRIN AMUNTS/INM-1; FORSCHUNGSZENTRUM JÜLICH AND ROXANA KOOIJMANS/NETHERLANDS INSTITUTE FOR NEUROSCIENCE/HUMAN BRAIN PROJECT
Molecular glue CELMoD compounds are

RESEARCH ARTICLES regulators of cereblon conformation
E. R. Watson et al.
Brain connectivity
514 Cellular neuroscience
Molecular basis of astrocyte diversity and INTRODUCTION
554 Volcanic plumes
morphology across the CNS in health and The January 2022 eruption of Hunga 486 No neuron is an island
disease F. Endo et al. Tonga-Hunga Ha’apai volcano reached the
RESEARCH ARTICLE SUMMARY; FOR FULL TEXT: mesosphere S. R. Proud et al. REVIEWS
DOI.ORG/10.1126/SCIENCE.ADC9020 488 Atlas-based data integration
PERSPECTIVE p. 475; BRAIN CONNECTIVITY
557 Quantum technology for mapping the connections
SECTION p. 486
Robust multi-qubit quantum network and architecture of the brain
node with integrated error detection T. B. Leergaard and J. G. Bjaalie
515 Neuroscience P.-J. Stas et al.
Dual-polarity voltage imaging of the PERSPECTIVE p. 473 493 Solving brain circuit function
concurrent dynamics of multiple neuron types and dysfunction with computational
M. Kannan et al. 560 Monkeypox modeling and optogenetic fMRI
RESEARCH ARTICLE SUMMARY; FOR FULL TEXT: Multiple lineages of monkeypox virus J. H. Lee et al.
DOI.ORG/10.1126/SCIENCE.ABM8797 detected in the United States, 2021–2022
C. M. Gigante et al. 500 Scale matters:
516 Synthetic biology The nested human connectome
Refactored genetic codes enable bidirectional M. Axer and K. Amunts
genetic isolation J. F. Zürcher et al.
DEPARTMENTS
505 The emergent properties of the
523 Neuroscience 455 Editorial connected brain
On-demand cell-autonomous gene therapy Measuring emissions to manage M. Thiebaut de Schotten and S. J. Forkel
for brain circuit disorders Y. Qiu et al. emissions By A. Gore
PERSPECTIVE p. 471; PODCAST ON THE COVER
570 Working Life
532 Arctic wildfires Our labor isn’t free By I. Bennett The human brain’s primary interhemi-
Unprecedented fire activity above the Arctic spheric connections have different axonal
calibers. This image, from the middle of
Circle linked to rising temperatures
the brain, represents a cross section of
A. Descals et al. Science Staff ..............................................452 the axons, with warmer colors (yellow)
PERSPECTIVE p. 470 Science Careers ........................................ 566 indicating thicker axons. These finely
attuned connections allow for the perfect
orchestration of the brain’s symphony.
See the special
section beginning
on page 486. Image:
Michel Thiebaut de
Schotten, Stephanie
Forkel/Neurofunctional
Imaging Group, CNRS,
CEA University of
Bordeaux, France
474 & 538

SEE ALSO PERSPECTIVE p. 475; RESEARCH
ARTICLE p. 514
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 © 2022 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): $2212; Foreign postage extra: Air assist delivery: $98. 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 4 NOVEMEBER 2022 • VOL 378 ISSUE 6619 451

Editor-in-Chief Holden Thorp, [email protected] BOARD OF REVIEWING EDITORS (Statistics board members indicated with S)
Executive Editor Valda Vinson Erin Adams, U. of Chicago Louise Harra, U. Coll. London Samuel Pfaff, Salk Inst.
Editor, Research Jake S. Yeston Editor, Insights Lisa D. Chong Managing Editor Lauren Kmec Takuzo Aida, U. of Tokyo Carl-Philipp Heisenberg, Julie Pfeiffer,
DEPUTY EDITORS Gemma Alderton (UK),Stella M. Hurtley (UK), Phillip D. Szuromi, Sacha Vignieri SR. EDITORS Caroline Ash (UK), Leslie Aiello, Wenner-Gren Fdn. IST Austria UT Southwestern Med. Ctr.
Michael A. Funk, Brent Grocholski, Pamela J. Hines, Di Jiang, Priscilla N. Kelly, Marc S. Lavine (Canada), Mattia Maroso, Deji Akinwande, UT Austin Janet G. Hering, Eawag Philip Phillips, UIUC
Yevgeniya Nusinovich, Ian S. Osborne (UK), L. Bryan Ray, Seth Thomas Scanlon (UK), H. Jesse Smith, Keith T. Smith (UK), Judith Allen, U. of Manchester Christoph Hess, Matthieu Piel, Inst. Curie
Jelena Stajic, Peter Stern (UK), Valerie B. Thompson, Brad Wible, Yuen Yiu ASSOCIATE EDITORS Bianca Lopez, Corinne Simonti, Marcella Alsan, Harvard U. U. of Basel & U. of Cambridge Kathrin Plath, UCLA
James Analytis, UC Berkeley Heather Hickman, NIAID, NIH Martin Plenio, Ulm U.
Yury V. Suleymanov, Ekeoma Uzogara LETTERS EDITOR Jennifer Sills LEAD CONTENT PRODUCTION EDITORS Chris Filiatreau, Harry Jach
Paola Arlotta, Harvard U. Hans Hilgenkamp, U. of Twente Katherine Pollard, UCSF
SR. CONTENT PRODUCTION EDITOR Amelia Beyna CONTENT PRODUCTION EDITORS Julia Haber-Katris, Nida Masiulis, Abigail Shashikanth,
Delia Baldassarri, NYU Janneke Hille Ris Lambers, Elvira Poloczanska,
Suzanne M. White SR. EDITORIAL MANAGERS Carolyn Kyle, Beverly Shields SR. PROGRAM ASSOCIATE Maryrose Madrid EDITORIAL
Nenad Ban, ETH Zürich ETH Zürich Alfred-Wegener-Inst.
ASSOCIATE Joi S. Granger SR. EDITORIAL COORDINATORS Aneera Dobbins, Jeffrey Hearn, Lisa Johnson, Shannon McMahon,
Christopher Barratt, Kai-Uwe Hinrichs, U. of Bremen Julia Pongratz,
Jerry Richardson, Hilary Stewart (UK), Alice Whaley (UK), Anita Wynn EDITORIAL COORDINATORS Maura Byrne, Alexander Kief,
U. of Dundee Deirdre Hollingsworth, Ludwig Maximilians U.
Ronmel Navas, Isabel Schnaidt, Qiyam Stewart, Brian White RESEARCH & DATA ANALYST Jessica L. Slater ASI DIRECTOR, OPERATIONS
Nandita Basu, U. of Waterloo U. of Oxford Philippe Poulin, CNRS
Janet Clements (UK) ASI SR. OFFICE ADMINISTRATOR Jessica Waldock (UK)
Franz Bauer, Randall Hulet, Rice U. Lei Stanley Qi, Stanford U.
News Editor Tim Appenzeller Pontificia U. Católica de Chile Auke Ijspeert, EPFL Trevor Robbins, U. of Cambridge
NEWS MANAGING EDITOR John Travis INTERNATIONAL EDITOR Martin Enserink DEPUTY NEWS EDITORS Shraddha Chakradhar, Elizabeth Ray H. Baughman, UT Dallas Gwyneth Ingram, ENS Lyon Joeri Rogelj, Imperial Coll. London
Culotta, Lila Guterman, David Grimm, Eric Hand (Europe), David Malakoff SR. CORRESPONDENTS Daniel Clery (UK), Jon Cohen, Jeffrey Carlo Beenakker, Leiden U. Darrell Irvine, MIT Amy Rosenzweig, Northwestern U.
Mervis, Elizabeth Pennisi ASSOCIATE EDITORS Jeffrey Brainard, Michael Price, Kelly Servick NEWS REPORTERS Adrian Cho, Jennifer Yasmine Belkaid, NIAID, NIH Akiko Iwasaki, Yale U. Mike Ryan, UT Austin
Couzin-Frankel, Jocelyn Kaiser, Rodrigo Pérez Ortega (Mexico City), Robert F. Service, Erik Stokstad, Paul Voosen, Meredith Wadman Philip Benfey, Duke U. Stephen Jackson, Miquel Salmeron,
INTERNS Zack Savitsky, Viviana Flores, Katherine Irving CONTRIBUTING CORRESPONDENTS Warren Cornwall, Andrew Curry (Berlin), Kiros T. Berhane, Columbia U. USGS & U. of Arizona Lawrence Berkeley Nat. Lab
Joseph J. Berry, NREL Erich Jarvis, Rockefeller U. Nitin Samarth, Penn State U.
Ann Gibbons, Sam Kean, Eli Kintisch, Kai Kupferschmidt (Berlin), Andrew Lawler, Mitch Leslie, Eliot Marshall, Virginia Morell, Dennis
Alessandra Biffi, Harvard Med. Peter Jonas, IST Austria Erica Ollmann Saphire,
Normile (Tokyo), Elisabeth Pain (Careers), Charles Piller, Gabriel Popkin, Michael Price, Joshua Sokol, Richard Stone, Emily
Chris Bowler, Johanna Joyce, U. de Lausanne La Jolla Inst.
Underwood, Gretchen Vogel (Berlin), Lizzie Wade (Mexico City) CAREERS Rachel Bernstein (Editor), Katie Langin (Associate Editor)
École Normale Supérieure Matt Kaeberlein, U. of Wash. Joachim Saur, U. zu Köln
COPY EDITORS Julia Cole (Senior Copy Editor), Morgan Everett, Cyra Master (Copy Chief) ADMINISTRATIVE SUPPORT Meagan Weiland
Ian Boyd, U. of St. Andrews William Kaelin Jr., Alexander Schier, Harvard U.
Creative Director Beth Rakouskas Malcolm Brenner, Baylor Coll. Dana-Farber Cancer Inst. Wolfram Schlenker, Columbia U.
DESIGN MANAGING EDITORS Marcy Atarod, Chrystal Smith GRAPHICS MANAGING EDITOR Chris Bickel MULTIMEDIA MANAGING PRODUCER of Med. Daniel Kammen, UC Berkeley Susannah Scott,
Kevin McLean WEB CONTENT STRATEGY MANAGER Kara Estelle-Powers PHOTOGRAPHY MANAGING EDITOR Emily Petersen Emily Brodsky, UC Santa Cruz Kisuk Kang, Seoul Nat. U. UC Santa Barbara
DESIGNER Christina Aycock SENIOR SCIENTIFIC ILLUSTRATOR Valerie Altounian SCIENTIFIC ILLUSTRATORS Austin Fisher, Kellie Holoski, Ron Brookmeyer, UCLA (S) Sabine Kastner, Princeton U. Anuj Shah, U. of Chicago
Ashley Mastin INTERACTIVE GRAPHICS EDITOR Kelly Franklin SENIOR GRAPHICS SPECIALISTS Holly Bishop, Nathalie Cary SENIOR PHOTO Christian Büchel, UKE Hamburg V. Narry Kim, Seoul Nat. U. Vladimir Shalaev, Purdue U.
EDITOR Charles Borst PHOTO EDITOR Kaitlyn Dolan SENIOR PODCAST PRODUCER Sarah Crespi VIDEO PRODUCER Meagan Cantwell Dennis Burton, Scripps Res. Robert Kingston, Harvard Med. Jie Shan, Cornell U.
SOCIAL MEDIA STRATEGIST Jessica Hubbard SOCIAL MEDIA PRODUCER Sabrina Jenkins WEB DESIGNER Jennie Pajerowski Carter Tribley Butts, UC Irvine Nancy Knowlton, Beth Shapiro, UC Santa Cruz
György Buzsáki, Smithsonian Institution Jay Shendure, U. of Wash.
NYU School of Med. Etienne Koechlin,
Chief Executive Officer and Executive Publisher Sudip Parikh Mariana Byndloss, École Normale Supérieure
Steve Sherwood,
U. of New South Wales
Publisher, Science Family of Journals Bill Moran Vanderbilt U. Med. Ctr. Alex L. Kolodkin, Johns Hopkins U.
Brian Shoichet, UCSF
DIRECTOR, BUSINESS SYSTEMS AND FINANCIAL ANALYSIS Randy Yi DIRECTOR, BUSINESS OPERATIONS & ANALYSIS Eric Knott DIRECTOR OF Annmarie Carlton, UC Irvine LaShanda Korley, U. of Delaware
Robert Siliciano,
ANALYTICS Enrique Gonzales MANAGER, BUSINESS OPERATIONS Jessica Tierney MANAGER, BUSINESS ANALYSIS Cory Lipman Simon Cauchemez, Inst. Pasteur Julija Krupic, U. of Cambridge
JHU School of Med.
BUSINESS ANALYSTS Kurt Ennis, Maggie Clark FINANCIAL ANALYST Isacco Fusi BUSINESS OPERATIONS ADMINISTRATOR Taylor Fisher Ling-Ling Chen, SIBCB, CAS Paul Kubes, U. of Calgary
Lucia Sivilotti, U. Coll. London
SENIOR PRODUCTION MANAGER Jason Hillman SENIOR MANAGER, PUBLISHING AND CONTENT SYSTEMS Marcus Spiegler Ib Chorkendorff, Denmark TU Chris Kuzawa, Northwestern U.
Emma Slack, ETH Zürich & U.
CONTENT OPERATIONS MANAGER Rebecca Doshi SENIOR CONTENT & PUBLISHING SYSTEMS SPECIALIST Jacob Hedrick Karlene Cimprich, Stanford U. Laura Lackner, Northwestern U.
of Oxford
SENIOR PRODUCTION SPECIALIST Kristin Wowk PRODUCTION SPECIALISTS Kelsey Cartelli, Audrey Diggs DIGITAL PRODUCTION MANAGER James J. Collins, MIT Gabriel Lander, Scripps Res. (S)
Richard Smith, UNC (S)
Lisa Stanford CONTENT SPECIALIST Kimberley Oster ADVERTISING PRODUCTION OPERATIONS MANAGER Deborah Tompkins DESIGNER, Robert Cook-Deegan, Mitchell A. Lazar, UPenn
John Speakman, U. of Aberdeen
CUSTOM PUBLISHING Jeremy Huntsinger SR. TRAFFIC ASSOCIATE Christine Hall SPECIAL PROJECTS ASSOCIATE Sarah Dhere Arizona State U. Hedwig Lee, Duke U.
Tara Spires-Jones, U. of Edinburgh
Virginia Cornish Columbia U. Luis Liz-Marzán, CIC biomaGUNE
ASSOCIATE DIRECTOR, BUSINESS DEVELOPMENT Justin Sawyers GLOBAL MARKETING MANAGER Allison Pritchard DIGITAL MARKETING Allan C. Spradling,
Carolyn Coyne, Duke U. Omar Lizardo, UCLA
MANAGER Aimee Aponte JOURNALS MARKETING MANAGER Shawana Arnold MARKETING ASSOCIATES Aaron Helmbrecht, Ashley Carnegie Institution for Sci.
Roberta Croce, VU Amsterdam Jonathan Losos,
Hylton, Mike Romano, Lorena Chirinos Rodriguez, Jenna Voris SENIOR DESIGNER Kim Huynh V. S. Subrahmanian,
Christina Curtis, Stanford U. Wash. U. in St. Louis
Northwestern U.
DIRECTOR AND SENIOR EDITOR, CUSTOM PUBLISHING Sean Sanders ASSISTANT EDITOR, CUSTOM PUBLISHING Jackie Oberst Ismaila Dabo, Penn State U. Ke Lu, Inst. of Metal Res., CAS
Ira Tabas, Columbia U.
PROJECT MANAGER Melissa Collins Jeff L. Dangl, UNC Christian Lüscher, U. of Geneva
Eriko Takano, U. of Manchester
DIRECTOR, PRODUCT & PUBLISHING DEVELOPMENT Chris Reid DIRECTOR, BUSINESS STRATEGY AND PORTFOLIO MANAGEMENT Sarah Whalen Chiara Daraio, Caltech Jean Lynch-Stieglitz,
A. Alec Talin, Sandia Natl. Labs
ASSOCIATE DIRECTOR, PRODUCT MANAGMENT Kris Bishop PRODUCT DEVELOPMENT MANAGER Scott Chernoff PUBLISHING TECHNOLOGY Nicolas Dauphas, U. of Chicago Georgia Inst. of Tech.
Patrick Tan, Duke-NUS Med. School
MANAGER Michael Di Natale SR. PRODUCT ASSOCIATE Robert Koepke PRODUCT ASSOCIATE Caroline Breul, Anne Mason SPJ ASSOCIATE Frans de Waal, Emory U. David Lyons, U. of Edinburgh
Sarah Teichmann,
MANAGER Samantha Bruno Fuller SPJ ASSOCIATE Casey Buchta
Claude Desplan, NYU Fabienne Mackay,
Wellcome Sanger Inst.
Sandra DÍaz, QIMR Berghofer
MARKETING MANAGER Kess Knight BUSINESS DEVELOPMENT MANAGER Rasmus Andersen SENIOR INSTITUTIONAL LICENSING MANAGER Rocio Titiunik, Princeton U.
U. Nacional de CÓrdoba Zeynep Madak-Erdogan, UIUC
Ryan Rexroth INSTITUTIONAL LICENSING MANAGER Marco Castellan, Claudia Paulsen-Young SENIOR MANAGER, INSTITUTIONAL LICENSING Samuel Díaz-Muñoz, UC Davis Anne Magurran, U. of St. Andrews
Shubha Tole,
OPERATIONS Judy Lillibridge SENIOR OPERATIONS ANALYST Lana Guz SYSTEMS & OPERATIONS ANALYST Ben Teincuff FULFILLMENT ANALYST Tata Inst. of Fundamental Res.
Ulrike Diebold, TU Wien Ari Pekka Mähönen, U. of Helsinki
Aminta Reyes Stefanie Dimmeler, Asifa Majid, U. of Oxford Maria-Elena Torres Padilla,
Goethe-U. Frankfurt Oscar Marín, King’s Coll. London Helmholtz Zentrum München
DIRECTOR, GLOBAL SALES Tracy Holmes US EAST COAST AND MID WEST SALES Stephanie O'Connor US MID WEST, MID ATLANTIC AND
Hong Ding, Inst. of Physics, CAS Charles Marshall, UC Berkeley Kimani Toussaint, Brown U.
SOUTH EAST SALES Chris Hoag US WEST COAST SALES Lynne Stickrod ASSOCIATE DIRECTOR, ROW Roger Goncalves SALES REP, ROW
Dennis Discher, UPenn Christopher Marx, U. of Idaho Barbara Treutlein, ETH Zürich
Sarah Lelarge SALES ADMIN ASSISTANT, ROW Victoria Glasbey DIRECTOR OF GLOBAL COLLABORATION AND ACADEMIC PUBLISHING RELATIONS,
Jennifer A. Doudna, David Masopust, U. of Minnesota Jason Tylianakis, U. of Canterbury
ASIA Xiaoying Chu ASSOCIATE DIRECTOR, INTERNATIONAL COLLABORATION Grace Yao SALES MANAGER Danny Zhao MARKETING MANAGER
UC Berkeley Geraldine Masson, CNRS Wim van der Putten, Netherlands
Kilo Lan ASCA CORPORATION, JAPAN Rie Rambelli (Tokyo), Miyuki Tani (Osaka)
Ruth Drdla-Schutting, C. Robertson McClung, Inst. of Ecology
DIRECTOR, COPYRIGHT, LICENSING AND SPECIAL PROJECTS Emilie David RIGHTS AND PERMISSIONS ASSOCIATE Elizabeth Sandler Matthew Vander Heiden, MIT
Med. U. Vienna Dartmouth
LICENSING ASSOCIATE Virginia Warren CONTRACT SUPPORT SPECIALIST Michael Wheeler Ivo Vankelecom, KU Leuven
Raissa M. D'Souza, UC Davis Rodrigo Medellín,
Bruce Dunn, UCLA U. Nacional Autónoma de México Judith Varner, UC San Diego
William Dunphy, Caltech C. Jessica Metcalf, Princeton U. Henrique Veiga-Fernandes,
MAIN HEADQUARTERS EDITORIAL PRODUCT ADVERTISING AAAS BOARD OF DIRECTORS
[email protected] & CUSTOM PUBLISHING Scott Edwards, Harvard U. Baoxia Mi, UC Berkeley Champalimaud Fdn.
Science/AAAS CHAIR Susan G. Amara
advertising.science.org/ Todd Ehlers, U. of TÜbingen Tom Misteli, NCI, NIH Reinhilde Veugelers, KU Leuven
1200 New York Ave. NW NEWS PRESIDENT Gilda A. Barabino
products-services Nader Engheta, UPenn Alison Motsinger-Reif, Bert Vogelstein, Johns Hopkins U.
Washington, DC 20005 [email protected] PRESIDENT-ELECT Keith Yamamoto Julia Von Blume, Yale School of Med.
[email protected] Karen Ersche, U. of Cambridge NIEHS, NIH (S)
SCIENCE INTERNATIONAL INFORMATION FOR AUTHORS TREASURER Carolyn N. Ainslie David Wallach, Weizmann Inst.
Beate Escher, UFZ & U. of Tübingen Danielle Navarro,
Clarendon House science.org/authors/ CLASSIFIED ADVERTISING CHIEF EXECUTIVE OFFICER
Barry Everitt, U. of Cambridge U. of New South Wales Jane-Ling Wang, UC Davis (S)
Clarendon Road science-information-authors advertising.science.org/ Sudip Parikh Vanessa Ezenwa, U. of Georgia Daniel Nettle, Newcastle U. Jessica Ware,
Cambridge, CB2 8FH, UK REPRINTS AND PERMISSIONS science-careers BOARD Cynthia M. Beall Toren Finkel, U. of Pitt. Med. Ctr. Daniel Neumark, UC Berkeley Amer. Mus. of Natural Hist.
science.org/help/ [email protected] Ann Bostrom Gwenn Flowers, Simon Fraser U. Beatriz Noheda, U. of Groningen David Waxman, Fudan U.
SCIENCE CHINA Chris Wikle, U. of Missouri (S)
reprints-and-permissions Janine Austin Clayton Peter Fratzl, Helga Nowotny,
Room 1004, Culture Square JOB POSTING CUSTOMER SERVICE
Max Planck Inst. Potsdam Vienna Sci. & Tech. Fund Terrie Williams, UC Santa Cruz
No. 59 Zhongguancun St.
MEDIA CONTACTS employers.sciencecareers.org Kaye Husbands Fealing Ian A. Wilson, Scripps Res. (S)
[email protected] Maria M. Klawe Elaine Fuchs, Rockefeller U. Pilar Ossorio, U. of Wisconsin
Haidian District, Beijing, 100872 [email protected] Jane Maienschein Jay Gallagher, U. of Wisconsin Andrew Oswald, U. of Warwick Hao Wu, Harvard U.
MULTIMEDIA CONTACTS
Daniel Geschwind, UCLA Isabella Pagano, Li Wu, Tsinghua U.
SCIENCE JAPAN [email protected] MEMBERSHIP AND INDIVIDUAL Robert B. Millard
SUBSCRIPTIONS Ramon Gonzalez, Istituto Nazionale di Astrofisica Wei Xie, Tsinghua U.
ASCA Corporation [email protected] Babak Parviz
science.org/subscriptions U. of South Florida Elizabeth Levy Paluck, Benjamin Youngblood, St. Jude
Sibaura TY Bldg. 4F, 1-14-5 INSTITUTIONAL SALES William D. Provine
Sandra González-Bailón, UPenn Princeton U. Yu Xie, Princeton U.
Shibaura Minato-ku AND SITE LICENSES MEMBER BENEFITS Juan S. Ramírez Lugo Jan Zaanen, Leiden U.
Nicolas Gruber, ETH Zürich Jane Parker,
Tokyo, 108-0073 Japan science.org/librarian aaas.org/membership/benefits Susan M. Rosenberg Hua Guo, U. of New Mexico Max Planck Inst. Cologne Kenneth Zaret,
Science serves as a forum for discussion of important issues related to the advancement of science by publishing material on Taekjip Ha, Johns Hopkins U. Giovanni Parmigiani, UPenn School of Med.
which a consensus has been reached as well as including the presentation of minority or conflicting points of view. Accordingly, Daniel Haber, Mass. General Hos. Dana-Farber Cancer Inst. (S) Bing Zhu, Inst. of Biophysics, CAS
all articles published in Science—including editorials, news and comment, and book reviews—are signed and reflect the individual Sharon Hammes-Schiffer, Yale U. 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. Wolf-Dietrich Hardt, ETH Zürich Ana Pêgo, U. do Porto Maria Zuber, MIT

Subscribe to News from Science for unlimited
access to authoritative, up-to-the-minute news
on research and science policy.
bit.ly/NewsFromScience
Building the Future,
One City at a Time
Our WINLAB is taking tomorrow’s tech out of the lab
and into the streets of New York City
Building on their extensive research with wireless testbeds, radio technology, Smart intersections that
can safely process vital
and mobile internet architecture, the Rutgers team at the Wireless Information
information in real time
Network Laboratory (WINLAB) is helping to implement COSMOS, a real-world
outdoor lab putting next-generation wireless technology to work.
Cloud-based autonomous
vehicles for reliable
A major partnership between WINLAB and other local innovation leaders,
driverless transportation
with funding from the National Science Foundation, COSMOS will be
deployed across a square mile in Manhattan’s vibrant, densely populated
Remote surgery enabling
West Harlem neighborhood.
doctors to operate on patients
across great distances
These urban wireless networks of the future will support data-intensive
applications for a smarter, safer world.
Augmented and virtual reality
to enhance interactivity and
provide new experiences
Excellence is Earned
excellence.rutgers.edu
ED ITORIA L
Measure emissions to manage emissions
I
n the 30 years since the world began negotiating submitted intermittently to the United Nations (UN). As
the reduction of greenhouse gas (GHG) emissions, of last month, no nation has submitted a complete ac-
no one has identified exactly where all that pollu- counting of its emissions for 2021. Indeed, 52 countries
tion is coming from. That will begin to change next have not submitted any emissions inventories covering
week when Climate TRACE (Tracking Real-Time the past 10 years. Moreover, the inventories that do exist
Atmospheric Carbon Emissions)—a nonprofit coali- often have large omissions and fail to provide the gran-
tion of artificial intelligence (AI) specialists, data ular data needed to make decisions. For example, last
scientists, researchers, and nongovernmental organiza- year, Climate TRACE found the actual emissions from Al Gore
tions—releases the first facility-level inventory of the global oil and gas production collectively were around
is a former vice
largest known individual sources of the 162 million tons double what was self-reported to the UN in 2020. Even
president of the
of GHG pollution emitted into the troposphere every more oil and gas emissions went uncounted by coun-
United States. He
day. With thousands of businesses, banks, investors, tries that are not required to report these data.
and 88 nation-states committed to reducing emissions The Climate TRACE coalition is working to advance was awarded the
to net zero by 2050, comprehensively tracking progress emissions monitoring closer to “the stage of science” Nobel Peace Prize,
toward that goal is essential. This is Kelvin once described—by training along with the
especially important given last week’s AI algorithms to fuse data across Intergovernmental
United Nations Emissions Gap Report multiple wavelengths and from more Panel on Climate
indicating that the world is far behind
pace for reducing emissions by 2030.
“…no one than 300 satellites; 11,100 air-, land-,
and sea-based sensors; and other
Change (IPCC),
in 2007. He is a
Scientists have long known the to-
tal amount of carbon dioxide (CO2) in
has identified data streams to identify all of the
largest point sources and track them
founding member
and among the
the atmosphere. The Keeling Curve—
a daily record of global atmospheric
exactly over time. In recent years, new satel-
lites have begun to measure methane
funders of Climate
TRACE. press@
CO2 concentration—leaped from 313
parts per million (ppm) in March 1958
where all that emissions for a limited set of sources.
But the atmosphere is so enriched
carthagegroup.com
to 2021’s staggering 414.72 ppm global

average. Likewise, the component
pollution is with CO2 that the “signal-to-noise ra-
tio” defies the ability to measure CO2
parts of the emissions puzzle are well
understood—the burning of fossil fu-
coming from.” point-source emissions directly, leav-
ing AI as the best recourse. The prog-
els, transportation, industry, conven- ress demonstrated by Climate TRACE
tional agriculture, deforestation, and mirrors the breakthroughs in AI that
other sources are continually adding to the accumula- are ushering in a new era of radical transparency in
tion that lingers in the atmosphere. But until now, it has multiple domains—making it possible to “see” specific
not been possible to precisely map and track the spe- land-use changes more clearly, and even to identify mi-
cific sources of GHG pollution. Climate TRACE’s global croscopic structures inside living human cells that had
database of the most substantial individual sources of been impossible to “see” previously. Climate TRACE is
emissions across two dozen sectors of the global econ- proving that AI can do the same for emissions monitor-
omy will be free and publicly available. ing—making the invisible, visible.
Why is this information so essential? As the eminent Lord Kelvin’s dictum is commonly translated in the
physicist Lord Kelvin said nearly 140 years ago, “When business world as “you can only manage what you
you can measure what you are speaking about, and ex- can measure.” Thanks to breakthroughs in AI and
press it in numbers, you know something about it; but other technologies, researchers, government officials,
when you cannot measure it, when you cannot express and business leaders can now manage emissions with
it in numbers, your knowledge is of a meagre and un- timely, granular, and actionable climate information at
satisfactory kind; it may be the beginning of knowledge, their fingertips. With no time left to wait as the world
but you have scarcely, in your thoughts advanced to the burns and drowns, we can now begin to measure emis-
stage of science.” Efforts to limit global temperature rise sions with the precision needed to better manage their
to 1.5°C are currently informed by rough estimates prin- reduction—quickly.
cipally derived from self-reported national inventories –Al Gore
PHOTO: JASON MYERS
10.1126/science.adf5788
SCIENCE science.org 4 NOVEMEBER 2022 • VOL 378 ISSUE 6619 455

NEWS
Edited by
IN BRI EF Jeffrey Brainard
NATURAL HISTORY
Science society lists Earth’s top ‘geoheritage’ sites
T
he International Union of Geological Sciences familiar ones, such as the Grand Canyon’s “great un-
last week marked its 60th anniversary by an- conformity,” a billion-year gap in the rock record erased
nouncing a list of 100 “geoheritage” sites that by erosion. More exotic examples include limestones
have substantially influenced understanding of in Germany that preserve Archaeopteryx, a feathered
Earth’s deep history. The global list, released in fossil that links dinosaurs to birds, and the Canary
collaboration with UNESCO, is meant to foster Islands’ Taburiente Caldera (above), which gave such
conservation and tourism. The sites include volcanic formations their name.
tested it or a placebo in 330 adult volun- research on the much larger International
Antimalaria antibody delivers teers. The research team is now testing a Space Station. Nine of Tiangong-3’s research
INFECT IOUS DISEASES | A single intrave- more powerful antibody in 6- to 10-year- projects involve international collaborators,
nous infusion of lab-produced monoclonal old children in Mali who are receiving many from developing countries, jointly
antibodies was up to 88% efficacious in the antibody by subcutaneous injection, a selected with a U.N. agency. These include
preventing malaria infection in Mali, method that is much more practicable with a four-nation experiment on gamma ray
PHOTO: PETER THOMPSON/HERITAGE IMAGES/GETTY IMAGES

researchers reported this week. The young children than intravenous infusion. bursts and a pair of infrared cameras to
study, in The New England Journal of In 2020, malaria killed 627,000 worldwide, study Earth’s humidity flows. Tiangong-3
Medicine, offers new hope for combat- two-thirds of whom were children younger or “Heavenly Palace” will house between
ting the disease, spread by parasite-laden than 5 in Africa. three and six astronauts. It succeeds two
mosquitoes. Malaria parasites have smaller craft, Tiangong-1 and -2, that are
developed resistance against many drugs, no longer orbiting.
and the mosquitoes that spread them have China completes space station
adapted to withstand some insecticides. | China this week launched
L A B O R AT O R I E S
Researchers at the U.S. National Institute the final module of its Tiangong-3 space Oil researcher wins courage prize
of Allergy and Infectious Diseases devel- station, providing the country its first S C I E N T I F I C P R I Z E S | A biochemist in
oped the antibody, and working with long-term platform for space experiments. Nigeria has won this year’s John Maddox
colleagues at the University of Sciences, U.S. restrictions on use of NASA funds Prize, which honors “standing up for
Techniques and Technology of Bamako, have blocked China from participating in science,” for her tenacity in braving

threats while working with oil companies
and local communities to clean up soil On Little Bahama Bank, a tiger
pollution near the country’s oil fields. shark swims over manatee
Eucharia Oluchi Nwaichi, a biochemist grass (Syringodium filiforme).
at the University of Port Harcourt, was
threatened by representatives of an oil
company who confiscated her recordings
and data, according to a statement by the
award’s organizers, the U.K. charity Sense
for Science and the journal Nature. She
also worked to defuse a dispute between
residents and a different oil company
about the harmful effects of liquid waste
on fish stocks, and to persuade both sides
to support research to test methods to
decontaminate soil, the statement says.
Petroleum production in Nigeria, one
of the world’s largest oil exporters, has
caused extensive pollution and led at
times to armed conflict.
Academy probe clears biologist

R E S E A R C H I N T E G R I T Y | A review by
the U.S. National Academy of Medicine
ECOLOGY
(NAM) has found no basis for sanctioning a
member whom three leading Republicans
accused of obscuring COVID-19’s ori-
Cameras on sharks help map vast seagrass meadows
R
esearchers mounted cameras on tiger sharks to help measure an expanse of
gin. Last year, members of the House of
seagrass meadows in the Bahamas that they estimate is the world’s largest,
Representatives Committee on Energy and
at 92,000 square kilometers. Seagrass provides a vital habitat for many other
Commerce asked NAM to suspend and
marine species, including commercially valuable fish, and the plants sequester
investigate conservation biologist Peter
large amounts of carbon. The new estimated size of the Bahamas’ seagrass
Daszak. Their complaint alleged that he vio-
meadows, reported this week in Nature Communications, is about 40 times larger
lated NAM’s code of conduct by refusing to
than the previous estimate and represents an area the size of Portugal. Tiger sharks
share data and answer their questions about
(Galeocerdo cuvier) often swim over seagrass meadows, where they hunt dugongs and
a grant awarded by the National Institutes
other herbivores. The research team fitted seven sharks with video cameras and track-
of Health to the nonprofit he runs, the
ing devices to map the seagrass habitat—the first use of this technique. The team used
EcoHealth Alliance, which in turn shared
these data and diver surveys to spot-check estimates of total extent gleaned from
funding with China’s Wuhan Institute of
satellite data, which can be imprecise. Measurements of carbon in seagrass sediments
Virology (WIV). Several Republicans have
suggest the Bahamas could contain as much as one-quarter of the carbon stored
argued that the pandemic originated in a
worldwide by seagrass.
lab leak at WIV. Many scientists not involved
in the grant have agreed with Daszak that
no direct evidence supports this assertion
whereas stronger evidence indicates the stunning, 200-million-year-old fossil was researchers have long sought such a trial to
pandemic virus first jumped from animals obliterated by a German bomb that hit the study whether suppressing the SARS-CoV-2
to humans at a Wuhan market. In an email London museum where it was displayed. coronavirus can reduce Long Covid’s debili-
to members last week, NAM said its probe Since then, paleontologists have relied solely tating symptoms, such as fatigue and brain
determined Daszak had not violated its code on a single scientific illustration of the fossil, fog. Participants will take either Paxlovid (a
of conduct. made in 1819. The two much more detailed combination of the antivirals nirmatrelvir
plaster casts were discovered in 2016 and and ritonavir) or a placebo for 15 days. The
PHOTO: A.J. GALLAGHER ET AL., NAT. COMMUN. 13, 6328 (2022)
2019 at Yale University’s Peabody Museum of Food and Drug Administration authorized
Casts of destroyed fossil found Natural History and Berlin’s Natural History the drug in December 2021 for people at
PA L E O N T O L O GY | In two museums’ Museum, the researchers reported this week high risk of severe illness from acute COVID-
storerooms, scientists have stumbled upon in Royal Society Open Science. 19. Paxlovid knocks back virus that’s rapidly
previously unknown casts of an ancient replicating. Researchers don’t know whether
marine reptile’s skeleton that was destroyed such replication is happening in people with
during World War II. The original fossil Long Covid pill trial set to begin Long Covid or whether their bodies harbor
represented the first complete ichthyosaur, | Plans for the first
C L I N I CA L R E S E A R C H virus reservoirs that help drive symptoms.
a creature that resembled a cross between test of whether Pfizer’s COVID-19 pill known The trial marks the first treatment trial
a crocodile and a dolphin. Famed fossil as Paxlovid can alleviate Long Covid were backed by RECOVER, the U.S. National
hunter Mary Anning exhumed the speci- unveiled last week, as organizers said they Institutes of Health’s research effort on Long
men from Jurassic-age rocks along southern expect to begin recruiting 1700 volun- Covid, which critics charge has been slow to
England’s coastline in 1818. In 1941, the teers in January 2023. Patient groups and test therapies.
SCIENCE science.org 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 457

IN DEP TH
ASTRONOMY
Should Webb telescope’s data be open to all?

NASA plans to end policy of giving some observers 1 year’s exclusive access to data
By Daniel Clery But others say instant access to data may before the telescope’s launch urged that its
clash with pressing efforts to improve diver- proprietary period last just 6 months. Any
T
he $10 billion James Webb Space sity, equity, and inclusion in astronomy. Sci- longer, the committee concluded, and most
Telescope (JWST) has been observ- entists at smaller institutions, many of which of the data collected during the first year of
ing for less than 4 months, but al- serve underrepresented groups, may need observing, known as cycle 1, would be un-
ready a storm is brewing over access more time for analysis compared with the available to astronomers trying to plan what
to its data. Webb images and spec- well-resourced teams at established research to look for in cycle 2, or even some of cycle
tra all end up in an archive at the universities. Eliminating proprietary time 3. For a mission then expected to only last
Space Telescope Science Institute (STScI) could “put people in a position where they 5 years, that was unsupportable in the view of
in Baltimore, yet most of them aren’t can’t succeed,” says astronomer Mercedes some committee members.
freely available until 1 year after the data Lopez-Morales of Harvard University, chair of In recent years, NASA has reduced the
were collected. This gives the researchers the influential JWST Users Committee, which proprietary time for the Hubble Space Tele-
who proposed the observations time to recently urged NASA to maintain a data em- scope and the Chandra X-ray Observatory
analyze them and publish results without bargo period but reduce it to 6 months. to 6 months. But it did not do the same for
being scooped. To find a way forward, STScI will begin to Webb, in part because the academic teams
But some astronomers question the poll more than 12,000 astronomers later this that built its four instruments and the Eu-
practice, arguing that data from federally month about their attitudes to proprietary ropean and Canadian space agencies that
funded projects should be free for all to time. “There are mixed feelings in the com- partnered in the project had 12 months’ pro-
use. NASA, Webb’s primary backer, is facing munity,” says Neill Reid, STScI’s associate prietary time written into their agreements.
an open data push from the White House director for science. “There’s definitely a feel- Webb planners did, however, adopt the
and may soon end the restriction. Having ing that open access is a good way to go, but committee’s suggestion to provide im-
so much Webb data locked away “doesn’t without disadvantaging [some observers]. mediate access to a key subset of data:
pass the smell test. It’s just not right,” says There’s not one size that fits all.” 500 hours of observations made during the
astronomer Garth Illingworth of the Uni- Astronomers have long been pioneers in first 5 months of cycle 1, covering all Webb’s
versity of California, Santa Cruz, who from putting data into open archives for anyone key science areas and employing all its in-
2009 to 2017 chaired a committee advising to use. Typically, data from automated sur- struments and observing modes. This Early
STScI on Webb’s future science operations. vey telescopes, which systematically scan Release Science (ERS) program amounts to
He and other proponents of an open data the sky night after night, are deposited about 5% of cycle 1’s total observing time
policy also make a practical argument. They straightaway. But proprietary time remains (see table, p. 459) and its data are put in the IMAGE: NORTHROP GRUMMAN
say so-called proprietary time prevents a tradition at general purpose telescopes, archive immediately, providing some mate-
other astronomers from using fresh data which give astronomers whose observing rial for scientists devising cycle 2 proposals.
to shape their own observing plans, reduc- proposals are chosen exclusive access to Webb’s successful launch has also tem-
ing the efficiency of a highly sought-after their data for as long as 18 months. pered the data-access debate somewhat be-
instrument that has consumed billions of After often-heated debate, the Webb ad- cause it required fewer course corrections
dollars of public money. visory committee that Illingworth chaired than expected, leaving a fuel reserve that

N E WS
could extend the telescope’s lifetime and al- groups: Fixed proprietary periods can help SOCIAL JUSTICE
low more time for observing. with this,” McCaughrean says.
But proprietary time still galls many as-
tronomers. On 15 November, STScI will put
out the call for cycle 2 proposals, for observa-
Illingworth says underrepresented groups
could be helped by making proprietary time
optional, with space in the proposal form for
Teeth record
tions to start in July 2023. Although the ERS
data are helping many plan future observa-
tions, most cycle 1 data are still locked up
researchers to request it and explain their
need. Marcia Rieke of the University of Ari-
zona, principal investigator of Webb’s near-
pneumonia—
and some will remain so until the second half
of 2024. NASA is also under pressure from
the White House. On 25 August, the White
infrared camera, favors the opposite: giving
everyone proprietary time with the expecta-
tion that those who don’t need it will give it
and racism
House’s Office of Science and Technology up, or at least release their data when they Research raises standards
Policy ordered departments and agencies to
move toward making the results of all feder-
publish. Proprietary time “encourages people
to do a good job analyzing data,” she says.
for working with
ally funded research freely and immediately “It’s complicated,” Lopez-Morales con- anthropological collections
available by 2026. cedes. On the one hand, she says, reducing
In response, NASA asked the JWST Users proprietary time could be “in conflict with By Rodrigo Pérez Ortega
Committee, which was meeting that same [NASA’s] other efforts” to widen the pool
I
day, to look again at the telescope’s propri- of astronomers. Yet instant access to data n the 1930s, a 23-year-old Black man
etary time. It, too, recommended moving to means the many groups who are unsuccess- was admitted to City Hospital #2 in St.
6 months. A year “won’t fly anymore,” Lopez- ful at applying for observing time—in cycle Louis and, according to his death cer-
Morales says. But the committee recognized 1, the rejection rate was 75%—can still get a tificate, died of pneumonia shortly after.
that its members, 12 senior astronomers, may crack at using the data. “We need to see what Without his consent—or his family’s—his
have a biased view, so they asked STScI to the community wants,” Lopez-Morales says, deidentified body was included in one of
consult with the wider community. “and we will advocate for that.” the United States’s most studied collections
Mark McCaughrean, an astronomer Michael New, NASA’s deputy associate di- of human remains, the Robert J. Terry Ana-
with the European Space Agency, says the rector for research, says even if the commu- tomical Skeletal Collection, which is now at
12-month embargo should remain in place nity comes down in favor of 6 months, that the Smithsonian Institution’s National Mu-
because it helps level the field. Astronomers will be a stepping stone to fully removing seum of Natural History (NMNH). Almost
at small universities often shoulder heavy proprietary time. A NASA policy, now be- a century later, a team of researchers has
teaching loads and other responsibilities. If ing finalized, will mandate instant access to been able to confirm the pathogen that ul-
their Webb observations were publicly re- instrument data in line with White House timately killed him by studying the plaque
leased straight away, they could be scooped policy. New says the Webb agreements with on his teeth, an achievement that opens new
by teams at large, well-resourced research other space agencies contain clauses allowing avenues for studying diseases of the past that
institutions before they had a chance to ana- proprietary time to be reduced once the tele- may leave no other mark after death.
lyze the data over the summer, for example. scope is operating and discussions to do so In their paper, published last month in
The situation is trickier still for European are already taking place. “It’s not if we do it, Communications Biology, the researchers
astronomers who don’t get automatic fund- but when,” he says. “Will everyone be happy? also took steps to address the complex issues
ing for postdocs and other expenses linked No, but we will listen to all sides.” of ethics and social justice that surround re-
to winning time on a space telescope as U.S. NASA is already working to provide positories such as the Terry collection. They
astronomers do. “We should be aiming to im- support for researchers who will be dis- reconstructed the context in which a now-
prove diversity by removing structural biases advantaged, New adds. If some are under- nameless man lived, including how struc-
against smaller institutions, early-career re- resourced, “solve that problem, don’t pre- tural racism contributed to his death—and
searchers, and historically disadvantaged serve proprietary time.” j helped build the collection itself.
“This may be a way of doing research with
CREDITS: (GRAPHIC) K. FRANKLIN/SCIENCE; (DATA) SPACE TELESCOPE SCIENCE INSTITUTE
the Terry collection ethically,” says Carlina

Private eye de la Cova, a biological anthropologist at
Most astronomers using the James Webb Space Telescope get a yearlong “proprietary” period to exclusively analyze the University of South Carolina, Columbia,
data from observing projects. About 75% of Webb’s first year of observations is not immediately open for use. who was not involved in the study. Analyz-
ing dental calculus, she adds, is a clever,
Early release science nondestructive way to learn more about an
Competitively chosen to individual’s past, and their disease profile.
provide representative data sets Proprietary Open “It should not be overlooked.”
“Previous research does not recognize
Guaranteed time observers
Time given to instrument builders,
the individual skeletons and the once-living
principal investigators, etc. individuals that they represent,” says Molly
Zuckerman, a bioarchaeologist and paleo-
General observer program pathologist at Mississippi State Univer-
Competitive, open to all sity, Starkville, and a co-author of the new
study. “Their humanity has been reduced
Cycle 1 total* or ignored, and they’re turned into just a
collection of objects for scientific research.”
Hours 0 2000 4000 6000 8000 10,000 As the field of biological anthropology
*Hours exceed 1 year in case some observations can’t be done for technical reasons. grapples with—and tries to correct for—a

N E WS | I N D E P T H
In 1923, City Hospital #2 in St. Louis, where the

St. Louis Individual died, was severely overcrowded
and underfunded, and had poor sanitary conditions.
complex history of racism, she and her col-

leagues aim to set a new ethical standard
for studying anthropological collections
with a controversial past. Rachel Watkins,
a biocultural anthropologist at American
University who was not involved in the
study, applauds the effort. “Offering up this
way of repositioning the Terry collection is
really significant,” she says.
The idea of using dental calculus—or hard-
ened plaque—to study diseases of the past is
not new. Accumulated over a lifetime, calcu-
lus doesn’t decay after death and holds onto
everything from traces of food, such as milk
proteins, to fragments of bacterial DNA. In
2015, molecular anthropologist Rita Austin,
then at NMNH and the University of Okla-
homa, decided to see whether calculus can
reveal causes of death, including so-called
invisible diseases such as pneumonia, which
don’t leave any markings in bones and are
often misdiagnosed in historical documents.
“There’s a lot of potential for thinking about St. Louis (WUSTL), went about building the healed poorly—a possible sign of physical
what kind of pathogens could end up in your collection that now bears his name. Taking violence—several cavities, missing teeth,
dental calculus,” says Austin, now at the Uni- advantage of the lack of legal protections abscesses in the mouth, gingivitis, and peri-
versity of Oslo. for marginalized people in the 19th and odontitis, all signs of poor health care. His
She decided to use the Terry collection early 20th century, he amassed thousands skeleton also shows signs of a TB infection,
because it contains more than 1700 skel- of bodies of people who died in public in- although the researchers didn’t find traces
etons of individuals, as well as their death stitutions in St. Louis—most of them poor of the bacterium that causes the disease in
certificates. If she and her team could find or institutionalized—and whose families the plaque.
evidence of disease in calculus, they could didn’t claim them within 36 hours after Still, “It was a little bit difficult to really
compare their findings with the recorded death. “These are individuals who went see this as a study of an individual, given
causes of death. through a process where their identity was how little access we had to the details of
Austin began by scraping the dental calcu- stripped from them,” de la Cova says. “From his life,” Ward says. So the researchers
lus of several individuals who had been di- the beginning, they were dehumanized.” supplemented their research with a review
agnosed with tuberculosis (TB), syphilis, and More than half of the individuals in the of historical documents. They used news-
pneumonia and sequencing the DNA traces Terry collection are Black. paper clippings to reconstruct the medical
it contained to make a profile of the bacte- “It was clear that we needed to spend context of 1930s St. Louis, including how
ria that once lived in their mouths. What she more time on this discussion of structural Black people would often delay seeking
found in one individual—whom the team racism in St. Louis, particularly in rela- medical care because of distrust of medi-
named the St. Louis Individual—stood out: tion to the medical profession,” says Geoff cal institutions and how conditions at City
a diverse collection of microbes, the most Ward, a historical sociologist at WUSTL Hospital #2 were reportedly worse than at
abundant of which, such as Acinetobacter who helped with the contextual work on the local zoo, according to a 1924 article in
nosocomialis and Klebsiella pneumoniae, the St. Louis Individual. In a lengthy trans- The Pittsburgh Courier. Jim Crow laws and
commonly cause pneumonia and hospital parency statement in the paper—the result local anti-Black political violence likely
infections even today. They were direct evi- of feedback from colleagues from different also worsened things for people like the
dence of the pneumonia that killed the man, scientific fields—the researchers outlined St. Louis Individual. His “life was likely so-
according to his death certificate. their efforts to “rehumanize” the St. Louis cially, mentally, and physiologically stress-
“Recovering biomolecular evidence Individual and show how structural racism, ful,” the researchers write, which could
that completely aligned with the recorded from poor-quality medical care for Black have made him vulnerable to a fatal pneu-
cause of death was just astounding for us,” people to physical violence, shaped his life monia, a rare cause of death for young
PHOTO: ST. LOUIS POST-DISPATCH/POLARIS

Zuckerman says. “It really changed our un- and death. They hoped to “bring some of the men even before widespread antibiotics,
derstanding of what dental calculus can tell St. Louis Individual’s life history, previously Zuckerman says. These very same injus-
us about the past.” obscured and erased, into broader recog- tices, the researchers say, allowed his body
But the researchers felt an obligation to nition, fostering a shared remembrance of to end up in the Terry collection.
probe further, into the life of the St. Louis In- him and others in the Terry collection with Sabrina Sholts, curator of biological
dividual. “We didn’t come to the Terry collec- similar experiences.” anthropology at NMNH, sees the study as
tion thinking that it was neutral,” Zuckerman The team found, for example, that his part of the field’s broader effort to move
says, given the way Robert J. Terry, a profes- skeleton showed many telltale signs of a “towards more ethical practices, and [have]
sor of anatomy at Washington University in hard life. He had a fractured jaw that had conversations and priorities on fairness and

RESOURCES
A cleaner way to mine

clean-energy mainstays
An electric field can reduce the need for
polluting chemicals to extract rare earth elements
By Dennis Normile electrokinetic technology might offer a
cleaner alternative. The approach, in which
E
lectric cars, wind turbines, and LED electrodes on the top and bottom of a vol-
lighting all help keep the environ- ume of soil induce an electric field, speeding
ment clean, but making them can be a the movement of the leaching agent and the
dirty business. The high-performance ions it extracts, is already used in soil reme-
magnets in motors and generators diation and has been proposed for copper
and the glowing phosphors in LEDs and gold mining. It has “the real potential
and flat screens all depend on substances to outperform traditional mining tech-
called rare earth elements (REEs). And niques in terms of efficiency, environmen-
capturing REEs from the clay deposits in tal impacts, and economics,” says Riccardo
which many are found requires leaching Sprocati, a specialist in the technology at
agents that pollute soil and groundwater. the Technical University of Denmark.
Now, a Chinese group has developed— The Chinese team started with a bench-
justice.” Other controversial collections, and tested on tons of soil—an approach top experiment, then scaled up to 20 kilo-
such as the Samuel George Morton Cranial called electrokinetic mining that relies on grams of material, and finally moved to
Collection at University of Pennsylvania electric currents to free the REEs, sharply a field test at an actual ion-adsorption
Museum of Archaeology and Anthropo- reducing the need for polluting chemicals. deposit, trying the technique on a 14-ton
logy, for example, have recently sparked a The strategy, described this week in Nature hunk of clay. The method extracted a higher
reckoning with the racism that built them Sustainability, could be “a game changer, percentage of the REEs more quickly than
(Science, 9 July 2021, p. 148). providing that it is feasible at a large scale,” conventional leaching and needed less am-
Austin, Zuckerman, and their colleagues says Anouk Borst, a geologist at KU Leuven. monium sulfate. It also left the soil cleaner
are now working with the Smithsonian and Despite their name, REEs are fairly and reduced contaminating elements in the
other institutions to create new guidelines abundant in Earth’s crust. It’s just hard to leachate, which could simplify processing.
for studying anthropological collections in a find deposits that are economical to mine. The team calculates that the process could
more ethical way, respecting both the indi- Heavy REEs—those with high atomic num- cut mining costs by about two-thirds.
viduals in the collections and their modern- bers, including dysprosium, yttrium, and Gareth Hatch, a rare earths expert at
day communities. Since last year, NMNH no terbium—are most commonly extracted the Strategic Materials Advisory, a consult-
longer approves new research requests on from masses of clay formed through eons ing firm in Manchester, England, notes
Black individuals in their collections, Sholts of weathering of igneous rocks such as that whether the technique can be scaled
says—which means the current study would granite. In these “ion-adsorption” deposits, up “remains to be seen.” The group’s next
now be off-limits. In May, the Smithsonian the elements are adsorbed—or stuck—to test will include about 2000 tons of soil,
also adopted a new policy that allows its the surface of clay particles. They are usu- where they will try “to optimize operating
museums to share ownership of or return ally extracted by pumping large quantities conditions,” says geochemist Hongping He,
entirely collections based on ethical consid- of ammonium sulfate or a similar solution director of GIG and co-author on the pa-
erations, such as lack of consent. “There’s so into the ground. The leachate pulls REEs per. They will divide the site into sections
much potential to possibly identify relatives from the clay and percolates down to bed- to keep the electric current and voltage low
of these individuals and to offer returns,” rock, where it is collected for processing. enough to avoid harming surrounding soil,
Sholts says. Even if families can’t be traced, All that can contaminate water and soil, vegetation, or fauna, He says. And they
Watkins says researchers can take action. not to mention lay waste to large tracts have struck an agreement with a major
“Nothing should happen before [the] de- of land. And roughly 80% of the world’s rare earth supplier to test the process on
scendant community is engaged—and it supply of heavy REEs come from ion- an even larger scale.
needs to be Black folks,” she says. adsorption deposits in southern China and “As with all mining methods, it will still
Although it’s too late for this for the St. adjacent parts of Myanmar, where environ- impact the environment,” says Henning
Louis Individual, whose anonymity will make mental regulations are poorly enforced and Prommer, an environmental engineer at
it nearly impossible to find his family, the re- illegal mining is common. (The light REEs— the University of Western Australia, Perth,
searchers are now planning to engage with cerium, neodymium, and praseodymium— whose group has worked on applying elec-
community members in St. Louis to raise are typically found in hard rock and are trokinetic mining to gold and copper. But,
awareness of their findings. As Zuckerman mined in a different way.) he says, “Given the crucial role that REEs
puts it, “We want to take the messages in this Gaofeng Wang of the Chinese Academy play in our ambitions for a renewable energy
paper to encourage everyone to remember of Sciences’s Guangzhou Institute of Geo- infrastructure, any reduction in the environ-
these are once living individuals.” j chemistry (GIG) and his colleagues thought mental impact of mining is welcome.” j

HARASSMENT
Prominent astronomer barred after complaints

Tim de Zeeuw had a pattern of “extremely unacceptable” behavior, Leiden University says
By Cathleen O’Grady became director of the Leiden Observatory in tling and intimidating behavior described
2003. From 2007 to 2017 he was director gen- by Leiden University was “just another little
A
stellar international career in astron- eral of ESO, a 16-member intergovernmental step,” she says.
omy ended in disgrace on 18 October organization headquartered in southern Ger- Leiden University reportedly asked its
when Leiden University said it had many that operates some of the world’s most Complaints Committee for Unacceptable Be-
“removed” a professor, later revealed advanced telescopes in Chile. haviour to launch an investigation after four
to be prominent theoretical astrono- Several sources who asked to remain women at the observatory filed a complaint
mer Tim de Zeeuw, for what it called anonymous told Science De Zeeuw has a with the dean of the Faculty of Science in
“extremely unacceptable” behavior toward reputation for being “power hungry” and May. Although the university has publicly an-
women colleagues. De Zeeuw, 66, will not be for behaving “weirdly” around women, with nounced the committee’s verdict, it has not
allowed to return to the university, but will reports of inappropriate behavior stretching mentioned De Zeeuw by name, referring to
keep his job and salary until he retires. back decades. An astronomer who worked at him as “a professor” instead. It wasn’t until
The Max Planck Institute for Extra- 25 October that the Dutch newspaper
terrestrial Physics, where he had an af- NRC first published De Zeeuw’s name.
filiation as well, has severed ties with The statement De Zeeuw’s lawyer sent
De Zeeuw and removed his profile to Science a day later confirmed he is
from its website; the European South- the accused scientist.
ern Observatory (ESO), which De The decision to withhold the name
Zeeuw led from 2007 to 2017, says it set off a week of uproar and specula-
has banned him from its premises and tion that led a few colleagues to pub-
will revoke access to his IT account. licly deny involvement. “If you wonder
Leiden University has not made whether I am the prof who was dis-
public a report from an indepen- missed at Leiden University, it is not
dent committee that investigated me,” Christoph Keller, director of sci-
complaints against De Zeeuw. But a ence at the Lowell Observatory and a
21 October online column by Execu- guest professor at the Leiden Observa-
tive Board President Annetje Ottow, tory, tweeted on 20 October. But in her
addressed to the academic commu- column, Ottow defended the secrecy as
nity, offered some details about the well as the decision not to fire the ac-
panel’s findings. “We are talking here cused. “There are good reasons for this
about … abuse of power, gender dis- relating to employment law,” Ottow
crimination and the systematic vilifi- wrote. “Make no mistake, this is a very
cation and belittling of staff,” Ottow drastic measure for him,” she added
wrote. “It also includes inappropriate in an interview in NRC. “He is placed
behaviour with an element of sexual Dutch astronomer Tim de Zeeuw—seen here at the inauguration of in total isolation; he is not allowed to
intimidation: ranging from comments the Extremely Large Telescope in Chile in 2017—acknowledged have contact with staff or to enter the
right down to unwelcome physical his behavior was “unpleasant and impatient in an old-fashioned way.” building.” Others say De Zeeuw came
contact with one of the members of out relatively well. The university’s
staff. All of this was under the constant threat ESO as a graduate student while De Zeeuw penalty amounts to an “unlimited sabbati-
of harming the complainants’ careers.” led the institute says he was flirtatious with cal,” Clements says.
De Zeeuw, in a statement sent to her and inappropriately touched her in social The Royal Netherlands Academy of Arts
Science via his lawyer on 26 October, ac- settings, including running his hand down and Sciences announced on 28 October that
knowledged having been “unpleasant and her back and hugging her from behind. The it had suspended De Zeeuw’s membership,
impatient in an old-fashioned way,” but researcher says she did not feel she could and he had resigned in response. An academy
said he disagreed with the university’s de- report the behavior but found that other spokesperson told Science in an email that
cision. “It has never been my intention to women had similar experiences. the resignation cut short a more thorough
hurt or harm people. I am very sorry that Another woman who worked closely with inquiry that would have determined whether
people have experienced my behaviour as De Zeeuw as a junior astrophysicist at ESO the suspension should be permanent.
negative,” he stated. His lawyer, Merienke more than a decade ago says he used demean- In her column, Ottow acknowledged the
PHOTO: MARIO RUIZ/NEWSCOM
Zwaan, added that media reports about De ing language, such as calling her a “good girl,” university had received signals about De
Zeeuw’s “sexually transgressive behaviour and that she frequently felt uncomfortable Zeeuw’s behavior before this spring, and said
towards women” were incorrect. when she was alone in the office with him. the university had to do better in the future.
De Zeeuw is a “major figure” in astronomy, Although she considers her own experience “It was noticed but regrettably, not enough
says Dave Clements, an astrophysicist at Im- to be relatively mild, she was not surprised was done about it,” she wrote. “For this to
perial College London. He was appointed as to hear about the independent panel’s find- have gone on for so long means that the work
a professor at Leiden University in 1990 and ings. Given his general demeanor, the belit- environment was not safe enough.” j

Master teachers in Houston, part of the Robert Noyce
Teacher Scholarship Program, learn to design science
experiments using balloons.
Black institution in Washington, D.C., who

recently won an MRI award for a single-
crystal x-ray diffractometer to analyze mo-
lecular structures. And if NSF responds by
providing smaller grants, he says, “institu-
tions may have to settle for cheaper instru-
ments. Maybe it’ll be something that’s good
for teaching but not good enough to sup-
port cutting-edge research.”
The CHIPS act is the latest congressio-
nal statement on cost sharing at NSF. For
decades cost sharing was a routine require-
ment in many NSF programs. But in 2004
the agency decided it was discriminatory
and dropped the requirement.
Congress thought that was too drastic
a step, however. In 2007 it mandated cost
U.S. SCIENCE POLICY sharing for the Noyce and MRI awards and
told NSF to take another look at the pros
NSF to end cost-sharing and cons of cost sharing. That review led
to a 2009 policy that limited cost sharing
to five agency programs, including its engi-
requirement for some grants neering research centers program and the
Established Program to Stimulate Competi-
tive Research, which earmarks funds for
Instrumentation and teacher training programs will no states that receive little agency funding.
longer ask institutions to provide matching funds The new law orders NSF to report back
in 5 years on how the waiver for MRI and
Noyce has affected participation and the
By Jeffrey Mervis ing U.S. competitiveness with China, which quality of the research, as well as whether it
President Joe Biden signed into law in Au- should become permanent. It also requires
S
everal U.S. government research pro- gust. Advocates say it’s long overdue. “I’m NSF to assess its impact on the demograph-
grams require financial buy-in from hoping that we’ll see an expanded land- ics of the applicant pool. (Legislators didn’t
institutions when their researchers scape of institutions getting these grants, have baseline data because NSF doesn’t
apply for a grant or new instrument. and that such diversity will strengthen sci- release information on applicants to indi-
The rationale for cost sharing—which ence everywhere,” says Cheryl Hayashi, pro- vidual programs.)
can amount to half of the size of the vost of science at the American Museum of One question is whether the waiver will
award—is to stretch federal dollars and Natural History in New York City, which in result in more proposals from institutions
guarantee that every grantee has a stake in the past has had to go to great lengths to that can’t currently afford to compete. The
the project. But many institutions, includ- line up the matching funds needed to win MRI program currently limits institutions
ing those serving rural areas and students MRI and Noyce awards. to submitting three applications per year,
from groups underrepresented in science, But some observers worry about possi- and top research universities, able to af-
can’t raise enough money to even compete ble negative repercussions. Requiring NSF ford the cost sharing, routinely hit that cap
for the grant. to foot the entire cost of MRI and Noyce to improve their odds of winning. In 2012,
So this year, Congress directed the Na- awards will mean fewer or smaller grants for example, the University of Oklahoma
tional Science Foundation (NSF) to elimi- unless Congress increases each program’s won three MRI awards.
nate the requirement in two of the five budget. (MRI’s $75 million budget now sup- “We didn’t expect it,” says meteorologist
agency programs that require cost shar- ports some 150 awards annually, and Noyce Kelvin Droegemeier, who at the time was
ing and see what happens. For the next makes about 60 to 70 grants a year from its vice president of research at the univer-
5 years, NSF will no longer require univer- $67 million budget. MRI awards range from sity, one of the 146 R1 institutions that
PHOTO: CHRIS WATTS/UNIVERSITY OF HOUSTON
sities and other organizations to cover 30% $100,000 to $4 million, and Noyce grants perform the most research. “But we were
of an award from its major research instru- can be as large as $3 million over 6 years.) able to come up with the required match”
mentation (MRI) program, which funds A more level playing field is also likely of $560,000, recalls Droegemeier, who led
new equipment. It is also ending the 50% to generate more applications, making for NSF’s earlier cost-sharing review by the
match required by two of the four funding even fiercer competition. Ironically, the re- National Science Board and later served
tracks in the Robert Noyce Teacher Schol- sult could be fewer awards to the very insti- as science adviser to former President
arship Program, which trains math and tutions the waiver is meant to aid. Donald Trump.
science teachers. “I think [the change] will open the flood- With roughly one-third of that sum to
The change was tucked into the new gates,” says Timothy Ramadhar, an organic meet cost-sharing requirements in any
CHIPS and Science Act aimed at improv- chemist at Howard University, a historically given year, Gerald Blazey, vice president

for research at Northern Illinois University SCIENCE AND ENVIRONMENT

(NIU), can’t afford to reach the cap. So NIU,
an R2 institution, has typically submitted
only one MRI proposal a year. It has won
five MRI awards in the program’s 30-year
After Lula’s win, ‘a huge relief!’
history, whereas some major research uni- Many in Brazil’s scientific community abhorred Jair
versities have snared more than 50.
Even without the matching require-
Bolsonaro. But his successor faces major challenges
ment, Blazey must still find the money to
pay for operations and maintenance of any By Sofia Moutinho fund’s largest contributor until it halted
new NSF-funded instrument. NSF imposed payments in 2019, after the Bolsonaro
I
the cap on applications because it feared t was “a huge relief!” That’s how Luiz administration unilaterally changed the
research projects might suffer or new in- Davidovich, a physicist at the Federal fund’s targets. About $480 million in fro-
strument might be underutilized if univer- University of Rio de Janeiro’s main cam- zen contributions will now be released.
sities overextended themselves. “We want pus and former president of the Brazil- Other countries welcomed Lula’s promises
to ensure that institutions are committed ian Academy of Sciences, describes his as well.
to the operations and maintenance of what feelings early on the evening of 30 Oc- But so far, details on Lula’s environmen-
are, typically, expensive, durable, shared- tober when it became clear Brazilian voters tal policies—or how he plans to balance
use instruments with a long life span,” says had ousted populist far-right President Jair them with bolstering the economy and
Alicia Knoedler, who leads the NSF office Bolsonaro and given his left-wing rival, for- meeting energy demands—have been scant.
that oversees the MRI program. mer President Luiz Inácio “Lula” da Silva, an Mercedes Bustamante, an ecologist at the
Some researchers hope NSF’s new rules, unprecedented third term. Federal University of Brasília and member
expected out later this year, will remove The sentiment is widely shared in Brazil’s of the Intergovernmental Panel on Climate
that cap. “I’d like to see an open competi- scientific community, where many feared a Change, says the government should adopt
tion, like NSF does for most of its programs, second term for Bolsonaro might be cata- a zero-tolerance policy for environmental
and then fund the strongest proposals,” strophic for issues they care about, includ- crimes such as illegal mining. It also needs
says Pamela Clarke, head of Howard’s re- ing support for science, climate policy, and to rebuild the institutions that monitor and
search development office. Howard, which deforestation. Lula ended up with 50.8% of fight deforestation, which were weakened
wants to regain R1 status by 2024, has won the vote in the second round, the thinnest by the previous administration, she says
an MRI award in each of the past 4 years, margin since Brazil’s return to democracy (Science, 27 May, p. 910).
she notes. in 1989. “Now we have the chance to start As to science, another 4 years of Lula
Another open question for NSF is rebuilding the country’s sci- may not be enough to repair
whether the waiver will encourage institu-
tions to ask for more money. “The size [of
ence and education, and to
resume protecting the Ama-
“Brazil and the the damage from the past
4 years, including obsolete
the proposal] is dictated by cost sharing,”
says Paige Evans, a science educator at the
zon forest,” Davidovich says.
Bolsonaro had stoked
world need lab equipment at public uni-
versities, a lack of research
University of Houston who in 2018 won a doubts about the election’s the Amazon alive.” grants, and a persistent brain
$2.8 million Noyce award to train two co- fairness for months, and Luiz Inácio “Lula” da Silva, drain, says Renato Janine,
horts of 15 high school teachers. Evans early this week, his support- Brazilian president-elect president of the Brazilian So-
notes she increased her request at the last ers blocked roads all over ciety for the Advancement of
minute after university officials said they the country to protest what they say was Science. And science will have to compete
were boosting the size of the match. a rigged result. In a very short speech on with other pressing needs—including pov-
The new policy could also affect how 1 November, Bolsonaro did not formally erty and hunger. Even though Brazil is a
researchers go about assembling propos- concede, but his chief of staff later said the major food producer, the United Nations
als. For example, before microbiologist transition of power would begin. in July added the country to its Hunger
Matthew Fields of Montana State University Deforestation in Brazil exploded under Map of countries where more than 2.5% of
could ask NSF to fund a new $1.1 million Bolsonaro, after dropping during Lula’s the population faces undernourishment,
digital fluorescent microscope for the Cen- first stint at the helm, between 2003 and 8 years after removing it from the list.
ter for Biofilm Engineering that he leads, 2011. In his victory speech, Lula said he “There are a lot of needs to tackle at the
Fields first had to convince a regional foun- would again prioritize environmental pro- same time,” Janine says.
dation, the M.J. Murdock Charitable Trust, tection, promised to crack down on illegal But with the economy faltering, an
to provide some matching funds. Such con- deforestation and mining, and signaled an already-approved budget for 2023 that in-
versations often make for sharper proposals openness to international cooperation to cludes cuts in research and education, and
and better science, Knoedler says. protect the rainforest. “Brazil and the world Bolsonaro’s Liberal Party holding the larg-
Congress wants to make sure institutions need the Amazon alive,” Lula said. “We are est number of seats in Congress, Lula’s
don’t skimp on in-house support for an ready to resume our leading role in the fight choices will be limited. “It is difficult to say
MRI instrument or Noyce grant they didn’t against the climate crisis.” what he will be able to achieve,” says Luiza
help pay for, say Democratic staffers on the Hours after the results were in, Norwe- Duarte, a political scientist and research
House of Representatives science commit- gian environment minister Espen Barth fellow at the Wilson Center in Washington,
tee, which wrote the CHIPS provision. That Eide said his country would resume con- D.C., “but he is a very skilled politician who
possibility also worries Droegemeier. tributions to the Amazon Fund for For- was able to make alliances with political en-
“You need to be careful not to set [institu- est Conservation and Climate Protection, emies before.” j
tions] up to fail,” he says. “You don’t want to which supports programs that prevent
have them fumble the money they receive.” j and combat deforestation. Norway was the Sofia Moutinho is a science journalist in Rio de Janeiro.

AAAS® and Science® are registered trademarks of the American Society for the Advancement of Science, USA. Eppendorf® and the Eppendorf Brand Design are registered trademarks of Eppendorf SE, Germany.
Congratulations
to Ann Kennedy, Ph.D.
Assistant Professor
Northwestern University
Feinberg School of Medicine
Chicago, USA
All rights reserved, including graphics and images. Copyright © 2022 by Eppendorf SE. Photography: Saverio Truglia
Meet the Winner 2022
Eppendorf & Science Prize for Neurobiology The annual US$25,000 Eppendorf & Science Prize
Congratulations to Ann Kennedy, Ph.D. on winning the for Neurobiology honors scientists, like Dr. Kennedy,
2022 Eppendorf & Science Prize for her work on the for their outstanding contributions to neurobiological
neural population dynamics that generate and maintain research. Ann Kennedy is the 21st recipient of this
internal motivational states. By comparing the activation international award. The winner and finalists will be
of neurons in multiple deep brain regions, Dr. Kennedy honored at a Prize Ceremony on November 13, 2022 at
can identify differences between them that point to the the Hard Rock Hotel in San Diego, USA. To register to
role each region plays in shaping survival behaviors such attend this event in person at 6:00 pm PST or to
as fighting or fleeing. Her work contributes to our watch virtually live at 6:45 pm PST, go to
understanding of how our emotional states arise and event.eppendorf.com/prizeceremony
how they influence our actions. or scan the QR code.
You could be next to win this prize.

If you are 35 years of age or younger and currently
performing neurobiological research, you could be next
to win the 2023 Prize. Deadline for entries is June 15, 2023.
Learn more at: eppendorf.com/prize

Chicago, USA
Congratulations
Assistant Professor
to Ann Kennedy, Ph.D.
Northwestern University
Feinberg School of Medicine
Meet the Winner 2022
Learn more at: eppendorf.com/prize

N E WS
FEATU RES
URBAN OASIS
In Berlin, pioneering research into urban ecology
has found surprising biodiversity in the city’s
green spaces By Gabriel Popkin, in Berlin
A
modest cemetery in the heart of says. The insects love the warmth radiat- To be sure, the relatively modest green
this 3.6-million-strong capital city ing from pavement and buildings—what’s spaces found in Berlin and other cities can-
is hardly a likely nature haven. Yet known as the “urban heat island.” And not compensate for the destruction of larger
it was here in the Domfriedhof, they thrive in the diverse plant communi- habitats, nor stop the extinction crisis that
in Berlin’s Mitte district, that ties found outside of manicured and often threatens an estimated 1 million species
Anita Grossmann, an ecologist at chemical-soaked fields and gardens. worldwide. But a growing chorus of eco-
the Technical University of Berlin “You don’t need a lot of space” if you’re logists and environmentalists says urban
(TU Berlin), turned up 19 wild a bee, Grossmann explains. “You just need habitats have been overlooked for too long,
bee species in a single parched pollen, nectar, and nesting space.” to the detriment of many of the plants and
and untended patch of grasses and flowers The findings are the latest to emerge animals conservationists are trying to pro-
on the graveyard’s edge. from one of the world’s longest series of tect. And the Berlin study, they say, high-
Additional surveys soon revealed this studies of a city’s ecosystems. Over more lights the kinds of intriguing discoveries that
metropolis buzzes with bee diversity. The than 6 decades, Berlin researchers have can reward scientists willing to trade pris-
PHOTO: MAURICE WEISS
researchers tallied 106 species living in helped launch urban ecology as a disci- tine forests for weedy lots and pocket parks.
49 grassy plots around and just outside pline, producing sometimes surprising Some of the findings have challenged
Berlin, they reported earlier this year. results that have helped promote conser- ecological orthodoxies: that small, isolated
Inner-city spaces such as the cemetery vation efforts and turn their city into one ecosystems cannot sustain meaningful
are “perfect for wild bees,” Grossmann of the world’s greenest capitals. plant and wildlife populations, and that

N E WS
CITIES ARE GENERALLY known for displacing Ingo Kowarik, who studied with Sukopp and
nature, not conserving it. Buildings, roads, later became a professor at his university.
and parking lots obliterate forests and But the results justified the peril. The re-
wetlands while cars, factories, and power searchers discovered, for example, that in
plants pollute the air and water. Night- areas dominated by nonnative species such
time lights disrupt foraging by nocturnal as black locust trees from North America,
animals. Wild creatures, it’s generally as- many organisms, such as ground beetles
sumed, can find better homes elsewhere. and spiders, were as abundant as they were
But much of elsewhere is becoming too in areas dominated by indigenous plants.
inhospitable. In Germany, for instance, Even many of Berlin’s red-listed species
the total biomass of flying insects in the were making a good living in these “novel
nation’s protected areas has plummeted ecosystems”—a term coined by Australian
by three-quarters since 1963, research- ecologist Richard Hobbs. (Less than 6% of
ers reported in 2017 (Science, 12 May 2017, Berlin’s endangered plants are threatened
p. 576). A follow-up study published in 2019 by nonnatives, according to a recent sur-
found similar results for insects in nearly vey by experts led by Kowarik.) The mix
300 forest and grassland sites. Both stud- of species often includes plants of vary-
ies pointed the finger largely at rural land ing heights and structures, Kowarik notes.
trends. Farms have expanded and consoli- That variation can provide more niches for
dated to blanket much of the landscape, insects and other animals than do many
and agricultural chemicals have wiped out managed landscapes.
native plants and insects, leading to knock- Starting in the 1980s, West Berlin began
on declines of birds and other creatures. to create legally protected nature reserves
Studies of Berlin’s urban ecology began in some of its Brachen. Then in 1989, the
well before such biodiversity crises hit the
headlines. In the 1950s, Herbert Sukopp was
a botanist at TU Berlin, located in the then-
divided city’s western sector, controlled by
the United States, the United Kingdom, and
France. (The Soviet Union controlled East
Berlin and East Germany, making it diffi-
cult for researchers such as Sukopp to leave
West Berlin.) As Sukopp began to investi-
gate his city’s forgotten spaces, he found
habitats unlike any others: World War II had
reduced much of Berlin to rubble and dra-
matically reduced the population of what
was once among the world’s largest cities—
leaving extensive areas abandoned, espe-
cially former industrial zones.
In Berlin’s Südgelände, a former railway At a time when most ecologists sought
yard has given way to a protected out remote, pristine ecosystems, Sukopp
forest that supports numerous species. and his colleagues honed in on these dis-
tinctive disturbed landscapes, known in Ecologist Ingo Kowarik has helped lead innovative
German as Brachen. They reminded him studies of Berlin’s urban ecosystems.
of volcanic rubble—ecological clean slates
nonnative species are always ecologically ripe for species to colonize—he recalled in fall of the Berlin Wall opened a new set
harmful. Others have shown Berlin’s a recent film. of wastelands for Western researchers to
“wastelands”—former industrial sites The Berliners were the first to study study. But it also triggered a multidecade
colonized by novel mixtures of nonnative how city habitats shaped the lives of their real estate boom, as Berlin, once again the
and native species—can harbor as much nonhuman residents. “This was a new idea capital of a united Germany, prospered. As
biodiversity as more natural sites. Grass- of urban ecology,” says Jürgen Breuste, an politicians and investors eyed the Brachen
hoppers, sand lizards, nightingales, and ecologist at the Paris Lodron University of and other areas for development, ecologists
skylarks that are declining or threatened Salzburg. “It developed really in Berlin and armed themselves with their biodiversity
elsewhere have been found thriving in the nowhere else in that format.” Sukopp went data and joined political battles, eventu-
city’s green spaces. Bees are just one of sev- so far as to create “red lists” of plants and ally helping persuade city officials to pro-
eral groups of organisms that often seem animals that were rare and threatened spe- tect key sites.
to prefer heavily urbanized areas. cifically in Berlin—possibly the first urban One such area is Südgelände, a former
“Cities are not these hostile places; cities are use of the extinction risk metric, which is railyard on the south side of the city. On a
PHOTO: MAURICE WEISS
islands of biodiversity,” says Sascha Buchholz, typically applied at larger scales. recent visit, Kowarik reveled in the 70-year-
an ecologist at the University of Münster The work carried risks. When the eco- old forest that has grown up in the park,
who has worked in Berlin for more than a logists tried to survey abandoned railyards, where he has intensively studied the inter-
decade. The world’s cities, he says, offer “a which were controlled by East Berlin, police play of natives and newcomers. “These spe-
chance for mitigating the biodiversity crisis.” sometimes chased them away, recalls ecologist cies evolved in different botanical realms

N E WS | F E AT U R E S
and now they have come together,” he said. Tegel Airport A buzzing metropolis
“That’s really exciting!” Domfriedhof Berlin’s parks and revitalized “wastelands,” such
A few kilometers north at the 300-hectare Gleisdreieck as former airports and railyards, have made it one
former Tempelhof Airport, which was park of the world’s greenest capitals. Ecologists have
decommissioned in 2008, grazing sheep long probed the biodiversity in these green spaces.
and hay-cutting farmers now maintain Tempelhof A recent study found wild bees (below, left) thrive
ICC Berlin Airport
a sprawling grassland. The site has be- in inner-city sites such as cemeteries and pocket
Südgelände
come a key refuge for Eurasian skylarks, a BERLIN parks. By contrast, hoverflies (below, right) prefer
bird that requires open fields that, in the less urban outlying areas. As cities around
countryside, have been largely overrun by Vollguter the globe expand and grow denser, such findings
crop monocultures. Grasslands are now Gemeinschaftsgarten could help planners preserve urban species.
among the most threatened ecosystems in
central Europe. Hoverfly indicator species
Wild bee indicator species
“To integrate so many unusual green
(Lasioglossum morio) (Helophilus trivittatus)
places into the city, with existing wild
vegetation—I think this is a unique feature
of Berlin,” Kowarik says.
THE BERLIN SCHOOL OF ECOLOGY, as it came

to be known, showed that sites like Süd-
gelände and Tempelhof support rich and
often surprising biodiversity. The next step
was to explain why. In 2016 Kowarik and
his colleagues, including Moritz von der
Lippe, Buchholz, and Grossmann, launched
a project called CityScapeLab Berlin to do 0 10
just that. Borrowing an ecological gradi- km
ent concept developed in the United States
and Australia, the researchers quantified Abundance
(individuals at site): 1–5 5–10 10–20 20–30 30–40 40–50 50–100
and mapped the degree of urbanization
throughout Berlin. Such maps can help re-
veal how species respond to “different types
of landscape configurations,” says Amy
Hahs, an ecologist at the University of Mel-
bourne who has helped adapt the method
for urban studies. The Berlin researchers
CREDITS: (GRAPHIC) K. FRANKLIN/SCIENCE; (DATA) A.K. GATHOF ET AL., OECOLOGIA 199, 165 (2022); (PHOTO) SASCHA BUCHHOLZ
also examined a suite of organisms, so their
data would not be biased by the preferences
of just one or a few species.
To generate statistically meaningful re-
sults, they focused on an ecosystem found
throughout sandy Berlin: dry grasslands.
Such areas can host rich mixes of grasses,
wildflowers, and legumes, in part because
their poor soils prevent any one species
from growing too vigorously and taking
over. The researchers ranked 56 grasslands
in and around Berlin by a range of factors
that might make life easier or harder for the
species living in them, such as the amount
of pavement nearby. Urban grasslands are a focus of a research project called CityScapeLab Berlin. Even green patches wedged
The researchers also pioneered the use between highways have been found to support a surprising diversity of native plants, bees, and beetles.
of a metric that describes how habitats are
connected to each other in 3D space. It con- (ground beetles) in pitfall traps, butterflies the rarest plants and beetles, von der Lippe
siders not just the horizontal distance be- in nets, and nocturnal moths with light says. “We found a lot of really amazing en-
tween patches, but also factors such as the traps. They painstakingly tallied plants dangered species on traffic greenery.”
height and number of buildings that could and logged climate data. The researchers discovered that native
hinder animals as they try to move from The work was often far from glamor- plants, wild bees, and ground beetles were
one habitat to another. ous. In a desiccated grassy patch next to at least as diverse in highly urban sites as
In 2017 and again in 2020, the research- a massive 1970s-era conference center, for those located elsewhere, von der Lippe
ers surveyed what lived on their study example, exhaust-spewing diesel trucks says—including many red-listed species.
sites. They put out brightly colored cups rumbled by just meters from where the One reason for the unexpected abundance
of formaldehyde-laced water to attract scientists collected field data. But such might be that—in contrast to intensive
and trap pollinators. They caught carabids unpromising places often harbored among farming—the regular but more moderate

ways that humans disturb urban habitats, eas. So is a less celebrated group of insects, need many more studies of tropical cities, of
such as occasional mowing, can actually parasitoid wasps, which lay their eggs in- cities in dry areas, of cities with very differ-
favor these native species. side other insects and spiders. (Hahs likens ent social-cultural conditions,” he says.
One surprise was that areas disturbed by them to “the large carnivores” of the urban Those studies can come none too soon,
wild boars, which are rampant in Berlin, environment.) many researchers say, as urban ecosystems
hosted more threatened grasshoppers and But these studies have also found discour- face growing peril. Demographers expect
sand lizards than undisturbed areas. Boars aging signs. A global sampling of urban soils, that, by 2050, two-thirds of all people will
are often considered environmentally de- for example, found that cities house a rela- live in cities. Housing shortages could be-
structive, but their rooting in the soil for tively limited set of bacteria, algae, amoebae, come severe, and builders and officials are
food seems to create valuable niches for cer- and fungi. “It seems to be more or less the eyeing green spaces for new development.
tain species, von der Lippe says—much as same community everywhere,” says Nico And as people gain wealth, they often spend
some human activity does. Eisenhauer, an ecologist at the German Cen- money in ways that harm biodiversity—
Another aid to Berlin’s biodiversity could tre for Integrative Biodiversity Research. using herbicides and pesticides in gardens,
be coming from the many trees and for example. Climate change is also
shrubs planted by city officials and predicted to make cities too hot
residents. The plants help animals and dry for many species that have
move within the urban matrix, von long lived in them. “We’re making a
der Lippe suspects, making habitat more and more hostile environment”
fragmentation less harmful to ur- for urban wildlife, says Wolfgang
ban wildlife than ecologists gener- Weisser, a TUM ecologist.
ally assume. Some ecologists are working to
Ecologists have also found evi- change that. In Berlin, Weisser has
dence that Berlin can serve as a ref- teamed with a landscape architect
uge for species from distant regions. to develop a “biodiversity-friendly
In July, for example, Monika Egerer, design” for a residential and com-
an ecologist at the Technical Uni- mercial district being planned for
versity of Munich (TUM), reported the Tegel grounds. The proposal,
the discovery of a bee in a Berlin which must be approved by the de-
community garden that was previ- velopers, builds on Weisser’s work
ously known only from the south of on two new apartment complexes in
Germany. The finding was part of a Munich; they provide nesting and
survey of 18 Berlin gardens in which feeding sites for species known to
Egerer and her colleagues found be harmed by urbanization, includ-
about 400 plant species, including ing a bat, a sparrow, a woodpecker,
four on Germany’s red list. Like the and a hedgehog.
Brachen, Egerer says, gardens can In other cities, researchers have
contain structures—such as dead, persuaded officials to mow lawns
decaying wood—that provide criti- less often to allow flowers to bloom
cal niches for wildlife. for pollinators, and at times en-
An abundance of gardens is part courage residents to do the same.
of what makes Berlin “an incred- Sand lizards thrive in some of Berlin’s wild spaces, especially those Tallamy urges city and suburban
ibly rich city from a conservation disturbed by wild boars rooting for food. dwellers to rewild their yards with
perspective,” she says. And that native plants.
richness can benefit city gardeners, too, And not everyone shares the Berlin re- But cities could be doing much more to
Buchholz adds, as abundant bees found in searchers’ optimism that novel urban eco- support biodiversity, Hahs says. Her team
cities can pollinate flowers and vegetables. systems offer refuges from the increasingly has found that when 30% of an urban green
The Berlin researchers have also turned perilous countryside. Doug Tallamy, an space is covered by shrubs and other mid-
up urban losers. Grossmann and her col- entomologist at the University of Delaware, story plants, more bats, birds, beetles, and
leagues found far fewer hoverflies, a ma- Newark, doubts that nonnative trees, such other animal species live there than in a less
jor group of pollinating insects, in highly as those the Berlin researchers celebrate, can complex habitat. But to incorporate a shrub
urban sites compared with more rural support the same mix of insects and birds layer into urban plantings, she says, green
ones. She suspects the small urban grass- that Europe’s native flora once did. His own space managers would need to accept land-
lands the team studied lacked the moisture studies in the mid-Atlantic United States have scapes that are not neat and tidy, and find
and structural complexity the flies need. found that native trees support larger and ways to allay concerns that bushes could
A study of birds found some species had more diverse populations of caterpillars— encourage crime.
declined in inner-city areas, and wetland essential food for many birds—than do non- “We have this really good body of knowl-
specialists such as dragonflies also fared native trees. A study he is now leading in edge now for ecology in and of cities,” Hahs
poorly, Buchholz says. Europe is producing similar results. “There says. “What’s missing is, how do we trans-
PHOTO: SASCHA BUCHHOLZ
are not different rules of ecology in Europe late that into ecology for cities?”
SIMILAR FINDINGS are now flowing from versus here,” he says. Even in Berlin, researchers say holding on
researchers studying other cities in North Breuste cautions against drawing general to the wildlife they have uncovered will re-
America, Australia, Central America, and conclusions. So far, he notes, nearly all urban quire an ongoing effort. “The city is rapidly
elsewhere. Wild bees, for example, appear ecology research has been done in relatively densifying,” Egerer says. “Are we going to be
to be thriving in many of these urban ar- wealthy cities in the temperate zone. “We able to conserve this high biodiversity?” j

INSIGHTS
PERSPECTIVES
Wildfires are becoming bigger in

the Siberian peatlands, with the
2019−2020 seasons seeing as many
wildfires as the preceding 40 years.
CLIMATE CHANGE
Arctic wildfires at a warming threshold

Bigger wildfires in the Siberian Arctic signal release of more carbon to the atmosphere
By Eric Post1 and Michelle C. Mack2 the 2019 and 2020 wildfire seasons in the terrestrial Arctic, but year-to-year records
Siberian Arctic and predict the extent of of its burned area are sparse.
V
ast amounts of organic carbon carbon-rich soils likely to burn in the area Descals et al. compiled multiple satellite-
are stored in Arctic soils. Much of with future warming. Critically, they sug- based estimates of the annual burned area
this is in the form of peat, a layer gest that even minor increases in tempera- for the Siberian Arctic from 1982 to 2020 to
of decomposing plant matter. Arc- ture above certain thresholds may promote analyze associations between burned area
tic wildfires release this carbon to increasingly larger wildfires. and several factors (see the figure). According
the atmosphere as carbon dioxide Assessment of the relationship between to their analysis across all sources of satellite
(CO2) (1) and contribute to global warm- climate warming and the frequency and data, 2019 and 2020 emerge as the biggest
ing. This creates a feedback loop in which extent of Arctic wildfires is complicated by fire years for the Siberian Arctic, account-
accelerated Arctic warming (2) dries peat- several factors. Satellite data of the annual ing for nearly half of the area burned for
land soils, which increases the likelihood area burned by wildfires in the Arctic may that region over the entire 39-year period
of bigger, more frequent wildfires in the require difficult-to-obtain ground-based and releasing nearly 150 million tonnes of
Arctic and releases more CO2, which fur- validation to improve accuracy. Moreover, carbon to the atmosphere. On 20 June 2020,
ther contributes to warming. Although multiple factors may interact with warm- the Russian town of Verkhoyansk set the re-
this feedback mechanism is qualitatively ing in complex ways to influence fire oc- cord for the highest single-day temperature
understood, there remain uncertainties currence, severity, and extent, such as measured above the Arctic Circle (38°C) (4).
about its details. On page 532 of this is- lightning strikes, rainfall, and fuel load On average, the Arctic region has warmed PHOTO: AP PHOTO/IVAN NIKIFOROV
sue, Descals et al. (3) analyze data from or vegetation cover. Add to this mix the faster than the rest of the globe. Northern
uncertainty that derives from gaps in the peatlands—including those in Asia, North
1
Department of Wildlife, Fish, and Conservation Biology, geographic representation of data across America, and Europe—currently account for
University of California, Davis, Davis, CA, USA. 2Center the Arctic and the challenges seem almost an annual carbon sink of ~100 million tonnes
for Ecosystem Science and Society and Department of
Biological Sciences, Northern Arizona University, Flagstaff, insurmountable. The Siberian Arctic, for (5). The enormous carbon release of 150 mil-
AZ, USA. Email: [email protected] example, represents as much as 70% of the lion tonnes from the 2019 and 2020 Siberian

fires demonstrates how quickly northern eco- test these hypotheses and their general ap- NEUROSCIENCE
systems can switch from carbon sinks to car- plicability to the Arctic region.
bon sources under the continuous warming
of the Arctic.
The authors started with individual sin-
It is worth considering the implications
of increasingly frequent and large wildfires
for the fate of carbon that is currently locked
Gene therapy
gle-predictor models, which mostly show
exponential increases in burnt area across
the Siberian Arctic for each of the individual
away in the permafrost soils and sediments
that underlie much of the Arctic. Increased
combustion of the insulating peat layer can
for epilepsy
drivers. These include the increases in tem- expose more permafrost and lead to the thaw On-demand inhibition
perature, vapor-pressure deficit (the ability
of the air to dry the land surface), climatic
and decomposition of an even larger reser-
voir of organic matter, releasing carbon that
of neuronal activity
water deficit (more water being evaporated has been stored underground for centuries or reduced spontaneous
relative to precipitation), and the number of even millennia (8). Larger and more intense
ignition events presumably related to light- wildfires could substantially accelerate the
seizures in mice
ning strikes. Building on the single-predictor release of permafrost carbon into the atmo-
models, the authors then created a multivari- sphere (9), but this interaction is not consid- By Kevin Staley
ate model, which revealed that some of the ered in current forecasts of Arctic feedback
E
single-predictor drivers can themselves be to global warming (10). Future studies that pilepsy is the predisposition for spon-
driven by an increase in temperature. For link rigorous assessment of wildfires with the taneous episodes of synchronous in-
example, warming can directly increase the dynamics of permafrost thaw in these remote creases in neuronal activity, called
number of ignition events and indirectly in- regions are therefore needed to better quan- seizures. The propensity for seizures
crease plant water stress by increasing the tify their impact on climate. j can be reduced with antiseizure medi-
vapor-pressure deficit. This in turn can dry cations. But many patients do not re-
RE FE RE NCES AND NOT ES
deeper soil layers and contribute to plant spond (1), develop unacceptable side effects
1. M. C. Mack et al., Nature 475, 489 (2011).
water stress. By linking these processes and 2. M. Rantanen et al., Commun. Earth Environ. 3,
(2), or respond initially but subsequently
identifying the direct and indirect effects of 168 (2022). develop tolerance (3). A closed-loop feed-
warming on increasing burn area, Descals et 3. A. Descals et al., Science 378, 532 (2022). back system provides transient, on-demand
al. provide insights into what the future of 4. M. Allen, Eos 102 (2021). treatment in response to a suprathreshold
5. G. Hugelius et al., Proc. Natl. Acad. Sci. U.S.A. 117,
Arctic wildfires may look like under acceler- 20438 (2020).
stimulus and is a potentially powerful so-
ating warming. 6. C. G. Collins et al., Nat. Commun. 12, 3442 (2021). lution to the problems of medication re-
According to their analysis, warming of 7. I. Myers-Smith et al., Nat. Clim. Chang. 10, 106 (2020). sistance, side effects, and tolerance (4). On
mean summer air temperature past a thresh- 8. E. A. G. Schuur et al., Nature 520, 171 (2015). page 523 of this issue, Qiu et al. (5) demon-
9. S. M. Natali et al., Proc. Natl. Acad. Sci. U.S.A. 118,
old of 10°C, or of mean summer surface strate a genetic closed-loop feedback system
e2100163118 (2021).
temperature above 17°C, would cause dis- 10. Intergovernmental Panel on Climate Change, Climate in mice that is designed to inhibit neurons
proportionately large increases in the extent Change 2021: The Physical Science Basis. Contribution that participate in seizure activity.
of carbon-rich soils burned in the Siberian of Working Group I to the Sixth Assessment Report of the Clinically available closed-loop feedback
Intergovernmental Panel on Climate Change (Cambridge
Arctic. However, patterns of both local Univ. Press, 2021). systems use electrographic seizure detec-
warming (2) and vegetation change (6, 7) are tors and intracranial electrodes; they are
highly variable across the Arctic. Therefore, ACK NOWLE DGM E NTS not capable of inhibiting neurons and can
additional studies in other regions of the The authors thank P. F. Sullivan for comments only be placed in a limited number of lo-
on the manuscript.
Arctic that harbor vast expanses of peatland, cations (6). Qiu et al. took a different ap-
such as Canada and Alaska (5), are needed to 10.1126/science.ade9583 proach to feedback: temporarily inhibiting
the activity of all the neurons that partici-
pated in the last seizure. To do this, they de-
The effect of Arctic wildfires on carbon release veloped a genetic strategy based on the Fos
Arctic wildfires accelerate the release of organic carbon from the soil into the atmosphere, which can gene, whose expression is up-regulated by
strengthen the feedback to warming. neuronal activity, including seizures (7, 8).
The activity-sensitive promoter region
of Fos was used to drive the expression
CO2 capture CO2 release Water CO2 capture CO2 release of a gene encoding an inhibitory protein,
voltage-gated potassium channel subunit
(Kv1.1, which is encoded by Kcna1). Kv1.1
is opened by moderate cytoplasmic mem-
brane depolarization (9). The open chan-
nel permits efflux of potassium ions and
hyperpolarization of the membrane, which
reduces action potential generation and
GRAPHIC: V. ALTOUNIAN/SCIENCE
Peat neurotransmitter release by the neuron. An

Surface
Permafrost vegetation adeno-associated virus (AAV) vector encod-
ing the Fos promoter and Kcna1 was used to
Past Present transfect neurons. When the gene therapy
Arctic peatlands, forests, and tundra are Higher temperatures dry the peat layer and drive more
generally carbon sinks. Cold temperatures active weather systems, which lead to more frequent
was expressed in cultures of mouse neurons
and wet soils keep the land relatively moist, lightning strikes, creating larger fires that release more exhibiting spontaneous synchronous activ-
which reduces wildfire activity carbon to the atmosphere. ity, multielectrode arrays demonstrated a
SCIENCE science.org 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 47 1

INSI GHTS | P E R S P E C T I V E S
substantial reduction in the activity of indi- the AAV Fos promoter-Kcna1 therapy. After in nonepileptic mice nor in mice subjected
vidual neurons as well as reduced synchro- 2 weeks for vector expression, seizures were to a seizure 24 hours previously (in whom
nous activation of populations of neurons. detected during another 2-week video EEG exogenous Kv1.1 should be maximally ex-
To test the system in the intact brain, recording, and the number of seizures was pressed). This finding is encouraging for
Qiu et al. injected the AAV vector into the significantly decreased. the proposed therapeutic strategy. It also
mouse hippocampus, a particularly seizure- It might be expected that the seizure fre- provides a new, if somewhat cryptic, clue
prone region of the brain that is also criti- quency decreased because transient, sei- about the nature of memory encoding: The
cal for memory formation. After seizures zure-induced expression of Kv1.1 increased memory engram is not influenced by the
that were induced by the convulsant pen- the interval between seizures. But it was suppression of excitability in the subset of
tylenetetrazol, they found that the Fos pro- not obviously the case that the intervals neurons expressing exogenous Kv1.1.
moter drove expression of a reporter gene increased in this manner. The reason may Would this system operate in human
throughout the hippocampus, particularly be that epilepsy is also characterized by neurons? Qiu et al. treated cultures of neu-
rons induced from human stem cells with
the AAV FOS promoter-KCNA1 gene ther-
Modulating neuron activity during seizures apy. Although these cultures did not gen-
The Fos promoter is engineered to drive the expression of voltage-gated potassium channel subunit (Kv1.1, erate seizures, convulsants increased neu-
which is encoded by Kcna1) in an adeno-associated virus (AAV) vector that is injected into the hippocampi of ronal activity, which drove FOS-mediated
mice to suppress seizure-inducing neuron activity. Kv1.1 expression and, in turn, decreased
activity. These findings indicate that the
FOS promoter-KCNA1 feedback system is
operative in human neurons. Whether the
Seizures occur through spontaneous episodes
of increased synchronous neuronal activity in vector is also selectively expressed in ex-
the hippocampus at random intervals. citatory human neurons and whether the
Time system can prevent seizures in the human
brain remain to be discovered.
The results of Qiu et al. describe a so-
If seizures drive Fos-mediated Kcna1 expression,
they are inhibited for as long as the potassium phisticated, localizable new strategy for
channel protein is in the membrane. Seizures closed-loop feedback to control seizures.
occurring after longer intervals are not prevented The group has been steadily improving
because Kv1.1 expression wanes (red). this strategy (9), and many elements of
Time the current system are more attractive
Frequent brief bursts of synchronous neuronal than hardware-based closed-loop feedback
activity, or spikes, can also drive Fos-mediated systems. Welcome improvements would
Kcna1 expression, which would result in more include the abilities to tune the threshold
constant Kv1.1 expression and consequently and duration of Kv1.1 expression, encode
Time consistent seizure inhibition.
cell-specific expression, and provide a
The graphs are illustrative, and y axes are not to scale. shut-off mechanism. It is also important to
find out more about which neurons need
in excitatory neurons. This was fortunate frequent, brief bursts of synchronous neu- to be inactivated—every neuron that is re-
because suppression of the activity of in- ronal activity that generate spikes on the cruited into a seizure, or only those that
hibitory interneurons might have promoted electroencephalogram (11). These spikes participate in seizure initiation (14)? j
rather than suppressed seizures. Fos-driven were sufficient to drive Fos activity in ani-
expression of Kv1.1 in the hippocampus re- mals that were not having seizures and
1. P. Kwan, S. C. Schachter, M. J. Brodie, N. Engl. J. Med.
duced the severity of seizures induced by so would be expected to also drive Kv1.1 365, 919 (2011).
pentylenetetrazole 24 hours after a prior expression. The populations of neurons 2. J. A. Witt, C. Helmstaedter, Epilepsy Behav. 26, 450
seizure that had activated Kv1.1 expression. that participate in spikes and seizures are (2013).
3. W. Woldman, M. J. Cook, J. R. Terry, Epilepsy Behav. 94,
Seizures were not entirely suppressed, but thought to overlap considerably, although 264 (2019).
this is expected: Systemic convulsant treat- the extent of this overlap varies with time 4. F. T. Sun, M. J. Morrell, R. E. Wharen Jr.,
ment broadly affects neurons in the brain, (12). Spike-driven Kv1.1 expression could Neurotherapeutics 5, 68 (2008).
5. Y. Qiu et al., Science 378, 523 (2022).
whereas the anticonvulsant treatment was unlink its expression from seizures, pro- 6. K. L. Dell, M. J. Cook, M. I. Maturana, Curr. Treat. Options
confined to the hippocampus. viding a more consistent inhibition of epi- Neurol. 21, 47 (2019).
Using conditions that more closely ap- leptic neurons (see the figure). 7. M. Sheng, M. E. Greenberg, Neuron 4, 477 (1990).
8. J. Y. Joo et al., Nat. Neurosci. 19, 75 (2016).
proximate human epilepsy, Qiu et al. injected The Fos gene has long been studied as 9. A. Snowball et al., J. Neurosci. 39, 3159 (2019).
the convulsant kainic acid into the hippo- a model of how the brain learns from ex- 10. F. Twele, K. Töllner, M. Bankstahl, W. Löscher, Epilepsia
campi of mice to render them chronically perience through altered gene expression Open 1, 45 (2016).
11. M. de Curtis, J. G. R. Jefferys, M. Avoli, in: Jasper’s Basic
epileptic—that is, subject to spontaneous sei- (7, 13). The findings of Qiu et al. support
Mechanisms of the Epilepsies [Internet], J. L. Noebels, M.
zures (10). Seizures were identified by using the idea that experience in the form of sei- Avoli, M. A. Rogawski, R. W. Olsen, A. V. Delgado-Escueta,
video electroencephalography (EEG) over a zures strongly activates the Fos promoter. Eds. (National Center for Biotechnology Information, ed GRAPHIC: N. CARY/SCIENCE
2-week baseline observation period, then the But would the exogenously expressed Fos 4, 2012).
12. E. C. Conrad et al., Brain 143, 554 (2020).
mice were injected in the hippocampus with promoter interfere with endogenous Fos- 13. F. T. Gallo, C. Katche, J. F. Morici, J. H. Medina, N. V.
driven experiential learning? Qiu et al. Weisstaub, Front. Behav. Neurosci. 12, 79 (2018).
found no evidence that seizure-induced 14. L. A. Lau, K. J. Staley, K. P. Lillis, Brain 145, 531 (2022).
Department of Neurology, Massachusetts General
Hospital, Harvard Medical School, Boston, MA, USA. expression of Kv1.1 interfered with the Fos
Email: [email protected] system for experience-dependent learning 10.1126/science.ade8836

QUANTUM TECHNOLOGY
An optical interface for quantum networks

A silicon atom embedded in diamond can be entangled with a photon
By Dorian Gangloff an efficient interface to convert between is long. Among them are the integration
photonic and matter “quantum bits,” or of diamond defect centers with photonics
C
ommunication networks that use “qubits”; it must maintain a high degree and electronics (7), their ability to be cre-
the quantum properties of photons of control over the matter qubit; and it ated and arranged in large arrays (8), their
and matter for transferring data are must be able to convert the information information storage time of up to a minute
fundamentally more secure than in the matter qubit to a memory qubit for (9), and their operation of a qubit that is
traditional networks. The physical longer-term storage. For practical reasons, robust to control and environmental-noise
implementation of such a quantum the device should also be able to operate errors (10).
network requires special devices that can close to room temperature instead of ex- Stas et al. show that the silicon-vacancy
convert stored information to a quantum tremely low temperatures, which are often center in diamond is a promising system
carrier—similar to how a normal computer required of systems that use superconduct- for producing and maintaining entangle-
converts bits on a hard drive to a fiber oping components. The entanglement be- ment with photons, which may enable ef-
tic signal (1). To enable this conversion, tween the photon and matter qubits must ficient quantum communication between
the device must be able to generate quan- also be producible on a large scale for prac- more end users and over longer distances
tum entanglement between the stationary tical applications, and the device must be (6). The term “silicon-vacancy center” may
quantum bit and the photon used for data able to maintain the stability of the matter seem like a misnomer, because it refers to
transfer. Once generated, the entangled qubit long enough for photonic qubits to a silicon atom sitting in a vacancy created
photon and quantum bit can then be used travel between nodes. by two missing carbon atoms in the dia-
to perform various tasks, such as to gener- A fully functional node that fulfills all mond lattice (11). The silicon-vacancy cen-
ate and send out an encryption key (2). On these requirements does not yet exist, ter is the perfect tool for converting quan-
page 557 of this issue, Stas et al. (3) pre- but several systems have come close, with tum information into photons because the
sent a design for a quantum network inter- varying degrees of success. These include photon energy is almost completely un-
face and storage device, called a quantum designs using atoms suspended in a vac- disturbed by the surrounding lattice. The
network node, by using embedded silicon uum (4), semiconductor nanostructures vacancy structure within the surrounding
atoms inside a diamond. (e.g. quantum dots) (5), and defect centers carbon lattice protects the silicon atom
There are several fundamental and in group IV semiconductor materials such from vibrations and the electrical noise of
practical requirements for a quantum as diamond (6). The list of achievements charging and discharging that pervades
networking node, where entanglement is based on the use of diamond defect cen- material surfaces, making it ideal for a
generated and stored. The node must have ters in prospective quantum technologies nanophotonic light-matter interface. Stas
et al. use a nanostructured photonic cavity
as an efficient interface to the silicon va-
A quantum network interface using a silicon vacancy cancy and its quantum bit consisting of an
A silicon-29 (29Si) nucleus in place of two missing carbon nuclei embedded in diamond can be used as an electronic spin—a fundamental quantum
interface between a single-photon quantum bit and the 29Si nucleus used as a nuclear spin quantum memory. property of particles related to magnetism
(see the figure) (12).
The authors selected a specific iso-
tope, silicon-29 (29Si), to be embedded at
Diamond the vacancy site. The magnetically active
nucleus in 29Si (compared to nonmagnetic
silicon-28, the most abundant isotope of
silicon) helps prolong the storage time by
C C Designed a thousand-fold, up to 2 s. Because of the
nanostructure
of the diamond coupling between the spin of the silicon
nucleus and that of the electron, the op-
tical property of the diamond nanostruc-
e– 29
Si Silicon vacancy Entanglement ture, which acts as the photonic interface,
depends on the state of the nuclear spin.
Photon
This chain of dependency involving the
Photon
photon, the silicon nucleus, and its elec-
tron, and the optical property of the dia-
GRAPHIC: A. FISHER/SCIENCE
mond nanostructure, enables the system to

be an effective device to entangle a photon
Before entanglement After entanglement and a long-lived storage element. Once the
An incoming photon traveling through the The optical property of the diamond nanostructure
nanostructured diamond toward the silicon depends on the state of the nucleus, which
vacancy interacts with the 29Si nucleus generates an entanglement between the outgoing Department of Engineering Science, University of Oxford,
through the electron quantum bit. photon and silicon atom. Oxford, UK. Email: [email protected]

entanglement is established, the photon

can be sent out and be used for quantum
communication. The entanglement be-
tween the photon and the silicon nucleus
can be maintained for ~2.5 ms, which is
among the longest times recorded, and al-
lows the photon to travel ~500 km before
losing entanglement.
To test their device in a practical set-
ting, Stas et al. created a working photon-
nucleus entangling (PHONE) quantum
operation and extended the ability of the
typical quantum operation to generate an
entangled state between a photon and the
nuclear spin of the silicon. During this pro-
cess, the electron spin provides an added ASTRONOMY
benefit to the system. In the ideal PHONE
gate, the logical state of this electron qubit
should remain unchanged. However, if its
state changes, it would mean that some-
Neutrinos unveil hidden
thing went wrong with the data transfer.
As a result, the electron qubit can be used
as a natural error detection scheme, and
galactic activities
Stas et al. show that by being able to moni- An obscured supermassive black hole may be
tor errors, the effective lifetime of the en- producing high-energy cosmic neutrinos
tangled state was extended from 1.5 ms to
the reported time of 2.5 ms.
By looking through nine silicon-vacancy By Kohta Murase1,2,3 hole jets, or in intergalactic space between
centers using optical spectroscopy and those objects and Earth. Revealing the ori-
A
picking out the two with the most strain— fter relying on nothing but light to gin of cosmic neutrinos and the relationship
i.e., where the nanostructure surrounding study the cosmos for centuries, the among neutrinos, gamma rays, and cosmic
the silicon atom is the most deformed— development of multimessenger rays is crucial to deciphering the fundamen-
Stas et al. were able to create a quantum astronomy in the past decade has tal processes that occur throughout the Uni-
node that operates at 4.3 K. An even larger revolutionized the field (1). The suc- verse (1). Neutrinos are also special because
strain surrounding the silicon vacancy cessful detections of the additional they can travel through dense environments
may reduce the temperature sensitivity messengers—i.e., information carriers other that are impassable for photons, which allow
of the spin qubit more and further extend than light—are only possible because of astronomers to “see” behind gas, dust, and
the storage lifetime (13). Although still an state-of-the-art, large-scale instruments, such radiation obstructions.
extremely low operating temperature, 4.3 as the detectors at the IceCube Neutrino Active galactic nuclei (AGNs), which ex-
K is 43 times warmer than previous sys- Observatory, which are located a kilometer ist at the centers of some galaxies, are en-
tems using silicon vacancy (14). The tem- deep inside the Antarctic ice. High-energy ergized by either the accretion of gas onto a
perature of 4.3 K also crosses a meaningful cosmic neutrinos have given rise to both sur- supermassive black hole and/or the spin of
bar at 4.2 K, which is the boiling point of prises and mysteries since their detections the black hole at the center of the galaxies.
liquid helium. This means that benchtop were reported in 2013 (2, 3). On page 538 of They are among the most promising candi-
cooling systems can be used to operate this issue, the IceCube Collaboration (4) pre- date sources of not only the highest-energy
this device, and the lower cost of entry to sents evidence for a neutrino source in the cosmic rays but also high-energy cosmic neu-
experimenting with these systems should data recorded between 2011 and 2020. The trinos observable from Earth (5, 6). In 2018,
enable more research in the future. j origin of the neutrinos has been traced to the the first candidate source of high-energy neu-
Messier 77 galaxy, also known as NGC 1068. trinos was reported (7, 8). The object, TXS
REFER ENCES A ND NOT ES
In space, high-energy neutrinos are mainly 0506+056, is a blazar, which is a type of AGN
1. S. Wehner, D. Elkouss, R. Hanson, Science 362,
eaam9288 (2018).
produced by cosmic rays—i.e., protons and with jets consisting of particles and magnetic
2. A. K. Ekert, Phys. Rev. Lett. 67, 661 (1991). atomic nuclei that are traveling at nearly the fields ejected to point at Earth with velocities
3. P.-J. Stas et al., Science 378, 557 (2022). speed of light—colliding with matter or ra- approaching the speed of light. The detection
4. D. P. Nadlinger et al., Nature 607, 682 (2022). diation, which creates subsequent particles of a high-energy neutrino and data from ra-
5. R. Stockill et al., Phys. Rev. Lett. 119, 010503 (2017).
6. M. Pompili et al., Science 372, 259 (2021).
that later decay into neutrinos and gamma dio, optical, ultraviolet, x-ray, and gamma-ray
7. N. H. Wan et al., Nature 583, 226 (2020). rays. Such collisions occur inside astro- observations enabled a comprehensive look
8. Y.-C. Chen et al., Optica 6, 662 (2019). physical objects, such as galaxies and black at the blazar jet (7). However, the observed
9. C. E. Bradley et al., Phys. Rev. X 9, 031045 (2019). x-ray and gamma-ray activities were less than
10. M. H. Abobeih et al., Nature 606, 884 (2022).
IMAGE: NASA/JPL-CALTECH
11. C. Hepp et al., Phys. Rev. Lett. 112, 036405 (2014). 1

what was expected based on the detected
Department of Physics, Department of Astronomy and
12. C. T. Nguyen et al., Phys. Rev. Lett. 123, 183602 (2019). Astrophysics, and Institute for Gravitation and the Cosmos, neutrinos (7, 8), and a consensus on how to
13. S. Meesala et al., Phys. Rev. B 97, 205444 (2018). Pennsylvania State University, University Park, PA, USA. explain this discrepancy has not yet been
2
14. M. K. Bhaskar et al., Nature 580, 60 (2020). School of Natural Sciences, Institute for Advanced Study, reached (1). So far, searches relying on the
Princeton, NJ, USA. 3Center for Gravitational Physics and
Quantum Information, Yukawa Institute for Theoretical population of gamma ray–detected blazars
10.1126/science.ade6964 Physics, Kyoto, Japan. Email: [email protected] have suggested that this class of AGN alone

The detection of neutrinos from the Messier 77 galaxy neutrino production is expected to occur NEUROSCIENCE
suggests that particles are accelerated by the black near the black hole. The excess of neutrino
hole at the galactic center, which is obscured by thick
clouds of gas and dust in this artist’s concept.
events observed from NGC 1068 may be ex-
plained by converting a part of the power ex-
tracted from accreting gas to cosmic rays (13,
Molecular
would not explain the amount of all cosmic
neutrinos (9) coming from every direction
in the sky (not every detected neutrino has
14), which then produce neutrinos in subse-
quent processes. This hypothesis proposed by
the theorists is consistent with the gamma-
diversity of
a well-defined source of origin). This has
prompted ideas for other classes of neutrino
sources besides blazars.
ray data, which have an energy flux much
lower than the reported neutrino energy flux
(4). The existence of such a “hidden” neu-
astrocytes
The IceCube Collaboration (4) found an trino source has previously been suggested The tissue environment
excess of 79 neutrinos over the background by analysis of the neutrino data from the influences astrocyte form
associated with NGC 1068 at a statistical sig- IceCube Neutrino Observatory and gamma-
nificance of 4.2s (or standard deviations). ray data from the Fermi Gamma-ray Space and function in health
The result strengthens the previous report Telescope (15). Radio-quiet AGNs, including and disease
of neutrinos originating from NGC 1068 (10). NGC 1068, and other low-luminosity AGNs,
This improved significance was driven by which are more abundant than blazars and
updates in the techniques for reconstructing radio-loud AGNs, might help explain the By Katherine T. Baldwin
detected neutrino events and data calibra- amount of all cosmic neutrinos observed by
A
tion methods, which made the location of the the IceCube Neutrino Observatory (9). strocytes are morphologically com-
neutrino excess more precisely aligned with More data are needed to firmly estab- plex glial cells that regulate essential
the location of NGC 1068. The improved data lish NGC 1068 and other AGNs as neutrino aspects of central nervous system
characterization methods enable more pre- sources. Observationally, to push the statistics (CNS) function, including synapse
cise estimates for the energy and direction of over the bar of 5s, next-generation planned formation (1), ion homeostasis, and
each detected neutrino observed in the sky. detectors, such as IceCube-Gen2 at the South neurovascular coupling (2). Once
The source of the detected neutrinos— Pole and Pacific Ocean Neutrino Experiment viewed as a homogeneous population of
NGC 1068—is among the brightest and clos- off the Pacific coast of Canada as well as cells, astrocytes are now known to exhibit
est AGNs from Earth. It is located in the con- KM3NeT in the Mediterranean Sea and Baikal molecular and functional heterogeneity at
stellation Cetus, at a distance of ~46 million Gigaton Volume Detector in Russia, will all the regional and cellular level (3). How this
light-years away. Optical images of NGC 1068 play important roles. In addition, a synthesis diversity arises is unclear. On page 514 of
show a rather normal barred spiral galaxy, of multiwavelength observations from radio this issue, Endo et al. (4) investigate the
with a bright starburst inner disk and star- up to gamma-ray wavelengths will continue molecular basis of astrocyte heterogene-
forming regions in the outer arms. The ga- to be crucial to reveal the multimessenger ity in mice, uncovering shared core fea-
lactic nucleus is classified as a type of AGN connections and underlying physics. There tures and identifying key differences that
that is characterized by emission lines in is also much progress in numerical simula- are specific to CNS regions. Their findings
the optical spectrum. It is also considered a tions for studying accretion flows, outflows suggest that astrocyte diversity arises pri-
radio-quiet AGN that does not have power- like jets and winds, and plasma processes in marily from differences in the tissue mi-
ful jets, as seen in blazars. Its nucleus is lu- AGNs (5). The continued collection of mul- croenvironment and highlight a key role
minous not only in the optical and infrared timessenger data will offer rich potential for for astrocyte morphological complexity in
bands, but also in ultraviolet and x-ray bands, understanding the detailed nature of plasma brain health and disease.
which are thought to largely come from the and particle acceleration in the vicinity of a Astrocytes are ubiquitous throughout
accretion disk and hot corona near the su- supermassive black hole. j the CNS. During development, astrocytes
permassive black hole at the galactic center. arise from neural progenitors in the ven-
The black hole is estimated to be ~15 million tricular zone (VZ) and migrate to differ-
1. P. Mészáros, D. B. Fox, C. Hanna, K. Murase, Nat. Rev.
times as massive as the Sun and is considered Phys. 1, 585 (2019). ent CNS regions, where they undergo local
to be one of the most obscured supermassive 2. IceCube Collaboration, Phys. Rev. Lett. 111, 021103 (2013). clonal expansion and morphogenesis (5).
black holes known, being blanketed by thick 3. IceCube Collaboration, Science 342, 1242856 (2013). Morphological complexity is a hallmark
4. IceCube Collaboration, Science 378, 538 (2022).
gas and dust. X-ray observations from the 5. R. Blandford, D. Meier, A. Readhead, Annu. Rev. Astron. feature of astrocytes. All astrocytes possess
Nuclear Spectroscopic Telescope Array indi- Astrophys. 57, 467 (2019). densely branched arbors that directly con-
cate that the x-rays are largely intervened by 6. F. W. Stecker, C. Done, M. H. Salamon, P. Sommers, Phys. tact neighboring cells and structures (6) to
Rev. Lett. 66, 2697 (1991).
the gas (11), and radio data from the Atacama 7. IceCube Collaboration, Science 361, eaat2890 (2018).
actively control the formation and function
Large Millimeter/submillimeter Array have 8. IceCube Collaboration, Science 361, 147 (2018). of neuronal circuits and execute a diverse
suggested a compact gaseous structure with 9. IceCube Collaboration, Phys. Rev. Lett. 125, 121104 array of homeostatic functions (7). Overall,
(2020).
a radius of ~10 light-years around the black 10. M. G. Aartsen et al., Phys. Rev. Lett. 124, 051103 (2020).
the morphological complexity of astro-
hole (12). Once the x-ray emission is corrected 11. A. Marinucci et al., Mon. Not. R. Astron. Soc. Lett. 456, L94 cytes is a prerequisite for their functional
by accounting for absorption by the gas, NGC (2016). complexity.
12. S. García-Burillo et al., Astrophys. J. Lett. 823, L12 (2016).
1068 has the brightest x-ray emission among Astrocytes in different regions of the
13. K. Murase, S. S. Kimura, P. Mészáros, Phys. Rev. Lett. 125,
radio-quiet AGNs in the sky region targeted 011101 (2020). mouse CNS show variations in gene ex-
by the IceCube Neutrino Observatory, which 14. Y. Inoue, D. Khangulyan, A. Doi, Astrophys. J. Lett. 891, L33 pression, morphology, and function (8).
helps explain the neutrino excess. (2020).
15. K. Murase, D. Guetta, M. Ahlers, Phys. Rev. Lett. 116,
Because the supermassive black hole at the 071101 (2016). Department of Cell Biology and Physiology, Neuroscience
center of NGC 1068 is obscured by gas and Center, University of North Carolina, Chapel Hill, NC, USA.
dust and is active with radiation, efficient 10.1126/science.ade4190 Email: [email protected]

Heterogeneity is also observed among astro- findings provide strong evidence that astro- Among the genes that correlated strongly
cytes within the same CNS region (9). What cyte regional heterogeneity is shaped by the with astrocyte territory size, Endo et al. iden-
is the source of astrocyte diversity? Some as- tissue microenvironment. tified several genes associated with increased
pects of heterogeneity are developmentally In addition to molecular heterogeneity, risk for Alzheimer’s disease (AD). In an AD
encoded. In the spinal cord, for example, Endo et al. characterized the morphologi- mouse model, the authors observed down-
astrocyte location and gene expression are cal features of astrocytes from the same 13 regulation of several genes that positively
determined by the site of origin of progeni- CNS regions and found considerable regional correlate with astrocyte territory size and up-
tor cells in the VZ (10). Heterogeneity may differences in territory size, sphericity, and regulation of genes that negatively correlate
also arise locally (11), although the extent branching complexity. For example, astro- with territory size. Accordingly, the territory
to which differences in the tissue microen- cytes in the motor cortex displayed the larg- volume of cortical astrocytes was reduced in
vironment influence regional astrocyte di- est territory size of all CNS regions, whereas AD-model mice. Additionally, many of these
versity is largely unexplored. Questions also striatal astrocytes were among the smallest. astrocyte territory-associated genes over-
remain regarding the relationship between Both motor cortex and striatal astrocytes lapped with genes associated with other
molecular diversity and morphological and had a relatively round shape, whereas as- CNS disorders, including Parkinson’s dis-
functional complexity, as well as the rel- trocytes in the cerebellum had the lowest ease, multiple sclerosis, amyotrophic lateral
evance of these features to disease. degree of circularity and the longest length sclerosis, and epilepsy. This observation
To address these knowledge gaps, Endo (see the figure). Intriguingly, astrocyte mor- suggests that changes in astrocyte morphol-
et al. analyzed anatomical, mo- ogy are a shared feature of neu-
lecular, and morphological fea- rological disorders.
tures of astrocytes across 13 dif- Region-specific features of astrocytes The findings of Endo et al. re-
ferent CNS regions from mice. Astrocytes in different regions of the central nervous system of mice show veal much about the heterogene-
Although neuronal density heterogeneity. The tissue microenvironments of these regions influence the ity of astrocytes, but it remains
differed substantially between molecular identity of astrocytes, which results in region-specific differences in unclear how local cues shape
regions, astrocyte density was gene expression and morphological complexity. For example, astrocytes in the astrocyte identity. The transcrip-
relatively constant, underscor- mouse motor cortex (blue) are larger in territory size than astrocytes in the tomic datasets will serve as a
ing their importance for brain striatum (orange), whereas cerebellar astrocytes (green) are the most elongated. starting point to explore how
homeostasis. Accordingly, as- transcription factors and up-
trocytes from all brain regions Tissue microenvironment Molecular identity Morphological complexity stream ligand-receptor signaling
shared expression of genes regulate astrocyte gene expres-
related to core astrocyte func- sion. These data can also be ap-
Astrocyte subsets
tions, including neurotrans- plied to investigate the cellular
mitter homeostasis, choles- mechanisms that underlie key
terol biosynthesis, and glucose astrocyte functions, such as se-
metabolism. The expression cretion of synaptogenic factors,
of many transcriptional regu- regulation of ion homeostasis,
Motor cortex Striatum Cerebellum
lators and ligand-dependent and lipid synthesis, in the con-
nuclear receptors was also text of the tissue microenviron-
shared among astrocytes across ment. Whether astrocyte het-
regions, suggesting that com- erogeneity is developmentally
mon mechanisms of upstream regulated and reversible is also
regulation may control core unknown. Exploring the up-
functional properties. stream regulation of astrocyte
To investigate the extent of heterogeneity, and to what ex-
astrocyte diversity across CNS tent astrocyte identity is plastic
regions, Endo et al. focused on in the context of development
astrocyte-enriched differentially expressed phological diversity correlated strongly with and aging, will have important implications
genes (DEGs) that were distinct to a partic- two gene modules from the region-specific for treating astrocyte dysfunction in neuro-
ular region or shared by a subset of regions. DEGs. The authors selected top morphology- logical disorders. j
DEGs from the 13 CNS regions clustered linked genes from these modules for further
into three broad regions based on anatomi- investigation. Deletion of morphology-linked
1. N. J. Allen, C. Eroglu, Neuron 96, 697 (2017).
cal proximity: cerebrum, brain stem–spinal genes from astrocytes in mice resulted in 2. A. Verkhratsky, M. Nedergaard, Physiol. Rev. 98, 239
cord, and cerebellum. Furthermore, astro- morphological and synaptic deficits and im- (2018).
3. M. Y. Batiuk et al., Nat. Commun. 11, 1220 (2020).
cyte region-enriched DEGs correlated with paired spatial memory. Together, these find- 4. F. Endo et al., Science 378, eadc9020 (2022).
region-enriched DEGs from bulk tissue, in- ings establish a functional link between gene 5. S. Clavreul et al., Nat. Commun. 10, 4884 (2019).
dicating that the identity of astrocytes re- expression, astrocyte morphology, and func- 6. B. Torres-Ceja, M. L. Olsen, Curr. Opin. Neurobiol. 74,
102550 (2022).
flects the tissue microenvironment. Single- tional output. 7. N. J. Allen, D. A. Lyons, Science 362, 181 (2018).
cell sequencing analysis of a subset of brain Astrocyte dysfunction is observed in many 8. H. Chai et al., Neuron 95, 531 (2017).
9. O. A. Bayraktar et al., Nat. Neurosci. 23, 500 (2020).
regions identified seven astrocyte subclus- neurological disorders (12). Although astro- GRAPHIC: K. HOLOSKI/SCIENCE
10. A. V. Molofsky et al., Nature 509, 189 (2014).
ters that were present at different propor- cytes respond to disease states by undergo- 11. Y. Xie et al., Cell Rep. 38, 110416 (2022).
tions in each brain region. Several extracel- ing molecular and functional changes that 12. M. Pekny et al., Acta Neuropathol. 131, 323 (2016).
13. M. A. Anderson, Y. Ao, M. V. Sofroniew, Neurosci. Lett.
lular signaling molecules were identified as may have both beneficial and detrimental 565, 23 (2014).
upstream regulators of each cluster, sug- outcomes on neuronal health (13), increasing 14. A. J. Gleichman, S. T. Carmichael, Neurobiol. Dis. 142,
104957 (2020).
gesting that local cues may shape the iden- evidence suggests that astrocyte dysfunction
tify of these subclusters. Collectively, these could also be a cause of neuropathology (14). 10.1126/science.ade9249

or the marine ecosystem and to account for
P O LICY FO RU M environmental change. Currently, fishing is
restricted to CCAMLR members (comprising
CONSERVATION 25 nation states plus the European Union),
and in principle any state can go through the
Protect global values of the process to join CCAMLR.
SOCIAL AND ECOLOGICAL TRADE-OFFS
Southern Ocean ecosystem Today, Southern Ocean fisheries are con-

ducted by only ~12 countries, do not con-
tribute to food security, and threaten the
Climate change and fishing present dual threats Southern Ocean ecosystem. The primary tar-
geted species include toothfish (Dissostichus
eleginoides and D. mawsoni), the region’s
By Cassandra M. Brooks1, David G. Ainley2, rich waters (2). These waters harbor some of top fish predator, and Antarctic krill, with
Jennifer Jacquet3, Steven L. Chown4, the healthiest large marine ecosystems left in a minor contribution of mackerel icefish
Luis R. Pertierra5, Elizabeth Francis6, Alex the world (e.g., Ross Sea, Weddell Sea), which (Champsocephalus gunnari). Although the
Rogers7, Vasco Chavez-Molina1, Louise Teh8, provide remarkable wilderness, scientific, Food and Agriculture Organization of the
U. Rashid Sumaila8,9 and ecological value. United Nations has suggested that Antarctic
Recent assessments have highlighted the krill are underfished, with potential to con-
T
he Southern Ocean, which comprises immense global value of the Southern Ocean tribute to food security (7), the krill caught
~10% of the global ocean, is critically and how those values are immediately threat- are not used as a direct food source for peo-
important to the homeostasis of the ened by climate change, calling for a critical ple but as fishmeal for farmed salmon and
Earth system, exhibits distinctive ma- evaluation of trade-offs in management, in- shrimp, as well as in premium priced krill
rine biodiversity, and has tremendous cluding a better understanding of the risks oil supplements (8–10). Overall, these prod-
scientific, diplomatic, and wilderness posed by fishing (1–5). Antarctic waters affect ucts make up <1% of the global fishmeal and
value. Yet, the region and its suite of global the Earth’s climate, moderate sea level, and supplements markets (3). Toothfish are pro-
values are critically threatened by climate play a strong role in global ocean circulation cessed for human consumption, but sold as
change, which is exacerbated by commercial and nutrient cycling. The Southern Ocean Chilean sea bass in high-end restaurants and
fishing, an activity that provides value for disproportionately absorbs global carbon stores in the United States, Europe, and Asia,
relatively few industrial actors and compro- dioxide and heat, thus helping to regulate with limited ability to reduce the food insecu-
mises the greater values that the Southern temperature and buffering global impacts rity of malnourished people worldwide (10).
Ocean ecosystem provides to the world. The of climate change (5). The Southern Ocean There are other social consequences. Like
Commission for the Conservation of Antarctic biosphere also contributes to climate regu- other fisheries operating in international
Marine Living Resources (CCAMLR), the arm lation and oxygen production through its waters, some toothfish operations support
of the Antarctic Treaty System responsible primary production of seasonal phytoplank- forced labor on their vessels, and accidents
for managing Southern Ocean marine living ton blooms (3, 5). In addition, the Antarctic with loss of life at sea are not uncommon (11).
resources, meets in October–November 2022 seafloor stores extensive amounts of carbon Finally, many of these fisheries continue to be
and is under pressure to strengthen fisher- (3). Antarctic krill (Euphausia superba) economically viable only on account of gov-
ies management, especially toward climate play a critical role both as a central species ernment subsidies, which, as shown through-
change resilience. We encourage improved in the Southern Ocean food web (5) and in out the world, can contribute to overfishing
management practices that account for the biogeochemical cycles, stimulating primary (9, 11).
environmental externalities arising from production and influencing the drawdown Fishing in the Antarctic benefits very few
trade-offs between fishing and the global of atmospheric carbon to the deep sea (6). A and comes at a high cost of increasingly com-
contribution of the Southern Ocean ecosys- host of seasonally migrating iconic marine promising the ecosystem. The Antarctic krill
tem, including under a changing climate. mammals and birds depend on the Southern fishery recently has seen record high catches
With sovereignty suspended under the Ocean to supply their energetic needs (4). (>450,000 tonnes) (12). Yet, though these
Antarctic Treaty, Antarctica and the Southern With no Indigenous Peoples in Antarctica, catch levels fall within the precautionary
Ocean serve as an international space with and no local fishing communities, exploita- total allowable catch, extraction is increas-
immense global value. The Antarctic is cel- tion of the waters surrounding Antarctica ingly concentrated in nearshore predator
ebrated for its rich history of exploration, has always been the result of industrial dis- “hotspots” frequented also by fish, birds,
vast wilderness and extraordinary austere tant water activities. Historically, prior to the and mammals (e.g., Scotia Sea and the west
beauty, exceptional governance prioritizing existence of CCAMLR, these activities led coast of the northern Antarctic Peninsula;
diplomacy and peace, and long history of to overexploitation and near extirpation of CCAMLR Area 48) (7, 13). Reductions in sea
international science collaboration (1). The whales, seals, and some finfish populations ice (due to climate change) in the prime krill
Southern Ocean surrounding Antarctica sup- (2). CCAMLR came into force in 1982 to regu- fishing grounds have further enabled vessel
ports exceptional marine biodiversity, includ- late Southern Ocean fisheries with a mandate access to previously inaccessible waters and
ing assemblages of endemic organisms and to ensure that fishing does not cause irrevers- allowed for fishing year round, leading to in-
large populations of predators that rely on its ible harm to target and nontarget populations creased competition with predators and re-
1
Department of Environmental Studies, University of Colorado Boulder, Boulder, CO, USA. 2H.T. Harvey & Associates Ecological Consultants, Los Gatos, CA, USA. 3Department of Environmental
Studies, New York University, New York, NY, USA. 4Securing Antarctica’s Environmental Future, School of Biological Sciences, Monash University, Victoria, Australia. 5Department of Plant
and Soil Sciences, University of Pretoria, Pretoria, South Africa. 6The Gallifrey Foundation, Geneva, Switzerland. 7REV Ocean, Lysaker, Norway. 8Institute for the Oceans and Fisheries and the
School of Public Policy and Global Affairs, University of British Columbia, Vancouver, Canada. 9Institute for Environment and Development (LESTARI), National University of Malaysia, Selangor,
Malaysia. Email: [email protected]

INSI GHTS | P O L I C Y F O RU M
moval of krill from the food web (13). Recent

research indicates that the cumulative catch
in some of these coastal regions can be
greater than the amount consumed by local
predators, and greater than the local replen-
ishable population of krill (7).
This localized depletion has substantial
potential consequences for predators, includ-
ing visiting whale populations, which are
still recovering from historic depletion (7,
13). As a result of direct competition at these
hotspots, in 2021, the krill fishery for the first
time incidentally killed three juvenile hump-
back whales (12). This was in addition to the
incidental killing of at least 16 seals and 60
petrels and other seabirds over the last two
seasons (12). CCAMLR’s mandate stipulates
maintenance of ecological relationships, such
that fishing should consider impacts on pred-
ators, including whales. Yet there are insuf-
ficient measures in place to ensure that the
fishery is not detrimentally competing with
predators or to prohibit (or even limit) inci-
dental catch of whales in the krill fishery.
Ecosystem and target species’ impacts
from fishing are further compounded by cli-
mate change. Parts of the Southern Ocean,
especially those adjacent to the western
Antarctic Peninsula (where krill fishing is
concentrated), are among the most rapidly
changing regions in the world, with poten-
tial cascading impacts on the ecosystem (2, Further, given that historic whale population). Fishing also stands to create diplomatic
3). Krill, in particular, owing to dependence tions would have enhanced nutrient cycling, conflict (e.g., current conflict over toothfish
on sea ice and vulnerability to acidification, promoted productivity, and facilitated the fisheries at South Georgia) and erode the wil-
are seen as extremely high risk from climate drawdown of carbon and nutrients to the derness value of the Antarctic, without con-
change (3). Benthopelagic-dwelling Antarctic seafloor, allowing whales to recover (by lim- tributing to food security. Although actions
toothfish, because of their limited thermal iting competition with fisheries) could lead must continue to be taken through national
tolerance, are also at risk (2, 3). Antarctic to higher Southern Ocean productivity, in- governments toward reducing global carbon
finfish have shown to be easily overexploited creasing its powers of carbon sequestration emissions, CCAMLR has the responsibility to
even without the stress of climate change. (15). In addition, the incredible remoteness embrace the ecosystem mandates in its con-
For example, marbled rockcod (Notothenia of the Antarctic means that fuel use by fish- vention and the competency to implement a
rossii), which were severely overfished in the ing vessels is high, leaving a disproportion- variety of tools to help immediately.
1970s, remain at a fraction of their estimated ate carbon footprint from fishing (8).
pre-exploitation levels, despite ongoing re- Manage for ecological and climate resilience
gional fishery closures (14); bycatch of larval PROTECTING THE GLOBAL VALUES For three decades, CCAMLR has discussed
fish in the krill fishery may be impeding their This is a critical moment for the Southern more sophisticated spatial management for
recovery (8, 14). The population status of Ocean, and for CCAMLR, which showed Antarctic krill (e.g., at smaller scales), yet
many toothfish fisheries remains unknown, tremendous leadership in 2016 by adopting has not been able to adopt this into practice.
including in some regions of the Southern the Ross Sea region marine protected area Thus, CCAMLR has not yet moved beyond
Ocean that were heavily targeted by illegal, (MPA), among the largest MPAs in the world, single-species stock assessments with a large-
unregulated, and unreported fisheries in the but which has since waned on adopting fur- scale static total allowable catch. Meanwhile,
1990s (12). Meanwhile, ecosystem impacts ther conservation initiatives beyond preexist- recent modeling has indicated that combined
of fishing toothfish remain largely unknown ing fisheries measures. This comes at a time effects of krill fishing and climate change
and debated. Fishing at current levels will when the global community, through the are potentially disastrous for populations of
likely exacerbate environmental impacts on United Nations, is close to negotiating a new penguins and other predators (13). Historical
toothfish and krill, along with the greater treaty for conserving biodiversity on the high trends suggest that the quantity of krill fished
Southern Ocean ecosystem, including com- seas. CCAMLR thus stands to either serve as will further increase and grow more spatially
peting birds and mammals (3, 13). a leading model or be surpassed. Continuing concentrated in predator hotspots.
Krill fishing, in particular, because of to authorize the removal of key prey and A suite of science-based tools exist for
PHOTO: JOHN B. WELLER
krill’s critical role in the ecosystem and predator species through fishing, by a rela- managing marine systems for ecological
biogeochemical cycles, is predicted to negatively small industrial fleet, will further com- and climate resilience. These include us-
tively affect Southern Ocean ecosystem ser- promise ecosystem and biosphere regulating ing environmental and ecosystem indices to
vices (1–3), including reducing the carbon services (as has been shown elsewhere, e.g., understand impacts of fishing and environ-
storage potentially provided by krill (6). Peru, and Benguela current forage fish extrac- mental change; considering future climate

An emperor penguin stands in front of an icebreaker Moratorium on fishing
on the frozen Ross Sea, Antarctica. Upon coming into force in 1982, CCAMLR
soon thereafter closed various regions to fin-
contain language that allows the closing of fish fishing to allow populations to recover,
areas to manage fisheries, and at times it has which is a powerful tool incorporated within
exercised that power. Moreover, CCAMLR its convention. The measure is showing ef-
committed to an ecologically representa- fectiveness, though slowly, in the Antarctic
tive network of Southern Ocean MPAs. After Peninsula region (14). Despite working to-
adopting two MPAs (in 2009 and 2016), it ward a more ecologically and spatially ex-
has since lacked the renewed political will plicit management framework, CCAMLR has
to move forward, with Russia and China not used this tool under the precautionary
not yet joining in consensus for the existing principle despite the growing evidence of
Southern Ocean MPA proposals under nego- the external costs from the fishing indus-
tiation. These MPA proposals are grounded try. Given the clear acceleration in climate
in extensive science, but likely need high- change, far more rapid now than when
level diplomacy, one of the key facilitators of CCAMLR was negotiated, we argue that
the adoption of the Ross Sea region MPA. We this conservation tool can no longer be ne-
encourage CCAMLR states to work together glected. We recommend considering a mora-
to find common ground on the MPAs, which torium (a temporary prohibition) on fishing
can serve as a tool for both conservation and while a more comprehensive management
international diplomacy, emblematic of the assessment is undertaken. This assessment
spirit of the Antarctic Treaty System. should account for all gains and losses, and
consider broader climate resilience with the
Consider the full suite of values inclusion of regional and global trade-offs
for future generations in ecosystem services. We acknowledge that
As an international space, the Southern among some CCAMLR member states, pro-
Ocean must be valued beyond its direct hibitions on fishing may seem unreasonable,
economic (extractive) value, which ben- although such actions are not without his-
efits a small number of nations (including torical precedent. Further, given the proven
the wealthiest on Earth) in the short term. vulnerability of Southern Ocean organisms
These activities are likely also harming the and ecosystems to climate change and ways
impacts on targeted species and the ecosys- wider values and functions of the Southern in which fishing is known to compromise
tem; and employing adaptive management. Ocean. A comprehensive ecosystem services ecosystem services, a real reassessment of
Consideration of recovering whales and the assessment and mapping of the Southern current practices may make an indefinite
provisioning needs of krill and toothfish pred- Ocean urgently needs to be completed, in- moratorium appropriate until the climate
ators must be accounted for in management, cluding analyses of social perceptions from crisis is better managed. j
including in more refined spatial manage- stakeholders and all nonmonetary values
ment and precautionary catch limits. All of (1). This should include its support of sea-
1. L. R. Pertierra et al., Ecosyst. Serv. 49, 101299 (2021).
these management ideas are being discussed sonally present marine mammals and birds, 2. A. D. Rogers et al., Annu. Rev. Mar. Sci. 12, 87 (2020).
among CCAMLR’s Scientific Committee and its role in the Earth system, along with 3. R. D. Cavanagh et al., Front. Mar. Sci. 7, 615214 (2021).
(which advises CCAMLR). However, data its value as a global wilderness, and its 4. E. J. Murphy et al., Front. Ecol. Evol. 9, 624451 (2021).
5. S. M. Grant, S. L. Hill, P. N. Trathan, E. J. Murphy, Antarct.
gaps and uncertainty, along with resistance history of peace and science. Further, the Sci. 25, 603 (2013).
to more stringent catch limits (spatially, and Southern Ocean needs to be valued for the 6. E. L. Cavan et al., Nat. Commun. 10, 4742 (2019).
overall), have prevented progress. In 2022, suite of services it provides to both current 7. P. N. Trathan et al., J. Mar. Syst. 225, 103598 (2022).
8. S. Nicol, J. Foster, in Biology and Ecology of Antarctic
the precautionary interim distribution of the and future generations. Improved manage- Krill, V. Siegel, Ed. (Springer, 2016), pp. 387–421.</edb>
“trigger” catch limit for krill expires. This ment of the Antarctic could boost global 9. R. Cappell, G. MacFadyen, Mar. Policy 143, 105200
presents a tremendous risk for fishing to be climate change mitigation and ensure (2022).
10. L. Schiller, M. Bailey, J. Jacquet, E. Sala, Sci. Adv. 4,
even more concentrated. Given the recent that Antarctica, as an exceptional global eaat8351 (2018).
research demonstrating the risk to krill and treasure, is not exploited for the benefit 11. E. Sala et al., Sci. Adv. 4, eaat2504 (2018).
their predators if the fishery continues under of short-term gain for few at the expense 12. CCAMLR, “Fishery Reports” (2022);
https://fisheryreports.ccamlr.org/.
its current rules, CCAMLR has an immense of many. Moreover, proximal Indigenous
13. G. M. Watters, J. T. Hinke, C. S. Reiss, Sci. Rep. 10, 2314
opportunity and responsibility to adopt the communities have ties to the Antarctic, (2020).
spatially explicit precautionary management though these communities have not yet 14. E. R. Marschoff, E. R. Barrera-Oro, N. S. Alescio, D. G.
that it has been developing for decades. had a voice at Antarctic governance meet- Ainley, Fish. Res. 125-126, 206 (2012).
15. A. J. Pershing et al., PLOS ONE 5, e12444 (2010).
ings. Worldviews that prioritize conserva-
Implement marine protection tion for future generations are consistent ACK NOWLE DGM E NTS
Within CCAMLR’s capabilities to exclude with the foundations of the Antarctic Treaty C.M.B., V.C.-M., U.R.S., and L.T. received support through
areas from fishing, MPAs, if well designed, System, including the CAMLR Convention, the Pew Charitable Trusts. L.R.P. received support through
Biodiversa ASICS. U.R.S. and L.T. also thank the University of
can potentially promote adaptation to cli- and need to be considered and incorporated British Columbia–based Solving Food-Climate-Biodiversity
mate change and enhance Southern Ocean into management. CCAMLR could evalu- research partnership sponsored by the Social Sciences and
ecosystem resilience. MPAs can provide cli- ate values, management, and trade-offs in Humanities Research Council of Canada.
mate refugia and facilitate adaptation by tandem with the wider Antarctic Treaty
alleviating the stressors on the ecosystem System, providing unified management for Published online 20 October 2022
(e.g., fishing). CCAMLR’s founding articles Southern Ocean and Antarctic values. 10.1126/science.add9480

A NASA-supported mobile health unit is prepared for
deployment to the Papago Indian Reservation in 1974.
of physician-patient relationships. By the

beginning of the 21st century, the telephone
had become a routine device used by med-
ical practitioners—a quarter of interactions
between physicians and patients took place
over the phone—yet it had done little to
resolve unequal health outcomes. The era-
sure of the debates surrounding the intro-
duction of previous forms of telehealth is a
kind of power that allows technologists to
ignore the pitfalls of past innovations, notes
Greene. “Forgetting,” he writes, “is an active
as well as a passive process.”
While popular histories of medicine
B O O KS e t al . tend to be told through a progressive lens,
Greene’s study reads more like an anti-
progress manifesto—or at least like a story
MEDICAL TECHNOLOGY of the roads not taken. Sifting through his
case studies, one is reminded not only of
Where telemedicine always the complex and at times circuitous ways

in which new telehealth technologies are
adopted but also how they are almost always
falls short accompanied by fears that the clinician will

be displaced from the diagnostic and thera-
peutic process.
Efforts to innovate our way to better health Greene’s study reminds readers that
consistently fail to deliver on equity promises the introduction of technologies aiming to
provide health care at a distance has long
been bound up with recurring promises to
By Vanessa Rampton can the provision of health care at a dis- improve health equity. And yet we know
tance replace in-person treatment? Jeremy that the preservation of physical distance
I
n 2015, the Australian startup CliniCloud Greene, an emergency care physician in can have a stabilizing function, in that it
promised to leverage technology to dis- Baltimore and a historian of medicine at prioritizes the stability of current systems
rupt medicine, making health care more Johns Hopkins University, is well placed to over social justice (1). It can thus also serve
accessible and affordable. The company, answer these questions. His new to reproduce social hierarchies.
which had created a digital thermom- book, The Doctor Who Wasn’t During the COVID-19 pandemic,
PHOTO: PETER RUIZ COLLECTION, ARCHIVES, HIMDAG KI, TOHONO O’ODHAM NATION CULTURAL CENTER AND MUSEUM
eter and stethoscope that paired with There, tells a fascinating story for example, telehealth technolo-
a smartphone app, sought to do this by al- of the hype, achievements, and gies created new barriers to care,
lowing patients to measure their vital signs inevitable shortcomings of tele- exacerbating the differences be-
and communicate with doctors from the health technology. tween those who could reliably
comfort of their own homes. Launched to The technologies Greene high- connect to new electronic inter-
much enthusiasm—its utility was endorsed lights range widely. They include faces and those who could not.
by The Lancet and the US Food and Drug the original landline telephone, Given that many of the most
Administration—the technology’s success used by physicians to communi- The Doctor Who pressing global health problems
Wasn’t There
was short-lived. By 2018, the company’s on- cate with patients and with one Jeremy A. Greene
remain unaddressed because of
line presence had disappeared, and the prod- another in the late 19th century, University of Chicago issues associated with resource
uct had joined a long list of promising medi- radio technologies that broad- Press, 2022. 336 pp. allocation rather than a lack
cal innovations that were soon forgotten. cast health data in the 1940s, of available treatments, why
CliniCloud’s rapid rise and fall raises sev- closed-circuit televisions that enabled early then do we continue to focus on achieving
eral questions: How do medical technologies telemedicine in the 1950s, and the systems breakthroughs rather than on using exist-
age, become outmoded, and retreat from that led to the creation of electronic medical ing knowledge and technical powers to put
our collective consciousness? Have technol- records in the 1960s. everyone on a more equal health footing?
ogies aimed at improving health equity and The humble telephone is a perfect case Why, in other words, should we believe that
democratizing health care access delivered study. Almost immediately after its inven- new technologies will succeed where old
on their creators’ promises? To what extent tion in the late 19th century, its “revolu- ones have failed? As Greene shows, science
tionary” medical potential was noted— and technology alone will never be able to
The reviewer is at the Department of Humanities, patients were going to be able to be heard deliver a more just and equitable society. j
Social and Political Sciences, ETH Zürich, 8092 Zürich, in their own homes, at their convenience,
Switzerland, and the Institute for Health and Social Policy RE FE RE NCES AND NOT ES
and the Department of Philosophy, McGill University,
and without risk of mutual infection. But it
1. M. Kolkenbrock, Kunst und Kirche 4, 4 (2020).
Montréal, Quebec H3A 1G1, Canada. also elicited considerable concerns, for ex-
Email: [email protected] ample, about the potential mechanization 10.1126/science.ade1246

I NSI GHTS
EDUCATION Masters of Health:

Racial Science and Slavery
Reifying racism in medicine in U.S. Medical Schools

Christopher D. E. Willoughby
University of North Carolina
Press, 2022. 282 pp.
False beliefs about racial differences have a long history
in physician training programs
By Suman Seth and French empires, which covered a siz- Willoughby concludes, “made it clear that
able portion of the globe by 1860. anatomical racial pedagogy was rigorous,
I
n 2015, a study published in JAMA The book’s core, part two, contains the standardized, and deeply influential.”
Pediatrics found that in close to a mil- strongest and most striking material. Here, We have long known that southern medi-
lion diagnosed cases of appendicitis, Willoughby tracks in detail the ways that cal schools repeatedly made use, without
Black children were found “less likely to early US medical students learned the racial consent, of the bodies of the enslaved for
receive any pain medication for moderate biology that would inform their practices. anatomical study. As one abolitionist news-
pain and less likely to receive opioids for First, they encountered claims about race paper put it in 1861: “What a convenience to
severe pain” (1). A survey conducted in 2016 in their textbooks. University of Pennsylvania possess for scientific purposes a class of pop-
by scholars at the University of Virginia of- professor Joseph Leidy’s An Elementary ulation sufficiently human to be dissected,
fered some clues for why this might be so, Treatise on Human Anatomy (1861), for ex- but not human enough to be supposed to
revealing that striking proportions of re- ample, introduced students to a section on take offence to it!”
spondents—including those with medical the brain by suggesting that “it is the larg- Willoughby makes clear that northern
training—held false beliefs about differences est in the white race, and…its bulk bears a schools may have been more circumspect,
between races (2). Fifty-eight percent of re- general relationship with the development of but physicians in New York and Philadelphia
spondents without medical training, 42% of intellect.” Such claims were further empha- “preyed” on their Black population as well.
medical students in their second year, and sized and elaborated on in lectures, which The goal was to source cadavers with mini-
25% of residents all believed, mal resistance or backlash
for example, that “Blacks’ skin is from local populations, and
thicker than whites’.” the wishes of free Black com-
One conclusion we might munities were the most easily
draw from these data is that ignored.
medical schools in the US are Where the North and South
not doing enough to eradicate differed more dramatically was
widely held, inaccurate ideas over the issue of medical experi-
concerning putative biologi- mentation, where the enslaved
cal differences between races. could not effectively resist. In
In his new book, Masters of research conducted for a the-
Health, historian Christopher sis submitted to the Medical
Willoughby suggests that this College of South Carolina in
failure might be rooted in medi- 1846, for example, C. Caldwell
cal education’s long tradition of Higgins described experiments
perpetuating and reifying pre- with no therapeutic purpose
IMAGE: WELLCOME COLLECTION, ATTRIBUTION-NONCOMMERCIAL 4.0 INTERNATIONAL (CC BY-NC 4.0)
cisely such racial ideologies. conducted on the bodies of a

The book is divided into lactating enslaved woman and
three parts. The first provides her child. The woman’s body
necessary background material Racial biases were often reinforced in 19th-century anatomy lectures. was covered with moist tobacco
on both the history of medi- for almost half an hour, until
cine and the history of race before the early also allowed more visual and even tactile she reported feeling sick. Higgins then had
19th century. Even if each of these narra- experiences, so that students could, for ex- the women suckle her infant: “In a short
tives is largely well known, it is nonetheless ample, look at, and even handle, skulls that time afterwards,” Higgins wrote, “the child
useful to have them told in tandem, so that illustrated the arguments being made. began to vomit. In fact for a while I was
the original and ongoing connections be- Anatomical museums rendered peda- afraid it was going into convulsions but luck-
tween medical institutions and the develop- gogical lessons even more authoritative. At ily for me it did not.”
ment of racial ideas come to the forefront. Harvard’s museum, skulls were ordered by Little about Masters of Health makes for
The book’s third part is dedicated to locat- race: white skulls on the top shelf, South and easy reading, but it is a text that those who
ing the US within a transnational history of East Asian skulls toward the middle, and have finished medical school—and those
racial medicine. In this section, Willoughby African skulls at the bottom. aspiring to enter it—should read with care
reveals that there was much borrowing of Willoughby also tracks how students and attention. j
thoughts and things between US medical wrote about race in their theses. Such texts
men and their counterparts in the British were rarely terribly novel, usually summa-
1. M. K. Goyal, N. Kuppermann, S. D. Cleary, S. J. Teach,
rizing extant literature rather than provid- J. M. Chamberlain, JAMA Pediatr. 169, 996 (2015).
ing original research. As such, they serve 2. K. M. Hoffman, S. Trawalter, J. R. Axt, M. N. Oliver,
The reviewer is at the Department of Science and Proc. Natl. Acad. Sci. U.S.A. 113, 4296 (2016).
Technology Studies, Cornell University, Ithaca, NY 14853, as excellent guides for the views of leading
USA. Email: [email protected] men of 19th-century medicine. These works, 10.1126/science.ade8481

Onlookers watch
as pumps drain
floodwaters in
Dadu, Pakistan on
19 October 2022.
LET TERS
Edited by Jennifer Sills Pakistan to increase by about 0.5°C since 2. United Nations Office for the Coordination of
1970 (8). Over the past three decades, the Humanitarian Affairs (OCHA), “Pakistan: 2022 monsoon
floods situation report no. 03 (as of 26 August 2022)”
Pakistan’s floods flow frequency of heat waves has increased five-
fold, resulting in prolonged and extreme
(2022).
3. Pakistan Meteorological Department, “Pakistan
from climate injustice summers and accelerating evaporation and
transpiration processes (8). Each 1°C rise
monsoon 2022 rainfall update” (2022); www.pmd.
gov.pk/cdpc/Monsoon_2022_update/Pakistan_
Monsoon_2022_Rainfall_Update.htm.
Recent flooding in Pakistan left one- in temperature allows the atmosphere to 4. OCHA, “Pakistan: 2022 monsoon floods situation report
third of the country under water (1). The hold 7% more water (9), making monsoon no. 9 (as of 14 October 2022)” (2022).
5. OCHA, “Revised Pakistan 2022 floods response plan”
floods, which affected at least 33 million seasons more intense. If global emissions (2022).
people in 72% of the country’s districts, continue to rise, the average annual tem- 6. Pakistan Bureau of Statistics, “Annual analytical report
were caused by an extreme monsoon perature in Pakistan could increase by 3° on external trade statistics of Pakistan (FY 2020-
21)” (Ministry of Planning, Development, & Special
season (2) that dropped up to five times to 6°C (compared with the projected global
Initiatives, External Trade Section, Karachi, 2022).
as much rain as the average for the past average of 1.5° to 5°C) by the end of this 7. Government of Pakistan, “Pakistan: Updated nation-
30 years (3). The floods led to more than century (8, 10), likely leading to worse dev- ally determined contributions 2021” UN Framework
1000 human casualties and the deaths of astation in the future. Convention on Climate Change, New York (2021).
8. Q. U. Z. Chaudhry, “Climate change profile of Pakistan”
1.2 million livestock (4). Housing, educa- Politicians, scientists, and the United (Asian Development Bank, 2017).
tion, and health infrastructures, as well Nations have spoken out against climate 9. D. Coumou, S. Rahmstorf, Nat. Clim. Change 2, 491
as between 2.6 and 3.8 million hectares injustice in Pakistan (11), which relies (2012).
10. J. Tollefson, Nature 580, 443 (2020).
of cropland, were affected (5), threaten- on foreign grants for climate mitiga- 11. UN Press Release SG/SM/21519 (2022); https://press.
ing food security in Pakistan as well as in tion and adaptation. A global framework un.org/en/2022/sgsm21519.doc.htm.
the countries to which Pakistan exported must be established to prevent devel-
10.1126/science.ade8490
US$4.4 billion worth of food in 2021 (6). oping countries such as Pakistan from
Communicable diseases are increasing becoming ground zero of climate change.
in communities of displaced people (5). Responsible nations must work to provide
Pakistan’s climate-driven humanitarian
crisis is the result of the emissions pro-
technical and economic assistance for cli-
mate change mitigation.
Restore Pakistan’s
duced by larger, more developed coun-
tries. The international community must
Muhammad Ahmed Waqas
Department of Agroecology, Aarhus University,
forests to prevent floods PHOTO: STRINGER/GETTY IMAGES
address climate injustice. DK-8830 Tjele, Denmark. Forests comprise less than 5% of
Email: [email protected]
Pakistan accounts for only 0.9% of global Pakistan’s total area (1), and every year
greenhouse gas emissions but is among the more forest area is destroyed by anthro-
top 10 countries most affected by climate pogenic and natural activities. Pakistan’s
1. European Space Agency, “Pakistan inundated” (2022);
change (7). Ongoing global warming has www.esa.int/ESA_Multimedia/Images/2022/09/ dwindling forest cover has exacerbated
caused the average annual temperature in Pakistan_inundated. the effects of climate change (2). In June,

I NSI GHTS
the country suffered devastating floods deforestation, stop illegal infrastructure Glacial lake outbursts combined with
that threatened valuable ecosystem development adjacent to floodplains heavy rainfall during the monsoon
resources, including agricultural products, and waterways, and control the rate of period are the main causes of flooding
as well as the socioeconomic conditions of human population growth. By investing in in the Indus River basin in Pakistan (3).
millions of people (3). Given the ability of ecological restoration and water manage- A cross-border international river, the
vegetation, native biodiversity, and con- ment, Pakistan and other at-risk countries Indus originates upstream of the Shigatse
nected waterways and floodplains to serve can mitigate the damage caused by future River on the Tibetan Plateau in China.
as buffers for storms, rains, and floods natural disasters. Flood prediction is conducted extensively
(4, 5) and mitigate climate change (6), Arshad Ali in Sindh and Punjab (the middle and
Pakistan and other Asian countries must Forest Ecology Research Group, College of Life lower reaches of Pakistan) (4, 5), but the
Sciences, Hebei University, Baoding, Hebei
increase their environmental budgets to extent to which the melting glaciers in
071002, China. Email: [email protected]
prioritize ecosystem restoration. the upper reaches will affect the middle
Preventing flooding in Pakistan is an RE FE RE NCES AND NOT ES and lower reaches of the river remains
ongoing challenge. In 2014, in response to 1. Food and Agriculture Organization of the United unclear. Global climate models, which
Nations, “Global forest resources assessment” (2020).
widespread flooding and climate change predict a likely long-term warming of 1.58
2. Food and Agriculture Organization of the United
(7), the country launched massive planta- Nations, “2019 Forestry sector review: Pakistan” or 28C (6), also fail to predict extreme
tion projects to restore degraded forest (2020). weather events.
lands and increase green job opportuni- 3. S. Mallapaty, Nature, 10.1038/d41586-022-02813-6 Pakistan, and all countries with rivers
(2022).
ties (8). Although the efforts made some 4. United Nations Environment Assembly, “Resolution affected by glacial meltwater (7), should
short-term gains, they were undermined 73/284: United Nations Decade on Ecosystem make predicting regional hazards and
by substandard plantation designs and Restoration (2021–2030)” (2019). extreme weather a priority. For basins
5. C. J. Zabin et al., Front. Ecol. Environ. 20, 310 (2022).
scientific planning, inadequate monitor- 6. B. B. N. Strassburg et al., Nature 586, 724 (2020). such as the Indus, joint observatories
ing, and weak community mobilization 7. K. Larkin, Nature, 10.1038/news.2010.409 (2010). and systematic simulations should be
activities (8, 9). Meanwhile, natural forest 8. A. Ullah et al., Sci. Tot. Environ. 772, 145613 (2021). established between countries to enable
9. S. Khan, “Pakistan: Environmentalists slam ‘10 bil-
degradation continued unabated (1, 2, 8). lion trees’ project,” DW Report (2021); https://p.
the forecasting of floods related to glacier
Pakistan and other countries in the dw.com/p/3zom8. melt driven by climate change. Rather
region should promote regional and 10. B. B. N. Strassburg, Science 374, 125 (2021). than focusing on national boundar-
11. C. Albert, J. H. Spangenberg, B. Schröter, Nature 543,
global cooperation between experts in 315 (2017).
ies, climate studies should be directed
ecological science, policy, and practice (10, 12. S. L. Stoffella et al., Ecosphere 13, e4223 (2022). toward naturally delineated basins. Aid
11) to improve on past efforts to conserve provided by international organizations
and restore natural ecosystems (particu- to developing countries should include
larly natural forests). Plantation locations not only relief supplies, but also open
should be selected based on the abiotic data sources, technology,
(soil, light, and water) and biotic (species
facilitation) interactions underpinning
Predicting Pakistan’s and researchers.
Climate change will continue to
targeted species. Practitioners should
promote native mixed-species plantations
next flood affect vulnerable countries for years to
come. With collaboration and invest-
with flood-tolerant species and hetero- In June, the largest flooding disaster of ment in early warning systems and
geneous stand structures, which would this century hit Pakistan, flooding farm- climate-resilient infrastructure, develop-
increase adaptability and resilience to land and inducing a severe food crisis (1). ing countries like Pakistan will be bet-
future extreme climatic events (5, 12). Over 30 million people were affected, and ter prepared to face cascading effects,
Restorations plans should also anticipate more than 1000 people were killed (2). including natural disasters.
the vulnerability of new ecosystem res- Developing countries such as Pakistan Yingying Yao1* and Zoon Ahmed Khan2
toration projects to natural disasters (5) remain at high risk of disasters arising 1Institute of Global Environmental Change, Xi’an Jiaotong
University, Xi’an, China. 2Belt and Road Strategy Institute,
and ensure that they are adequately pro- from climate change. To mitigate the dam-
Tsinghua University, Beijing, China.
tected. In addition to these improvements, age, countries facing such threats should *Corresponding author. Email: [email protected]
developing countries in Asia should cooperate to improve disaster prediction
continue longstanding efforts to reduce capability and infrastructure. RE FE RE NCES AND NOT ES
1. Q. K. Afridi, “Pakistan’s disastrous floods uproot
refugees and citizens,” UN Refugee Agency (2022);
www.unhcr.org/news/stories/2022/9/6311c7f54/
NEXTGEN VOICES: SUBMIT NOW pakistans-disastrous-floods-uproot-refugees-citizens.
html.
Favorite failures 2. P. Fihlani, “Pakistan floods: Time running out for fami-
lies in Sindh,” BBC News (2022); www.bbc.com/news/
world-asia-62757605.
Add your voice to Science! Our new NextGen Voices survey is now open:
3. S. Mallapaty, Nature, 10.1038/d41586-022-02813-6
Have you ever looked back and felt grateful that you experienced a setback? Describe a (2022).
failure or mistake in your science career and explain how it led to a revelation, opportunity, 4. M. Aslam, Mehran Univ. Res. J. Eng. Technol. 37, 297
or decision that would not have happened otherwise. (2018).
5. I. Khan, H. Lei, A. A. Shah, I. Khan, I. Muhammad,
Environ. Sci. Pollut. Res. 28, 29720 (2021).
To submit, go to www.science.org/nextgen-voices
6. P. M. Forster, A. C. Maycock, C. M. McKenna, C. J. Smith,
Deadline for submissions is 18 November. A selection of the best responses will be pub- Nat. Clim. Change 10, 7 (2020).
lished in the 6 January 2023 issue of Science. Submissions should be 100 words or less. 7. T. Yao et al., Nat. Rev. Earth Environ. 3, 618 (2022).
Anonymous submissions will not be considered.

INSI GHTS
PRIZE ES SAY
GRAND PRIZE WINNER
Ann Kennedy
Ann Kennedy
received
undergradu-
ate degrees
in Biomedical
Engineering and
Biology from
Johns Hopkins University and
a PhD in Neurobiology and Be-
havior from Columbia University.
After completing her postdoc-
toral fellowship at Caltech, Ann
started her laboratory in the
Department of Neuroscience
at Northwestern University
Feinberg School of Medicine in
2020. Her laboratory develops
computational methods to
characterize the structure of
complex behavior and model its
control by the brain. www.science.
org/doi/10.1126/science.ade2128
NEUROSCIENCE
Boiling over
An emergent encoding of aggressive motivation
in neurons of the hypothalamus
By Ann Kennedy the opponent, with an estimate of the oppo-
nent’s strength weighed against one’s fitness
A
ggression between members of a spe- to determine the next move. But this behav-
cies takes many forms, but the rule- ior is seen in the animal kingdom from flies
book that guides it is ancient. Fighting to fiddler crabs to ferrets. Its importance for
may help resources and mates go to survival of a species means that aggressive
the fittest members of a species, as strategies are under strong selective pressure.
PHOTOS: (LEFT TO RIGHT) COURTESY OF ANN KENNEDY; PETE OXFORD/MINDEN PICTURES

Darwin posed (1). This aggression can But what is that pressure acting on?
also serve as a repelling force to distribute In 1928, Hess found that electrostimula-
animals evenly over an environment (2). In tion of a deep brain area could elicit displays
social species, aggression creates hierarchies of rage (5). Further stimulation work, as well
that shape the health and behavior of group as the rare tumor patient with aberrant ag-
members (3). But picking a fight is costly be- gressive behavior, pointed to the existence
cause even the winner may walk away with of a “hypothalamic attack area” (6, 7). With
injuries. If time allows, it is safer to begin the advent of optogenetics, this area found
with threat displays and posturing and only an identity: the ventrolateral portion of the
escalate to outright attack if neither side will ventromedial hypothalamus (VMHvl) (8).
back down (4). Artificial activation of VMHvl drives mice
This cautious, calculated escalation sounds to attack and chase whatever other mouse
cerebral: an integration of tactical observa- is present; conversely, silencing VMHvl
tions that gradually reduce uncertainty about interrupts naturally occurring attacking
but does not affect other behaviors, such
as mating (9). Neighboring hypothalamic
Laboratory for Theoretical Neuroscience and Behavior, nuclei play similar dramatic roles in other
Department of Neuroscience, Northwestern University
Feinberg School of Medicine, Chicago, IL, USA. behaviors—eliciting feeding, drinking, mat-
Email: [email protected] ing, or defensive reactions when stimulated.

A red-legged seriema threat display. Some FINALIST to baseline, but in eager fighters, activity
aggressive encounters end after opponents Kevin Guttenplan quickly grew and remained high.
face off through threat displays, whereas Kevin Guttenplan We could not identify a clear attack “deci-
others further escalate to outright fighting. received his under- sion”; during pauses between attacks, neural
graduate degree in activity remained high. The signal looked
The combined effect of these studies paints Neuroscience and like a level of aggressive motivation rather
the hypothalamic nuclei as a great internal Mathematics from than a trigger for a specific motor act. This
switchboard, administered by some homun- Pomona College. He signal echoed our other data from the dorso-
cular operator to conduct the fundamental then went to Stan- medial VMH (VMHdm), a region implicated
motor acts of survival. ford University for his PhD, where he in predatory defense (13). Predator cues or
To see how these switches were flipped worked first in the laboratory of B. auditory threats activate VMHdm neurons
and try to understand their interaction, my Barres and then, after Barres’s pass- for tens of seconds, during which animals
collaborators and I used head-mounted mining, in the laboratory of A. Gitler and exhibit freezing and defensive behaviors. We
iaturized microendoscopes to capture the studied the role of astrocytes in dis- found that this VHMdm activity was nec-
activity of several dozen to a few hundred ease and injuries of the nervous sys- essary to keep animals in a defensive mo-
hypothalamic neurons, while allowing mice tem. He is now a Helen Hay Whitney tivational state. When we showed a mouse
to freely interact with another male or fe- postdoctoral fellow in the laboratory a predator for a few seconds, removed the
of M. Freeman in the Vollum Institute
male (10). We studied several nuclei, includ- predator, then silenced VMHdm, the animal
at Oregon Health & Science Univer-
ing VMHvl, to look for neural signatures of quickly reverted to its baseline behavior—
sity, studying the role of astrocytes
an animal’s decision to attack, and one of rather than persisting to freeze and hide as
in neuronal circuits. www.science.org/
VMHvl’s tightly coupled partners, the medial doi/10.1126/science.ade2129
mice would normally do.
preoptic area (MPOA). Our studies suggest an alternative view
Neurons in MPOA were clearly and FINALIST
of hypothalamic nuclei: not as switches for
strongly modulated by animals’ moment-to- specific behaviors, as optogenetic studies sug-
moment actions, particularly reproductive Filipa Cardoso gested, but as distinct computational units
behaviors. As the imaged male investigated, Filipa Cardoso (12). MPOA does indeed seem to reflect many
fought, or mated with other mice, distinct received her moment-to-moment actions of the animal,
behavior-specific ensembles of neurons bachelor’s degree and our dynamical systems analysis failed to
fired—their activity linked with the animal’s in Biochemistry and find any persistent modes of its activity. But
master’s degree
actions. By contrast, VMHvl activity during VMH is more akin to a volume dial than a
in Health Sciences
social interactions was a morass—although switch. VMH reminded us of Konrad Lorenz’s
from Universidade
many neurons showed a strong response to description of behavioral drives as a leaky
do Minho where she was introduced
the introduction of another mouse, activ- to the field of immunology, working in
bucket that slowly fills with water, pushing
ity during interactions showed only weak the immune response to tuberculosis the animal to greater heights of aggression as
correlation with the animal’s behavior (11). infection and to colitis models. After the pressure builds (14, 15).
Instead, neurons were persistently active her master’s, she moved for a PhD in Our observations that some animals
over the duration of a social encounter, with H. Veiga-Fernandes’s laboratory to de- showed more VMH persistence, and more
modest fluctuations in the intensity of their velop research in how the nervous and aggression, suggests that this “leak” rate is
activity as animals interacted. the immune systems interact to con- tunable—perhaps modifying behavior in re-
We wondered whether we knew what we trol metabolism. After completing the sponse to memories of past defeat or what
were looking for. We had defined “attack” PhD, Filipa joined a biopharmaceutical is at stake in the current fight. We know
through manual, frame-by-frame annota- company, LiMM Therapeutics, which many ways to tune persistence in models of
tion—a subjective measure that may have lit- stemmed from the Veiga-Fernandes biological neural populations. Now we have
tle to do with when the mouse itself decided laboratory. Currently, she is striving only to ask what molecular machinery links
to start fighting. We thus opted to discard our to translate the knowledge obtained that tuning to experience in a living, and
annotations and looked at the neural activity from basic research to clinical ap- evolving, brain. j
by itself; we fit a dynamical system model to plication by harnessing the molecular
obtain a low-dimensional summary of how cross-talk between neuronal and REFER E NCES & NOTES
PHOTOS: (TOP TO BOTTOM) AARON BIELECK; COURTESY OF FILIPA CARDOSO
innate lymphoid cells within peripheral 1. C. Darwin, The Descent of Man (D. Appleton, 1871).
neurons’ firing rates changed over time. 2. K. Lorenz, On Aggression (Routledge, 1966).
tissues. www.science.org/doi/10.1126/
When we plotted neural activity in this 3. E. O. Wilson, Sociobiology: The New Synthesis (Harvard
science.ade2132 Univ. Press, 1975).
summarized space, we saw the same strong
4. J. Archer, F. Huntingford, in The Dynamics of Aggression
response when another mouse was intro- (Psychology Press, 2013), pp. 21–50.
duced to the imaged animal, followed by a signal captured about 20% of the variance 5. W. R. Hess, Physiologie 42, 554 (1928).
gradual change in activity that evolved over in the data. When the level of activity along 6. M. R. Kruk et al., Brain Res. 260, 61 (1983).
7. A. G. Reeves, F. Plum, Arch. Neurol. 20, 616 (1969).
tens of seconds (12). As we compared rep- this dimension was low, mice investigated 8. D. Lin et al., Nature 470, 221 (2011).
resentations across nine animals, a trend or ignored each other. As activity in the di- 9. H. Lee et al., Nature 509, 627 (2014).
emerged. In every animal, there was one mension increased, investigatory behavior 10. T. Karigo et al., Nature 589, 258 (2021).
11. R. Remedios et al., Nature 550, 388 (2017).
“persistent” dimension of population activ- transitioned to low-intensity aggressive ac- 12. A. Nair et al., bioRxiv 10.1101/2022.04.19.488776
ity: a pattern of firing across neurons that tions such as dominance mounting, and as (2022).
would gradually grow in prevalence and it saturated, animals began to show bouts 13. A. Kennedy et al., Nature 586, 730 (2020).
14. K. C. Berridge, Physiol. Behav. 81, 179 (2004).
remain present for long periods of time. We of outright attack. Dynamics of this activa-
15. K. Lorenz, P. Leyhausen, Motivation of Human
reexamined our recordings and were able tion pattern mirrored animals’ motivation to and Animal Behavior: An Ethological View (Van
to point to a subset of neurons that showed fight. In mice that rarely fought, any activity Nostrand-Reinhold,1973).
the same dynamics; overall, this ramping along this dimension quickly decayed back 10.1126/science.ade2128

INSI GHTS
NEUROSCIENCE
PRIZE ES SAY
FINALIST
Kevin Guttenplan
Why do neurons die?
Kevin
Astrocytes emerge as key mediators of neurodegeneration
Guttenplan
received his
undergraduate By Kevin Guttenplan drives reactive astrogliosis in diverse hu-
degree in Neu- man diseases and established genetic tools
W
roscience and hy do neurons die in neurode- to investigate the influence of astrocytes on
Mathematics generative disease? Despite dec- disease pathogenesis.
from Pomona College. He then ades of research, this question Next, we asked what role reactive astro-
went to Stanford University for remains a fundamental and un- cytes play in disease and injury. We started
his PhD, where he worked first in solved mystery of neurobiology. with a simple mouse model of axonal injury.
the laboratory of B. Barres and One clue came when 19th-century Retinal ganglion cells (RGCs) are neurons
then, after Barres’s passing, in scientists noticed that astrocytes, cells that that project from the eye into the brain,
the laboratory of A. Gitler and normally support the health of neurons, and acutely injuring RGC axons leads to the
studied the role of astrocytes in dramatically change morphology after neu- death of these neurons. Surprisingly, our tri-
disease and injuries of the ner- ronal damage (1), a response termed “re- ple-knockout mice that lack the reactive as-
vous system. He is now a Helen active astrogliosis.” Almost 150 years later, trocyte response were completely resistant
Hay Whitney postdoctoral fellow researchers found that astrocytes become to this injury, and none of their RGCs died
in the laboratory of M. Freeman actively neurotoxic in mouse models of after axon damage (8). This demonstrated
in the Vollum Institute at Oregon
amyotrophic lateral sclerosis (ALS) (2, 3), that factors that drive reactive astrocyte for-
Health & Science University,
hinting that reactive astrocytes might play mation also drive neurodegeneration after
studying the role of astrocytes
an important role in neurodegeneration. acute injury and that loss of these factors
in neuronal circuits. www.science.
org/doi/10.1126/science.ade2129
Despite this evidence, researchers have prevents neuronal death.
been historically hampered by a lack of We then explored a mouse model of glau-
tools to determine how astrocytes become coma in which elevated intraocular pres-
reactive and drive neuronal death (4). To sure leads to progressive RGC death. Just
unravel the mysterious connection between as in acute injury, inactivating the factors
astrocytes and neurodegeneration, my col- that induce reactive astrocytes prevented
leagues and I have pioneered methods to the death of neurons in this chronic in-
track and prevent astrocyte reactivity and jury model (8). Further, morphological and
uncovered a mechanism underlying astro- electrophysiological analysis of surviving
cyte-mediated toxicity. neurons revealed that they retained elec-
To better understand what reactive as- trical responses to light stimulation and
trocytes do, we first sought to determine dendritic targeting in the retina. Together,
what factors make nonreactive astrocytes these results raise hope that damaged neu-
become reactive. We characterized the gene rons could be reincorporated into their en-
expression profile of reactive astrocytes in dogenous neural circuits and, in the case of
vivo and screened for molecules that could glaucoma, restore vision (8).
replicate this reactivity signature (5, 6). We Finally, we tested the role of astrocytes in
found that the cytokines interleukin-1a a mouse model of ALS. By inhibiting reactive
(IL-1a), tumor necrosis factor–a (TNFa), astrocytes using our triple-knockout strat-
and complement component 1q (C1q) to- egy in the SOD1G93A mouse model of ALS, we
gether are able to instruct astrocytes to slowed the progression of motor impairment
become reactive. To test this model in vivo, and extended the overall life span of the mice
we generated a triple-knockout mouse lack- by more than 50% (9). This result represents
ing IL-1a, TNFa, and C1q and found that one of the longest preservations of life span
the reactive-astrocyte response after neu- ever reported in SOD1G93A mice. Combined
ronal injury was eliminated. Importantly, with our data in mouse models of acute and
this molecular cascade is highly conserved chronic injury, this protection highlights as-
in humans: We discovered a marker of this trocytes as drivers of neurodegeneration and
type of reactive astrocyte in our mouse could propel therapeutic strategies that tar-
models and observed that the same marker get reactive astrogliosis.
was substantially up-regulated in brain tis- Once we knew that astrocytes were in-
sue from patients with Alzheimer’s disease, volved in the brain’s response to acute in-
Parkinson’s disease, ALS, and multiple scle- jury and neurodegeneration, we wondered
PHOTO: AARON BIELECK
rosis (5, 7). Through these experiments, what mechanism they used to drive neu-
we revealed a molecular mechanism that ronal death. Our first clue was that media
conditioned by reactive astrocytes, but not
Vollum Institute, Oregon Health & Science University, by nonreactive astrocytes, was actively toxic
Portland, OR, USA. Email: [email protected] to neurons (2, 3, 5). This effect indicated
485-A 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE

that reactive astrocytes were secreting 5. S. A. Liddelow et al., Nature 541, 481 (2017).
something that could kill neurons. To iden- 6. J. L. Zamanian et al., J. Neurosci. 32, 6391 (2012).
7. L. Barbar et al., Neuron 107, 436 (2020).
tify this factor, we biochemically fraction-
8. K. A. Guttenplan et al., Cell Rep. 31, 107776 (2020).
ated reactive astrocyte–conditioned media 9. K. A. Guttenplan et al., Nat. Commun. 11, 3753 (2020).
and identified candidate toxic proteins by 10. K. A. Guttenplan et al., Nature 599, 102 (2021).
mass spectrometry (10). Unexpectedly, all
toxic fractions contained apolipoprotein E 10.1126/science.ade2129
(ApoE) and apolipoprotein J (ApoJ). ApoE
and ApoJ normally shuttle lipids between
cells in structures called lipoparticles, and
mutations in these genes are known to in-
crease the risk of developing Alzheimer’s
disease. Notably, using antibodies to deplete
ApoE- and ApoJ-containing lipoparticles
from reactive-astrocyte media blocked tox-
icity. Although ApoE- and ApoJ-containing
lipoparticles appeared to be required for
astrocytes to kill neurons, additional ex-
periments suggested that the lipoproteins
themselves were not toxic. Instead, our data
suggested that the toxic factor was a lipid
carried within these lipoparticles.
What, then, is the lipid responsible for
astrocyte toxicity? We performed mass
spectrometry to compare lipids secreted
by reactive versus control astrocytes and
found an increase in abundance of long-
chain saturated free fatty acids and phos-
phatidylcholines. These lipids are normally
present at very low levels in the cell and
can induce a form of cell death known as
lipoapoptosis. To our surprise, we found
that these lipids are largely responsible for
astrocyte toxicity—we could kill neurons in
vitro with long-chain saturated lipids and
could prevent astrocyte toxicity by elimi-
nating mediators of lipoapoptosis. We then
inactivated an enzyme required for the syn-
thesis of toxic long-chain lipids specifically
in astrocytes in mice and showed that it re-
duced the death of RGCs after axon injury,
establishing the same mechanism of lipid-
mediated neuronal death in vivo.
Through this work, we developed tools
to track and inhibit reactive astrogliosis, a
common feature of neurodegeneration. We
used these tools to reveal that reactive as-
trocytes drive neuronal death and disease
progression in mouse models of acute in-
jury and neurodegenerative disease. Finally,
we discovered a long-sought-after and unex-
pected mechanism—the secretion of a lipid
rather than a protein—by which astrocytes
can kill neurons. These findings highlight
that reactive astrocytes and the secretion of
toxic lipids are powerful mediators of neu-
rodegeneration in vitro and in vivo and are
promising targets for future therapies. j

1. J. Grimm, Virchows Arch. 48, 445 (1869).
2. F. P. Di Giorgio, M. A. Carrasco, M. C. Siao, T. Maniatis,
K. Eggan, Nat. Neurosci. 10, 608 (2007).
3. M. Nagai et al., Nat. Neurosci. 10, 615 (2007).
4. K. A. Guttenplan, S. A. Liddelow, J. Exp. Med. 216, 71 (2019).
SCIENCE science.org 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 485-B

INSI GHTS
NEUROSCIENCE
PRIZE ES SAY
FINALIST
Filipa Cardoso
The brain-fat connection
Filipa Cardoso
Type 2 innate lymphoid cells shape metabolism
received her through a brain-body circuit
Bachelor’s
degree in Bio-
chemistry By Filipa Cardoso to regulate adipose tissue metabolism (5).
and Master’s As a result, we identified a brain-adipose
A
degree in Health dipose tissue is not merely a fat stor- circuit that controls adipose tissue physi-
Sciences from Universidade do age site but is, in effect, an active ology through neuro-mesenchymal units,
Minho where she was introduced neuroendocrine organ that supports which modulate ILC2 function. In these cir-
to the field of immunology, essential body functions, ranging cuits, sympathetic neurons activate the β2-
working in the immune response from reproduction to muscle activity. adrenergic receptor of nearby adipose mes-
to tuberculosis infection and to These neuroendocrine functions are enchymal cells, inducing the expression of
colitis models. After her Master’s, only possible, thanks to a complex commu- glial-derived neurotrophic factor (GDNF).
she moved for a PhD in H. nication network between the brain and fat Consequently, GDNF controls the activity of
Veiga-Fernandes’s laboratory to tissue. Adipose tissue communicates with ILC2s in gonadal fat, which affects energy
develop research in how the ner- the brain through the release of adipokines expenditure, glucose levels, and susceptibil-
vous and the immune systems and afferent innervation outputs, whereas ity to obesity.
interact to control metabolism. the brain talks back to the tissue through Our findings indicate that synergy be-
After completing the PhD, Filipa
systemic hormones and efferent innerva- tween the nervous and immune systems
joined a biopharmaceutical
tion inputs (1). Obesity and obesity-associ- is required for adipose tissue physiology.
company, LiMM Therapeutics,
ated conditions stem from the malfunction Neuro-mesenchymal units stimulate type
which stemmed from the Veiga-
Fernandes laboratory. Currently,
of this fine-tuned regulation between the 2 immune responses, which disseminate
she is striving to translate the brain and fat tissue. and prolong the metabolic effects of sym-
knowledge obtained from basic One of the main axes of adipose tissue pathetic activation on adipocytes working
research to clinical application regulation is the sympathetic nervous sys- in tandem to limit adiposity (5). This inter-
by harnessing the molecular tem (SNS), which conveys brain-derived cues cellular axis may also help to explain why,
cross-talk between neuronal and in the form of catecholamines to adipocytes in the context of obesity, dysregulation of
innate lymphoid cells within pe- and other adipose tissue cells to regulate me- the immune landscape is difficult to repair.
ripheral tissues.www.science.org/ tabolism. SNS-derived signals control many Although in a typically sized adipose depot,
doi/10.1126/science.ade2132 adipose tissue functions, including thermo- the SNS nerve fibers are densely distrib-
genesis, neutral lipolysis, and immune cell uted throughout the tissue, in an enlarged
function. Sympathetic output, in the form of adipose tissue, sympathetic innervation
catecholamines, can limit adiposity by acting is less dense because of the physical con-
directly on adipocytes (2). straints imposed by expanding adipocytes.
Obesity-induced metabolic syndrome is In the healthy adipose tissue, sympathetic
driven by a state of systemic and chronic cues promote thermogenesis, lipolysis, and
low-grade inflammation that originates in type 2 immune responses. The disruption
the adipose tissue. Adipose tissue is rich in of cross-talk between the SNS and ILC2s,
immune cells, which play a vital role in its however, can likely limit type 2 cytokine
physiology. Under homeostatic conditions, production, which contributes to metabolic
type 2 innate lymphoid cells (ILC2s) pro- imbalance (5, 6).
mote a type 2 immune environment that Retrograde tracing and chemical, surgi-
protects the adipose tissue from harmful cal, and chemogenetic neuronal manipu-
inflammation. The ILC2s achieve a protec- lation were used to unravel the anatomi-
tive environment through the production of cal basis for this neuroimmune regulation
interleukin-5 and interleukin-13, while also axis. We identified a previously unappreci-
promoting energy expenditure through the ated prevertebral sympathetic aorticore-
release of met-enkephalin (3, 4). In obesity, nal ganglion that connects to the para-
a gradual shift in immune mediators to- ventricular nucleus of the hypothalamus
ward a type 1 inflammatory response leads (PVH) and other higher-order brain areas.
to systemic inflammation and promotes Manipulation of this neuronal circuit had
PHOTO: COURTESY OF FILIPA CARDOSO

metabolic dysfunction. similar effects on ILC2s as manipulating
Adipose tissue biology is tightly con- systemic sympathetic output, paving the
trolled by both the SNS and the immune way to potential therapeutic interventions
system, and my colleagues and I aimed to for limiting visceral adipose tissue forma-
discover how these two systems interact tion. Notably, the PVH not only coordinates
whole-body metabolism through systemic
Champalimaud Research, Champalimaud Foundation, Lisboa, leptin activity (7), but it also contributes to
Portugal. Email: [email protected] depot-specific metabolic regulation through
485-C 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE

the control of local SNS output to adipose
tissue. Altogether, our study identified a
neuro-mesenchymal unit that converts the
activity of a brain-body circuit into adipose-
resident ILC2 function, thus shaping sys-
temic metabolism.
The integration of neuroscience and im-
munology techniques allowed us to add a
previously unknown axis to the repertoire
of neuroimmune interactions (8) and, most
notably, to identify mesenchymal cells as the
essential intermediary for this communica-
tion. This work emphasizes that the SNS is
not a one-dimensional regulatory axis of
adipose tissue, solely promoting lipolysis
and thermogenesis through adrenergic sig-
naling on adipocytes (2). Instead, an inter-
organ sympathetic circuit integrates neuro-
nal and mesenchymal signals to orchestrate
ILC2 activation and to control adipose tis-
sue physiology.
Historically, obesity treatments have tar-
geted adipocytes (9), central nervous sys-
tem mechanisms of metabolic regulation, or
nutrient uptake. However, these treatments
are not effective for all patients, and most
have considerable side effects. In particu-
lar, dietary interventions have a high failure
rate because of poor long-term compliance
(10). Thus, nutritional education and life-
style interventions are likely to be insuffi-
cient to tackle the current obesity epidemic.
Accordingly, research on mechanisms of pre-
venting fat accumulation or promoting its
loss is of vital importance to control obesity
and its associated comorbidities before they
develop. Our work unravels the complexity of
cellular interactions and brain-body mecha-
nisms that coordinate adipose tissue metab-
olism. It is an enticing prospect that this pre-
viously unappreciated neuronal network will
open new possibilities for fine-tuned control
of obesity and related disorders. j
1. X. Ding et al., J. Endocrinol. 231, 35 (2016).
2. W. Zeng et al., Cell 163, 84 (2015).
3. A. B. Molofsky et al., J. Exp. Med. 210, 535 (2013).
4. J. R. Brestoff et al., Nature 519, 242 (2015).
5. F. Cardoso et al., Nature 597, 410 (2021).
6. X. Meng, W. Zeng, Immunity 54, 2191 (2021).
7. A. K. Sutton, M. G. Myers Jr., D. P. Olson, Annu. Rev.
Physiol. 78, 207 (2016).
8. C. Godinho-Silva, F. Cardoso, H. Veiga-Fernandes, Annu.
Rev. Immunol. 37, 19 (2019).
9. M.-W. Lee et al., Cell 160, 74 (2015).
10. K. D. Hall, S. Kahan, Med. Clin. North Am. 102, 183 (2018).
SCIENCE science.org 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 485-D

SPECI AL SECTIO N
REVIEWS
Atlas-based data integration for
mapping the connections and
architecture of the brain p. 488
Solving brain circuit function and
dysfunction with computational
modeling and optogenetic fMRI p. 493
Scale matters: The nested human
connectome p. 500
The emergent properties of
the connected brain p. 505
RELATED ITEMS
PERSPECTIVE p. 475
RESEARCH ARTICLE p. 514
NO NEURON IS
AN ISLAND By Peter Stern
T
he brain is so much more than its con- of neurons in animals and noninvasively mea-
stituent cells. Each neuron in the brain sure how they activate other parts of the brain,
connects with thousands of other whether near or far. Advances in brain imaging
ROXANA KOOIJMANS/NETHERLANDS INSTITUTE FOR NEUROSCIENCE/HUMAN BRAIN PROJECT

neurons—but instead of a cacophony reveal anatomical projections and functional
IMAGE: MARKUS AXER AND KATRIN AMUNTS/INM-1; FORSCHUNGSZENTRUM JÜLICH AND

of connections, we have a synchro- connectivity patterns, allowing us to see their ac-
nized symphony. tivation in real time. The first digital brain atlas-
Coordination of the body’s myriad es of the mouse and rat, for example, provide as-
functions, behaviors, and thoughts tonishing insights into the connectivity of cells.
requires large numbers of neurons to With a better understanding of the complexity
act cooperatively and not as isolated entities. of normal brain connections, we learn more about
The outcomes are driven by the connections what goes wrong when they are disrupted. And the
between the neurons, whether that involves view of connectivity patterns in various organisms
communicating with a neighboring nerve cell begins to reveal the steps involved in the evolution
or sending and receiving signals to and from from the simplest neuronal nets to the multilay-
distant areas of the brain. ered, multinucleate mammalian brain.
Innovative neuroscientific techniques allow re- Without connections that run smoothly, the
searchers to specifically stimulate select groups brain is nothing more than a pile of neurons.
Details of a human brain section, which show the architecture of fibers down to single axons in cortical and subcortical structures, are revealed by
three-dimensional (3D) polarized light imaging. The colors represent 3D fiber orientations and highlight pathways of individual fibers and tracts.
486
487
B R A IN CO N N EC T I VI T Y
REVIEW dimensional (3D) digital brain atlases (3–6)

offer new opportunities for extensive data in-
Atlas-based data integration for mapping the tegration aimed at improving our understand-
ing of the organization of the brain. These
connections and architecture of the brain integration efforts are accelerated by the use
of tools for the registration of heterogeneous
Trygve B. Leergaard* and Jan G. Bjaalie data types to the atlases in combination with
computerized workflows for subsequent auto-
Detailed knowledge about the neural connections among regions of the brain is key for advancing our mated analyses of large data collections. Fo-
understanding of normal brain function and changes that occur with aging and disease. Researchers cusing on the rodent brain as a model system
use a range of experimental techniques to map connections at different levels of granularity in rodent for basic neuroscience, we review approaches
animal models, but the results are often challenging to compare and integrate. Three-dimensional for the mapping of neural connections and
reference atlases of the brain provide new opportunities for cumulating, integrating, and reinterpreting atlas-based solutions for integrating and ana-
research findings across studies. Here, we review approaches for integrating data describing neural lyzing data and discuss future directions for
connections and other modalities in rodent brain atlases and discuss how atlas-based workflows can facilitate advancing the field.
brainwide analyses of neural network organization in relation to other facets of neuroarchitecture.
Mapping brain connections
A variety of techniques are available for the
he brain is composed of vast numbers of work also depend on the physiological and mapping of neural connections at different
T neurons, glia, and vasculature, encased

within a solid skull. It processes and
stores information; generates memories,
thoughts, and ideas; performs planning;
and effectuates a wide range of behaviors.
Dendrites and axons allow neurons to transmit
neurochemical properties of neurons, includ-
ing their detailed local cellular relationships
to other neurons (microcircuitry) and to sup-
porting cells with sustaining or modulatory
functions. Comprehensive knowledge about
how the brain exerts its functions thus requires
levels of granularity. The overall trajectories
of fiber bundles in the brain can be visual-
ized with myelin staining (7, 8) (Fig. 2B) or po-
larized light imaging in histological sections (9)
(Fig. 2B) and whole-brain diffusion magnetic
resonance imaging (MRI) methods (8, 10, 11)
signals across shorter or longer distances, and integration of knowledge about all these fea- (Fig. 2B). The neuron-by-neuron connections
axons profusely branch into terminal fields with tures. We argue that recently introduced three- are mapped with the use of high-resolution
multiple synaptic contacts to other neurons. The
functions performed by neurons are determined
to a high degree by their connections with other A B Cerebral cortex
neurons within and across brain regions.
Large ensembles of widely distributed neu- Striatum
Cerebral cortex D2 D1 Thalamus
rons make up complex neural networks. The Cerebellum
networks are highly organized, typically with DCN SNc
different cell types distributed in layers or Thalamus GP
clusters. Within the network, populations of CPu
SNr STN EPN/SNr DCN PN
neurons exert specific excitatory, inhibitory,
GP EPn PPN / PN
or modulatory influences on other parts of Cerebellar
the network, and variations in the strengths STN PN cortex
Brain stem/spinal cord
and spatial distributions of the connections,
including specific patterns of divergence and D Cerebral cortex
E1
convergence, influence how the network sub- C D E3
E2
serves its functions. Overall, knowledge about
the organization of the networks—the wiring
patterns of the brain—is critical for understand-
ing normal brain function and is typically em- Thalamus
SC E i Thalamus
bedded in network models aimed at elucidating
CPu
and studying a variety of brain functions (1). TN
Insight into the detailed organization of wir-
ii CPu iii PN
ing patterns is also key to understanding and
PN
treating brain disorders. One example is the
use of knowledge about the wiring of deep
brain structures (Fig. 1, A and B) for treat-
ment of neurological disease—e.g., the use of Fig. 1. Wiring patterns in the brain. (A and B) Schematic visualizations of basal ganglia and cerebellar
electrical stimulation targeting the subthalamic neural networks, topologically drawn (A) in an oblique slice through the WHS rat brain atlas (RRID:
nucleus or specific parts of the thalamus for SCR_017124) and as a box diagram (B). (C) Illustration of tract-tracing experiments in which an anterograde
ameliorating symptoms of Parkinson’s disease tracer is placed in the cerebral cortex, taken up by groups of neurons, and transported along the axons
and medication-resistant tremor (2). and their branches to visualize projections to intracortical and subcortical regions. (D and E) Microscopic
Although the patterns of neural wiring direct images from a tract-tracing experiment [data from (41)] in which an axonal tracer (visualized as black
how neural signals are distributed through a labeling) was injected into the primary somatosensory cortex (D), giving anterograde labeling of dense axonal
network, the functional characteristics of a net- plexuses in the thalamus [(E), i], caudoputamen [(E), ii], and pontine nuclei [(E), iii]. CPu, caudoputamen;
DCN, deep cerebellar nuclei; EPn, entopeduncular nucleus; GP, globus pallidus; SNc, substantia nigra,
Neural Systems Laboratory, Institute of Basic Medical
Sciences, University of Oslo, Oslo, Norway. pars compacta; SNr, substantia nigra, pars reticulata; PN, pontine nuclei; PPN pedunculopontine nucleus;
*Corresponding author. Email: [email protected] TN, trigeminal nuclei; STN, subthalamic nucleus; SC, superior colliculus. Scale bars, 1 mm (D) and 200 mm (E).

microscopic techniques that allow the imaging A Waxholm Space rat brain atlas v4 with integrated microscopic data
of tracer-filled individual neurons (12, 13) or by
the use of serial electron microscopy visualiza-
tion of cellular and synapse ultrastructure (14).
The current foundation for whole-brain map-
ping of neuronal connections is, however, pro-
vided by invasive tract-tracing experiments in
wild-type and transgenic animals (Figs. 1 and 2)
(15, 16). These methods are highly suitable for B Specific axonal connections and general fiber architecture
S1 whisker S1 forelimb Myelin Polarized light imaging
describing connections at the level of groups of
neurons, demonstrating patterns that are per-
sistent and reproducible among individuals and
therefore useful for inferring functional proper-
ties and disease-related changes [for a discussion
of different levels of connectivity analysis, see (17)].
In classical tract-tracing experiments, a tracer
substance is deposited in a specific location in
the brain and taken up by groups of neurons
(Fig. 1, C to E). The tracer is transported along
the axons of the neurons anterogradely, from
neuronal cell bodies to their axonal terminal
fields; retrogradely, from axonal terminal fields
to the neurons of origin; or in both directions
(15, 16). Depending on the tracer used, some of C Atlas-based comparison of location
the morphologies of the labeled neurons are
revealed, or the tracer is transported across
synapses, which allows for the identification of PO
pre- and postsynaptic connections in a net- VPM
work (16). Creation of new animal models has VPL
also made it possible to use advanced cell type–
specific tracing paradigms with genetically con- IC
trolled expression signals (18).
A key methodological innovation for the Fig. 2. WHS rat brain atlas with integrated microscopic data. The WHS rat brain atlas is enriched with
tract-tracing methods has been serial two- spatially registered microscopic image data, which allow comparison across different experiments and
photon tomography (19). By allowing block- data types. (A) Illustration of tract tracing and myelin-stained microscopic images of coronal sections
face acquisition of high-resolution microscopic registered to the brain at the level of the thalamus, indicated with a frame. (B) Overview and details from four
images, this technology has been successfully different coronal microscopic images taken from the same anteroposterior level of the thalamus, including
used by the Allen Institute to generate large anterogradely labeled corticothalamic projections from whisker [data from (41)] and forelimb [data from
volumes of microscopic, 3D tract-tracing im- (42)] representations in the primary somatosensory cortex (S1), tissue fiber orientations visualized by myelin
age data showing brainwide connections in the staining [data from (43)], and polarized light imaging [data from (44)]. (C) WHS rat brain atlas superimposed
mouse brain (5, 18). Using a similar approach, on the tract-tracing image shown in (B), providing a starting point for interpreting the spatial location of
the MouseLight project of Janelia Research axonal labeling across subregions of the thalamus. IC, internal capsule; PO, posterior thalamic nucleus; VPL,
Campus has created high-resolution volumet- ventral posterolateral thalamic nucleus; VPM, ventral posteromedial thalamic nucleus. Scale bars, 1 mm.
ric reconstructions of individual axonal trajec-
tories across entire mouse brains (20, 21).
brain experimental studies revealed substan- For the mouse brain, the atlases developed
Atlases for brainwide mapping of connections tial differences in the research design, inter- by the Allen Institute for Brain Science have
and related features pretation of results, and reproducibility among become widely used resources. The most re-
Traditionally, experimental tract-tracing studies studies (25). The main challenges were related cent version is defined in a high-resolution
have focused on one or a few brain regions at a to the use of different parcellation schemes and image volume constructed by interpolation of
time, yielding precise information about the the lack of precision in the reporting of how serial two-photon tomography (STPT) images
connections among a few selected regions of observations would map onto a given ana- from 1675 adult mice (the Allen Mouse Brain
interest. Extensive literature-mining efforts tomical scheme. Although atlases for mouse Common Coordinate Framework, CCFv3). In
have aggregated information from available and rat brains (26–28) have assisted research- this population-averaged image volume, many
publications into databases to attain a more ers for decades in assigning location to their anatomical delineations were defined using
complete understanding of the connections observations, their utility is limited in that information from a large body of multimodal
between brain regions (22, 23). However, these they are 2D and lack efficient and standardized image data registered to the CCFv3 template (6).
valuable resources are limited by the diversity tools for the registration of observations to the The underlying images included the collection
of the methods used, the variable levels of pre- atlases. These limitations have been overcome of STPT data created for the Allen mouse brain
cision in the reporting of brain location, and with a new generation of open-access 3D at- connectivity atlas (5). Additional images were
the lack of access to the underlying data (17, 24). lases, which integrate information from multi- readily integrated into the atlas volume using
Regarding the reporting of the anatomical lo- ple anatomical parcellation schemes and have the same high-throughput imaging and infor-
cation of observations, a recent analysis of prac- powerful tools for the registration of data to matics pipeline, as achieved with the inclusion
tices and precision from 120 different rodent atlases and atlas-based analyses. of more than 1000 STPT image volumes from

Input • Microtomy
• Experiment Serial section
• Section staining
• Brain extraction images
Brain • Image acquisition
Mouse/rat
Atlas registration
Numbers/statistics
• Affine registration Location metadata Quantification per region
Image processing
• Non-linear refinement Atlas maps
• Serial order • Pixel quantification
Machine-
Analysis • Orientation, scale readable image Feature extraction • Coordinate conversion
files • Sorting by atlas regions
• Unique IDs • Signal detection/
classification Segmented Point coordinate
images lists
• Segmentation
Visualization
Use Group comparisons
in silico experiments
Fig. 3. Workflow for data integration and atlas-based analysis. Diagram steps are the spatial registration of images to a volumetric reference atlas and
showing key processing steps (blue) with inputs and outputs (gray) of a generic the extraction of labeling signals from background, providing input combined
workflow for the integration of rodent brain experimental data in volumetric brain in (automated) analyses extracting quantitative measures and 3D point
reference atlases and atlas-based analysis, yielding quantitative data and 3D coordinates representing selected labeling features. The workflow output can be
point coordinate data sorted to atlas-defined brain regions. The input to the visualized and used in statistical analyses for characterizing and comparing
workflow is provided by experimental procedures resulting in serial microscopic labeled parameters. Point coordinates representing labeled neuronal elements
images of tissue sections showing neural labeling. Preprocessing steps include are suitable for interactive 3D visualization and exploration of spatial distribution
validation of image order and orientation and assignment of unique serial patterns and hypothesis-driven in silico experiments visualizing selected
identifiers to create machine-readable image files. Important parallel analytic combinations of data. Usage example data are taken from (45–49).
tract-tracing experiments conducted in trans- of images to an atlas, (ii) sharing of registered analyzing electrode tracts (34, 35), and others
genic Cre-dependent mice (18). images with viewing and navigation of images suitable for many data modalities (36).
For the rat brain, the Waxholm Space (WHS) in atlas space, and (iii) extraction of features
atlas is now available in a version 4 with brain- from the images followed by quantification of Sharing, viewing, and navigating images
wide parcellation (RRID: SCR_017124; https:// the distribution of features within and across in atlas space
www.nitrc.org). This atlas is based on a single brain structures. Below, we review the princi- In addition to having spatial metadata, data
high-resolution ex vivo structural and diffusion ples and practical implementation of the three are integrable only if they are properly an-
MRI volume, in which brain regions have been steps, taking a starting point in the EBRAINS notated with metadata and other structured
identified and delineated by manual interpreta- Atlases services (https://ebrains.eu/services/ information that help us understand the data.
tion of the underlying MRI data, enriched with atlases) for the mouse and rat brain and the Furthermore, the data will have to be discover-
multiple microscopic image data showing dif- EBRAINS Data and Knowledge services for able and accessible. The approach taken by the
ferent facets of the neuroarchitecture (4, 29, 30) data sharing and access (https://ebrains.eu/ EBRAINS Share Data service (https://ebrains.
(see also Fig. 2). The atlas has seen broad in- services/data-and-knowledge). eu/service/share-data) is to provide procedures
terest, as reflected in close to 25,000 downloads, for annotation of the data with metadata ac-
more than 300 citations, and inclusion in sev- Registration of images to an atlas cording to a metadata standard and to provide
eral services or products (e.g., EBRAINS re- Data integration requires that the data are made descriptions and other information required to
search infrastructure, https://ebrains.eu, and comparable, as if data of different origins were make the data interpretable and reusable. Fol-
the Neuroinfo software of MBF Bioscience). To part of a single dataset. Thus, the registration lowing a curation of the metadata, the data are
facilitate comparisons of different atlas parcel- of data of different origins to the same atlas stored in the EBRAINS data repository, whereas
lation schemes, seven versions of the Paxinos framework is a key step toward integration. In the metadata are ingested in a knowledge graph
and Watson rat brain atlases and four versions EBRAINS, series of 2D histological images of that makes the data findable through a search
of the Swanson rat brain atlas were registered to mouse or rat brains are spatially registered user interface (https://search.kg.ebrains.eu) or
the WHS rat brain atlas (31, 32). to the atlases by an initial landmark-based a programmatic access route (https://ebrains.
anchoring of sections followed by a nonlinear eu/service/find-data/). A search points the user
Atlas-based data integration and analysis adjustment. The EBRAINS tools supporting this to data cards with information about the data,
The 3D reference atlas spaces provided by step are the QuickNII tool for affine registra- access to the datasets, and links to a virtual
the Allen Mouse Brain Common Coordinate tion [RRID: SCR_016854; (33)] and the VisuAlign microscopy viewer for inspection of the image
Framework (CCFv3 and earlier versions) and tool for nonlinear registration adjustments (RRID: data integrated in the atlas.
the WHS rat brain atlas are useful frameworks SCR_017978). The output of the registration pro- With use of the EBRAINS tools, a broad range
for integrating heterogeneous data originat- cess is referred to as spatial metadata: a set of of data generated with different methods and
ing from different researchers and research anchoring vectors and deformation fields that in different animals have been registered to the
projects. Recently introduced tools and work- together define the transformations between atlases. Figure 2 shows examples of histological
flows designed for use with the mouse and rat the image data and the atlas. A range of other images registered to the WHS rat brain atlas
atlases support a three-step process for data registration tools are available, with some having v4. In the thalamus, as the selected region of
integration and analysis (Fig. 3): (i) registration been developed for particular use cases, such as interest, data on specific connections of the

A B C D Corticopontine topography
PN
PN
Topography of rat corticopontine projections mapped in atlas space Wild-type mice
E M2: unaltered topography F S1: altered topography G H

M2
S1
Neuron 1 Neuron 2 S1
S1 S1 mouth S1 bfd
Nex-cKO S1 mouth 1 2
Neuron 1 Neuron 2
Control S1 bfd PN
Neuron 1
Neuron 2
PN
Transgenic mice lacking Nr2f1 Whole-neuron tracing Atlas-integrated data

Fig. 4. 3D analysis of spatially integrated connections data. Spatial secondary motor cortex. (G and H) Data from different sources can be integrated
organization in the first link of the rat and mouse cerebro-cerebellar circuit, and compared in atlas space. (G) Whole-brain reconstructions of two projection
based on tract-tracing data integrated and covisualized in the WHS rat brain atlas neurons [from (21)] located in the mouth and whisker representations of the
[(A) to (C)] and the Allen Mouse Brain Common Coordinate Framework, CCFv3 primary somatosensory cortex, with a variable amount of profusely branching
[(E) to (G)]. (A to C) Data from several experiments in which an axonal tracer axons in several subcortical regions, including the pontine nuclei. (H) A
was placed at separate locations in the cerebral cortex. (A) Anterogradely labeled comparison of corticopontine projections from single neurons and anterogradely
fibers were semiquantitatively represented as point coordinates, spatially labeled projections from two populations of neurons in the mouth and
registered within the pontine nuclei. (B) The WHS rat brain atlas in view from whisker representations in the primary somatosensory cortex [data from (5)]
ventral. (C) The outer boundaries of the pontine nuclei and descending indicating the trajectory of single-cell projections relative to the point cloud
corticofugal fiber tract as gray transparent surfaces and axonal labeling from representing the spatial distribution of projections from a larger number of
different experiments as color-coded point clouds. Shifts in location of tracer neurons in the same part of the cerebral cortex. The data in (A) to (C) are from
injection sites from anterolateral to dorsomedial locations (A) correspond to (45) (cases R113-BDA, R118-BDA, and R124-BDA), (46) (case D55-FR), (47)
topographic shifts from internal to external lamellar subspaces in the pontine (cases M27-BDA and M27-FR), and (48) (cases R409-BDA, R412-BDA, and
nuclei (C). (D to F) Data from a study of the impact of the cortical area– R413-BDA). The data in (E), (F), and (H) are from (51) (wild-type cases:
patterning gene Nr2f1 on topographical organization of corticopontine projections 100141780, 112229814, 112952510, 126908007, 127084296, 127866392,
in mice (40). (D) The distribution of corticopontine projections from different 141602484, 141603190, and 585025284; Nex-cKO cases: 11643_17 and 19423_7;
cortical locations in wild-type mice shows a similar inside-out topographical littermate control case: 18035_1). The reconstructions in (G) are from (20, 21)
organization as that in rats (50) [(A) to (C)]. (E and F) Comparison of the [https://www.janelia.org/project-team/mouselight: neuron AA0945 (neuron 1)
topographical distribution of pontine projections arising from corresponding and neuron AA1049 (neuron 2)]. Atlas surfaces and data points were covisualized
locations in the secondary motor cortex (E) and the primary somatosensory using the MeshView web application, RRID: SCR_017222, https://www.nitrc.org.
cortex (F) of Nex-cKO transgenic mice lacking Nr2f1 and control animals Neuron reconstructions in (G) were visualized using the Scalable Brain Atlas
demonstrates that corticopontine projections from the primary somatosensory Composer [https://sba-dev.incf.org/composer (52)]. bfd, barrel field; M2,
cortex are altered in Nex-cKO mice to resemble the projections from the secondary motor cortex.
primary somatosensory cortex are available Feature extraction, quantification, and distribution gions of interest, filtrating artifacts, and applying
together with images showing bundles of Finally, features from the images can be extracted, quality control steps. It also allows for the export
axons and their orientation with myelin stain- sorted by brain region, and displayed and fur- of coordinates to other tools for 3D visualization
ing and polarized light imaging. Figure 2C ther analyzed in three dimensions. To this end, of the distribution of the selected features, from
demonstrates some of the basic functional- various approaches are used, as exemplified in the whole brain or selected regions. Figure 4
ities of the virtual microscopy viewer that is the EBRAINS workflow for automation of sev- shows examples of regional analysis of brain con-
available through the data cards. The viewer eral of the steps (37, 38). This workflow consists nection features extracted from images registered
supports web-based pan and zoom of high- of a registration tool, a machine learning–based to the WHS rat brain atlas and the Allen Mouse
resolution images with overlays of the atlas tool for extraction of selected features in the im- Brain Common Coordinate Framework.
parcellation map generated by a volumetric atlas ages [ilastik (39)], and a tool for quantifying the Figure 4 shows examples of regional analyses
slicer. The user can inspect the images at cellular features per atlas region [Nutil (38)]. The work- of brain connection features extracted from im-
resolution and observe brain regions, names, flow allows the users to customize their analysis ages registered to the WHS rat brain atlas or the
and boundaries and annotate points of inter- in many aspects, including choosing the granu- Allen Mouse Brain Common Coordinate Frame-
est to extract atlas coordinates. larity level of the atlas, defining their own re- work. The data originate from different research

projects and data repositories but are integrated anatomical location and are embedded in soft- spatially registered to the Waxholm Space atlas of the rat brain,
and made comparable by registration to the ware tools for integration and analysis. Data- data set, Human Brain Project Neuroinformatics Platform (2019);
https://doi.org/10.25493/486N-966.
same atlas spaces. Specific combinations of tract- sharing services, such as those delivered by 33. M. A. Puchades, G. Csucs, D. Ledergerber, T. B. Leergaard,
tracing data showing terminal fields of axons in EBRAINS, organize the data and help standard- J. G. Bjaalie, PLOS ONE 14, e0216796 (2019).
the corticopontine projection system have been ize the metadata, including metadata about 34. P. Shamash, M. Carandini, K. Harris, N. Steinmetz, bioRxiv
447995 [Preprint] (2018). https://doi.org/10.1101/447995.
selected for analysis of topographical organiza- the location of research data from the brain. 35. J. Paglia, P. Saldanha, J. G. Fuglstad, J. R. Whitlock, bioRxiv
tion (Fig. 4, A to D) and identification of changes Through the atlas frameworks, data from 2021.10.01.462770 [Preprint] (2021). https://doi.org/10.1101/
in topography resulting from lack of specific individual researchers published in research 2021.10.01.462770.
36. D. Fürth et al., Nat. Neurosci. 21, 139–149 (2018).
gene expression (Fig. 4, E and F). Furthermore, articles and integrated in the atlases are made 37. S.C. Yates et al., Front. Neuroinform. 13, 75 (2019).
tract-tracing data showing corticopontine pro- directly comparable to data from large-scale 38. N. E. Groeneboom, S. C. Yates, M. A. Puchades, J. G. Bjaalie,
jections from large groups of cortical neurons mapping efforts, such as the Allen Mouse Brain Front. Neuroinform. 14, 37 (2020).
39. S. Berg et al., Nat. Methods 16, 1226–1232 (2019).
have been combined with 3D reconstructions Connectivity Atlas and the MouseLight project.
40. C. Tocco, M. Øvsthus, J. G. Bjaalie, T. B. Leergaard, M. Studer,
showing individual neurons and their exten- The new paradigm for research on brain con- Development 149, dev200026 (2022).
sive branching patterns, including branching nections and brain architecture in general is to 41. I. Zakiewicz, Y. Van Dongen, T. B. Leergaard, J. G. Bjaalie,
to multiple target clusters within the pontine bring the research data into the same reference Axonal projections from the D3 whisker representation of the
rat primary somatosensory cortex (case R604), data set,
nuclei and elsewhere in the brain stem (Fig. 4, G space, share the data, and prepare the data for Human Brain Project Neuroinformatics Platform (2019);
and H), illustrating opportunities for parallel systematic reanalysis and reinterpretations of https://doi.org/10.25493/9MNV-Y97.
processing and neural circuit complexity. our understanding of the brain. With these new 42. I. Zakiewicz, Y. Van Dongen, T. B. Leergaard, J. G. Bjaalie,
Axonal projections from the forelimb representation of the rat
approaches being introduced in neuroscience, primary somatosensory cortex (case R605), data set,
Conclusion and outlook literature mining can be supplemented with Human Brain Project Neuroinformatics Platform (2019);
Online repositories containing large collections powerful mining of the data underlying the https://doi.org/10.25493/ZF26-DZK.
43. T. B. Leergaard, S. Lillehaug, A. Dale, J. G. Bjaalie, Atlas of
of experimental data integrated in an open- interpretations included in publications. normal rat brain cyto- and myeloarchitecture, data set,
access volumetric reference atlas have proven Human Brain Project Neuroinformatics Platform (2018);
RE FERENCES AND NOTES
successful for mouse brain research, as made https://doi.org/10.25493/C63A-FEY.
1. O. Sporns, Nat. Neurosci. 17, 652–660 (2014). 44. N. Schubert et al., Wistar rat brain fibre orientation model, data
evident from the impressive data and results 2. T. Wichmann, M. R. DeLong, Neurotherapeutics 13, 264–283 set, Human Brain Project Neuroinformatics Platform (2019);
provided, e.g., by the Allen Institute and the (2016). https://doi.org/10.25493/F9RX-65U.
3. K. Amunts et al., Science 340, 1472–1475 (2013).
Janelia Research Campus (5, 18, 20, 21). Open- 45. T. B. Leergaard et al., Three-dimensional somatotopic organization
4. E. A. Papp, T. B. Leergaard, E. Calabrese, G. A. Johnson, of rat pontine projections from major body representations of the
access sharing of standardized data mapped in J. G. Bjaalie, Neuroimage 97, 374–386 (2014). primary somatosensory cortex, data set, EBRAINS (2020);
an appropriate anatomical context makes it pos- 5. S. W. Oh et al., Nature 508, 207–214 (2014). https://doi.org/10.25493/XQ7V-QN9.
sible to find and efficiently use new combina- 6. Q. Wang et al., Cell 181, 936–953.e20 (2020). 46. T. B. Leergaard, K. D. Alloway, J. J. Mutic, J. G. Bjaalie, Three-
7. T. B. Leergaard et al., PLOS ONE 5, e8595 (2010). dimensional somatotopic organization of corticopontine
tions of data, suitable for characterizing and 8. N. S. White, T. B. Leergaard, H. D’Arceuil, J. G. Bjaalie, projections from whisker representations in the rat primary
investigating many aspects of brain connections. A. M. Dale, Hum. Brain Mapp. 34, 327–346 (2013). somatosensory (barrel) cortex, data set, EBRAINS (2020);
However, despite the impressive amounts of 9. N. Schubert, M. Axer, U. Pietrzyk, K. Amunts, in High-Resolution https://doi.org/10.25493/NECX-22S.
Neuroimaging: Basic Physical Principles and Clinical Applications, 47. T. B. Leergaard et al., Three-dimensional topography of
data presented, attempts to use these generous A. M. Halefoğlu, Ed. (IntechOpen, 2017). corticopontine projections from the rat sensorimotor
resources may also reveal that data coverage 10. S. Mori, P. C. van Zijl, NMR Biomed. 15, 468–480 (2002). cortex (S1, S2, M1), data set, EBRAINS (2020); https://doi.org/
may be insufficient to answer challenging ques- 11. B. Jeurissen, M. Descoteaux, S. Mori, A. Leemans, NMR 10.25493/1SSX-606.
Biomed. 32, e3785 (2019). 48. T. B. Leergaard, J. G. Bjaalie, Topography of the complete
tions, as exemplified in the recent study by 12. R. Parekh, G. A. Ascoli, Neuroscientist 21, 241–254 (2015). corticopontine projection, data set, EBRAINS (2020);
Tocco et al. (40), which demonstrated that studies 13. M. Wang et al., Nat. Commun. 13, 1531 (2022). https://doi.org/10.25493/ZM3M-YNJ.
of topographic organization in the corticopontine 14. N. L. Turner et al., Cell 185, 1082–1100.e24 (2022). 49. E. R. Cullity, I. E. Bjerke, K. Kjelsberg, T. B. Leergaard, J. H. Kim,
15. J. L. Lanciego, F. G. Wouterlood, J. Chem. Neuroanat. 42, Distribution of dopamine 1 receptor positive neurons in the
projection require precisely corresponding tracer 157–183 (2011). adult male mouse brain, data set, EBRAINS (2020);
injections to detect subtle changes occurring in 16. J. L. Lanciego, F. G. Wouterlood, Brain Struct. Funct. 225, https://doi.org/10.25493/AVRZ-4JB.
transgenic animals (Fig. 4, E and F). Similarly, 1193–1224 (2020). 50. T. B. Leergaard, J. G. Bjaalie, Front. Neurosci. 1, 211–223 (2007).
17. J. W. Bohland et al., PLOS Comput. Biol. 5, e1000334 (2009). 51. C. Tocco et al., Impact of area patterning genes on corticopontine
attempts to compare the individual axonal mor- projection topography: Point coordinate data representing spatial
18. J. A. Harris et al., Nature 575, 195–202 (2019).
phologies in the pontine nuclei (using data from 19. T. Ragan et al., Nat. Methods 9, 255–258 (2012). distribution of corticopontine projections in Nr2f1 conditional
the MouseLight project at Janelia Research Cam- 20. M. N. Economo, J. Winnubst, E. Bas, T. A. Ferreira, knockout mice and controls (v1), data set, EBRAINS (2022);
J. Chandrashekar, J. Comp. Neurol. 527, 2190–2199 (2019). https://doi.org/10.25493/AJVE-B8G.
pus) to the terminal fields visualized in tract-
21. J. Winnubst et al., Cell 179, 268–281.e13 (2019). 52. R. Bakker, P. Tiesinga, R. Kötter, Neuroinformatics 13, 353–366
tracing experiments (using data from the Allen 22. M. Bota, S. Talpalaru, H. Hintiryan, H. W. Dong, L. W. Swanson, (2015).
Mouse Brain Connectivity Atlas) reveal that data J. Comp. Neurol. 522, 3160–3176 (2014).
23. N. M. van Strien, N. L. Cappaert, M. P. Witter, Nat. Rev. AC KNOWLED GME NTS
mapped to the same atlas allow interesting ob-
Neurosci. 10, 272–282 (2009). We thank the current and past members of the Neural Systems
servations (Fig. 4, G and H), but these attempts 24. I. M. Zakiewicz, Y. C. van Dongen, T. B. Leergaard, J. G. Bjaalie, Laboratory for their contributions to the development, application,
also indicate that more data are needed for com- PLOS ONE 6, e22669 (2011). and validation of new methods to experimental research in small
plete mapping of neural networks. Attempts to 25. I. E. Bjerke et al., Front. Neuroanat. 12, 82 (2018). animal models and colleagues in the EBRAINS research
26. G. Paxinos, C. Watson, The Rat Brain in Stereotaxic Coordinates infrastructure for their contributions to the delivery of services.
find, visualize, and compare such data are also (Academic Press, 1982). Funding: This work was in part supported by the European Union’s
hampered by technical challenges related to lack 27. L. W. Swanson, Brain Maps: Structure of the Rat Brain (Elsevier, 1992). Horizon 2020 Framework Program for Research and Innovation
of tool interoperability. In the rat, large data col- 28. K. B. J. Franklin, G. Paxinos, The Mouse Brain in Stereotaxic under the specific grant agreement no. 945539 (Human Brain
Coordinates (Academic Press, 1996). Project SGA3) and the Research Council of Norway under grant
lections on neural connections are not available, 29. L. J. Kjonigsen, S. Lillehaug, J. G. Bjaalie, M. P. Witter, T. B. Leergaard, agreement no. 269774 (INCF Norwegian Node). Competing
and, so far, few attempts have been made to sys- Neuroimage 108, 441–449 (2015). interests: J.G.B. is a member of the Management Board of the
tematically map brainwide connections (24). For 30. K. K. Osen, J. Imad, A. E. Wennberg, E. A. Papp, T. B. Leergaard, EBRAINS AISBL, Brussels, Belgium. The authors declare no other
Neuroimage 199, 38–56 (2019). competing interests. License information: Copyright © 2022 the
these reasons, adding more data and tools will
31. I. E. Bjerke, U. Schlegel, M. Puchades, J. G. Bjaalie, T. B. Leergaard, authors, some rights reserved; exclusive licensee American
be critical for attaining an increasingly complete Paxinos & Watson’s “The Rat Brain in Stereotaxic Coordinates” Association for the Advancement of Science. No claim to original
overview of the organization of brain connections (7th edition) spatially registered to the Waxholm Space atlas of the US government works. https://www.science.org/about/science-
and other features of rodent brain architecture. rat brain, data set, Human Brain Project Neuroinformatics Platform licenses-journal-article-reuse
(2019); https://doi.org/10.25493/APWV-37H.
The atlases play a key role in this endeavor. 32. I. E. Bjerke, U. Schlegel, M. Puchades, J. G. Bjaalie, T. B. Leergaard, Submitted 21 June 2022; accepted 18 August 2022
They provide a standardized representation of Swanson’s “Brain Maps: Structure of the Rat Brain” (4th edition) 10.1126/science.abq2594

REVIEW ofMRI technologies. It has been shown that lo-
cation, polarity, and temporal shape of neural
Solving brain circuit function and dysfunction with activity can be accurately inferred from the
ofMRI signal (3, 13, 14), and that neural ac-
computational modeling and optogenetic fMRI tivity can be measured by ofMRI across mul-
tiple synapses (11, 15). Stimulation cell type,
Jin Hyung Lee1,2,3,4*, Qin Liu1†, Ehsan Dadgar-Kiani1,2 location, frequency (14, 15), and intensity (11)
were shown to substantially change the loca-
Can we construct a model of brain function that enables an understanding of whole-brain circuit tion and shape of activities throughout the
mechanisms underlying neurological disease and use it to predict the outcome of therapeutic brain. It was also made clear that whole-brain
interventions? How are pathologies in neurological disease, some of which are observed to have spatial neural dynamics as measured by ofMRI can
spreading mechanisms, associated with circuits and brain function? In this review, we discuss approaches accurately predict distinct behaviors (2, 14, 15).
that have been used to date and future directions that can be explored to answer these questions. Many studies have used ofMRI to improve
By combining optogenetic functional magnetic resonance imaging (fMRI) with computational modeling, our understanding of fundamental circuitries
cell type–specific, large-scale brain circuit function and dysfunction are beginning to be quantitatively associated with behavior, memory, and cog-
parameterized. We envision that these developments will pave the path for future therapeutics nition. For instance, Liu et al., Weitz et al., and
developments based on a systems engineering approach aimed at directly restoring brain function. Leong et al. identified that frequency-dependent
thalamic activities drive distinct whole-brain
function in circuits associated with arousal, atten-
C
an we systematically design treatments can be combined to enable cell type–specific, tion, and somatosensory function (14, 16, 17).
for brain disorders such as Parkinson’s large-scale modeling of brain function at the Weitz et al. and Takata et al. revealed distinct
disease? To enable this we need a full single-cell-spiking level. In addition, although dorsal and ventral hippocampal control of brain-
algorithmic description beyond simple restoring brain function is the ultimate goal wide function (15, 18), Jung et al. identified cir-
correlations of how specific cells or brain of neurological disease treatment, understand- cuits associated with cell type–specific targeting
regions cooperatively contribute to a given ing how prominent pathology relates to brain of the somatosensory cortex (19), and Choe et al.
behavior or symptom in the context of whole- function is also of critical importance. In this revealed cerebellar cortex functional control
brain networks. What does it take to generate review, we will discuss approaches used to date over forebrain and midbrain (20). Another study
such algorithms? Cell types cannot be ignored and approaches that may be employed in the revealed brain-wide dynamics that govern how
because even neurons in the same location in future to achieve such goals. the medial prefrontal cortex regulates reward-
the brain can drive completely opposite func- related behaviors through distant regions such as
tions and resulting behaviors (1, 2). Neurons Bridging scales with optogenetic fMRI the striatum (21). A recent study has tried fMRI
often interact with large networks across the ofMRI is a technology that combines opto- with cell type–specific activation of astrocytes (22).
whole brain. A limited field of view within the genetic stimulation with fMRI readouts. Op- These studies have shown that the observed hemo-
brain is thus insufficient to understand these togenetics (5, 6) enables cell type–specific, dynamic activities are closely tied to neuronal
algorithms. Therefore, to obtain the data nec- millisecond-scale, activity modulation using activities using either simultaneous or follow-up
essary to reconstruct these algorithms of be- light whereas high-field fMRI measures the re- electrophysiology. Furthermore, although most
havioral control we need an imaging system sulting hemodynamic responses in live subjects ofMRI studies have been conducted in rodents it
that can measure cell type–specific, whole- across the whole brain. In the initial proof- has also been applied to nonhuman primates, in
brain function. Optogenetic functional mag- of-concept study (3), motor cortex excitatory which both saccade latencies and whole-brain
netic resonance imaging (ofMRI) (3) has begun neurons triggered fMRI responses that could activity were found to be dependent on spe-
to achieve this goal. With ofMRI, we can select be measured throughout the brain with sub- cific neuronal targets in the motor cortex (23).
cell type–specific modulation targets while second temporal resolution. To accelerate The ability to probe and read out whole-brain
monitoring the outcome of such modulation scientific discovery with ofMRI, several tech- activity with ofMRI has also advanced our under-
across the whole brain, in vivo with high spa- nological innovations were made, including standing of dysfunctional circuitry associated
tiotemporal resolution. This has opened a win- development of real-time imaging with robust- with neurological disorders. In studies of epi-
dow into the study of brain function. We can ness to the live subject’s motion which achieves lepsy ofMRI provided an advantage as it could
see how modulating specific elements of the data acquisition, reconstruction, and motion be used to optogenetically induce seizures with
brain leads to specific behaviors of interest correction (7), as well as analysis of three- precise origins on demand while measuring the
while also directly observing the inner work- dimensional (3D) images with high accuracy in resulting whole-brain activities with simultaneous
ings of the brain that lead to such behaviors. approximately 12 ms. To resolve cortical layer electrophysiology recordings. This enabled studies
Through computational modeling of ofMRI and subnuclei–specific responses, novel com- that can generate models to predict and classify
signals measured across the whole brain (4), pressed sensing (8–10) and machine learning– seizures using its early activity markers (15, 24).
cell type–specific, large-scale brain function based fMRI technology was developed, which Furthermore, longitudinal effects of seizures on
can be quantitatively described at the regional achieved substantial reduction in voxel volume. global brain function could be measured to un-
level. Once regional interaction maps are recon- MR-compatible optrodes and electrodes were derstand how the disease progresses and how
structed we envision that biophysical modeling also developed for simultaneous electrophysio- seizures are generated and maintained (25).
logical recordings to validate the neural basis of These advances aid our understanding of circuit
1
Department of Neurology and Neurological Sciences, the ofMRI hemodynamic signal (11, 12). These mechanisms underlying seizures and help
Stanford University, Stanford, CA 94305, USA. 2Department
can achieve simultaneous acquisition of electro- design intervention parameters such as stimula-
of Bioengineering, Stanford University, Stanford, CA 94305,
USA. 3Department of Neurosurgery, Stanford University, physiology recordings during fMRI sessions and tion location and frequency to effectively inhibit
Stanford, CA 94305, USA. 4Department of Electrical provide information with higher temporal reso- seizures. Optogenetic fMRI can also elucidate
Engineering, Stanford University, Stanford, CA 94305, USA. lution in regions of interest identified by ofMRI. mechanisms behind existing therapies, such as
*Corresponding author. Email: [email protected]
†Present address: Brain Simulation Section, Charité – The capabilities and precision of ofMRI have post stroke recovery. Activation as measured by
Universitätsmedizin Berlin, Berlin, Germany. been extensively tested using these advanced ofMRI was highly predictive of the degree of

A Selective stimulation of D1/D2 MSNs B Rotational Behavior C Optogenetic fMRI Setup

D1-MSN stim.
D1-MSN stim. D2-MSN stim.
Light delivery
Ipsiversive
15
Rotations per minute

10 **
5 n.s.
D2-MSN stim. 0
Contraversive
D1/D2 MSNs 5
10
*
Laser OFF
15 Laser ON
***
D
iMC Cg 2π iMC Cg 2π
cMC cMC
Orb Olf iOrb cOrb
Olf iAcc iCPu iCPu
iAcc cAcc cCPu
iSC cSC
Phase
Phase
Ins
Spt
Spt cCPu iGPe cGPe Thal iGPi cGPi iGPe cGPe Thal iGPi cGPi
iHp cHp Rsp SC iHp cHp Rsp Vis SC
Au Vis Au
Ent PRh cHp

STN Amy SNR DpME iHp
2 mm
STN SNR 2 mm
0 0
E Segmented regional time series F Validation of fMRI signal using electrophysiology Thalamus
Anterior Caudate Putamen Anterior Caudate Putamen
1.2 D1-MSN stim. D2-MSN stim.
0.6 *** ***
0 12
-0.6 16
% BOLD change
Firing rate (Hz)
Firing rate (Hz)

Thalamus Thalamus
1 12 8
20 15
0
-1 15 8
10
10
4
Motor Cortex Motor Cortex 4
0.8 5
5
0 0 0 0 0
Stim
Stim
Pre
Pre
Post
Post
-20 0 20 40 -20 0 20 40
-0.8
Time (sec) Time (sec) 20 s Time post stim. onset (sec)
Fig. 1. Optogenetic functional magnetic resonance imaging bridges scales. stimulations with fMRI readout. (D) ofMRI with selective stimulation of D1- and
(A) Optogenetics enable cell type–specific stimulations, such as the selective D2-MSN results in distinct brain-wide activities associated with distinct behavior
targeting of D1- or D2-MSNs in the striatum. (B) With selective stimulations (n = 12 animals). Group-wise phase maps, which were thresholded to only
of D1- or D2-MSNs in the striatum, mice show contraversive or ipsiversive rotations, active voxels within the brain, depict the heterogeneity in the temporal dynamics
respectively (D1-MSN stim: n = 12 animals, mean ± SEM, ***P < 0.001, two-tailed of the evoked responses. (E) The time series of any region can be extracted
paired t test; D2-MSN stim: n = 11 animals, mean ± SEM, *P < 0.05, **P < 0.005, from the 4D fMRI data. (F) Electrophysiology recordings mirror fMRI response in
two-tailed paired t test). (C) Optogenetic fMRI technology combines optogenetic polarity of neural activity change. This figure is based on Lee et al. (2).
recovery and was useful in helping identify D2-MSNs, Lee et al. performed whole-brain underlying neuronal activity. For example, the
sensory circuits involved in this process (26). fMRI during repeated 20-s periods of optoge- fMRI time series in the thalamus exhibited
We can now start to reveal detailed circuit netic D1- or D2-MSN stimulations (2) (Fig. 1C). robust and reliable increases and decreases upon
mechanisms that were challenging to under- Active voxels were identified as those signif- D1- and D2-MSN stimulations, respectively (Fig. 1,
stand before. As an example, we will review icantly synchronized to the repeated stim- D and E). Indeed, single units exhibited an in-
studies uncovering how D1- and D2-receptor– ulations (Fig. 1D). The local signal at the site of crease in firing rate during D1-MSN stimulations
expressing medium spiny neurons (MSNs) stimulation was positive for both inhibitory and a decrease in firing rate during D2-MSN
dynamically regulate global brain function D1- and D2-MSN stimulations (Fig. 1E), re- stimulations (Fig. 1F). These results demonstrate
and dysfunction (Fig. 1). The cortico-basal- solving a widely debated issue of whether in- that ofMRI can detect cell type–specific multi-
ganglia-thalamus circuit is implicated in many hibitory neuron activity evokes a positive or synaptic activity changes across the whole brain
important brain functions including motor negative fMRI signal. In most regions of the with high spatiotemporal precision and that
control and reward mechanisms. Neurolog- ipsilateral cortico-basal-ganglia-thalamus electrophysiological recordings alongside ofMRI
ical disorders such as Parkinson’s disease, network, the evoked response in a given re- can support these findings.
Huntington’s disease, addiction, and autism gion exhibited qualitatively different temporal ofMRI has caveats for future improvements
involve this network. The caudate putamen— profiles between D1- and D2-MSN stimula- as well as potential pitfalls to avoid. Channel-
which seats the D1- and D2-MSNs (Fig. 1A)— tions (Fig. 1, D and E). Given the diversity of rhodopsin (ChR2) is known to evoke synchro-
is the key node that separates the direct and BOLD responses evoked by D1- and D2-MSN nized neuronal activity upon light stimulation.
indirect pathways (Fig. 1B). To assess the brain- stimulations, using single-unit recordings we Therefore, before launching an ofMRI investi-
wide dynamics driven by inhibitory D1- and sought to verify whether the responses reflected gation the behavioral impact of the optogenetic

A Basal ganglia network nodes B Direct pathway Indirect pathway C Model connectivity
Excitatory CPu
CTX CTX
Inhibitory GPe
CTX
CPu (D1) GPi
Source
CPu (D2)
THL STN
CPu
STN GPe THL STN GPe SNr
SNr THL
THL
GPe GPi/SNr CTX
GPi/SNr
GPi STN
r
L
i
u
X
e
GP
SN
TH
CP
GP
CT
ST
Target
D Observed vs. model reconstructed signaling E Model connectivity estimates
Observed
Predicted
D1-MSN stim. D2-MSN stim. D1-MSN stim. D2-MSN stim.
Anterior Caudate Putamen Anterior Caudate Putamen
0.5
Connectivity Estimates
0.5
0 0
-0.5 -0.5
% BOLD change
Thalamus Thalamus
0.5 0.5
0 0 E
-0.5 -0.5
Motor Cortex Motor Cortex

0.5 0.5
0 0
-0.5 -0.5
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Time (sec)
F Motor cortex network (D1) G Motor cortex network (D2)

+ connectivity + connectivity
– connectivity – connectivity
CPu CPu
GPe GPe
CTX CTX
Source
GPi
Source
GPi
STN STN
CPu THL CPu THL
SNr SNr
SNr THL SNr THL
CTX CTX
GPe GPe STN
GPi STN
r
L
ST i
GP u
X
e
GP
SN
r
L
ST i
GPi
TH
GP u
X
e
CP
GP
SN
CT
TH
CP
CT
Target Target
Fig. 2. Computational modeling of ofMRI data reveals brain-wide functional (E) DCM used ofMRI data to estimate the causal influence (effective connectivity)
interaction dynamics. (A) The cortico-basal-ganglia-thalamus network involves among regions of interest during D1- and D2-MSN stimulations, respectively.
a large number of network nodes across the brain. (B) The anatomical connectivity (F and G) The graph and matrix representations of effective connectivity network
of direct and indirect pathways involve a large number of common anatomical for (F) D1-MSN, and (G) D2-MSN stimulations, respectively. Significant and
regions with distinct cell types in caudate putamen (CPu). (C) Anatomical close-to-significant represent parameters with P < 0.05 and 0.05 ≤ P < 0.10,
connections were used as an a priori generative network model. In addition to respectively (one-sample t test, multiple comparison correction across con-
the direct and indirect pathways shown in (B), other established anatomical nections with FDR P < 0.10). CPu, caudate putamen; GPe, external globus pallidus;
connection such as the hyper-direct pathway were also included. (D) DCM- GPi, internal globus pallidus; STN, subthalamic nucleus; SNr, substantia nigra;
generated fMRI time series closely match experimental ofMRI time series. THL, thalamus; CTX, cortex. This figure is based on Bernal-casas et al. (4).
stimulations should first be investigated as a amplify existing spontaneous activity, which cortical regions but its depth coverage is limited
means to ensure that the behavior generated is expands the application range of ofMRI to to superficial layers of cortex. Combining fMRI
of interest in either a normal physiological or more physiological conditions. There are many and widefield calcium imaging has been shown
pathological context. For example, in our D1- aspects of ofMRI that can be further im- to mitigate the limitations of both methods
and D2-MSN stimulation ofMRI experiments, proved. For example, ofMRI will benefit greatly (34). Therefore, future ofMRI studies could be
increased contralateral and ipsilateral rota- from higher spatio-temporal resolution, real- integrated with other technologies for com-
tions were observed, respectively (2) (Fig. 1B). time feedback–based stimulation control, and plementary strengths.
This shows that the two separate stimulations imaging during behaviors (7, 8).
result in opposite behaviors known to be asso- Other technologies have also been developed Cell type–specific modeling of large-scale
ciated with movement disorders. The fact that for the investigation of brain circuitries, includ- brain function
ChR2 evokes synchronization upon light stim- ing high-speed volumetric calcium imaging Large-scale neural network models (35–39)
ulation also makes it suitable for studying (30, 31), probes for high-density electrophys- using experimental data from positron emission
pathological oscillations in a number of dis- iology recordings (32, 33), and widefield cal- tomography (40), fMRI (41), and electroenceph-
ease models and contexts. Excessive beta-band cium and voltage imaging (34). Compared with alography (EEG)/magnetoencephalography
oscillations in Parkinson’s disease have been ofMRI, these advances offer higher spatiotem- (MEG) (42) have made considerable contribu-
extensively explored with ChR2-induced os- poral resolution although they are limited by tions to understanding of brain function. How-
cillations (27, 28). Some newer opsins such as recording depths and field of view. For exam- ever, although these modeling efforts are based
stabilized step function opsin (29) modulate ple, widefield “whole-cortex” calcium imaging on data from carefully designed experiments
target neurons by increasing the excitability to enables up to 30 Hz simultaneous recording of that attempt to isolate specific brain functions,

without the capability to untangle contributions dynamical systems (MDS) causal modeling was single-cell-level spiking control and modulation
from different cell types across the whole brain, used with ofMRI to estimate causal brain inter- of brain function and dysfunction (57–59).
models have been limited in their capabilities. actions (50). Similar to DCM, MDS models both In addition to having the ability to capture
The development of ofMRI technology opens intrinsically and experimentally induced causal and model the macroscale and mesoscale in-
an opportunity in terms of whole-brain compu- couplings in a large-scale brain network. teractions between separate regional popula-
tational modeling because it measures cell In our previous study (4) spectral DCM tions of neurons in the brain with ofMRI and
type–specific whole-brain dynamics. (36)—a variation of DCM that enables large- DCM, there is an added advantage in modeling
For example, the cortico-basal-ganglia- scale networks to be modeled with computa- the interactions between individual neurons
thalamus network features a large number of tional efficiency—was used to investigate the of varying cell types. Neurological disorders
network nodes (Fig. 2A) and distinct cell types interactions within the cortico-basal-ganglia- may differentially affect specific cell popula-
that have characteristic long-range projections. thalamus network with ofMRI data from D1- tions within one brain region. Optogenetically
According to the canonical functional model and D2-MSN optogenetic stimulations. One stimulating one type of cell population while
of basal ganglia (43, 44), movement initiation recent study reported consistent results using inhibiting another type in the external globus
and suppression are mediated by direct and DCM with D1- and D2-MSN stimulation ofMRI pallidus (GPe) prolonged therapeutic effects in
indirect pathways, respectively (Fig. 2B). The data (51). As illustrated in Fig. 2, combining a mouse model of Parkinson’s disease (50). It
direct pathway promotes movements through DCM or equivalent modeling schemes with is widely assumed and supported by optoge-
D1-MSN excitation in the striatum followed by ofMRI data can accurately reveal brain-wide netics studies that Parkinson’s disease causes
excitation in the thalamus and consequently regional interactions. The time series is also hyperactivity of D2-MSNs and hypoactivity of
the motor cortex. On the other hand, the in- accurately reproduced by DCM, closely match- D1-MSNs, thus impairing the balance between
direct pathway is assumed to transfer the ac- ing experimental ofMRI time series (Fig. 2D). the direct and indirect pathways (44, 45, 60).
tivation of D2-MSN in the striatum through Fig. 2, E to G show the DCM estimations of DYT1 dystonia—a genetic early onset dystonia—
complex interactions within the basal ganglia, between-region effective connectivity (4) using is related to cholinergic interneuron dysfunc-
resulting in inhibition of the thalamus and ofMRI data. DCM results verified the direct tion and altered D2 receptor function in the
consequently the motor cortex. However, de- pathway activation during D1-MSN stimula- striatum (61, 62). Firing pattern alterations of
spite the wide usage and success of the canon- tion and indirect pathway activation during one cell type may also contribute to large-scale
ical model in explaining experimental results, D2-MSN stimulation. The defining connec- changes. Optogenetic stimulation in striatal
recent experimental evidence argues against tions of the direct pathway model are statisti- cholinergic interneurons—a subpopulation
the model and finds the interactions between cally significant (CPu to SNr, GPi to thalamus, constituting less than 2% of the striatum—
the two pathways still puzzling. For exam- SNr to thalamus) or close to significant (CPu could generate broadband oscillations in the
ple, the direct and indirect pathways operate to GPi) with D1-MSN stimulations. In the motor network (27). With cell type–specific,
simultaneously during movement onsets (45). D2-MSN stimulation network, significant con- single-cell-spiking level modeling, it is easier
As demonstrated earlier with ofMRI, we can nections included those of indirect pathways to address the heterogeneity and rich micro-
decompose the operation of the cortico-basal- (CPu to GPe, GPe to STN, STN to GPi/SNr, and scopic interactions within one region with
ganglia-thalamus network by optogenetically GPi to thalamus). The existence of cortical feed- biophysical details. ofMRI-based DCM or other
activating D1- and D2-MSNs selectively and backs can also be observed (Fig. 2, F and G). By regional brain dynamics models can serve as
directly observing the corresponding brain- contrast, the effective connectivity estimates by a bridge between whole-brain dynamics and
wide dynamics (2) (Fig. 1). However, the degree DCM also suggest several positive projections single-cell-spiking level activity, enabling
of connectivity in the cortico-basal-ganglia- that are anatomically inhibitory and cannot construction of large-scale, cell type–specific
thalamus network is too high (Fig. 2, B and C) to be explained by the canonical direct/indirect biophysical models that can test neuronal-
identify regional causal influences based on the pathway model, such as the projections from level hypotheses.
time series alone (Fig. 2D). Through computa- GPi to thalamus during D1-MSN stimulations, Cell type–specific, single-cell-spiking level
tional modeling—e.g., dynamic causal modeling which matches several experimental reports of models that can accurately predict circuit func-
(DCM)—with the precision afforded by ofMRI GPi-thalamus paradoxical coactivation (2, 52–54). tion and dysfunction can be very powerful tools
data, cell type–specific causal linkages among Understanding the mechanism underlying such for designing or optimizing therapy. Thus far,
regions of interest (ROIs) in the cortico-basal- paradoxical connections requires further micro- many large-scale models without cell type spec-
ganglia-thalamus network could be identified. scopic investigations into the specific synaptic ificity have been constructed to test various
DCM is a modeling scheme that estimates interactions with techniques such as single-cell- hypotheses underlying deep-brain stimulation
the causal coupling (effective connectivity) spiking level modeling and single-unit record- (DBS), an existing therapy for Parkinson’s dis-
in a multiregion network based on neuroim- ings, as we will discuss next. ease (63–65). However, because each cell type
aging data (fMRI, MEG/EEG) (46–48). The es- in the basal ganglia has distinct synaptic and
timations are fitted to empirical results using Cell type–specific, single-cell-spiking level physiological properties, accurate models would
Bayesian techniques. One major strength of modeling of brain-wide function need cell type–specific parameters that can fit
DCM is that the estimated regional connec- One key strength of computational modeling is cell type–specifically acquired experimental
tivities are directional, which is especially that it can bridge spatial scales, from whole-brain data. We envision that ofMRI, DCM, and elec-
valuable for networks with many reciprocal dynamics to single-cell activity and explain data trophysiological recordings could be combined
connections and feedbacks similar to the from different experimental modalities, from to build large-scale biophysical models that
cortico-basal-ganglia-thalamus network. ofMRI fMRI to extracellular recordings (55, 56). DCM can model brain-wide activity at the single-
can also be combined with other modeling and other macroscale and mesoscale brain cell-spiking level. Figure 3A illustrates this
schemes. For example, Salvan et al. (49) opto- models commonly use neural mass models or vision. With ofMRI and electrophysiology ex-
genetically modulated the entorhinal cortex meanfield models as the basic unit describing periments providing multiscale data and DCM
and combined hidden Markov modeling with the collective neural activity in a brain region or providing brain-wide interaction information,
ofMRI data to study how entorhinal cortex cortical column (40–42). By contrast, single-cell- single-cell level biophysical models, with accu-
drives frequency-dependent brain-wide dy- spiking models computationally depict the mi- rate cell type–specific, large-scale context, can
namic states. In another study, multivariate croscopic biophysical features that allow for be built and validated (Fig. 3B). As illustrated

A B Single-neuron biophysical model
CTX Model validation
CTX Simulated spiking

TH
CPu TH CPu
Stimulation THL
SNr
GPi STN Stimulation Observed spiking
GPe GPi STN
CPu GPe
GPi STN SNr
Single-unit recordings
stim. Single Neuron
C Comparison of measurements with model spiking

D1-MSN Stim CPu GPi Thalamus STN D
Experiment
Experimental
D1-MSN Stim
10 10 10 10
Simulation
12
5 5 5 5
10
Firing rate (Hz)

0 0 0 0
8
6
Simulated
10 10 10 10 4
5 5 5 5
2
0 0 0 0
0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60
0
Time (sec) CPu GPi THL STN
D2-MSN Stim CPu GPi Thalamus STN
D2-MSN Stim
Experimental
16
10 10 10 10 14
5 5 5 5 12
Firing rate (Hz)

0 0 0 0 10
8
6
Simulated
10 10 10 10 4
5 5 5 5 2
0 0 0 0 0
0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60
CPu GPi THL STN
Time (sec)
Fig. 3. Brain circuit function modeling at the single-cell-spiking level can be by directly comparing simulated spiking trains with experimental data. (C) Models
made possible through a multiscale approach. (A) Locations of optogenetic allow for direct comparison of experimental single-unit recording data and
stimulation and in vivo extracellular recordings for the cortico-basal-ganglia- simulated data. (D) Models should be designed so that the spike rates of all ROIs
thalamus network study are schematically illustrated. (B) Single-cell-spiking level simulated by the single-cell spiking level biophysical model statistically match
modeling with ofMRI and single-unit recording data are exemplified. A large-scale experimental data. ofMRI combined with biophysics modeling can enable successful
model built with ofMRI data is expanded to a single-neuron level biophysical reproduction of the single-cell-spiking level dynamics induced by cell type–specific
model. Each ROI consists of many simulated single neurons. The model is validated optogenetic stimulations such as D1- and D2-MSN stimulations.
single-cell level biophysical models, with accu- envision that in the future, ofMRI-enabled brain- ing mechanisms of these proteins remain un-
rate cell type–specific, large-scale context, can wide models with single-cell-spiking level pre- known, several recent studies have shown that
be built and validated (Fig. 3B). As illustrated cision will ultimately enable systematic design whole-brain spreading patterns are highly
in Fig. 3, C and D, a single-cell-spiking level of neuromodulation therapy with predictable dependent on the inoculation site (68), and
model can be designed to reproduce experi- outcomes. that Anatomical connectivity is highly predic-
mental neuronal activity and group dynamics tive of susceptible regions after induced seed-
with high precision. With such models, we can Modeling whole-brain pathology dynamics and ing of a-syn pathology (69).
imagine simulating brain-wide spiking activ- their relationship with brain function To understand the linkage between anatom-
ity from “virtual neuromodulations” without Brain circuitry is relevant to neurological dis- ical connections, pathology, and function, it
needing to do in vivo experiments. We would orders beyond use in modeling local and global is important to reliably measure whole-brain
then be able to use these results to better op- brain function. It is also important for under- pathological states. The past decade has seen
timize therapeutic targets and parameters standing the underlying pathology of many several prominent advancements in whole-brain
in which simulated spiking matches our de- disorders. In the case of Parkinson’s disease tissue clearing technology. These range from
sired response. For example, a recent study (66) and other synucleinopathies, the seeding and hydrogel-based techniques such as CLARITY
employed optogenetics and single-cell level gradual accumulation of pathological alpha- (70)—which enables multiplexed brain imag-
computational modeling within three differ- synuclein (a-syn) causes dopaminergic neuro- ing but potentially requires long incubation
ent cell types of the motor cortex to show that degeneration, which ultimately leads to striatal times—to solvent-based techniques such as
DBS on cortical somatostatin interneurons imbalance and the cardinal Parkinsonian signs 3DISCO (71) that are generally faster but may
can ameliorate Parkinsonian symptoms. We (67). Although the specific biochemical traffick- bleach fluorescent signals. Combining these

tissue-clearing methods with whole-brain im- gates automatically segmented and registered circuitry as a directed graph with neuroana-
munolabeling and light-sheet microscopy to a standardized atlas. As depicted in Fig. 4B, tomical regions as nodes and axonal pathways
allows for high-resolution examination of whole- longitudinal analysis of pathology can demas edges between these nodes (Fig. 4C). These
brain pathology using biochemical reporters onstrate a highly dynamic, brain-wide process models have found wide applicability in pre-
(72, 73). As these 3D whole-brain histological of spreading after seeding of a-syn, consistent dicting the spread of a-syn pathology in both
datasets now provide micron resolution and with numerous histological studies (67, 69). rodent and human imaging studies (69, 78),
are exceeding terabytes in size, it is necessary Registration to a standardized atlas can pro- as well as in human dementia (79) and supra-
to have automated registration and segmen- vide an enormous advantage enabling sys- nuclear palsy neuroimaging studies (80). In
tation techniques (74) to capture the rich in- tematic statistical comparisons and analysis Fig. 4C, we present an example of how a whole-
formation provided by the data. Figure 4A of disease pathology alongside connectomic brain model can accurately reconstruct the
depicts how a Parkinsonian disease model in (75), genetic (76), and vascular (77) databases, regional variability in pathology. Compared
which the injection of a-synuclein PFFs are which all reside in the same reference coordi- with previous models dependent upon serial
used to trigger whole-brain pathology can nate space. Most notably, the Allen Connectivity histological sectioning (69) or in vivo human
be systematically analyzed using a computa- Atlas (75) has been crucial in the development imaging data (78), this type of modeling can
tional pipeline. Brains can be immunolabeled of models describing the spread of pathology provide a more comprehensive, higher resolu-
and cleared at each time point using the through the connectome (69). For instance, tion description of pathology dynamics that
iDISCO method (72), with the imaged aggre- network diffusion models represent whole-brain includes all brain regions, which is important
Fig. 4. To understand pathology A Segmentation Atlas registration Whole brain pathology

function interaction, whole-brain
Isocortex
pathology dynamics can be modeled
Raw
alongside ofMRI. (A) a-synuclein
Brains immunolabeled and
PFFs injected into seed locations
TH HY MB HB CB
100 µm
cleared at each time point
induce pathology at various time points
Segmented
post-injection, which can then be 2W 2M 4M 6M 8M 12M 18M
captured by iDISCO tissue clearing
and light-sheet fluorescent microscopy 1 mm
(LSFM). Machine-learning–based,
automatic segmentation and B α-syn aggregation at months post-injection (MPI)
registration techniques can 0.5 MPI 2.0 MPI 4.0 MPI 6.0 MPI 8.0 MPI
Aggregate count
streamline the quantification of each 1 mm 150
pathological marker within the 100

Allen Reference Atlas (ARA). (B) Com-
50
parisons of averaged heatmaps
0
across cohorts can depict whole-
brain pathology changes over many
months after injection. (C) Modeling
C Allen reference 1 06
D Connectome reweighted by E Colocalization of
atlas gene transcription levels treatment & activity
of longitudinal data based on whole-brain
Predicted count
∆ a-syn aggregation
1 04
Decrease Increase
anatomical connectivity can capture
# 25-um voxels
10000
regional differences in pathology. 1 02

6000
(D) Reweighting the connectivity 2000
1 00
matrix can allow for encoding of genetic
Negative Positive
contribution within a model. (E) Whole- 1 00 1 02 1 04 1 06
Actual count Optogenetic fMRI activity
brain colocalization analysis between
optogenetic-stimulation induced F
alpha-synuclein pathology change α-syn aggregation (treated vs. untreated) ofMRI with stimulation at treatment location
4.00 11
and optogenetic fMRI activity can 1 mm 1 mm
reveal the relationship between 3.75 10
pathology and function. In this

example, positive activity is 3.50
9
colocalized with decreases in

8
pathology, whereas negative 3.25
activity is highly colocalized with
z-score
7
increases in pathology. (F) Montages 3.00
of the modulated alpha-synuclein 6
pathology and ofMRI brain activity 2.75
maps show a high degree of 5
colocalization with opposite polarity.

2.50
4
2.25 3
2.00 2
p < 0.05 (corrected) Increase Decrease Negative Positive

for quantitative model descriptions. Given that could be tailored to patients on the basis of 37. X. J. Wang, Physiol. Rev. 90, 1195–1268 (2010).
transgenic mice such as those with modified their current pathological state and the ex- 38. G. Deco, M. L. Kringelbach, Neuron 84, 892–905 (2014).
39. A. Fornito, A. Zalesky, M. Breakspear, Nat. Rev. Neurosci. 16,
LRRK2 (69, 81) or SNCA (67) expression dem- pected impact of a given neuromodulation 159–172 (2015).
onstrate significantly altered spreading pat- parameter on pathology. Taking it one step 40. M. A. Tagamets, B. Horwitz, Cereb. Cortex 8, 310–320 (1998).
terns, models can additionally incorporate the further, therapy could even be designed while 41. K. E. Stephan et al., Neuroimage 42, 649–662 (2008).
42. S. J. Kiebel, O. David, K. J. Friston, Neuroimage 30, 1273–1284
brain’s inherent genetic or cell type–specific taking the expected future states of pathol- (2006).
differences. As depicted in Fig. 4D, reweighting ogy into consideration. Altogether, the abil- 43. R. L. Albin, A. B. Young, J. B. Penney, Trends Neurosci. 12,
the connectivity matrix by regional expression ity to readout and model pathology with brain 366–375 (1989).
44. M. R. DeLong, Trends Neurosci. 13, 281–285 (1990).
of a gene of interest can allow for evaluation clearing while measuring whole-brain net- 45. J. G. Parker et al., Nature 557, 177–182 (2018).
of that gene’s relevance in disease spread. work function with optogenetic fMRI will allow 46. K. J. Friston, L. Harrison, W. Penny, Neuroimage 19, 1273–1302
Henderson et al. weighted a network diffusion for the development of circuit-based models (2003).
47. K. J. Friston, J. Kahan, B. Biswal, A. Razi, Neuroimage 94,
model with SNCA expression, demonstrating that bridge pathology and function.
396–407 (2014).
that in silico simulation of circuits and genes 48. K. E. Stephan et al., Neuroimage 49, 3099–3109 (2010).
can recover observed a-synuclein pathology (69). Conclusion 49. P. Salvan et al., Cell Rep. 37, 109954 (2021).
Accurate whole-brain models of pathology These recent advancements in cell type–specific 50. S. Ryali et al., Neuroimage 132, 398–405 (2016).
51. C. Grimm et al., Cell Rep. 37, 110161 (2021).
dynamics that can predict both future and past neuromodulation, whole-brain functional im- 52. J. Kim et al., Neuron 95, 1181–1196.e8 (2017).
states of these progressive disorders will have aging, and computational modeling are start- 53. B. C. Schwab et al., PLOS Biol. 18, e3000829 (2020).
important clinical implications. For example, ing to pave a path for a major turning point in 54. J. H. Goldberg, M. A. Farries, M. S. Fee, Trends Neurosci. 36,
695–705 (2013).
a computational model that takes an arbitrary neuroscience. We aim to establish new ap- 55. G. Deco, V. K. Jirsa, P. A. Robinson, M. Breakspear, K. Friston,
pathological state and predicts future states proaches to simulating brain function that PLOS Comput. Biol. 4, e1000092 (2008).
could guide interventions dependent upon the can replicate and predict behaviors of inter- 56. M. Schirner, A. R. McIntosh, V. Jirsa, G. Deco, P. Ritter, eLife 7,
e28927 (2018).
predicted localization of pathology. Similarly, est. This will transform the treatment of a
57. J. J. J. Hjorth et al., Proc. Natl. Acad. Sci. U.S.A. 117,
a model that can iterate backward and com- wide array of neurological diseases including 9554–9565 (2020).
pute previous states, even back to the initial Parkinson’s disease. 58. X. Zhang, S. Santaniello, Proc. Natl. Acad. Sci. U.S.A. 116,
seeding site, can further aid in disease pro- 13592–13601 (2019).
RE FERENCES AND NOTES 59. M. J. Bezaire, I. Raikov, K. Burk, D. Vyas, I. Soltesz, eLife 5,
gression classification for diagnosis purposes.
1. P. Calabresi, B. Picconi, A. Tozzi, V. Ghiglieri, M. Di Filippo, e18566 (2016).
Because neurodegenerative disorders such 60. A. V. Kravitz et al., Nature 466, 622–626 (2010).
Nat. Neurosci. 17, 1022–1030 (2014).
as Parkinson’s disease, Lewy body dementia, 2. H. J. Lee et al., Neuron 91, 412–424 (2016). 61. K. L. Eskow Jaunarajs, P. Bonsi, M. F. Chesselet, D. G. Standaert,
and multiple system atrophy have long been 3. J. H. Lee et al., Nature 465, 788–792 (2010). A. Pisani, Prog. Neurobiol. 127-128, 91–107 (2015).
4. D. Bernal-Casas, H. J. Lee, A. J. Weitz, J. H. Lee, Neuron 93, 62. G. Sciamanna et al., Cell Rep. 31, 107644 (2020).
hypothesized to collectively form a neuropa- 63. J. M. Meier et al., Exp. Neurol. 354, 114111 (2022).
522–532.e5 (2017).
thological spectrum (82), modeling the var- 5. F. Zhang, A. M. Aravanis, A. Adamantidis, L. de Lecea, 64. W. J. Neumann et al., Brain 141, 2655–2669 (2018).
ious circuits that drive different pathologies K. Deisseroth, Nat. Rev. Neurosci. 8, 577–581 (2007). 65. E. M. Adam, E. N. Brown, N. Kopell, M. M. McCarthy, Proc. Natl.
6. A. M. Aravanis et al., J. Neural Eng. 4, S143–S156 (2007). Acad. Sci. U.S.A. 119, e2120808119 (2022).
and clinical manifestations will play a large 66. S. Valverde et al., Nat. Commun. 11, 2388 (2020).
7. Z. Fang, J. H. Lee, J. Neurosci. Methods 218, 184–195
part in precisely defining these disorders. (2013). 67. K. C. Luk et al., Science 338, 949–953 (2012).
Although functional and histological read- 8. Z. Fang, N. Van Le, M. Choy, J. H. Lee, Magn. Reson. Med. 76, 68. Y. J. Guo, H. Xiong, K. Chen, J. J. Zou, P. Lei, Mol. Psychiatry
440–455 (2016). 27, 758–770 (2022).
outs provide distinct information regarding 69. M. X. Henderson et al., Nat. Neurosci. 22, 1248–1257 (2019).
9. D. J. Holland et al., Magn. Reson. Med. 70, 1634–1643
the brain and disease, their colocalization is (2013). 70. K. Chung, K. Deisseroth, Nat. Methods 10, 508–513 (2013).
likely to be important in both furthering our 10. X. Zong, J. Lee, A. John Poplawsky, S. G. Kim, J. C. Ye, 71. A. Ertürk et al., Nat. Protoc. 7, 1983–1995 (2012).
Neuroimage 92, 312–321 (2014). 72. N. Renier et al., Cell 159, 896–910 (2014).
understanding of neural circuitry and devel- 73. T. Liebmann et al., Cell Rep. 16, 1138–1152 (2016).
11. B. A. Duffy, M. Choy, M. R. Chuapoco, M. Madsen, J. H. Lee,
oping novel therapeutics. For instance, com- Neuroimage 123, 173–184 (2015). 74. N. Renier et al., Cell 165, 1789–1802 (2016).
bining ofMRI and whole-brain pathological 12. M. R. Chuapoco et al., Neuroimage 186, 806–816 (2019). 75. S. W. Oh et al., Nature 508, 207–214 (2014).
76. E. S. Lein et al., Nature 445, 168–176 (2007).
labeling could aid in the development of ther- 13. B. Byers et al., Neuroimage 114, 328–337 (2015).
14. J. Liu et al., eLife 4, e09215 (2015). 77. C. Kirst et al., Cell 180, 780–795.e25 (2020).
apeutics that treat the underlying pathology, 78. S. Pandya et al., Neuroimage 192, 178–194 (2019).
15. A. J. Weitz et al., Neuroimage 107, 229–241 (2015).
such as a-syn pathology in Parkinson’s disease. 16. A. J. Weitz, H. J. Lee, M. Choy, J. H. Lee, Neuron 104, 79. A. Raj, A. Kuceyeski, M. Weiner, Neuron 73, 1204–1215 (2012).
We give an example of spatial colocalization be- 1153–1167.e4 (2019). 80. S. Pandya, C. Mezias, A. Raj, Front. Neurol. 8, 692 (2017).
17. A. T. Leong et al., Proc. Natl. Acad. Sci. U.S.A. 113, E8306–E8315 81. G. Bieri et al., Acta Neuropathol. 137, 961–980 (2019).
tween optogenetic stimulation–induced changes 82. J. W. Langston, Ann. Neurol. 59, 591–596 (2006).
(2016).
and ofMRI-measured functional changes ob- 18. N. Takata et al., PLOS ONE 10, e0121417 (2015).
served during delivery of the same stimulation 19. W. B. Jung, H. Jiang, S. Lee, S. G. Kim, Proc. Natl. Acad. Sci. AC KNOWLED GME NTS
(Fig. 4, E and F). This optogenetic stimulation U.S.A. 119, e2113313119 (2022). Funding: J.H.L., Q.L., and E.D.-K. were supported by NIH/NIMH
20. K. Y. Choe, C. F. Sanchez, N. G. Harris, T. S. Otis, P. J. Mathews, RF1 MH114227, NIH/NINDS DP1 NS116783, and NIH S10 OD025176.
paradigm reduces the a-synuclein aggregate Neuroimage 173, 370–383 (2018). J.H.L. and E.D.-K. were also supported by NIH/NIA R01 AG064051 and
count at the site of stimulation and in various 21. E. A. Ferenczi et al., Science 351, aac9698 (2016). NIH/NIBIB R01 EB030884. Author contributions: J.H.L. planned and
downstream regions. When colocalized with 22. N. Takata et al., Glia 66, 2013–2023 (2018). wrote the paper and made the figures, and also mentored and supported
23. A. Gerits et al., Curr. Biol. 22, 1722–1726 (2012). all personnel involved in the projects presented and manuscript writing.
brain activity, their similar spatial locations but 24. B. A. Duffy, M. Choy, J. H. Lee, Cell Rep. 30, 2540–2554.e4 Q.L. and E.D.-K. made the figures and wrote sections of the paper.
opposing polarities are apparent, revealing that (2020). Competing interests: J.H.L. is a founder, consultant, and board
positive activity is colocalized with decreases 25. M. Choy et al., Neuron 110, 221–236.e4 (2022). member of LVIS. Q.L. and J.H.L are inventors on a pending patent
26. S. Vahdat et al., Sci. Adv. 7, eabd9465 (2021). application, titled “Joint Dynamic Causal Modeling and Biophysics
in pathology, whereas negative activity is highly 27. B. Crompe et al., Nat. Commun. 11, 1570 (2020). Modeling to Enable Multi-Scale Brain Network Function Modeling.”
colocalized with increases in pathology. 28. K. Kondabolu et al., Proc. Natl. Acad. Sci. U.S.A. 113, Data and materials availability: All data and materials necessary to
This ability to target pathology with neuro- E3159–E3168 (2016). understand and assess these conclusions are available in the manuscript.
29. O. Yizhar et al., Nature 477, 171–178 (2011). License information: Copyright © 2022 the authors, some rights
modulation, while predicting subsequent brain- reserved; exclusive licensee American Association for the Advancement
30. J. Demas et al., Nat. Methods 18, 1103–1111 (2021).
wide changes in pathology could provide an 31. V. Voleti et al., Nat. Methods 16, 1054–1062 (2019). of Science. No claim to original US government works. https://www.
alternative way of thinking about neuromod- 32. N. A. Steinmetz et al., Science 372, eabf4588 (2021). sciencemag.org/about/science-licenses-journal-article-reuse
ulation therapy for Parkinson’s disease and 33. R. Hultman et al., Cell 173, 166–180.e14 (2018).
34. E. M. R. Lake et al., Nat. Methods 17, 1262–1271 (2020).
related disorders. For instance, parameters 35. O. Sporns, Networks of the brain. (The MIT Press, 2016). Submitted 28 May 2022; accepted 19 August 2022
such as duration and site of neuromodulation 36. A. R. McIntosh, Neural Netw. 13, 861–870 (2000). 10.1126/science.abq3868

REVIEW their relationship to function and dysfunction,

an early approach that viewed the human brain
Scale matters: The nested human connectome as a multiscale system with a nested design.
Although this work of the Vogts has largely
Markus Axer1,2* and Katrin Amunts1,3 fallen into oblivion, their view of cortical areas
forming nested groups and hierarchies has
A comprehensive description of how neurons and entire brain regions are interconnected is fundamental received ample support from recent studies
for a mechanistic understanding of brain function and dysfunction. Neuroimaging has shaped the of connectivity (7), cyto expression, receptor
way to approaching the human brain’s connectivity on the basis of diffusion magnetic resonance imaging expression, and gene expression (8). Precisely
and tractography. At the same time, polarization, fluorescence, and electron microscopy became what these hierarchies look like, how they are
available, which pushed spatial resolution and sensitivity to the axonal or even to the synaptic level. defined, and how they map to each other are
New methods are mandatory to inform and constrain whole-brain tractography by regional, high- ongoing topics of research (9).
resolution connectivity data and local fiber geometry. Machine learning and simulation can provide Fibers may form polysynaptic pathways, col-
predictions where experimental data are missing. Future interoperable atlases require new concepts, laterals, and feedback connections (5). Also
including high-resolution templates and directionality, to represent variants of tractography targets of intensive research are the inter-
solutions and estimates of their accuracy. play of electrical and molecular-biochemical
mechanisms of signal transduction at synapses;
synaptic plasticity; lifelong reorganization;
C
ognitive abilities and behavior are close- Axonal architecture and nerve fibers the relevance of fiber architecture for support-
ly related to the connectome, which is Axons are branches of neurons that transmit ing a concrete cognitive function; the specific
described as “a comprehensive descrip- signals from one neuron to the next, sometimes and dynamic consequences of variations in
tion of how neurons and brain regions over many centimeters (5). They are only a few brain organization, including cytoarchitecture,
are interconnected. It is the indispens- micrometers thick, with a length-to-caliber myeloarchitecture, and chemoarchitecture; and
able foundation for understanding how brain ratio in the range of 100,000:1, which makes interregional connectivity (10).
dynamics and function emerge from their it challenging to trace a single axon over its en- Two major lines of empirical research can
underlying structural (neural) substrate” (1). tire extent. Many axons have a myelin sheath (a be used to study structural connectivity in the
The human brain has ~86 billion neurons, stack of lipid bilayers), and a myelinated axon human brain: diffusion MRI-based neuroim-
each with up to 10,000 synapses. Neurons form is what we call a fiber. Dendrites represent the aging (dMRI) and methods targeting connectiv-
hundreds of cortical areas and subcortical other type of branches and integrate signals ity at higher spatial resolution in postmortem
nuclei, which are connected by nerve fibers. from other neurons. The spatially organized brains.
Signal propagation along these fibers is an neuronal cell bodies (6) and the neuropil (includ-
electrochemical process that includes a vari- ing, e.g., dendrites and glial cell processes) make In vivo diffusion MRI
ety of neurotransmitters, as well as postsyn- up the gray matter of the brain, which includes dMRI allows us to approach the connectome
aptic excitatory or inhibitory potentials, and the cerebral cortex and subcortical nuclei. in both living subjects and in postmortem brains.
is supported by the activity of glial cells. Con- The wiring of the cerebral cortex encompasses In combination with tractography, functional
sequently, a comprehensive understanding of both short-range and long-range axonal con- MRI (fMRI), modeling, and (graph) theoretical
the connectome encompasses the molecular nections between neurons (Fig. 1). Short-range approaches, it enables us to reveal fiber tracts,
and the cellular level up to the macro level—that connections form local circuits (such as the quantify network characteristics, and predict
is, it requires addressing a multiscale system. U-shaped fibers connecting neighboring gyri), function.
To approach the entire connectome was un- whereas long-range connections link different The Human Connectome Project (HCP) en-
til now only possible in the smaller brains of brain regions in the same hemisphere (long abled substantial progress in the area of in vivo
invertebrates such as Caenorhabditis elegans association fibers) or between them (commis- neuroimaging for measures of structural and
(2) and rodents such as the mouse (3). Map- sural fibers). Ascending and descending con- functional brain connectivity (11). dMRI was
ping the mouse brain at the synaptic level now nections link the cortex to subcortical nuclei established as the tool of choice to study struc-
seems to be in reach (4). For the human brain, and the spinal cord (projection fibers). Long- tural connectivity at the macroscale, pushed
a comparable resolution can be achieved in range axonal connections cluster into large, by new scanner designs and innovative white
small tissue blocks and slabs (e.g., using op- dense bundles of fibers making up the brain’s matter reconstruction methods. The HCP started
tical methods with clearing) but not yet whole- white matter (stained dark in Fig. 1A). with a focus on healthy young adults, explored
brain wide. Fibers within the cortex are less densely relationships with behavior and lifestyle, and
Here, we focus on fiber architecture because packed and show a high degree of structured- freely shared the imaging data, protocols, and
of progress made in human neuroimaging, ness in what is referred to as the myeloarchi- software tools with the scientific community.
in particular anatomical and diffusion mag- tecture. Myeloarchitectonic research was driven This has become a blueprint for other large-
netic resonance imaging (MRI), which illustrates by Oskar and Cécile Vogt, who systematically scale cohort projects (Lifespan HCP, ABCD, UK
approaches to bridging the gap to higher- studied differences between cortical regions, re- Biobank, etc.) (11). These projects collect, in a
resolution optical methods, and discuss the sulting in a map of >150 areas (6). This map, systematic manner, a large amount of diffusion-
respective challenges posed by the specific however, was not a mere mosaic of areas. Rather, based connectivity data often complemented by
characteristics of the human brain. it was an early attempt to group the regions on functional imaging, questionnaires, and neuro-
1
the basis of their myeloarchitectonic similarities logical surveys, and they are the basis for a large
Institute of Neurosciences and Medicine (INM-1), Research
(e.g., the expression of the stripes of Baillarger; body of research worldwide.
Centre Jülich, Jülich, Germany. 2Department of Physics, School
of Mathematics and Natural Sciences, Bergische Universität Fig. 1A) into families and superfamilies and to dMRI is sensitive to the random microscopic
Wuppertal, Wuppertal, Germany. 3Cécile and Oskar Vogt build “nested” representations. In addition, the motion or diffusion of water molecules. By
Institute for Brain Research, Medical Faculty, University Hospital Vogts studied myeloarchitecture in combina- measuring dMRI signals along various ori-
Düsseldorf, Heinrich-Heine University Düsseldorf, Düsseldorf,
Germany. tion with cellular architecture, neurophysiol- entations, distributions of local fiber orienta-
*Corresponding author. Email: [email protected] ogy, anatomy, and even genetics to understand tions can be derived. These orientations are

used to computationally infer trajectories of Klingler’s dissection
white matter pathways (tractography) to esti- With the introduction of dMRI and tractog-
mate connectivity (Fig. 2, A to C) (12). Methods raphy, there has been renewed interest in
for tractography can be grouped according Klingler’s dissection because it enables physi-
to their most distinguishing features, such as cally tracing white matter bundles in postmor-
deterministic and probabilistic, local and global, tem human brains (23). The process of freezing
shortest-path, or machine learning–based algo- and thawing of formalin-fixed tissue facilitates
rithms, with specific applications in research the dissection of fine fiber bundles. Digitiza-
and clinical contexts (12–14). tion poses a particular challenge. Laser scan-
Challenges arise in the context of interpreta- ning was proposed to obtain surface images
tion and quantification. Reconstructed trajecto- to be registered to corresponding MRI scans
ries are modeled entities and do not necessarily (24). Dissection studies are performed on small
represent physical nerve fibers (14). Basic dif- numbers of specimens following selected long-
fusion metrics represent inferences that are range white matter tracts, but whole-brain or
based on local diffusion properties, which are cortical connectivity analyses are currently out
not direct measures of tissue properties, nor of reach.
do they provide information about the direc-
tionality of the connections. Tractography Imaging with polarized light
algorithms may result in false-positive and Polarization microscopy reveals fibers and even
false-negative pathways, e.g., due to limited single axons at the level of a few micrometers.
spatial and angular resolution (15). Given that It does not use contrast agents and relies ex-
a single voxel 1 mm in size contains hundreds clusively on birefringence, which is an optical
of thousands of single fibers, different fiber property of anisotropic material usually caused
configurations within, such as bending, fan- by orderly arranged molecules, atoms, or spa-
ning, crossing, and kissing, can result in the tially repeating configurations, such as those
same signal, making it nearly impossible to found in nervous tissue.
distinguish between the different cases at the Polarization microscopy has been used to
voxel level. Trajectories tend to “travel” par- study nerve fibers in normal and pathological
allel to the cortical surface, largely avoiding tissue for >100 years and experienced a con-
sulcal walls and fundi and terminating prefer- siderable boost when techniques moved from
entially on gyral crowns. Sharp fiber turns from two dimensions (2D) to 3D. When polarized
long-distance fibers and U-shaped fibers light passes through a thin brain section, al-
are not resolvable (14, 16). This makes cross- terations in the polarization state of the light
validation an important topic. generate contrast between fibers of different
orientation and birefringence strength (re-
Postmortem diffusion MRI flecting myelin density). Matrix optics enables
dMRI enables the study of postmortem tissues estimation of 3D orientations of fibers. 3D–
with higher resolution, down to a few hundred polarized light imaging (3D-PLI) was intro-
micrometers. It has been shown to provide duced more than a decade ago (25). In this
meaningful measures of axonal properties such method, fiber orientations are displayed as
as diameters and densities (17) to aid in our a color-coded fiber orientation map (Fig. 3, A
understanding of the biological correlates of to F) or as glyphs when combined over a neigh-
dMRI measures (18). borhood to describe the fiber orientation
Postmortem imaging can also harness the distribution (Fig. 3C) (26). Fiber orientation
advantages of higher magnetic field and gra- distributions can be upscaled to larger voxel
dient strengths, more sensitive radiofrequency sizes for validating distribution functions ob-
coils, and longer scan times (19). Using tailored tained with dMRI.
acquisition pulse sequences, isotropic spatial 3D-PLI can resolve the fine-grained fiber ar-
resolutions of 300 to 500 µm have been achieved. Fig. 1. Historical approaches to studying human chitecture of the cerebral cortex, as well as
This allowed the study of the connectivity of brain connectivity showing association, commis- fibers around and within nuclei. At the same
anatomical structures such as the hippocampus sural, and projection fibers. (A) Myelin-stained time, it enables following projection, associ-
(Fig. 2, D and E) (20), as well as whole brains fibers in the primary visual cortex with the external ation, and commissural pathways over large
(21). Pushing the resolution of a whole-brain band of Baillarger, the stria of Gennari. (Section from distances (Fig. 3). By contrast, conventional
dataset to this resolution leads inevitably to the brain collection of the Cécile and Oskar Vogt myelin staining stays within 2D and does not
terabytes of data and has consequences for soft- Institute for Brain Research, Heinrich Heine University resolve fiber orientations in the densely packed
ware and hardware demands. A human whole- Düsseldorf, Germany.) (B) Drawing of white matter white matter. Moreover, 3D-PLI contrast is gen-
brain tractogram may consist of ~100 million tracts in the region of the central sulcus. Shown are erated by myelinated and unmyelinated axons,
trajectories, posing a serious challenge for vi- projection fibers (thick lines), subcortical U-shaped which results in rich information where myelin
sualization and connectivity assessment. fibers, and intersecting callosal fibers (image repro- staining fails [e.g., mossy fibers in the hippo-
The possibility of using the same tissue for duced with permission from the archive of the Cécile campus (27)].
dMRI and for microscopy opens the possibil- and Oskar Vogt Institute of Brain Research, Heinrich Limitations of 3D-PLI include time- and labor-
ity of validating the results of tractography and Heine University Düsseldorf, Germany). (C) Drawing intensive laboratory work and inaccuracies in
providing biologically meaningful connectivity of a myelin-stained coronal human brain section the 3D reconstruction of sections with dis-
information (22). [image reproduced from (51)]. tortions caused by histological processing.

Registration workflows have been developed tricate connectivity of brain regions such as ferent species to reveal similarities in brain
to reduce distortions and to improve the align- the corpus callosum (28), sagittal stratum structure (31, 32).
ment (19). High-throughput scanning and (29), and brainstem (30). A recent hippocam- Polarization-sensitive optical coherence to-
the use of supercomputing-based workflows pus study investigated the full topography mography (PSOCT) is another approach to
allowed the reconstruction and analysis of large of the different components of the perfo- probing birefringence and determining the
amounts of data (in the terabyte to petabyte rant pathway, a key player in learning and orientation of fibers by means of polarized light
range) and provided new insights into the in- memory (27). 3D-PLI can be applied in dif- (33). This technique relies on the backscatter-
ing of light from a block of tissue, analogous to
ultrasound technologies. PSOCT does not re-
quire the tissue to be sectioned before it is im-
aged, which makes volumetric fiber visualization
possible without large reconstruction artifacts.
However, the technique is currently limited to
cubic centimeter–sized tissue blocks, which still
prevents a whole-brain approach.
Imaging at the subcellular level

Fluorescence microscopy is a rapidly growing
and widely applied optical imaging technology.
Key elements are tissue clearing or refractive
index matching combined with labeling (e.g.,
lipophilic dyes or immunohistochemistry of
myelin-specific proteins) and light-sheet fluo-
rescence microscopy, two-photon fluorescence
microscopy, or confocal fluorescence micros-
copy (19). These techniques are excellent for
studying neural microstructures, mostly in
small tissue samples. Automated fluorescence-
based approaches with integrated sectioning
provide distortion-free 3D reconstructions, ren-
dering investigations of a larger number of
(small) brain samples possible.
An emerging field of imaging uses the phys-
ics of scattering. Although the scattering of
visible light resolves complex fiber constella-
tions (34), small-angle x-ray (35) and neutron
scattering (36) were shown to also quantify
layers of myelin using Bragg’s law of diffrac-
tion. The latter two require access to large ac-
celerators such as the Deutsches Elektronen
Synchroton (DESY) and those at the Euro-
pean Synchrotron Radiation Facility (ESRF)
and the Stanford National Accelerator Lab-
oratory (SLAC).
Electron microscopy can image nanometer-
scale structures from tissues labeled with heavy
metals (e.g., osmium) to reveal the detailed mor-
phology of neurons and glia, myelin sheaths,
synapses, microtubules, or mitochondria with-
in the axoplasm (37). Various methods for 3D
electron microscopy have been developed, dif-
fering in tissue processing and image acquisi-
tion. They are usually confined to sample sizes
smaller than a dMRI voxel. The multibeam se-
rial electron microscope (38) is the first device
that provides nanometer-resolved images at
Fig. 2. Diffusion MRI and tractography. (A) Vector field of local predominant fiber orientations and two of its the millimeter range.
trajectories depicted on a coronal view of the human brain. The blue trajectory is part of the corticospinal tract, Classical tracing, the gold standard for proving
and the red one is part of the corpus callosum. (B and C) Whole-brain tractogram (B) and virtual dissection of synaptic connectivity, is methodically highly
multiple fiber bundles thereof (C). Bundles are selected as trajectories that pass expert-defined regions of interest. challenging in the human brain, and only a
(D and E) Orientation estimation in a postmortem human hippocampus at 300-μm resolution with color-coded few applications exist. For example, the Nauta
diffusion directions sliced along the axial (D) and the coronal (E) planes. Orange arrowheads indicate the fimbria, method revealed callosal connectivity in early
yellow the alveus, and pink the lacunosum-molecular layer. [Images in (A) to (C) were modified from (14)/CC BY 4.0, visual areas at their cytoarchitectonic borders
and images in (D) and (E) were modified from (20)/CC BY 4.0.] (39), and carbocyanine-based tracing has been

Fig. 3. Human brain fiber
A architecture. (A) Fiber orienta-
B tion map of a coronal section
through a human hemisphere
C obtained with 3D-PLI at 1.3 ×
1.3 × 70 μm3 resolution (color
B sphere encodes 3D fiber orienta-
m
tions). CC, corpus callosum;
1m
CSO, centrum semiovale;
IC, internal capsule; ILF, inferior
longitudinal fascicle; LGN, lateral
C geniculate nucleus; MD, medio-
dorsal thalamic nucleus; VL,
CSO ventrolateral thalamic nucleus.
(B) Details of projection and
CC long association fibers (continuous
D lines) and U-shaped fibers
(dashed-dotted lines) from a
region of interest at higher
IC magnification. (C) Radial and
VL
MD horizontal cortical fiber directions
for a 1 × 1 mm2 region of
interest (also visualized as a
fiber orientation distribution).
D (D) Fanning and splitting of
LGN LGN fascicles (red/yellow) and fibers
in the region of the subcortical
E nucleus. (E) Human hippocampus
ILF showing dentate fibers, mossy
fibers (asterisk), the endfolial
pathway (dashed line), and
E Schaffer collaterals (dotted line).
(F) 3D reconstruction of serial
fiber orientation maps of the
F occipital lobe of a vervet monkey.
(G) Myelin-stained hippocampus.
The asterisk indicates the
mossy fiber region, which is
weakly stained, in contrast to
visualization by 3D-PLI in (E).
(Section from the brain collection
of the Cécile and Oskar Vogt
Institute for Brain Research,
G
Heinrich Heine University
Düsseldorf, Germany.)
compared with diffusion-based tractography modalities, and sources into an atlas framework. block was imaged using anatomical and dif-
in the same tissue and has revealed a high de- There is a growing number of data integration fusion MRI, 3D-PLI, and two-photon fluo-
gree of concordance (40). efforts; for example, a recent concordance map rescence microscopy and shared through the
of human brain architecture combines histol- EBRAINS research infrastructure (https://
Mapping connectivity and sharing data ogy, immunohistochemistry, and MRI (41). doi.org/10.25493/JQ30-E08). Such multimodal
To study the connectome at multiple levels re- Cellular images have also been analyzed for datasets provide essential insights into the mi-
quires integration of data from different scales, fiber courses (42). One human hippocampus crostructural characteristics of complex fiber

bundles, helping to improve our understand- data have been developed by different consor- of fibers (e.g., fanning, crossing, converging,
ing of MRI measurements and the reliability tia such as HIBALL (https://bigbrainproject. etc.). As a consequence, tracing algorithms need
of tractography (43, 44) and thus stimulating org/hiball.html), EBRAINS (https://ebrains.eu), to be empowered to work at different hierar-
the field of quantitative MRI. JANELIA (https://www.janelia.org), and the chies of discretization (e.g., scalable anisotropic
Tools to spatially organize connectivity data Allen Brain Map (https://portal.brain-map. voxels) during runtime.
across scales are required. Mutual landmarks org). Repositories need to be interoperable
have to be identified and described in multi- to enable scaling up efforts to approach the REFERENCES AND NOTES
modality images at different scales. Gross human connectome. 1. O. Sporns, G. Tononi, R. Kötter, PLOS Comput. Biol. 1, e42 (2005).
2. S. J. Cook et al., Nature 571, 63–71 (2019).
anatomical structures (e.g., fiber bundles, gyri, Research infrastructures ideally bring together 3. M. N. Economo et al., eLife 5, e10566 (2016).
sulci, and axon distributions) are suitable for insights and data from different fields to further 4. L. F. Abbott et al., Cell 182, 1372–1376 (2020).
large-volume intersubject alignment, whereas develop brain theory, which then can drive both 5. K. S. Rockland, Brain Struct. Funct. 225, 1327–1347 (2020).
6. C. Vogt, O. Vogt, Allgemeine Ergebnisse unserer Hirnforschung
microstructure features (e.g., vasculature, indi- empirical research and simulation. In particular, (JA Barth, 1919), vol. 25.
vidual fibers, and cell bodies) are suited for the principles of brain topography as observed 7. D. J. Felleman, D. C. Van Essen, Cereb. Cortex 1, 1–47 (1991).
high-precision, intrasubject alignment of small by studies of cyto-, myelo-, and chemoarchitec- 8. D. Zachlod, S. Bludau, S. Cichon, N. Palomero-Gallagher,
K. Amunts, Neuroimage 257, 119286 (2022).
tissue (sub-)samples. Knowing the precise lo- ture and interregional connectivity are likely to 9. C. C. Hilgetag, A. Goulas, Philos. Trans. R. Soc. Lond. B Biol.
cation of each (sub-)sample is indispensable to ultimately influence brain dynamics, including Sci. 375, 20190319 (2020).
preserving brain region–specific connectivity the local ratio of excitatory to inhibitory cell 10. G. Deco et al., Curr. Biol. 28, 3065–3074.e6 (2018).
11. J. S. Elam et al., Neuroimage 244, 118543 (2021).
characteristics. MRI or en face scans accom- activity, resulting in a variable balance across
12. F. C. Yeh, A. Irimia, D. C. A. Bastos, A. J. Golby, Neuroimage
panying tissue preparation can act as individ- different brain regions (10, 47, 48). Studies in- 245, 118651 (2021).
ual references to be aligned with an atlas. creasingly benefit from integrating the differ- 13. F. Zhang et al., Neuroimage 249, 118870 (2022).
New microanatomical connectivity models ent facets under a common infrastructure (49). 14. B. Jeurissen, M. Descoteaux, S. Mori, A. Leemans, NMR Biomed.
32, e3785 (2019).
[equivalent to the BigBrain model (45)] would 15. K. H. Maier-Hein et al., Nat. Commun. 8, 1349 (2017).
be helpful when used as reference data of an Conclusion 16. C. Reveley et al., Proc. Natl. Acad. Sci. U.S.A. 112,
atlas, accommodating a broad range of formats Mastering the empirical, theoretical, and com- E2820–E2828 (2015).
17. Y. Assaf et al., Neuroimage 80, 273–282 (2013).
(e.g., scalars, vectors, glyphs, and tractograms). putational challenges for bridging the differ- 18. T. B. Dyrby, G. M. Innocenti, M. Bech, H. Lundell, Neuroimage
Because the results of tractography depend ent spatial (and temporal) scales will open 182, 62–79 (2018).
on several constraints, there is a need to pro- new perspectives for a deeper understanding 19. A. Yendiki et al., Neuroimage 256, 119146 (2022).
20. J. Beaujoin et al., Brain Struct. Funct. 223, 2157–2179 (2018).
vide variants of tractogram solutions. This is of the connectome and its impact on brain 21. F. J. Fritz et al., Neuroimage 202, 116087 (2019).
already routine in particle physics, where an function and disease. Diffusion-based imaging 22. I. Shamir, O. Tomer, R. Krupnik, Y. Assaf, Brain Struct. Funct.
entire set of probable particle trajectories are addresses long-distance connections and can 227, 2153–2165 (2022).
23. T. A. Dziedzic, A. Balasa, M. P. Jeżewski, Ł. Michałowski,
provided to the community. Variations also be applied to study healthy subjects and pa- A. Marchel, Brain Struct. Funct. 226, 13–47 (2021).
exist at the level of brains (intersubject var- tients, whereas the finer fiber and axonal ar- 24. I. Zemmoura et al., Neuroimage 103, 106–118 (2014).
iability), which should be part of such frame- chitecture cannot be resolved. By contrast, 25. M. Axer et al., Neuroimage 54, 1091–1101 (2011).
26. A. Alimi et al., Med. Image Anal. 65, 101760 (2020).
works. Thus, multiscale connectivity data can microscopy and optical imaging provide supe-
27. M. M. Zeineh et al., Cereb. Cortex 27, 1779–1794 (2017).
easily reach petabyte sizes (46). rior contrast in the gray matter, but they are 28. J. Mollink et al., Neuroimage 157, 561–574 (2017).
Joint efforts would be beneficial to set up less powerful to disentangle fibers in the white 29. S. Caspers, M. Axer, D. Gräßel, K. Amunts, Brain Struct. Funct.
research infrastructures and establish stan- matter. To take advantage of both worlds and 227, 1331–1345 (2022).
30. D. J. H. A. Henssen et al., Brain Struct. Funct. 224, 159–170 (2019).
dards. Several large brain initiatives, multi- to reduce limitations, strategies are evolving 31. M. Stacho et al., Science 369, eabc5534 (2020).
partner projects, and repositories (e.g., HBP, to combine methods with their respective 32. H. Takemura et al., eLife 9, e55444 (2020).
HCP, UK biobank, etc.) provide a broad scope strengths and to fill in the gaps in experimen- 33. H. Wang et al., Neuroimage 165, 56–68 (2018).
34. M. Menzel et al., Neuroimage 233, 117952 (2021).
of different datasets and tools. Web-based tools tal knowledge. Digital tools such as modeling, 35. M. Georgiadis et al., Nat. Commun. 12, 2941 (2021).
for the visualization and analysis of connectivity simulation, and machine learning–based ap- 36. S. Maiti et al., Sci. Rep. 11, 17306 (2021).
proaches are capable of predicting missing in- 37. S. Loomba et al., Science 377, eabo0924 (2022).
38. A. L. Eberle, D. Zeidler, Front. Neuroanat. 12, 112 (2018).
formation (50). To constrain and validate them, 39. S. Clarke, J. Miklossy, J. Comp. Neurol. 298, 188–214 (1990).
“gold standards” from anatomical and physio- 40. A. K. Seehaus et al., Cereb. Cortex 23, 442–450 (2013).
Box 1. Topics of future research. logical work are mandatory, including cross- 41. A. Alkemade et al., Sci. Adv. 8, eabj7892 (2022).
42. R. Schurr, A. A. Mezer, Science 374, 762–767 (2021).
validation in the same tissue imaged with 43. S. Schiavi et al., Neuroimage 249, 118922 (2022).
Connect the 3D myelo- (or axonal) architecture as a
description of the gray matter with pathways of the different methods (19). 44. R. Schurr et al., Neuroimage 181, 645–658 (2018).
45. K. Amunts et al., Science 340, 1472–1475 (2013).
white matter. It is an intriguing concept to approach the 46. K. Amunts, T. Lippert, Science 374, 1054–1055 (2021).
Assess the directionality of connections. connectome not only as a multiscale system in 47. M. L. Kringelbach et al., Proc. Natl. Acad. Sci. U.S.A. 117,
which each scale (e.g., neurons, microcircuits, 9566–9576 (2020).
Correlate the axonal and network architecture with neu- 48. D. Jancke, S. Herlitze, M. L. Kringelbach, G. Deco, FEBS J. 289,
ronal, cellular architecture. and networks) has distinct features, but also as 2067–2084 (2022).
a system that has repetitive properties. How- 49. International Brain Initiative, Neuron 105, 212 (2020).
Elucidate relationships of structural connectivity with
ever, to reveal the principles of connectivity 50. D. M. Camacho, K. M. Collins, R. K. Powers, J. C. Costello,
brain dynamics and plasticity.
J. J. Collins, Cell 173, 1581–1592 (2018).
within the experimentally accessible range of 51. J. J. Déjerine, Anatomie des Centres Nerveux (Rueff, 1895).
Build microscopical templates of connectivity, interoper-
able with existing atlases and maps. scales (from axons to pathways), in other words,
to describe the human brain’s “nestedness,” AC KNOWLED GME NTS
Supplement the axonal by the dendritic architecture.
requires critically revisiting the methodology, This work was supported by the European Union’s Horizon 2020
Predict connectivity on the basis of sparse regional infor- Framework Programme for Research and Innovation (grant no.
including tractography. Future tractography 945539: “Human Brain Project” SGA3). License information:
mation from high-resolution optical methods and elec-
tron microscopy, comparative studies, and modeling and could, in a stepwise manner, move from lower Copyright © 2022 the authors, some rights reserved; exclusive
simulation. to higher resolution, being informed by both licensee American Association for the Advancement of Science. No
claim to original US government works. https://www.science.org/
Develop tractography algorithms able to reproduce nested regional properties of the next higher resolution
about/science-licenses-journal-article-reuse
connectivity patterns. (e.g., myeloarchitecture, cortical columns, and
types of neurons) and the underlying geometry 10.1126/science.abq2599

REVIEW onstrated for identifying the visual word form
area, even before the acquisition of literacy
The emergent properties of the connected brain (14). Typically, the main trends of brain con-
nectivity capture the spatial organization of
Michel Thiebaut de Schotten1,2* and Stephanie J. Forkel2,3,4,5 functions in the brain (15). Deprived of its con-
nections, a brain region will prune its remain-
There is more to brain connections than the mere transfer of signals between brain regions. Behavior ing dendrites and synapses, and its neurons
and cognition emerge through cortical area interaction. This requires integration between local and will wither or die (16). Consequently, this re-
distant areas orchestrated by densely connected networks. Brain connections determine the brain’s gion’s network breaks down functionally [i.e.,
functional organization. The imaging of connections in the living brain has provided an opportunity to identify dysconnection (17)] and structurally (i.e., dis-
the driving factors behind the neurobiology of cognition. Connectivity differences between species and connection) and no longer can contribute to
among humans have furthered the understanding of brain evolution and of diverging cognitive profiles. Brain a function. This phenomenon, also known as
pathologies amplify this variability through disconnections and, consequently, the disintegration of cognitive diaschisis (18), demonstrates the critical im-
functions. The prediction of long-term symptoms is now preferentially based on brain disconnections. This portance of connections in maintaining the
paradigm shift will reshape our brain maps and challenge current brain models. integrity of distant brain regions and their func-
tioning. Lastly, animal studies that surgically
swap connections between sensory cortices
ew ideas often emerge through close in- man brain using neuroimaging methods that have further demonstrated the leading and
N
teraction between minds. Accordingly, unveil their structure (i.e., axon bundles) and decisive role of white matter connections with-
these ideas do not solely belong to any their function (i.e., synchronization of regional in the brain’s functional organization. Specif-
of these individual minds but rather are activity). Structurally, diffusion-weighted im- ically, forcing visual inputs onto the auditory
the fruit of their integration—an emer- aging (DWI) measures water diffusion along cortex alters it to acquire many of the cyto-
gent property of mutual exchange and inter- the directions of axons and can derive proper- architectonic and functional properties of the
action. The concept behind integration comes ties such as trajectory (i.e., tractography), den- visual cortex (19) that are associated with nor-
from emergentism, which postulates that “the sity, caliber, and dispersion (8). Functionally, mal visual behavior (20). This indicates that
whole is something besides the parts” (1) and the communication between brain regions can cell properties are mediated by their connec-
that “no complex system can be understood be extrapolated by measuring the coherence tions through their interaction with the rest
except through careful analysis; however, the between distant areas’ activity, mainly through of the brain, and moreover, those brain func-
interactions of the components must be con- functional magnetic resonance imaging (fMRI). tions are an emerging property of these inte-
sidered as much as the properties of grative mechanisms. Thus, as previously
the isolated components” (2). In neuro- hypothesized (21), this sum of evidence
science, there is a growing consensus
that functions are an emerging prop-
“…circuits create networks by stringing demonstrates that brain connections sup-
port the latent mechanisms that determine
erty of the interaction between brain together many brain regions to orchestrate the function of the brain and cognition as
areas (3). Thus, function-specific brain we know it.
activity involves the integrative effort a brain symphony conducted by finely This shift away from considering iso-
of several brain regions (4). White mat-
ter connections support this integra-
attuned connections with variable caliber lated regions in favor of an integrated
anatomical-functional network led to the
tion by interconnecting brain regions
(Fig. 1). With cortical expansion, these
and myelination…” reevaluation of functional activations with
regard to their white matter connections.
connections have evolved to preserve Accordingly, brain connections revealed
interactions between distant regions (5). These Mathematical models applied to fMRI can es- by the highest-resolution tractography (22) can
connections support local, intra- and interlobar timate communication strength (i.e., functional be used to systematically decipher human ac-
associations, projection, and interhemispheric connectivity) or directionality (i.e., effective con- tivation networks and recently led to the first
commissural circuits (6). These circuits create nectivity) (9). Because regions that fire together functional white matter atlas (23). This atlas
networks by stringing together many brain wire together (10), measuring communication identifies the joint contribution—or integra-
regions to orchestrate a brain symphony con- between regions indicates their connections, tion—of structurally connected brain areas by
ducted by finely attuned connections with var- albeit these regions are not always directly a statistical association of fMRI and diffusion
iable caliber and myelination tailored to their connected (11). data (Fig. 2B). The result also revealed a leftward
functional role (7). asymmetry between the known granularity (i.e.,
The integrative functional role of brain connectivity the level of detail) of functions of the left and the
Measuring brain connectivity in the living brain There is, however, more to brain connections right hemispheres. This asymmetry reflects an
Today, the connections between brain regions than the mere transfer of signals between brain epistemological imbalance (i.e., we know more
are noninvasively measured in the living hu- regions. Connections can amplify or reduce about the cognition of the left than the right
1
brain signals (11) and determine the brain’s hemisphere), leading to a publication bias and
Groupe d’Imagerie Neurofonctionnelle, Institut des Maladies
Neurodégénératives UMR 5293, CNRS, CEA, University of
cortical structure and function. Specifically, triggering the necessity for more dedicated
Bordeaux, Bordeaux, France. 2Brain Connectivity and there is a similarity between the synchronized cognitive explorations of the right hemisphere,
Behaviour Laboratory, Sorbonne University, Paris, France.
3
communication between brain regions at rest which for a long time was considered the mi-
Donders Centre for Brain Cognition and Behaviour, Radboud and their activity during tasks at the group nor or nondominant hemisphere.
University, Nijmegen, Netherlands. 4Centre for Neuroimaging
Sciences, Department of Neuroimaging, Institute of (12) and the individual (13) levels (Fig. 2A). It is The mechanisms that sustain the lateraliza-
Psychiatry, Psychology, and Neuroscience, King’s College also possible to predict where a function will tion of brain functions are related to interhemi-
London, London, UK. 5Department of Neurosurgery, arise in the developing brain on the basis of spheric connections. Some cortical regions show
Technical University of Munich School of Medicine, Munich,
Germany. the cortical projections of white matter tracts increased functional asymmetries for cognitive
*Corresponding author. Email: [email protected] alone. For instance, the latter has been dem- functions such as language, perception and

action, emotion, and decision-making. These cies (25, 26). In particular, the axonal con- connectivity across species (25). Taken together,
cortical regions are less connected with the ductive properties can change (27), and axon these studies suggest that during evolution,
contralateral hemisphere via the corpus cal- calibers correlate with the interhemispheric brain size expansion may have led to functional
losum (Fig. 2C) (24). DWI has revealed that speed of conduction (Fig. 2D) (28). Further, a lateralization to avoid a disproportionate corpus
the structure of the corpus callosum changes trade-off in the number of connections exists callosum or excessive conduction delays across
along with brain size across and within spe- between interhemispheric and intrahemispheric hemispheres.
Mapping evolution through connectivity

A B As structural connectivity can be used to de-
cipher the brain’s functional organization, multi-
2 ple studies compared different primate species
1 to understand human uniqueness and shed
light on the mechanisms involved in its evolu-
tion. For instance, human language capacity
parallels the extraordinary expansion of the ar-
cuate fasciculus in the left hemisphere (Fig. 3A)
Modular language model Hierarchical language model
(29). The anatomical delineation of white matter
Broca’s area for articulation 1 Direct processing route tracts (e.g., arcuate fasciculus or corpus cal-
Geschwind’s area for concepts
2 Indirect processing route losum) allowed for extracting corresponding
Wernicke’s area for comprehension connectivity profiles across species. This point
of comparison permitted the computation of
deformation fields between species’ brains
C
(Fig. 3B) (30). These deformation fields define
Integrative language model
1 2 similarities and differences across species.
1 Motor speech production Comparative studies assume that similarities
2 Semantic processing between species can be traced back to a com-
mon ancestor and account for the preservation
3 Auditory speech comprehension
of specific functions across evolution. Recent
4 Verbal working memory comparative work revealed one of the first
3 4 Inactive comprehensive maps of the phylogenetic or-
Active ganization of brain regions (30). Such tech-
nical advances in comparative neuroimaging
will allow for targeted studies that better match
X Y Z human brain mechanisms to their phylogenetic
1 0 0 counterparts. These advances may also help
discover and mimic neuroprotective mecha-
1 1 0
D nisms in animals that could potentially trans-
1 0 1
late to improving human disease models and
X Y Z 0 1 0 therapeutics. For instance, frontoparietal dis-
Brain states
1 0 0 X Y Z 0 1 1 connection is a very sensitive (85%) and specific

0 1 0 1 1 1 0 0 1 (95%) biomarker for persistent disorders of
0 0 1 1 0 1 1 1 1 visual neglect (31). Whereas most humans with
this disconnection will fail to recover from
E visual neglect, monkeys with the same discon-
120 nection will recover within a few days (32). Hence,
Models
there is a distinctive mechanism in monkeys
Brain states
90 Modular
that facilitates brain recovery. However, this
60 Hierarchical
mechanism has yet to be identified, and its
Integrative
30 translatability to humans needs to be explored.
One frequent limitation of comparative studies
0 is the small number of brains per species used
3 4 5 6 7
Brain areas (n) (usually fewer than 10), which fails to fully cap-
ture interindividual variability. While the
Fig. 1. The superior flexibility of the integrative brain model compared with other classical models, amount of connectivity variability (i.e., mag-
as exemplified by simplified language models. Brain models determine the flexibility of brain states nitude) is proportional to the brain size, its
(i.e., functions). In a modular system (A), one region performs one function without cross-talk, and the pattern of variability is similar between hu-
number of possible brain states increases linearly with the number of regions. In the hierarchical system (B), mans and other primates. Accordingly, brain
functions emerge from the sequential activation of regions. Accordingly, the repetition of words or areas that have recently evolved increasingly
sentences would rely on the temporal-parietal-frontal propagation from auditory-to-motor processes. In differ between individuals, whereas evolution-
contrast to (A) and (B), the integrative model (C) offers the highest computational flexibility, allowing for the arily older areas tend to be more stable (33).
high complexity and flexibility of language processes as we know them. Each model can be translated Intraspecies variability in brain connections
into different brain state patterns (D) across brain areas X, Y, and Z. Each line indicates a brain state might therefore be a novel dimension to our
(function) (courtesy of Chris Foulon). (E) Illustration of the interaction between the brain model and the understanding of evolutionary mechanisms
number of areas involved in the number of brain states a system can take. (Fig. 3C).

A B
Frontal
R
0.8
Ax Ay Az
A1
Temporal
Actual
A2 0.5
A 200
Occipital
Z
-6 6 Predicted
Right hemisphere
C D
µm
1.0
Hemisphere 6 T
Left
Degree of functional lateralization
80
Speed of conduction (m•s-1)

0.9 Right
T
0.8 V V
60
0.7 1
40
0.6
0.5
20 Modality
0.4 Visual (V)
Tactile (T)
0.3
0.00 0.07 0.18 0.36 0.64 0 1 2 3 4 5 6
Interhemispheric probability of connection (log scale) Interhemispheric axonal diameter (µm)
Fig. 2. Functional integration through brain connections as the latent mecha- allows projection of the processes involved in the fMRI tasks onto the brain
nisms determining the function of the brain. (A) Functional connectivity (top left) connections [modified from (23)]. R, goodness of fit. (C) The relationship
can be summarized in a connectivity matrix (bottom left) that is specific to each between functional lateralization and interhemispheric connectivity for the left (blue)
individual (35). Main trends of functional connectivity can predict individual patterns and the right (red) hemispheres [modified from (24)]. (D) Interhemispheric speed
of task-related activations [right; reproduced with permission from (13)]. A, area. of conduction for visual (gray) and tactile (black) modalities is correlated with
(B) Similarly, main trends of structural connectivity can predict task-related patterns the strength of their interhemispheric connections [i.e., axonal diameter; modified
of cortical activations. The statistical association between these two modalities with permission from (28)].
Connectivity unveils interindividual variability cognition [i.e., connectome-based predictive its degeneration has been associated with in-
This connectivity variability, or “neurovari- modeling (34)]. These fingerprints of connec- creased symptom severity (41). These obser-
ability,” is critical to the individuals we are. tivity are specific to individuals and predict vations have led to new anatomical-cognitive
What we know, who we are, and how we com- fluid intelligence (35) and creativity (36) with models 150 years after the first descriptions
municate with others are ascribed through impressive accuracy. These correlations with of aphasia (i.e., language disorder) as a discon-
integrative brain mechanisms. Therefore, to cognition also extend to the differences in the nection syndrome.
understand the origin of our identity, we need structure of specific brain connections [see Connectivity profiles can change across a life
to decipher how connections between brain Fig. 4A and (37)]. For instance, a stronger left span, leading to increasingly divergent molec-
regions orchestrate our brain functions at the arcuate fasciculus seems to facilitate the learn- ular and circuit-level changes that develop over
individual level. Accordingly, the strength of ing of new words (38). These differences affect weeks, months, years, and even decades owing to
communication between brain regions de- healthy brain performance and extend to the environmental and learning-induced plasticity
rived from functional connectivity can pre- severity of neurodevelopmental, psychiatric, and mechanisms. A good example is literacy, which
dict individual differences in brain activation neurological symptoms (37). Developmental brain alters brain connections between the visual and
during tasks (Fig. 2A) (13). Preliminary work connectivity patterns have assisted in diagnos- the auditory system even when acquired later
has already created “fingerprints” of the brain’s ing neurodevelopmental learning deficits (39). in life (42). Indeed, plasticity mechanisms can
functional connectivity patterns in healthy In neurology, a stronger arcuate fasciculus limit long-term cognitive predictions, which are
volunteers that correlate with behavior and facilitates recovery after stroke (40), whereas based on the fingerprints of brain connections.

A B T 1 T 2T 3 T 4 …
V1 Tracts
Macaques
Left hemisphere
Strength
V2
Right hemisphere T1
V3
Macaques
Inferior frontal gyrus V4 T2
…
Superior temporal gyrus
Arcuate fasciculus Common T3
space T 1 T 2T 3 T 4 …
Chimpanzees
V1 T4
Strength
V2
Chimpanzees
V3 T5
V4
…
Common
space
T 1 T 2T 3 T 4 …
V1
Humans
Strength
V2
Humans
V3
V4
…
1000 2000 3000 25 6 Today
Size (streamlines) Phylogenetic distance (millions of years)
Humans (mm) Chimpanzees (mm) Macaques (mm)
2 4 0.75 2.5 0.5 2
Fig. 3. Connectivity sheds light on mechanisms of brain evolution. (A) A (i.e., as a surrogate for connectivity strength) for each tract of interest. T, tract;
comparison of the connections between frontal and temporal brain regions V, voxel; Strength, connectivity strength. (C) Connectivity variability (i.e., average
in humans, chimpanzees, and macaques reveals the remarkable expansion and white matter deformation required to match a common species-specific template)
lateralization of the arcuate fasciculus (29). (B) The extraction of a reliable reveals that the same variability that makes us individually different from
connectivity profile across species permits the computation of common spaces each other is also at the root of our differences from our ancestors and our
allowing for an approximation of our ancestors’ brains (30). In this context, closest evolutionary relatives [modified from (33), thanks to data openly available
the connectivity profile is defined as the number of streamlines per voxel at http://www.chimpanzeebrain.org]. mm, millimeters.
Importantly, while the identification of predic- pathologies and can induce long-lasting func- and understand the critical contribution of
tive biomarkers holds vast potential for changing tional symptoms. For instance, a disconnec- connections to the realization of functions.
the health of individuals and populations, it also tion between visual and language networks These new methods can even demonstrate
bears fundamental ethical risks and moral chal- leads to irremediable alexia [i.e., inability to that clinical-anatomical lesion studies in neu-
lenges (e.g., treatment of predicted brain dis- read (43)]. Although advanced neuroimaging roscience’s most famous cases can be extended
ease that may never manifest or withholding can identify disconnections, these methods to a disconnection paradigm. This new para-
treatment on the basis of recovery predictions). are not yet systematically available across the digm shows the networks considered function-
Overall, these recent studies put forth evidence clinical sector. Therefore, new indirect methods ally engaged for emotion and decision-making
demonstrating that new behavioral patterns that use a priori knowledge of connections (DWI (Phineas Gage), language production (Louis
and cognitive functions can arise from even or fMRI) derived from the highest-resolution Victor Leborgne), and declarative memory
small changes in the interaction between brain datasets to estimate disconnection profiles after (Henry Molaison) in the healthy population
regions via their connections. a brain injury are needed to reliably and statis- (Fig. 4B) (45). Hence, consideration of brain
tically map the association between discon- connections appears to reconcile brain lesion
Functional disintegration through disconnection nection and symptoms (44). In doing so, it is studies with functional neuroimaging in
Drastic disruptions of brain interactions [i.e., possible to reevaluate classical clinical neuro- healthy volunteers and provides a more com-
disconnections (18)] manifest secondary to anatomical phenomena within brain networks prehensive biological interpretation of clinical

A B C
Hemisphere Left Right Unknown Right motor functions Visuospatial attention Verbal memory
Left motor functions Visuospatial memory Language
Cingulum Sickness impact profile
Symptoms
Memory
Language
Executive
Attention
Motor
Visual
Sleep
Mood Phineas Gage
Social
Addiction
Arcuate segments
Symptoms
Memory
Language
Executive
Attention Anterior
Motor Long
Auditory Posterior
Sleep Louis Victor Leborgne
Uncinate
Symptoms
Memory
Language
Executive
Attention
Motor
Auditory
Mood
Social
2 10 20 30 40 50 Henry Molaison
Number of correlations
Fig. 4. Disintegration of brain functions through disconnections. neurosciences showing the tracts disconnected in classic neurobehavioral
(A) Systematic review of the relationship between variability in brain connectivity, syndromes [from (45)]. (C) Neuropsychological white matter atlas highlighting
cognitive profiles, and clinical symptoms [modified from (37); https://youtu.be/ the disconnection patterns causing loss of function [courtesy of (47);
JVOegYHT21w]. (B) Connectivity revisiting neuroscience’s most famous cases of http://disconnectomestudio.bcblab.com].
manifestations with regard to the disinteg- the brain is in the interaction between mul- regions. However, regardless of the quality
ration of brain processes. Further, brain dis- tiple areas, and the best guide to its consti- of the dataset used to build these priors, axons
connection results usually achieve a higher tutional (as opposed to phasic, task-related) are ∼1 to 6 mm in diameter (11), and routine
explanatory power than lesion localization interactions is its connecting infrastructure. neuroimaging has a minimum resolution of
alone (46). The disconnection framework 1 mm3 in vivo and 200 mm postmortem (50).
has recently been extended to the entire brain Where do we go from here? Similarly, monosynaptically connected areas
to provide the first clinical map of symptoms Despite the current progress in estimating brain are synchronized with a delay of 2 to 3 ms (51),
associated with specific brain disconnections connectivity, new challenges have emerged that leading to high-frequency synchronization.
(Fig. 4C) (47). Therefore, it would prove will only be tackled by relying on synergetic Yet standard functional connectivity relies
beneficial if measures of brain connectivity efforts. First and foremost, no known method on acquisitions with a temporal resolution
were translated into advanced standard op- can directly measure the activation of connec- of 1 s at best. Hence, some level of approx-
erating procedures for personalized neuro- tions in the healthy living human brain. In- imation exists in the estimation of the orien-
science (48) that focus on rehabilitation and stead, recent indirect approaches statistically tation of fiber populations and the estimation
that support the prediction of symptom recov- project the functional signal from the cortex of the communication between brain regions,
ery while providing new targets for pharma- onto the white matter (49). These indirect meth- leaving room for improvement to estimate
cological treatments. The evidence presented ods rely on a priori knowledge of the group- the connectome (i.e., whole-brain connectiv-
demonstrates that the key to understanding level probability of connections between brain ity) accurately.

for cytoarchitecture measures and cortical thick- advanced neuroimaging methods, personal- 32. J. W. Marshall, H. F. Baker, R. M. Ridley, Behav. Brain Res. 136,
ness. Non-neuroimaging methods may get us ized anatomical models, and clinical impact. 257–265 (2002).
33. P. L. Croxson, S. J. Forkel, L. Cerliani, M. Thiebaut de Schotten,
closer to that goal. Advanced polarized light Hence, it will take an integrative system of Cereb. Cortex 28, 3829–3841 (2018).
imaging (53) or Nissl-based structure tensor people (research consortium) to decipher this 34. X. Shen et al., Nat. Protoc. 12, 506–518 (2017).
(54), for example, can estimate axonal orien- complex system within us all and discover the 35. E. S. Finn et al., Nat. Neurosci. 18, 1664–1671 (2015).
36. M. Ovando-Tellez et al., Sci. Adv. 8, eabl4294 (2022).
tation from postmortem tissues. However, emergent properties of the brain. 37. S. J. Forkel, P. Friedrich, M. Thiebaut de Schotten, H. Howells,
these postmortem approaches are limited to Brain Struct. Funct. 227, 529–544 (2022).
two dimensional (2D) in-plane reconstruc- 38. D. López-Barroso et al., Proc. Natl. Acad. Sci. U.S.A. 110,
13168–13173 (2013).
tion and require tremendous effort to pro- 1. Aristotle, Metaphysics, C. D. C. Reeve, Transl. (Hackett 39. D. E. Astle, J. Bathelt, J. Holmes; CALM Team, Dev. Sci. 22,
duce a whole 3D human brain comparable Publishing Company, 2016). e12747 (2019).
2. E. Mayr, What Makes Biology Unique? Considerations on the
to BigBrain. Preliminary work already dem- 40. S. J. Forkel et al., Brain 137, 2027–2039 (2014).
Autonomy of a Scientific Discipline (Cambridge Univ. Press, 41. S. J. Forkel et al., Neurology 94, e594–e606 (2020).
onstrates that such an attempt is possible in 2004). 42. M. Thiebaut de Schotten, L. Cohen, E. Amemiya, L. W. Braga,
macaques, and efforts are underway in hu- 3. L. Pessoa, The Entangled Brain: How Perception, Cognition, and S. Dehaene, Cereb. Cortex 24, 989–995 (2014).
mans (55). Future progress will serve as a Emotion Are Woven Together (MIT Press, 2022). 43. L. Cohen, S. Dehaene, S. McCormick, S. Durant, J. M. Zanker,
4. G. Herbet, H. Duffau, Physiol. Rev. 100, 1181–1228 (2020). Neuropsychologia 91, 36–49 (2016).
cornerstone for improving structural and 5. D. N. Pandya, M. Petrides, B. Seltzer, P. B. Cipolloni, Cerebral 44. M. D. Fox, N. Engl. J. Med. 379, 2237–2245 (2018).
functional connectivity methods. These new Cortex: Architecture, Connections, and the Dual Origin Concept 45. M. Thiebaut de Schotten et al., Cereb. Cortex 25, 4812–4827
datasets can validate current neurobiological (Oxford Univ. Press, 2015). (2015).
6. K. Rojkova et al., Brain Struct. Funct. 221, 1751–1766 46. V. Pacella et al., eLife 8, e46075 (2019).
approximations—such as axonal diameter, (2016). 47. L. Talozzi et al., Research Square [Preprint] (2021). https://
neurite density, the extent of myelination, in- 7. B. Zalc, F. Rosier, Myelin: The Brain’s Supercharger (Oxford doi.org/10.21203/rs.3.rs-1181593/v1.
formation flow, or synaptic complexity at the Univ. Press, 2018). 48. T. D. Satterthwaite, C. H. Xia, D. S. Bassett, Neuron 98,
8. Y. Assaf, H. Johansen-Berg, M. Thiebaut de Schotten, NMR 243–245 (2018).
whole-brain level—derived from neuroimag- Biomed. 32, e3762 (2019). 49. V. Nozais, S. J. Forkel, C. Foulon, L. Petit,
ing. This step is fundamental to ensure that 9. K. J. Friston, Brain Connect. 1, 13–36 (2011). M. Thiebaut de Schotten, Commun. Biol. 4, 1035 (2021).
connectivity matrices are reliably translatable 10. S. Löwel, W. Singer, Science 255, 209–212 (1992). 50. S. B. Sébille, A.-S. Rolland, M.-L. Welter, E. Bardinet,
11. G. M. Innocenti et al., Prog. Neurobiol. 208, 102186 M. D. Santin, J. Neurosci. Methods 311, 222–234 (2019).
to the clinical realm and lead to advances in (2022). 51. B. L. Sabatini, W. G. Regehr, Annu. Rev. Physiol. 61, 521–542
therapeutic interventions. 12. S. M. Smith et al., Proc. Natl. Acad. Sci. U.S.A. 106, (1999).
Going forward, we need to consolidate these 13040–13045 (2009). 52. K. Amunts et al., Science 340, 1472–1475 (2013).
13. I. Tavor et al., Science 352, 216–220 (2016).
new concepts by developing dedicated soft- 53. M. Menzel et al., Sci. Rep. 12, 4328 (2022).
14. Z. M. Saygin et al., Nat. Neurosci. 19, 1250–1255 (2016). 54. R. Schurr, A. A. Mezer, Science 374, 762–767 (2021).
ware, which provides us with additional re- 15. D. S. Margulies et al., Proc. Natl. Acad. Sci. U.S.A. 113, 55. A. Alkemade et al., Sci. Adv. 8, eabj7892 (2022).
sources to sharpen our observations. These 12574–12579 (2016).
developments will include, for example, tools 16. S. A. Capurso et al., J. Neurosci. 17, 7372–7384 (1997). AC KNOWLED GME NTS
17. K. E. Stephan, T. Baldeweg, K. J. Friston, Biol. Psychiatry 59,
that can represent high-dimensional data in 929–939 (2006).
The authors thank the members of the Neurofunctional Imaging
Group (GIN), particularly V. Pacella, L. Talozzi, and E. Allard
the same latent space as anatomical informa- 18. E. Carrera, G. Tononi, Brain 137, 2408–2422 (2014).
but also L. Sakuma, T. Orset, P. Pouget, P. Bartolomeo, P. Nachev,
tion across scales (cells, voxels, circuits), across 19. J. Sharma, A. Angelucci, M. Sur, Nature 404, 841–847
L. Labache, C. Foulon, L. Verhagen, R. Mars, L. Pessoa, and
(2000).
species (macaque, chimpanzee, human), and 20. L. von Melchner, S. L. Pallas, M. Sur, Nature 404, 871–876
S. Sesack, and three anonymous reviewers for their helpful feedback
across imaging modalities (tracing, tractogra- and supporting material. Funding: This work was supported by
(2000).
the European Union’s Horizon 2020 research and innovation
phy, polarized light imaging), while account- 21. R. E. Passingham, K. E. Stephan, R. Kötter, Nat. Rev. Neurosci.
programme under European Research Council (ERC) Consolidator
3, 606–616 (2002).
ing for neurovariability at the individual level 22. A. T. Vu et al., Neuroimage 122, 318–331 (2015).
grant agreement no. 818521 (M.T.d.S., DISCONNECTOME);
to capture the factor interactions. To do so, University of Bordeaux’s IdEx “Investments for the Future”
23. M. Thiebaut de Schotten, C. Foulon, P. Nachev, Nat. Commun.
program RRI “IMPACT,” which received financial support from the
researchers ought to create professional net- 11, 5094 (2020).
French government; Marie Skłodowska-Curie grant agreement
24. V. R. Karolis, M. Corbetta, M. Thiebaut de Schotten, Nat. Commun.
works, integrate ideas, and share data openly. 10, 1417 (2019).
no. 101028551 (S.J.F., PERSONALISED); and the Donders Mohrmann
Together, these endeavors will push brain con- Fellowship (S.J.F., NEUROVARIABILITY). Competing interests:
25. Y. Assaf, A. Bouznach, O. Zomet, A. Marom, Y. Yovel, Nat. Neurosci.
The authors have no competing interests to declare. License
nectivity research to integrate across imaging 23, 805–808 (2020).
information: Copyright © 2022 the authors, some rights reserved;
26. L. Yang et al., J. Neurosci. 42, 3599–3610 (2022).
modalities (e.g., functional white matter), for- exclusive licensee American Association for the Advancement of
27. C. Sampaio-Baptista, H. Johansen-Berg, Neuron 96, 1239–1251
mulate new frameworks (e.g., neurovariability), Science. No claim to original US government works. https://www.
(2017).
science.org/about/science-licenses-journal-article-reuse
and foster our understanding of brain devel- 28. A. Horowitz et al., Brain Struct. Funct. 220, 1777–1788
opment and evolution (comparative neuro- (2015).
29. F. Balezeau et al., Nat. Neurosci. 23, 611–614 (2020).
imaging and neuroecology). This joint effort 30. R. B. Mars et al., eLife 7, e35237 (2018). Submitted 18 May 2022; accepted 12 August 2022
will push our current frontiers and lead to 31. M. Thiebaut de Schotten et al., Cereb. Cortex 24, 691–706 (2014). 10.1126/science.abq2591

RESEARCH
STRUCTURAL BIOLOGY
Closed complex for
protein degradation
The ability to target endoge-
nous proteins for degradation
with small molecules has
opened up avenues for treat-
IN S CIENCE JOU RNAL S ing a wide range of diseases.
Watson et al. investigated how
Edited by Michael Funk a class of these molecules
called CELMoDs (CRBN E3
ligase modulatory drugs) alter
the conformational landscape
of cereblon (CRBN), the
enzyme that recognizes and
marks substrates for deg-
radation. The authors found
that CELMoD compounds
trigger a conformational
rearrangement of CRBN from
an entirely open and defunct
form to an active closed form,
PLANT SCIENCE but only for a subset of CRBN
Floral sex determination proteins. Next-generation
CELMoDs, which are more
I
n melons, flowers initially
develop bisexually, but further effective therapies, appear
growth brings arrest of either to activate many more CRBN
carpel or stamen development, proteins by promoting closure
refining mature flowers into much more efficiently, and
male or female. Only the female future drug development
flowers produce melons. Zhang may be able to harness this
et al. have identified the genes allosteric effect to improve
involved in turning the bisexual efficacy. —MAF
primordial flower into either a male Science, add7574, this issue p. 549
or female flower. The zinc finger
transcription factor WIP1 interferes
with carpel development, allow- QUANTUM TECHNOLOGY
ing male flower development to A robust quantum
proceed. Conversely, expression of
an enzyme involved in producing
network node
the hormone ethylene, perhaps The ability to develop
supported by auxin signaling, sup- quantum networks and
ports and promotes female flower communicate quantum infor-
development. —PJH mation over long distances
requires quantum memory
Science, add4250, this issue p. 543
nodes with efficient optical
interfaces and long memory
A melon vine with male (top) and times. Because of their long
female (bottom) flowers, both of which coherence times and efficient
begin as bisexual buds optical interface, color centers
in diamond are promising
candidates to achieve this
goal. Stas et al. used silicon
SYNTHETIC BIOLOGY and innovation between exchange functional genetic vacancies in diamond and
diverse cells. Biologists have information with other organ- integrated the properties
Changing the genetic long made use of the common isms. This altered genetic into a single device (see the
code’s structure code for engineering microbes, code offers protection to Perspective by Gangloff). The
The genetic code, the rules by crops, and animals. Zürcher valuable engineered organ- authors demonstrate a quan-
PHOTO: ZHANG ET AL.
which genetic information is et al. permuted the rules isms from natural invaders tum memory with a lifetime
encoded in genes within DNA, by which genetic informa- and safeguards the natural exceeding 2 seconds and full
is nearly universal across all tion is encoded in DNA and world from engineered genetic optical control of the quan-
kingdoms of life, enabling the thereby created genetically material. —DJ tum states of the two-qubit
transfer of genetic information isolated organisms that cannot Science, add8943, this issue p. 516 register. With the capability of

RESEARC H | I N S C I E N C E J O U R NA L S
built-in error detection, such of transcripts encoding

a platform is promising for matrix-degrading enzymes
the development of scalable downstream of intracellular IN OTHER JOURNALS
quantum networks. —ISO signaling pathways that were
Science, add9771, this issue p. 557; distinct from those by which Edited by Caroline Ash
see also ade6964, p. 473 Sema4D regulates cell mor- and Jesse Smith
phology and motility. —AMV
Sci. Signal. 15, eabl5304 (2022).
PLASMA CELLS
Long-lived plasma cell VOLCANIC PLUMES
production Reaching new heights
High-affinity antibodies that Large, explosive volcanic erup-
provide durable protective tions can loft material such
immunity are produced by as ash, gases, and water all
long-lived bone marrow the way into the stratosphere,
plasma cells that differentiate with measurable impacts on
from germinal center (GC) atmospheric composition
B cells. However, the kinetics and climate. Proud et al. used
with which these plasma cells geostationary satellite images
occupy survival niches in the of the January 2022 Hunga-
bone marrow after immuni- Tonga volcano eruption, one
zation is poorly understood. of the largest eruptions ever
Robinson et al. constructed a recorded, to show that its
genetic time-stamping mouse volcanic cloud reached an
model to mark plasma cells altitude of 57 kilometers, well
generated at specific stages past the stratosphere and into
of an immune response, which the mesosphere and higher
enabled them to track the fate than any volcanic plume previ-
of plasma cells over time. The ously recorded. This is the first
formation of new long-lived time a plume has been seen BIOGEOGRAPHY
plasma cells started early in to penetrate the stratopause.
the GC response after immu- —HJS
Imperial affiliations for alien flora
O
nization and progressed at a Science, abo4076, this issue p. 554 ne legacy of empires is the alien plant life that was
linear rate. These findings intentionally spread around the world by colonists, often
suggest that vaccine formu- for economic reasons. Some of these alien species have
lations that can extend the MONKEYPOX since become damagingly invasive, and floristic homog-
duration of the GC response enization and ecosystem degradation have become
will boost the number of long-
Deaminase drives widespread. Lenzner et al. used zeta diversity to capture the
lived antigen-specific plasma viral evolution contribution of rare and widespread species to compositional
cells generated after immuni- Monkeypox cases have similarity and turnover of a flora. Species turnover was deter-
zation. —IRW occurred sporadically around mined largely by climate and modified by the length of time of
Sci. Immunol. 7, eabm8389 (2022). the world for several decades. colonization. Given its size and relatively recent expansion, the
However, May 2022 saw a British empire has had the greatest influence on alien invaders.
surge of reports of monkey- Despite increased global plant trade since World War II, the
PHYSIOLOGY pox infections outside of specific signatures of pre-war colonial power in a region can still
endemic regions and concen- be traced in the composition of its alien flora. —CA
Semaphorin 4D as trated in the community of Nat. Ecol. Evol. 10.1038/s41559-022-01865-1 (2022).
cartilage destroyer men who have sex with men.
Various proinflammatory cyto- Fortunately, modifications Captain James Cook’s ship Endeavour (a modern replica of the barque is
kines contribute to cartilage of the smallpox vaccine are shown) carried the botanist Joseph Banks, who was a towering figure in
destruction in joint diseases. effective against monkeypox. plant collection and dispersal around the British empire.
Murakami et al. found that Gigante et al. identified two lin-
the cytokine Semaphorin 4D eages occurring in the United
PHOTO: PATRICK EDEN/ALAMY STOCK PHOTO

(Sema4D) was released by States showing evidence of
inflamed macrophages and the antiviral activity of host EVOLUTION and almonds. Amygdalin is a
stimulated cartilage matrix apolipoprotein B editing cyanogenic glycoside that is
degradation in articular complex (APOBEC3) cytosine
Why are worms not afraid poisonous to many animals,
chondrocytes and cartilage deaminase. APOBEC3 editing of cyanide? but bacteria and plants have
explants. In a mouse model of is unusual for poxviruses, but Amygdalin, the Greek word for enzymes that can detoxify
inflammatory arthritis, loss of it looks to be driving recent “almond,” is a natural chemi- cyanide. The nematode
Sema4D protected against car- monkeypox virus evolution. cal compound found in many Caenorhabditis elegans, a
tilage degeneration. Sema4D —CA plants, including in the seeds tiny roundworm that, when it
stimulated the expression Science, add4153, this issue p. 560 of apples, peaches, cherries, is not being used as a model

organism, normally lives as the long-term partner, the multiplexed QKD receiver that of Black versus white drivers
in temperate soils and is lower the CRP levels recorded. can achieve a key distribution was most biased. Among those
able to tolerate amygdalin. Simply spending time together of more than 12 megabits per officers, effects were exacer-
Interestingly, this worm seems appears to be a good expla- second. Readily scalable by add- bated when rallies mentioned
to have acquired resistance to nation for how high-quality ing more wavelength channels, racial issues. —BW
the toxin by horizontal trans- relationships sustain physical the architecture presents a viable Q. J. Econ. 10.1093/
fer of two genes, one of which health among adult humans by route to developing secure, high- qje/qjac037 (2022).
encodes a detoxifying enzyme, reducing systemic inflamma- rate transmission of information
from green algae. In nature, C. tion. —MM across communication networks.
FERROELECTRICS
elegans lives within rotten fruit, Brain Behav. Immun. —ISO
where it likely encounters plant- 107, 132 (2022). Optica 9, 1121 (2022). The best way to relax
derived cyanide. —DJ Relaxor ferroelectrics have
Curr. Biol. 10.1016/ both a polarization and a strain
j.cub.2022.09.041 (2022). QUANTUM KEYS SOCIAL SCIENCE response to an external electric
field, making them of great
Chip-based quantum key Political rallies can bias interest for various applica-
SOCIAL RELATIONSHIPS distribution policing tions. However, the origin of this
Stay well together The flow of information between Campaign rallies by Donald effect has been challenging to
Inflammation is thought to be sender and receiver is typically Trump influenced law enforce- fully resolve. Kim et al. looked at
an important neurobiological encrypted with secret keys to ment behavior toward racial the evolution of nanoscale polar
mediator between romantic ensure privacy. With the develop- minorities for up to 2 months nanodomains under varying
emotions and physical vigor, ment of quantum computers, afterward. Grosjean et al. used epitaxial strains using diffused
but this connection is contro- however, it has been shown that data from 35 million US traffic x-ray scattering, microdiffrac-
versial. Jolink et al. measured classical encryption methods are stops from 2015 to 2017, includ- tion, electron microscopy, and
blood levels of an inflamma- susceptible to hacking. Quantum ing in 141 counties that hosted molecular dynamics simula-
tory marker, C-reactive protein key distribution (QKD) is a prov- rallies. In these counties, there tions. The authors found that
(CRP), in romantically attached ably robust method for ensuring was a nearly 6% post-rally structural transition between
couples. The authors confirmed security against such quantum increase in the probability that polar nanodomain configura-
that just being in the presence hackers, but the generation and a stopped driver was Black. This tions not only forms the basis
of the other partner affected transmission of the quantum was not due to any changes in for rotation of polarization but
systemic inflammation. Indeed, keys have been impractically the behavior or racial composi- also for large electromechanical
the more time (measured in slow. Beutal et al. present a tion of drivers. Effects were coupling. —BG
minutes per 24 hours) spent chip-based, fully integrated most pronounced among Nat. Phys. 10.1038/
within the same physical space four-channel wavelength-division officers whose prior treatment s41567-022-01773-y (2022).
PLANT SCIENCE
Adaptive diversity
I
nsight into local adaptation of long-lived trees, which are often foundational
species in their habitats, can aid restructuring of forest populations that
have been overtaken by climate change. Capblancq et al. studied the inter-
action between distribution and genomic adaptation for the red spruce
(Picea rubens), which is found from Nova Scotia to North Carolina, areas
that include a range of precipitation, moisture, and temperature extremes.
PHOTO: BLICKWINKEL/ALAMY STOCK PHOTO
The authors found that red spruce seedlings grew more poorly the farther
they were transplanted from the site and environmental conditions local
to their mother tree because of genetic selection for local conditions. —PJH
New Phytol. 10.1111/nph.18465 (2022).
Red spruce grows along a wide latitudinal gradient in the eastern United States,
across which it displays genomic adaptation for local conditions.

RESEARC H
ALSO IN SCIENCE JOURNALS Edited by Michael Funk
CELLULAR NEUROSCIENCE first comprehensive studies of and Mack). They project that addition of the recombinant
the real-time, millisecond-scale near-term climatic warming will MANF protein improved heart
Astrocyte diversity and interactions between neurons of cause an exponential increase in function. Treating cardio-
morphology distinct genetic classes. —PRS burned area in Arctic carbon-rich myocytes derived from human
Astrocytes are the major type of Science, abm8797, this issue p. 515 soils before mid-century. —HJS induced pluripotent stem cells
glia, displaying complex bushy Science, abn9768, this issue p. 532; with estrogen or estrogen recep-
morphologies as their defin- see also ade9583, p. 470 tor b agonist induced MANF
ing feature. Endo et al. studied NEUROSCIENCE and HSPA5 expression. Treating
astrocyte similarities, diversity, female tumor-bearing mice with
and morphology across the
Selectively silencing NEUTRINO ASTROPHYSICS estrogen receptor b agonist
mouse central nervous system abnormal neurons Nearby active galaxy during ICI treatment reduced
(CNS) (see the Perspective Nearly one-third of epilepsy T cell infiltration and preserved
by Baldwin). They identified patients do not respond to cur- emits neutrinos heart function. Hormone therapy
gene networks related to rently available anti-epileptic Observations have shown a dif- could thus potentially limit ICI
astrocyte morphology between drugs. Several gene therapy fuse background of high-energy myocarditis by promoting the
CNS regions, several of which approaches have been sug- neutrinos, which is known protective effects of MANF. —CC
unexpectedly included genes gested, but these methods tend to be of extragalactic origin. Sci. Transl. Med. 14, eabo1981 (2022).
related to risk of Alzheimer’s to indiscriminately target all However, it has been difficult to
disease (AD). When expression neurons in a given brain region. identify individual sources that
of these genes was reduced in Qiu et al. developed a gene contribute to this background. ANIMAL BEHAVIOR
mice, astrocyte morphological therapy strategy that self-selects The IceCube Collaboration
complexity was diminished, and neurons that are pathologically reanalyzed the arrival direc-
Complex cognition in
mice exhibited impairments in a overactive and down-regulates tions of astrophysical neutrinos songbirds
cognitive test. Remarkably, the their excitability in a closed loop and then searched for point A cornerstone of symbolic cogni-
same genes are down-regulated (see the Perspective by Staley). sources (see the Perspective tion is recursion, the capacity to
in both a mouse model of AD and They put the KCNA1 gene, which by Murase). They identified embed structures within similar
in human AD and several other encodes a potassium chan- evidence for neutrino emission structures. This capacity has
CNS disorders. These findings nel, under the control of an from NGC 1068 (also known only been identified in primates,
suggest that loss of astrocyte immediate early gene promoter as Messier 77), a nearby active but Liao et al. found that carrion
morphology may be therapeuti- with activity that is switched galaxy. Its properties are quite crows (Corvus corone), a song-
cally targetable in diverse CNS on by intense neuronal firing. different from TXS 0506+056, bird family that demonstrates
disorders. —SMH The treatment switches itself which was found to be a neutrino complex cognition, can parse,
Science, adc9020, this issue p. 514; off once brain circuit activity source in 2018, leading the inves- learn, and generate recursive
see also ade9249, p. 475 has returned to baseline. This tigators to suggest that there sequences of visual stimuli. The
approach could in principle be might be more than one popula- birds reached a level of perfor-
used to treat any neuropsy- tion contributing to the neutrino mance comparable to that of
NEUROSCIENCE chiatric disorder in which only background. —KTS 3- to 4-year-old human children
a subpopulation of neurons is Science, abg3395, this issue p.538; and superior to that of nonhu-
Real-time voltage imaging pathologically overactive. —PRS see also ade4190, p. 474 man primates after only a few
of neurons Science, abq6656, this issue p. 523; learning trials. Given the parallels
How different subpopulations see also ade8836, p. 471 between songbird communica-
of neurons coordinate their MYOCARDITIS tion and human language, these
activity patterns in real time to findings shed light on the fun-
control network output is still
ARCTIC WILDFIRES Of myocarditis and damental cognitive abilities that
not fully understood. This is in Getting burned women underlie complex communica-
part because of our inability to Global warming is exacerbat- Inflammation of the heart tion between conspecifics. —AF
simultaneously measure the ing the conditions that cause muscle, or myocarditis, is an Sci. Adv. 10.1126/
voltage dynamics within and wildfires in many regions, includ- off-target effect of anticancer sciadv.abq3356 (2022).
across targeted cell ensembles. ing the Arctic, where extensive treatments such as immune
Kannan et al. developed a suite peatlands hold large amounts of checkpoint inhibitor (ICI) ther-
of mutually compatible, geneti- carbon. However, is the extent apy. Studying sex differences
cally encoded voltage indicators of wildfires there increasing as in ICI myocarditis, Zhang et al.
for high-speed voltage imaging would be expected given the observed greater T cell infiltra-
of multiple genetically identi- changing conditions? Descals et tion and cardiac dysfunction in
fied neurons in awake behaving al. found that during the summer female mice with this disorder
animals. These indicators can be of 2020, which was the warm- and identified down-regulation
uniquely combined for simul- est in four decades, Arctic fires of two genes, MANF and HSPA5,
taneous recordings of spiking burned an unprecedentedly in the heart. In a mouse model,
by ensembles of neurons. This large area of carbon-rich soils cardiac depletion of Manf wors-
technical advance enables the (see the Perspective by Post ened ICI myocarditis, whereas
513-B 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE

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

Learn more at Science.org/journals
RES EARCH
◥ We mined our data to identify gene networks

RESEARCH ARTICLE SUMMARY related to astrocyte morphology, and unexpect-
edly discovered several AD risk genes. We de-
CELLULAR NEUROSCIENCE veloped astrocyte-specific CRISPR/Cas9–based
gene knockdown to reduce the expression of
Molecular basis of astrocyte diversity and such genes in the hippocampus. We found a re-
duction of astrocyte morphology with changes
morphology across the CNS in health and disease in a cognitive task after knockdown of key
morphology-related genes, indicating that as-
Fumito Endo, Atsushi Kasai, Joselyn S. Soto, Xinzhu Yu, Zhe Qu, Hitoshi Hashimoto, trocyte morphological changes affect neural cir-
Viviana Gradinaru, Riki Kawaguchi, Baljit S. Khakh* cuit function. In addition to AD risk genes that
were astrocyte morphology related, we found
a significant association between morphology-
INTRODUCTION: The central nervous system gene expression, single-cell gene expression, related genes and those related to several other
(CNS) comprises a large population of non- astrocyte morphology, and interrelationships common CNS disorders.
neuronal cells called glia. The predominant between gene expression and morphology
type of glia are astrocytes, which were discov- using bioinformatic methods across the CNS. CONCLUSION: We provide comprehensive mo-
ered ~140 years ago. Astrocytes tile the entire The data were mined to identify genes and lecular data that will allow many new types of
CNS, serve critical homeostatic functions, and pathways related to astrocyte morphological experiments to explore core features of astro-
display complex, “bushy” morphologies—their complexity, which we explored experimentally cytes across the CNS, in particular, those unique
defining feature. Unlike CNS neurons, which using gene knockdowns. We then compared to specific regions and how they relate to the
are highly diverse, astrocytes have historically these morphology-related genes with astro- dynamics and biophysics of neural circuits
been considered largely homogenous, serving cytic differentially expressed genes from an in specific CNS regions. We used our data to
as a type of omnipresent glue that is equivalent Alzheimer’s disease (AD) mouse model dis- identify the molecular underpinnings of astro-
between CNS areas. Although recent studies playing reduced astrocyte morphology, as well cyte morphology between CNS regions and to
have challenged this belief, there has been no as in relation to gene expression in AD and show that reduced astrocyte morphological
broad assessment of astrocyte diversity, sim- other CNS disorders. complexity, and the attendant loss of tissue
ilarity, or morphology across the CNS of any support, is common to diverse CNS disorders.
species. An important goal, therefore, is to RESULTS: We found several hundred genes This raises the prospect that restoring astrocyte
rigorously understand the molecular sim- that were enriched within and shared among morphology, and thus the ensuing neuronal
ilarities and differences between astrocytes astrocytes in the CNS, with functions related and tissue support functions, may be ther-
in the CNS, to determine how they affect as- to metabolism, cholesterol, and neurotrans- apeutically beneficial in disease. Our findings
trocyte morphology, and to determine how mitter uptake and biosynthesis. These genes provide the molecular basis to explore such
these properties relate to normal and dis- represent the core functions of astrocytes, but strategies and to unravel the nascent under-
ease conditions. little is known about approximately a third of lying biology determining how astrocytes con-
them, implying that understanding the fun- tribute to CNS function and dysfunction.
RATIONALE: It is critical to understand all cell damental physiology of astrocytes across CNS
▪
types of the CNS in detail as part of our quest areas is an important ongoing experimen-
to explore fundamental biology and to develop tal goal. Our data also reveal that astrocytes The list of author affiliations is available in the full article online.
*Corresponding author. Email: [email protected]
new therapeutic strategies to treat CNS dis- have molecular features and functions spe-
Cite this article as F. Endo et al., Science 378, eadc9020
orders. We studied astrocyte regional diversity cific to CNS regions, and that these region- (2022). DOI: 10.1126/science.adc9020
and major signaling pathways using whole- specific functions arise from the local tissue
brain imaging, astrocyte-neuron density, marker environment and by variable representation READ THE FULL ARTICLE AT
expression, astrocyte-specific and bulk tissue of seven distinct astrocyte subclusters. https://doi.org/10.1126/science.adc9020
1 Astrocyte morphology and 2 Molecular signatures 3 Morphology-related 4 Astrocyte genes related

diversity across the CNS and origins of diversity gene networks to CNS diseases
Branches
Soma
Territory
Whole brain imaging, morphology, Astrocyte RNA-seq, bulk Bioinformatic analyses Hypothesis testing in
and marker expression RNA-seq, scRNA-seq mice with CRISPR-Cas9
Astrocyte diversity and morphology across the CNS. (1) Astrocyte morphology and gene expression was measured across CNS regions. (2) Gene expression
data revealed the molecular signatures of astrocyte diversity and its origins. (3) Gene expression and morphological data identified gene networks related to complex
morphology. Causal roles were explored. (4) Several astrocyte morphology genes were Alzheimer’s disease risk genes, a disorder in which astrocytes displayed
reduced morphology.

RES EARCH
◥ morphology to gene expression and astrocyte

RESEARCH ARTICLE morphological changes in an Alzheimer’s dis-
ease (AD) mouse model and to human CNS dis-
CELLULAR NEUROSCIENCE orders, including AD.
Molecular basis of astrocyte diversity and Cellular and anatomical evaluation of astrocytes
We mapped astrocyte density across 13 CNS
morphology across the CNS in health and disease regions by using whole-brain imaging by block-
face serial microscopy tomography (FAST) (19)
Fumito Endo1, Atsushi Kasai2, Joselyn S. Soto1, Xinzhu Yu1†, Zhe Qu3, Hitoshi Hashimoto2,4,5,6,7, in mice expressing tdTomato in astrocytes (20)
Viviana Gradinaru3, Riki Kawaguchi8, Baljit S. Khakh1,9* (from Aldh1l1 locus; Fig. 1B). We observed
tdTomato+ astrocytes throughout the brain
Astrocytes, a type of glia, are abundant and morphologically complex cells. Here, we report astrocyte molecular (movie S1) and spinal cord, with densities that
profiles, diversity, and morphology across the mouse central nervous system (CNS). We identified shared varied between regions but overlapped with
and region-specific astrocytic genes and functions and explored the cellular origins of their regional diversity. Sox9 (21) by 95 ± 2% (Fig. 1C and figs. S1 and
We identified gene networks correlated with astrocyte morphology, several of which unexpectedly contained S2). Although Aldh1l1 and Sox9 are reliable
Alzheimer’s disease (AD) risk genes. CRISPR/Cas9–mediated reduction of candidate genes reduced astrocyte markers of astrocytes (20, 21), no marker labels
morphological complexity and resulted in cognitive deficits. The same genes were down-regulated in human all astrocytes, and 100% overlap was therefore
AD, in an AD mouse model that displayed reduced astrocyte morphology, and in other human brain disorders. not expected. We quantified tdTomato+ astro-
We thus provide comprehensive molecular data on astrocyte diversity and mechanisms across the CNS and on cyte density and found significant but modest
the molecular basis of astrocyte morphology in health and disease. twofold differences between areas (Fig. 1, C
and D; N = 3 to 6 mice; all regions in fig. S1).
Next, we labeled neurons and most astrocytes
o explore how the brain performs its Unlike neurons, which are extremely di- by immunohistochemistry (IHC) for S100b;
T
functions, it is critical to understand its verse within neural circuits, astrocytes have and glial fibrillary acidic protein (GFAP) in
diverse cell populations in detail, as arti- traditionally been viewed as being homoge- mice expressing GCaMP6f in astrocytes (fig.
culated by the BRAIN Initiative (1, 2). Fur- neous. However, recent gene expression and S3). The astrocyte-neuron ratios varied almost
thermore, it is important to explore both functional studies using different methods in 10-fold, being highest in the ventral spinal
neuronal and non-neuronal cells such as glia the context of health and brain tumors have cord and lowest in the cerebellum, where our
(1–3). shown that astrocytes are separable between evaluations included Bergmann glia (Fig. 1E;
Astrocytes, the most populous glial cells, several central nervous system (CNS) regions N = 8 mice). These differences were not driven
were documented at the same time as neu- (9–18). This has led to the realization that by major variation in astrocytes but by differ-
rons, but for a long time their role was rele- astrocytes may be diverse, an attribute that, ences in neurons (Fig. 1, F and G; N = 8 mice).
gated to a type of “glue” that held the brain together with their highly complex morphol- There was no one-to-one scaling between neu-
together (4). With recent technical advances ogy, may allow them to mediate their multiple ron and astrocyte density across regions, al-
(5), these fascinating cells are emerging as es- roles in different parts of the CNS. However, though the two parameters were correlated
sential contributors of brain physiology, ani- there has been no broad molecular assessment (r = 0.495, P = 0.118; Fig. 1H). These data show
mal behavior, and disease (6, 7). Astrocytes are of astrocyte diversity, similarity, or morphol- that the density of astrocytes per region did
ubiquitous, morphologically complex, bushy ogy across the CNS of any species. not scale simply with neuron density, but that
cells that make extensive contacts with other astrocytes tiled the CNS with modest varia-
brain cells. They serve diverse roles, including Results tion, likely reflecting homeostatic functions.
ion and neurotransmitter homeostasis, syn- Strategy Large variations of the astrocyte-neuron ratios
apse formation and removal, synaptic mod- We studied astrocytes from 13 regions of the between CNS regions were driven by neurons.
ulation, and contributions to neurovascular adult mouse CNS (Fig. 1A): the olfactory bulb Because astrocytes form part of the neurovas-
coupling and the blood–brain barrier. Astro- (OB), motor cortex (MCX), somatosensory cor- cular unit, we also mapped astrocyte-endothelial
cytes are widely implicated in neurological tex (SCX), visual cortex (VCX), hippocampus cell density and ratios across CNS regions
and psychiatric diseases (7, 8). (HIP), striatum (STR), thalamus (TH), hypo- (fig. S4, A to D). We found that endothelial
thalamus (HY), cerebellum (CB), midbrain (MB), cell and astrocyte density were strongly cor-
1
Department of Physiology, University of California Los Angeles, hindbrain (HB), ventral spinal cord (VSC), and related (fig. S4E), reflecting critical neurovas-
Los Angeles, CA, USA. 2Laboratory of Molecular
Neuropharmacology, Graduate School of Pharmaceutical dorsal spinal cord (DSC). First, we assessed cular unit biology.
Sciences, Osaka University, Suita, Osaka, Japan. 3Division of astrocyte-neuron density and astrocyte marker We performed IHC for GFAP, S100b, and
Biology and Biological Engineering, California Institute of expression in different CNS regions. Next, we GCaMP6f (from the Aldh1ll locus) to deter-
Technology, Pasadena, CA, USA. 4Molecular Research Center for
Children’s Mental Development, United Graduate School of
assessed molecular profiles and signaling path- mine whether these markers were equivalent
Child Development, Osaka University, Suita, Osaka, Japan. ways using astrocyte-specific RNA sequencing or dissimilar between CNS regions (20). There
5
Division of Bioscience, Institute for Datability Science, Osaka (RNA-seq), bulk tissue RNA-seq, and single- was an ~500-fold variation between CNS re-
University, Suita, Osaka, Japan. 6Open and Transdisciplinary
Research Initiatives, Osaka University, Suita, Osaka, Japan.
cell RNA-seq (scRNA-seq). Additionally, we gions for these established markers (fig. S5;
7
Department of Molecular Pharmaceutical Science, Graduate documented the morphology of astrocytes and N = 8 mice; movie S2). Although the density
School of Medicine, Osaka University, Suita, Osaka, Japan.
8
used weighted gene coexpression network of astrocytes varied little (Fig. 1), quantitative
Center for Neurobehavioral Genetics, Semel Institute for
analyses (WGCNA) to identify gene networks IHC experiments (fig. S5) provided evidence
Neuroscience and Human Behavior, University of California Los
Angeles, Los Angeles, CA, USA. 9Department of Neurobiology, correlated with morphological complexity. We for the diversity between CNS regions.
University of California Los Angeles, Los Angeles, CA, USA. also tested the consequences of reducing the
*Corresponding author. Email: [email protected] expression of genes predicted to affect astro- Shared CNS-wide astrocyte genes and pathways
†Present address: Department of Molecular and Integrative
Physiology, Beckman Institute, University of Illinois at cyte morphology using CRISPR/Cas9. Then, To evaluate astrocyte gene expression, we
Urbana-Champaign, Urbana, IL 61801, USA. we related molecular signatures of astrocyte performed astrocyte-specific RNA-seq (9).
Endo et al., Science 378, eadc9020 (2022) 4 November 2022 1 of 12

RES EARCH | R E S E A R C H A R T I C L E
Fig. 1. Astrocyte density and astrocyte-to-neuron ratios across the (Aldh1l1 tdTomato) mouse brains imaged by FAST. Magnified images show
mouse CNS. (A) Experimental design for investigating astrocytes in 13 CNS tdTomato+ astrocytes in the MCX, HIP, and STR; the other 10 regions are
regions. CB, cerebellum; DSC, dorsal spinal cord; HB, hindbrain; HIP, shown in fig. S1. The color images show exemplar Sox9- and Aldh1l1-driven
hippocampus; HY, hypothalamus; MB, midbrain; MCX, motor cortex; OB, tdTomato images (see fig. S2). (D) Astrocyte density measurements by
olfactory bulb; SCX, somatosensory cortex; STR, striatum; TH, thalamus; VCX, FAST in 13 CNS regions (N = 6 mice for spinal cord; N = 3 mice for brain).
visual cortex; VSC, ventral spinal cord. (B) Illustration and representative (E) Plot showing the astrocyte-to-neuron ratio in 13 CNS regions (N = 8
images of whole-brain astrocyte imaging by FAST. A, anterior; L, lateral; mice). (F and G) Plot of astrocyte and neuron numbers in 13 CNS regions.
V, vertical. The brains were from Aldh1l1-Cre/ERT2 x Ai14 (Aldh1l1 tdTomato) (H) Correlation between astrocyte and neuron numbers. Data are shown as
mice. (C) Representative coronal sections of Aldh1l1-Cre/ERT2 x Ai14 means ± SEM.

Whole-tissue (input) and astrocyte RNA (immu- Region-specific astrocyte genes and pathways they shared the same anatomical area (Fig. 2,
noprecipitated) was extracted from 13 regions We analyzed astrocyte differentially expressed E and F) and are widely studied in the field.
(Fig. 2A; N = 3 to 5 mice). Multidimensional genes (DEGs) in each CNS area relative to the We asked several questions. First, do subclus-
scaling showed separation among input, im- other 12 regions and identified 3500 that were ters of astrocytes exist within these regions?
munoprecipitation, and CNS areas (Fig. 2B). region enriched (Fig. 2E). Hierarchical cluster- Next, do different combinations of these sub-
Immunoprecipitated fractions were enriched ing of these region-enriched DEGs revealed clusters contribute to the regional nature of
with astrocyte markers (fig. S6A). a trend consistent with anatomical proximity astrocytes? Moreover, are regional differences
We identified 4314 astrocyte-enriched genes (Fig. 2F). Thus, astrocytes from the cerebrum between astrocytes related to a metric of the
[relative to input; log2 ratio > 1; false discovery (OB, MSC, SCX, VCX, HIP, and STR) were local tissue in which they reside?
rate (FDR) < 0.05, fragments per kilobase of tran- clustered distinctly from broad regions of the We analyzed scRNA-seq data of 89,553 cells
script per million mapped reads (FPKM) > 1], brain stem and spinal cord (TH, HY, MB, VSC, from the cortex, hippocampus, and striatum,
and of these, 825 (~20%) were common across and DSC) and the cerebellum (Fig. 2, E and F). (Fig. 3, A and B, and fig. S14A; N = 13 mice).
the 13 regions and called “shared” (fig. S6, B to The number of region-enriched genes varied Astrocytes accounted for ~15% (fig. S14B),
D). A total of 939 genes (~22%) were “unique” (fig. S9A), but regional differences were mainly which is consistent with expectations (4).
to a particular region, and the remaining 2588 driven by up-regulated genes (fig. S9A). Over- However, this percentage should be treated
genes (~60%) were “partially shared” in two lap of the top 100 astrocyte region–enriched cautiously because it results from tissue dis-
to 12 regions (fig. S6, B to D). To explore the DEGs is illustrated in fig. S9B for the cere- sociation, but the percentage of Sox9 + and
pathways and signaling mechanisms shared brum, brain stem and spinal cord, and cere- tdTomato+ astrocytes relative to 4′,6-diamidino-
by astrocytes across all regions, we assessed bellum. Clustering for the bulk tissue–specific 2-phenylindole–positive (DAPI+) cells (e.g., from
the molecules represented by the 825 genes. DEGs, with or without astrocyte-enriched fig. S2) in the cortex, hippocampus, and stria-
About half were related to enzymatic and transgenes for the 13 CNS regions, showed a similar tum was 10 ± 1, 11 ± 1, and 13 ± 1%, respectively
porter activity or transcriptional regulation pattern as astrocyte DEGs (fig. S9, C to E), (N = 3). Further analysis showed that these
(fig. S6E). Figure 2C shows the top 50 astrocyte- suggesting that the regional identity of as- CNS regions shared seven astrocyte subclusters
enriched genes shared across 13 regions in trocytes reflects the region-specific micro- (AST1 to AST7; Fig. 3C) representative of the
order of enrichment, with the color scale repre- environment (as reflected in the bulk RNA-seq cortex, hippocampus, and striatum (Fig. 3D).
senting abundance and annotations illustrat- data), as well as astrocyte lineage–related gene Genes known to be enriched in cortical (Chrdl1),
ing the major signaling mechanisms. Recalling expression (as reflected in the immunopreci- hippocampal (Gfap), and striatal (Crym) as-
IHC (Fig. 1 and fig. S2), Aldh1l1 and Sox9 were pitated RNA-seq data). trocytes showed high expression within their
abundant across regions (Fig. 1C) and known Because astrocytes from 13 CNS regions cognate region (Fig. 3E). By analyzing the sub-
astrocyte genes (e.g., Kcnj10, Slc1a2, and Apoe) clustered into three broad anatomical classes, clusters, we identified the top 100 genes en-
were among the top 50 highly expressed (Fig. we identified the top 10 marker genes from the riched within each (Fig. 3F). Furthermore, the
2C). Unbiased analyses for the 825 shared cerebrum, brain stem and spinal cord, and seven subclusters represented different pro-
astrocyte-enriched genes showed that the top cerebellum (Fig. 2G). Furthermore, we iden- portions of astrocytes per brain region, im-
pathways were related to neurotransmitter ho- tified region-specific astrocyte marker genes plying that regional diversity may be cast
meostasis, cholesterol biosynthesis, and glucose for each of the 13 CNS areas (Fig. 2H) and partly from variable representation of the sub-
metabolism (Fig. 2D). classified the top 20 on the basis of their func- clusters (Fig. 3G). For example, subcluster
We performed several analyses for astro- tions (fig. S10). Some of the genes we identified AST2 was dominant in the cortex, subcluster
cytes across 13 regions (figs. S7 and S8). For K+ as being regionally enriched within astrocytes AST1 was dominant in the hippocampus, and
homeostasis, the most highly expressed genes (e.g., Chrdl1 and Crym) were known (9, 14, 22), subclusters AST5 and AST6 were dominant in
were Kcnj10, Kcnj16, and Atp1a2, encoding K+ but many were new. We validated expression the striatum (Fig. 3G). Upstream analysis for
channels and Na+/K+ ATPase (fig. S7A). Astro- of the genes for which reliable antibodies were the seven subclusters revealed differences and
cytes highly expressed genes related to neuro- available: IHC for proteins encoded by Sphk1, similarities in the top five upstream transcrip-
transmitter transport and metabolism but Ccnd1, Crym, and Sept4 confirmed regional lo- tional regulators (fig. S14C). We also found
they lacked or expressed at very low levels the cation at the protein level (Fig. 2, I to L; N = 4 upstream regulators related to external cues
genes related to Ca2+-dependent vesicular re- mice). Additional analyses for the astrocyte such as neurotransmitters [e.g., γ-aminobutyric
lease (fig. S7, B to E). Shared functional pro- region–enriched genes across the 13 CNS re- acid (GABA)], growth factors [e.g., epider-
perties of astrocytes may be cell-autonomously gions revealed differences and similarities in mal growth factor (EGF)], and cytokines [e.g.,
regulated because >50% of all upstream regu- transcription factors that may partly explain interleukin-9 (IL-9); fig. S14D], implying that
lators were categorized as transcriptional reg- the interregional differences of astrocytes (fig. astrocyte regional differences were related to
ulators and ligand-dependent nuclear receptors S11A). Gene ontology analysis for the region- tissue-dependent profiles within the local mi-
(fig. S8, A and B). The top 20 genes across enriched DEGs identified putative astrocyte croenvironment. We assessed functional fea-
13 regions reflecting functional classes of mo- region–specific functions (fig. S11B). We also tures of the astrocyte subclusters (fig. S14E)
lecules such as enzymes, transporters, kinases, assessed genes related to astrocyte functions and performed candidate gene analyses (fig.
receptors, peptidases, and ion channels are such as phagocytosis, synapses, reactivity (fig. S15). Genes related to glucose homeostasis
provided in fig. S8, C to J. For example, the G S12), and Ca2+ signaling (fig. S13), noting dif- were up-regulated in AST2, AST4, AST5, and
protein–coupled receptor (GPCR) genes Gpr37l1, ferences between regions for transmembrane AST6 (fig. S15A). Related to K+ homeostasis,
S1pr1, Ntsr2, Ednrb, Smo, Adora2b, Olfr287, Ca2+ flux pathways and GPCR signaling (fig. Kcnj10 was up-regulated in AST2, whereas
Gpr146, Agtrap, Fzd1, Fzd9, and Npr2 were S13, E and F) that foreshadow functional studies. Atp1a2 and Atp1b2 were up-regulated in AST1
expressed in astrocytes across the CNS. These (fig. S15B). Subcluster-specific enrichment was
evaluations of shared genes and pathways re- Bioinformatic evaluation of astrocyte diversity observed for genes related to cholesterol ho-
present core features of CNS astrocytes (Fig. 2, To explore how differences between astrocytes meostasis, neurotransmitter homeostasis, do-
C and D, and table S1). Little is known about from different CNS regions may arise, we per- pamine metabolism, synapse function, and
the physiology of many astrocyte genes within formed scRNA-seq for the cortex, hippocam- phagocytosis (fig. S15), indicating that combi-
the top 50 (Fig. 2C). pus, and striatum, which we chose because nations of subclusters between CNS regions

Fig. 2. Shared and unique

astrocyte CNS molecular
signatures and mechanisms.
(A) Approach for astrocyte-
specific RNA-seq. (B) Multi-
dimensional scaling plot
for input and immunoprecipi-
tated (N = 3 to 5 mice)
RNA-seq data. (C) Heat map
showing the log2 FPKM values
of the top 50 most enriched
astrocyte genes shared in
the 13 CNS regions. The bar
graph shows the average ratio
of immunoprecipitation versus
input. Colored arrowheads
indicate functional class.
(D) Bar graph showing the top
25 ingenuity pathway analysis
(IPA) canonical pathways
for the 825 shared astrocyte-
enriched genes with a
threshold of P value < 0.05.
The colored arrowheads
indicate functional class.
(E) Heat map showing
3500 astrocyte region–
enriched DEGs in 13 CNS
regions with a cluster
dendrogram. Astrocyte
region–specific DEGs were
defined by comparing
immunoprecipitation in
one CNS region with the
average of immunoprecipi-
tated samples from the other
12 regions with a threshold
of log2 ratio > 1, FDR < 0.05,
and FPKM > 1. From the
dendrogram, astrocytes could
be classified into three broad
groups. Genes defining
OB, MCX, SCX, HIP, and STR
(cerebrum) are outlined in
green (“a”). Genes highly
expressed within and defining
TH, HY, MB, HB, VSC,
and DSC (brain stem or spinal
cord) are outlined in pink
(“b”). Genes highly expressed
within and defining the CB are
outlined in orange (“c”) and
include Bergmann glia genes.
(F) Schematic of anatomical grouping of astrocytes in 13 CNS regions based on RNA-seq. (G) Heat map showing the log2 ratio of immunoprecipitation versus the
averaged immunoprecipitation from the other 12 regions of the top 10 astrocyte-specific DEGs for anatomically relevant CNS regions. (H) Heat map showing the top
region-specific astrocyte genes from RNA-seq. (I to L) Representative (N = 4 mice) IHC images for Sphk (I), Ccnd1 (J), Crym (K), and Sept4 (L), which reproduced regional
expression in OB, MCX, STR, and CB, respectively.
imparted separable functional features to as- necrosis factor (TNF), and IL-5; fig. S14D] Fig. 2, astrocyte RNA-seq data for the 13 CNS
trocytes in different CNS regions. suggested that regional astrocyte diversity regions could be hierarchically clustered into
The existence of several extracellular species may reflect the tissue microenvironment, three broad anatomical regions. This trend
in the top 10 upstream regulators for the astro- which we explored using astrocyte-specific was also observed when we compared the per-
cyte subclusters [e.g., GABA, IL-4, EGF, tumor and bulk RNA-seq data (Fig. 2). As shown in centage of shared genes between all 13 regions

Fig. 3. Origins of astrocyte diversity from regional and scRNA-seq data. subclusters to illustrate the three different brain regions examined. (E) UMAP
(A) Experimental procedure for scRNA-seq. Single cells were dissociated from plots from (C) showing the expression of Chrdl1, Gfap, and Crym. (F) Heat map
the cortex, hippocampus, and striatum of wild-type mice (N = 4 to 7 mice). shows the top 100 genes that were up-regulated in each astrocyte subcluster
Data for striatal cells were partly from our recent study (34) and combined with relative to the others. Example genes are identified. (G) Proportional bar graph
new cortical, hippocampal, and striatal single-cell data. (B) Uniform manifold showing the percentage of each astrocyte subcluster representation in the
approximation and projection (UMAP) plot of 89,553 brain cells from three CNS three CNS regions from scRNA-seq data. (H) Correlations between astrocyte
regions grouped by expression similarity identified 26 major cell populations region–specific RNA-seq data (left), bulk RNA-seq data (middle), and bulk
(10 major cell types). (C) UMAP plot of cluster analysis for 8898 astrocytes RNA-seq minus astrocyte region–specific RNA-seq data (right) that were
[AST1 and AST2 in (B)] from three CNS regions. The cell numbers of astrocyte gathered as part of Fig. 2. The correlation plots show that astrocytes grouped
subclusters are shown as AST1 to AST7. (D) UMAP plot of the astrocyte into three broad anatomical areas (“a,” “b,” and “c”).

in a pairwise manner for astrocyte RNA-seq ules (“cyan” and “lightyellow”) and astrocyte multiple mechanisms contributed to astrocyte
and bulk RNA-seq data (Fig. 3H). Next, we territory size and with Feret’s minimum and morphology, as reflected within the WGCNA
compared the percentage of shared genes be- maximum (Fig. 4, E and F). We analyzed genes modules.
tween the astrocyte region–specific RNA-seq within these modules using stringent criteria of We also performed experiments using the
data with the bulk RNA-seq data (minus astro- more than fourfold enrichment within astro- object location memory test as a measure of
cyte enriched genes) and found a similar trend cytes for both hippocampus and cortex (MCX, cognition in the Fermt2 and Ezr knockdown
(Fig. 3H). Because the astrocyte-specific RNA- SCX, and VCX) and by focusing on the top 5% mice and noted significant impairments rela-
seq data were correlated with the equivalent of genes displaying the highest significance tive to controls (Fig. 5, L and M; N = 10 mice).
bulk RNA-seq data per region, these data (Fig. 4, G and H). Using these criteria, within At a cellular level, these impairments were
provide evidence that region-specific features the darkmagenta module, we identified genes accompanied by elevated activity-dependent
were shaped, at least in part, by the local tis- Rorb, Fermt2, Gstm5, Slc9a3r1, Slc1a2, Limd1, immediate early gene (cFos) expression in
sue environment (Fig. 3H) and by variable and Ppara. Within the turquoise module, we neurons of the CA1 pyramidal cell layer (Fig. 5,
representation of the subclusters in a region- identified genes Tank, Tspan7, Ezr, Crot, Oaf, N and O; N = 5 to 7 mice) and reduced co-
specific manner (Fig. 3G). This recalls the find- Psat1, Mertk, and Celsr1. We also assessed localization between the pre- and postsynapse
ing that many postnatal cortical astrocytes morphology-related genes (top 5%) in astro- markers PSD-95 and VGLUT1 (Fig. 5, P and Q),
arise from local proliferation and are likely cyte subclusters from scRNA-seq (Fig. 4I). Most which was accompanied by reduced immu-
tuned to their environment (23). Our analy- of the genes positively correlated with astrocyte nostaining for these molecules (fig. S19). Thus,
ses provide the rationale and data to explore territory size were highly expressed in sub- these data show that reducing expression of
such relationships when specific genetic tools cluster AST2 and depleted in subcluster AST6. genes predicted from our WGCNA modules
become available. Because astrocytes in the cortex (MCX, SCX, to contribute to astrocyte morphology led to
and VCX) displayed larger territory size than reduced astrocyte morphology and to cogni-
Astrocyte morphology–related gene networks the striatum (Fig. 4, B and C, and fig. S16) and tive, cellular, and synaptic deficits (Fig. 5).
A key feature of astrocytes is their morpho- because subcluster AST2 was dominant in the
logical complexity. To exploit transcriptomic cortex, whereas subcluster AST6 was dominant Astrocyte morphology–related genes in AD
datasets, we documented the two-dimensional in the striatum (Fig. 3G), our data suggest that We identified several known AD risk genes (e.g.,
(2D) morphological features of astrocytes from distinct astrocyte subclusters contribute mor- Apoe, Clu, and Fermt2) among the WGCNA
13 CNS regions and related these to RNA-seq phological features of astrocytes. module genes that were highly correlated with
data using WGCNA. To quantify the morphol- astrocyte territory size (Fig. 4, F to H). These
ogy of astrocytes, we expressed GCaMP6f sparse- Evaluation of astrocyte morphology–related were enriched in astrocytes from the cortex
ly in astrocytes, obtained diffraction-limited 2D genes with CRISPR/Cas9 and hippocampus, areas that are affected in
images from flattened confocal volumes using From analyses of astrocyte territory size–related AD (fig. S20A). To explore a possible asso-
IHC (Fig. 4, A and B, and fig. S16), and sub- modules (Fig. 4), our attention was drawn to ciation between morphology-related genes
jected them to identical evaluations (Fig. 4A the proteins encoded by Fermt2 and Ezr, i.e., and AD, we performed scRNA-seq on 10- to
and fig. S16). For example, astrocytes in MCX FERMT2 (FERM Domain Containing Kindlin 2) 11-month-old APP/PS1 AD model mice (29)
displayed the largest territory size and showed and ezrin, because they are known putative relative to controls to identify astrocyte DEGs.
a relatively round shape, whereas astrocytes in actin-interacting proteins that may contribute We sequenced 45,391 cortical cells, and found
CB (Bergmann glia) showed the most elongated to astrocyte morphology. To test this possibility, 640 astrocyte DEGs (Fig. 6, A and B). Eleven
shape (fig. S16, B to I). We found no differences we used astrocyte-specific CRISPR/Cas9–based DEGs were up-regulated relative to controls
in the number of major branches per astrocyte gene knockdown by expressing Staphylococcus and 629 were down-regulated (P < 0.05, log2-
from the 13 CNS regions (average: 4.0 to 5.6; aureus Cas9 (SaCas9) in astrocytes together fold change > 0.1 or < –0.1; Fig. 6B). As shown
fig. S16J). The fractal dimension, a parameter with guide RNAs (gRNAs) to reduce the expres- from the volcano plots, several of the down-
reflecting morphological complexity, revealed sion of candidate genes in astrocyte reporter regulated genes in AD were related to those
significant differences between astrocytes in mice to image astrocyte morphology (Aldh1l1- correlated with astrocyte morphology, im-
13 CNS regions (fig. S16, K to N). Astrocytes Cre/ERT2 x Ai95; Fig. 5, A to C, and fig. S18). plying that astrocyte territory size may be
in each region showed distinct morphologi- We focused on Fermt2 and Ezr, achieving cell- reduced in AD. We also assessed genes related
cal features, which were summarized with a specific SaCas9 expression within astrocytes to astrocyte reactivity and found no significant
z score (Fig. 4C). By assessing the correlation (Fig. 5, D and E; N = 3 mice) and 72 and 75% trend in scRNA-seq of APP/PS1 mice or snRNA-
coefficient for the morphological z score of reduction in Fermt2 and Ezr expression rela- seq of human AD (fig. S20C), recalling recent
astrocytes in 13 CNS regions, we found strong tive to controls, respectively (Fig. 5, F to I, and work (30).
concordance among astrocytes from four close- fig. S18, A and B; N = 6 to 7 mice). Having We determined how many of the astrocyte
ly related forebrain regions (Fig. 4D), recalling established a significant reduction of Fermt2 territory size–related genes (from the top 5%;
RNA-seq (Fig. 2). and Ezr, we assessed the morphology of sparse- Fig. 4, F to H) were also significantly up- or
We performed WGCNA for astrocyte RNA- ly labeled hippocampal CA1 astrocytes (Fig. 5, J down-regulated DEGs in APP/PS1 mice (ad-
seq data of the 13 CNS regions and identified and K, and fig. S18C). We found that astrocytes justed P < 0.05). We found that 40 genes that
62 module eigengenes (fig. S17). We compared with reduced Fermt2 or Ezr showed signifi- were positively correlated with astrocyte terri-
morphological parameters and the 62 modules cantly reduced territory sizes and Feret’s di- tory size were down-regulated, and five genes
for astrocytes from the 13 CNS regions and mensions by ~20% (22% for Fermt2 and 19% that were negatively correlated with astrocyte
found significant correlations between the for Ezr; Fig. 5, J and K, and fig. S18, D and E), territory size were up-regulated in APP/PS1
two (Fig. 4E). We identified several modules showing that they are involved in the main- mice, and this trend was similar in human AD
with high correlation to astrocyte morpholog- tenance of astrocyte morphology, supporting snRNA-seq data (Fig. 6C) (31). To further ex-
ical parameters (red box in Fig. 4E). Of these, the predictions from WGCNA (Fig. 4). The plore these associations, we performed sub-
we focused on two modules, “darkmagenta” observed ~20% reductions in astrocyte mor- cluster analysis for astrocytes in APP/PS1 mice
and “turquoise” (Fig. 4F), because they dis- phology (P < 0.0001) are consistent with single- and found that wild-type (WT) and transgenic
played the highest correlation with other mod- gene evaluations (24–28) and suggest that APP/PS1 mice shared seven astrocyte subclusters

Fig. 4. Astrocyte morphology–related

gene networks. (A) Procedure for
morphological analysis of astrocytes
(an artificial red color was used).
(B) Representative images of single
astrocytes from 13 CNS regions.
(C) Heat map of aggregate z score of
the astrocyte morphological parameters
across 13 CNS regions (raw data are
provided in fig. S16). (D) Heat map
showing the correlation coefficient
along with a cluster dendrogram of
morphological parameter z scores
of astrocytes in 13 CNS regions.
Four regions (MCX, SCX, VCX, and HIP)
that are the most correlated are
outlined in yellow. (E) Heat map showing
correlation coefficients along with
cluster dendrograms among 62 module
eigengenes (MEs) in WGCNA and the
morphological parameters of astrocytes
from 13 CNS regions. Histogram in
the color scale represents the coefficient
values. The WGCNA modules that were
highly correlated with morphological
parameters, including for territory size,
are outlined in red (positive correlation) or
blue (negative correlation). (F) Multi-
dimensional scaling plot of MEs signifi-
cantly correlated with astrocyte territory
size (P < 0.05). Size of the node indicates
P value, color of the node indicates
correlation coefficient with astrocyte
territory size, and color of the line
indicates correlation coefficient between
two MEs. (G and H) Scatter plots showing
log2 ratio (immunoprecipitation versus
input) for the average value from
MCX, SCX, VCX, and HIP and –log10
(P value) for correlation of genes within
the darkmagenta (G) and turquoise
(H) modules. Several genes are named,
and the top 5% are emphasized in the
graphs with background coloring. (I) Heat
map showing the expression of the top
5% morphology-related genes within
astrocyte subclusters from the three CNS
regions identified in Fig. 3.
in distinct proportions. Subcluster AST1 was top 5% of morphology-related genes in astro- AST1 (Fig. 6E), indicating that astrocyte ter-
significantly increased and subcluster AST6 cyte subclusters from APP/PS1 mice. Most of ritory size may be reduced in the context of
was significantly decreased in APP/PS1 mice the genes that were positively correlated with AD. To test this, we injected AAV PHP.eB:
compared with wild-type mice (Fig. 6D and astrocyte territory size were highly expressed GfaABC1D Lck-GFP (32) into the retro-orbital
fig. S20D). We assessed the expression of the in subcluster AST6 and depleted in subcluster sinus of 9-month-old wild-type and APP/PS1

Fig. 5. Evaluation of astrocyte morphology–related genes with CRISPR/Cas9. CA1 layers of Aldh1l1-Cre/ERT2 x Ai95 mice injected with AAV for astrocyte-selective
(A) AAVs used for astrocyte-specific CRISPR-Cas9 gene knockdown. (B) Schematic SaCas9 and AAV for control gRNAs, Fermt2 gRNAs, or Ezr gRNAs (83 to 117
of morphology analysis of hippocampal astrocytes using astrocyte-specific astrocytes from N = 3 mice per group). (L and M) Object location memory test to
CRISPR/Cas9 gene knockdowns. (C) Exemplar images of brain sections immuno- assess control, Fermt2, and Ezr knockdown mice. Schematic, representative traces,
stained for SaCas9-HA (green) and mCherry (red). (D) Images of hippocampal and average data are shown (N = 10 mice; males and females are shown). The
astrocytes in 2- to 3-month-old wild-type mice injected with AAV for astrocyte- differences were significant for both sexes (P = 0.015 for males, P = 0.0076 for
specific SaCas9 immunostained for SaCas9-HA (green), S100b (red, left), or females) and for pooled data (P < 0.0001) using ANOVA; sex differences were not
NeuN (red, right). (E) Percentage of SaCas9-HA+ astrocytes or neurons in the our focus, so average data are shown. (N and O) Representative images and average
hippocampus of wild-type mice injected with AAV for SaCas9 (N = 3). (F) Exemplar data for cFos+ neurons (NeuN) in the CA1 region of the hippocampus from
images of sections immunostained with Fermt2 antibody. Brains were microinjected control, Fermt2, and Ezr knockdown mice (N = 5 to 7 mice). (P and Q) Representative
with AAV for astrocyte-specific SaCas9 and with AAV for control gRNAs or AAV images and average data for VGLUT1 and PSD-95 co-localization in the CA1 region
for Fermt2 gRNAs in the hippocampus. Loss of Fermt2 immunostaining was observed of the hippocampus from control, Fermt2, and Ezr knockdown mice (25 to 35 images
in the hippocampus. (G) Knocking-down efficiency for Fermt2 protein in the from N = 5 to 7 mice per group). The graph shows the percentage of PSD95 puncta
hippocampus of Aldh1l1-Cre/ERT2 x Ai95 mice injected with astrocyte-specific AAV co-localized with VGLUT1. In (Q), the points represent the average co-localization
for SaCas9 and AAV for Fermt2 gRNAs (N = 6 to 7). (H and I) As in (F) and (G) between VGLUT1 and PSD-95 puncta for each image (see fig. S19 for additional
but for Ezr (N = 6 to 7). (J and K) Images and data for astrocytes in the hippocampal analyses). Data are shown as means ± SEM.

Fig. 6. Astrocyte morphology–related genes in an AD mouse model, human showing the expression of the morphology-related genes (top 5% from Fig. 4) within
AD, and for other CNS disorders. (A) UMAP plot (45,391 cells) from the cortex of astrocyte subclusters in APP/PS1 mice. Genes positively correlated with astrocyte
wild-type and APP/PS1 mice grouped by expression similarity identified 10 cell morphology were enriched in cluster 6, which was reduced in APP/PS1 mice. By
populations (fig. S20B). Astrocytes were identified by the expression of Gja1. contrast, genes positively correlated with morphology were depleted in cluster 1,
(B) Volcano plot showing DEGs within astrocytes of APP/PS1 versus wild-type mice which was increased in APP/PS1 mice. (F) Procedure for morphology analysis
(adjusted P value < 0.05, log2-fold change > 0.10 or < –0.10); several genes are of cortical astrocytes in APP/PS1 mice. (G) Images of single astrocytes from the cortex
named. (C) Top graphs show the number of territory size–related genes from the labeled with Lck-GFP in WT and APP/PS1. (H) Analysis of Lck-GFP–labeled astrocyte
up- and down-regulated DEGs in astrocytes that were altered in scRNA-seq data territory size in the cortex of WT and APP/PS1 mice (N = 99 to 111 astrocytes from
from APP/PS1 mice and in human AD snRNA-seq data. The heat maps show the top N = 3 mice per group). (I) Hypergeometric heat map showing the enrichment
10 territory size–related genes that displayed high-log10P values in the WGCNA of astrocyte territory–related genes within those associated with CNS disorders.
analyses (from the top 5% from WGCNA). The WGCNA modules that the genes *Significance with FDR < 0.05. (J) Top 10 astrocyte territory size–related genes that
belong to are on the right. (D) UMAP plot of cluster analysis (3808 astrocytes) from overlapped with genes associated with AD, amyotrophic lateral sclerosis (ALS),
WT and APP/PS1 mice in relation to the bar graph showing cluster frequency of cerebral infarction (CI), multiple sclerosis (MS), schizophrenia (SCZ), bipolar disorder
astrocyte subclusters for WT and APP/PS1 mice (N = 9 to 10 mice). (E) Heat map (BD), and obsessive-compulsive disorder (OCD). Data are shown as means ± SEM.

mice and measured territory sizes of sparsely Discussion astrocyte RNA-seq. To test the hypothesis that
labeled astrocytes at ~10 months age (Fig. 6F). There is renewed interest in understanding the astrocyte morphology–related genes con-
Astrocyte territory sizes were significantly astrocyte biology in health and disease, neces- tributed to morphology, we developed astrocyte-
reduced in the cortex of APP/PS1 mice rela- sitating the exploration of astrocytes with specific CRISPR/Cas9–based gene knockdown
tive to controls (Fig. 6, G and H, and fig. S20E high-dimensional molecular data in different to reduce the expression of core genes in the
and F), which was consistent with our pre- regions (9–18). By comparing astrocytes across hippocampus, where astrocytes showed the
dictions (Fig. 5) and notable because astro- 13 CNS regions, we found ~850 genes that highest morphological complexity. We found a
cytes were assessed irrespective of subcluster were shared, with functions related to metab- significant reduction of hippocampal astrocyte
identity. Furthermore, because physiological olism, cholesterol biosynthesis, and neuro- territory sizes after the reduction of Fermt2
functions of astrocytes are implicated in the transmitter uptake and biosynthesis. These and Ezr proteins. We also observed concom-
pathophysiology of AD (7, 33), we performed represent the core functions of astrocytes, i.e., itant changes in a cognitive task, cFos expres-
candidate gene analyses for the astrocyte to maintain tissue homeostasis, and this re- sion in neurons, and co-localization of pre- and
subclusters and found that genes related to alization is important because many of these postsynapse markers. This suggested that the
cholesterol homeostasis, K+ homeostasis, neu- pathways are disrupted in brain disorders. astrocyte morphological changes that accom-
rotransmitter homeostasis, and synapse func- However, of the top genes shared between pany disease states have causal effects on neural
tion were enriched in subcluster AST6 and astrocytes, little is known about nearly a third circuit and synaptic function, thus contributing
depleted in subcluster AST1 (fig. S20, G to J). of them, implying that fully understanding the to disease phenotypes. Our molecular studies
scRNA-seq data lack spatial information, which core functions of astrocytes across CNS areas portend physiological studies to explore these
could be evaluated by mapping of astrocyte is an important experimental goal. For example, relationships systematically for AD across age
types in relation to plaque markers in AD K+ homeostasis is a core function of astrocytes, and sex.
model mice and in human tissue. Our data on and both Kir4.1 and ATP1A2 contribute (38). Unexpectedly, we identified several known
disease-specific astrocyte subclusters (AST1 Our data show that both Kcnj10 and Atp1a2 AD risk genes, Apoe, Clu, and Fermt2, as being
and AST6) suggest these as potential thera- are highly expressed in astrocytes across the significant astrocyte morphology–related
peutic targets for AD. CNS. Further studies are needed to explore genes. This is notable because astrocyte mor-
which dominates under specific settings. Across phology changes in neurological and psychiat-
Astrocyte morphology–related genes in other all regions, we found little evidence for the ric diseases, including AD (43–46). Moreover,
CNS disorders molecular machinery supporting astrocyte disease-associated astrocytes displayed higher
Our analysis of astrocytes in wild-type mice Ca2+-dependent vesicular glutamate release. Gfap expression in an AD mouse model (47).
identified genes associated with astrocyte We found abundant evidence that cholesterol By exploring associations between astrocyte
territory sizes (Fig. 4), which we tested with metabolism and fatty acid β-oxidation were morphology–related genes and AD, we found
CRISPR/Cas9–mediated knockdown (Fig. 5). core functions of astrocytes. that positively correlated morphology-related
Furthermore, these genes were down-regulated Our data provided strong evidence to in- genes were predominantly down-regulated. A
in human AD and in a mouse model of AD dicate that astrocytes have molecular features similar trend was also observed for human AD
that displayed reduced astrocyte territory sizes and functions specific to CNS regions, and (31) and for mouse and human HD (28, 34, 35).
(Fig. 6). To further explore such associations, suggested that these region-specific functions Because our findings predicted that astrocyte
we evaluated astrocytes from Huntington’s arise from the local tissue environment and by territory size might be reduced in the context
disease (HD), because they displayed reduced variable representation of subclusters AST1 to of AD, we assessed astrocyte territory sizes in
territory sizes (27, 28, 34, 35). In accord with AST7 in a region-specific manner. Once spe- APP/PS1 mice and found that astrocyte terri-
our AD findings, we found down-regulation cific genetic tools become available, it will be tory sizes were indeed significantly reduced in
of genes positively correlated with astrocyte possible to disentangle these two contribu- the cortex of these mice. In subcluster analysis
territory sizes, and up-regulation of genes tions to astrocyte diversity. Our data also pro- for astrocytes, we identified seven astrocyte
negatively correlated with astrocyte terri- vide molecules that are highly enriched within subclusters in distinct proportions. Subcluster
tories in a HD mouse model and postmortem astrocytes from specific brain regions. For ex- AST1 was significantly increased and subclus-
human HD tissue (fig. S21) (36). To explore ample, Chrdl1 is highly enriched in cortical as- ter AST6 was significantly decreased in APP/
the association between astrocyte territory- trocytes and regulates synaptic function (22). PS1 mice. The increased subcluster AST1 pop-
related genes and other common CNS dis- Crym, encoding m-crystallin, was unique to ulation was characterized by higher Gfap and
orders, we identified disease-associated genes striatal astrocytes, but its functional relevance reduced homeostatic and morphology-related
from Phenopedia (37) for AD, Parkinson’s dis- has been unexplored. Region-specific genes genes in APP/PS1 mice. Conversely, subcluster
ease, amyotrophic lateral sclerosis, cerebral suggest that it may be possible to target as- AST6, which was decreased, was characterized
infarction, multiple sclerosis, epilepsy, schiz- trocytes in a brain region–specific manner to by lower Gfap and by up-regulated homeo-
ophrenia, major depression, bipolar disorder, investigate neural circuits and for therapeu- static and morphology-related genes. These
and obsessive-compulsive disorder and per- tic strategies. data indicate dynamic alterations of astro-
formed hypergeometric analyses. Significant A cardinal feature of astrocytes is their bushy cytes in response to early pathological states in
enrichment of astrocyte territory–related genes morphology. We mined our data to explore the an amyloid AD model, reflecting morphology
was found for multiple sclerosis, obsessive- molecular underpinning of morphological dif- and cellular composition.
compulsive disorder, bipolar disorder, AD, ferences between astrocytes from different CNS We provide comprehensive molecular data
schizophrenia, cerebral infarction, and amy- regions. Several previously proposed molecules permitting new physiological and hypothesis-
otrophic lateral sclerosis (Fig. 6I). Figure 6J were detected, including Ntrk2 (39), Hepacam, driven experiments for astrocytic genes and
shows the top 10 astrocyte territory size–related Gja1 (24), Gjb6 (40), Nlgn1-3 (25), β-integrin pathways that are core features across the CNS
genes that overlapped with the genes associ- signaling genes (41), and Mertk, Chrdl1, and and for those unique to specific regions. Our
ated with these CNS disorders, indicating that Cyp4f15 (14, 42). We then identified gene net- findings identify the molecular basis of com-
astrocyte morphological changes may be a works related to astrocyte morphology across plex astrocyte morphology and indicate that
common underappreciated feature of diverse 13 CNS regions by assessing the correlation of reduced astrocyte morphological complexity,
CNS disorders. 10 morphological parameters to modules from and the attendant loss of homeostatic and

metabolic tissue support, is a common feature Tukey’s or Kruskal-Wallis tests were used. In 20. R. Srinivasan et al., New transgenic mouse lines for
of CNS disorders that is potentially amenable the figures, P values are indicated as follows: selectively targeting astrocytes and studying calcium
signals in astrocyte processes in situ and in vivo. Neuron 92,
to therapeutic manipulation. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 1181–1195 (2016). doi: 10.1016/j.neuron.2016.11.030;
0.0001. When P values were > 0.05, this was pmid: 27939582
Methods summary considered nonsignificant (n.s.). The number 21. W. Sun et al., SOX9 is an astrocyte-specific nuclear marker in
the adult brain outside the neurogenic regions. J. Neurosci. 37,
All animal experiments were conducted in ac- of replicates (n values), exact P values, sample 4493–4507 (2017). doi: 10.1523/JNEUROSCI.3199-16.2017;
cordance with the National Institutes of Health’s numbers, and details of statistical analyses in pmid: 28336567
Guide for the Care and Use of Laboratory each case are provided in table S4 for every 22. E. Blanco-Suarez, T. F. Liu, A. Kopelevich, N. J. Allen,
Animals and approved by the Chancellor’s experiment. Astrocyte-secreted chordin-like 1 drives synapse maturation
and limits plasticity by increasing synaptic GluA2 AMPA
Animal Research Committee at the University receptors. Neuron 100, 1116–1132.e13 (2018). doi: 10.1016/
of California, Los Angeles (UCLA). Mice were RE FERENCES AND NOTES
j.neuron.2018.09.043; pmid: 30344043
housed in the vivarium managed by the Divi- 23. W. P. Ge, A. Miyawaki, F. H. Gage, Y. N. Jan, L. Y. Jan, Local
1. C. Bargmann et al., “BRAIN 2025: A scientific vision” (National
generation of glia is a major astrocyte source in postnatal
sion of Laboratory Animal Medicine at UCLA Institutes of Health, 2014); https://braininitiative.nih.gov/
cortex. Nature 484, 376–380 (2012). doi: 10.1038/
with a 12-hour light/dark cycle and ad libitum sites/default/files/pdfs/brain2025_508c.pdf.
nature10959; pmid: 22456708
2. T. R. Insel, S. C. Landis, F. S. Collins, The NIH BRAIN Initiative.
access to food and water. All animals were Science 340, 687–688 (2013). doi: 10.1126/science.1239276;
24. K. T. Baldwin et al., HepaCAM controls astrocyte self-
healthy with no obvious behavioral phenotype organization and coupling. Neuron 109, 2427–2442.e10 (2021).
pmid: 23661744
doi: 10.1016/j.neuron.2021.05.025; pmid: 34171291
at the start of the experiments other than those 3. B. A. Barres, The mystery and magic of glia: A perspective
25. J. A. Stogsdill et al., Astrocytic neuroligins control astrocyte
on their roles in health and disease. Neuron 60,
expected, were not involved in previous studies, 430–440 (2008). doi: 10.1016/j.neuron.2008.10.013;
morphogenesis and synaptogenesis. Nature 551, 192–197
and were sacrificed during the light cycle. Mice (2017). doi: 10.1038/nature24638; pmid: 29120426
pmid: 18995817
26. B. Zhou et al., Astroglial dysfunctions drive aberrant
were assigned randomly to the experiments 4. N. J. Allen, D. A. Lyons, Glia as architects of central nervous
synaptogenesis and social behavioral deficits in mice with
system formation and function. Science 362, 181–185 (2018).
on the basis of genotype. WT C57BL/6NTac neonatal exposure to lengthy general anesthesia. PLOS Biol. 17,
doi: 10.1126/science.aat0473; pmid: 30309945
mice were maintained in an in-house breeding 5. X. Yu, J. Nagai, B. S. Khakh, Improved tools to study astrocytes.
e3000086 (2019). doi: 10.1371/journal.pbio.3000086;
pmid: 31433818
colony or purchased from Taconic Biosciences. Nat. Rev. Neurosci. 21, 121–138 (2020). doi: 10.1038/
27. R. Jiang, B. Diaz-Castro, L. L. Looger, B. S. Khakh,
Ai95 [B6;129S-Gt(ROSA)26Sortm95.1(CAG- s41583-020-0264-8; pmid: 32042146
Dysfunctional calcium and glutamate signaling in striatal
6. J. Nagai et al., Behaviorally consequential astrocytic regulation
GCaMP6f)Hze/J, JAX stock no. 028865] or of neural circuits. Neuron 109, 576–596 (2021). doi: 10.1016/
astrocytes from Huntington’s disease model mice.
J. Neurosci. 36, 3453–3470 (2016). doi: 10.1523/
Ai14 [B6;129S6-Gt(ROSA)26Sortm14(CAG- j.neuron.2020.12.008; pmid: 33385325
JNEUROSCI.3693-15.2016; pmid: 27013675
tdTomato)Hze/J, JAX stock no. 007908] mice 7. H. G. Lee, M. A. Wheeler, F. J. Quintana, Function and
28. J. C. Octeau et al., An optical neuron-astrocyte proximity assay
therapeutic value of astrocytes in neurological diseases.
were acquired from JAX and crossed with Nat. Rev. Drug Discov. 21, 339–358 (2022). doi: 10.1038/
at synaptic distance scales. Neuron 98, 49–66.e9 (2018).
Aldh1l1-Cre/ERT2 BAC mice [B6N.FVB- doi: 10.1016/j.neuron.2018.03.003; pmid: 29621490
s41573-022-00390-x; pmid: 35173313
29. J. L. Jankowsky et al., Mutant presenilins specifically elevate
Tg(Aldh1l1-Cre/ERT2)1Khakh/J, JAX stock no. 8. K. W. Kelley, H. Nakao-Inoue, A. V. Molofsky, M. C. Oldham,
the levels of the 42 residue beta-amyloid peptide in vivo:
Variation among intact tissue samples reveals the core
029655] fully backcrossed to C57BL/6NTac transcriptional features of human CNS cell classes.
Evidence for augmentation of a 42-specific gamma secretase.
background in an in-house breeding colony. Hum. Mol. Genet. 13, 159–170 (2004). doi: 10.1093/hmg/
Nat. Neurosci. 21, 1171–1184 (2018). doi: 10.1038/
ddh019; pmid: 14645205
RiboTag mice (B6N.129-Rpl22tm1.1Psam/J, s41593-018-0216-z; pmid: 30154505
30. Z. Jiwaji et al., Reactive astrocytes acquire neuroprotective as
JAX stock no. 011029) were acquired from JAX 9. H. Chai et al., Neural circuit-specialized astrocytes:
well as deleterious signatures in response to Tau and Aß
Transcriptomic, proteomic, morphological and functional
and crossed with Aldh1l1-Cre/ERT2 mice. APP/ pathology. Nat. Commun. 13, 135 (2022). doi: 10.1038/
evidence. Neuron 95, 531–549.e9 (2017). doi: 10.1016/
s41467-021-27702-w; pmid: 35013236
PS1dE9 mice [B6;C3-Tg(APPswe,PSEN1dE9) j.neuron.2017.06.029; pmid: 28712653
31. H. Mathys et al., Single-cell transcriptomic analysis of
85Dbo/Mmjax, JAX MMRRC stock no. 034829- 10. E. S. Akdemir et al., Lunatic fringe-GFP marks lamina-specific Alzheimer’s disease. Nature 570, 332–337 (2019).
astrocytes that regulate sensory processing. J. Neurosci. 42,
JAX] were acquired from the JAX Mutant 567–580 (2022). doi: 10.1523/JNEUROSCI.1392-21.2021;
doi: 10.1038/s41586-019-1195-2; pmid: 31042697
32. R. C. Challis et al., Systemic AAV vectors for widespread
Mouse Regional Resource Center (MMRRC) pmid: 34872929 and targeted gene delivery in rodents. Nat. Protoc. 14,
and maintained in an in-house breeding col- 11. A. Y. Huang et al., Region-specific transcriptional control of 379–414 (2019). doi: 10.1038/s41596-018-0097-3;
astrocyte function oversees local circuit activities. Neuron 106,
ony. APP/PS1 mice carry human transgenes 992–1008.e9 (2020). doi: 10.1016/j.neuron.2020.03.025;
pmid: 30626963
33. B. De Strooper, E. Karran, The cellular phase of Alzheimer’s
for both APP bearing the Swedish mutation pmid: 32320644 disease. Cell 164, 603–615 (2016). doi: 10.1016/
and PSEN1 bearing the deletion mutation of 12. C. C. John Lin et al., Identification of diverse astrocyte j.cell.2015.12.056; pmid: 26871627
exon 9, both under the control of the mouse populations and their malignant analogs. Nat. Neurosci. 20, 34. X. Yu et al., Context-specific striatal astrocyte molecular
396–405 (2017). doi: 10.1038/nn.4493; pmid: 28166219 responses are phenotypically exploitable. Neuron 108, 1146–1162.
prion promoter. 13. B. Lozzi, T. W. Huang, D. Sardar, A. Y. Huang, B. Deneen, e10 (2020). doi: 10.1016/j.neuron.2020.09.021; pmid: 33086039
All statistical tests were performed in Regionally distinct astrocytes display unique transcription 35. A. Benraiss et al., Cell-intrinsic glial pathology is conserved
GraphPad Prism 9. Power analyses were per- factor profiles in the adult brain. Front. Neurosci. 14, 61 (2020). across human and murine models of Huntington’s disease.
doi: 10.3389/fnins.2020.00061; pmid: 32153350 Cell Rep. 36, 109308 (2021). doi: 10.1016/j.celrep.2021.109308;
formed in GraphPad StatMate 2.0. The graphs
14. O. A. Bayraktar et al., Astrocyte layers in the mammalian pmid: 34233199
were created in OriginPro 2018b and assembled cerebral cortex revealed by a single-cell in situ transcriptomic 36. H. Lee et al., Cell type-specific transcriptomics reveals
in CorelDraw 2018. Data are presented as map. Nat. Neurosci. 23, 500–509 (2020). doi: 10.1038/ that mutant huntingtin leads to mitochondrial RNA release and
means ± SEM along with the individual data s41593-020-0602-1; pmid: 32203496 neuronal innate immune activation. Neuron 107, 891–908.e8
15. A. V. Molofsky et al., Astrocyte-encoded positional cues (2020). doi: 10.1016/j.neuron.2020.06.021; pmid: 32681824
points. In some of the graphs, the error bars maintain sensorimotor circuit integrity. Nature 509, 189–194 37. W. Yu, M. Clyne, M. J. Khoury, M. Gwinn, Phenopedia and
representing the SEM were smaller than the (2014). doi: 10.1038/nature13161; pmid: 24776795 Genopedia: Disease-centered and gene-centered views of the
symbols representing the mean. OriginPro 16. M. Y. Batiuk et al., Identification of region-specific astrocyte evolving knowledge of human genetic associations. Bioinformatics
subtypes at single cell resolution. Nat. Commun. 11, 1220 26, 145–146 (2010). doi: 10.1093/bioinformatics/btp618;
2018b was used to determine whether the data
(2020). doi: 10.1038/s41467-019-14198-8; pmid: 32139688 pmid: 19864262
were normally distributed. Parametric tests 17. M. M. Boisvert, G. A. Erikson, M. N. Shokhirev, N. J. Allen, The 38. N. MacAulay, Molecular mechanisms of K+ clearance and
were used for normally distributed datasets, aging astrocyte transcriptome from multiple regions of the extracellular space shrinkage-Glia cells as the stars. Glia 68,
and nonparametric tests were applied to data mouse brain. Cell Rep. 22, 269–285 (2018). doi: 10.1016/ 2192–2211 (2020). doi: 10.1002/glia.23824; pmid: 32181522
j.celrep.2017.12.039; pmid: 29298427 39. L. M. Holt et al., Astrocyte morphogenesis is dependent
not normally distributed or to data with a 18. L. Morel et al., Molecular and functional properties of regional on BDNF signaling via astrocytic TrkB.T1. eLife 8, e44667
small sample size. For comparisons of two astrocytes in the adult brain. J. Neurosci. 37, 8706–8717 (2019). doi: 10.7554/eLife.44667; pmid: 31433295
groups, paired and unpaired two-tailed Stu- (2017). doi: 10.1523/JNEUROSCI.3956-16.2017; 40. U. Pannasch et al., Connexin 30 sets synaptic strength
pmid: 28821665 by controlling astroglial synapse invasion. Nat. Neurosci. 17,
dent’s t tests or two-tailed Mann-Whitney 549–558 (2014). doi: 10.1038/nn.3662; pmid: 24584052
19. K. Seiriki et al., High-speed and scalable whole-brain imaging in
tests were used. For comparisons of more than rodents and primates. Neuron 94, 1085–1100.e6 (2017). 41. S. Cho, A. K. Muthukumar, T. Stork, J. C. Coutinho-Budd,
three groups, one-way ANOVA tests followed by doi: 10.1016/j.neuron.2017.05.017; pmid: 28641108 M. R. Freeman, Focal adhesion molecules regulate astrocyte

morphology and glutamate transporters to suppress seizure- discussions, S. Cheong for help counting astrocytes (Fig. 1D), the figures with help from F.E. B.S.K. wrote the paper with
like behavior. Proc. Natl. Acad. Sci. U.S.A. 115, 11316–11321 D. Geschwind for discussion about gene expression analysis, A. Huang help from F.E., and all authors approved the final version.
(2018). doi: 10.1073/pnas.1800830115; pmid: 30327343 and L. Wu for comments, and N. Okabe for IHC guidance. The Competing interests: The authors declare no competing interests.
42. D. Lanjakornsiripan et al., Layer-specific morphological and sequencing core was supported partly by the Genetics, Genomics Data and materials availability: RNA-seq data are available at the
molecular differences in neocortical astrocytes and their and Informatics Core of the Semel Institute of Neuroscience. Funding: Gene Expression Omnibus (GEO) under GSE198032 subseries
dependence on neuronal layers. Nat. Commun. 9, 1623 (2018). B.S.K., J.S., F.E., and X.Y. were supported by the National Institutes GSE19824 and are also provided in tables S1 to S3. scRNA-seq
doi: 10.1038/s41467-018-03940-3; pmid: 29691400 of Health (NIH grants R01DA047444, R35NS111583, and R01AG0759655), data are available at the GEO under GSE 198032 subseries
43. A. M. Arranz, B. De Strooper, The role of astroglia in an Allen Distinguished Investigator Award, a Paul G. Allen Frontiers GSE198027. License information: Copyright © 2022 the authors,
Alzheimer’s disease: Pathophysiology and clinical implications. Group advised grant of the Paul G. Allen Family Foundation, some rights reserved; exclusive licensee American Association
Lancet Neurol. 18, 406–414 (2019). doi: 10.1016/S1474-4422 and the Ressler Family Foundation (B.S.K.). J.S.S. was supported for the Advancement of Science. No claim to original US
(18)30490-3; pmid: 30795987 by the National Science Foundation Graduate Research government works. https://www.science.org/about/science-
44. M. Pekny et al., Astrocytes: A central element in neurological Fellowship Program (NSF-GRFP; DGE-2034835) and by a UCLA licenses-journal-article-reuse
diseases. Acta Neuropathol. 131, 323–345 (2016). Eugene V. Cota-Robles Fellowship. A.K. and H.H. were supported
doi: 10.1007/s00401-015-1513-1; pmid: 26671410 by JSPS KAKENHI grants JP21K19335 (H.H.), JP20H00492
45. M. Pekny, M. Pekna, Astrocyte reactivity and reactive (H.H.), and JP20H03391 (A.K.); MEXT KAKENHI grant JP18H05416 SUPPLEMENTARY MATERIALS
astrogliosis: Costs and benefits. Physiol. Rev. 94, 1077–1098 (H.H.); AMED grant JP21dm0207117 (H.H.); and the Takeda Science science.org/doi/10.1126/science.adc9020
(2014). doi: 10.1152/physrev.00041.2013; pmid: 25287860 Foundation, Japan (A.K., H.H.). V.G. was supported by the NIH Materials and Methods
46. C. Escartin et al., Reactive astrocyte nomenclature, definitions, (Pioneer Award DP1OD025535), the Vallee Foundation, the Figs. S1 to S21
and future directions. Nat. Neurosci. 24, 312–325 (2021). CZI Neurodegeneration Challenge Network, and the Beckman References (48–62)
doi: 10.1038/s41593-020-00783-4; pmid: 33589835 Institute for CLARITY, Optogenetics and Vector Engineering MDAR Reproducibility Checklist
47. N. Habib et al., Disease-associated astrocytes in Alzheimer’s Research (CLOVER) for technology development and dissemination. Tables S1 to S4
disease and aging. Nat. Neurosci. 23, 701–706 (2020). Author contributions: F.E. performed most experiments and Movies S1 and S2
doi: 10.1038/s41593-020-0624-8; pmid: 32341542 analyzed data. A.K. and H.H. performed FAST. X.Y. performed
striatal scRNA-seq. J.S. conducted stereotactic AAV microinjections
ACKN OW LEDG MEN TS and made SaCas9 plasmids. Z.Q. and V.G. made PHP.eB:GfaABC1D
We thank F. Gao and the UCLA Neuroscience Genomics Core for Lck-GFP AAV. R.K. guided RNA-seq analyses. B.S.K. conceived Submitted 7 May 2022; accepted 16 September 2022
assistance with RNA-seq, M. Gangwani for AAV delivery (Fig. 6F) and the study, planned and directed the experiments, and produced 10.1126/science.adc9020

RESEAR CH
◥ in the voltage dynamics of SST- and vasoactive

RESEARCH ARTICLE SUMMARY intestinal peptide (VIP)–expressing interneurons.
In the hippocampus, dual-polarity voltage
NEUROSCIENCE imaging uncovered the differential contribu-
tions of distinct cell classes to the local field
Dual-polarity voltage imaging of the concurrent potential (LFP). Spiking in entorhinal cortex
(EC)–projecting neurons, but not in SST in-
dynamics of multiple neuron types terneurons, consistently phase-locked to the
LFP. Moreover, using dual-color voltage im-
Madhuvanthi Kannan*†, Ganesh Vasan*†, Simon Haziza†, Cheng Huang, Radosław Chrapkiewicz, aging, we found that anterior cingulate cortex
Junjie Luo, Jessica A. Cardin, Mark J. Schnitzer*, Vincent A. Pieribone* (ACC)– and EC-projecting excitatory subclasses
differ in their responses to state change; the
former uniformly increase their firing rates,
INTRODUCTION: Signal processing in the brain proved voltage sensitivities. We also identified but the latter do not.
involves the concerted dynamics of multiple the respective reverse response-polarity variants, Finally, by combining dual-polarity and dual-
distinct cell subtypes that belong to excit- pAce and pAceR, which exhibit fluorescence color imaging using three mutually compatible
atory and inhibitory neuron populations. To increases during APs. We reasoned that the sensors from our suite, we extracted the si-
uncover how they synergize their activities four new indicators make up a suite of mutually multaneous, real-time voltage dynamics of
in real time, it is important to record the volt- compatible tools for multipopulation voltage three neuron types. These subtypes belong to
age dynamics of multiple identified subtypes recordings in behaving animals. When tar- excitatory and inhibitory interneuronal sub-
simultaneously. Genetically encoded fluores- geted to distinct neuron types, the indicators populations in the visual cortex and hippo-
cent voltage indicators (GEVIs) are ideally enable unambiguous cell-type identification campus of running mice.
suited to reveal the millisecond-scale activa- from the voltage readout based on their fluo-
tion dynamics of one or more genetically iden- rescence spectra, optical spike polarity, or both. CONCLUSION: Our work reveals functional
tified neuron types in awake animals. correlations between targeted neuron types
RESULTS: Our improved indicators exhibit en- in visual cortical circuits during altered be-
RATIONALE: We previously introduced the green hanced voltage sensitivities to enable low- havioral states as well as the contributions of
and red fluorescence resonance energy transfer power (<25 mW mm−2) recordings of spiking distinct cell subclasses to hippocampal LFPs in
(FRET)–opsin GEVIs Ace-mNeon and VARNAM, and transmembrane voltage dynamics from running mice. Together, our findings bench-
which are fusion proteins of voltage- sensitive >50 individual neurons at a time in running mark multipopulation voltage imaging for
rhodopsin from Acetabularia with bright fluo- mice and ~30-min continuous imaging in flies. investigations of the concerted dynamics of
rescent proteins mNeonGreen and mRuby3, re- Ace-mNeon2 voltage imaging from ~1200 neu- multiple cell classes in behaving animals.
spectively. Ace-mNeon and VARNAM exhibit rons revealed visual cortical cell type–specific
▪
fluorescence decreases during voltage depolar- responses to behavioral-state transitions in
The list of author affiliations is available in the full article online.
ization events such as action potentials (APs) awake mice. Cortical interneurons expressing *Corresponding author. Email: [email protected] (M.K.);
and together constitute a pair of mutually com- somatostatin (SST) exhibited the most pro- [email protected] (G.V.); [email protected] (M.J.S.);
patible GEVIs that can be genetically targeted found activation during a quiescence-to-arousal [email protected] (V.A.P.)
†These authors contributed equally to this work.
to two distinct neuron types for simultaneous transition. Further, using simultaneous dual-
Cite this article as M. Kannan et al., Science 378,
dual-color voltage recordings. Using protein polarity voltage recordings with Ace-mNeon2 eabm8797 (2022). DOI: 10.1126/science.abm8797
engineering and high-throughput voltage screen- and pAce, we examined the functional dynamics
ing, we uncovered the second-generation GEVIs of pairs of neuron types. We unveiled a time- READ THE FULL ARTICLE AT
Ace-mNeon2 and VARNAM2, which have im- varying, state-dependent mutual antagonism https://doi.org/10.1126/science.abm8797
Dual-polarity imaging of the simul- Negative polarity Positive polarity Viral injection of Dual-polarity voltage
taneous voltage dynamics of two voltage indicator voltage indicator positive and negative imaging using a single
neuron types in running mice. S81D
polarity indicators color channel
Ace N81S R78K in two distinct 505-nm
Newly characterized negative- and
neuron types LED
positive-polarity GEVIs enable
unambiguous identification of neu-
ron types based on the optical spike D92N W178F
mNeon
polarity in single-color voltage imag- Green
ing. Using this technique, we uncover
a state-dependent antagonism
between visual cortical SST and VIP Primary visual cortex Hippocampus
interneurons as well as relative con-
Excitatory Inhibitory LFP
Visual cortical Quiescence Arousal Hippocampus

tributions to the LFP from hippocam- microcircuit LFP
pal excitatory and inhibitory input
SST
VIP
neurons. N81S, Asn81→Ser; R78K,
Arg78→Lys; S81D, Ser81→Asp; D92N, SST
output
Asp92→Asn; W178F, Trp178→Phe;
VIP
LED, light-emitting diode; PN, PN Inhibitory

pyramidal neuron. 500 ms Air puff Excitatory 500 ms

RES EARCH
◥ in excitable human embryonic kidney (HEK)

RESEARCH ARTICLE cells (Fig. 1, A and B) (18, 30, 31). Our screening
platform scores GEVIs for brightness, voltage
NEUROSCIENCE sensitivity, and kinetics in a plate-based format,
achieving a capacity of 120 unique variants per
Dual-polarity voltage imaging of the concurrent day with a sample content of ~2000 cells per
variant (18).
dynamics of multiple neuron types In Ace-mNeon and VARNAM, depolarization-
dependent quenching by Ace of fluorescence
Madhuvanthi Kannan1,2*†‡, Ganesh Vasan1,2*†‡, Simon Haziza3,4†, Cheng Huang3, from the FP leads to depolarization-evoked
Radosław Chrapkiewicz3,4, Junjie Luo3,4,5, Jessica A. Cardin6,7,8, fluorescence decreases as the readout of volt-
Mark J. Schnitzer3,4,5*, Vincent A. Pieribone1,2,6* age changes. To improve voltage sensitivity,
we sought to enhance the FRET transfer by
Genetically encoded fluorescent voltage indicators are ideally suited to reveal the millisecond-scale optimizing the Ace-FP linker (32). In Ace-
interactions among and between targeted cell populations. However, current indicators lack the requisite mNeon, we also performed site-saturation
sensitivity for in vivo multipopulation imaging. We describe next-generation green and red voltage mutagenesis at Ace N81 (Asn81) and screened a
sensors, Ace-mNeon2 and VARNAM2, and their reverse response-polarity variants pAce and pAceR. Our library of N81X positional mutants (where X
indicators enable 0.4- to 1-kilohertz voltage recordings from >50 spiking neurons per field of view in is any amino acid) (Fig. 1A and fig. S1) (18).
awake mice and ~30-minute continuous imaging in flies. Using dual-polarity multiplexed imaging, we Screening of mutagenic libraries uncovered
uncovered brain state–dependent antagonism between neocortical somatostatin-expressing (SST+) and variants with increased response amplitudes
vasoactive intestinal peptide–expressing (VIP+) interneurons and contributions to hippocampal field (Fig. 1B and fig. S2). Ace-mNeon N81S G229Y
potentials from cell ensembles with distinct axonal projections. By combining three mutually compatible (Asn81→Ser and Gly229→Tyr) and VARNAM
indicators, we performed simultaneous triple-population imaging. These approaches will empower D228W D229R G231I (deletion of Trp228 and
investigations of the dynamic interplay between neuronal subclasses at single-spike resolution. Arg229, and Gly231→Ile) (henceforth, Ace-mNeon2
and VARNAM2, respectively) exhibited signif-
icantly improved fluorescence responses to
he dynamic interplay of multiple excit- concurrent, targeted recordings from two or depolarizing voltages compared with their
T
atory and inhibitory neuronal subtypes more neuron types in behaving animals. predecessors in patch-clamp recordings from
is central to how the brain processes in- We previously introduced the green and red transfected cells [the ratio of fluorescence
formation. During motor or sensory pro- fluorescence resonance energy transfer (FRET)– change to baseline fluorescence (DF/F) values
cessing, the external environment and opsin GEVIs Ace-mNeon and VARNAM (17, 18). for 120-mV steps were (mean ± SEM) −31.3 ±
an animal’s internal state can affect the ac- FRET-opsin GEVIs are fusions of a bright 0.7% for Ace-mNeon2 (n = 13 cells) versus
tivity of neocortical interneurons, which in fluorescent protein (FP; the FRET donor) to a −13.9 ± 0.6% for Ace-mNeon (n = 11 cells), P <
turn modulate excitatory projection neurons transmembrane opsin with a voltage-sensitive 0.0001, Mann-Whitney test; and −22.9 ± 0.7%
(1–5). Likewise, in the hippocampus, differ- light absorption spectrum (the FRET acceptor) for VARNAM2 (n = 13 cells) versus −14.1 ± 0.5%
ent ensembles of projection neurons exhibit (25, 26). This modular design enables voltage for VARNAM (n = 8 cells), P < 0.0001, Mann-
state-dependent changes in spiking activity imaging in live animals with a high dynamic Whitney test] (Fig. 1, C and D; figs. S3 to S5;
related to learning and memory (6–9). How- range of fluorescence signaling but at far- and table S1).
ever, how different neural subclasses coordi- lower illumination levels (~25 mW mm−2) than We next screened a library of Ace-mNeon2
nate their dynamics in real time remains poorly those (1 to 10 W mm−2) required by opsin volt- D92X positional mutants, owing to the highly
understood, owing to an inability to measure age indicators (16, 19, 22, 23, 27). Imaging conserved role of the N81/D92 (D, Asp) pair
the spiking activity of two or more genetically studies using FRET-opsins have generally mon- in opsin proton-pumping and photosensing
defined or projection-targeted neural ensembles itored the time-dependent emissions of the (33, 34). We uncovered several variants with
simultaneously. bright FRET donor but not those of the opsin response polarity reversals that exhibited flu-
Genetically encoded voltage indicators (GEVIs) FRET acceptor, because the latter’s fluores- orescence increases during depolarization (fig.
enable targeted recordings of spiking and sub- cence is too dim to augment the detection of S6). Although the best-performing positive var-
threshold activity with submillisecond tempo- voltage signals. Thus, only one fluorescence iant Ace-mNeon2 D92N exhibited slower kinetics,
ral precision (10–12). Although extensive protein channel is needed to image a FRET-opsin reciprocal S81X saturation mutagenesis re-
engineering has spawned a host of new sensors GEVI, with excitation and emission wave- vealed kinetics rescue mutants (figs. S6 and S7).
(13–24), existing GEVIs are not optimized for length bandwidths matching those of the bright Among them, the S81D mutant also provided
donor (17, 18, 25). Ace-mNeon and VARNAM the largest voltage sensitivity. The 92N/81D
1
comprise a fusion of the Acetabularia (Ace) reversed polarity phenotype is transposable
The John B. Pierce Laboratory, New Haven, CT 06519, USA.
2
Department of Cellular and Molecular Physiology, Yale
rhodopsin (17, 18) to the bright FP mNeonGreen across other Ace-based indicators, including
University, New Haven, CT 06520, USA. 3James H. Clark (28) or mRuby3 (29), respectively. Notably, VARNAM and Voltron (13, 18).
Center, Stanford University, Stanford, CA 94305, USA. 4CNC Ace-mNeon and VARNAM have demonstra- For further engineering, we targeted resi-
Program, Stanford University, Stanford, CA 94305, USA.
5
Howard Hughes Medical Institute, Stanford University,
ted optical compatibility for concurrent re- dues lining Ace’s photosensing helix C, amino
Stanford, CA 94305, USA. 6Department of Neuroscience, Yale cordings in live flies (18). acids involved in stabilizing its voltage-sensitive
University, New Haven, CT 06520, USA. 7Kavli Institute of intermediate state, and those near the chromo-
Neuroscience, Yale University, New Haven, CT 06520, USA. Protein engineering of negative- and phore (Fig. 1A and fig. S1) (34–36). Voltage
8
Wu Tsai Institute, Yale University, New Haven, CT 06520, USA. positive-polarity FRET-opsins
*Corresponding author. Email: [email protected] (M.K.); screening of mutagenic libraries uncovered
[email protected] (G.V.); [email protected] (M.J.S.); To improve the sensitivities of Ace-mNeon positive variants with enhanced sensitivities
[email protected] (V.A.P.) and VARNAM for recordings in awake mice, (fig. S8). Ace-mNeon2 R78K S81D D92N W178F
†These authors contributed equally to this work.
‡Present address: Department of Neuroscience, University of we performed protein engineering and high- and VARNAM2 R78E S81D D92N, henceforth
Minnesota, Minneapolis, MN 55455, USA. throughput screening of voltage responsivity pAce (positive Ace) and pAceR (positive Ace in
Kannan et al., Science 378, eabm8797 (2022) 4 November 2022 1 of 20

A B 40 pAce
C Ace II Linker FP D 40 pAce
Total performing = 2,592
∆F/F (%)
20
∆F/F (%)
Ace opsin 20 Ace-mNeon2
0 0
S189
1S
A122 SY mNeon TS
E199 Ace-mNeon Ace-mNeon
N8
-20 Green -20
E D F G pAce
Ace-mNeon2 -40
Ace-mNeon2
-40
S8 K
2N
8F
D9 D
-150 -100 -50 0 50
8
1
17
R7
D207 0 500 1000 1500 2000 2500 40 pAceR
W
W178
40 Total performing = 828 pAceR
∆F/F (%)
∆F/F (%)
C B A 20
R78...I94 20 VARNAM2 ∆W∆R
0
0 SI
mRuby3 TS VARNAM
VARNAM -20
-20 pAceR VARNAM2
VARNAM2 -40
0 200 400 600 800 -150 -100 -50 0 50
S8 E
2N
D9 D
8
No. of variants Voltage (mV)
1
R7
E Split-Gal4 > K NDNF-Cre+/AAV-CAG-DIO-Ace-mNeon2 in V1
UAS-GEVI 22 46 27
9 15 31
2 26 14 18 25
11 38 4 37 34
PPL1-γ2α’1
28 32 39 3 16
10 36 45 41
50 12
F 5% IAA G 60 1 29 40
Ace-mNeon2
5% IAA 51 5 17
∆F/F 49 33 21
8 13
48 23 43
30 42
35 44
4%
7 20
0 19
47 24 6 30
60
pAce
4%
30
Spikes /s
51
0
50
VARNAM2
60 49 51
2%
30 48
47
0
46 49
60
45
30
pAceR
2%
44
0 43 46
42
1s 1s
41
H I 40
39
45
V1 PNs VIP-Cre+/AAV-CAG-DIO-GEVI in V1 38
37 44
Ace-mNeon2
∆F/F 36
35
15%
34
33 40
32
pAce
31
10%
30
36
29
28
VARNAM2
27
10%
26 32
25
24
31
23
pAceR
5%
22
21
1s 50 ms 5 ms 20 20
J 19
18
AAV-CaMKII-Cre/AAV-CAG-DIO-Ace-mNeon2 in V1 17
16 18
15
14
Cell # 17
13
1 12
11 16
10
2 9
8 9
3 7
6
5 7
4 4
-10% ∆F/F
10% ∆F/F
3
2
6
1
1s 2s 0.5 s
Fig. 1. Design and characterization of Ace-mNeon2, VARNAM2 and reverse obtained using whole-cell recordings and concurrent fluorescence imaging of
polarity variants pAce and pAceR. (A) Crystal structure of Ace [Protein Data HEK cells transfected with Ace-mNeon, Ace-mNeon2, and pAce (top) and
Bank (PDB) ID 3AM6], showing residues targeted for site-directed saturation VARNAM, VARNAM2, and pAceR (bottom). Values represent mean ± SEM.
mutagenesis on Ace-mNeon and VARNAM. The seven transmembrane helices (E) In vivo voltage imaging (left) through a transparent surgical window
are labeled A to G. Amino acids R78 through I94 on helix C and specified loci on implanted on a fly head (graphic on right). Each of the indicators was selectively
other helices were targeted. A, Ala; W, Trp; R, Arg; I, Ile; S, Ser; E, Glu; D, Asp. expressed in a PPL1-g2a′1 dopaminergic neuron using a split-GAL4 system (center).
(B) Distribution of fluorescence responses to field stimulation acquired from spiking The yellow dashed box indicates the axonal region of PPL1-g2a′1. Scale bars,
HEK cells expressing Ace-mNeon (top) and VARNAM (bottom) variants obtained 200 mm (left) and 20 mm (right). (F) Representative recordings from PPL1-g2a′
on the high-throughput platform. (C) Schematic representations of Ace-mNeon2, 1 using the four GEVIs, showing odor-evoked spiking elicited by 5-s delivery
pAce, VARNAM2, and pAceR constructs, depicting the point mutations in Ace and (gray shading) of 5% isoamyl acetate (IAA). (G) Spiking rates (mean ± SEM) during
the Ace-FP linker. SI is Ser, Ile; SY is Ser, Tyr; and TS is Thr, Ser. (D) DF-DV curves odor delivery and raster plots of individual trials in (F) (n = 10 trials, two trials

per fly). (H) Confocal images from coronal cortical slices of V1 PNs expressing drifting gratings presented at eight different orientations, acquired using Ace-
the indicated sensors. Scale bar, 5 mm. (I) Example fluorescence traces showing mNeon2 in an awake mouse. Arrows indicate stimulus orientation. Gray shading
spontaneous spiking obtained from awake, head-restrained mice selectively denotes stimulus periods. (K) Representative epifluorescence image of a single field
expressing the indicators in V1 VIP+ interneurons (left); the boxed areas on the of view from a NDNF-Cre+ mouse expressing Cre-dependent soma-targeted Ace-
traces shown at an expanded time scale (center); and optical spike waveforms mNeon2 in V1 (scale bar, 50 mm) (top) and DF/F traces showing spontaneous
(mean ± SEM) (right) [n = 5 neurons per condition ranked in order of decreasing activity from the numbered cells (bottom). Fluorescence traces are inverted for
signal-to-noise ratios, three mice each (Ace-mNeon2 and pAce) and two mice each visualization. The region enclosed by the dashed box is shown at an expanded time
(VARNAM2 and pAceR)]. (J) Example visual responses from layer 2/3 PNs to scale on the right for select cells. Gray ticks denote identified spikes.
Red), exhibited DF/F values of 36.6 ± 0.7% For studies in live mice, our epifluorescence Ace-mNeon2 to neuron-derived neurotrophic
(mean ± SEM) and 33.4 ± 1.1% (n = 11 cells imaging methodology was designed to allow factor–expressing (NDNF+) interneurons, ~50
each), respectively, per 120 mV in cultured cellular resolution recordings in cortical brain to 70 mm below the pia (41, 47, 48) (Fig. 2B),
HEK cells (Fig. 1, B to D; and figs. S5, S9, and tissue and to reduce both tissue autofluo- and recorded spiking in up to 51 cells at once
S10) (E, Glu; F, Phe; K, Lys). Because these pos- rescence and background fluorescence from (Fig. 1K) for 4 to 5 min continuously (fig. S16).
itive variants have the same signaling dynam- scattered or out-of-focus indicator emissions. Similarly, using other Cre- or Flp-driver lines,
ic range as Ace-mNeon2 and VARNAM2 but Specifically, we (i) were able to use modest illu- we respectively targeted Ace-mNeon2 or pAce
in the opposite direction, they have slightly mination intensities (typically, 25 mW mm−2), to layer 2/3 somatostatin-expressing (SST+)
lower resting fluorescence intensities than the owing to the high brightness of our indicators; interneurons [~200 to 350 mm deep (40)] or
latter and reduced resting-state photobleach- (ii) achieved relatively sparse fluorescence la- VIP+ interneurons [~50 to 300 mm below the
ing, similar to other positive GEVIs (fig. S11) beling (<370 labeled cells per mm2) by target- pia (49)] to capture voltage signals from multi-
(14, 37). ing specific cell classes through virally mediated ple cells of each neuron type (Fig. 2, C and D;
All indicators except pAceR showed sub- projection targeting or using cell type–specific and fig. S17). Across these different cortical
millisecond kinetics at room temperature Cre- or Flp-driver mouse lines (39–42); and subtypes, d′ values ranged from 9.9 to 17.5,
(fig. S5 and table S1). Overall, Ace-mNeon2, (iii) generated soma-targeted versions of Ace- implying spike detection error rates ranging
VARNAM2, and pAce exhibited the largest mNeon2, VARNAM2, pAce, and pAceR, which from 0.04 spike min–1 to infinitesimal values
effective DF/F (%) for single action poten- mainly trafficked to cell body membranes (fig. S18) (38). This high level of spike detection
tials (APs) compared to prior negative- and (Fig. 1H) (43). With this labeling and traffick- fidelity allowed us to extract spike trains and
positive-polarity GEVIs in 1-ms impulse re- ing approach, <4.2% (and usually much less) of spike timing estimates that were invariant to
sponse computations (fig. S12 and table S1). the area of each imaging plane was occupied the use of different cell-extraction algorithms
by labeled cell bodies. Further, for data analy- (fig. S15, A, B, D, and G).
Sensor characterization in awake ses, we used automated algorithms designed to
flies and mice extract neuronal activity traces with minimal Voltage imaging of visual cortical ensembles
For voltage imaging in live animals for all contamination from physiologic artifacts or during behavioral state transitions
studies described in this paper, we used high- background activity (44–46) (fig. S15 and Ma- In neocortex, interneuron activity is strongly
speed (400 or 600 Hz in mice, 1 kHz in fruit terials and methods). influenced by an animal’s internal state and,
flies), widefield one-photon epifluorescence We expressed each of our four GEVIs in in turn, modulates excitatory cells (1–5). How-
microscopy. We first expressed each of our vasoactive intestinal peptide–expressing (VIP+) ever, the extent to which different interneuron
four GEVIs in the PPL1-γ2a′1 dopaminergic interneurons in primary visual cortex (V1), in types respond to state transitions and how dis-
neuron in flies, in which they reported spon- which they successfully reported single action tinct brain states affect the millisecond-scale in-
taneous and odor-evoked spiking with com- potentials (Fig. 1, H and I). The DF/F values teractions between subtypes remain unknown.
parable baseline- and evoked spike rates (Fig. per spike (mean ± SEM) were −16.5 ± 1.9% for To address these questions, we selectively
1, E to G). Compared with their red counter- Ace-mNeon2 (n = 127 spikes), 10.7 ± 1.4% for targeted Ace-mNeon2 to NDNF+, VIP+, or SST+
parts, the green GEVIs exhibited greater re- pAce (n = 87 spikes), −7.3 ± 1.2% for VARNAM2 interneurons in area V1, or to V1→AM proj-
sponse amplitudes and values of d′, a signal (n = 49 spikes), and 3.2 ± 0.5% for pAceR (n = ecting neurons, by using the same mouse lines
detection theory metric that characterizes how 48 spikes). and viruses as described earlier in the text
the ratio of fluorescence signals to background To record stimulus-evoked activity, we used (Fig. 2A) (40, 41). Voltage imaging in V1 of
fluctuations sets the spike detection fidelity axonal projection targeting to express our best- awake, head-restrained mice revealed sponta-
(38) [DF/F, d′ (mean ± SEM): 3.1 ± 0.2%, 14.5 ± performing GEVI, Ace-mNeon2, in a sparse neous spiking rates (mean ± SEM) of 4.72 ±
1.4 (Ace-mNeon2); 4.0 ± 0.1%, 13.4 ± 0.8 (pAce); subset of V1 layer 2/3 pyramidal neurons 0.01 Hz for NDNF+ neurons (n = 192 cells,
1.1 ± 0.1%, 7.4 ± 0.4 (VARNAM2); 0.9 ± 0.03%, (PNs) ~100 to 200 mm beneath the pia (Fig. seven mice), 6.52 ± 0.23 Hz for VIP+ neurons
6.5 ± 0.3 (pAceR)]. We also compared the per- 2E). Specifically, we targeted PNs projecting to (n = 49 cells, five mice), 4.71 ± 0.21 Hz for SST+
formance of all four GEVIs in the PPL1-a′2a2 the anteromedial visual cortical area (AM) neurons (n = 43 cells, five mice), and 0.93 ±
neuron, where they had greater d′ values (39, 42) by injecting an adeno-associated virus 0.34 Hz for PNs (n = 53 cells, five mice). These
than their parent indicators Ace-mNeon and (AAV) expressing Cre-dependent Ace-mNeon2 values are consistent with estimated spike
VARNAM [~10.8 (Ace-mNeon2), ~8.7 (pAce), into area V1 of wild-type mice along with a retro- rates of V1 layer 2/3 PNs and SST+ cells ob-
~7.0 (VARNAM2), and ~6.7 (pAceR) versus re- grade AAV expressing Cre under the CaMKII tained using electrophysiological recordings
ported values of ~6.9 (Ace-mNeon) and ~6.6 promoter into the AM. Voltage imaging of these in awake mice (4, 50). Likewise, the higher
(VARNAM)] (fig. S13) (18). The higher DF/F V1→AM projecting neurons during presenta- spontaneous firing rates of VIP+ versus SST+
values and superior photostability of pAce tions of drifting grating visual stimuli revealed interneurons have been observed across dis-
enabled extended duration (30 min) contin- the cells’ orientation-selective visual responses tinct cortical areas (51, 52).
uous recordings with stable d′ values through- (Fig. 1J). To induce a change in brain state by in-
out imaging sessions with illumination powers To showcase imaging of neural ensemble creasing arousal, we delivered a brief air puff
of ~5 mW mm−2 (fig. S14). dynamics, we used Cre-driver mice to target to the mouse’s back while monitoring its pupil

A B C D E
NDNF-Cre+ VIP-Cre+ SST-Cre+ AAV-CaMKII-Cre
Intracranial viral a. L1 a. L1 L2/3 a. L1
L2/3
a. L1 L2/3
delivery of indicators L2/3
AAV/CAG-DIO-
Ace-mNeon2 in V1
Cre+
b. b. b. b.
Z-score 10
High-speed (0.4 kHz)
voltage imaging
sCMOS
camera Pupil
tracking
505 nm
LED
camera c. 1 1
c. 1 1
c. 1 1
c. 1 1
101
Air puff 101
21
201
Cell #
0.5 0.5 101 0.5 41 0.5

301
201
401 61
501 301
0 0 201 0 81 0
d. 25
63%
9% unchanged d. 25 13% d. 25 9% d. 25 5%
13%
Running
Spike rate (Hz)
20 elevated 28% 20 53% 20 20 39%

suppressed 34% 78% 56%
wheel
15 15 15 15
574 cells 330 cells 213 cells 82 cells
10 10 10 10
5 5 5 5
0 0 0 0
e. e. e. e.
0.3 mm
Pupil
diameter (∆d)
f. f. f. f.
2 cm s-1
Locomotion
-2 -1 0 1 2 3 -2 -1 0 1 2 3 -2 -1 0 1 2 3 -2 -1 0 1 2 3
Time from airpuff onset (s) Time from airpuff onset (s) Time from airpuff onset (s) Time from airpuff onset (s)
F Mean ± S.E.M.: G Mean ± S.E.M.:
H Mean ± S.E.M.:
I Mean ± S.E.M.:
0.4 0.07 ± 0.004 (before) vs. 0.4 0.19 ± 0.01 vs. 0.21 ± 0.008 0.4 0.16 ± 0.01 vs. 0.26 ± 0.01 0.4 0.097 ± 0.04 vs. 0.12 ± 0.03
0.08 ± 0.003 (after)
0.3 before air puff 0.3 0.3 0.3
Fraction
Fraction
Fraction
Fraction
after air puff
0.2 0.2 0.2 0.2
0.1 0.1 0.1 0.1
0 0 0 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
Corr. coefficient Corr. coefficient Corr. coefficient Corr. coefficient
Fig. 2. Ace-mNeon2 voltage imaging unveils state-dependent modulation increased (dark blue), suppressed (cyan), or unchanged (white) spike rates
of spontaneous firing in excitatory and inhibitory cell types in V1. after air puff. (e) Pupil diameter (mean ± SEM). (f) Locomotor speed (mean ±
(A) Schematic of AAV injection and experimental setup. LED, light-emitting diode; SEM). (C) Same as (B) for VIP+ interneurons (n = 330 cells, five mice).
sCMOS, scientific complementary metal oxide semiconductor. (B) Ace-mNeon2 (D) Same as (B) for SST+ interneurons (n = 213 cells, six mice). (E) Same as (B)
voltage imaging of NDNF+ neurons during air puff delivery. (a) Confocal for PNs (n = 82 cells, four mice). (F) Distribution of pairwise correlation
images of soma-targeted Ace-mNeon2 expression in NDNF+ interneurons. Scale coefficients of mean firing rates across a sliding 20-ms window, as determined
bar, 50 mm. (b) Representative fluorescence-time traces from individual cells for 3-s intervals, before (black) and after (red) air puff for NDNF+ interneurons
(inverted for visualization purposes). The vertical dashed line in red indicates (2300 cell pairs in the same fields of view, six mice; P = 0.0021; Wilcoxon
air-puff onset. Gray ticks denote identified spikes. (c) Z-scored firing rate matched-pairs signed rank test for before versus after air puff). The mean ± SEM
for all NDNF+ interneurons (n = 574 cells, six mice). Cells are arranged in order correlation coefficient values are shown above the distribution plots. (G) Same
of decreasing spike modulation indices (Materials and methods). (d) Firing as (F) for VIP+ interneurons (n = 442 pairs, five mice; P = 0.0011). (H) Same
rate (mean ± SEM) aligned to air-puff onset for activated (dark blue) and as (F) for SST+ interneurons (n = 334 pairs, six mice; P < 0.0001). (I) Same as (F)
suppressed (cyan) fractions. Pie chart insets indicate the percentage of cells with for PNs (n = 51 pairs, four mice; P = 0.03).
diameter, which is a commonly used, simple garding both the fractions of cells that increased substantial state-dependent modulation, with
metric of arousal (Fig. 2A) (53, 54). The evoked or decreased their spiking and the magnitudes 78% of SST+ cells increasing their spontaneous
arousal was often accompanied by the onset of of spike rate modulation (Fig. 2, B to E). spiking rates and only 13% decreasing their
locomotion and robust modulations of spiking We imaged the spiking dynamics of ~1200 spiking rates (Fig. 2D). SST+ cells also under-
across all neuron classes studied (Fig. 2, B to E). neurons across the different cell types. Upon went the greatest increases in spike rates among
However, the distinct cell classes differed re- arousal, SST+ interneurons exhibited the most the cell types imaged, from 2.5 ± 0.4 Hz (mean ±

SEM) under baseline conditions to 22.5 ± 1.2 Hz We performed dual-polarity multiplexed volt- SST+ and VIP+ cells had positive intrapopula-
after air puff (Fig. 2D). Sixty-three percent of age imaging (DUPLEX) in awake mice, where tion correlations, whereas SST+-VIP+ pairs had
NDNF+ interneurons also increased their spik- we targeted Ace-mNeon2 and pAce either to negatively correlated dynamics, and that the
ing, from 2.6 ± 0.3 to 16.8 ± 0.7 Hz, whereas two distinct inhibitory or to one excitatory and dynamics of NDNF+ cells were only weakly
28% decreased their spiking after air puff (Fig. one inhibitory cell type by using Cre- and Flp- correlated with those of other NDNF+ cells and
2B). VIP+ neurons and PNs exhibited compa- dependent expression strategies (Fig. 3A and VIP+ cells. These results are consistent with
rable state-dependent responses, with half of Materials and methods). We virally expressed paired recordings in live-brain slices that have
each population showing increased spiking Cre-dependent Ace-mNeon2 and Flp-dependent revealed a bidirectional inhibition between
(53 and 56%, respectively) and a third show- pAce in area V1 of NDNF-Cre+/VIP-Flp+ or SST+ and VIP+ neurons and nonoverlapping
ing suppressed spiking (34 and 39%, respec- SST-Cre+/VIP-Flp+ double-transgenic mice (Fig. 3, synaptic inputs to the two subtypes (55, 62).
tively) after air puff. These latter two classes B to G). Alternatively, we injected recombinase-
also had similar spike rate increases during dependent Ace-mNeon2 and pAce viruses along DUPLEX imaging in hippocampal area CA1
arousal (from 3.8 ± 0.5 to 13.7 ± 0.8 Hz for with AAV-CaMKII-Cre into VIP-Flp+ mice (fig. Past studies using GEVIs have assessed rela-
VIP+ cells and 0.9 ± 0.3 to 11.8 ± 1.6 Hz for S21). High-speed epifluorescence imaging in tionships between specific neuron classes and
PNs) (Fig. 2, C and E). That each neuron type a single color channel revealed the joint dy- the local field potential (LFP) in hippocampal
that we studied exhibited a subset of cells namics of NDNF+ and VIP+ cells, SST+ and VIP+ area CA1 (23, 63). Using DUPLEX imaging, we
that decreased their spiking argues against cells, or PNs and VIP+ cells, respectively, in these tracked the concurrent dynamics of two CA1
the possibility that the net spiking increases mice while they were awake but head restrained neuron classes by expressing our two red
were motion-related artifacts stemming from (Fig. 3, B to G; and fig. S21). GEVIs in distinct cell types (fig. S23) or by
the locomotion that often occurred after the To estimate the degree of functional con- using our two green GEVIs in tandem while
air puff. Further supporting this point, dur- nectivity within and across cell types, we com- also recording LFPs (Fig. 4A and fig. S24). In
ing air-puff trials in which there was no loco- puted pairwise correlation coefficients for the a running mouse expressing Ace-mNeon2 in
motor increase, we still observed comparable time (t)–varying activity traces, DF/F(t), as well entorhinal cortex (EC)–projecting excitatory
modulations in neural activity (fig. S19). Among as for the spike rates of cell pairs in the same neurons in the stratum pyramidale (~150 to
the cell types imaged, the air puff–evoked fields of view. NDNF+ and VIP+ interneurons 200 mm below a glass window) and pAce in
rise in spiking among SST+ cells returned the had positive intrapopulation correlations but SST+ interneurons in the stratum oriens (~25
fastest toward baseline levels, whereas VIP+ NDNF+- VIP+ pairs had virtually uncorrelated to 150 mm below a glass window), we tracked
and NDNF+ interneurons, which receive in- dynamics (Fig. 3, D, H, and I). The mean ± SEM 10 excitatory and 18 SST+ cells in the same
hibitory inputs from SST+ cells (47, 55), re- correlation coefficients were 0.19 ± 0.02 (DF/F) field of view and distinguished the two cell
turned more gradually to baseline spiking and 0.15 ± 0.02 (spike rates) for NDNF+-NDNF+ types by the opposing polarities of their op-
levels (Fig. 2, B to D). pairs, 0.31 ± 0.04 (DF/F) and 0.29 ± 0.09 (spike tical spike waveforms (Fig. 4, B to D). Spike
To characterize the degree of spiking syn- rates) for VIP+-VIP+ pairs, and −0.09 ± 0.04 detection fidelity indices (d′) in CA1 neurons
chronization between the cells of each type, (DF/F) and 0.02 ± 0.02 (spike rates) for NDNF+- ranged from 10.6 to 17.6, implying spike de-
we computed pairwise correlation coefficients VIP+ pairs (Fig. 3I), in accord with reports that tection error rates of 0.007 spike min–1 or less
for the time-dependent spike rates of each cell the two subtypes may receive distinct inputs or (fig. S18) (38).
pair and compared the distributions of these not strongly target each other (47, 48). SST+ interneurons had greater baseline spike
coefficients as determined before versus after By comparison, DUPLEX imaging of SST+ rates and burstiness than the EC-projecting
air puff. Both SST+ and VIP+ cells exhibited and VIP+ cells revealed slightly negatively cor- neurons [19.8 ± 1.8 versus 14.7 ± 0.5 Hz (mean ±
positive intrapopulation coupling of sponta- related intertype dynamics, whereas the in- SEM), P < 0.01] (Fig. 4E). We assessed the ex-
neous activity (56), which rose slightly during tratype dynamics were positively correlated tent to which spiking by the excitatory and
arousal, especially among SST+ cells. By com- [mean ± SEM correlation coefficients were inhibitory neurons was phase-locked to mem-
parison, NDNF+ interneurons and PNs showed 0.37 ± 0.04 (DF/F) and 0.11 ± 0.03 (spike rates) brane potential oscillations in these specific
weaker intrapopulation correlations (57), which for SST+-SST+ pairs, 0.51 ± 0.05 (DF/F) and cell types or to CA1 LFP signals. Across a total
were unaltered by air puff (Fig. 2, F to I). 0.30 ± 0.07 (spike rates) for VIP+-VIP+ pairs, of 157 neurons, spiking from both cell types
and −0.22 ± 0.04 (DF/F) and −0.04 ± 0.01 (spike was locked with greater precision to intrapop-
Dual polarity multiplexed voltage imaging rates) for SST+-VIP+ pairs] (Fig. 3, G, J, and K). ulation subthreshold oscillations than to the
Although arousal increases the net spontane- DUPLEX imaging of PNs and VIP+ cells re- LFP; notably, SST+ cells and EC-projecting PNs
ous activity of multiple cell subtypes (Fig. 2), vealed very low correlations between the two spiked at distinct phases of their intrinsic
how it alters the fine-scale dynamic relation- neuron types [−0.009 ± 0.03 (DF/F) and 0.06 ± theta (4 to 9 Hz) membrane potential dynam-
ships between two or more subtypes remains 0.02 (spike rates)], whereas the intrapopulation ics [13.8° ± 2.6° and −1.3 ± 4.8°, respectively
unknown. Thus, we turned to simultaneous correlation coefficients were higher [0.23 ± (mean ± SEM, P < 0.01)] (Fig. 4, F to H). Spik-
multipopulation voltage imaging using our 0.10 (DF/F) and 0.15 ± 0.08 (spike rates) for ing in EC-projecting cells but not SST+ cells
improved GEVIs. PN pairs; 0.32 ± 0.07 (DF/F) and 0.16 ± 0.05 also consistently phase-locked to the LFP (P <
Multipopulation recordings using fluores- (spike rates) for VIP+-VIP+ pairs] (Fig. 3, L and 0.001 and P = 0.82, respectively) (Fig. 4, G
cent Ca2+ or voltage sensors were previously M; and fig. S21). and H).
done by targeting spectrally orthogonal sen- To examine how spiking by one interneuron We next assessed the relationships between
sors to distinct cell types (18, 58–61). We rea- subtype relates to membrane potential dy- the LFP and the cell type–specific subthresh-
soned that a pair of GEVIs operating in the namics of cells in the same or another inter- old dynamics, as well as those between the
same spectral channel but with opposite re- neuron subtype, we computed cross-correlation subthreshold dynamics of the two subtypes.
sponse polarities should be mutually compa- functions between one cell’s spike rate and We found strong subthreshold coherence
tible, allowing studies of two cell types per the subthreshold part of another cell’s time- between both cell types and the LFP in the
color channel. The cells of each type can be varying activity trace, DF/F(t), averaged over theta frequency band (P < 0.0001 and P <
distinguished by the directionality of their all pairs of cells of the two subtypes in the same 0.0001, respectively) (Fig. 4, I to L; and fig.
optical spike waveforms (fig. S20). field of view (fig. S22). The results verified that S24, A to D). We also found strong intra- and

A Intracranial delivery of 0.4 kHz DUPLEX imaging E SST-Cre+/DIO-Ace-mNeon2; VIP-Flp+/fDIO-pAce G

opposite polarity GEVIs in the green channel Raw
AAV/CAG-DIO-Ace-mNeon2 1 1
AAV/CAG-fDIO-pAce
Non-overlapping expression n 505 nm
of the GEVIs in two neuron types
pes LED
4 0
Cre+/Flp+ Activity mask 3
V1 2 6
5 4
1
7 -1
7
F SST VIP
NDNF-Cre+/DIO-Ace-mNeon2; VIP-Flp+/fDIO-pAce 7
B Raw D
1 6
1
5
11
Activity mask 19 9 26
4
25
21
0 4
24 6 13
23 17
16 28
5 10 18 21
2 8 1 3
20 11
3 12
7 15 22 -1
27
14 2
NDNF VIP
5% ΔF/F
1
28
1s 200 ms
27 NDNF-NDNF SST-SST PN-PN
VIP-VIP VIP-VIP VIP-VIP
26 H NDNF-VIP J SST-VIP L PN-VIP
0.4 ΔF/F 0.4 ΔF/F 0.4 ΔF/F
25
24
0.2 0.2 0.2
23
22
0 0 0
21 -1 0 1 -1 0 1 -1 0 1
20
0.4 Firing rate 0.4 Firing rate 0.4 Firing rate
19
Fraction
18
0.2 0.2 0.2
17
16
0 0 0
15 -1 0 1 -1 0 1 -1 0 1
Corr. coefficient
14
13
I ΔF/F K ΔF/F M ΔF/F
1.0 0.004 1.0 0.004 1.0
12 n.s. 0.05
0.016
11 0.5 <0.001 0.04 0.5 0.5
10 0.004 n.s.
0.0 0.0 0.0
9
8
-0.5 -0.5 -0.5
7
Firing rate Firing rate Firing rate
6 1.0 0.004 1.0 1.0
0.008 n.s.
Corr. coefficient
5
<0.001 n.s.
4 0.5 0.5 0.5 n.s.
0.016 0.008
0.01
3
0.0 0.0 0.0
2
5% ΔF/F
1
-0.5 -0.5 -0.5
1s 200 ms Intra- and inter-type
Fig. 3. DUPLEX captures the concerted activation dynamics between cell the DF/F traces (upper triangle) and spiking rates (lower triangle), determined
types in V1. (A) Schematic of AAV injections and experimental setup. using a 20-ms sliding window for the cells in (C) and (F), respectively.
(B to G) Example DUPLEX recordings from two targeted cortical cell types. (B) and (H) Distribution of pairwise correlation coefficients for DF/F traces (top) and
(E) show raw epifluorescence (top) and activity-mask (bottom) images of single spiking rates (bottom) across all mice for NDNF/VIP recordings (n = 1795 NDNF-
fields of view from a (B) NDNF-Cre+/VIP-Flp+ mouse expressing Ace-mNeon2 in NDNF pairs, 77 VIP-VIP pairs, and 373 NDNF-VIP pairs; 208 NDNF-neurons and
NDNF+ interneurons (green) and pAce in VIP+ cells (blue) and (E) SST-Cre+/ 54 VIP-neurons, five mice). (I) Mean ± SEM pairwise correlation coefficients for DF/F
VIP-Flp+ mouse expressing Ace-mNeon2 in SST+ interneurons (green) and pAce in traces (top) and spiking rates (bottom) for the NDNF/VIP dataset in (H) (n = 17 fields
VIP+ interneurons (blue). Active cells are numbered. Scale bar, 50 mm. (C) and of view). P values are italicized (one-sample Wilcoxon test against zero). n.s., not
(F) show DF/F traces from the cells numbered in (B) and (E), respectively. Ace- significant. (J and K) Same as (H) and (I) for SST/VIP recordings (n = 28 SST-SST
mNeon2 traces are inverted for visualization purposes. The regions enclosed pairs, 33 VIP-VIP pairs, and 67 SST-VIP pairs; 25 SST+ neurons and 29 VIP+ neurons,
by the dashed boxes in (C) and (F) are shown at an expanded time scale to the right. 10 fields of view, three mice). (L and M) Same as (H) and (I) for PN/VIP recordings
Gray ticks indicate identified spikes. (D) and (G) show intra- and interpopulation (n = 17 PN-PN pairs, 34 VIP-VIP pairs, and 61 PN-VIP pairs; 20 PNs and 22 VIP+
correlation coefficient matrices of pairwise, zero time-lag correlation coefficients of neurons, eight fields of view, three mice).

A Concurrent 0.6 kHz DUPLEX imaging

B SST-Cre+/DIO-Ace-mNeon2; CaMKII-pAce D E
SST EC-projecting
and LFP recordings in hippocampal CA1
1
AAV/CAG-DIO-Ace-mNeon2 (CA1)
Cumul. Prob.
AAV/Retro-CaMKII-pAce (EC)
Density
16x Obj.
WD 3 mm 0.5
NA 0.8
LFP
SST-Cre+ glue 32
4 11
gel 12
1 6 14
3 8 0
7 4 8 0 0.2 0.4 0.6
10 17 13
5 2 10 9 ∆F/F (%) ISI (s)
CA1 15 1 18
16 9 6 5 7
C LFP (theta power V2/Hz) Theta power (mV2)
0 0.05 0.1
Speed (cm/s) Speed (cm/s)
0 10 20
SST EC-projecting
∆F/F
1 -6.8
-4.4
-6
-4.3
5 -8.8
-5.1
-8.8
-6
-4.3
10 -6.3
-5.1
-6
-5.7
-4.6
15 -3.5
-4.4
-3.3
-2.4
1 3.7
4.7
4.3
3.7
5 5.1
4.8
6.5
3.8
6.8
7.2
10
2s 200 ms
F Raw
G 90
H 90
500 (µV) 200

120 60 120 0.3 60
0.1 0.2
150 30 150 30
LFP
0.05
0.1
5 Hz - 9 Hz filtered
180 0 0 180 0 0
210 330 210 330

8.8 (% ∆F/F) 2.5
240 300 240 300

270 270
SST
90 90
120 60 120 0.3 60
0.06
150 30 150 0.2 30
0.04
0.02 0.1
EC-proj.
180 0 0 180 0 0
7.2
210 330 210 330

1.9
240 300 240 300

270 270
500 ms
LFP/Vm Coherence (%) LFP/Vm Coherence (%)
I 0.4 0.3 0.2 0.1 0 J K 0.25 0.2 0.15 0.1 0.05 0 L
Coherence (%)
Coherence (%)
15 15
SST EC-projecting
Excitatory
Inhibitory
LFP/Vm
LFP/Vm
10 Shuffle LFP 10 Shuffle LFP
5 5
0 0
5 10 15 20 25 30 5 10 15 20 25 30 5 10 15 20 25 30 5 10 15 20 25 30
Frequency (Hz) Frequency (Hz) Frequency (Hz) Frequency (Hz)
Fig. 4. DUPLEX uncovers cell class–specific contributions to the LFP in fields of view. Note the opposite polarities within similar dynamic ranges for
hippocampal CA1. (A) Schematic of AAV injections and experimental setup. Ace-mNeon2 and pAce. (E) Cumulative distributions of interspike intervals (ISIs)
(B) Representative raw epifluorescence image (top) and spatial footprint and firing rates (inset; sp/s, spikes per second) for all recorded neurons
of negative-polarity signals (green) and positive-polarity signals (blue) of 18 and (n = 55 SST+ interneurons and 102 projection neurons, six fields of view, one
10 identified neurons, respectively (bottom). Scale bar, 50 mm. (C) DF/F traces mouse). The shaded area represents the 95% confidence interval (CI). (F) Phase
(bottom) for all the neurons in (B) along with wheel speed and LFP theta relationships between spike and theta oscillations extracted from the LFP or
(top). Ace-mNeon2 traces are inverted for visualization purposes. The first 500 ms cellular transmembrane voltage oscillations for projection neurons and SST+ cells
are shown at the expanded time scale to the right. Gray ticks denote identified recorded simultaneously. Shown here are the raw LFP trace (top) and the theta-
spikes. (D) Fluorescence trace amplitude distributions for all neurons across all filtered (5 to 10 Hz) (center) and unfiltered (bottom) excitatory and inhibitory

traces. (G) Spike-theta phase relationship for the two neurons in (F). Color and gray timing relative to the theta cycle was computed using the circular mean. (I to L) LFP-
represent the optically acquired theta transmembrane voltage (Vm) and theta LFP, subthreshold coherence of all neurons in (C), showing that a fraction of [(I) and
respectively. Note that both neurons are phase-locked to their own Vm and to the LFP (J)] SST+ interneurons and [(K) and (L)] EC-projecting excitatory neurons are phase-
but with approximately opposite phases. (H) Polar histogram of the probability density locked to theta LFP. For (I) and (K), raster plots are sorted in decreasing order of
of the average spike-theta phase relationship for all 157 neurons, computed against theta-band coherence strength. For (J) and (L), LFP coherence is averaged across
their respective theta Vm (top) or theta LFP (bottom). For each neuron, average spike all neurons in (I) and (K). The shaded area represents the 95% CI.
interpopulation coherence in the beta fre- SST+ cells were positively modulated by the air projecting cells did [log2 fold change (mean ±
quency band (15 to 25 Hz) (fig. S24, E to H), in puff [mean ± SEM; median values: 0.43 ± 0.05; SEM): 0.0 ± 0.2 for EC-projecting cells versus
contrast to the lack of beta frequency band 0.57 (P < 0.0001, one-sample Wilcoxon test)], 0.9 ± 0.1 for ACC-projecting cells; P = 0.59 and
coherence with the LFP (fig. S24, A to D). whereas VIP+ cells imaged concurrently at the P < 0.0001; rank sum test between real data
same tissue depth and in the same fields of and spike trains circularly permuted in time]
DUPLEX imaging in live flies view more commonly had negative values of the (Fig. 5, O to Q; and fig. S26). We also observed
To showcase the use of DUPLEX for multi–cell spike modulation index [mean ± SEM; median greater coherence in subthreshold activity after
type recordings across species, we expressed values: −0.07 ± 0.05; −0.12 (P = 0.14)] (Fig. 5H). the behavioral transition, but interestingly,
pAce in the g-aminobutyric acid–signaling DUPLEX studies in PNs and VIP+ cells (~150 each subclass synchronized at distinct frequen-
(GABAergic) MBON-g1pedc>a/b neuron and to 200 mm below the pia) revealed uncorrelated cies (7.3 ± 0.3 and 4.2 ± 0.2 Hz, respectively, for
Ace-mNeon2 in the PPL1-a′2a2 dopaminergic dynamics for these cell classes before and after ACC- and EC-projecting cell pairs; P < 0.0001;
neuron in fruit flies (fig. S25A). Both these cells air puff (mean ± SEM correlation coefficients rank sum test computed for run epochs against
receive excitatory olfactory inputs from mush- were 0.04 ± 0.02 before versus 0.05 ± 0.02 after levels estimated from neurons belonging to
room body Kenyon cells, and MBON-g1pedc>a/b air puff) (Fig. 5, I to K). The spike modulation different fields of view) (Fig. 5, R to T).
also exerts feedback inhibition onto PPL1-a′ indices revealed a small positive modulation
2a2 axons (64). When we exposed the double- of both populations by the air puff [mean ± Simultaneous voltage imaging of
labeled flies to odorants, DUPLEX imaging SEM; median values: 0.19 ± 0.09; 0.44 (P = 0.06, three cell populations
of PPL1-a′2a2 axons and MBON-g1pedc>a/b one-sample Wilcoxon test) and 0.08 ± 0.05; 0.07 Together, our indicators enable discrimina-
dendrites revealed that MBON-g1pedc>a/b (P = 0.14), respectively] (Fig. 5L). tions of distinct targeted subtypes or projection-
was strongly activated throughout exposure to specific cell subclasses based on the polarity of
an attractive odor [apple cider vinegar (ACV)] Dual-color voltage imaging of hippocampal optical spike waveforms as well as spectral or-
but was more transiently activated by a repul- projection neurons thogonality. Thus, in principle, the four GEVIs
sive odor [benzaldehyde (BEN)]. By contrast, Although DUPLEX can dissect activity patterns are mutually compatible and allow recordings
PPL1-a′2a2 was strongly inhibited by ACV among nonoverlapping populations, such as from up to four distinct neuron types at once.
but weakly excited by BEN, in accord with discrete excitatory and interneuron subtypes, To demonstrate dual polarity and dual-color
the known inhibitory connection from MBON- when the level of overlap between two neuronal multiplexing, we performed proof-of-concept
g1pedc>a/b to PPL1-a′2a2 (fig. S25, B and C). subclasses is unknown, their activities can in- voltage imaging from three distinct cell pop-
stead be distinguished by simultaneous dual- ulations simultaneously.
Interpopulation correlations of spiking color voltage imaging. We used Ace-mNeon2 and In an SST-Cre+/VIP-Flp+ double-transgenic
dynamics during state transitions VARNAM2 to examine the extent to which two mouse, we injected AAVs encoding both Cre-
We next used DUPLEX to examine how state subclasses of CA1 projection neurons, target- dependent Ace-mNeon2 and Flp-dependent
transitions affect the interactions between cor- ing distinct cortical regions, modulate their spik- pAce into V1, along with a retrograde AAV ex-
tical subtypes. In green DUPLEX recordings ing during a rest-to-run behavioral transition. We pressing CaMKII-VARNAM2 in area AM to
from NDNF+ and VIP+ interneurons in layer 1 of expressed Ace-mNeon2 and VARNAM2 in an- label a subset of excitatory projection neurons
V1 (~50 to 100 mm below the pia), spike rates terior cingulate cortex (ACC)– and EC-projecting (Fig. 6A). We then performed voltage imaging
were uncorrelated across the two populations CA1 PNs, respectively, using retrograde AAV in V1 using dual-color, widefield epifluorescence
under baseline conditions (Fig. 5, A to C), and labeling. We then performed simultaneous microscopy to record from SST+, VIP+, and ex-
air puff–induced arousal had no impact on this dual-color voltage imaging in an injected citatory neurons concurrently (Fig. 6, B and C).
uncorrelated behavior [mean ± SEM correla- mouse, which was head restrained on a run- To image three neuron classes at once in
tion coefficients: 0.07 ± 0.02 before air puff ver- ning wheel, and used a brief air puff to elicit a the hippocampus, we retrogradely labeled EC-
sus 0.10 ± 0.03 after] (Fig. 5C). To quantify the rest-to-run behavioral transition (Fig. 5M). and lateral septum (LS)–projecting CA1 PNs
effects of air puff on the spiking of individual cells, We imaged a total of 101 ACC-projecting and SST+ interneurons, using viruses to express
we also computed a spike modulation index and 34 EC-projecting neurons from five fields Ace-mNeon2, pAce, and pAceR, respectively, in
from −1 to 1 (Material and methods). Spiking by of view in the same animal, with a single field a SST-Cre+ mouse (Fig. 6D). With simultaneous
NDNF+ and VIP+ cells was significantly positive- of view containing as many as 30 and 13 cells dual-color imaging, we captured the concurrent
ly modulated by air puff [mean ± SEM; median of the two subclasses, respectively (Fig. 5, M spiking and subthreshold activity of as many as
values: 0.27 ± 0.02; 0.32 (P < 0.0001, one-sample and N; and fig. S26). Both Ace-mNeon2 and 57 cells in one field of view belonging to the
Wilcoxon signed rank test against zero) and VARNAM2 captured a diversity of cell dynam- three subpopulations (21 EC-projecting, 15 LS-
0.11 ± 0.03; 0.08 (P = 0.006), respectively] (Fig. 5D). ics, including isolated spikes, spike bursts, and projecting, and 21 SST+ cells). These ensembles
In DUPLEX recordings from SST+ and VIP+ subthreshold oscillations. comprised nonoverlapping cell populations
interneurons (~200 to 250 mm below the pia), We assessed whether the ACC- and EC- (Fig. 6E), and all three GEVIs reported spiking
spiking by the two subtypes became significantly projecting cell ensembles had distinct supra- and at high signal-to-noise ratios (Fig. 6F).
more anticorrelated after air puff (mean ± SEM subthreshold dynamics across the behavioral
correlation coefficients: –0.02 ± 0.01 before air state transition (6–9). Whereas EC-projecting Conclusion
puff versus −0.14 ± 0.03 after) (Fig. 5, E to G). The cells did not significantly change their spike Our suite of mutually compatible GEVIs allows
spike modulation indices also revealed that most rates after the onset of locomotion, ACC- simultaneous high-speed voltage imaging of

Fig. 5. Dual population record- A B 0 5 cm s C -1

before puff D
Speed after puff 0.3
0.4
0.3 mm
ings in V1 and CA1 during NDNF-Cre+/DIO-Ace-mNeon2+;VIP-Flp+/fDIO-pAce Pupil diameter (∆d)
0.6
908 NDNF-VIP pairs n.s. NDNF
Raw VIP
state transitions. DUPLEX record- 0.2
Corr. coefficient
8
0.2
Fraction
Fraction
7 0.4
ings in V1 are shown in (A) to 6 0.0
(L). (A) Representative raw Activity mask 2

7
4
5
0.2
0.1
1 5 4 -0.2
epifluorescence (top) and activity- 8 6
3 3
0
mask (bottom) images of a NDNF
2
-1 0 1
-0.4 0
-1 -0.5 0 0.5 1
VIP 1
Corr. coefficient Spike modulation index
single field of view from a 1s
E F Speed G H
NDNF-Cre+/VIP-Flp+ mouse
0.3 mm
SST-Cre+/DIO-Ace-mNeon2+;VIP-Flp+/fDIO-pAce
Pupil diameter (∆d) 0.4 0.3
+
expressing Ace-mNeon2 in NDNF Raw 0.6
242 SST-VIP pairs SST
6 VIP
0.2
interneurons (green) and pAce in
Corr. coefficient
0.0156* 0.2
Fraction
Fraction
5
0.4
VIP-cells (blue). Scale bar, 50 mm. Activity mask 4 4
0.0
(B) DF/F traces from the cells 6

1
5 3 0.2 -0.2
0.1
2 3
numbered in (A) aligned to air-puff SST
2
0 -0.4 0
onset (vertical dashed line). Ace- VIP 1 -1 0 1 -1 -0.5 0 0.5 1
mNeon2 traces are inverted for I J 1s
K L
Speed
0.3 mm
visualization purposes. Also shown AAV/CaMKII-Cre/DIO-Ace-mNeon2; VIP-Flp+/fDIO-pAce
Pupil diameter (∆d)
0.4 0.3
12 183 PN-VIP pairs PN
at the top are air-puff onset, Raw 11 0.6
VIP
10 0.2 n.s.
locomotion speed, and pupil diam-
Corr. coefficient
0.2
Fraction
Fraction
9
8 0.4
eter. (C) Distribution of pairwise Activity mask
3 12 7
0.0
2 4 6
intertype correlation coefficients of 10 7 5
1
5 0.2
-0.2
0.1
11
8 4
spike rates (left) determined using 6 9
3
PN 2 0 -0.4 0
a 20-ms sliding window for a 3-s VIP 1
-1 0 1 -1 -0.5 0 0.5 1
interval before (black) or after (red) 1s
M Simultaneous 0.6 kHz dual-channel N 0 18 cm s O -1
air puff (n = 908 NDNF-VIP pairs voltage imaging using green and red GEVIs Speed
+ + AAV/Retro-CaMKII- ACC-projecting EC-projecting
from 393 NDNF and 188 VIP Ace-mNeon2
20
AAV/Retro-CaMKII- Speed 1
cells, seven mice) and mean ± SEM
Speed (cm/s)
VARNAM2 2
16x Obj.
WD 3 mm 3
correlation coefficients by mouse ACC
NA 0.8 4 10
5
(right) (n = 7 mice; one-tailed EC 6
7
CA1
Wilcoxon matched-pairs signed 8
9
0
0 5 10 15
AP +0.8 mm AP -4.8 mm
rank test). (D) Distribution of CaMKII-Ace-mNeon2 (ACC-projecting PNs);
10
11
Time (s)
spike modulation indices for all CaMKII-VARNAM2 (EC-projecting PNs) 12

13 P
NDNF+ and VIP+ neurons in (C). Raw green channel 14 ACC
Firing Rate (spikes/s)

15 EC
20
(E to G) Same as (A) to (C) 16
17
shuffle
18
for DUPLEX recordings in SST- 19 10
Green spatial filters
Cre+/VIP-Flp+ mice (n = 242 SST- 20
21
VIP pairs from 103 SST+ and 79

22
23 0
0 5 10 15
Time (s)
VIP+ cells, six mice; *P < 0.05, one-
24
25
Raw red channel
26
tailed Wilcoxon matched-pairs 27
28 Q 0.6
signed rank test). (H) Same as 29 ACC
30
+ + EC
(D) for all SST and VIP neurons. Red spatial filters
1
shuffle
Prob. Density
2 0.4
(I to K) Same as (A) to (C) for 3
4
DUPLEX recordings in AAV-CaMKII- 5
6
0.2
Cre/VIP-Flp+ mice (n = 183 PN-VIP Activity mask 9

7
7
8
20
pairs from 75 PNs and 67 VIP+

8 15 2 12 5
3 14 16 9
24 4 61 12 17 0
13 1 8 3 29 -4 -2 0 2 4
5 10 10
7 11 6 28 11
Fold Change
cells, four mice; one-tailed Wilcoxon 23
27
4
19
18
2
22 25 21 13 26
10
9 30
11
12
matched-pairs signed-rank test). 13
2s
(L) Same as (D) for all PNs and R run rest ACC shuffle S run rest EC shuffle T run rest ACC/EC shuffle
VIP+ neurons. Dual-color voltage 0.5 0.5 * 0.5
Coherence
Coherence
Coherence
* *
recordings from hippocampal 0.4 0.4 0.4
*
projection neurons are shown in 0.3 0.3 0.3
(M) to (T). (M) Schematic of 5 10 15 5 10 15 5 10 15

Frequency (Hz) Frequency (Hz) Frequency (Hz)
the experimental approach (top)
and representative field of view of
the red and green channels in grayscale, their respective overlay of spatial filters estimated by EXTRACT for each spiking neuron in each spectral channel, and
the overlay of the spatial filter contours for all 30 and 13 identified projection specific neurons, respectively (bottom). Scale bar, 50 mm. (N) DF/F traces for all neurons
in (M), aligned to the rest-run transition. Also shown at the top are the onset of five consecutive air puffs and wheel speed. (O) Average mouse speed for five
trials (black). Individual trials are shown in gray. (P) Average firing rate (250-ms window) for each projection-specific class. Shuffle is computed by random circular
permutations of each neuron’s spike train. The shaded area represents the 95% CI. (Q) Fold-change in the cells’ spiking rates, computed across 2-s intervals
before and after air puff. Shuffle is computed using a random circular permutation of each neuron’s spike train. (R to T) Average pairwise coherence of the
subthreshold dynamics of neurons belonging to the ACC-projecting subclass [(R); n = 101 neurons], the EC-projecting subclass [(S); n = 34 neurons], and across
the two subclasses [(T); n = 135 neurons, five fields of view, one mouse]. P < 0.0001 (rank sum test computed during epochs of running, against levels estimated from
neurons belonging to different fields of view). Shaded areas represent the 95% CI.

A Dual-polarity/dual-color voltage imaging in V1 F 1

2
AAV/CAG-DIO-Ace-mNeon2 AAV/Retro-CaMKII-
+AAV/CAG-fDIO-pAce VARNAM2 3
AM 20x Obj. 4
V1
WD 2 mm
NA 1.0 5
SST-Cre+/VIP-Flp+
6
mouse V1
7
8
AP -3.5 mm AP -3 mm 9
B 10
EC-projecting
Raw Activity mask
11
channel
Green
6 4 8 5
9 3 12
7 10 13
14
channel
2
Red
15
1
16
Z-score 14
17
2
Overlay
6 4 5 18
8
1 9 3
7 10 19
20
C 21
10 1
Z-score 19
9 2
3
8
4
SST
7 5
6 6
LS-projecting
7
5
8
4 9
VIP
3 10
11
2
12
PN
1 13
1s 14
15
1
D Dual polarity/dual color voltage imaging in CA1
2
AAV/Retro-CaMKII- AAV/Retro- 3
AAV/CAG-DIO-pAceR Ace-mNeon2
Z-score 15
CaMKII-pAce 4
16x Obj.
WD 3 mm 5
NA 0.8
CA1 6
7
EC LS
8
CA1
SST-Cre+ 9
mouse
AP -2 mm AP -4.8 mm AP 0.3 mm 10
SST
11
E 12
channel
channel
Green
13
Red
14
15
17 9
Projecting
5 21 17
1 18 16
8 3 1
SST
13 8
EC-
19 7 2 12 19 11
15 1114
10 12 5 7 20 6 15 17
4 3 20 2 21 6 16 18 10 9 13
16 4 14
18
8 2 12
Projecting
7
Overlay
4 13 19
14 15
LS-
5 9 1
11 10 3 20
6
21
2s
Fig. 6. Simultaneous dual-polarity and dual-color imaging capture the voltage Ace-mNeon2 and VARNAM2 traces are inverted for visualization purposes.
dynamics of three distinct targeted subtypes or cell subclasses in awake mice. (D) Schematic of AAV injections for three–cell class voltage imaging in
(A) Schematic of AAV injections for three–cell type V1 imaging. (B) Raw CA1. (E) Representative field of view of the red and green channels with the
epifluorescence (left) and activity-mask (right) images of a single field of view respective spatial footprint of the identified neurons belonging to each of
in the green and red channels and overlay. Scale bar, 50 mm. (C) DF/F the three cell classes. Scale bar, 50 mm. (F) DF/F traces for all neurons in (E),
traces from the cells numbered in the mask overlay image in (B) and representing EC- and LS-projecting excitatory neurons and SST+ interneurons
representing SST+ interneurons, VIP+ interneurons, and PNs expressing expressing Ace-mNeon2 (green), pAce (blue), and pAceR (light red),
Ace-mNeon2 (green), pAce (blue), and VARNAM2 (dark red), respectively. respectively. Ace-mNeon2 traces are inverted for visualization purposes.

multiple genetically identified ensembles in To access cells in deeper cortical layers, the and VIP+ interneurons during arousal. In the
awake animals. Notably, our indicators can pro- use of patterned illumination strategies to re- hippocampus, DUPLEX revealed cell type–
vide large datasets for quantitative characteriza- duce background fluorescence may improve specific subthreshold activity, which allowed
tions of the spiking patterns of targeted neuron imaging depths (16, 22, 24, 68). Alternatively, us to characterize how the dynamics of excit-
types in behaving flies and mice. This provides a imaging of voltage dynamics in infragranular atory and inhibitory populations relate to those
starting point for future studies in which the cortical layers or deep brain areas with a micro- of the LFP. Our results fit with the hypothesis
fine-scale dynamics between targeted cell types prism (69) or gradient index (GRIN) microendo- that excitatory neurons contribute more strong-
can be examined toward dissecting the circuit scope (70) should also be feasible, especially ly to the LFP because of their spatial organi-
basis of animal behavior. The FRET-opsin GEVIs if high numerical aperture lenses (71) for en- zation and temporal synchronization (74).
introduced here are bright and can be used with hanced fluorescence collection are used in- Whereas DUPLEX allows recordings from
illumination levels far lower than those needed stead of those typically used for Ca2+ imaging nonoverlapping cell types, our green and red
for imaging opsin GEVIs (16, 22, 23, 27). They (72). Further, although the densities of labeled indicators can be combined for studies of two
are hence compatible with both widefield epi- cells in our recordings (<370 cells mm−2) were cell populations that overlap or for which the
fluorescence microscopy and specialized micro- comparable to those of past voltage-imaging degree of overlap is unknown. The ACC- and
scopy methods that use sculpted illumination studies (13, 16, 22, 23), they were low com- EC-projecting CA1 ensembles represent spa-
patterns (16, 22, 24), facilitating broad dissem- pared with the total neuronal density in the tially distinct subpopulations, as seen from
ination of our voltage imaging approaches. mammalian cortex. Thus, our imaging ap- the nonoverlapping expression of Ace-mNeon2
Our V1 data, acquired from ~1200 cells proach may not generalize to studies requiring and VARNAM2 (Fig. 5M). This anatomical
using Ace-mNeon2, reveal that SST+ interneu- pan-neuronal or other dense labeling strat- distinction and the separate dynamical attrib-
rons exhibit a major increase in spontaneous egies. That said, recent work has aimed to utes of the two cell classes (Fig. 5, P to T) are in
firing during arousal, recapitulating the results delineate the large number of mammalian line with the idea that there are heterogeneous
of targeted patch recordings (4). Although neuron types in the cortex (73), so we expect parallel modules in the hippocampus (75).
our results do not confirm the VIP+ neuron- that upcoming studies that seek to differen- Looking ahead, that our two green GEVIs
centered PN disinhibition model (VIP ⊣ SST ⊣ tiate the functional attributes of these dis- of opposite polarities can be further combined
PN), which was proposed based on Ca2+ im- tinct cell types will require targeted labeling with our red GEVIs should enable many dif-
aging data (2, 65, 66), the distinct findings strategies such as those used here. With con- ferent voltage-imaging studies of three or four
might stem from key experimental differences, tinued progress in scientific-grade image sen- cell classes simultaneously (Fig. 6). Thus, the
such as the presence or absence of visual sor chips, we also expect improved high-speed new indicators collectively empower neuro-
stimuli (67). However, past interpretations of cameras that will enable voltage imaging over scientists to unravel intercellular interactions
Ca2+ imaging data might be confounded by larger fields of view and of more cells than within and between targeted neuron classes in
subtype-specific differences in intracellular those studied here. awake behaving animals.
Ca2+ dynamics or cytosolic Ca2+ release, which We quantified the veracity of optically de-
in turn may be influenced by state-dependent tected spike trains using the d′ metric, but Materials and methods
neuromodulatory effects (67). A prior alterna- measuring the fidelity of optically detected, Plasmids
tive to Ca2+ imaging was to perform targeted subthreshold membrane potential changes is For high-throughput screening in HEK cells,
whole-cell patch electrode recordings, but the more challenging because the relevant fluo- VARNAM and Ace-mNeon were inserted into
invasiveness of the technique and the avail- rescence signals and noise fluctuations are a modified pCAGGS backbone (18). A nuclear
ability of fewer healthy cells for successful not just frequency-dependent but also non- localization sequence (NLS)–tagged FP re-
patching makes it challenging to sample large stationary. Although the subthreshold compo- porter (mCerulean for VARNAM and mCherry
numbers of cells from sparse populations nents of the estimated neural activity traces, for Ace-mNeon) was fused to the C terminus of
(e.g., VIP+ interneurons) (2). The use of GEVIs DF/F(t), did vary to a modest extent with the the indicator interspersed by a T2A ribosomal
avoids such limitations of electrodes and use of different cell-extraction algorithms (fig. skipping signal. For in vivo voltage imaging, the
Ca2+ imaging, and our suite of four compati- S15), likely due to different efficacies in com- indicators were cloned in AAV backbones. pAAV-
ble GEVIs enables unprecedented studies of putationally removing background fluores- CaMKII-Ace-mNeon2, pAAV-CaMKII-VARNAM2,
spiking in multiple identified neuron classes cence contaminants or heartbeat artifacts, the and pAAV-CaMKII-pAce were cloned by insert-
at once. biological conclusions obtained about cell class ing the respective indicator sequences in place
Notwithstanding, our imaging methodology interactions using spike rate correlations and of enhanced green fluorescent protein (EGFP)
does have limitations. Our widefield one- subthreshold correlations were in general agree- between the Kpn I and Hind III sites of pAAV-
photon epifluorescence imaging studies in the ment with each other (Fig. 3, H to M) and did CaMKII-EGFP (Addgene #50469). Cre-dependent
mouse neocortex, like other recent voltage- not depend on the analytic approach to cell ex- constructs for Ace-mNeon2 and pAceR were
imaging studies (13, 16, 23, 24), were limited to traction (fig. S22). Nonetheless, further progress generated by cloning the respective inverted
supragranular cortical layers, relatively sparse in cell-extraction algorithms that are tailored for open reading frame (ORF) sequences in pAAV-
fluorescence labeling patterns, and moder- voltage-imaging studies remains important. CAG-Flex-EGFP (Addgene #59331) by replacing
ately sized fields of view. Similarly, our studies Traditionally, multipopulation imaging has the floxed EGFP cassette between the Asc I and
of the CA1 hippocampal area were limited been achieved using spectrally orthogonal in- Nhe I sites. Likewise, Flp-dependent VARNAM2
to its two most superficial layers. In these re- dicators (18, 58–61). Yet, likely due to the lim- and pAce constructs were cloned by introducing
gions, by combining high brightness indica- ited set of mutually compatible sensors or the the respective inverted ORF sequences in place
tors with cell type–specific and soma-targeted need for tailored optical hardware, simulta- of the floxed mNeonGreen between the Asc I
labeling strategies, we achieved high signal- neous multichannel imaging is not widely per- and Nhe I sites in pAAV-CAG-fDIO-mNeon-
to-background fluorescence and exception- formed in live animals. DUPLEX offers a Green (Addgene #99133). All AAV constructs
al d′ values for accurate spike detection and simple alternative to distinguish between two above included a Golgi–endoplasmic reticu-
timing estimation (fig. S18), independent of neuron types using one fluorescence channel. lum export sequence and a Kv2.1 proximal
the cell-extraction algorithm used (fig. S15, Here, DUPLEX unveiled anticorrelated dy- restriction and clustering sequence (43) at
B, D, and G). namics among pairs of visual cortical SST+ the C terminus of the indicator sequence for

improved membrane trafficking and periso- 3269) (30, 31) were maintained in DMEM/F12, Fluorescence recordings were obtained using
matic localization, respectively. 10% FBS, 1% penicillin (100 U ml−1), strepto- an Olympus upright microscope. Cells were
mycin (100 µg ml−1), geneticin (500 µg ml−1), visualized under a 1.0 NA 60× water-immersion
Fly stocks and puromycin (2 µg ml−1) as described previ- objective lens. Ace-mNeon and pAce were im-
To create transgenic flies carrying the four ously. About 24 hours before transfection, the aged using a 505-nm light-emitting diode
improved FRET-opsin GEVIs, we synthesized cells were plated on poly L-lysine–coated 12-mm (LED) (Thorlabs) and a filter set composed of a
codon-optimized transcripts for Ace-mNeon2, coverslips or 96-well glass-bottomed plates in 509/22-nm excitation filter, 526-nm dichroic
VARNAM2, pAce, and pAceR using a com- antibiotic-free media. Plasmid DNA (0.5 µg mirror, and 544/24-nm emission filter (Semrock).
mercial gene synthesis service (GenScript) and per 12-mm coverslip or 0.2 µg per well of a VARNAM and pAceR were illuminated using
then inserted them in place of green fluorescent 96-well plate) was transfected using Lipofect- a 565-nm LED (Thorlabs), 560/40-nm excita-
protein (GFP) in the pJFRC7-20×UAS-IVS- amine (ThermoFisher), and cells were imaged tion filter, 585-nm dichroic mirror, and 630/
mCD8::GFP (Addgene #26220) and pJFRC19- 1 to 2 days after transfection. Both cell lines 75-nm emission filter (Semrock). The illumi-
13×LexAop2-IVS-myr::GFP backbones (Addgene served solely as a system for the expression and nation intensity at the sample plane for all
#26224). After verifying the sequences of the characterization of voltage indicators rather indicators was 15 to 20 mW mm−2. Fluores-
plasmids, we inserted the plasmids into the than the subject of investigation. cence time-series images (1 to 5 kHz) were
attP40 or VK00027 phiC31 docking site to captured using a NeuroCCD camera, con-
generate 20×UAS-Ace-mNeon2, 13×LexAop- High-throughput semi-automated voltage screening trolled via NeuroPlex software (RedShirtI-
Ace-mNeon2, 20×UAS-pAce, 13×LexAop-pAce, High-throughput voltage screening (Fig. 1B) maging, GA). Fluorescence response traces
20×UAS-VARNAM2, 13×LexAop-VARNAM2, was performed on a custom-built platform de- were extracted by ranking each pixel by their
20×UAS-pAceR, and 13×LexAop-pAceR flies scribed previously (18). A white-light source signal-to-noise ratios, where the noise was cal-
using a commercial transformation service pE-4000 (CoolLED, UK) was used for sample culated as the root mean square value of the
(Bestgene Inc.). illumination. For imaging Ace-mNeon, we used baseline fluorescence fluctuations. The top 25%
We obtained MB058B, MB296B, and MB085C a 472/30-nm excitation filter, 495-nm dichroic of the SNR-ranked pixels were used to calcu-
split-GAL4 lines from the FlyLight team at mirror, and 520/35-nm emission filter (Semrock). late the DF/F values shown in Fig. 1D and fig.
Janelia Research Campus and R82C10-LexA For VARNAM and mCherry (reporter fluoro- S5. Probe kinetics for activation (τON) and de-
(#54981) from the Bloomington Stock Center. phore in Ace-mNeon constructs), we used a activation (τOFF) were calculated from 5-kHz
We raised all flies on standard cornmeal agar 560/40-nm excitation filter, 585-nm dichroic recordings obtained in HEK cells, by fitting
media under a 12-hour light-dark cycle at 25°C mirror, and 630/75-nm emission filter (Chroma the first 50 ms of the step responses using a
and 50% relative humidity. Technologies Corporation). mCerulean re- double exponential equation as described pre-
porter in the VARNAM constructs was imaged viously (76).
Site-directed saturation mutagenesis using a 455/40-nm excitation filter, 458-nm
Mutagenic libraries were generated on Ace- dichroic mirror, and 480/30-nm emission fil- Surgical dissection of live flies
mNeon-T2A-NLS-mCherry and VARNAM- ter (Semrock). After identifying transfected We created an optical window on the fly head
T2A-NLS-mCerulean backbones using a cells in the reporter channel, time-series flu- using an ultraviolet (UV) laser microsurgery
preestablished protocol (18). Briefly, a set of orescence images were captured in the respec- system, as described previously (77). In brief,
four forward primers containing degenerate tive GEVI channels using an ORCA Flash4.0 we anesthetized the flies by placing them on
codons WKC, NMC, VWG, or DGG at the tar- sCMOS camera (Hamamatsu) at 50 Hz. A sin- ice for 1 min and then transferring them to a
get site was pooled with a single, partially gle pulse of 60 V/0.5 ms was applied using a cooled surface (∼4°C) consisting of an alumi-
overlapping reverse primer, and mutageniz- Grass S48 Stimulator, 1 s after baseline flu- num thermoelectric cooling block. We then
ing polymerase chain reaction (PCR) reac- orescence acquisition (F0). The percent change glued the fly’s thorax with a 125-mm-diameter
tions were set up using CloneAmp polymerase in fluorescence at time t was obtained using fused silica optical fiber (PLMA-YDF-10/125,
(Clontech). After Dpn I treatment to digest the formula Nufern) on a custom-made plastic fixture. To
unmutated template, the linear PCR products DF F ðtÞ F0 minimize head motion, we glued the fly’s head
were circularized using InFusion ligase (Clontech) %¼ 100 to the thorax using a UV light curing epoxy
F0 F0
and transformed in TOP10 competent cells (NOA 89, Norland). After transferring the
(Invitrogen). To obtain up to 19 unique amino Platform automation, sample illumination, mounted fly to the surgery station, we created
acid substitutions at a single target site, 48 col- data acquisition and parallel data analysis were an optical window in the cuticle by laser-
onies were picked and cultured in 96–deep- controlled using custom-written virtual instru- drilling a 150-mm-diameter hole (30 to 40 pulses,
well culture plates. Plasmid DNA was isolated ments in LabView. delivered at 100 Hz, 36 mJ per pulse, as mea-
using Nucleospin 96 Plasmid kit (Macherey- sured at the specimen plane). Immediately
Nagel) on an epMotion 5075 liquid handling Fluorescence voltage imaging in cultured HEK cells after surgery, we applied 1 ml of UV epoxy
workstation (Eppendorf), and purified DNA Whole-cell voltage-clamp recordings were (NOA 68, Norland; refractive index: 1.54; trans-
was collected in 96-well plates. The libraries performed between 22° and 25°C. The bath mission 420 to 1000 nm: ~100%) and cured it
were sequenced after voltage screening to solution contained (in mM) 125 NaCl, 2 KCl, for ~30 s to seal the cuticle opening.
identify the individual mutations and en- 10 HEPES, 30 glucose, 3 CaCl2, 1 MgCl2 (~310
sure that at least 19 variants were obtained mOsm liter−1, pH 7.3). The intracellular solution Visual cortical surgeries
at every site. consisted of (in mM) 125 K-gluconate, 8 NaCl, Animal experiments were performed accord-
0.6 MgCl2, 0.1 CaCl2, 1 EGTA, 4 Mg2-ATP, ing to the guidelines of the National Institutes
Maintenance and transfection of HEK and 0.4 Na-GTP, and 10 HEPES. Patch pipettes had of Health and approved by The John B. Pierce
excitable HEK cells tip resistances of 4 to 5 megaohms, yielding series Laboratory Animal Care and Use Committee.
HEK293 cells (ATCC® CRL-1573TM) were main- resistances of <20 megaohms. In the voltage-
tained in high-glucose Dulbecco’s minimum clamp mode, voltage steps in increments of Intracranial AAV injections
essential medium (DMEM)/10% fetal bovine 20 mV/ 0.5 s were applied using a MultiClamp C57BL/6J wild-type and VIP-Cre, SST-Cre,
serum (FBS). Excitable HEK cells (ATCC CRL- amplifier controlled by pClamp10 software. VIP-Flp, and NDNF-Cre transgenic mice were

purchased from The Jackson Laboratory (JAX fDIO-VARNAM2 were mixed at a 1:1 (v/v) ratio for The coordinates for CA1 injections in Figs. 4
#664, #31628, #13044, #28578, #28536). NDNF- injections in an SST-Cre+/VIP-Flp+ background. and 6, D to F, and figs. S23 and S24, were (in
Cre+/+ or SST-Cre+/+ mice were crossed with mm from Bregma; AP, ML, DV) −1.8, −2.5, −1.1
VIP-Flp+/+ to generate the NDNF-Cre+/VIP- Histology and confocal imaging and −1.3, −1.3, −1.1. The coordinates for retro-
Flp+ and SST-Cre+/VIP-Flp+ double driver lines. For confocal imaging of the cell type–specific grade labeling in the lateral septum (LS) (Fig.
The following AAV vectors were custom- expression of Ace-mNeon2 in area V1 (Fig. 2, 6, D to F), entorhinal cortex (EC) (Figs. 5, M to
produced at the University of North Carolina B to E, subpanel a), injected mice were trans- T and 6, D to F; and figs. S23, S24, and S26),
Vector Core at high titers (>1 × 1013 GC ml−1): cardially perfused with ice-cold Sorenson’s and anterior cingulate cortex (ACC) (Fig. 5, M
AAV/DJ-CAG-DIO-Ace-mNeon2, AAV2/Retro- buffer (pH 7.4) followed by 4% paraformal- to T; and fig. S26) were, respectively (in mm
CaMKII-Ace-mNeon2, AAV/DJ-CAG-fDIO-pAce, dehyde. The brains were isolated and post- from Bregma; AP, ML, DV) 0.3, 0.3, −3; −4.8,
AAV2/Retro-CaMKII-pAce, AAV/DJ-CAG-fDIO- fixed overnight at 4°C. Forty- to 60-µm-thick 3.3, −3.5; and 0.8, 0.2, −1.5. We used the viruses
VARNAM2, AAV2/Retro-CaMKII-VARNAM2, coronal sections comprising the visual cortical AAV/DJ-CAG-DIO-Ace-mNeon2 and AAV2/
and AAV/DJ-CAG-DIO-pAceR. All AAV vec- areas were prepared on a Leica vibratome and Retro-CaMKII-pAce to label SST+ neurons and
tors above encode soma-targeted versions of mounted on gelatin-coated glass microscope EC-projecting neurons, respectively (Fig. 4);
the indicators. AAV5-CaMKII-mCherry-Cre and slides using ProlongGold (Invitrogen). Con- AAV2/Retro-CaMKII-VARNAM2 and AAV2/
AAV2/Retro-EF1A-mCherry-IRES-Cre viruses focal micrographs were obtained using Zeiss Retro-CaMKII-Ace-mNeon2 to label EC-
were purchased from the UNC Vector Core LSM 710 under a 20× air objective. projecting and ACC-projecting neurons,
and Addgene (#55632), respectively. respectively (Fig. 5); AAV2/Retro-CaMKII-Ace-
Mice were used without regard to sex; we Headpost and cranial window implantation mNeon2, AAV2/Retro-CaMKII-pAce, and AAV/
used a total of 21 males and 23 females in Headpost and cranial windows were implanted DJ-CAG-DIO-pAceR to label EC-projecting and
this study. Stereotactic AAV injections were between 25 and 30 days after injection with LS-projecting neurons and SST+ neurons,
performed in 3- to 4-month-old mice under some modifications to a preestablished pro- respectively (Fig. 6); and AAV/DJ-CAG-DIO-
1.5% (v/v) isoflurane anesthesia. Each animal tocol (78). Mice were anesthetized using a pAceR and AAV2/Retro-CaMKII-VARNAM2
received five injections in area V1. The coor- mixture of ketamine (100 mg/kg) and xylazine to label SST neurons and EC-projecting neu-
dinates for the V1 injections (Figs. 1, H to K; 2; (16 mg/kg). A custom-designed titanium head rons, respectively (fig. S23). All viral vectors
3; 5, A to L; and 6, A to C; and figs. S16, S17, S19, post was affixed to the skull using C&B had titers of >1012 genome copies ml−1 and
and S21) were [in mm from Bregma; (AP, ML, Metabond (Parkell), after which a 3-mm- were custom produced at the UNC Vector Core.
and DV): 2.5, 2.5, and 0.2 (site 1); 2.8, 2.0, and diameter circular craniotomy was made over
0.2 (site 2); 3.0, 2.5, and 0.2 (site 3); 3.4, 2.0, the injection area in V1. A sterile double-glass Cannular implant fabrication and
and 0.2 (site 4); 3.6, 2.75, and 0.2 (site 5). For window, assembled by gluing one each of preparation of LFP electrodes
PN projection labeling in the anteromedial 3-mm- and 5-mm-diameter circular cover- Cannula implants were prepared under a
(AM) visual cortical area (Figs. 1J and 2E and slips (Warner), was secured in place using stereomicroscope (Leica MZ7-5) with a fully
fig. S21), the coordinates were (AP, ML, and Vetbond, as previously described (78). C&B equipped solder station. The hippocampal
DV): 3.0, 0.96, and 0.4. Injections were per- Metabond was applied over the window mar- cannula consisted of a circular borosilicate
formed using beveled glass micropipettes de- gins and the remainder of the exposed skull. cover glass (3 mm in diameter, 170 ± 5 mm
livering a total of 50 to 100 nl virus per site at Carprofen (5 mg/kg, subcutaneously, every thickness, Schott) glued to the end of a 3-mm
a rate of 100 nl/min. 24 hours for 48 hours) and buprenorphine outer diameter (OD), 1.5-mm-long stainless-
For targeted recordings from identified neu- (100 µg/kg, subcutaneously, every 12 hours steel ring, using UV-curable adhesive (Loctite
ron types (Figs. 1, H to K, and 2; and figs. S16, for 48 hours) were administered as part of 3105) (Figs. 5, M to T, and 6; and fig. S23).
S17, and S19), AAV/DJ-CAG-DIO-Ace-mNeon2 postoperative analgesia. Before awake imag- For the dual-modality optical-LFP hippo-
(or AAV/DJ-CAG-fDIO-pAce) was injected in ing experiments, the headposted mice were campal recordings (Fig. 4), we affixed three
Cre- (or Flp)-recombinase–expressing trans- handled and habituated to run on a custom ~1-cm-long, ~50-µm-diameter tungsten wires
genic driver lines. For DUPLEX imaging (Figs. 3 wheel every day for up to 2 weeks in 10- to coated with single polyimide insulation (M215580,
and 5, A to L), AAV/DJ-CAG-DIO-Ace-mNeon2 20-min training sessions. Imaging began once California Fine Wire) to the optical cannula at
and AAV/DJ-CAG-fDIO-pAce viruses were the animal was fully wheel trained. three different axial positions, spaced in incre-
mixed at a 1:3 (v/v) ratio for injections in ments of ~50 µm. The tip of one of the three
NDNF-Cre+/VIP-Flp+ or at a 1:1 ratio for injec- Hippocampal CA1 surgeries tungsten wires was placed flush with the op-
tions in SST-Cre+/VIP-Flp+ mice. For DUPLEX The Stanford Administrative Panel on Lab- tical window, and the tips of the other two
recordings from PNs with VIP+ interneurons oratory Animal Care (APLAC) approved all extended beyond the optical window. This set
in VIP-Flp+ mice (Figs. 3, L and M, and 5, I to L; procedures involving animals, and we com- of electrode placements was chosen to max-
and fig. S21), the AAV/DJ-CAG-DIO-Ace-mNeon2 plied with all ethical regulations. imize the chances that at least one of the elec-
virus was mixed with the AAV5-CaMKII-mCherry- trodes would provide high-quality LFP signals,
Cre virus for Ace-mNeon2-labeling of PNs and Hippocampal AAV injections given that our surgical placement of the
injected together with an equal volume of the C57BL/6J wild-type and SST-Cre transgenic implant was performed without fine visual
AAV/DJ-CAG-fDIO-pAce virus for pAce-labeling mice were purchased from The Jackson Lab- feedback about the exact location of the
of VIP+ interneurons. To lower PN labeling den- oratory (JAX #664 and #013044). Mice (aged 8 ~100-µm-thick CA1 pyramidal cell layer. We
sities by labeling a subset of neurons project- to 16 weeks at start) underwent two surgi- glued these tungsten wires onto the optical
ing to a higher-order visual cortical area (Figs. cal procedures under isoflurane anesthesia cannula with UV light–cured adhesive, which
1J and 2E), the AAV/DJ-CAG-DIO-Ace-mNeon2 (1.5 to 2% in O2). In the first procedure, we constrained the electrodes to be laterally off-
virus was injected locally in V1, whereas the injected AAVs to express the fluorescent volt- set by 1.5 mm from the center of the imag-
AAV2/Retro-EF1A-mCherry-IRES-Cre virus was age indicators. In the second procedure, per- ing window. Next, we soldered a gold-plated
delivered to the AM cortical area. formed about a week after viral injection, we pinhead (NC0102069, WPI) onto the end of
For dual-color imaging (Fig. 6, A to C), AAV/ inserted a cannular implant for hippocampal each tungsten wire projecting away from the
DJ-CAG-DIO-Ace-mNeon2 and AAV/DJ-CAG- imaging. imaging window. The reference electrode was

composed of a ~5-mm-long, ~127-µm-diameter, water-immersion objective (XLUMPlanFL, trial. We then performed a spline interpola-
uncoated, stainless-steel wire (#791600, A-M Olympus). For Ace-mNeon and pAce imaging, tion (10-ms intervals) of the mean waveform
Systems) soldered to a gold-plated pinhead we used a 503/20-nm excitation filter (Chroma), and, from the resultant, determined the spike
(NC0102069, WPI). Separately, we fabricated a 518-nm dichroic (Chroma), and 534/30-nm emis- amplitude and duration. Determinations of
four-channel connector using four gold-plated sion filter (BrightLine). We illuminated the sam- the spike detection fidelity for studies in live
sockets (NC1456862, WPI) and a connector ple using the 500-nm wavelength module of a flies were performed using the same formu-
board (EIB8, Neuralynx). solid-state light source (Spectra X, Lumencor), las as were used for the studies in mice (see
with 5 to 7 mW of optical power at the specimen section Determinations of the spike detec-
Hippocampal cannular implantation plane. For VARNAM and pAceR imaging, we tion fidelity index, d′).
To implant the optical cannula, we resected used a 559/25-nm excitation filter (Semrock),
the skin and scalp to expose the dorsal surface 580-nm dichroic (Chroma), and 630/50-nm emis- Single- and dual-channel voltage imaging in V1
of the cranium and removed any remaining sion filter (Semrock). We illuminated the sam- in running mice
connective tissue by cleaning the exposed skull ple using the 550-nm wavelength module of the All data from brain area V1 (Figs. 1, H to K; 2,
with hydrogen peroxide (H2O2). To ensure light source, with 8 to 10 mW of optical power B to E; 3; 5, A to L; and 6, A to C; and figs. S16,
strong adhesion between the cannula implant at the specimen plane. We acquired images at S17, S19, S21, and S22) were acquired at the
and the skull, we slightly roughened the skull 1000 Hz, using a scientific-grade CCD camera John B. Pierce Laboratory. Before the awake
surface using a drill bit and then rinsed the (Zyla 4.2, Andor) and 2 pixel–by–2 pixel binning. animal experiments, head-posted mice were
cranium with Ringer’s solution. We drilled a handled and acclimated to a custom 15-cm-
3-mm-diameter opening above the right dorsal Odor delivery diameter 3D-printed wheel for ~2 weeks in
CA1, sized to allow the cannula implant to fit in To image odor-evoked activity, we perfused training sessions of 10 min per day.
snugly. We aspirated the cortex above dor- odors to the flies’ antennae using a custom-
sal CA1 using a 30-gauge blunt needle, under built olfactometer that delivered constant air- Instrumentation and imaging
irrigation with ice-cold Ringer’s solution. We flow (200 ml min−1) to either the control path For in vivo neocortical voltage imaging, mice
proceeded with cortical aspiration until the (mineral oil) or the odor path (odorant dis- implanted with cranial windows were head-
light-scattering, myelinated axon bundles of solved in mineral oil). We presented 5% isoamyl fixed by securing the head post to two points
the corpus callosum became visible. Of the acetate (CAS# 123-92-2, Sigma-Aldrich Inc.), on the custom wheel with thumbscrews. Neu-
three superimposed layers of myelinated axons, 3% benzaldehyde (CAS# 100-52-7, Sigma-Aldrich rons expressing soma-targeted indicators at
which were visually distinguishable by their Inc.), or apple cider vinegar (Bragg Inc.) to the cortical depths of ~50 to 200 mm below the
distinct orientations, we removed the top two fly using a probe needle [1.7 mm inner diameter pial surface (Fig. 2, B to E, subpanel a) were
layers with gentle air suction, while making (ID), Grainger Inc.], ~3 mm in front of the an- imaged on a custom-built dual-channel up-
sure to leave intact the stratum oriens layer of tenna. Each odor delivery trial lasted for 5 s. right fluorescence microscope equipped with
the hippocampus. (We excluded from further an epi-illuminator module (WFA2001, Thor-
study any mice with surgical damage to CA1, Spike extraction labs), an EYFP/mCherry (59026, Chroma) dual-
because such damage can lead to local epilep- To extract the voltage data, we motion- bandpass filter set (comprising a 59026×
tiform activity.) The cannula implant was then corrected the raw videos using the Turboreg excitation filter with passbands of 503/12
inserted and affixed to the skull using UV- algorithm for image registration (http:// and 575/15, a 69008bs dichroic mirror with
curable adhesive (Loctite 3105). bigwww.epfl.ch/thevenaz/turboreg/) (79). We passbands of 470/20, 540/25 and 630/50,
For studies in which LFP recordings were to then selected pixels whose mean fluorescence and a 59026m emission filter with passbands
be performed concurrently with imaging, we intensity, averaged over the entire movie, was of 535/15 and 635/35), and a 20X 1.0 NA Olym-
drilled a 0.5-mm-diameter hole into the skull in the top 10% of all pixels across the field of pus XLUMPLFLN water immersion objective.
above the cerebellum, inserted the reference view; we defined the union of these high- The illumination from a pair of LEDs, emit-
electrode into cerebellar tissue and then af- ranking pixels as the region of interest (ROI). ting light of 505- and 565-nm wavelengths
fixed it to the skull with UV-curable adhesive We corrected for photobleaching by fitting (M505L4 and M565L3, Thorlabs, respectively),
(Loctite 3105). Carprofen (5 mg kg−1) was ad- a double exponential function to the spatially was collimated with a pair of aspheric con-
ministered 30 min before the end of the sur- averaged fluorescence across the ROI, F(t), denser lenses (ACL2520U-A), one for each
gery to mitigate pain. To enable head-fixation and then dividing F(t) by the fitted double LED. For dual-channel imaging (Fig. 6, A
during in vivo imaging, we secured a custom exponential trace. We then computed time- to C), illumination from the two LEDs was
stainless steel head bar to the mouse’s skull using dependent changes in the relative fluorescence combined using a 550-nm long-pass dichroic
blue light-cured resin (Flow-it ALC, Pentron). intensity, DF(t)/F = (F′(t)–F0)/F0, where F ′(t) mirror (T550 LPXR, Chroma). The illumina-
Carprofen (5 mg kg−1) was administered up to was the time-dependent fluorescence intensity tion intensity at the sample plane in each of the
2 days after surgery as part of postoperative trace across the ROI after the photobleaching red and green channels was ~25 mW mm−2.
analgesia. Mice recovered for ~2 weeks before correction and F0 was the time-averaged base- Fluorescence emissions returning from the
the start of voltage imaging sessions. line intensity in the ROI after the photobleach- sample plane were directed by a 585-nm long-
ing correction. To identify individual spikes, pass dichroic mirror (T585 LPXR, Chroma) into
Voltage imaging in awake flies we high-pass filtered the DF(t)/F trace by sub- the two fluorescence detection arms of the mi-
The fly data was acquired at Stanford Univer- tracting a median-filtered (40-ms window) ver- croscope. Emissions from Ace-mNeon2 and/
sity. The optical instrumentation described sion of the trace and identified spikes as local or pAce passed through a 520/35-nm bandpass
below was used to obtain the data in Fig. 1, E peaks that surpassed a set threshold (3 SD for filter (FF01-520/35-25, Semrock). Emissions
to G; and figs. S13, S14, and S25. PPL1-DANs, 2 SD for MBON-g1pedc>a/b). We from VARNAM2 and/or pAceR passed through
estimated the instantaneous, time-dependent a 630/75-bandpass filter (ET630/75m, Chroma).
Instrumentation and imaging spike rate by tabulating the number of spikes Within each fluorescence detection arm, a 0.75×
To image voltage dynamics in flies, we used a within a 100-ms sliding window. camera tube lens (WFA4101, Thorlabs) focused
custom-built upright epi-fluorescence micro- To determine the mean spike waveform, we the fluorescence onto a high-speed sCMOS
scope and a 1.0 numerical aperture (NA) 20× averaged the waveforms of all spikes across a camera (Orca Flash 4.0 v3, Hamamatsu).

For the dual-color, dual-polarity imaging monitor was placed ~15 cm from the animal, ing (Fig. 6, D to F), we used a custom-built
studies of V1 using Ace-mNeon2, pAce, and perpendicular to the surface of the right eye microscope with a 16× 0.80 NA, 3.0-mm working
VARNAM2 in Fig. 6, A to C, both the 505- and (contralateral to the injection area). To map distance (WD), water immersion objective lens
565-nm LEDs were ON concurrently, and we the receptive field of an imaged neuron, static (Nikon N16XLWD-PF). For red-DUPLEX (fig.
acquired data simultaneously from both cam- bright and dark stimuli were presented individ- S23), we used a 25× 1.0 NA, 4.0-mm WD, water
eras. For synchronized registration of the im- ually in a pseudorandom sequence within each immersion lens (Olympus XLPLN25XSVMP2).
ages, the two cameras were operated in their square of a 3-by-4 grid extending across the en- The recording depth for all hippocampal data
external trigger mode, allowing us to trigger tire video monitor. Once the cell’s receptive field was ~50 to 100 µm, as set by the intrinsic cel-
image acquisition on both cameras by sending location was approximately mapped, to charac- lular organization of the dorsal CA1 area. A
them a single, common transistor-transistor terize the cell’s orientation-selective responses, motorized stage (Sutter MP-285) was used
logic (TTL) pulse. We further verified that im- we sequentially presented a set of drifting sinu- for fine axial focusing.
age acquisition by the two cameras was syn- soidal gratings at one of eight orientations, Illumination from two LEDs (UHP-F-470 and
chronized by using the time stamps on the two spaced by increments of 45° and presented in a UHP-F-545, Prizmatix) was combined using a
videos. To ensure that images from the two pseudorandom sequence. The spatial frequency long-pass dichroic mirror (511 nm, Prizmatix)
channels were spatially aligned, we calibrated of the grating was 0.04 cycles per division, and and then filtered using a dual-bandpass clean-
images against an alignment target and con- the temporal frequency was 2 Hz. Each trial up filter (Semrock, FF01-482/563-25, with pass-
firmed this during each imaging session based lasted 16 s, with interstimulus intervals of 500 ms. bands of 482/18 and 563/9 nm). A dual-band
on reference points, e.g., the location of land- dichroic mirror (Semrock, Di01-R488/561-25x36,
mark vasculature. Induced state transitions and pupillometry with passbands of 525/40 and 690/108 nm)
For all experiments in Figs. 1, H to K; 2, B In Figs. 2, B to E, and 5, A to L, to induce a reflected the combined illumination toward
to E; 3; 5, A to L; and 6, A to C; and figs. S16, change of state, a brief 50-ms air puff was the specimen and transmitted the returning
S17, S19, S21, and S22, fluorescence time-series delivered to mice at rest using a small tube fluorescence emissions. The illumination inten-
images (2048 pixels by 512 pixels) were col- aimed at the back of the animal. Wheel posi- sity at the sample plane from each LED was
lected at 400 Hz with 4-by-4 binning. We typ- tion was tracked using a programmable angle ~25 to 75 mW mm−2.
ically acquired 3 or 4 trials of data from each sensor (KMA210, Digikey). A data acquisition In the emission pathway, to separate fluo-
imaging field of view, and we imaged ~6 to device (NI USB-6259, National Instruments) rescence in the two color channels, we used a
8 fields of view per session. Over the course registered the wheel’s position and delivered 550-nm short-pass dichroic mirror (70 mm by
of several days, but in some cases over 3 to TTL pulses to initiate image acquisition. Soft- 100 mm, custom designed, Alluxa) and a pair
4 weeks, we revisited and reimaged each field ware code that was custom written in Lab- of emission filters, each of which was specific
of view two or three more times, using vas- View 2019 (National Instruments) controlled for one of the two color channels: 520/41 nm
cular landmarks to identify the approximate data acquisition. (70-mm diameter, custom designed, Alluxa) for
tissue locations of prior imaging sessions. For the state-transition experiments in Figs. Ace-mNeon2 and pAce and 609/62 nm (Semrock,
We inferred the cortical depth of the imaged 2, B to E, and 5, A to L, an infrared camera FF01-609/62-50) for VARNAM2 and pAceR.
neurons based on the laminar organization, as (Basler A602f) was used to measure the pupil Each arm of the fluorescence collection path-
reported in the published scientific literature, diameter at a frame rate of 15 Hz. Pupil diam- way was composed of a tube lens of 85-mm
of the different cortical cell types that we ex- eter was measured in real time using custom effective focal length (Canon, EF 85 mm f/1.2L II
amined. Specifically, we inferred that we re- code written in LabView. An annulus region of USM) and a high-speed scientific-grade CMOS
corded neurons in cortical layers 1, 2, and 3 interest was drawn over the pupil, with the camera (Hamamatsu ORCA Fusion Digital CMOS
based on published evidence that (i) cortical inner circle located at the center of the eye and camera; Hamamatsu Photonics K.K., C14440-
NDNF + interneurons are layer 1 specific the outer circle extending past the edge of the 20UP). Data acquisition was controlled by
(41, 47, 48) and that (ii) cortical PNs, selectively pupil. The edge-detection functionality of the custom software with a field programmable
targeted using CaMKII, and SST+ interneu- LabView Vision Development Module was used gate array (FPGA)–based (Xilinx Z-7010) back
rons, selectively targeted using the published to detect dark-to-light transition points along a end and a front end written in LabView 64-bit
SST-Cre transgenic line (40) are both absent set of radial measurement lines drawn from the 2019 (National Instruments). Note that although
in cortical layer 1. PN projection neurons from inner circle and extending to the outer circle. our labs were acquainted with and therefore
the AM area reside in cortical layer 2/3, be- The pupil diameter was calculated by fitting a used LabView to implement the hardware con-
tween 100 and 200 mm below the pial surface circle to the set of detected dark-to-light tran- trol, other programming languages (e.g., C++,
(39, 42). SST+ interneurons reside at depths sition points. The threshold level for edge detec- MATLAB, Python) can also be used to control
>200 mm beneath the pial surface (40), and VIP+ tion and the number of transition points were data acquisition if the camera manufacturer
interneurons reside at depths >50 to 100 mm manually optimized for each individual mouse. provides an appropriate software develop-
below the pia (49). Confocal micrographs show- ment package.
ing cell subtype–specific labeling in V1 (Fig. 2, B Single- and dual-channel voltage recordings in As with dual-channel imaging in V1, dual-
to E, subpanel a) further confirmed the laminar CA1 in running mice channel imaging in CA1 (Figs. 5, M to T, and
organization of the imaged cortical cell classes All imaging studies of hippocampal area CA1 6, D to F) was performed with both the 470-
in the supragranular layers and the relative cor- were performed at Stanford University. The and 545-nm illumination sources turned ON
tical depths of the distinct neocortical subclasses. optical instrumentation described below was (or OFF) simultaneously at the beginning (or
used to acquire the CA1 data shown in Figs. end) of each trial. Frame acquisitions were
Visual stimulus presentations 4; 5, M to T; and 6, D to F; and figs. S23, S24, synchronized across the two cameras using an
In Fig. 1J, visual stimuli were generated using and S26. external electronic trigger generated by the
the Psychophysics toolbox on MATLAB and FPGA. We achieved subpixel spatial alignment
were presented to awake, stationary mice on Instrumentation and imaging between the fields of view of the two cameras
an LCD video monitor (1280 pixels by 1024 For green DUPLEX studies (Fig. 4), simulta- by registering them with an alignment target.
pixels, 20 inches by 16 inches, 60-Hz refresh neous dual-color (Fig. 5, M to T; and figs. S24 We verified proper registration using image
rate, mean luminance of 250 cd m−2). The and S26), and triple-population voltage imag- landmarks, e.g., large blood vessels.

Images were collected at 600 Hz with no filtered the LFP traces (the filter’s 3-dB high- fied as threshold crossing events, and the spike
spatial binning. Live data from each camera frequency cutoff was 500 Hz and the 3-dB waveforms are averaged to obtain a spike tem-
were stored as native DCAM image files (.dcimg) low-frequency cutoff was 0.1 Hz). Then, we inter- plate. A matched filter (82) is then derived from
on two separate 2-TB SSD drives. The data were polated the LFP traces and the squared syncTTL this spike template and from additive white
batch-processed offline using the custom pipe- signal so that the sampling rates of the resulting Gaussian noise, with the goal of identifying a
line SpikeImagingAnalysis (see section Data traces matched the 600-Hz sampling frequency template waveform that can be applied to a
extraction pipeline and downstream analyses of of the optical data. We used the MATLAB func- noisy signal. To denoise traces, the matched
hippocampal CA1 datasets for details), which tion interp1 with the “spline” method to inter- filter is applied to ts and spikes are detected
we wrote in MATLAB (Mathworks). polate the continuously valued LFP trace. To again. Lastly, the reconstructed fluorescence
interpolate the binary- valued syncTTL signal, trace representing the spike train trec is ob-
Rest-to-run state transition we used the interp1 function with the “nearest” tained by convolving the spike template and
For the studies in Fig. 5, M to T, we used a method. Finally, we temporally aligned the the inferred spike train. (Step 2) To refine each
mouse’s running wheel (8 cm wide, 13 cm in optical and electrical traces according to the cell’s spatial mask, its reconstructed trace trec,
diameter) that was 3D-printed in PLA plastic. temporal registration provided by the paired is regressed onto the fluorescence movie using
The wheel surface was covered with a self- set of syncTTL recordings. ridge regression.
adherent wrap (Coban, 3M). Angular displace-
ments of the wheel were tracked using a rotary Data extraction pipeline and downstream Automated assignment of cell classes in
encoder (Optical AB Phase Quadrature En- analyses of V1 datasets DUPLEX recordings
coder 600P/R, Amazon). A data acquisition Extraction of individual neurons and their To identify the polarity of the neural spike
interface (BNC-2090A, National Instruments) fluorescence activity traces for imaging studies waveforms in DUPLEX recordings (Figs. 3; 5,
controlled from LabView was used to acquire in area V1 (Figs. 1, H to K; 2; 3; 5, A to L; and A to L; and 6, A to C; and figs. S20 to S22), for
the wheel displacement data. 6, A to C; and figs. S15 and S17 to S22) was each cell, we identified putative spikes in the
Mice were habituated on the wheel for at performed at the John B. Pierce Laboratory trec trace as negative- or positive-going that
least 2 days, for at least 15 min per day, be- using the open-access Python-based auto- deviated >3 SD from baseline noise levels. We
fore brain-imaging sessions began. To trigger mated pipeline VolPy (44), which consisted of identified the polarity of the indicator ex-
rest-to-run behavioral state transitions, a brief the following steps: pressed in the cell according to whether it
air puff was delivered to the mouse’s back. The 1) File loading. We modified the VolPy code exhibited more negative- or positive-going
air puff intensity was ~30 psi and was admin- to support the .dcimg file format of the data events (fig. S20). Because VolPy extracts
istered with a 20-gauge blunt needle placed taken with the Hamamatsu camera. action potentials (APs) through template-
1 to 2 cm away from the mouse. For each field- 2) Motion correction. Each movie was motion- matching (44), only typical AP waveforms
of-view, we performed high-speed voltage imag- corrected using the NoRMCorre algorithm were considered during polarity assignment.
ing across a 30-s trial (18,000 frames at 600 Hz), (80), which is built into the VolPy pipeline. Traces with no spikes were not assigned and
starting with the mouse at rest. Within each trial, 3) Cell segmentation. Steps 3 and 4 are in- were removed manually from downstream
a TTL signal (five pulses at 5 Hz) triggered the air cluded as part of the updated SpikePursuit analyses (see analysis pipeline, fig. S20).
puff delivery after 10 or more seconds of base- algorithm (13) that is built into the VolPy pipe-
line data had been acquired. Each experimen- line (44) for spike extraction and denoising. Spike rate, spike modulation index,
tal session involved imaging six fields of view, Whereas the original version of SpikePursuit and correlational analyses
with one air-puff trial per field of view. used manual input parameters to identify the We exported data from VolPy as MATLAB
location of each neuron (13), in the updated files for downstream analyses. Each cell’s fir-
LFP recordings version, neuron identification and segmenta- ing rate (FR) was estimated by sliding a rec-
We amplified the electrophysiological signals tion are performed automatically using a con- tangular window (of 100 ms for Fig. 2 and fig.
using a digital recording head-stage (RHD 16- volutional neural net known as “Mask R-CNN” S19 or 20 ms for Figs. 3, H to M, and 5, C, G, and
Channel, Part #C3334, Intantech) that was at- (81), which provides binary masks covering K; and fig. S17) along the spike train followed by
tached to our custom-designed connector. We each identified cell. smoothing with a Gaussian kernel of 100 ms
then acquired the digital data at a sampling rate 4) Trace denoising and spike extraction. using the smooth function in MATLAB. The air
of 2 kHz using a USB data acquisition board (RHD Fluorescence activity traces are extracted from puff–induced spike modulation indices in Fig.
USB interface board, Part #C3100, Intantech). the regions of the movie covered by spatial 2, B to E (subpanel d) and 5, D, H, and L; and
To synchronize the optical and electrical masks (ROIs) from step 3. and are then cor- fig. S19 were calculated as (FRarousal – FRbaseline)/
recording instruments, we generated a syn- rected for photobleaching using a high-pass (FRarousal + FRbaseline), where FR is the mean
chronizing digital signal (hereafter called filter (third-order Butterworth filter, 1/3 Hz firing rate over a 750-ms time window that was
“syncTTL”), which was composed of a square cut-off). For each ROI, the following two steps either before or after the air puff. In Fig. 2, B to
wave signal with a 50% duty ratio and a 1-s- are then executed in a loop for three iterations. E (subpanels c and d), we temporally aligned
long on-state but with intervals of random The first step estimates spike times and the the spike rates to the air-puff onset. Pairwise
duration between pulses, to enable unambig- second refines the spatial filter. (Step 1) To correlation coefficients and cross-correlation val-
uous temporal alignments. The syncTTL signal estimate spike times, the first eight principal ues in Figs. 2, F to I; 3, H to M; and 5, C, G, and K;
was externally and independently generated components of background pixels (at >12 pixels and figs. S17, S21, and S22 were calculated using
by a data acquisition system (USB-6003, Na- from the ROI) are extracted and removed the corrcoef and xcorr functions in MATLAB.
tional Instruments) and was recorded concur- from the fluorescence movie using ridge regres-
rently by both the electrical and the optical sion. The resulting trace of fluorescence activity Data extraction pipeline and downstream
data acquisition systems (at 600 Hz and 2 kHz, within the ROI after background removal is analyses of hippocampal CA1 datasets
respectively). then high-pass-filtered (fifth-order Butterworth Trace extraction for the hippocampal data (Figs.
To temporally align the optical and electri- filter, 1 Hz cut-off) to obtain a trace, ts, that 4; 5, M to T; and 6, D to F; and figs. S18, S23,
cal physiological signals (recorded at 600 Hz contains activity at frequencies typical of neu- S24, and S26) was performed at Stanford Uni-
and 2 kHz, respectively), we first bandpass- ral oscillations and spikes. Spikes are identi- versity using a custom pipeline written in

MATLAB (SpikeImagingAnalysis, https://github. the activity traces, {T}. A validation procedure rest-to-run transition, which we detected as a
com/sihaziza/SpikeImagingAnalysis_public). then checks each putative cell against a set of discrete change in the running speed (thresh-
This pipeline consisted of the following steps: predetermined quality metrics and removes olded at >2 cm s−1). We then computed the
1) File loading and motion correction. Raw any cell with metrics below predetermined cells’ time-varying spiking rates using a 100-ms
16-bit .dcimg video files from voltage-imaging threshold values. This three-step process runs sliding window. For each neuron type studied,
sessions were loaded into memory, underwent for a fixed number of iterations, and EXTRACT we averaged the time-varying spike rate across
spatial (2 by 2) binning, and saved in .h5 data outputs the final estimates of each cell’s spatial all cells to determine the mean spike rate (Fig.
format. Movies underwent motion-correction profile and activity trace. 5P). To assess the rest-to-run fold change in
using NoRMCorre (79) and were then cropped For DUPLEX studies that involved a pair of instantaneous spiking rates (Fig. 5Q), we com-
so as to remove parts of the images that did similarly colored GEVIs but with opposite po- puted the spiking rate in the 2-s intervals im-
not contain fluorescently labeled brain tissue. larities, we ran EXTRACT twice: once on the mediately before and immediately after the
The NoRMcorre parameter shift_method was detrended video, F(t)/F0, to identify active neu- state transition.
set to “cubic,” instead of the default “fft.” rons expressing the positive polarity indicator,
2) Detrending. To account for fluorescence and once on a sign-inverted version of the de- Analyses of subthreshold activity
photobleaching, at each time point in the fluo- trended video, F(t)/F0, to identify active neu- In Fig. 5, R to T, neuronal traces from all fields
rescence video, F(t), we normalized each pixel rons expressing the negative polarity indicator. of view were temporally aligned to the rest-to-
by F0(t), the time-varying value of the pixel’s 4) Spike timing estimation. To precisely es- run transition. To compute coherence of sub-
baseline fluorescence, which we computed timate each cell’s spike train based on its fluo- threshold activity between pairs of cells, in
using a temporal low-pass filter (fourth-order rescence activity trace, we used the statistical either the running or resting states, we first
Butterworth filter, 0.5-Hz cut-off frequency). method MLspike (85), which outputs the spike estimated the subthreshold dynamics of each
The resultant is the detrended movie, F(t)/F0(t). train that is most likely to have led to the ob- cell by computing low-pass filtered versions
3) Cell extraction. We extracted individual served fluorescence trace. The algorithm re- (fourth-order Butterworth low-pass filter, cut-
neurons and their time traces of membrane quires five input parameters: the saturation S, off frequency of 50 Hz) of the “drift” output of
voltage activity using the automated cell- single-spike amplitude A, single-spike time ML spike for each cell. Next, we computed the
extraction algorithm EXTRACT (45, 46). constant τ, baseline fluctuation “drift,” and magnitude-squared coherence between the
EXTRACT is generally agnostic about cells’ noise level σ. The saturation parameter, de- two low-pass-filtered time traces by using
morphologies and activity patterns and there- fined as the number of spikes at which the in- the MATLAB function mscohere, which we
fore has broad applicability to both one- and dicator provides a full-amplitude response, applied to 1-s-long segments of data, chosen
two-photon fluorescence imaging studies of was set at S = 1. We estimated A and τ using such that temporally successive segments from
neural activity. It is based on the mathemat- a double-exponential fit to the mean spike the traces overlapped by 0.8 s. To estimate the
ical framework of robust statistics, which gen- waveform, which we computed using only noise level in such computations, we computed
erally allows it to surpass other widely used temporally isolated spikes. We empirically es- the pairwise coherence of cells belonging to
cell-extractions algorithms, such as CNMF (83) timated the parameter, “drift,” by computing two separate fields of view. In Fig. 4, I to L, we
and PCA/ICA (84), in the removal of back- peak-to-peak amplitude variation of a low-pass- computed the neuron-LFP coherence in a pair-
ground fluorescence contaminants and cross- filtered (fourth-order Butterworth filter, 50-Hz wise manner, between the subthreshold activ-
talk between overlapping cells (46). cut-off frequency) version of the cell’s activity ity of each cell and the LFP, determined in the
In brief, EXTRACT is designed to automat- trace that retained subthreshold activity up to same way as for cell pairs in Fig. 5.
ically identify each neuron’s spatial profile, S, the gamma frequency band but not spiking
and retrieve its activity trace, T, using an itera- activity. σ was estimated as the standard de- Comparisons of cell-extraction methods
tive process that is applied to individual cells, viation of the residual signal, defined as the We compared the results from VolPy and
one at a time. This “cell finding” process is fol- difference between the raw and a denoised EXTRACT to a simple ROI-based approach
lowed by an iterative “cell refinement” process. version of the cell’s fluorescence activity trace for cell extraction. For this comparison, we
During cell finding, EXTRACT finds a seed (we obtained by applying a wavelet denoising used DUPLEX data providing joint record-
pixel that attains the movie’s maximum fluo- of degree one). ings of NDNF+ and VIP+ interneurons in V1
rescence intensity and initializes a candidate (Fig. 3, B and C; and fig. S15).
cell image at that pixel. The estimates of S and Computations of oscillatory phase Unlike ROI-based cell extraction, both VolPy
T are iteratively updated in an alternating In Fig. 4, G and H, to calculate the phase of and EXTRACT performed comparably in terms
manner by using a robust regression at each spiking with respect to theta oscillations (5 to of removal of cellular cross-talk (fig. S15, A to
step and a one-sided Huber loss function (46). 10 Hz), we first band-pass-filtered both the H), neuropil contamination (fig. S15, I to N),
Once the estimates of S and T stabilize for an optical voltage trace and the LFP trace at the and hemodynamic artifacts (fig. S15, O to T).
individual cell, the cell’s inferred fluorescence theta frequencies using the butter and filtfilt Moreover, the inferred biological conclusions
contribution is subtracted from the movie, and MATLAB functions (fourth-order Butterworth were independent of the algorithm used, in
the entire set of steps is repeated for another filter, cut-off frequency of 5 to 10 Hz). We then that the results obtained using both VolPy and
cell. This process continues until no more puta- computed the phase of each filtered trace by EXTRACT were identical and indicated posi-
tive cells are found, which is determined to occur performing a Hilbert transform (hilbert and tively correlated activity within each interneu-
when the peak value for the activity trace of angle functions). For Fig. 4G, we calculated the ronal population but no significant correlation
the seed pixel falls below a predefined thresh- timing of each spike relative to the period of of the voltage signals between NDNF+ and
old value (>3× SD of background noise). that oscillation cycle (in degrees). For each VIP+ interneurons (fig. S15, F to H). However,
During the iterative process of “cell refine- cell, we then computed the circular mean to traces extracted using an ROI-based approach
ment,” at each iteration, the set of estimated generate Fig. 4H. showed spurious pairwise correlated activity,
traces {T} is updated using robust estimation suggestive of increased cellular cross-talk,
while holding fixed the set of cells’ spatial pro- Spike-rate analyses and the presence of hemodynamic artifacts
files. Next, the spatial profiles {S} are updated In Fig. 5, O to Q, all traces (spike rasters and in the subthreshold traces. In addition, the
using robust estimation while holding fixed speed traces) were temporally aligned to the data obtained from ROI-extracted traces were

inconsistent with prior observations as they a fluorescence detector, F = (F1, F2, …, FN), to the decision boundary for classifying instances
showed highly positively correlated activity detect spikes, the distribution of F follows of H(0) and H(1) was set midway between the
not only within populations but also anom- Poisson statistics in the shot-noise-limited re- two log-likelihood probability distributions
alously between NDNF+ and VIP+ interneur- gime. Given a specific observation, F, the Pois- [see, e.g., figure 1 of (38)]. In other words,
ons (47, 48). son distribution of F enables one to estimate false-negative and false-positive spikes were
the a posteriori likelihood of two mutually ex- considered equally deleterious. When there
Computations of GEVI impulse responses clusive hypotheses: the null hypothesis, H(0), are abundant fluorescence photons, the two
In fig. S12, activation and deactivation ker- which posits the absence of a spike; and the log-likelihood distributions are each approx-
nels for each GEVI were determined using a alternative, H(1), which posits that a spike oc- imately Gaussian, with means that differ by d′
biexponential function from the empirically curred. d′ can then be calculated as in units of σL. We used the complementary
measured values of each indicator’s fast and error function erfc in MATLAB to compute the
ð1Þ ð0Þ
slow time constants and the relative ampli- d ′ ¼ mL mL =σL ð1Þ proportion of the area under each log-likelihood
tudes of the fast and slow relaxation processes probability distribution that was on the oppo-
associated with the indicator’s activation and in which mL and σL respectively represent the site side of the decision boundary from the
deactivation responses. The empirical values mean and variance of the (approximately mean of the distribution [i.e., the proportion of
used and the publications from which they normal) probability distributions of the log- cases representing spike detection errors; see
are taken are indicated in the caption for likelihood ratio, L(f), of the two hypotheses, in figure 1 of (38)].
table S1. the cases when there actually is or is not a
We determined each GEVI’s impulse re- spike. The mean, mL, and variance, σL, of these Statistics
sponse function (fig. S12, A and B) by con- log-likelihood ratio distributions are given by For both the V1 and CA1 datasets, we per-
volving its activation kernel with the rising formed statistical tests using standard MATLAB
edge of a 1-ms-duration square voltage pulse ð0Þ F 0 XN functions or built-in tests in Prism 8 (GraphPad).
mL ¼ n¼1
logð1 þ sn Þ
(+100-mV depolarization, starting from −70 mV) n For two-sample comparisons of a single vari-
and its deactivation kernel with the falling F0 XN able, we always used nonparametric tests, as
s
n¼1 n
ð2Þ
edge of the square pulse, and then summing n described in this section.
the two resultants at a sampling rate of 5 kHz. In Fig. 1D, to test for significant differences
To isolate the kinetic aspects of the indicator ð1Þ F0 XN in the mean DF/F values of the indicators, we
response, we defined the excursion rate (fig. mL ¼ n¼1
ð1 þ sn Þlogð1 þ sn Þ used Mann-Whitney test. In Figs. 2, F to I,
n
S12A) as the time-dependent fluorescence emis- F0 XN and 5, C, G, and K, for comparisons of pairwise
sion rate response to the 100-mV voltage s
n¼1 n
ð3Þ correlation coefficients of spike rates for time
n
pulse, normalized by the indicator’s steady- bins before versus after air puff, we performed
state fluorescence emission rate in response to Wilcoxon matched-pairs signed rank test. In
a maintained 100-mV depolarization (i.e., a rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
F0 XN Fig. 3, I, K, and M, to assess whether neuron
ð1Þ ð0Þ
100-mV step function). To account for the σL ≈σL ≈σL ¼ n¼1
log2 ð1 þ sn Þ ð4Þ pairs exhibited positively versus negatively
n
different response amplitudes of different correlated activity, we compared the mean
GEVIs, we performed the same calculation for Where ν is the sampling rate, F0 represents the pairwise correlation coefficients to zero using
each GEVI but without normalizing by the baseline fluorescence intensity in time one-sample Wilcoxon test. In Fig. 4H, to assess
steady-state response amplitude; the result- periods that contain no neural spike, and sn statistical differences in the polar plots, we
ing plots, termed the “effective 1× AP-DF/F′” is the mean fluorescence signal at each time computed the mean and standard deviation
plots (fig. S12B), show each indicator’s esti- bin within a time period that contains the of each data distribution and made statisti-
mated response to a 1-ms action potential of optical waveform of the identified spikes for cal comparisons between different distribu-
100-mV amplitude. To evaluate how different each neuron. For studies in flies (figs. S13 and tions using circular statistics. Specifically,
GEVIs respond to a burst of action potentials S14C), we determined each neuron’s d′ value we used the Hodges-Ajne test to test the hy-
in quick succession (fig. S12C), we computed by first computing its mean spike waveform, pothesis of nonuniformity of circular data
indicator responses to a set of three spikes, sn, over an interval of [−25 ms, +25 ms] relative and the Watson-Williams multisample test
with each spike modeled as a 1-ms-duration to the spike occurrence time, and then using to test the hypothesis of equal means. In Fig.
square pulse of 100-mV depolarization, and Eq. 1 to 4 to compute d′. 5, D, H, and L, to assess whether neuronal
with the spikes spaced in 3-ms increments. To For studies in mice (figs. S16, F and H; and spike rates were positively versus negatively
assess the capability to resolve action potential S18), we first used Eq. 1 to 4 to compute a d′ modulated by the air puff, we tested whether
bursts under typical imaging conditions, we value for each action potential fired by an in- mean spike modulation indices were signif-
then down-sampled the computed traces to dividual cell and then determined the mean d′ icantly different from zero using a one-sample
1 kHz (fig. S12C). value averaged across all spikes in the cell’s Wilcoxon test.
spike train. We used only spikes that were tem- We did not use power analysis or statistical
Determinations of the spike detection fidelity porally separated from other action potentials methods to estimate sample sizes but instead
index, d′ by >10 ms, so as to avoid computational biases estimated useful sample sizes based on pub-
To compute d′, we used a signal detection that could be introduced by spike bursts, e.g., lished reports and our prior experiences with
theory framework that accounts for both the by instances when a cell does not return to its voltage imaging. Data exclusion criteria were
duration and intensity of fluorescence wave- resting membrane potential between spikes. not preestablished. However, recordings with
forms in optical recordings of action potentials For each isolated spike, we estimated d′ using no spiking neurons were excluded.
and that characterizes the ability to correctly data within a time interval of [−10 ms, +20 ms]
distinguish instances of a spike from back- relative to the spike’s occurrence time.
REFERENCES AND NOTES
ground noise fluctuations within the fluores- To determine how the spike detection error
1. K. A. Ferguson, J. A. Cardin, Mechanisms underlying gain
cence trace (17, 38). As described in our prior rate depended on the imaging frame rate (fig. modulation in the cortex. Nat. Rev. Neurosci. 21, 80–92
work (38), if we use N successive samples from S18, A and B), we treated the case in which (2020). doi: 10.1038/s41583-019-0253-y; pmid: 31911627

2. Y. Fu et al., A cortical circuit for gain control by behavioral 25. Y. Gong, M. J. Wagner, J. Zhong Li, M. J. Schnitzer, Imaging 47. E. Abs et al., Learning-related plasticity in dendrite-targeting
state. Cell 156, 1139–1152 (2014). doi: 10.1016/ neural spiking in brain tissue using FRET-opsin protein voltage layer 1 interneurons. Neuron 100, 684–699.e6 (2018).
j.cell.2014.01.050; pmid: 24630718 sensors. Nat. Commun. 5, 3674 (2014). doi: 10.1038/ doi: 10.1016/j.neuron.2018.09.001; pmid: 30269988
3. C. C. Y. Lee, E. Kheradpezhouh, M. E. Diamond, E. Arabzadeh, ncomms4674; pmid: 24755708 48. K. Cohen-Kashi Malina et al., NDNF interneurons in layer 1
State-dependent changes in perception and coding in the 26. P. Zou et al., Bright and fast multicoloured voltage reporters gain-modulate whole cortical columns according to an
mouse somatosensory cortex. Cell Rep. 32, 108197 (2020). via electrochromic FRET. Nat. Commun. 5, 4625 (2014). animal’s behavioral state. Neuron 109, 2150–2164.e5 (2021).
doi: 10.1016/j.celrep.2020.108197; pmid: 32997984 doi: 10.1038/ncomms5625; pmid: 25118186 doi: 10.1016/j.neuron.2021.05.001; pmid: 34038743
4. P.-O. Polack, J. Friedman, P. Golshani, Cellular mechanisms of 27. D. R. Hochbaum et al., All-optical electrophysiology in 49. M. He et al., Strategies and tools for combinatorial targeting of
brain state-dependent gain modulation in visual cortex. mammalian neurons using engineered microbial rhodopsins. GABAergic neurons in mouse cerebral cortex. Neuron 92,
Nat. Neurosci. 16, 1331–1339 (2013). doi: 10.1038/nn.3464; Nat. Methods 11, 825–833 (2014). doi: 10.1038/nmeth.3000; 555 (2016). doi: 10.1016/j.neuron.2016.10.009;
pmid: 23872595 pmid: 24952910 pmid: 27764676
5. J. Schiemann et al., Cellular mechanisms underlying behavioral 28. N. C. Shaner et al., A bright monomeric green fluorescent 50. H. Adesnik, W. Bruns, H. Taniguchi, Z. J. Huang, M. Scanziani,
state-dependent bidirectional modulation of motor cortex protein derived from Branchiostoma lanceolatum. Nat. Methods A neural circuit for spatial summation in visual cortex. Nature 490,
output. Cell Rep. 11, 1319–1330 (2015). doi: 10.1016/ 10, 407–409 (2013). doi: 10.1038/nmeth.2413; 226–231 (2012). doi: 10.1038/nature11526; pmid: 23060193
j.celrep.2015.04.042; pmid: 25981037 pmid: 23524392 51. L. Mesik et al., Functional response properties of VIP-
6. B. K. Hulse, E. V. Lubenov, A. G. Siapas, Brain state 29. B. T. Bajar et al., Improving brightness and photostability of expressing inhibitory neurons in mouse visual and auditory
dependence of hippocampal subthreshold activity in awake green and red fluorescent proteins for live cell imaging and cortex. Front. Neural Circuits 9, 22 (2015). doi: 10.3389/
mice. Cell Rep. 18, 136–147 (2017). doi: 10.1016/ FRET reporting. Sci. Rep. 6, 20889 (2016). doi: 10.1038/ fncir.2015.00022; pmid: 26106301
j.celrep.2016.11.084; pmid: 28052244 srep20889; pmid: 26879144 52. J. Yu, H. Hu, A. Agmon, K. Svoboda, Recruitment of GABAergic
7. P. Rajasethupathy et al., Projections from neocortex mediate 30. J. Park et al., Screening fluorescent voltage indicators with interneurons in the barrel cortex during active tactile
top-down control of memory retrieval. Nature 526, 653–659 spontaneously spiking HEK cells. PLOS ONE 8, e85221 (2013). behavior. Neuron 104, 412–427.e4 (2019). doi: 10.1016/
(2015). doi: 10.1038/nature15389; pmid: 26436451 doi: 10.1371/journal.pone.0085221; pmid: 24391999 j.neuron.2019.07.027; pmid: 31466734
8. K. Q. Shan, E. V. Lubenov, M. Papadopoulou, A. G. Siapas, 31. H. Zhang, E. Reichert, A. E. Cohen, Optical electrophysiology 53. M. Vinck, R. Batista-Brito, U. Knoblich, J. A. Cardin, Arousal
Spatial tuning and brain state account for dorsal hippocampal for probing function and pharmacology of voltage-gated ion and locomotion make distinct contributions to cortical activity
CA1 activity in a non-spatial learning task. eLife 5, e14321 channels. eLife 5, e15202 (2016). doi: 10.7554/eLife.15202; patterns and visual encoding. Neuron 86, 740–754 (2015).
(2016). doi: 10.7554/eLife.14321; pmid: 27487561 pmid: 27215841 doi: 10.1016/j.neuron.2015.03.028; pmid: 25892300
9. C. Xu et al., Distinct hippocampal pathways mediate 32. L. Stryer, Fluorescence energy transfer as a spectroscopic 54. J. Reimer et al., Pupil fluctuations track fast switching of
dissociable roles of context in memory retrieval. Cell 167, ruler. Annu. Rev. Biochem. 47, 819–846 (1978). doi: 10.1146/ cortical states during quiet wakefulness. Neuron 84, 355–362
961–972.e16 (2016). doi: 10.1016/j.cell.2016.09.051; annurev.bi.47.070178.004131; pmid: 354506 (2014). doi: 10.1016/j.neuron.2014.09.033; pmid: 25374359
pmid: 27773481 33. T. Kikukawa et al., Photochemistry of Acetabularia rhodopsin II 55. C. K. Pfeffer, M. Xue, M. He, Z. J. Huang, M. Scanziani,
10. M. Kannan, G. Vasan, V. A. Pieribone, Optimizing strategies from a marine plant, Acetabularia acetabulum. Biochemistry Inhibition of inhibition in visual cortex: The logic of connections
for developing genetically encoded voltage indicators. 50, 8888–8898 (2011). doi: 10.1021/bi2009932; between molecularly distinct interneurons. Nat. Neurosci. 16,
Front. Cell. Neurosci. 13, 53 (2019). doi: 10.3389/ pmid: 21905737 1068–1076 (2013). doi: 10.1038/nn.3446; pmid: 23817549
fncel.2019.00053; pmid: 30863283 34. T. Wada et al., Crystal structure of the eukaryotic light-driven 56. U. Knoblich, L. Huang, H. Zeng, L. Li, Neuronal cell-subtype
11. T. Knöpfel, C. Song, Optical voltage imaging in neurons: proton-pumping rhodopsin, Acetabularia rhodopsin II, from specificity of neural synchronization in mouse primary visual
Moving from technology development to practical tool. marine alga. J. Mol. Biol. 411, 986–998 (2011). doi: 10.1016/ cortex. Nat. Commun. 10, 2533 (2019). doi: 10.1038/s41467-
Nat. Rev. Neurosci. 20, 719–727 (2019). doi: 10.1038/s41583- j.jmb.2011.06.028; pmid: 21726566 019-10498-1; pmid: 31182715
019-0231-4; pmid: 31705060 35. D. Maclaurin, V. Venkatachalam, H. Lee, A. E. Cohen, 57. L. Cossell et al., Functional organization of excitatory synaptic
12. H. H. Yang, F. St-Pierre, Genetically encoded voltage indicators: Mechanism of voltage-sensitive fluorescence in a microbial strength in primary visual cortex. Nature 518, 399–403
Opportunities and challenges. J. Neurosci. 36, 9977–9989 rhodopsin. Proc. Natl. Acad. Sci. U.S.A. 110, 5939–5944 (2013). (2015). doi: 10.1038/nature14182; pmid: 25652823
(2016). doi: 10.1523/JNEUROSCI.1095-16.2016; pmid: 27683896 doi: 10.1073/pnas.1215595110; pmid: 23530193 58. J. Akerboom et al., Genetically encoded calcium indicators for
13. A. S. Abdelfattah et al., Bright and photostable chemigenetic 36. L. Zimányi et al., Pathways of proton release in the multi-color neural activity imaging and combination with
indicators for extended in vivo voltage imaging. Science bacteriorhodopsin photocycle. Biochemistry 31, 8535–8543 optogenetics. Front. Mol. Neurosci. 6, 2 (2013). doi: 10.3389/
365, 699–704 (2019). doi: 10.1126/science.aav6416; (1992). doi: 10.1021/bi00151a022; pmid: 1327104 fnmol.2013.00002; pmid: 23459413
pmid: 31371562 37. J. Platisa, G. Vasan, A. Yang, V. A. Pieribone, Directed evolution 59. H. Dana et al., Sensitive red protein calcium indicators for
14. A. S. Abdelfattah et al., A general approach to engineer of key residues in fluorescent protein inverses the polarity of imaging neural activity. eLife 5, e12727 (2016). doi: 10.7554/
positive-going eFRET voltage indicators. Nat. Commun. 11, voltage sensitivity in the genetically encoded indicator eLife.12727; pmid: 27011354
3444 (2020). doi: 10.1038/s41467-020-17322-1; ArcLight. ACS Chem. Neurosci. 8, 513–523 (2017). 60. M. Inoue et al., Rational design of a high-affinity, fast, red
pmid: 32651384 doi: 10.1021/acschemneuro.6b00234; pmid: 28045247 calcium indicator R-CaMP2. Nat. Methods 12, 64–70 (2015).
15. S. Chamberland et al., Fast two-photon imaging of subcellular 38. B. A. Wilt, J. E. Fitzgerald, M. J. Schnitzer, Photon shot noise doi: 10.1038/nmeth.3185; pmid: 25419959
voltage dynamics in neuronal tissue with genetically encoded limits on optical detection of neuronal spikes and estimation of 61. M. Inoue et al., Rational engineering of XCaMPs, a multicolor
indicators. eLife 6, e25690 (2017). doi: 10.7554/eLife.25690; spike timing. Biophys. J. 104, 51–62 (2013). doi: 10.1016/ GECI suite for in vivo imaging of complex brain circuit
pmid: 28749338 j.bpj.2012.07.058; pmid: 23332058 dynamics. Cell 177, 1346–1360.e24 (2019). doi: 10.1016/
16. L. Z. Fan et al., All-optical electrophysiology reveals the role of 39. M. E. Garrett, I. Nauhaus, J. H. Marshel, E. M. Callaway, j.cell.2019.04.007; pmid: 31080068
lateral inhibition in sensory processing in cortical layer 1. Cell Topography and areal organization of mouse visual cortex. 62. M. M. Karnani et al., Opening holes in the blanket of inhibition:
180, 521–535.e18 (2020). doi: 10.1016/j.cell.2020.01.001; J. Neurosci. 34, 12587–12600 (2014). doi: 10.1523/ Localized lateral disinhibition by VIP interneurons.
pmid: 31978320 JNEUROSCI.1124-14.2014; pmid: 25209296 J. Neurosci. 36, 3471–3480 (2016). doi: 10.1523/
17. Y. Gong et al., High-speed recording of neural spikes in 40. H. Taniguchi et al., A resource of Cre driver lines for genetic JNEUROSCI.3646-15.2016; pmid: 27013676
awake mice and flies with a fluorescent voltage sensor. Science targeting of GABAergic neurons in cerebral cortex. Neuron 71, 63. W. Akemann et al., Imaging neural circuit dynamics with a
350, 1361–1366 (2015). doi: 10.1126/science.aab0810; 995–1013 (2011). doi: 10.1016/j.neuron.2011.07.026; voltage-sensitive fluorescent protein. J. Neurophysiol. 108,
pmid: 26586188 pmid: 21943598 2323–2337 (2012). doi: 10.1152/jn.00452.2012; pmid: 22815406
18. M. Kannan et al., Fast, in vivo voltage imaging using a red 41. B. Tasic et al., Adult mouse cortical cell taxonomy revealed by 64. F. Li et al., The connectome of the adult Drosophila mushroom
fluorescent indicator. Nat. Methods 15, 1108–1116 (2018). single cell transcriptomics. Nat. Neurosci. 19, 335–346 (2016). body provides insights into function. eLife 9, e62576 (2020).
doi: 10.1038/s41592-018-0188-7; pmid: 30420685 doi: 10.1038/nn.4216; pmid: 26727548 doi: 10.7554/eLife.62576; pmid: 33315010
19. K. D. Piatkevich et al., A robotic multidimensional directed 42. J. Zhuang et al., An extended retinotopic map of mouse cortex. 65. J. Jackson, I. Ayzenshtat, M. M. Karnani, R. Yuste, VIP+
evolution approach applied to fluorescent voltage reporters. eLife 6, e18372 (2017). doi: 10.7554/eLife.18372; pmid: 28059700 interneurons control neocortical activity across brain states.
Nat. Chem. Biol. 14, 352–360 (2018). doi: 10.1038/s41589- 43. S. T. Lim, D. E. Antonucci, R. H. Scannevin, J. S. Trimmer, J. Neurophysiol. 115, 3008–3017 (2016). doi: 10.1152/
018-0004-9; pmid: 29483642 A novel targeting signal for proximal clustering of the jn.01124.2015; pmid: 26961109
20. V. Villette et al., Ultrafast two-photon imaging of a high-gain Kv2.1 K+ channel in hippocampal neurons. Neuron 25, 66. J. Reimer et al., Pupil fluctuations track rapid changes in
voltage indicator in awake behaving mice. Cell 179, 1590–1608. 385–397 (2000). doi: 10.1016/S0896-6273(00)80902-2; adrenergic and cholinergic activity in cortex. Nat. Commun.
e23 (2019). doi: 10.1016/j.cell.2019.11.004; pmid: 31835034 pmid: 10719893 7, 13289 (2016). doi: 10.1038/ncomms13289; pmid: 27824036
21. H. H. Yang et al., Subcellular imaging of voltage and calcium 44. C. Cai et al., VolPy: Automated and scalable analysis pipelines 67. J. M. P. Pakan et al., Behavioral-state modulation of inhibition
signals reveals neural processing in vivo. Cell 166, 245–257 for voltage imaging datasets. PLOS Comput. Biol. 17, e1008806 is context-dependent and cell type specific in mouse visual
(2016). doi: 10.1016/j.cell.2016.05.031; pmid: 27264607 (2021). doi: 10.1371/journal.pcbi.1008806; pmid: 33852574 cortex. eLife 5, e14985 (2016). doi: 10.7554/eLife.14985;
22. Y. Adam et al., Voltage imaging and optogenetics reveal 45. H. Inan, M. A. Erdogdu, M. J. Schnitzer, “Robust estimation of pmid: 27552056
behaviour-dependent changes in hippocampal dynamics. neural signals in calcium imaging” in NIPS’17: Proceedings of 68. S. Xiao, H. A. Tseng, H. Gritton, X. Han, J. Mertz, Video-rate
Nature 569, 413–417 (2019). doi: 10.1038/s41586-019-1166-7; the 31st International Conference on Neural Information volumetric neuronal imaging using 3D targeted illumination.
pmid: 31043747 Processing Systems, U. von Luxburg, I. Guyon, S. Bengio, Sci. Rep. 8, 7921 (2018). doi: 10.1038/s41598-018-26240-8;
23. K. D. Piatkevich et al., Population imaging of neural activity H. Wallach, R. Fergus, S. V. N. Vishwanathan, R. Garnett, Eds. pmid: 29784920
in awake behaving mice. Nature 574, 413–417 (2019). (Curran Associates Inc., 2017). 69. M. L. Andermann et al., Chronic cellular imaging of entire cortical
doi: 10.1038/s41586-019-1641-1; pmid: 31597963 46. H. Inan et al., Fast and statistically robust cell extraction from columns in awake mice using microprisms. Neuron 80, 900–913
24. S. Xiao et al., Large-scale voltage imaging in behaving mice large-scale neural calcium imaging datasets. bioRxiv (2013). doi: 10.1016/j.neuron.2013.07.052; pmid: 24139817
using targeted illumination. iScience 24, 103263 (2021). 2021.2003.2024.436279 [Preprint] (2021); doi: 10.1101/ 70. J. C. Jung, A. D. Mehta, E. Aksay, R. Stepnoski, M. J. Schnitzer,
doi: 10.1016/j.isci.2021.103263; pmid: 34761183 2021.03.24.436279 In vivo mammalian brain imaging using one- and two-photon

fluorescence microendoscopy. J. Neurophysiol. 92, 3121–3133 J. Comput. Neurosci. 38, 439–459 (2015). doi: 10.1007/ conceived the project; designed, screened and characterized
(2004). doi: 10.1152/jn.00234.2004; pmid: 15128753 s10827-015-0547-7; pmid: 25652689 the sensors; and performed molecular biology and rAAV cloning.
71. R. P. Barretto, B. Messerschmidt, M. J. Schnitzer, In vivo 83. E. A. Pnevmatikakis et al., Simultaneous denoising, M.K. performed visual cortical surgeries. G.V. built the in vivo
fluorescence imaging with high-resolution microlenses. deconvolution, and demixing of calcium imaging data. Neuron setup and wrote the software for V1 imaging. M.K. and
Nat. Methods 6, 511–512 (2009). doi: 10.1038/nmeth.1339; 89, 285–299 (2016). doi: 10.1016/j.neuron.2015.11.037; G.V. conceived and performed in vivo V1 experiments and data
pmid: 19525959 pmid: 26774160 analyses. S.H. conceived and performed the hippocampal
72. R. P. Barretto, M. J. Schnitzer, In vivo optical microendoscopy 84. E. A. Mukamel, A. Nimmerjahn, M. J. Schnitzer, Automated experiments, wrote the SpikeImagingAnalysis pipeline, and
for imaging cells lying deep within live tissue. Cold Spring analysis of cellular signals from large-scale calcium imaging performed data analyses for CA1 studies. C.H. performed and
Harb. Protoc. 2012, 1029–1034 (2012). doi: 10.1101/ data. Neuron 63, 747–760 (2009). doi: 10.1016/ analyzed the fly experiments. R.C. built the optical setup used
pdb.top071464; pmid: 23028071 j.neuron.2009.08.009; pmid: 19778505 for CA1 imaging. J.L. designed the transgenic flies. J.A.C. shared
73. Allen Institute for Brain Science, Cell types: Overview of the data, 85. T. Deneux et al., Accurate spike estimation from noisy the running wheel and provided advice on V1 imaging. M.K.
Allen cell types database; https://celltypes.brain-map.org/. calcium signals for ultrafast three-dimensional imaging of and G.V. wrote the manuscript with contributions from all
74. G. Buzsáki, C. A. Anastassiou, C. Koch, The origin of extracellular large neuronal populations in vivo. Nat. Commun. 7, 12190 authors. M.J.S. developed data analyses, edited the manuscript,
fields and currents—EEG, ECoG, LFP and spikes. Nat. Rev. Neurosci. (2016). doi: 10.1038/ncomms12190; pmid: 27432255 and supervised work at Stanford. V.A.P. supervised work at
13, 407–420 (2012). doi: 10.1038/nrn3241; pmid: 22595786 86. M. Kannan et al, Dual-polarity voltage imaging of the John B. Pierce and Yale. Competing interests: The authors
75. I. Soltesz, A. Losonczy, CA1 pyramidal cell diversity enabling concurrent activation dynamics of multiple neuron types. declare no competing interests. Data and materials
parallel information processing in the hippocampus. Zenodo (2022); doi: 10.5281/zenodo.6361012 availability: All sequence information is available on GenBank
Nat. Neurosci. 21, 484–493 (2018). doi: 10.1038/s41593-018- (accession numbers OM687163 to OM687166 for soma-targeted
0118-0; pmid: 29593317 ACKN OWLED GMEN TS Ace-mNeon2, pAce, VARNAM2, and pAceR, respectively).
76. L. Jin et al., Single action potentials and subthreshold electrical Plasmids and rAAVs will be made available through Addgene
We thank Pierce scientific staff members C. Gardiner, ([email protected]). Fly stocks have been deposited at the
events imaged in neurons with a fluorescent protein
M. Izydorsczak, P. O’Brien, X. Liu, T. Liu, and R. O’Brien for Bloomington Drosophila Stock Center, and all data and codes
voltage probe. Neuron 75, 779–785 (2012). doi: 10.1016/
technical assistance; Pierce workshop members J. Buckley are available at Zenodo (86). License information: Copyright ©
j.neuron.2012.06.040; pmid: 22958819
and A. Wilkins; G. Lur (University of California, Irvine) 2022 the authors, some rights reserved; exclusive licensee
77. C. Huang et al., Long-term optical brain imaging in live adult
and K. Ferguson (Yale School of Medicine, New Haven) for American Association for the Advancement of Science. No claim
fruit flies. Nat. Commun. 9, 872 (2018). doi: 10.1038/
advice on V1 imaging; and J. Verhagen (Yale Neuroscience) to original US government works. https://www.science.org/
s41467-018-02873-1; pmid: 29491443
for comments on the manuscript. We further thank members about/science-licenses-journal-article-reuse. This research was
78. G. J. Goldey et al., Removable cranial windows for long-term
of the Schnitzer laboratory: J. Li for animal husbandry and funded in whole or in part by the Howard Hughes Medical
imaging in awake mice. Nat. Protoc. 9, 2515–2538 (2014). genotyping, Y. Zhang for virus handling and surgical advice, Institute, a cOAlition S organization. The authors will make the
doi: 10.1038/nprot.2014.165; pmid: 25275789 G. Delamare for instrumentation support, and F. Dinc and J. Li Author Accepted Manuscript (AAM) version available under a
79. P. Thévenaz, U. E. Ruttimann, M. Unser, A pyramid approach to for computational consultations. All schematics were created CC BY public copyright license.
subpixel registration based on intensity. IEEE Trans. Image Process. using BioRender. Funding: This study was supported
7, 27–41 (1998). doi: 10.1109/83.650848; pmid: 18267377 by NIH BRAIN Initiative grants U01NS103517 (V.A.P.), SUPPLEMENTARY MATERIALS
80. E. A. Pnevmatikakis, A. Giovannucci, NoRMCorre: An online U01NS120822 (M.J.S. and G.V.), UF1NS107610 (M.J.S. and
algorithm for piecewise rigid motion correction of calcium science.org/doi/10.1126/science.abm8797
H. Zheng), and U19NS104590 (M.J.S. and I. Soltesz); and NSF
imaging data. J. Neurosci. Methods 291, 83–94 (2017). Figs. S1 to S26
NeuroNex grant DBI-1707261 (M.J.S. and K. Deisseroth). This
doi: 10.1016/j.jneumeth.2017.07.031; pmid: 28782629 Table S1
research was funded in part by the Defense Advanced Research
References (87, 88)
81. K. He, G. Gkioxari, P. Dollar, R. Girshick, Mask R-CNN. IEEE Projects Agency (DARPA) of the United States, contract nos.
MDAR Reproducibility Checklist
Trans. Pattern Anal. Mach. Intell. 42, 386–397 (2020). N6600117C4012 (NESD) and N6600119C4020 (N3) (V.A.P.).
doi: 10.1109/TPAMI.2018.2844175; pmid: 29994331 The views, opinions and/or findings expressed are those of the
82. F. Franke, R. Quian Quiroga, A. Hierlemann, K. Obermayer, authors and should not be interpreted as representing the
Bayes optimal template matching for spike sorting - official views or policies of the Department of Defense or Submitted 22 October 2021; accepted 3 October 2022
combining fisher discriminant analysis with optimal filtering. the US government. Author contributions: M.K. and G.V. 10.1126/science.abm8797

RES EARCH
◥ vious observations (18), SpecR WT did not

RESEARCH ARTICLES confer spectinomycin resistance to Syn61D3
cells (Fig. 1).
SYNTHETIC BIOLOGY We created a recoded spectinomycin re-
sistance gene [recSpecR (DTCG, TCA)], with
Refactored genetic codes enable bidirectional the compressed genetic code used in the Syn61
genome. recSpecR (DTCG, TCA) conferred
genetic isolation spectinomycin resistance to Syn61D3 cells
(Fig. 1). The recSpecR (DTCG, TCA) gene also
Jérôme F. Zürcher1, Wesley E. Robertson1, Tomás Kappes2, Gianluca Petris1, Thomas S. Elliott1, conferred spectinomycin resistance to cells
George P. C. Salmond2, Jason W. Chin1* that read the canonical genetic code; this was
expected because the compressed genetic code
The near-universal genetic code defines the correspondence between codons in genes and amino uses a subset of the codons used in the canon-
acids in proteins. We refactored the structure of the genetic code in Escherichia coli and created ical genetic code. We made similar observations
orthogonal genetic codes that restrict the escape of synthetic genetic information into natural life. with hygromycin resistance genes written in
We developed orthogonal and mutually orthogonal horizontal gene transfer systems, which permit the canonical and compressed codes (fig. S1).
transfer of genetic information between organisms that use the same genetic code but restrict the These experiments demonstrated that ge-
transfer of genetic information between organisms that use different genetic codes. Moreover, we netic information written in the canonical code
showed that locking refactored codes into synthetic organisms completely blocks invasion by mobile can be read in cells that decode the canonical
genetic elements, including viruses, which carry their own translation factors and successfully code, but not in cells with genome-wide code
invade organisms with canonical and compressed genetic codes. compression and cognate tRNA deletion. How-
ever, code-compressed genes can be read in
both cells with cognate tRNA deletion and cells
he near-universal genetic code defines supports arginine codon reassignment in that decode the canonical code. Therefore,
T
the correspondence between codons in bacilli (12). there is no barrier limiting the flow of genetic
genes and amino acids in proteins (1, 2). Genome synthesis (13, 14) and editing pro- information from engineered organisms with
Because all forms of life use essentially vide the opportunity to rewrite the genetic compressed genetic codes to natural forms of
the same genetic code, evolutionary in- code and create organisms with new properties life. Creating orthogonal genetic codes that
novation can be shared through horizontal (14–19). We synthesized a 4-Mb Escherichia actively restrict the transfer of genetic informa-
gene transfer (HGT) between organisms (3, 4), coli genome in which we replaced all annotated tion from engineered biological systems to nat-
and this is a major driver of evolution (5). occurrences of the TCG and TCA serine codons ural systems is an important and unaddressed
However, the near-universal genetic code with the synonymous AGC and AGT codons challenge.
is also a liability for organisms; mobile ge- using defined recoding rules (20); we also
netic elements (or selfish genetic elements)— replaced the TAG stop codon with TAA. This tRNAs enable invasion of codon-compressed
including transposons, viruses, and plasmids— created Syn61, an organism with a compressed organisms
exploit the universality of the code and co-opt genetic code (14). We further evolved the strain A WT F plasmid [F (WT)], written in the ca-
the host cell’s machinery to read their genes and deleted the genes for the tRNAs that de- nonical genetic code, was efficiently transferred
and propagate at the expense of host organ- code TCG and TCA codons (serU, tRNACGASer; to cells that read the canonical code. By con-
isms. There is a clear tension between main- serT, tRNAUGASer) and the gene for RF-1 (prfA) trast, F (WT) was not transferred to Syn61D3
taining a common genetic code, to allow the that terminates protein synthesis at the TAG (Fig. 1 and data file S1), as expected. However,
acquisition of beneficial innovation through stop codon. The resulting organism, Syn61D3, upon selecting for the conjugation of F (WT)
HGT, and excluding selfish genetic elements cannot read all the codons in the near-universal from cells that read the canonical code into
that exploit the common code for their own genetic code and therefore cannot read hori- Syn61D2 cells (Syn61 cells deleted for serU and
ends (3, 6). zontally transferred genes that contain the serT but containing prfA), we obtained two
Several deviations from the standard genetic codons deleted from its genome, as exem- viable colonies in which recipient cells had
code have been documented in mitochondria plified by resistance to a range of bacterio- received F (WT) (fig. S2). These colonies cor-
and chloroplasts, and most characterized phage (18). responded to rare events, which appeared
code reassignments involve stop codons (7–9). It has been widely hypothesized that re- at a frequency 106-fold less than the colonies
Known sense codon reassignments in the nu- factoring the structure of the genetic code, resulting from conjugation of F (WT) into cells
clear genome are rare. The “CTG yeast” de- through the reassignment of sense codons to that read the canonical code. Sequencing the
codes the CUG codon (which encodes leucine distinct canonical amino acids, would create two clones revealed that they had acquired
in the standard code) primarily as serine organisms with new properties and could sequences that contained serT from the donor
(97%, with the remaining 3% still assigned create a genetic firewall to limit the escape of cell. This provided direct experimental evidence
to leucine) (10). Viruses for the CTG yeasts genetic information from synthetic organisms that selection for transfer of a mobile genetic
are essentially unknown, which suggests that to natural organisms (4, 6, 21–23). In this work, element that uses the canonical code to recip-
sense codon reassignment may protect against we tested these hypotheses. ients that cannot read the entire canonical
viruses (11). There are no experimentally val- code can enable selection for recipients that
idated examples of sense codon reassignment Compressed codes are nonorthogonal acquire the tRNA genes necessary to read the
in bacteria, although computational evidence A spectinomycin resistance gene written in canonical genetic code.
the canonical genetic code [SpecR wild type To follow the effects of introducing serT
(WT)] was correctly read in, and conferred into recipient cells in a reproducible sys-
1
Medical Research Council Laboratory of Molecular Biology, spectinomycin resistance to, cells that con- tem, we created the mobile genetic element F
Cambridge, UK. 2Department of Biochemistry, University of
Cambridge, Cambridge, UK. tain the full complement of tRNAs to read the (WT + serT), a variant of F (WT) that contains
*Corresponding author. Email: [email protected] canonical code. However, consistent with pre- serT. We demonstrated that F (WT + serT)

RESE ARCH | R E S E A R C H A R T I C L E S
A B
C E
Fig. 1. Compressed genetic codes are nonorthogonal. (A) The relationship HGT outcomes from mobile genetic elements (MGEs) and recipient cells with the
between the TCG and TCA codons in genes, the decoders for these codons indicated decoders and codons in essential genes. (C) A mobile genetic
in cells with canonical (WT) decoding and Syn61D3 decoding (DTCG, DTCA), and the element encoding its genes according to the canonical genetic code, in which
corresponding protein sequence synthesized. The anticodon of the tRNAs that TCG and TCA encode serine, cannot be horizontally transferred to Syn61D3 cells
read TCG or TCA codons is indicated (decoder). The amino acid (aa) used that have no decoders for TCG and TCA codons. Translation will stall at TCG
by the tRNA is indicated. Gray for a decoder indicates that the tRNA is loaded and TCA codons, and no full-length protein will be synthesized from the essential
with serine. Gray for a codon indicates that the codon is within a non–codon- genes within the mobile genetic element that contain TCG and TCA codons.
compressed gene, and its decoding as serine will make the correct protein (D) A mobile genetic element encoding its genes according to the canonical
sequence. Pink for a decoder–amino acid pair indicates that the tRNA is genetic code that also carries a gene for a tRNA decoding TCG and TCA codons
deleted. Pink for a codon indicates the codon is absent from the gene because can be horizontally transferred to Syn61D3 cells. The tRNA encoded on the
the gene has been designed with codon compression. (B) Functional mobile genetic element can rescue decoding of TCG and TCA codons within
assessment of SpecR WT and codon-compressed spectinomycin resistance essential genes in the mobile genetic element to make the correct MGE proteins.
[recSpecR (DTCG, TCA)] genes in (left) cells that use the full complement (E) Transfer of mobile genetic elements through conjugation. Colony count
of tRNAs to decode all the codons in the reading frame (Syn61 WT) and indicates successful transconjugants received from ~106 cells. A WT mobile
(right) cells in which the tRNAs that decode TCG and TCA codons have been genetic element (F WT) can be transferred into cells that read the canonical
deleted (Syn61D3). Cells were spotted on agar plates in the presence or code (Syn61 WT) but not into cells that lack tRNAs decoding TCG and TCA
absence of spectinomycin and incubated overnight. The growth of cells in the (Syn61D3). A WT mobile genetic element that encodes a tRNA decoding
presence of spectinomycin indicates that the indicated SpecR gene is TCG and TCA codons as serine [F (WT + serT)] can be transferred to both Syn61
functional in the indicated strain. (C and D) Predicted protein synthesis and WT and Syn61D3.

RES EARCH | R E S E A R C H A R T I C L E S
Fig. 2. Sense codon

reassignment generates
genetic codes. (A) Total
synthesis of a codon-
compressed genome followed
by tRNA and release factor
deletion yielded Syn61D3.
The discovery of tRNAs that
direct the incorporation
of distinct canonical amino
acids in response to TCG or
TCA codons enables sense
codon reassignment to
create new genetic codes.
(B) Isoacceptor tRNAs for the
indicated amino acids with
anticodons altered to the
Watson-Crick complement of
TCG or TCA codons were
introduced into Syn61D3 in
the indicated pairwise
combinations. We read out
the identity of the amino
acid incorporated into each
codon using GFP genes
with TCG or TCA codons
at position 3 and electrospray
ionization mass spectrometry.
When pairs of isoacceptors
for distinct amino acids
were used, each codon led to
the specific incorporation
of the amino acid attached to
the Watson-Crick–paired
isoacceptor. The secondary
peak measured in the proline
incorporations results from
incomplete methionine cleave
at the N terminus. A complete
list of found and expected
masses are provided in data
file S2. MW, molecular
weight. (C) Sixteen genetic
codes in which TCG and
TCA codons are reassigned
to alanine, histidine, leucine,
and proline.
can be transferred to Syn61D3 cells and that deletion in Syn61D3. These experiments high- Refactoring code structure
this transfer is dependent on serT (Fig. 1). We light that creating systems that actively ob- We found that chimeric tRNAs for alanine
conclude that acquisition of serT is sufficient struct invasion by mobile genetic elements (tRNACGAAla, tRNAUGAAla), histidine (tRNACGAHis,
to circumvent the genetic isolation that is pro- that carry their own decoders is an impor- tRNAUGAHis), leucine (tRNACGALeu, tRNAUGALeu),
vided by code compression and cognate tRNA tant challenge. and proline (tRNACGAPro, tRNAUGAPro) specifically

Fig. 3. Orthogonal genetic systems. (A) Relationship between the TCG and (B) Functional assessment of SpecR WT (written in the canonical genetic code)
TCA codons in genes; the decoders for these codons in cells with canonical (WT) and O-SpecR (TCG-Ala, TCA-His), which is codon-compressed according to the
decoding and decoding by tRNACGAAla, tRNAUGAHis in Syn61D3; and the Syn61 recoding scheme and has alanine codons replaced with TCG and histidine
corresponding protein sequence synthesized. The anticodon of the tRNAs that codons replaced with TCA. The genes are read in cells that read the canonical
read TCG or TCA codons is indicated (decoder). The amino acid (aa) used by the code (Syn61 WT), and cells where TCG is decoded as alanine and TCA is decoded
tRNA is indicated. Gray for a codon indicates that its decoding as serine will as histidine [Syn61D3 (tRNACGAAla, tRNAUGAHis)]. Cells were spotted on agar
make the correct protein sequence. Yellow for a codon indicates that its plates in the presence or absence of spectinomycin and incubated overnight. The
decoding as alanine will make the correct protein sequence. Green for a codon growth of cells in the presence of spectinomycin indicates that the indicated
indicates that its decoding as histidine will make the correct protein sequence. SpecR gene is functional in the indicated strain.
direct the incorporation of the amino acid ical amino acids encoded at specific sense make the correct protein product in cells that
defined by the parent isoacceptor tRNA in codons with respect to both the canonical decode the TCG and TCA codons to incorpo-
response to their cognate codon (TGC or TCA) code and the other 15 codes we have created rate the correct amino acid. However, these
at position 3 in superfolder green fluorescent (Fig. 2C). Our reassignment strategy is analo- synthetic genes will yield an incorrect, likely
protein (sfGFP) or position 11 in ubiquitin in gous to models that have been proposed for nonfunctional protein product in cells that
Syn61D3 (Fig. 2, figs. S3 to S13, and data files codon capture in natural evolution (24). read the canonical genetic code (Fig. 3).
S2 and S3) and produce good yields of protein Overall, we have refactored the structure We converted all 27 GCN codons (which en-
(fig. S3). The fidelity of tRNAUGALeu was lower of the genetic code. Our genetic codes ex- code alanine in the canonical code) to TCG
than that of other tRNAs (data file S3 and fig. pand the number of codons used to encode codons and all six CAT/C codons (which en-
S5). We investigated alanyl and leucyl-tRNAs alanine and proline (from four to six), double code histidine in the canonical code) to TCA
because their anticodons are not identity ele- the number of codons used to encode histi- codons in recSpecR (DTCG, TCA). This created
ments for their cognate aminoacyl-tRNA syn- dine from two to four, and increase the number the orthogonal resistance gene O-SpecR (TCG-
thetases, and they were therefore expected of codons used to primarily encode leucine Ala, TCA-His). We demonstrated that O-SpecR
to be permissive to anticodon mutation; the from six to eight. These experiments also (TCG-Ala, TCA-His) can be decoded in, and
other tRNAs were identified through a screen show that the UCN codon box, which en- confer spectinomycin resistance to, Syn61D3
(fig. S14). We found that unlike tRNACGASer codes serine in the canonical code, can be (tRNACGAAla, tRNAUGAHis) cells, in which TCG
and tRNAUGASer, our chimeric tRNAs specif- split to encode additional canonical amino is read as alanine and TCA is read as histidine.
ically decode the Watson-Crick complement acids. We further demonstrated that O-SpecR (TCG-
of their anticodon sequence; for example, Ala, TCA-His) did not confer spectinomycin
tRNACGAAla decodes TCG codons in preference Orthogonal code–orthogonal decoder pairs resistance to cells that read the canonical ge-
to TCA codons, and tRNAUGAAla decodes TCA Genes that are written by using the canonical netic code. Last, we demonstrated that SpecR
codons in preference to TCG codons (figs. S3 genetic code, in which TCG and TCA codons WT, in which serine is encoded by using TCG
and S5). These tRNAs are also specific with encode serine, will make the correct protein and TCA codons, cannot confer spectinomycin
respect to other TCN codons (figs. S6 and S7), product in natural cells that read these codons resistance to Syn61D3 (tRNACGAAla, tRNAUGAHis)
and most reassigned strains grew comparably as serine. However, these genes will yield the cells (Fig. 3). We extended this approach to
with parental strains (fig. S15). Our data show incorrect, likely nonfunctional protein product six other reassignment schemes, as well as to
that we can independently reassign the TCA in cells that decode these codons to incorporate other genes (fig. S16).
and TCG codons to alanine, histidine, leucine, amino acids other than serine. These experiments demonstrated that we
or proline in Syn61D3 and thereby create 16 Similarly, synthetic genes—in which we com- can create a genetic code–decoder pair for
new genetic codes (Fig. 2B, figs. S3 and S5, and press the genetic code using the Syn61 recoding synthetic genes that is functionally orthogonal
data files S2 and S3). In each of these genetic scheme and replace codons for specific natural with respect to the canonical genetic code–
codes, we changed the identity of the canon- amino acids with TCG and TCA codons—will decoder pair for natural genes. The orthogonal

Fig. 4. Orthogonal and mutually orthogonal HGT systems. (A) HGT cells where TCG is reassigned to alanine and TCA is reassigned to
between organisms that use distinct, orthogonal genetic codes is prohibited histidine [Syn61D3 (tRNACGAAla, tRNAUGAHis)]. An orthogonal mobile genetic
(dashed gray arrows), whereas HGT can occur between cells that share element (O-F1), in which alanine codons were replaced with TCG and
a common genetic code (solid arrows). (B) Orthogonal horizontal transfer histidine codons were replaced with TCA, was transferred into Syn61D3
of mobile genetic elements. Colony count indicates the number of (tRNACGAAla, tRNAUGAHis) but not into Syn61 WT. (C) Mutually orthogonal
transconjugants received from ~106 donor cells that bear the indicated HGT systems. Colony count (orange heat map) indicates successful
mobile genetic element. A WT mobile genetic element [F (WT)] was transferred transconjugants received from ~106 donor cells bearing the indicated
into cells that read the canonical genetic code (Syn61 WT) but not into mobile genetic element.
code, written in synthetic genes, is correctly by using computational approaches that lever- canonical code can transfer a WT mobile ge-
read by the cognate orthogonal decoder but age evolutionary sequence- and/or structural netic element between themselves but can-
not by the canonical decoder. The canonical information (26–29). Although the composi- not transfer the WT mobile genetic element
code, written in natural genes, is correctly tion of natural genes is fixed, the codon usage to cells that contain orthogonal decoders.
read by the canonical decoder but not by the in synthetic genes—written in the standard Cells that contain O-HGT systems can trans-
orthogonal decoder. code or any orthogonal code—can be simply fer their mobile genetic element to cells that
The functional orthogonality of genes in designed to maximize the number of codons contain a compatible orthogonal decoder but
cells with altered decoders will depend on the that are subject to reassignment, which may cannot transfer their mobile genetic element
frequency of reassigned codons and the func- maximize the functional orthogonality of syn- to cells that contain an incompatible orthog-
tional consequences of codon reassignments. thetic genes. onal decoder or to cells that read the canon-
The consequences of amino acid substitutions— ical code.
a result of codon reassignment—may globally Orthogonal HGT A mobile genetic element [F (WT)] written
and crudely correlate with differences in amino Next, we created orthogonal HGT (O-HGT) in the canonical code was transferred to cells
acid polarity and hydrophobicity (6, 25). The systems composed of an orthogonal decoder that read the canonical code, as expected. We
consequences of amino acid substitutions at and a mobile genetic element that uses an also showed that F (WT) could not be trans-
particular sites in proteins may be predicted orthogonal genetic code. Cells that read the ferred to Syn61D3 (tRNACGAAla, tRNAUGAHis)

Fig. 5. Orthogonal code locking blocks invading codes. (A and B) Predicted histidine. In addition, essential genes in the host cell, in which TCG is used to
protein synthesis and HGT outcomes from mobile genetic elements (MGEs) and encode alanine and TCA is used to encode histidine, will be missynthesized.
recipient cells with the indicated decoders, and codons in essential genes. This is predicted to further restrict HGT. (C) HGT of a WT mobile genetic
(A) Transfer of a WT mobile genetic element that encodes for a tRNA decoding element [F (WT + serT)] is ablated in cells that use a refactored genetic code in
TCG and TCA codons as serine into a cell where TCG is reassigned to alanine essential genes. Colony count indicates successful transconjugants received
and TCA is reassigned to histidine. Essential genes in the WT mobile genetic from ~106 cells. Recipient cells and spectinomycin resistance gene variant
element, which contain TCG and TCA codons, will be missynthesized, with (SpecR gene) in the recipient cell are indicated. Correctly reading the indicated
each TCG and TCA codon in the gene being stochastically decoded as serine or SpecR gene in the recipient cell is made essential by addition of spectinomycin.
alanine/histidine. This is predicted to attenuate HGT. (B) Transfer of a WT (D) T4-like phage encoding a seryl-tRNAUGA infect Syn61D3 but not cells
mobile genetic element that encodes for a tRNA decoding TCG and TCA that bear orthogonal genetic codes. Plaque count indicates the number of
codons as serine into a cell where TCG is reassigned to alanine and TCA is successfully replicating phage obtained from infection with 1.1 × 1010 plaque-
reassigned to histidine. Essential genes in the WT mobile genetic element, forming units (PFU)/ml (phage 12) and 7.5 × 109 PFU/ml (phage 6). Cells
which contain TCG and TCA codons, will be missynthesized, with each TCG and contain cognate spectinomycin resistance genes, as in (C); all experiments
TCA codon in the gene being stochastically decoded as serine or alanine/ were performed in the presence of spectinomycin.
cells, in which TCG codons are read as alanine in the canonical code) and CAT/C codons zontally transferred to cells that read the
nine and TCA codons are read as histidine (which encode histidine in the canonical code) canonical genetic code (Fig. 4). These exper-
(Fig. 4). were converted to TCG and TCA codons, re- iments demonstrated that we can create
Next, we investigated HGT for mobile genetic spectively, within the trfA gene; this gene is O-HGT systems. We created additional HGT
elements written in altered genetic codes. We essential for the replication of the mobile ge- systems that are orthogonal to the natural
synthesized the mobile genetic element O-F1 netic element (30). genetic system and mutually orthogonal to
(TCG-Ala, TCA-His). The genetic code in all O-F1 (TCG-Ala, TCA-His) was horizon- each other (Fig. 4 and fig. S17). Overall, we
annotated open reading frames of this F tally transferred to Syn61D3 (tRNA CGA Ala , demonstrated the scalability of our approach
plasmid was compressed by using the Syn61 tRNAUGAHis) cells. We further demonstrated through the creation of five mutually orthog-
scheme, and GCN codons (which encode ala- that O-F1 (TCG-Ala, TCA-His) was not hori- onal HGT systems.

Blocking invading codes ture of the genetic code and its robustness and 3. R. J. Hall, F. J. Whelan, J. O. McInerney, Y. Ou,
We hypothesized that reassigning TCA and evolvability (21, 22), it has been impossible to M. R. Domingo-Sananes, Front. Microbiol. 11, 1569 (2020).
4. K. Vetsigian, C. Woese, N. Goldenfeld, Proc. Natl. Acad. Sci.
TCG codons to specific natural amino acids and investigate the resulting hypotheses through U.S.A. 103, 10696–10701 (2006).
replacing codons for specific natural amino experiments in living cells. Refactoring the 5. S. M. Soucy, J. Huang, J. P. Gogarten, Nat. Rev. Genet. 16,
acids in essential genes with TCA and TCG structure of the genetic code directly alters the 472–482 (2015).
6. E. V. Koonin, A. S. Novozhilov, IUBMB Life 61, 99–111
codons in Syn61D3 (Fig. 5) would obstruct number and types of amino acids that can be (2009).
the serT-mediated HGT we observed in this accessed by point mutations in live cells (figs. 7. M. Kollmar, S. Mühlhausen, BioEssays 39, 1600221 (2017).
strain (Fig. 1). S23 and S24) and provides opportunities to 8. J. Ling et al., Nucleic Acids Res. 42, 499–508 (2014).
9. A. L. Borges et al., Nat. Microbiol. 7, 918–927 (2022).
Transfer of F (WT + serT) to Syn61D3 experimentally test how altered codes affect 10. M. A. Santos, A. C. Gomes, M. C. Santos, L. C. Carreto,
(tRNACGAAla, tRNAUGAHis, O-SpecR [TCG-Ala, the robustness and evolvability of protein and G. R. Moura, C. R. Biol. 334, 607–611 (2011).
TCA-His]) was obstructed (104-fold) in the ab- cellular function. In future work, we will aim 11. D. J. Taylor, M. J. Ballinger, S. M. Bowman, J. A. Bruenn,
PeerJ 1, e50 (2013).
sence of spectinomycin because tRNACGAAla to leverage genetic code refactoring to accel-
12. Y. Shulgina, S. R. Eddy, eLife 10, e71402 (2021).
and tRNAUGAHis compete with tRNAUGASer in erate directed evolution. 13. D. G. Gibson et al., Science 319, 1215–1220 (2008).
the recipient cell to decrease the production of Competition between pools of genotypes 14. J. Fredens et al., Nature 569, 514–518 (2019).
functional proteins from the mobile genetic written in different genetic codes may have 15. M. J. Lajoie et al., Science 342, 357–360 (2013).
16. D. J. Mandell et al., Nature 527, 264–264 (2015).
element. However, this obstruction was not led to a universal code being fixed in extant 17. A. J. Rovner et al., Nature 527, 264 (2015).
sufficient to completely ablate transfer of F life. Future work may exploit organisms with 18. W. E. Robertson et al., Science 372, 1057–1062 (2021).
(WT + serT). Upon the addition of spectino- refactored and mutually orthogonal codes 19. D. de la Torre, J. W. Chin, Nat. Rev. Genet. 22, 169–184 (2021).
20. K. Wang et al., Nature 539, 59–64 (2016).
mycin, which makes O-SpecR (TCG-Ala, TCA- to experimentally investigate competition 21. G. Pines, J. D. Winkler, A. Pines, R. T. Gill, mBio 8, e01654-17
His) an essential gene in the cell, the decoding between codes and the role of HGT in fix- (2017).
of TCG codons as alanine and the decoding ing and maintaining a universal genetic code 22. J. Calles, I. Justice, D. Brinkley, A. Garcia, D. Endy, Nucleic
Acids Res. 47, 10439–10451 (2019).
of TCA codons as histidine become essential (4, 36). 23. P. Marliere, Syst. Synth. Biol. 3, 77–84 (2009).
and “locked in.” Under these conditions, the Shielding synthetic organisms from envi- 24. S. Osawa, T. H. Jukes, J. Mol. Evol. 28, 271–278 (1989).
transfer of F (WT + serT) was completely ab- ronmental genetic elements may be valuable 25. M. Schmidt, V. Kubyshkin, ChemBioChem 22, 1268–1284
(2021).
lated (Fig. 5). Similar results were obtained for biotechnological applications on an in- 26. J. Frazer et al., Nature 599, 91–95 (2021).
with other refactored codes and other essen- dustrial scale, for which contamination with 27. S. Teng, A. K. Srivastava, C. E. Schwartz, E. Alexov, L. Wang,
tial genes (figs. S18 and S19). mobile genetic elements, including viruses, can Int. J. Comput. Biol. Drug Des. 3, 334–349 (2010).
28. V. Parthiban, M. M. Gromiha, D. Schomburg, Nucleic Acids Res.
To extend our approach to viral infection, cause financial losses and disrupt vital supply
34, W239–W242 (2006).
we identified pools of phage from the River chains (37). Synthetic organisms with decoders 29. P. C. Ng, S. Henikoff, Annu. Rev. Genomics Hum. Genet. 7,
Cam that can infect Syn61D3 (data file S4 for an orthogonal code and essential genes that 61–80 (2006).
and supplementary materials, materials and lock the cognate orthogonal code into the 30. I. Konieczny, K. S. Doran, D. R. Helinski, A. Blasina, J. Biol.
Chem. 272, 20173–20178 (1997).
methods). From these pools, we isolated two organism exhibit resistance to mobile genetic 31. V. A. Delesalle, N. T. Tanke, A. C. Vill, G. P. Krukonis,
individual phage (12 and 06, both T4-like elements that are written in the canonical code. Bacteriophage 6, e1219441 (2016).
phage), which carry an identical tRNAUGASer This resistance extends to mobile genetic ele- 32. M. Bailly-Bechet, M. Vergassola, E. Rocha, Genome Res. 17,
1486–1495 (2007).
gene and infect Syn61D3 (fig. S20); some vi- ments that carry tRNA genes that otherwise 33. B. A. Renda, M. J. Hammerling, J. E. Barrick, Mol. Biosyst. 10,
ruses are known to carry their own tRNAs and allow correct reading of the canonical code in 1668–1678 (2014).
other translation factors to augment the cel- their genes. Our work defines a paradigm for 34. G. Moratorio et al., Nat. Microbiol. 2, 17088 (2017).
35. J. R. Coleman et al., Science 320, 1784–1787 (2008).
lular pool of translation factors and assist in creating organisms that actively resist invasion
36. V. Kubyshkin, C. G. Acevedo-Rocha, N. Budisa, Biosystems
the translation of codons within their own by foreign codes. 164, 16–25 (2018).
genes (9, 31, 32). As expected, expression of Our refactored genetic codes limit the 37. P. W. Barone et al., Nat. Biotechnol. 38, 563–572 (2020).
this tRNA in Syn61D3 is sufficient to confer transfer of genetic information from syn- 38. P. Marlière et al., Angew. Chem. Int. Ed. 50, 7109–7114
(2011).
susceptibility to infection by (otherwise non- thetic organisms to natural organisms and
39. J. W. Lee, C. T. Y. Chan, S. Slomovic, J. J. Collins, Nat. Chem.
infectious) T4 phage (fig. S21). We demon- may form the basis of genetic firewalls that Biol. 14, 530–537 (2018).
strated that unlike Syn61D3, several refactored, isolate synthetic genetic systems from the
code-locked strains were completely resistant environment. Such genetic firewalls com- AC KNOWLED GME NTS
to infection with phage 06 and phage 12 (Fig. plement strategies for controlling the sur- We thank J. Fredens (National University of Singapore) for
advice on conjugation assays; S. Grazioli and A. Kleefeldt for help
5 and fig. S22). vival and growth of synthetic organisms, to with de novo assembly of phage genomes; and the LMB mass
Our results demonstrate that cells with re- realize biocontainment (16, 38, 39). This is spectrometry facility, especially S.-Y. Peak-Chew, for performing
factored genetic codes resist invasion by mob- especially important when considering appli- mass spectrometry experiments. Funding: This work was supported
by the Medical Research Council (MRC), UK (MC_U105181009,
ile genetic elements that use competing codes. cations of engineered organisms outside the MC_UP_A024_1008 to J.W.C.); G.P. was supported by a Marie
Locking the refactored codes into essential laboratory. Skłodowska-Curie European Postdoctoral Fellowship (897663).
genes in the cell enhances this resistance and The strategies that we have described Work in the G.P.C.S. laboratory was supported by award BB/
W000105/1 from the Biotechnology and Biological Sciences
can ablate transfer of mobile genetic elements should be generally applicable to any gene Research Council, UK (UKRI). Author contributions: J.F.Z. and
with competing codes. or genetic system added to the synthetic or- J.W.C. conceptualized and planned the project, with input from
ganism. Because the canonical genetic code W.E.R.; J.F.Z. demonstrated unidirectional genetic isolation
Discussion is near universally conserved, we anticipate and loss of isolation through tRNA import. J.F.Z., T.S.E., G.P., and
W.E.R. designed and characterized tRNAs for genetic code
Previous work has shown that the choice of that the principles we have established may refactoring. J.F.Z. demonstrated bidirectional genetic isolation
synonymous codons in individual genes and be applied to a broad range of organisms. and (mutually) orthogonal HGT systems. J.F.Z. demonstrated
improved genetic isolation through code refactoring and
viruses can alter their robustness and evolv-
genetic code locking. J.F.Z. and T.K. performed phage
ability (33–35), but such approaches are lim- experiments under supervision of G.P.C.S. J.W.C. and J.F.Z. wrote
ited to exploring subsets of the canonical code. the manuscript, with input from all authors. J.W.C. supervised
1. F. H. Crick, L. Barnett, S. Brenner, R. J. Watts-Tobin, Nature the project. Competing interests: The MRC has filed a
Although a large body of theoretical work and
192, 1227–1232 (1961). provisional patent application related to this work on which J.F.Z.,
a limited number of in vitro experiments have 2. M. W. Nirenberg, J. H. Matthaei, Proc. Natl. Acad. Sci. U.S.A. W.E.R., and J.W.C. are listed as inventors. J.W.C. has a
considered the relationship between the struc- 47, 1588–1602 (1961). commercial interest in Constructive Bio. Data and materials

availability: The GenBank accession nos. for all the plasmids to original US government works. https://www.sciencemag.org/ References (40–44)
described in the text are provided in data file S5, and the authors about/science-licenses-journal-article-reuse Data Files S1 to S5
agree to provide any materials and strains used in this study MDAR Reproducibility Checklist
upon request. All data are available in the main text or SUPPLEMENTARY MATERIALS
supplementary materials. License information: Copyright © science.org/doi/10.1126/science.add8943 Submitted 12 July 2022; accepted 7 October 2022
2022 the authors, some rights reserved; exclusive licensee Materials and Methods Published online 20 October 2022
American Association for the Advancement of Science. No claim Figs. S1 to S24 10.1126/science.add8943
NEUROSCIENCE dsGFP expression dynamically follows epilep-

tiform network activity in vitro. Primary neu-
On-demand cell-autonomous gene therapy for ronal cultures were infected at 7 days in vitro
and grown either on multielectrode arrays
brain circuit disorders (MEAs) to record network activity or on cov-
erslips for imaging (Fig. 1, C to F). At 18 days
Yichen Qiu1, Nathanael O’Neill1, Benito Maffei1, Clara Zourray1,2, Amanda Almacellas-Barbanoj1, in vitro, we blocked g-aminobutyric acid type A
Jenna C. Carpenter1, Steffan P. Jones1, Marco Leite1, Thomas J. Turner1, Francisco C. Moreira1, (GABAA) receptors with picrotoxin (PTX) to
Albert Snowball1, Tawfeeq Shekh-Ahmad1†, Vincent Magloire1, Serena Barral2, Manju A. Kurian2,3, mimic an epileptic seizure and then recorded
Matthew C. Walker1, Stephanie Schorge4, Dimitri M. Kullmann1*, Gabriele Lignani1* network activity or, in parallel, fixed a subset
of coverslips, at 2, 6, 24, and 48 hours in the
Several neurodevelopmental and neuropsychiatric disorders are characterized by intermittent episodes of continued presence of PTX. MEA recordings
pathological activity. Although genetic therapies offer the ability to modulate neuronal excitability, a limiting revealed an increase in network activity that
factor is that they do not discriminate between neurons involved in circuit pathologies and “healthy” peaked 6 hours after addition of PTX and then
surrounding or intermingled neurons. We describe a gene therapy strategy that down-regulates the excitability returned to baseline after 48 hours (Fig. 1D
of overactive neurons in closed loop, which we tested in models of epilepsy. We used an immediate early and fig.S1). c-Fos expression and dsGFP fluo-
gene promoter to drive the expression of Kv1.1 potassium channels specifically in hyperactive neurons, rescence followed a similar time course (Fig. 1,
and only for as long as they exhibit abnormal activity. Neuronal excitability was reduced by seizure-related E and F, and fig. S2).
activity, leading to a persistent antiepileptic effect without interfering with normal behaviors. Activity- We next placed a transgene that decreases
dependent gene therapy is a promising on-demand cell-autonomous treatment for brain circuit disorders. neuronal and network activity under the same
cfos promoter. We used a codon-optimized
version of KCNA1, which encodes the potas-
any neurodevelopmental and neuro- In the case of epilepsy, subpopulations of neu- sium channel Kv1.1, with an Ile400Val mutation
M psychiatric circuit disorders are char- rons exhibit stereotypical patterns of patho- to bypass the need for posttranscriptional
acterized by intermittent episodes of logical discharges during seizures (19, 20). editing and facilitate recovery from inactivation
pathological activity (1–3). Pharmaco- Targeting these neurons specifically would be (15). We refer to this engineered potassium
therapy is focused on reducing the fre- an important step toward a rational treatment (K+) Channel as EKC. Kv1.1 overexpression
quency and/or severity of such episodes, but with minimal side effects. decreases both neuronal excitability and synaptic
often with limited success. For example, 30% of Activity-dependent promoters, typically of neurotransmitter release (26). Kv1.1 overexpres-
people with epilepsy are refractory to pharma- immediate early genes (IEGs), respond rapidly sion using KCNA1, EKC, or transcriptional up-
cological treatment, even with drugs that act to increases in neuronal activity (21–23) and regulation has also been used in experimental
in a use-dependent manner on their molecular have been used to identify hyperactive cells re- antiepileptic gene therapies targeting excit-
targets (4). Genetic therapies are promising cruited during seizures (24). Although seizures atory neurons (11, 12, 15, 27, 28). Because of
strategies to treat these pathologies because of often start and terminate abruptly, they can the innate propensity of primary neuronal
their ability to modulate neuronal excitability occur in clusters with intervals comparable with cultures in vitro to exhibit bursting activity as
in a region-specific and cell type–specific manner the kinetics of IEGs (25). We therefore asked they mature, we transduced neurons plated on
(5–9). However, current experimental genetic whether an activity-dependent promoter could MEAs at 7 days in vitro with either cfos-dsGFP
therapies do not discriminate between neu- be used to drive a therapeutic transgene to or AAV9 cfos-EKC (cfos-EKC) and recorded net-
rons involved in triggering circuit paroxysms reduce neuronal excitability in closed loop. work activity at 21 days in vitro (Fig. 2, A
and healthy surrounding or intermingled Epileptiform hyperactivity should reduce their and B). As hypothesized, cultures treated with
neurons (10–16). There is a need to develop likelihood to fire while leaving other neurons cfos-EKC showed less network activity than
methods that select and treat only those neu- unaffected (Fig. 1A). Once seizures resolve, the that of cultures treated with cfos-dsGFP (Fig.
rons involved in the generation of crises (17, 18). genetic therapy should switch itself off, unless 2C). Spike frequency, burst frequency, and
and until excessive activity occurs again (Fig. 1B). number of spikes per burst were all signifi-
cantly lower in cultures treated with cfos-EKC
1
Department of Clinical and Experimental Epilepsy, UCL Results (Fig. 2D and fig. S3). We also asked whether
Queen Square Institute of Neurology, University College Activity-dependent gene therapy follows
London, London, UK. 2Department of Developmental
an experimental perturbation aimed at tran-
neuronal dynamics, decreases network siently increasing activity is followed by a rapid
Neurosciences, Zayed Centre for Research Into Rare Disease
in Children, GOS–Institute of Child Health, University College hyperexcitability in vitro decrease in neuronal excitability, as would be
London, London, UK. 3Department of Neurology, Great
We used an adeno-associated virus (AAV) to expected from the activity-dependent promoter
Ormond Street Hospital for Children, London, UK.
4
Department of Neuroscience, Physiology and Pharmacology, express a destabilized version of the fluorescent driving EKC expression (fig. S5). We increased
University College London, London, UK. reporter green fluorescent protein (dsGFP) spontaneous activity by adding PTX for 30 min
*Corresponding author. Email: [email protected] (G.L.); under the promoter of an extensively charac- and then silenced network activity for 2 hours
[email protected] (D.M.K.)
†Present address: Department of Pharmaceutics, The Hebrew terized IEG, c-Fos (AAV9 cfos-dsGFP; short- with tetrodotoxin (TTX) before washing off
University of Jerusalem, Jerusalem, Israel. ened to cfos-dsGFP) (21–24), and verified that the blockers and recording neuronal firing. We

Fig. 1. Activity-dependent gene therapy paradigm. (A) Cartoon illustrating test versus 0 hours]. (E) dsGFP expression in primary neurons transduced
the activity-dependent strategy. (B) Schematic sequence of activity-dependent with cfos-dsGFP 0, 6, 24, and 48 hours after addition of PTX. Scale bars, 100 mm.
promoter and transgene expression. The star indicates the beginning of the (F) dsGFP-positive neurons expressed as a percentage of all 4′,6-diamidino-2-
pathological event. (C) Timeline of MEA recordings from primary cultures. phenylindole (DAPI)–positive neurons at different time points after PTX application
(D) Network mean firing rate after disinhibition with PTX application [*P < 0.05; (**P < 0.01; ***P < 0.001; one-way ANOVA followed by Bonferroni multiple-
one-way analysis of variance (ANOVA) followed by Bonferroni multiple-comparison comparison test versus 0 hours).
observed a net decrease in firing rate in dsGFP- promoter-transgene combinations were effec- The other AAV9 expressed the cfos promoter
positive neurons treated with cfos-EKC com- tive in reducing spiking and/or bursting, none driving the tetracycline trans-activator rtTa
pared with cfos-dsGFP (fig. S5). was as consistent across the different electro- linked to enhanced GFP (EGFP) with a T2A
We asked whether the results with cfos-EKC physiological measures as cfos-EKC (fig. S6). peptide, as well as a U6 promoter driving either
would generalize to other activity-dependent We sought to determine whether cfos-EKC an sgRNA targeting the endogenous Kcna1
promoters and transgenes (fig. S6). We com- dampens the effect of a proconvulsant manip- promoter or an sgRNA designed to target the
pared the promoters of several IEGs (cfos, arc, ulation in vitro. However, the lower baseline yeast LacZ gene as control (Fig. 2E) (14, 34).
and egr1) as well as synthetic activity–dependent activity in cfos-EKC–treated cultures (Fig. In this system, endogenous Kcna1 overexpres-
promoters [enhanced synaptic activity-responsive 2D) precluded a simple comparison with cfos- sion was designed to be switched on only when
element (ESARE) and NPAS4 robust activity dsGFP. Instead, we modified a previously de- all three of the following conditions are met:
marker (NRAM)], which have different proper- scribed method, CRISPR activation (CRISPRa), (i) neurons are cotransduced with both AAV9s,
ties (21, 29–31) (table S2), in combination with based on dual-AAV9 delivery of a single guide (ii) doxycycline is applied, and (iii) activity is suf-
three transgenes: the control reporter dsGFP; RNA (sgRNA) and a nuclease-defective “dead” ficient to activate the cfos promoter. Primary
EKC as above; or another potassium chan- Cas9 (dCas9) controlled by doxycycline (14, 34). cultures plated on MEAs were treated with
nel gene, KCNJ2, which encodes the muscle One AAV9 carried a TeT-ON promoter driv- both AAVs at 7 days in vitro. We then obtained
inward-rectifier Kir2.1 (32, 33). Although some ing dCas9 fused to a transcriptional activator. an initial (baseline) recording of network

Fig. 2. Activity-dependent gene therapy

decreases epileptiform activity in vitro.
(A) Timeline of MEA recordings in primary cultures.
DIV, days in vitro. (B) Primary cultures were
transduced with either cfos-dsGFP as control or
cfos-EKC. (C) MEA traces. (D) Effect of cfos-EKC
compared with cfos-dsGFP on spike frequency,
mean bursting rate, and average number of spikes
per burst. Student’s t test corrected for multiple
comparisons, a = 0.02. (E) Timeline of MEA
recordings in primary cultures transduced with dual
AAVs controlled by doxycycline: one expressing
an inducible dCAS9 fused to a transcriptional
activator, and the other carrying an sgRNA target-
ing either the Kcna1 promoter or a LacZ sequence.
(F) Spike frequency before PTX addition was
similar between neurons transduced with cfos-
CRISPRa_LacZ and cfos-CRISPRa_Kcna1. (G) Spike
frequency over time normalized to network
activity recorded at 14 days in vitro (*P < 0.05;
two-way ANOVA followed by Bonferroni multiple-
comparison test; interaction time × sgRNA,
P = 0.006). (F) and (G) use the same color scheme.
activity at 14 days in vitro without doxycy- doxycycline to activate CRISPRa and recorded later. This revealed a clear difference between
cline treatment. There was no difference in from the same cultures 4 days later. We again cultures treated with cfos-CRISPRa_Kcna1
network excitability between cultures trans- observed no significant difference in network and cfos-CRISPRa_LacZ. Indeed, PTX failed to
duced with either cfos-CRISPRa_Kcna1 or cfos- activity. Last, we added PTX and recorded increase network activity in cultures treated
CRISPRa_LacZ (Fig. 2F). We then added from the same cultures 2, 6, 24, and 48 hours with cfos-CRISPRa_Kcna1 (Fig. 2G and fig. S7).

Fig. 3. Activity-dependent gene therapy is effective in decreasing excitatory neurons positive or negative for dsGFP, of excitatory neurons positive or
neuronal excitability and in protecting against sequential chemoconvulsant negative for dsGFP, and of dsGFP positive neurons identified as excitatory or
challenges. (A) Illustrative timeframe for electrophysiology and immuno- inhibitory (n = 3 animals) (***P < 0.001 two-way ANOVA followed by Bonferroni
fluorescence analysis. (B) Evoked trains of action potentials recorded in neurons multiple-comparison test). Scale bar, 100 mm. (E) Illustrative timeline for
transduced with either cfos-dsGFP or cfos-EKC. (C) Neuronal excitability repeated PTZ injections. OFF, basal cfos-driven transgene expression; ON,
parameters of neurons expressing either cfos-dsGFP or cfos-EKC after a seizure-induced cfos-driven expression. (F) Raw Racine scale data in animals
single generalized seizure. (Left) Number of action potentials evoked by transduced with either cfos-dsGFP or cfos-KCNA1 and then injected with
increasing current steps. (Middle) Maximum number of evoked action potential. PTZ at three time points (0 hours, 24 hours, and 2 weeks). (G) Averaged Racine
(Right) Current and voltage thresholds. Input/output, two-way ANOVA; other scores of data shown in (F). Racine score of >4 was considered as a generalized
graphs, Student’s t test corrected for multiple comparisons, a = 0.02. seizure. Two-way ANOVA followed by Bonferroni multiple-comparison test.
(D) Immunofluorescence images and analysis of the percentage of inhibitory (H) Percentage of mice experiencing generalized seizures from (F) (c2 test).

Activity-dependent gene therapy decreases

neuronal excitability in a transient,
cell-autonomous, on-demand fashion
We next asked whether the strategy is effective
in vivo. In the first set of experiments, we asked
how the cfos promoter behaves in response
to chemoconvulsant-evoked seizures. Mice
were injected with either cfos-dsGFP or AAV9
CaMKII-dsGFP (shortened to CaMKII-dsGFP)
in the visual cortex. The CaMKII promoter was
chosen as a control because it has previously
been used to bias expression to forebrain prin-
cipal cells in constitutive antiepileptic gene
therapies (10, 15). Two weeks later, we injected
either pilocarpine into the same brain region
to induce a focal seizure (35) or saline as a
control (fig. S8) and sacrificed the mice 2 hours
later. Animals treated with CaMKII-dsGFP
showed widespread expression of GFP whether
saline or pilocarpine was injected. By con-
trast, mice treated with cfos-dsGFP showed
focal expression of GFP only after pilocarpine
injection (fig. S8).
To test whether activity-dependent expres-
sion of EKC driven by the cfos promoter can
affect neuronal excitability, we injected either
cfos-dsGFP or cfos-EKC in the hippocampus
of adult mice. Two weeks later, we elicited a
single generalized seizure with an intraperi-
toneal pentylenetetrazole (PTZ) injection. Mice
were monitored to verify that they reached
stage 5 on a revised Racine scale (equivalent
to a generalized tonic-clonic seizure) (36) and
were sacrificed after 2 hours to prepare acute
hippocampal slices for electrophysiological
recordings from dsGFP-positive CA1 pyram-
idal neurons (Fig. 3A). dsGFP-positive neurons
treated with cfos-EKC exhibited a profound
decrease in excitability when compared with
cfos-dsGFP treated neurons (Fig. 3, B and C),
with robust decreases in the maximal firing
frequency and in the slope that relates action
potential frequency to depolarizing current
injection. There was also an increase in current
threshold for triggering the first action poten-
tial and an increase in the size of the after-
hyperpolarization, but other parameters were
not affected (Fig. 3C and fig. S9). The effect
of activity-dependent EKC expression was
Fig. 4. Activity-dependent gene therapy does not affect normal behavior. (A) Schematic of the CFC test.
more pronounced than previously observed
(B) Percentage of freezing time for naïve or PTZ-treated mice injected with either cfos-dsGFP (n = 7 and 8,
with constitutive expression (11, 14). We were
respectively) or cfos-EKC (n = 7 and 9, respectively) during fear conditioning, fear recall, and in a new
unable to measure the effect of cfos-EKC in
context. Two-way ANOVA followed by Bonferroni multiple-comparison test. (C) Open field test. Shown is
the absence of seizures because we could not
thigmotaxis (fraction of time spent in the periphery) and distance traveled before and after either cfos-dsGFP or
identify the transduced neurons without acti-
cfos-EKC injection in both hippocampi. (D) Spontaneous T-maze alternation test. Shown are the
vation of the fluorescent reporter.
spontaneous alternation rates before and after either cfos-dsGFP or cfos-EKC injection.
We also tested other activity-dependent
promoter-transgene combinations. These re-
vealed a consistent effect of EKC overexpres- changes in firing (figs. S10 and S11 and sup- different IEG promoters driving dsGFP in
sion on action potential firing, independent plementary text). the hippocampus (Fig. 3D and figs. S12 to S17).
of the promoter (figs. S10 and S11). When the To better understand which neurons were dsGFP driven by cfos was mainly expressed in
KCNJ2 transgene was used, however, neurons activated after a single generalized PTZ-evoked excitatory neurons, with a very low percent-
exhibited a hyperpolarized resting membrane seizure, we performed immunohistochemical age of inhibitory neurons or of parvalbumin-
potential, as expected from the intrinsic prop- analysis in different hippocampal regions positive neurons expressing the reporter (Fig.
erties of this channel (32), but inconsistent (CA1/2, CA3, and hilus) in animals treated with 3D and fig. S15). Although ESARE-driven dsGFP

Fig. 5. Activity-dependent gene therapy decreases spontaneous generalized either cfos-dsGFP or cfos-KCNA1. Spike rates were normalized to the
seizures and interictal spikes and protects against a subsequent maximum for each animal (yellow, maximum; blue, minimum spike rate).
chemoconvulsant challenge. (A) Schematic of the preclinical trial. (B) Seizures (Middle) Spike frequency normalized to baseline (before viral injection)
(vertical bars) over time for each mouse. Gray boxes indicate the viral (Student’s t test). (Right) Weighted cumulative plot normalized by the
injection, either with cfos-dsGFP or cfos-EKC, followed by a 2-week period total interictal spike count before viral treatment (two-way ANOVA). (F) Percentage
to allow viral expression. (C) Percentage change in spontaneous generalized change in coastline normalized to baseline (before viral injection). Student’s
seizures after cfos-EKC treatment compared with baseline, normalized to the t test. (G) Percentage power change versus baseline in different frequency
same percentage of change in cfos-dsGFP treated mice. One Sample t test bands (two-way ANOVA). (H) Chronic epileptic animals or naïve animals
versus 0 hours. (D) Weighted cumulative plot normalized by the total seizure (injected with cfos-dsGFP) received an intraperitoneal PTZ injection at
count before treatment. Two-way ANOVA. (E) (Left) Number of spikes per the end of the study. Kaplan-Meier plot showing survival rate. Log-rank
hour plotted against time. Red dotted lines indicate the viral injection, with (Mantel-Cox) test.

Fig. 6. Activity-dependent gene therapy decreases epileptiform activity were fixed and stained for dsGFP, and the density of dsGFP-positive neurons
in hCAs. (A) Schematic representation for the generation of human forebrain was plotted. Scale bars, (I) and (J), 100 mm; (K) 10 mm. Student’s t test.
assembloids after fusion of hCSs and hSSs. (B) Inhibitory neurons migrate (L) Schematic representation for a test of activity-dependent gene therapy
from hSS to hCS, recapitulating human cortical development. Scale bar, 100 mm. in hCAs. (M) Local field potential showing a decrease in activity after addition
(C and D) Trains of evoked action potentials and spontaneous inhibitory of KCl (orange line) in hCAs transduced with cfos-EKC compared with cfos-
postsynaptic events recorded in excitatory neurons in cortical assembloids dsGFP. For the purpose of illustration, the voltage range illustrated is ± 0.5 mV.
at the latest stage of differentiation (d240). (E) Cortical assembloids were (N) (Top) Baseline LFP amplitude spectra for hCAs treated with cfos-dsGFP
maintained either in control medium or in a medium with 4AP and PTX. (F) Local or cfos-EKC. (Bottom) Amplitude changes across frequencies after KCl.
field potential showing an increase in activity after addition of 4AP and PTX (O) Percentage change in the median standard deviation (SD), normalized to
(orange triangle). (G) Immunofluorescence images of cfos obtained 4 hours after baseline activity, as a proxy for network activity. Student’s t test. (P) Time
control medium (left) or medium supplemented with 4AP and PTX (right). course of SD during the experiment. Two-way ANOVA. (Q) Change in SD in
Scale bar, 100 mm. (H) Cortical assembloids were transduced with cfos-dsGFP hCA transduced with either cfos-dsGFP or cfos-EKC divided into two periods:
and then, 12 days later, moved to either control medium or medium <2000 s before cfos activation and >2000 s after cfos activation (two-way
supplemented with 4AP and PTX. (I to K) After 4 hours, cortical assembloids ANOVA followed by Bonferroni multiple comparison test).

expression was mainly in excitatory neurons, 2 weeks later (Fig. 4A). We tested this paradigm spontaneous seizures and interictal spike ac-
dsGFP driven by a minimal arc promoter (mArc) both in naïve animals and in animals that had tivity, suggesting a broader impact of the treat-
showed a significantly higher percentage of ex- received a single PTZ injection 24 hours before ment than only raising the seizure threshold.
pression in inhibitory neurons (≈12%) (figs. S14, fear conditioning (leading to a stage 5 seizure Indeed, treatment efficacy was reflected in sig-
S16, and S17). With cfos-dsGFP, there was a trend in all such animals). In the naïve animals, no nificant decreases in ECoG coastline (a measure
for a lower percentage of excitatory neurons in difference in freezing behavior was observed of overall activity) and in total power, especially
CA1/2 to express dsGFP compared with CA3 between cfos-dsGFP– and cfos-EKC–treated at higher frequencies (Fig. 5, F and G).
and hilus (Fig. 3D). groups, either during fear conditioning or We also asked whether the activity-dependent
We hypothesized that expression of EKC 24 hours later during fear recall, or when ani- transgene expression was maintained in the
in response to an evoked generalized seizure mals were exposed to a new context (fig. S18). absence of overt seizures. cfos-dsGFP was in-
should attenuate the effect of a second chemo- In animals that had received a PTZ injection, jected bilaterally in the dentate gyrus. Two
convulsant challenge if delivered during the we observed a nonsignificant trend for greater weeks later, KA was injected unilaterally, also
time that potassium channel overexpression freezing during fear recall in the cfos-EKC in the dentate gyrus. After 5 days, at which
persists, but that this protective effect should group than in the cfos-dsGFP group (Fig. 4B). stage interictal spikes occur without seizures
fade with time. We therefore performed three When animals were tested in several other (and dsGFP induced by KA has been degraded),
consecutive intraperitoneal PTZ injections behavioral tasks, cfos-EKC–expressing ani- we sliced the brain and observed ipsilateral
in the same animals (Fig. 3E). We first ex- mals again performed similarly to cfos-dsGFP– cfos-dsGFP expression (fig. S21). This finding
pressed either cfos-dsGFP or cfos-EKC in both expressing animals: No differences were suggests that interictal activity is sufficient for
hippocampi and, after 2 weeks, gave the first observed before or after viral injection (Fig. 4, activation of the treatment.
PTZ injection. This injection triggered a sin- C and D, and fig. S19) in an open field test to At the end of the chronic epilepsy study, we
gle generalized seizure with no difference monitor anxiety and locomotor activity (Fig. asked whether the treatment protected animals
in severity between the two groups (Fig. 3, E 4C), a T-maze test to assess working memory from a lethal dose of PTZ (Fig. 5H). We observed
to H), which was as expected because the cfos (Fig. 4D), or an olfactory discrimination test a significant survival advantage in animals in-
promoter is normally inactive under baseline (fig. S19). jected with cfos-EKC compared with cfos-
condition. We then administered a second PTZ dsGFP, with a survival rate close to that of
injection 24 hours later. This led to markedly Activity-dependent gene therapy suppresses cfos-dsGFP–transduced naïve animals injected
attenuated seizures in the animals injected spontaneous seizures and interictal activity with the same PTZ dose (85%) (Fig. 5H).
with cfos-EKC compared with those injected in a chronic model of intractable epilepsy We also tested the effect of cfos-KCNJ2 on
with cfos-dsGFP, with the majority of animals Although treatment with cfos-EKC is effec- spontaneous seizures in the same chronic
failing to reach a Racine score of 4 (Fig. 3, E tive in protecting against a second chemo- epilepsy model (fig. S22). In contrast to cfos-
to H). Last, to test whether this effect persists, convulsant challenge delivered shortly after EKC, no impact on seizure frequency was
we gave a third PTZ injection after 2 weeks, by a first, it remained to be determined whether observed in animals injected with cfos-KCNJ2
which time we anticipated that overexpression it could modify the disease course in epilepsy, when compared with cfos-dsGFP (fig. S22),
of Kv1.1 channels had returned to baseline, as in which seizures occur sporadically. We there- implying that transient Kir2.1 expression does
expected from their membrane persistence fore switched to a model of chronic limbic not have an antiepileptic action.
(37). At this time point, no difference was ob- epilepsy. We used a clinically relevant model Last, we asked whether treatment with cfos-
served between the two groups (Fig. 3, E to of drug-resistant epilepsy that consists of an EKC affected performance in behavioral tests
H). Activity-dependent induction of EKC ex- intra-amygdala kainic acid (KA) injection to that are affected by chronic epilepsy. Having
pression thus protects against generalized induce a period of status epilepticus (14, 42, 43). quantified several behavioral defects in chronic
convulsions, which are associated clinically Mice started to exhibit spontaneous seizures epileptic animals compared with naïve ani-
with mortality and morbidity, in a time-limited after 2 weeks and were implanted with wire- mals, we injected either cfos-dsGFP or cfos-
manner. less electrocorticogram (ECoG) transmitters EKC in epileptic animals and waited 4 weeks
(Fig. 5A). We recorded spontaneous general- to measure the same behavioral parameters
Activity-dependent gene therapy does not affect ized seizures for 2 weeks and then randomized (figs. S23 to S25). We observed no worsening
normal behavior animals for injection with either cfos-dsGFP in animals treated with cfos-EKC compared
A potential limitation of cfos-EKC treatment or cfos-EKC in both hippocampi. Two weeks with cfos-GFP in either T-maze, open field,
is that it might interfere with normal brain later, after waiting for viral expression, the or olfactory discrimination tests (figs. S23
function. Indeed, the cfos promoter is recruited ECoG was recorded for 2 more weeks (Fig. 5A). to S25).
during physiological behaviors such as mem- cfos-EKC–treated animals showed a robust
ory formation and learning (38). We therefore decrease in the number of spontaneous sei- Activity-dependent gene therapy suppresses
asked whether salient sensory and aversive zures when compared with animals receiving epileptiform activity in human neurons
stimuli could lead to the overexpression of cfos-dsGFP (Fig. 5, B to D). We asked whether cfos can be activated by
Kv1.1 sufficient to interfere with normal be- We also analyzed interictal spike activity in epileptiform activity in human neurons. We
havior. Cfos activation has previously been the same animals (Fig. 5E). We observed a net used an induced pluripotent stem cell line to
harnessed to manipulate a memory trace or decrease in spike activity in animals treated derive human cortical assembloids (hCAs) by
engram, with behavioral readout in a con- with cfos-EKC compared with cfos-dsGFP fusing cortical spheroids (hCSs) with subpallial
textual fear conditioning (CFC) task: Chemo- (Fig. 5E), with a trend to more consistently spheroids (hSSs) (Fig. 6A) (44–47). Immuno-
genetic or optogenetic modulation of neurons follow a circadian rhythm (fig. S20). Despite a fluorescence analysis at different time points
“tagged” in a cfos-dependent manner during slight decrease in seizure frequency with time showed expression of hCS and hSS region-
this test can alter fear recall 24 hours after in cfos-dsGFP–treated animals, interictal spike specific transcription factors [forkhead box
memory consolidation (39–41). We repeated frequency increased, suggesting that this phe- G1 (FOXG1), paired box 6 (PAX6), and NK2
this behavioral paradigm after bilateral intra- nomenon is a natural evolution of this chronic homeobox 1 (NKX2.1)] and mature neuronal
hippocampal expression of either cfos-dsGFP model. By contrast, animals treated with cfos- markers [microtubule associated protein 2
or cfos-EKC and performed the CFC test EKC showed significant decreases in both (MAP2), NeuN, and GABA)] (fig. S26). We

demonstrated successful fusion of hCSs and iors and as a test of causality in CFC (39–41). other neuropsychiatric diseases are character-
hSSs, with inhibitory neurons migrating to, The combination of cfos with EKC proved ized by circuit hyperactivity, including early
and integrating with, the cortical side (Fig. highly effective in down-regulating neuronal stages of schizophrenia [in which anterior hip-
6B). We then performed electrophysiological excitability after chemoconvulsants and in pocampal hyperactivity has been implicated
recordings in neurons located in the hCS area suppressing spontaneous seizures, without (50)], Parkinson’s disease [subthalamic nucleus
of acute hCA slices and showed that neurons deleterious effects in a battery of behavioral (51)], migraine and cluster headache [posterior
can fire trains of action potentials and exhibit tests. We observed less consistent effects on hypothalamus (52)], and obsessive-compulsive
spontaneous PTX-sensitive inhibitory post- neuronal excitability with other promoters. disorder [anterior cingulate cortex (53)]. Early
synaptic currents, confirming that interneu- Why other IEG promoters, and synthetic pro- hyperactivity has also been documented in
rons integrate in the network (Fig. 6, C and D). moters, were less effective is unclear. Among mouse models of Alzheimer’s disease and is
We treated mature hCAs with 4-aminopyridine factors determining their potential utility are associated with high levels of arc expression
(4AP) and PTX to increase network excitability the different time courses of activation and (54). In this case, arc or ESARE could represent
and recorded epileptiform activity using local deactivation, and the triggering mechanisms appropriate therapeutic activity-dependent
field potentials (Fig. 6, E and F). Four hours for each promoter. For example, mArc activa- promoters.
later, we observed cfos-positive neurons that tion has been linked to synaptic transmission
were not present when hCAs were treated with and plasticity processes rather than simply
a control solution (Fig. 6G). Last, we transduced responding to increases in neuronal activity,
1. P. Fusar-Poli et al., JAMA Psychiatry 70, 107–120
hCAs with AAV9 cfos-dsGFP and after 12 days as is the case for cfos, although the interme- (2013).
incubated them in either a control solution diate intracellular signaling cascades overlap 2. S. Harvey, M. D. King, K. M. Gorman, Front. Neurol. 12, 659064
or 4AP and PTX (Fig. 6H). No dsGFP-positive (21). We also found that KCNJ2 could not (2021).
3. K. Staley, Nat. Neurosci. 18, 367–372 (2015).
neurons were present in the control condition be substituted for EKC, even though it pro- 4. M. J. Brodie, CNS Drugs 31, 527–534 (2017).
(Fig. 6I). By contrast, we observed dsGFP- foundly hyperpolarized neurons when driven 5. D. M. Kullmann, S. Schorge, M. C. Walker, R. C. Wykes,
positive neurons in hCAs treated with 4AP in an activity-dependent manner (fig. S4). If Nat. Rev. Neurol. 10, 300–304 (2014).
6. T. J. Turner, C. Zourray, S. Schorge, G. Lignani, J. Neurochem.
and PTX, suggesting that this minimal pro- anything, animals treated with AAV9 cfos- 157, 229–262 (2021).
moter is also activated in human neurons after KCNJ2 had a worse outcome in the chronic 7. K. M. Tye, Neuron 83, 1259–1261 (2014).
epileptiform activity (Fig. 6, J and K). Last, we epilepsy study when compared with animals 8. R. C. Wykes, G. Lignani, Neuropharmacology 132, 108–117
(2018).
tested the ability of cfos-EKC to suppress injected with cfos-dsGFP (fig. S10). A possible
9. B. E. Deverman, B. M. Ravina, K. S. Bankiewicz, S. M. Paul,
epileptiform activity in hCAs. We transduced explanation for the efficacy of Kv1.1 overex- D. W. Y. Sah, Nat. Rev. Drug Discov. 17, 641–659
hCAs with a lentivirus expressing either cfos- pression is that it reduces not only excitability (2018).
dsGFP or cfos-EKC and after 10 days treated but also neurotransmitter release through a 10. A. Lieb et al., Nat. Med. 24, 1324–1329 (2018).
11. R. C. Wykes et al., Sci. Transl. Med. 4, 161ra152 (2012).
them for 1 hour and 45 min with 55 mM KCl modulatory effect on the presynaptic action po- 12. D. Kätzel, E. Nicholson, S. Schorge, M. C. Walker,
(fig. S27). Four hours later, we recorded base- tential waveform (supplementary text) (26, 48). D. M. Kullmann, Nat. Commun. 5, 3847 (2014).
line local field potential activity and then added It remains to be determined whether the 13. F. M. Noe, A. T. Sorensen, M. Kokaia, A. Vezzani, in Jasper’s
Basic Mechanisms of the Epilepsies, J. L. Noebels, M. Avoli,
55 mM KCl a second time (Fig. 6L). We ob- powerful antiepileptic effect of cfos-EKC de- M. A. Rogawski, R. W. Olsen, A. V. Delgado-Escueta, Eds.
served a clear reduction in overall network pends on suppressing the excitability of a small (Oxford Univ. Press, 2012).
activity in hCAs transduced with cfos-EKC number of “detonator” or “hub” neurons re- 14. G. Colasante et al., Brain 143, 891–905 (2020).
15. A. Snowball et al., J. Neurosci. 39, 3159–3169
compared with cfos-dsGFP (Fig. 6, M to Q). sponsible for triggering seizures, or of a network (2019).
of neurons whose activity is prone to reaching 16. A. Nanobashvili et al., Gene Ther. 26, 65–74 (2019).
Discussion a critical point (20, 49). The data are therefore 17. P. Kwan, S. C. Schachter, M. J. Brodie, N. Engl. J. Med. 365,
919–926 (2011).
We have reported an activity-dependent, cell- silent on whether successive seizures arise from 18. F. Tang, A. M. S. Hartz, B. Bauer, Front. Neurol. 8, 301
autonomous, on-demand gene therapy that a stereotypical or variable pattern of neuronal (2017).
follows network dynamics, driving the expres- recruitment in an ictogenic network. 19. W. Truccolo et al., Nat. Neurosci. 14, 635–641 (2011).
20. M. Wenzel, J. P. Hamm, D. S. Peterka, R. Yuste, Cell Rep. 19,
sion of a transgene that in turn decreases Because patients with focal epilepsy often ex-
2681–2693 (2017).
neuronal excitability (Fig. 1, A and B). This perience two or more seizures within 24 hours 21. S. W. Flavell, M. E. Greenberg, Annu. Rev. Neurosci. 31,
approach is specific for neurons that partic- (25), cfos-EKC would be expected to abort 563–590 (2008).
ipate in pathological network activity but is such clusters. However, we observed the per- 22. E. L. Yap, M. E. Greenberg, Neuron 100, 330–348 (2018).
23. T. Kawashima, H. Okuno, H. Bito, Front. Neural Circuits 8, 37
also time-limited in that transgene expres- sistence of the antiepileptic treatment in the (2014).
sion persists only for as long as neurons are chronic model despite evidence from the PTZ 24. N. Dabrowska et al., Brain 142, 2336–2351 (2019).
hyperactive (Fig. 3, E to H). In a chronic model injections that the effect of a single evoked 25. S. R. Haut, R. B. Lipton, A. J. LeValley, C. B. Hall, S. Shinnar,
Neurology 65, 1313–1315 (2005).
of epilepsy, we achieved a greater decrease in seizure faded by 2 weeks (Figs. 3 and 5). A 26. J. H. Heeroma et al., Dis. Model. Mech. 2, 612–619
seizure frequency than previously reported for possible explanation is that in the chronic (2009).
gene therapy with constitutive promoters or epilepsy model, interictal discharges repeat- 27. A. L. Lee et al., J. Neurochem. 86, 1079–1088 (2003).
28. H. J. Wenzel et al., Epilepsia 48, 2023–2046 (2007).
with widely used antiseizure drugs (11–13, 43), edly reactivated the cfos promoter, so that the 29. T. Kawashima et al., Nat. Methods 10, 889–895
and without deleterious effects on normal be- epileptiform activity remained below thresh- (2013).
haviors (Figs. 4 and 5). Given that the components old for generalization. This is supported by the 30. A. T. Sørensen et al., eLife 5, e13918 (2016).
31. X. Sun et al., Cell 181, 410–423.e17 (2020).
of activity-dependent gene therapy are normal findings that interictal activity is sufficient to 32. N. A. Hager, C. K. McAtee, M. A. Lesko, A. F. O’Donnell,
mammalian genetic elements packaged in a activate cfos-driven dsGFP expression (fig. S21) Front. Cell Dev. Biol. 9, 796136 (2022).
well-tolerated viral vector that is already in and that treated animals were relatively pro- 33. C. R. Yu et al., Neuron 42, 553–566 (2004).
34. G. Colasante et al., Mol. Ther. 28, 235–253 (2020).
the clinic, there is a relatively straightforward tected from the effects of a chemoconvulsant
35. V. Magloire, J. Cornford, A. Lieb, D. M. Kullmann, I. Pavlov, Nat.
path to first-in-human studies. The translational challenge despite a prolonged absence of overt Commun. 10, 1225 (2019).
potential is underlined by the preliminary evi- seizures (Fig. 5H). 36. J. Van Erum, D. Van Dam, P. P. De Deyn, Epilepsy Behav. 95,
dence of effectiveness in hCAs (Fig. 6). A cell-autonomous self-regulated tool that 51–55 (2019).
37. E. F. Fornasiero et al., Nat. Commun. 9, 4230 (2018).
Cfos tagging has been used to track the re- normalizes network dynamics has potential 38. F. T. Gallo, C. Katche, J. F. Morici, J. H. Medina,
cruitment of neurons in pathological behav- clinical applications beyond epilepsy. Several N. V. Weisstaub, Front. Behav. Neurosci. 12, 79 (2018).

39. K. Minatohara, M. Akiyoshi, H. Okuno, Front. Mol. Neurosci. 8, ARCTIC WILDFIRES

78 (2016).
40. K. Z. Tanaka et al., Neuron 84, 347–354 (2014).
41. A. R. Garner et al., Science 335, 1513–1516 (2012). Unprecedented fire activity above the Arctic Circle
42. E. M. Jimenez-Mateos et al., Nat. Med. 18, 1087–1094
(2012).
43. P. J. West et al., Exp. Neurol. 349, 113954 (2022).
linked to rising temperatures
44. S. P. Paşca, Nat. Neurosci. (2018).
45. S. A. Sloan, J. Andersen, A. M. Pașca, F. Birey, S. P. Pașca, Adrià Descals1,2*, David L. A. Gaveau3, Aleixandre Verger1,2,4, Douglas Sheil5,6,
Nat. Protoc. 13, 2062–2085 (2018). Daisuke Naito6,7, Josep Peñuelas1,2
46. C. Zourray, M. A. Kurian, S. Barral, G. Lignani, Front. Mol.
Neurosci. 15, 839366 (2022).
47. F. Birey et al., Nature 545, 54–59 (2017). Arctic fires can release large amounts of carbon from permafrost peatlands. Satellite observations reveal
48. U. Vivekananda et al., Proc. Natl. Acad. Sci. U.S.A. 114, that fires burned ~4.7 million hectares in 2019 and 2020, accounting for 44% of the total burned area
2395–2400 (2017).
49. L. A. Lau, K. J. Staley, K. P. Lillis, Brain 145, 531–541
in the Siberian Arctic for the entire 1982–2020 period. The summer of 2020 was the warmest in
(2022). four decades, with fires burning an unprecedentedly large area of carbon-rich soils. We show that factors
50. M. McHugo et al., Am. J. Psychiatry 176, 1030–1038 of fire associated with temperature have increased in recent decades and identified a near-exponential
(2019).
relationship between these factors and annual burned area. Large fires in the Arctic are likely to recur
51. M. Jahanshahi, I. Obeso, C. Baunez, M. Alegre, P. Krack,
Mov. Disord. 30, 128–140 (2015). with climatic warming before mid-century, because the temperature trend is reaching a threshold in which
52. P. J. Goadsby et al., Physiol. Rev. 97, 553–622 (2017). small increases in temperature are associated with exponential increases in the area burned.
53. K. D. Fitzgerald et al., Biol. Psychiatry 57, 287–294
(2005).
54. M. Przybyla et al., Brain 143, 1889–1904 (2020).
missions from Arctic wildfires jeopar- 2020 in the Siberian Arctic have provided new
ACKN OW LEDG MEN TS
We thank I. Zalivina and L. Ussingkaer for the help in preparing
AAVs; M. Weston for help in behavioral assessment; I. Colombi
and M. Chiappalone for the MEA analysis software; and T. Smart,
M. Hausser, and B. Clark for facilitating the CFC experiments.
We are thankful to all members of the Experimental Epilepsy Group
at UCL and C. Bernard for helpful discussions. The figures were
created with Biorender.com under a license granted to G.L.
E dize global climate goals (1). The Arctic
is warming rapidly because of a climate
change–related phenomenon known as
“Arctic amplification” (2); annual mean
temperature has already increased more than
2°C compared with that of the preindustrial
empirical observations between climatic fac-
tors and burn extent and may already be indi-
cating the changes in fire regimes expected by
the end of the century. The fire seasons of 2019
and 2020, however, raised two uncertainties—
first, whether the annual burned area above the
era (3) and is expected to reach 3.3° to 10°C Arctic Circle was actually increasing. Satellite-
Funding: This work was supported by Epilepsy Research UK
Emerging Leader Fellowship F1701 (G.L.), Medical Research Council above the 1985–2014 average by 2100 (4). These derived burned-area products tend to under-
New Investigator Project Grant MR/S011005/1 (G.L.), Medical increased temperatures result in thawing of estimate the true extent of burning (12), and
Research Council PhD studentship for Y.Q. (D.M.K. and G.L.), UCL permafrost and deterioration of peatlands with rigorous validation techniques are required
Technology Fund UTF-20-005 (G.L. and D.M.K.), Sir Jules Thorn
Trust Studentship for C.Z. (G.L., S.B., M.A.K., and S.S.), emissions of carbon dioxide and methane (5–7). (16). Second, even if the burned areas in 2019
Wellcome Trust Investigator Award 212285/Z/18/Z (D.M.K.), High-latitude peatlands are expected to be- and 2020 were the largest yet observed, the
Sir Jules Thorn Trust Award for Biomedical Research (M.A.K.), come a net carbon source as a consequence of links to other trends required evaluation.
NIHR Research Professorship NIHR-RP-2016-07-019) (M.A.K.), and
NIHR Great Ormond Street Hospital Biomedical Research 17BR09 global warming (8). The release of carbon cre- We assessed annual burned area in the
(S.B.). Author contributions: Conceptualization: G.L., D.M.K., ates positive feedback with additional emis- Siberian Arctic (latitudes >66.5°N) for 1982–
and S.S. Funding acquisition: G.L., D.M.K., S.S., M.A.K., and S.B. sions contributing to further warming and 2020 using six satellite-derived maps of burned
Methodology: Y.Q., N.O., B.M., C.Z., A.A.-B., J.C.C., S.P.J., F.C.M.,
M.L., A.S., T.S.-A., V.M., S.S., G.L., and D.M.K. Investigation: Y.Q.,
thawing with further peatland degradation and areas (fig. S1). We investigated the Siberian
N.O., B.M., C.Z., A.A.-B., J.C.C., S.P.J., M.L., T.J.T., A.S., and emissions. In this context, the numerous fires Arctic because it is where most burning occurs
T.S.-A. Formal analysis: Y.Q., N.O., B.M., C.Z., A.A.-B., S.P.J., F.C.M., identified by satellite thermal sensors in east- above the Arctic Circle and fire frequency ap-
M.L., and G.L. Supervision: G.L., D.M.K., S.S., M.C.W., M.A.K.,
S.B., V.M., and J.C.C. Writing – original draft: G.L. and D.M.K.
ern Siberia in 2020 (9) raise particular concerns peared to be increasing (9). We investigated
Writing – review and editing: G.L. and D.M.K. Competing because of the resulting emissions (10). 10 factors associated with the likelihood of
interests: G.L., D.M.K., S.S., Y.Q., and M.C.W. are listed as Wildfires are common in the Arctic and Sub- fire: six climatic variables [air and surface tem-
inventors on patent WO2021191474A1. G.L., D.M.K., S.S.,
arctic (11), but their size, frequency, and inten- perature, total precipitation, wind speed and
and M.C.W. have equity in a company that aims to bring epilepsy
gene therapy to the clinic. Data and materials availability: All sity are expected to increase as the climate direction, and vapor-pressure deficit (VPD)],
Python and Matlab scripts are available at https://github.com/ warms (12). Extreme weather, such as that in three variables describing the vegetation condi-
KullmannLab/Qiu_et_al_2022. PyEcog software is available here: 2020 in the Siberian Arctic (13), is expected tions [length of the growing season, mean nor-
https://www.pyecog.com. All the vectors used in this manuscript
will be made available on a suitable platform. All data are stored in to become more severe as Arctic oscillations malized difference vegetation index (NDVImean),
the UCL cloud database. All raw data supporting the main and weaken over time (14). Previous research in and climatic water deficit (CWD)], and the
supplementary figures are available here: https://doi.org/10.5522/ the Alaskan tundra suggests that the annual number of ignitions, a direct factor associated
04/20867117. License information: Copyright © 2022 the
authors, some rights reserved; exclusive licensee American burned area might be two times greater than with the likelihood of fires. We evaluated how
Association for the Advancement of Science. No claim to original in the 1950–2010 period by the end of the cen- these factors have varied over the past four
US government works. https://www.science.org/about/science- tury as warmer and drier conditions coin- decades and their relationships with satellite-
licenses-journal-article-reuse
cide more frequently (15). The conditions that derived estimates of annual burned areas. Lastly,
affected the Arctic fire seasons of 2019 and we investigated the future trends of annual
SUPPLEMENTARY MATERIALS burned area and fire emissions under future
1
science.org/doi/10.1126/science.abq6656 CREAF, Centre de Recerca Ecològica i Aplicacions Forestals,
Cerdanyola del Vallès, 08193 Barcelona, Catalonia, Spain.
Representative Concentration Pathways (RCPs).
Materials and Methods
2
Supplementary Text CSIC, Global Ecology Unit CREAF-CSIC-UAB, Bellaterra,
Figs. S1 to S27 08193 Barcelona, Catalonia, Spain. 3TheTreeMap; Bagadou Results
Tables S1 and S2 Bas, 46600 Martel, France. 4CIDE, CSIC-UV-GV, 46113 Trends of burned area for 1982–2020
References (55–77) València, Spain. 5Forest Ecology and Forest Management
MDAR Reproducibility Checklist Group, Wageningen University and Research, Wageningen, Between 1982 and 2020, the satellite burned-
Netherlands. 6Center for International Forestry Research area products indicate that 12.97 million hec-
(CIFOR), Bogor 16000, Indonesia. 7Graduate School of tares (Mha) burned in the circumpolar region
Agriculture, Kyoto University, Kitashirakawa Oiwake-cho,
Submitted 27 April 2022; accepted 14 September 2022 Sakyo-ku, Kyoto 606-8502, Japan. (latitudes >66.5°N). The Siberian Arctic, a re-
10.1126/science.abq6656 *Corresponding author. Email: [email protected] gion with continuous permafrost, accounted

A B
Peatland carbon storage (kg C m -2 )

Arctic Circle Burned 2001-2018
Siberian Arctic Burned 2019-2020 0 >50
Fig. 1. Maps of burned area for 2001–2020 and peatland carbon storage in 2019 and 2020. Areas that burned at least once in both periods, in 2001–2018
the circumpolar region. (A) Extent of the burns for 2001–2018 is from the and 2019–2020, are also depicted in red; these areas represent only 3% of total
FireCCI51 product, and the extent for 2019 and 2020 is the union of the C3SBA10 burning above the Arctic Circle during the 2001–2020 period. We show the
product and the Sentinel-2 burned-area map developed in this study. The annual burned area from 2001 to 2020, which is the period when the occurrence
Siberian Artic is the area inside the blue outline. Black represents areas that of fires accelerated. (B) Estimated storage of organic carbon in peatlands from
burned at least once for 2001–2018, and red represents areas that burned in a reference dataset (8).
3.0
A Annual burned area in the Siberian Arctic
FireCCILT11
2.5
MCD64A1
FireCCI51
Burned area (Mha)
2.0
C3SBA10
Landsat-7/-8
1.5
Sentinel-2
1.0
0.5
1982 1985 1990 1995 2000 2005 2010 2015 2020
0.8
B Annual burned area in carbon-rich peatlands (organic carbon storage >20 kg C m-2)
Burned area (Mha)
0.6
0.4
0.2
1982 1985 1990 1995 2000 2005 2010 2015 2020

Year
Fig. 2. Annual burned area in the Siberian Arctic and in carbon-rich peatlands for 1982–2020. (A) Annual burned area in the Siberian Arctic derived from remotely
sensed data from six products. (B) Annual burned area in carbon-rich peatlands; >20 kg C m−2 in storage of organic carbon obtained from a reference dataset (8). The
annual burned area in carbon-rich peatlands represents the median burned area for the available satellite products. Satellite burned-area products contain no data for 1994.

for 71% of this burned area. The years 2019 burned area (9.24 Mha) in the region from MCD64A1, C3SBA10, Landsat, and Sentinel-2,
and 2020 had the greatest mapped burned 1982 to 2020. The burned area mapped in respectively.
area in Siberia above the Arctic Circle (Fig. 1A) the Siberian Arctic varied between the satellite The sampling-based burned area in 2020,
(see supplementary text A for consistency of products, most notably the MCD64A1 product based on an assessment of errors of omis-
the time series of the burned area and fig. S2), for 2019 and 2020 (Fig. 2A). The burned areas sion and commission (16), was nearly 3 Mha
which represents 44% of the total mapped for 2020 were 1.71, 2.38, 2.59, and 2.62 Mha for (MCD64A1 = 2.83 ± 0.26 Mha, C3SBA10 = 2.92 ±
Air temperature Vapor-pressure deficit Mean NDVI Climatic water deficit

0.45
12 R 2 = 0.44 R 2 = 0.31 R 2 = 0.65 R 2 = 0.23
Air temperature (°C)
Slope = 0.652** Slope = 0.021** 40 Slope = 2.699**

11 0.4 0.65 Slope = 0.020**
Mean NDVI
CWD (mm)
VPD (kPa)
10 30
0.35
9 0.60
0.3 20
8
7 0.25 0.55
10
80
85
90
95
00
05
10
15
20
80
85
90
95
00
05
10
15
20
80
85
90
95
00
05
10
15
20
80
85
90
95
00
05
10
15
20
19
19
19
19
20
20
20
20
20
19
19
19
19
20
20
20
20
20
19
19
19
19
20
20
20
20
20
19
19
19
19
20
20
20
20
20
Year Year Year Year
Surface temperature Total precipitation Length of season Number of ignitions

130
Surface temperature (°C)
19 Total precipitation (mm) 8

R 2 = 0.16 R 2 = 0.00 R 2 = 0.28 R 2 = 0.09
Length of season (d)
Number of ignitions
Slope = 0.765 Slope = 0.005 Slope = 2.611** 400 Slope = 46.211
18
7 120
17 300
16 6 110
200
15
5 100 100
14
80
85
90
95
00
05
10
15
20
80
85
90
95
00
05
10
15
20
80
85
90
95
00
05
10
15
20
80
85
90
95
00
05
10
15
20
19
19
19
19
20
20
20
20
20
19
19
19
19
20
20
20
20
20
19
19
19
19
20
20
20
20
20
19
19
19
19
20
20
20
20
20
Year Year Year Year
Fig. 3. Trends of eight fire factors in the Siberian Arctic during 1982–2020. Factors are the mean summer air and surface temperature, mean VPD, total
summer precipitation, mean CWD, mean NDVI depicting vegetation green biomass, the length of the growing season, and the number of detected ignitions. The red
lines are linear regressions; slopes are estimated on a decadal time scale. *p < 0.05; **p < 0.01.
Air temperature Vapor-pressure deficit Mean NDVI Climatic water deficit

2 2
R 2 = 0.87** 2020 R = 0.89** 2020 R 2 = 0.78** 2020 R = 0.92** 2020
Burned area (Mha)
Burned area (Mha)

2.5 2.5
Burned area (Mha)
Burned area (Mha)
2.5 2.5
2 2 2 2
2019 2019 2019 2019
1.5 1.5 1.5 1.5
2001 2001 2001 2001
1 1 1 1
2018 2018 2018 2018
0.5 0.5 0.5 0.5
0 0 0 0
6 8 10 12 0.25 0.3 0.35 0.4 0.55 0.6 0.65 10 20 30 40
Air temperature (°C) VPD (kPa) Maximum NDVI CWD (mm)
Surface temperature Total precipitation Length of season Number of ignitions

R 2 = 0.86** 2020 2020 R 2 = 0.15* R 2 = 0.67** 2020 2
R = 0.87** 2020
Burned area (Mha)
Burned area (Mha)
2.5 2.5
Burned area (Mha)
2.5
Burned area (Mha)
2.5
2 2 2 2
2019 2019 2019 2019
1.5 1.5 1.5 1.5
2001 2001 2001 2001
1 1 1 1
2018
0.5 2018 0.5 0.5 2018 0.5
2018
0 0 0 0
14 15 16 17 18 5 6 7 100 110 120 100 200 300 400
Surface temperature (°C) Total precipitation (mm) Length of season (d) Number of ignitions
Fig. 4. Regression between the annual burned area and eight fire factors products. The factors are the mean summer air and surface temperature, mean
in the Siberian Arctic during 1982–2020. Solid lines are the best regression VPD, total precipitation, mean CWD, mean NDVI depicting green biomass,
(linear or exponential), based on the coefficient of determination (R2; *p < 0.05; the length of the growing season, and the number of ignitions. Red solid lines
**p < 0.01). The best regression model was the exponential for all the factors. depict a fit with a significant correlation (p < 0.01). The dashed lines are the
The annual burned area is the median burned area for the available satellite 95% prediction limits of the regressions.

0.17 Mha, Landsat = 2.92 ± 0.15 Mha, and

Sentinel-2 = 2.99 ± 0.14 Mha) (see full assess- Precipitation
ment of accuracy in table S1 and a description -0.09
of the results in supplementary Text B). The 0.75**
area estimate for 2019 and 2020 amounts to VPD
Plant water
~4.7 Mha. The mapped burned area is less than 0.12 stress 0.46* n = 20
the estimated burned area for all four products 0.93** Length of R 2 = 0.82
because the omission errors of the “burned” season Burned
0.04 0.43*
class (ranging from 15.5 to 53.7%) are higher 0.66** area
than the commission errors (ranging from 3.2 0.60**
Temperature NDVI
to 23.0%). Our estimates of carbon emissions 0.48**
from burning were 55.3 and 90.4 Tg C for 2019
0.49* Ignitions
and 2020, respectively, which is 156.7 and
256.1 Tg CO2-eq (including CO2 and CH4) (fig. S3).
Fires in 2020 damaged a wide area (0.71 Mha) Fig. 5. Causality networks for the association among factors of fire in the Siberian Arctic for 2001–2020.
of carbon-rich peatlands (organic carbon stor- Variables are categorized as climate (mean summer surface temperature, total precipitation, and mean VPD)
age >20 kg C m−2), indicated with a reference (yellow), vegetation (mean summer NDVI depicting green biomass, the length of the growing season, and plant
map of soil carbon storage (Fig. 1B) (8). The water stress measured by mean summer CWD) (green), and fire (number of detected ignitions and annual
area of carbon-rich peatlands affected by fires burned area) (light red). Factor loadings between variables are shown next to lines (*, p < 0.05; **p < 0.01).
has also recently expanded: 70% of total burned The width of the lines depicts the magnitude of the effect, and dashed lines represent nonsignificant
area occurred in these areas within the past effects. R2 is the variance explained for the annual burned area.
8 years of the record, and 30% occurred in
2020 (Fig. 2B).
rently with high CAPE values in 2002, 2005, controlling other factors that affect the extent
Trends of the fire factors for 1982–2020 2013, and 2018. of burning (Fig. 5 and fig. S8). Temperature
Various factors that may exacerbate the risk of showed significant positive relationships with
fire have increased significantly over the past Sensitivity of the burned area to the fire factors the lengthening of the growing season (0.66),
four decades in the Siberian Arctic (Fig. 3 and Linear and exponential regressions were used the vegetation green biomass represented by
fig. S4). Air temperature, NDVI, the length of to analyze the best association between the NDVImean (0.60), and atmospheric dryness
the growing season, and VPD have steadily annual burned area (aggregated with the me- measured by VPD (0.93). We hypothesized
risen. The average increase in summer air tem- dian across available satellites for each year) that these temperature-regulated factors and
perature was 0.66°C per decade. In 2019 and and the factors of fire regime. An exponential total precipitation would influence plant water
2020, the mean summer air temperature was regression was the best regression model stress, measured by CWD, but only VPD showed
11.35° and 11.53°C, which was 2.65° and 2.82°C (Fig. 4); the annual burned area accelerated a significant effect (0.75) for the low number
higher than the 1982–2020 average, respectively. when specific thresholds were exceeded. For of observations (n = 20). Despite this, the hy-
CWD, a proxy of plant water stress defined as example, the four years with the largest mapped pothesized relationships displayed the expected
the difference between potential and actual burned areas (2001, 2018, 2019, and 2020) had sign. Temperature and CWD had a positive rela-
evapotranspiration, also increased between a mean summer air temperature >10°C. The tionship with the number of detected ignitions
1982 and 2020, although the linear trend likely best fit was for CWD, which explained 92% of (0.49 and 0.43, respectively). Annual burned
began in the 2000s. More surprising, however, the interannual variability in the burned area. area presented an R2 of 0.82 and was directly
was the abrupt increase in CWD in 2019 and Other factors with a high coefficient of deter- explained by the number of detected ignitions
2020. The estimated number of ignitions, total mination (R2) were summer air temperature (0.48) and the CWD (0.46).
precipitation, and wind speed all had strong (87%), VPD (89%), and number of ignitions (87%). Climate factors may differ locally and through-
interannual variations, and the slope of their The annual burned area was correlated most out the fire season. An additional analysis based
trends was not significantly different from zero. weakly with total precipitation (15%). We also on local weather conditions during the burn-
The annual number of detected ignitions detrended the fire factors using the linear re- ing revealed that ignitions affecting areas larger
was relatively consistent, with a median of 143, gression shown in Fig. 3 before determining than 4000 ha occurred with average hourly
but high counts were observed in specific years, the correlation with the annual burned area to maximum temperatures of 28.6°C (SD = 3.4°C)
peaking at 423 in 2020. Seventy-two percent reduce the potential of spurious correlations. and mean wind direction from the northeast
of these 2020 ignitions were detected within The detrended correlations (fig. S7) confirmed (fig. S9). Thirty-day preignition precipitation
20 days, between 13 June and 3 July, reaching the high R2 for CWD (90%), air temperature was 0.37 mm (SD = 0.81 mm), and mean wind
Siberian Arctic regions as far north as 72.9° (80%), VPD (51%), and number of ignitions (86%), speed was 0.96 m s−1 (SD = 0.55 m s−1). Igni-
(fig. S5). Notably, these ignitions coincided with but the correlation decreased for NDVImean tions that lead to burned areas larger than
anomalously high values of convective available (from 78 to 11%). 4000 ha represent only 10% of all counts but
potential energy (CAPE) (fig. S6), an indicator We further examined the potential relation- account for 81% of all burned areas that were
of convective storms and lightning. Between ships among the fire-related factors in a mapped between 2001 and 2020.
13 June and 3 July, satellite thermal sensors structural equation modeling (SEM) (the
registered a rapid increase in the number of rationale of the proposed relationships is Projections of annual burned area and carbon
active fire detections, which accounts for described in the materials and methods). emissions under warming scenarios
40.6% of all hot spots detected in 2020. By The hypothesized causal model outperformed Annual burned area in 2018, 2019, and 2020
contrast, hot spots detected before 13 June the model validity analysis (p > 0.05 in the chi- more than doubled the long-term average,
represented only 1.1%. Similar peaks in the square test; details on the covariances and re- which was 0.24 Mha for the period 1982–2020
number of detected ignitions, preceding high siduals in the model are shown in table S2). in the Siberian Arctic. Summer 2001, with a
rates of active fire detection, occurred concur- The SEM supported the role of temperature in mean temperature nearing 10°C, was the first

A B average and damaged an unprecedented area of

18 peatlands. We found that temperature-related
Mean summer temperature (°C)

10 Historical + RCP 8.5
ERA5-Land Carbon-rich peatlands
factors of fire regime have increased significantly
16 RCP 8.5 8 over the past four decades and identified an ex-
Other
Burned area (Mha)

RCP 4.5
6 ponential relationship between these factors and
14 Historical
4 annual burned area, accounting for the unprec-
Year 2020
12 Year 2020
edented extent of the burns in 2019 and 2020.
2 Year 2001
The SEM results confirmed the positive as-
Year 2001 0
10 sociation between higher temperatures, longer
Historical + RCP 4.5
growing season, and greener vegetation. Higher
8 4 Year 2020
temperatures account for the earlier snow-
2 Year 2001 melt, permitting vegetation growth (17) and
6
0 increased green biomass (18), which increases
fuel availability. This earlier start of the grow-
60
80
00
20
40
60
80
00
60
80
00
20
40
60
80
00
19
19
20
20
20
20
20
21
19
19
20
20
20
20
20
21
ing season, also reported more widely (19),
Year Year
modifies water use and availability such that
C D plants may also experience water stress earlier
0
00
in the season (20). According to the SEM re-
40
Siberian Arctic Carbon-rich peatlands
14
sults, the lengthening of the growing season and

00
Historical + RCP 8.5 Historical + RCP 8.5

Emissions CO2-eq (Tg)
Emissions CO2-eq (Tg)

12
Historical + RCP 4.5 Historical + RCP 4.5

increasing green biomass of vegetation were
0
30
associated with increased plant water stress,
00
10
but the association was not significant, likely

0
80
owing to the limited number of observations.
0
20
The increasing vulnerability to drought is
0
60
exacerbated by extreme heatwaves, as in 2020,

0
which can potentially desiccate plants and re-

40
0
10
duce moisture in peat and thus increase the
0
20
severity of burning (21). This is reflected by the

high influence of atmospheric dryness, mea-
80 0
80 0
00
20
40
60
80
00
00
20
40
60
80
00
sured by VPD, on plant water stress, represented
19
20
20
20
20
20
21
19
20
20
20
20
20
21
Year Year by CWD, and its high correlation with annual
burned area. Furthermore, CWD encompasses
Fig. 6. Projected temperatures, annual burned areas, and emissions from fire in the Siberian Arctic. climatic factors, the water balance, and pheno-
(A) Mean summer air temperatures from climate reanalysis (ERA-5 Land) during 1982–2020 and historical logical changes that influence the susceptibility
and projected temperatures under the RCP 4.5 and 8.5 scenarios based on HadGEM2-CC model. (B) Annual of vegetation to fire, so the interconnection of
burned areas for the historical period and under the RCP 4.5 and 8.5 scenarios for carbon-rich peatlands the fire factors with CWD may explain why CWD
(organic carbon storage >20 kg C m−2) and other regions of the Siberian Arctic. (C and D) Projected CO2-eq was best correlated with the annual burned area.
emissions under the RCP 4.5 and 8.5 in the Siberian Arctic (C) and carbon-rich soils (D). Climate warming and extreme weather may
also account for the increase in the number
of ignitions for specific years. The year 2020
year on record to have a mapped burned area a mean annual emission of 37.8 (SD = 14.4) Tg C had record-breaking temperatures and caused
over twice that of the long-term average. The year−1 and 107.0 (SD = 40.7) Tg CO2-eq year−1 drought conditions early during the growing
exponential regression between the burned under RCP 8.5 between 2030 and 2050 (Fig. season (13). Recent warm winters, such as 2020,
area and temperature (Fig. 4) indicated that 6C), of which 27.6% would come from carbon- appear associated with abnormal circulation
an annual burn of 0.5 Mha occurred at a mean rich peatlands (Fig. 6D). Large fires of the patterns that also favor the early spring snow
summer temperature of 10.2°C. The 10°C magnitude observed in 2020 (burned area > melt and lower albedo that maintain warm
threshold also indicated the rapid growth of 2.5 Mha) might recur on a yearly basis at the conditions (22). Heatwaves and, more espe-
the annual burned area in 2018, 2019, and end of the century under RCP 8.5, with annual cially, increased surface temperatures are as-
2020. This indicates that small increases in mean carbon emissions of 135.0 (SD = 69.0) sociated with convective storms and ignitions,
summer mean temperature above the 10°C Tg C year−1 and 382.5 (SD = 195.6) Tg CO2-eq as confirmed with the SEM. Whereas light-
threshold tend to be associated with exten- year−1 (27.9% from carbon-rich peatlands). Under ning remains infrequent at high latitudes, it is
sive annual burned areas. the RCP 4.5 scenario, annual carbon emissions expected to increase as the climate warms (23).
The linear trend of mean summer air tem- would stabilize [51.7 (SD = 18.8) Tg C year−1] in Climatic warming thus has a dual effect on fire
perature (Fig. 3) indicated that temperatures the second half of the century, and fires such as regimes; warming increases the susceptibility
would reach 10.2°C by 2024 and reach the lev- those in 2020 (>2.5 Mha) would become less of vegetation and peatlands to fire and increases
els in 2020 by 2045 if mean summer temper- frequent in the Siberian Arctic, with a 10-year the number of lightning-caused ignitions.
atures continued to increase linearly at the return interval if greenhouse gas concentrations Increased winter warmth, as seen in 2020,
current rate. (Fig. 6A). The RCP 4.5 and 8.5 stabilize by the middle of the century. reflects changes in circulation that draw more
scenarios also indicated an increase in temper- heat and moisture from lower latitudes (24).
atures that could substantially expand the Discussion Circulation in eastern Siberia draws air and
burned area in the Siberian Arctic; annual The Siberian Arctic burned at the highest rates moisture from all directions of the compass
burned area could range from 0.5 to 2.5 Mha in 2019 and 2020, based on the burning trends (25) and is not immune to the so-called cir-
before the middle of the century under RCP over four decades of satellite data. Burning was culation “blocking” (26) seen elsewhere across
4.5 and RCP 8.5 (Fig. 6B). This would result in sevenfold higher in 2020 than the 1982–2020 the continent (27). Nonetheless, most burning

occurs during relatively gentle winds blowing of the record. This increase in annual burned Climate Change Service (C3S) Climate Data Store (CDS) (2019).
from the northeast, indicating that the pro- area suggests that the Arctic is already expe- https://doi.org/10.24381/cds.f333cf85.
38. Z. Wan, S. Hook, G. Hulley, MOD11A2 MODIS/Terra land
cesses that promote flammability may be dis- riencing a change in fire regimes caused by surface temperature/emissivity 8-day L3 global 1km SIN
tinct from those that promote the subsequent climatic warming. The burned areas in 2019 grid V006, NASA EOSDIS Land Processes DAAC (2015).
burning. and 2020 might be exceptional occurrences, but https://doi.org/10.5067/MODIS/MOD11A2.006.
39. J. Muñoz Sabater, ERA5-Land monthly averaged data from 1950
Our ignition detection method indicated that the recent temperature trend and projected to present, Copernicus Climate Change Service (C3S) Climate
numerous fires started near simultaneously scenarios indicate that temperatures are reach- Data Store (CDS) (2019). https://doi.org/10.24381/cds.68d2bb30.
across a vast region during a period of atmo- ing a threshold in which small increases above 40. J. T. Abatzoglou, S. Z. Dobrowski, S. A. Parks, K. C. Hegewisch,
Monthly climate and climatic water balance for global
spheric instability in the 2020 fire season, 10°C can alter fire-related factors and result terrestrial surfaces from 1958-2015, University of Idaho (2017).
from which we speculate that lightning was in exponentially increasing burned area and ttps://doi.org/10.7923/G43J3B0R.
the main cause of ignition, but local observa- associated fire emissions in the next decades. 41. H. Wouters, J. Berckmans, R. Maes, E. Vanuytrecht, K. De Ridder,
Global bioclimatic indicators from 1950 to 2100 derived from climate
tions are required to verify this supposition. An Forthcoming fires can potentially affect peat- projections, Copernicus Climate Change Service (C3S) Climate
alternative, or additional, explanation is that lands and deteriorate the permafrost, which in Data Store (CDS) (2021). https://doi.org/10.24381/cds.a37fecb7.
fires emerge from smoldering material that turn will exacerbate the carbon emissions from 42. K. Didan, MOD13Q1 MODIS/Terra vegetation indices 16-day L3
global 250m SIN grid V006, NASA EOSDIS. Inf. Serv. Land
has persisted through the winter to reemerge carbon-rich soils. Processes DAAC (2015). https://doi.org/10.5067/MODIS/
when conditions permit a broader conflagra- MOD13Q1.006.
tion (28, 29). We also found that satellite ther-
1. S. M. Natali et al., Proc. Natl. Acad. Sci. U.S.A. 118, e2100163118 AC KNOWLED GME NTS
mal sensors showed that fires spread quickly (2021). Funding: This work was funded by the Ministry of Agriculture,
after high CAPE values and midseason igni- 2. M. C. Serreze, J. A. Francis, Clim. Change 76, 241–264 (2006). Forestry and Fisheries (MAFF) of the government of Japan and the
tions, which suggests that most of the annual 3. E. Post et al., Sci. Adv. 5, eaaw9883 (2019). Center for International Forestry Research (“CIFOR”), through
4. Arctic Monitoring and Assessment Programme, Arctic Climate the project “Transitions to Climate Resilient Landscapes: Reducing and
burned area is caused by fires that started Change Update 2021: Key Trends and Impacts. Summary for Policy- Mitigating Boreal and Tropical Forest Fires to Promote Sustainable
during that time. Makers (Arctic Monitoring and Assessment Programme, 2021). Rural Livelihoods.” We acknowledge funds from the Spanish
The link we see between fires and temper- 5. E. A. Schuur et al., Nature 520, 171–179 (2015). Government grant PID2019-110521GB-I00, the Fundación Ramón
6. K. Dutta, E. Schuur, J. Neff, S. Zimov, Glob. Change Biol. 12, Areces grant CIVP20A6621, and the Catalan Government grant
ature suggests that severe fire years, such as 2336–2351 (2006). SGR 2017-1005. This work represents a contribution to CSIC
2020, will become increasingly common and 7. M. R. Turetsky et al., Nat. Geosci. 13, 138–143 (2020). PTI-TELEDETECT. Author contributions: Conceptualization: A.D.,
resulting carbon emissions will rise. The mag- 8. G. Hugelius et al., Proc. Natl. Acad. Sci. U.S.A. 117, D.L.A.G., and J.P. Methodology: A.D. and D.L.A.G. Investigation:
20438–20446 (2020). A.D. and D.L.A.G. Visualization: A.D., D.L.A.G., and J.P. Supervision:
nitude of future fires and carbon emissions, 9. A. Witze, Nature 585, 336–337 (2020). J.P. Writing – original draft: A.D. and D.L.A.G. Writing – review
however, remains uncertain. First, although the 10. M. C. Mack et al., Nature 475, 489–492 (2011). and editing: A.D., D.L.A.G., A.V., D.S., D.N., and J.P. Competing
frequency of lightning strikes appears likely to 11. N. H. French et al., Int. J. Wildland Fire 24, 1045–1061 (2015). interests: Authors declare that they have no competing interests.
12. J. L. McCarty et al., Biogeosciences 18, 5053–5083 (2021).
increase as temperatures rise (23), the scale Data and materials availability: All source code and the Sentinel-
13. A. Ciavarella et al., Clim. Change 166, 9 (2021). 2 and Landsat-8 burned-area maps are available at Zenodo
of any resulting fires depends on specific local 14. O. V. Churakova Sidorova, R. T. W. Siegwolf, M. V. Fonti, (33). Other data that support the findings of the study are openly
weather and vegetation conditions, which re- E. A. Vaganov, M. Saurer, Sci. Rep. 11, 19010 (2021). available. The burned area time series derived from AVHRR
main challenging to predict. Second, we only 15. F. S. Hu et al., Front. Ecol. Environ. 13, 369–377 (2015). sensors (FireCCILT11) (34) is available at https://doi.org/10.5285/
16. P. Olofsson et al., Remote Sens. Environ. 148, 42–57 (2014). 62866635ab074e07b93f17fbf87a2c1a. The burned area time series
considered direct emissions from burning and 17. A. Descals et al., Sci. Total Environ. 742, 140637 (2020). derived from MODIS sensors are available at https://doi.org/
disregarded indirect emissions, although these 18. L. T. Berner et al., Nat. Commun. 11, 4621 (2020). 10.5067/MODIS/MCD64A1.006 for the MCD64A1 product (35) and
are not necessarily negligible. Burning removes 19. S. Piao et al., Glob. Chang. Biol. 25, 1922–1940 (2019). https://doi.org/10.5285/58f00d8814064b79a0c49662ad3af537
20. J. Peñuelas, T. Rutishauser, I. Filella, Science 324, 887–888 (2009). for the FireCCI51 (36) product. The burned area time series
the peat that insulates permafrost, exposing 21. M. R. Turetsky, W. F. Donahue, B. W. Benscoter, Nat. Commun. derived from Sentinel-3 (C3SBA10) (37) is available at https://doi.
it to thawing, which promotes soil respiration 2, 514 (2011). org/10.24381/cds.f333cf85. The Landsat-7 and -8 surface
and the production of carbon dioxide and meth- 22. A. B. M. Collow et al., J. Clim. 35, 3075–3090 (2022). reflectance datasets are available at https://earthexplorer.usgs.gov,
23. Y. Chen et al., Nat. Clim. Chang. 11, 404–410 (2021). and the Sentinel-2 Level-2A surface reflectance at https://scihub.
ane (30). Estimates from field studies in two 24. Q. You et al., Earth Sci. Rev. 217, 103625 (2021). copernicus.eu. The active fire datasets obtained from MODIS
different boreal forests in Alaska suggest that 25. J. Piao, W. Chen, S. Chen, H. Gong, Q. Zhang, J. Clim. 33, is available at https://firms.modaps.eosdis.nasa.gov/active_fire.
post-fire carbon emissions range from one- 3883–3899 (2020). Fire emission datasets were downloaded from the FIRECAM
26. C. You, M. Tjernström, A. Devasthale, D. Steinfeld, Geophys. platform https://globalfires.earthengine.app/view/firecam. Land
third to more than double those that occur Res. Lett. 49, e2022GL097899 (2022). surface temperature from MODIS (MOD11A2) (38) is available
during burning (31). Furthermore, permafrost 27. T. Nakamura, T. Sato, Environ. Res. 209, 112881 (2022). at https://doi.org/10.5067/MODIS/MOD11A2.006. Climate
prevents deeper burning in peatlands (21). As 28. R. C. Scholten, R. Jandt, E. A. Miller, B. M. Rogers, reanalysis ERA5-Land Monthly (39) is available at https://doi.org/
S. Veraverbeke, Nature 593, 399–404 (2021).
permafrost retreats, high temperatures and 10.24381/cds.68d2bb30, and TERRACLIMATE dataset (40) is
29. J. L. McCarty, T. E. Smith, M. R. Turetsky, Nat. Geosci. 13, available at https://doi.org/10.7923/G43J3B0R. The downscaled
drying conditions may favor higher combustion 658–660 (2020). air temperature projections from the HadGEM2-CC model (41)
rates (32). We used combustion rates ranging 30. J. E. Holloway et al., Permafr. Periglac. Process. 31, 371–382 are available at https://doi.org/10.24381/cds.a37fecb7. The
from 2.0 kg C m−2 for tundra to 3.4 kg C m−2 for (2020). storage of organic carbon from peatlands is available at
31. S. Veraverbeke et al., Curr. Opin. Environ. Sci. Health 23, https://bolin.su.se/data/hugelius-2020. Lastly, the NDVI time
boreal forests (31), but dry peatlands can release 100277 (2021). series from the GIMMS3g dataset is available at https://www.nasa.
up to 16.8 kg C m−2 (21), indicating that much 32. M. R. Turetsky et al., Nat. Geosci. 8, 11–14 (2015). gov/nex/data, and the NDVI from the MOD13Q1v6 product
higher emissions are credible. 33. A. Descals, adriadescals/ArcticFires: Arctic Fire repository, (42) at https://doi.org/10.5067/MODIS/MOD13Q1.006. License
Zenodo (2022). https://doi.org/10.5281/zenodo.6907765. information: Copyright © 2022 the authors, some rights
A previous study proposed temperature and 34. E. Chuvieco, M. L. Pettinari, G. Otón, ESA Fire Climate Change reserved; exclusive licensee American Association for the
rainfall thresholds for the annual burned areas Initiative, (Fire_cci): AVHRR-LTDR Burned Area Grid product, Advancement of Science. No claim to original US government
in the Alaskan tundra (15). The extensive area Centre for Environmental Data Analysis, Version 1.1 (2020). works. https://www.sciencemag.org/about/science-licenses-
http://dx.doi.org/10.5285/62866635ab074e07b93f17fbf87a2c1a journal-article-reuse
burned in 2019 and 2020 corroborated the 35. L. Giglio, C. Justice, L. Boschetti, D. Roy, MCD64A1 MODros. Inf.
proposed curve-growth relationship between Serv./Terra+ aqua burned area monthly L3 global 500m SIN grid SUPPLEMENTARY MATERIALS
annual burned area and climate-related fac- V006, NASA EOSDIS. Inf. Serv. Land Processes DAAC (2015).
science.org/doi/10.1126/science.abn9768
https://doi.org/10.5067/MODIS/MCD64A1.006.
tors for the Alaskan tundra. Hu et al. (15) fore- 36. E. Chuvieco, M. Pettinari, J. Lizundia-Loiola, T. Storm,
casted that the annual burned area would Supplementary Text
M. Padilla Parellada, ESA Fire Climate Change Initiative, (Fire_cci):
Figs. S1 to S9
double in the Alaskan tundra by the end of the MODros. Inf. Serv. Fire_cci burned area pixel product, Centre for
Tables S1 and S2
Environmental Data Analysis, version 5.1 (2018). https://doi.org/
century. We found, however, that the annual References (43–65)
10.5285/58f00d8814064b79a0c49662ad3af537.
burned area in the Siberian Arctic already dou- 37. Copernicus Climate Change Service, Fire burned area from Submitted 7 January 2022; accepted 13 September 2022
bled the long-term average in the past 3 years 2001 to present derived from satellite observations, Copernicus 10.1126/science.abn9768

NEUTRINO ASTROPHYSICS The measured flux of astrophysical neutrinos

(16) is largely isotropic, equally distributed among
Evidence for neutrino emission from the nearby neutrino flavors, and can be described by a sin-
gle power-law energy distribution that extends
active galaxy NGC 1068 from ~10 TeV to peta–electron volt energies
(17, 18). A specific source of high-energy cosmic
IceCube Collaboration*† neutrinos was reported after the spatial and
temporal coincidence of a high-energy IceCube
A supermassive black hole, obscured by cosmic dust, powers the nearby active galaxy NGC 1068. Neutrinos, neutrino (19) with a gamma-ray flaring blazar,
which rarely interact with matter, could provide information on the galaxy’s active core. We searched for TXS 0506+056 (20–22). TXS 0506+056 con-
neutrino emission from astrophysical objects using data recorded with the IceCube neutrino detector between tains a typical accretion disk and a dusty torus,
2011 and 2020. The positions of 110 known gamma-ray sources were individually searched for neutrino which emits high-energy radiation and, possi-
detections above atmospheric and cosmic backgrounds. We found that NGC 1068 has an excess of 79þ22 20
bly, cosmic rays (22). Neutrinos detected using
neutrinos at tera–electron volt energies, with a global significance of 4.2s, which we interpret as associated IceCube were correlated with a catalog of 110
with the active galaxy. The flux of high-energy neutrinos that we measured from NGC 1068 is more than an known gamma-ray emitters, with a signifi-
order of magnitude higher than the upper limit on emissions of tera–electron volt gamma rays from this source. cance of 3.3s (23). The individual sources that
made the largest contribution to the total sig-
nificance of that catalog were the active gal-
O
bservations of high-energy cosmic rays of sight, the AGN is observed as a blazar (10). axy NGC 1068 and the blazars TXS 0506+056,
(protons and atomic nuclei from space), AGNs are potential neutrino emitters (11, 12); if PKS 1424+240, and GB6 J1542+6129. The signif-
up to 1019 to 1020 eV (1–3), have demon- a plasma jet is present, it might dominate the icance of the neutrino excess from the direc-
strated that powerful cosmic particle emission (13, 14). tion of NGC 1068 was reported as 2.9s, which
accelerators must exist, but their nature The IceCube Neutrino Observatory (15) is is insufficient to claim a detection (23).
and location remain unknown. Interstellar mag- based at the Amundsen-Scott South Pole Sta-
netic fields change the direction of charged tion in Antarctica and has been operating since Searching for point-like neutrino emission
cosmic particles during their propagation to 2010. The observatory uses 1 km3 of optically We analyzed data collected with IceCube be-
Earth, concealing their sources. High-energy transparent glacial ice as a detection medium tween 13 May 2011 and 29 May 2020. This period
photons and neutrinos are not deflected, so to measure Cherenkov light—ultraviolet and begins with the installation of the full 86-string
they could be used to locate the cosmic accel- blue photons emitted by charged secondary detector configuration. Previous searches for
erators. Both travel along straight paths and particles traveling at a speed above the phase cosmic neutrino sources (23) included data
are produced wherever cosmic rays interact velocity of light in the ice. These relativistic collected with the incomplete detector with
with ambient matter or light, in or near the (close to the speed of light) secondary particles fewer strings going back to 2008 and the full
acceleration sites (4, 5). Depending on the en- are produced when neutrinos interact with detector up to the spring of 2018. We only used
vironment in which these interactions occur, nuclei in or near the instrument. A total of the full detector data because our methods de-
gamma rays could rapidly lose energy through 5160 digital optical modules (DOMs) are in- pend on uniformly processed data. The IceCube
several processes, including pair-production stalled on 86 vertical cables (strings), spaced dataset we used (24) has consistent selection
in interactions with lower-energy photons. 125 m apart to form a three-dimensional array criteria (25). We reprocessed these data uni-
Above tera–electron volt energies, gamma rays in the ice. Each DOM records the number of formly to remove data sample fragmentation,
are strongly absorbed over cosmological dis- induced photoelectrons (charges) as a func- align different data-taking conditions and cal-
tances through interactions with the extragalac- tion of time. ibrations, and improve event reconstructions
tic background light and the cosmic microwave
background (6). Neutrinos are not affected by
intergalactic absorption, so they could poten-
tially be used to probe tera–electron volt cos-
mic accelerators.
Active galaxies, those that host an active ga-
lactic nucleus (AGN) (7), are characterized by a
very bright central region powered by the ac-
cretion of material onto a supermassive black
hole (SMBH). The accretion flow of matter into
the SMBH is usually surrounded by an obscur-
ing, dusty torus, causing the observable char-
acteristics of an AGN to depend on the viewing
angle from Earth. For example, Seyfert II gal-
axies (8) are thought to be viewed edge on, with
the line of sight passing directly through the
obscuring torus (9). In some cases, the AGN
can launch a strong, narrow jet of accelerated
plasma. If such a jet is oriented close to the line Fig. 1. Sky map of the scan for point sources in the Northern Hemisphere. The color scale indicates the
logarithm of the local P value (Plocal) obtained from our maximum likelihood analysis, evaluated (with the
spectral index as a free parameter) at each location in the sky. The map is shown in equatorial coordinates on
*Corresponding authors: [email protected]; F. Halzen a Hammer-Aitoff projection. The black circles indicate the three most significant objects in the source list
([email protected])
†IceCube Collaboration authors and affiliations are listed in the search, which are labeled. The circle around NGC 1068 contains the most significant location in the Northern
supplementary materials. Hemisphere, shown in higher resolution in Fig. 2A.

(26). We applied the directional track recon-

struction method SPLINERECO (26, 27, 28) to all Table 1. Summary of final P values. For each of the three tests performed, we report the most
events in our dataset (26). We incorporated ad- significant local and global P values.
ditional calibration information in the extrac-
tion of the charges at each DOM and in the
corresponding arrival times of Cherenkov pho- Pretrial P value, Plocal Posttrial P value, Pglobal
Test type
tons. Compared with previous work (23), this (local significance) (global significance)
introduces small changes in the reconstructed Northern Hemisphere scan 5.0 × 10−8 (5.3s) 2.2 × 10−2 (2.0s)
.....................................................................................................................................................................................................................
event energies and some reconstructed event List of candidate sources, single test 1.0 × 10 (5.2s) −7
1.1 × 10−5 (4.2s)
.....................................................................................................................................................................................................................
directions (26). To ensure a uniform detector List of candidate sources, binomial test 4.6 × 10−6 (4.4s) 3.4 × 10−4 (3.4s)
.....................................................................................................................................................................................................................
response, the DOMs of the DeepCore subarray,
intended to study ≲100‐GeV neutrinos, were
excluded (25). Our resulting dataset, which
is optimized for track-like h events
i induced
by muon (anti-)neutrinos ðn Þ , has a total ex-
m
posure time of 3186 days.
We restricted our searches to the Northern
Hemisphere from declination d = −3° to 81°,
where IceCube is most sensitive to astrophys-
ical sources. IceCube uses Earth as a passive
cosmic muon shield and as a target material
for neutrinos. Hence, by selecting only upward-
going events, we reduced the atmospheric muon
background, which contributes <0.3% to our
final event sample (25). Declinations higher
than 81° are excluded because low-energy
events from those directions are closely aligned
with the strings of IceCube, complicating our
distinction between the signal and background
(26). The resulting loss of sky coverage is <1%.
A total of ~670,000 neutrino-induced muon
tracks pass the final event selection criteria
(25). However, only a small fraction of these
events originate from neutrinos produced in
astrophysical sources. Most arise from the de-
cay of particles (specifically mesons) that are Fig. 2. High-resolution scan around the most significant location. (A) High-resolution scan around the
produced in the interaction of cosmic rays most significant location marked by a white cross, with contours showing its 68% (solid) and 95% (dashed)
with nuclei in Earth’s atmosphere. To discrim- confidence regions. The red dot shows the position of NGC 1068, and the red circle is its angular size in
inate neutrinos that originate from individual ^ 2, between NGC 1068 and
the optical wavelength (61). (B) The distribution of the squared angular distance, y
astrophysical sources from the background of the reconstructed event directions. We estimated the background (orange) and the signal (blue) from
atmospheric and diffuse astrophysical neutrinos, Monte Carlo simulations, assuming the best-fitting spectrum at the position of NGC 1068. The superposition
we used a maximum-likelihood method and of both components is shown in gray and the data in black. This representation of the result ignores the
likelihood ratio hypothesis testing, based on the energy and angular uncertainty of the events.
estimated energy, direction, and angular uncer-
tainty of each event (26). The median angular
resolution of each neutrino arrival direction, tuple of mns and g fully determines the flux of peated under the same three spectral index
composed of reconstruction uncertainty and muon neutrinos, Fnm þnm , at any given energy. hypotheses as the sky scan.
the kinematic angle between the parent neu- We performed three different searches (26). All analysis methods, including the selection
trino and the muon, is 1.2° at 1 TeV, 0.4° at The first search consists of three discrete scans of the hypotheses to be tested, were formu-
100 TeV, and 0.3° at 1 PeV. We assume any of the Northern Hemisphere to identify the lated a priori. The performance of each method
point source emits a neutrino flux Fnm þnm de- location of the most statistically significant was evaluated using simulations and random-
scribed by a generalized power-law energy excesses of high-energy neutrino events. These ized experimental data (26). The local P values
spectrum, Fnm þnm ðEn Þ ¼ F0 ·ðEn =E0 Þ g , with scans use three different hypotheses for the are determined as the fraction of background-
normalization energy E0 = 1 TeV, where En is spectral index: g as a free parameter, g fixed to only simulations that yield a test statistic greater
the neutrino energy and the spectral index g 2.0, and g fixed to 2.5. The other two searches than (or equal to) the test statistic obtained
and the flux normalization F0 are free parame- use a list of 110 preselected astronomical ob- from the experimental data. The global P values
ters (26). This corresponds to two correlated jects, all located in the Northern Hemisphere: are determined from the smallest local P value
model parameters that we express as a pair The second search is for the most significant after correcting for testing multiple locations
(mns, g), where mns is the mean number of as- candidate neutrino source in the list, whereas (the look-elsewhere effect) (26). We use this
trophysical neutrino events associated with a the third search consists of a binomial test to global value to assess the evidence that the
given point in the sky. Using the energy- and evaluate the significance of observing an ex- data provide against a background-only null
declination-dependent effective area of the de- cess of k sources with local P values below or hypothesis (that the data consist purely of at-
tector and assuming a spectral index g, mns can equal to a chosen threshold, with k being an mospheric background and isotropic cosmic
be directly converted to F0 (26). Hence, the index from 1 to 110. The binomial test is re- neutrinos).

declination range we analyzed, −3° ≤ d ≤ 81°.

The number of objects in the list was chosen
such that a local 5s detection corresponds to
a global detection >4s after accounting for
the number of trials. The 110 astronomical
objects were selected from the gamma-ray flux
weighted by the IceCube sensitivity. Using the
gamma-ray flux above 1 GeV (32), the list con-
tains 95 blazars, 5 AGNs, and 9 other types of
galaxies. One galactic source is added be-
cause of its tera–electron volt gamma-ray emis-
sion (26, 33). We do not assume or require any
relationship between the observed gamma-ray
flux and the hypothesized neutrino flux during
our hypothesis testing (26).
Of the 110 astronomical objects tested,
NGC 1068 is the most significant with a local
P value of 1 × 10−7 (5.2s); it has best-fitting val-
ues of spectral index ^g ¼ 3:2þ0:2 0:2 and mean
number of signal events m ^ ns ¼ 79þ22 20. NGC 1068
is contained within the 68% confidence re-
gion around the most significant location in
the sky scan, offset by 0.11° (less than the direc-
tional uncertainty), consistent with neutrino
emission from NGC 1068 (Fig. 2). After cor-
Fig. 3. Source parameters of NGC 1068. The best-fitting flux values are shown as the black cross, with recting for having tested the 110 sources in the
the solid and dashed contours representing the 68% and 95% confidence levels, respectively. These were catalog, the global P value for NGC 1068 is 1.1 ×
derived from Wilks’ theorem—i.e., by using the log-likelihood ratio (color scale), 2D log L, between any 10−5, corresponding to a significance of 4.2s.

fixed values for the two flux parameters g; F1nmTeV
þnm and the overall best-fitting point marked by the cross.
Binomial test
The side panels show the corresponding one-dimensional projections (profile likelihoods) as black solid lines,
with the gray shaded areas showing the one-dimensional 68% confidence levels (as reported in the text).
We performed a binomial test to investigate
All contours include only statistical uncertainties.
how likely it is to find an excess of k sources
from the catalog with local P values below or
equal to a threshold (26), as defined a priori.
The discrete scan of the sky maximizes the a local significance of 5.3s. After including a By scanning the P value threshold, we find
signal and background likelihood functions penalty for the multiple tests performed—i.e., the smallest background probability for an ex-
at each point on an ~(0.2° × 0.2°) grid, deter- how likely it is to observe an equal or larger cess of k = 3 sources, under the free spectral
mining the local value of the test statistic, the significance when scanning many independent index hypothesis, with local P values ≤4.6 ×
best-fitting value of the mean number of as- sky positions under the three spectral index 10−6. This is equivalent to a local significance
trophysical neutrino events ðm ^ ns Þ , and the hypotheses (26)—the global P value corresponds of 3.7s, a small increase from the previous
mean spectral index ð^g Þ. The scan is repeated (26) to 2.0s and is therefore not significant search (23). After correcting for having tested
with fixed spectral indices g = 2.0 and g = 2.5 without additional prior information. The high- three different spectral index hypotheses, we
to reduce the impact of background fluctua- resolution scan around the most significant obtain a posttrial significance of 3.4s for the
tions, which arise from atmospheric neutrinos. location is shown in Fig. 2A. binomial test. Besides NGC 1068, the other
The atmospheric spectrum, produced by the We searched a posteriori for astrophysical two objects contributing to the excess are the
decay of charged pions and kaons as well as the counterparts in close proximity to the direc- blazars PKS 1424+240 and TXS 0506+056
muons produced when pions and kaons decay, tion of the five most significant locations in (Fig. 1), for which we found potential neu-
follows a power law with g ≈ 3.7 (29), and thus each of the three sky maps. The nearby active trino emission with local significances of 3.7s
those fluctuations typically yield ^g ≳ 3. Fixing galaxy NGC 4151 (8) is located at ~0.18° from and 3.5s, respectively. This significance of
the spectral index therefore increases the sen- the fourth most significant location in the map TXS 0506+056 relates to a time-integrated sig-
sitivity to astrophysical sources with g < 3. obtained with g = 2.5. nal over the duration of our dataset, whereas
For the 20 most significant locations in each previous analyses have found evidence for tran-
of the three searches, we scan the 1.5° × 1.5° Source list search sient emission (20, 21, 34). A search for transient
vicinity using increased resolution of ~(0.03° × Searching the entire Northern Hemisphere re- emission has a lower effective background than
0.03°) on a square grid. We find the most sig- quires a strong numerical penalty because of our time-integrated searches. The total num-
nificant location identified in the first scan (with testing multiple locations. This trials factor can ber of contributing candidate sources (k = 3)
free spectral index) at right ascension 40.69° be reduced by restricting the search to a list in the binomal test is lower than the four pre-
and declination 0.09° (J2000 equinox) with of a priori selected positions based, for exam- viously reported (23). Although the local sig-
^g ¼ 3:2 and ^ m ns ¼ 81. ple, on known gamma-ray emission (30). To nificance of PKS 1424+240 increased from 3.0s
avoid confirmation bias, we maintain the pre- to 3.7s, the local significance of GB6 J1542+61
Sky scan search vious selection method (23), which was adopted decreased from 2.9s to 2.2s, falling below
The full sky map of the discrete scan is shown before there was any indication for neutrino the threshold corresponding to the best-fitting
in Fig. 1. At the most significant location, the emission from NGC 1068 (31). This approach binomial P value. The results of all three searches
local P value is 5 × 10−8, which corresponds to identified a total of 110 objects within the (26) are summarized in Table 1.

Neutrino emission from NGC 1068 IceCube (this work) Electromagnetic observations (26)
The high-resolution scan around the most sig- Theoretical ν model (52,55) 0.1 to 100 GeV gamma-rays (40,41)
nificant location in the Northern Hemisphere
Theoretical ν model (53) > 200 GeV gamma-rays (42)
is shown in Fig. 2A, with NGC 1068 located
inside the 68% confidence region. The posi- 10 −9
tion of NGC 1068 produced m ^ ns ¼ 79þ22
20 more
events than expected from the atmospheric
E 2 Φ [TeV cm− 2 s− 1 ]
10 −10
and diffuse astrophysical neutrino backgrounds.
Figure 2B shows the distribution of the angu-
lar separation of these events from NGC 1068. 10 −11
Among the 79 most contributing events, 63
were included in a previous analysis (23). The 10 −12
systematic uncertainty on m ^ ns is ~2 events (26).
The measured spectral index is ^g ¼ 3:2þ0:2 0:2
with an estimated systematic uncertainty of 10 −13
±0.07 (26), consistent with previous results
(23). We estimate these systematic uncertain- 10 −14
ties by analyzing simulated data, assuming a 10 −15 10 −12 10 −9 10 −6 10 −3 10 0 10 3 10 6
source with flux equal to the one measured for
Energy [GeV]
NGC 1068 but varying assumptions about the
detector response (26). Systematic uncertainties Fig. 4. Multimessenger spectral energy distribution of NGC 1068. Gray points show multifrequency
arise mainly from the modeling of the photon observations (data sources listed in table S1). Dark and light green points indicate gamma-ray observations
propagation in the glacial ice—e.g., scattering at 0.1 to 100 GeV (40, 41) and >200 GeV (42), respectively. Arrows indicate upper limits, and error bars
and absorption—and the efficiency with which are 1s confidence intervals. The solid, dark blue line shows our best-fitting neutrino spectrum with the
photons are detected by the IceCube optical dark blue shaded region indicating the 95% confidence region. We restrict this spectrum to the range
modules. Systematic uncertainties are smaller between 1.5 and 15 TeV, where the flux measurement is well constrained (26). Two theoretical predictions
than statistical uncertainties for directional track are shown for comparison: The light blue shaded region and the gray line show the NGC 1068 neutrino
reconstructions (26) but have a nonnegligible emission models from (52, 55) and (53), respectively. The shaded region covers possible values of the
effect on the energy reconstructions. gyrofactor 30 ≤ hg ≤ 104 used to describe uncertainty in the efficiency of the underlying particle acceleration
The properties of the source spectrum are (55). All fluxes F are multiplied by the energy squared E2.
shown in Fig. 3, which shows the likelihood
as a function of the model parameters (F0, g)
evaluated at the coordinates of NGC 1068.
The conversion of m ^ ns to the flux F0 accounts
for the contribution from tau neutrino in-
teractions (which produce muons) assuming
an equal neutrino flavor ratio. The best-fitting
flux averaged over the data-taking period,
at a neutrino energy of 1 TeV, is F1Tev nm þnm ¼
ð5:0 T 1:5stat T 0:6sys Þ 10 11 TeV 1 cm 2 s 1.
This systematic uncertainty was estimated by
varying the flux normalization under differ-
ent ice and detector properties, such that we
reproduce the observed values of ^g and ^ m ns in
the median case.
Our analysis assumed that the spectrum fol-
lows an unbroken power law over the entire
energy range of the dataset. However, our re-
sults show that the main contribution to the
excess (and thus the measured spectral index
and flux normalization) comes from neutrinos
in an energy range from 1.5 to 15 TeV, which Fig. 5. Comparison of point-source fluxes with the total diffuse astrophysical neutrino flux. Fluxes
contributes 68% to the total test statistic. Out- for NGC 1068 (blue line, this work), TXS 0506+056 (orange line, this work), and the diffuse neutrino
side this energy range, the data do not strong- background [brown data points and gray band (17, 25)] are given for a single flavor of neutrinos and
ly constrain the inferred flux properties. Our antineutrinos. All fluxes Fvþv are multiplied by the neutrino energy squared E2n . For the conversion of the
results strengthen the suggestion (23) that diffuse astrophysical flux measured from the nent channel (17), we assume an equal flavor ratio. Shaded
NGC 1068 could be a neutrino source; we find regions and dashed lines indicate 68% confidence intervals. Downward arrows are 68% upper limits.
a higher statistical significance for this result
(4.2s versus 2.9s).
Incrementally removing the most contribut- an accumulation of neutrinos (26). We visually of unexpected contamination or anomalies
ing neutrino events one by one from the vicinity inspected all neutrino events contributing to (26). Out of the 20 events contributing the
of NGC 1068 shows that the excess persists, the excess from NGC 1068, finding typical, well- most to the test statistic, 19 were included in the
which indicates that it is not dominated by reconstructed, horizontal, and approximately previous analysis (23). Although the location is
one or a few single events but is the result of tera–electron volt–energy tracks with no sign therefore dominated by the same neutrinos, the

differences in the data processing modify their fluxes from the AGN NGC 1068 and the blazar 10. C. M. Urry, P. Padovani, Publ. Astron. Soc. Pac. 107, 803
contribution to the likelihood function. These TXS 0506+056 with the overall diffuse flux (1995).
11. D. Eichler, Astrophys. J. 232, 106 (1979).
changes are small but consistently cause the of astrophysical neutrinos. These two sources 12. V. S. Berezinsky, V. L. Ginzburg, Mon. Not. R. Astron. Soc. 194,
reconstructed neutrinos to align more closely contribute ~1% of the total diffuse flux in their 3–14 (1981).
with the direction of NGC 1068, thus strength- observed energy ranges. For NGC 1068, the ob- 13. K. Mannheim, P. L. Biermann, Astron. Astrophys. 221, 211–220
(1989).
ening the association (26). served energy range does not strongly overlap 14. F. Halzen, E. Zas, Astrophys. J. 488, 669–674 (1997).
with the diffuse flux, and its best-fitting spectral 15. M. G. Aartsen et al., J. Instrum. 12, P03012 (2017).
Astrophysical interpretation index of ^g ¼ 3:2þ0:20:2 is higher than the diffuse
16. IceCube Collaboration, Science 342, 1242856 (2013).
The flux we measured from the direction of flux with g ¼ 2:53þ0:07
0:07 (17). Nonetheless, it is evi-
17. M. G. Aartsen et al., Phys. Rev. Lett. 125, 121104 (2020).
18. R. Abbasi et al., Phys. Rev. D 104, 022002 (2021).
NGC 1068 was 79þ22 20 muon-neutrino events dent that the diffuse flux could contain additional 19. M. G. Aartsen et al., Astropart. Phys. 92, 30–41 (2017).
over a period of 3186 days, corresponding to bright and nearby sources, like NGC 1068, a 20. IceCube Collaboration et al., Science 361, eaat1378 (2018).
F1nmTev
þnm ¼ ð5:0 T 1:5stat Þ 10
11
TeV 1 cm 2 s 1 . large population of faint sources, or both. For 21. IceCube Collaboration, Science 361, 147–151 (2018).
22. P. Padovani, F. Oikonomou, M. Petropoulou, P. Giommi,
If all neutrino flavors contribute equally, as example, the nearby Seyfert I galaxy NGC 4151 E. Resconi, Mon. Not. R. Astron. Soc. 484, L104–L108
expected for sources dominated by pion decay has previously been suggested as a potential (2019).
and neutrino oscillations over cosmic distances neutrino source (55, 56, 57). Depending on the 23. M. G. Aartsen et al., Phys. Rev. Lett. 124, 051103 (2020).
24. M. G. Aartsen et al., Eur. Phys. J. C 79, 234 (2019).
(35, 36), the all-flavor flux is a factor of 3 higher. luminosity function and cosmological evolution,
25. M. G. Aartsen et al., Astrophys. J. 833, 3 (2016).
We adopt a distance of 14.4 Mpc for NGC 1068 the contribution from many faint sources at 26. Materials and methods are available as supplementary materials.
(37), although values in the literature range large distances ðredshift z ≳ 1Þ could be large 27. J. Ahrens et al., Nucl. Instrum. Methods Phys. Res. A 524,
from 10.3 ± 3 Mpc (38) to 16.5 Mpc (39). For (58, 59). 169–194 (2004).
28. R. Abbasi et al., J. Instrum. 16, P08034 (2021).
comparison, assuming isotropic emission, this We cannot characterize the underlying source 29. R. Abbasi et al., Astropart. Phys. 34, 48–58 (2010).
leads to a redshift-corrected equivalent neu- population with so few known point sources. 30. T. K. Gaisser, R. Engel, E. Resconi, Cosmic Rays and Particle
trino luminosity Ln, from 1.5 to 15 TeV and at However, our observation of neutrino emis- Physics: Second Edition (Cambridge Univ. Press, 2016).
31. M. Aartsen et al., Astrophys. J. 835, 151 (2017).
14.4 Mpc, of Ln = (2.9 ± 1.1stat) × 1042 erg s−1. sion from NGC 1068 can place constraints on 32. S. Abdollahi et al., Astrophys. J. Suppl. Ser. 247, 33
This is higher than the equivalent gamma-ray the density r of AGNs with similar or greater (2020).
luminosity of 1.6 × 1041 erg s−1 observed in the neutrino luminosity in the local Universe, 33. S. P. Wakely, D. Horan, International Cosmic Ray Conference,
vol. 3 of International Cosmic Ray Conference (2008),
energy range 100 MeV to 100 GeV (40, 41) and independent of the precise emission process pp. 1341–1344.
higher than the upper limits reported above or processes. If there is one source within a 34. P. Padovani et al., Mon. Not. R. Astron. Soc. 480, 192–203
200 GeV (42) (Fig. 4). spherical volume of radius 14.4 Mpc (37), the (2018).
35. M. Bustamante, J. F. Beacom, W. Winter, Phys. Rev. Lett. 115,
High-energy neutrinos are generated in or distance to NGC 1068, then r ≲ 10 4 Mpc 3 . 161302 (2015).
near astronomical sources as decay products This density estimate relies on the observation 36. C. A. Argüelles, T. Katori, J. Salvado, Phys. Rev. Lett. 115,
of charged mesons, which are themselves of a single source, which may or may not be 161303 (2015).
37. J. Bland-Hawthorn et al., Astrophys. Space Sci. 248, 9–19
produced in proton-proton interactions (4), or representative of the population, so this is an
(1997).
interactions between protons and low-energy upper limit. Simulations of populations with 38. B. F. Madore, in 27th Rencontres de Moriond: Physics of Nearby
ambient radiation (5). Gamma rays are also various luminosity functions (60) have an un- Galaxies: Nature or Nurture?, T. X. Thuan, J. Tran Thanh Van,
produced in the same processes, through the certainty of more than an order of magnitude. C. Balkoski, Eds. (Ed. Frontieres, 1992), pp. 25–28.
39. Gravity Collaboration, O. Pfuhl et al., Astron. Astrophys. 634,
decay of neutral mesons. Neutrinos escape the Nevertheless, given the differences in spectrum A1 (2020).
site without further interactions, whereas pho- and distance between NGC 1068 and TXS 0506 40. M. Ajello et al., Astrophys. J. 892, 105 (2020).
tons may experience additional interactions +056, which is ~100 times farther away, we 41. S. Abdollahi et al., Astrophys. J. Suppl. Ser. 247, 33 (2020).
42. V. A. Acciari et al., Astrophys. J. 883, 135 (2019).
depending on the optical depth of the envi- suggest that there are at least two populations 43. C. Ricci et al., Astrophys. J. Lett. 815, L13 (2015).
ronment. Therefore, the connection between of neutrino sources, which could differ in both 44. A. Marinucci et al., Mon. Not. R. Astron. Soc. 456, L94–L98
neutrinos and gamma rays is highly dependent density and luminosity by orders of magnitude. (2016).
45. A. Zaino et al., Mon. Not. R. Astron. Soc. 492, 3872–3884
on the astrophysical environment. NGC 1068 We have observed evidence for neutrino emis- (2020).
(8) is a nearby Seyfert II galaxy that hosts a sion from the active galaxy NGC 1068 at the 4.2s 46. A. B. Romeo, K. Fathi, Mon. Not. R. Astron. Soc. 460,
Compton-thick AGN (43–45), has vigorous level of significance, which we interpret as a 2360–2367 (2016).
47. B. Eichmann, J. B. Tjus, Astrophys. J. 821, 87 (2016).
starburst activity (46, 47) and outflows (48), signature of hadronic particle acceleration. The
48. G. Cecil, J. Bland, R. B. Tully, Astrophys. J. 355, 70 (1990).
and for these reasons has previously been dis- observation is consistent with phenomenolog- 49. R. Silberberg, M. M. Shapiro, Neutrinos as a Probe for the
cussed as a potential source of high-energy ical models (52, 53) that predicted neutrino Nature of and Processes in Active Galactic Nuclei, vol. 10 of
16th International Cosmic Ray Conference, S. Miyake,
neutrinos (49–51). Neutrino emission models emission and absorption of gamma-ray photons
N. G. Kaigi, N. B. Gakkai, Eds. (Institute for Cosmic Ray
(52, 53) have predicted the production of neu- in this source. Like the blazar TXS 0506+056 Research, 1979), p. 357.
trinos and gamma rays within the heavily (21), the active galaxy NGC 1068 is a point source 50. T. M. Yoast-Hull, J. S. Gallagher III, E. G. Zweibel, J. E. Everett,
obscured environment around the NGC 1068 of high-energy neutrinos. Astrophys. J. 780, 137 (2014).
51. A. Lamastra et al., Astron. Astrophys. 596, A68 (2016).
AGN (49, 54). In those models, the SMBH in 52. Y. Inoue, D. Khangulyan, A. Doi, Astrophys. J. Lett. 891, L33
the AGN provides the acceleration conditions, RE FERENCES AND NOTES (2020).
with x-ray photons generated through photon 1. J. Abraham et al., Nucl. Instrum. Methods Phys. Res. A 523, 53. K. Murase, S. S. Kimura, P. Mészáros, Phys. Rev. Lett. 125,
50–95 (2004). 011101 (2020).
Comptonization from the accretion disk in the
2. The Pierre Auger Collaboration, Nucl. Instrum. Methods Phys. 54. R. Silberberg, M. M. Shapiro, in Composition and Origin of
hot plasma above the disk, called the corona, Res. A 798, 172–213 (2015). Cosmic Rays, M. M. Shapiro, Ed., vol. 107 of NATO Advanced
providing the conditions for the production 3. T. Abu-Zayyad et al., Nucl. Instrum. Methods Phys. Res. A 689, Study Institute (ASI) Series C (Springer, 1983), pp. 231–244.
of neutrinos and absorption of gamma rays. 87–97 (2012). 55. Y. Inoue, D. Khangulyan, S. Inoue, A. Doi, Astrophys. J. 880, 40
4. S. R. Kelner, F. A. Aharonian, V. V. Bugayov, Phys. Rev. D 74, (2019).
Figure 4 compares our neutrino flux measure- 56. F. W. Stecker, C. Done, M. H. Salamon, P. Sommers, Phys. Rev.
034018 (2006).
ment with the model predictions (52, 53). 5. S. R. Kelner, F. A. Aharonian, Phys. Rev. D 78, 034013 (2008). Lett. 66, 2697–2700 (1991).
The evidence of neutrino emission from 6. A. De Angelis, G. Galanti, M. Roncadelli, Mon. Not. R. Astron. 57. R. Silberberg, M. M. Shapiro, C. H. Starr, in Particle Astrophysics
Soc. 432, 3245–3249 (2013). and Cosmology, M. M. Shapiro, R. Silberberg, J. P. Wefel, Eds.,
NGC 1068 suggests that AGNs could make a
7. P. Padovani et al., Astron. Astrophys. Rev. 25, 2 (2017). vol. 394 of NATO Advanced Study Institute (ASI) Series C
substantial contribution to the diffuse neu- 8. C. K. Seyfert, Astrophys. J. 97, 28 (1943). (Springer, 1993), p. 53.
trino flux. Figure 5 compares the neutrino 9. R. Antonucci, Annu. Rev. Astron. Astrophys. 31, 473–521 (1993). 58. K. Murase, E. Waxman, Phys. Rev. D 94, 103006 (2016).

59. F. Capel, D. J. Mortlock, C. Finley, Phys. Rev. D 101, 123017 J. G. Gonzalez (b, c, d, e), D. Grant (c, d, i), S. Griswold (b, d), A. Schneider (b, d, e), J. Schneider (d), F. G. Schröder (b, c, d), L.
(2020). C. Günther (g, i), P. Gutjahr (d), C. Haack (a, b, e, f, g, h), A. Hallgren Schumacher (a, b, e, f, i), G. Schwefer (g, i), S. Sclafani (b, d), D. Seckel
60. C. F. Tung et al., J. Open Source Softw. 6, 3194 (2021). (a, b, c, d), R. Halliday (b, d), L. Halve (g, i), F. Halzen (a, b, c, d, e, h, i), (a, b, c, d), S. Seunarine (d), A. Sharma (d), S. Shefali (d), M. Silva
61. G. Paturel et al., Astron. Astrophys. 412, 45–55 (2003). M. Ha Minh (e, i), K. Hanson (b, c, d, e), J. Hardin (b, d, e), A. A. Harnisch (b, d, e), B. Skrzypek (d), B. Smithers (b, d, e), R. Snihur (b, d, e),
(d, e), A. Haungs (c, d), K. Helbing (d), F. Henningsen (e, i), S. Hickford J. Soedingrekso (d, e), D. Soldin (b, c, d, e), C. Spannfellner (i),
(d), C. Hill (d), G. C. Hill (a, b, c, d), K. D. Hoffman (c, d), R. Hoffmann (d), G. M. Spiczak (d), C. Spiering (d), J. Stachurska (d), M. Stamatikos (d),
ACKN OW LEDG MEN TS K. Hoshina (b, d, e), M. Huber (a, e), T. Huber (d), K. Hultqvist (c, d), T. Stanev (a, d), R. Stein (d), J. Stettner (a, b, e, g, i), A. Steuer (d),
We dedicate this work to our recently deceased colleagues, M. Hünnefeld (d), R. Hussain (b, d, e), K. Hymon (d), S. In (d), N. Iovine R. Stokstad (a, c, d), T. Stürwald (d), T. Stuttard (b, e), G. W. Sullivan
Thomas K. Gaisser and P. Buford Price. The IceCube (d), A. Ishihara (d, e), M. Jansson (d), M. Jeong (d), M. Jin (d), (b, c, d), I. Taboada (c, d, i), S. Tilav (a, b, c, d, e, g), F. Tischbein
Collaboration acknowledges contributions to this manuscript from B. J. P. Jones (b, c, d, e), D. Kang (d), W. Kang (d), X. Kang (d), (g, i), K. Tollefson (d), C. Tönnis (d), S. Toscano (d), D. Tosi (b, d, e),
H. Niederhausen and T. Glauch. Funding: The authors gratefully A. Kappes (c, d), D. Kappesser (d), L. Kardum (d), T. Karg (d), A. Trettin (d), M. Tselengidou (d), C. F. Tung (d), A. Turcati (a, e),
acknowledge support from the following agencies and institutions. M. Karl (a, e, i), A. Karle (a, b, c, d, e, i), U. Katz (c, d), M. Kauer (b, d, e), R. Turcotte (d), J. P. Twagirayezu (d), M. A. Unland Elorrieta (d),
USA: US National Science Foundation–Office of Polar Programs, M. Kellermann (g, i), J. L. Kelley (b, c, d, e), A. Kheirandish (d, i), N. Valtonen-Mattila (b, d), J. Vandenbroucke (b, c, d, e, h, i),
US National Science Foundation–Physics Division, US National K. Kin (d), T. Kintscher (d), S. R. Klein (c, d), R. Koirala (d), H. Kolanoski N. Van Eijndhoven (c, d), D. Vannerom (d, e), J. van Santen (d),
Science Foundation–EPSCoR, Wisconsin Alumni Research (d), T. Kontrimas (a, b, e, g, h, i), L. Köpke (a, c, d), C. Kopper (b, c, e, f, i), C. Walck (d), C. Weaver (b, d, e), P. Weigel (d, e), A. Weindl (b), J. Weldert
Foundation, Center for High Throughput Computing (CHTC) at the D. J. Koskinen (d), P. Koundal (d), M. Kovacevich (d), M. Kowalski (d), (d), C. Wendt (b, c, d, e), J. Werthebach (d), M. Weyrauch (d),
University of Wisconsin–Madison, Open Science Grid (OSG), T. Kozynets (d), N. Kurahashi (b, c, d, e), N. Lad (d), C. Lagunas Gualda N. Whitehorn (c, d, e, i), C. H. Wiebusch (a, c, f, g, i), D. R. Williams
Extreme Science and Engineering Discovery Environment (XSEDE), (d), M. J. Larson (b, c, d, e), F. Lauber (d), J. P. Lazar (b, d), K. Leonard (c, d), M. Wolf (a, b, c, d, e, f, g, h), G. Wrede (d), J. Wulff (e, g),
the Frontera computing project at the Texas Advanced Computing (b, d, e), A. Leszczyńska (d), Q. R. Liu (b, d, e), E. Lohfink (d), S. Yoshida (d, e), S. Yu (d), T. Yuan (b, c, d, e), and P. Zhelnin (d).
Center, US Department of Energy–National Energy Research C. J. Lozano Mariscal (d), L. Lu (b, d, e), F. Lucarelli (a, b), A. Ludwig (e), Competing interests: There are no competing interests to
Scientific Computing Center, the Particle Astrophysics research W. Luszczak (b, d, e), Y. Lyu (d), W. Y. Ma (d), J. Madsen (c, d, e), declare. Data and materials availability: Numerical results for all
computing center at the University of Maryland, Institute for Y. Makino (b, d), S. Mancina (b, d, e), I. C. Mariş (d), I. Martinez-Soler 110 sources are listed in table S3. All data and software used in
Cyber-Enabled Research at Michigan State University, and the (d), K. Mase (e), F. McNally (d), J. V. Mead (d), K. Meagher (b, d, e), this paper, including the event list and analysis scripts, are
astroparticle physics computational facility at Marquette S. Mechbal (d), A. Medina (d, e), M. Meier (d), S. Meighen-Berger (e, i), available from the IceCube data archive at https://icecube.wisc.
University. Belgium: Funds for Scientific Research (FRS-FNRS and J. Micallef (d, e), D. Mockler (d), T. Montaruli (a, b, c, d, e, f, i), edu/science/data/neutrino-emission-from-ngc-1068. License
FWO), FWO Odysseus and Big Science programs, and Belgian R. Morse (a, b, c, d, e), M. Moulai (b, d, e), R. Naab (d), R. Nagai (d), information: Copyright © 2022 the authors, some rights reserved;
Federal Science Policy Office (Belspo). Germany: Bundesministerium U. Naumann (d), J. Necker (d), H. Niederhausen (a, b, c, d, e, f, g, h, i), exclusive licensee American Association for the Advancement of
für Bildung und Forschung (BMBF), Deutsche Forschungsgemeinschaft M. U. Nisa (d), S. C. Nowicki (d), A. Obertacke Pollmann (d), Science. No claim to original US government works. https://www.
(DFG), Helmholtz Alliance for Astroparticle Physics (HAP), Initiative M. Oehler (d), A. Olivas (b, c, d, e), E. O’Sullivan (c, d), H. Pandya (d), science.org/about/science-licenses-journal-article-reuse
and Networking Fund of the Helmholtz Association, Deutsches N. Park (a, d, i), G. K. Parker (b, d, e), E. N. Paudel (d), L. Paul
Elektronen Synchrotron (DESY), and the High Performance (b, d), C. Pérez de los Heros (a, b, c, d), L. Peters (b, g, i), S. Philippen SUPPLEMENTARY MATERIALS
Computing cluster of the RWTH Aachen. Sweden: Swedish Research (a, b, e, g, i), S. Pieper (d), A. Pizzuto (b, c, d, e), M. Plum (b, d),
science.org/doi/10.1126/science.abg3395
Council, Swedish Polar Research Secretariat, Swedish National Y. Popovych (d), M. Prado Rodriguez (b, c, d, e), C. Raab (d),
Infrastructure for Computing (SNIC), and Knut and Alice Wallenberg J. Rack-Helleis (d), K. Rawlins (b, c, d, e), A. Rehman (d), R. Reimann
Supplementary Text
Foundation. Australia: Australian Research Council. Canada: Natural (a, b, e, g, i), G. Renzi (d), E. Resconi (a, b, c, f, h), S. Reusch (d),
Figs. S1 to S30
Sciences and Engineering Research Council of Canada, Calcul Québec, W. Rhode (c, d), B. Riedel (b, c, d, e), E. J. Roberts (d), S. Robertson
Tables S1 to S4
Compute Ontario, Canada Foundation for Innovation, WestGrid, (d), G. Roellinghoff (d), M. Rongen (c, d), C. Rott (d), T. Ruhe (c, d),
IceCube Collaboration Author List
and Compute Canada. Denmark: Villum Fonden and Carlsberg D. Ryckbosch (c, d), D. Rysewyk Cantu (d), I. Safa (b, d, e), J. Saffer
References (62–98)
Foundation. New Zealand: Marsden Fund. Japan: Japan Society for (d), S. E. Sanchez Herrera (d), A. Sandrock (d, e), J. Sandroos (d),
Promotion of Science (JSPS) and Institute for Global Prominent M. Santander (c, d, i), S. Sarkar (Oxford U.) (d), K. Satalecka (d), Submitted 19 November 2021; accepted 22 September 2022
Research (IGPR) of Chiba University. Korea: National Research M. Schaufel (g, i), H. Schieler (d), S. Schindler (d), T. Schmidt (b, d, e), 10.1126/science.abg3395
Foundation of Korea (NRF). Switzerland: Swiss National Science
Foundation (SNSF). UK: the Department of Physics of the University of
Oxford. Author contributions: The IceCube Collaboration designed,
constructed, and now operates the IceCube Neutrino Observatory.
Data processing and calibration, Monte Carlo simulations of the PLANT SCIENCE
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 presented here. The manuscript was The control of carpel determinacy pathway leads to
reviewed by the entire collaboration before publication, and all authors
approved the final version. Contributions listed below are in the
following areas: (a) Conceptualization, formal analysis, and
sex determination in cucurbits
methodology; (b) Data curation and resources; (c) Funding
acquisition and project administration; (d) Investigation; Siqi Zhang1†, Feng-Quan Tan1†, Ching-Hui Chung1, Filip Slavkovic1, Ravi Sureshbhai Devani1,
(e) Software; (f) Supervision; (g) Validation; (h) Writing – original
draft and visualization; and (i) Writing – review and editing. Christelle Troadec1, Fabien Marcel1, Halima Morin1, Céline Camps1, Maria Victoria Gomez Roldan1,
Specific contributions are as follows: R. Abbasi (d), M. Ackermann Moussa Benhamed1, Catherine Dogimont2, Adnane Boualem1, Abdelhafid Bendahmane1*
(d), J. Adams (c, d), J. A. Aguilar (c, d), M. Ahlers (d), M. Ahrens
(d), J. M. Alameddine (d, e), A. A. Alves Jr. (d), N. M. Amin (d),
K. Andeen (b, c, d), G. Anton (c, d), C. Argüelles (b, d), Y. Ashida
Male and female unisexual flowers evolved from hermaphroditic ancestors, and control of flower sex is
(d), S. Axani (b, d, e), X. Bai (b, c, e), A. Balagopal V. (d), useful for plant breeding. We isolated a female-to-male sex transition mutant in melon and identified
B. Bastian (d), V. Basu (d), S. Baur (d), J. J. Beatty (d), K.-H. Becker the causal gene as the carpel identity gene CRABS CLAW (CRC). We show that the master regulator
(d), J. Becker Tjus (c), C. Bellenghi (a, b, e, g, h, i), S. BenZvi (b, c, d, i),
of sex determination in cucurbits, the transcription factor WIP1 whose expression orchestrates male
D. Berley (c), E. Bernardini (d), D. Bindig (d), E. Blaufuss (b, c, d, e, i),
S. Blot (d, i), F. Bontempo (d), J. Borowka (g, i), S. Böser (c, d), flower development, recruits the corepressor TOPLESS to the CRC promoter to suppress its expression
O. Botner (a, b, c, d, i), J. Böttcher (b, e, g, i), E. Bourbeau (d), through histone deacetylation. Impairing TOPLESS-WIP1 physical interaction leads to CRC expression,
F. Bradascio (d), J. Braun (b, d, e), J. Brostean-Kaiser (d), S. Browne (d), carpel determination, and consequently the expression of the stamina inhibitor, the aminocyclopropane-1-
A. Burgman (d), R. T. Burley (d), R. S. Busse (d), M. A. Campana (d),
E. G. Carnie-Bronca (d), C. Chen (d), D. Chirkin (b, d, e), B. A. Clark carboxylic acid synthase 7 (CmACS7), leading to female flower development. Our findings suggest that sex
(b, d, e, i), L. Classen (d), A. Coleman (b, d, e), G. H. Collin (b, d, e), genes evolved to interfere with flower meristematic function, leading to unisexual flower development.
J. M. Conrad (b, c, d), P. Coppin (d), P. Correa (d), R. Cross (b, d),
C. Dappen (b, g, i), P. Dave (d), C. De Clercq (d), J. J. DeLaunay (d),
D. Delgado López (d), H. Dembinski (b, d, e), K. Deoskar (d), A. Desai
o increase genetic diversity and fitness of unisexuality have been associated with the
T
(d), P. Desiati (b, c, d, e), K. D. de Vries (d), G. de Wasseige (d),
T. DeYoung (b, c, d), A. Diaz (b, d, e), J. C. Díaz-Vélez (b, d, e), offspring, flowering plants have developed emergence of sex determination genes (1, 2).
M. Dittmer (d), H. Dujmovic (d), M. A. DuVernois (b, d), E. Dvorak
(b, e), T. Ehrhardt (b, d), P. Eller (b, e, i), R. Engel (c), H. Erpenbeck
several systems to prevent self-fertilization.
(b, g, i), J. Evans (d), P. A. Evenson (b, d), K. L. Fan (d, e), A. Fedynitch As a result of sex determination pro- 1
Université Paris-Saclay, CNRS, INRAE, Université Evry,
(d), N. Feigl (d), S. Fiedlschuster (d), C. Finley (a, c, d, i), L. Fischer (d), cesses, unisexual male and female flowers Institute of Plant Sciences Paris-Saclay (IPS2); 91190 Gif sur
A. Franckowiak (d), E. Friedman (b, d, e), A. Fritz (d), P. Fürst
may develop on the same plant (monoecy) or Yvette, France. 2INRAE, Génétique et Amélioration des Fruits
(b, e, g, i), T. K. Gaisser (a, b, c, d), E. Ganster (b, e, g, i), A. Garcia et Légumes (GAFL), 84143 Montfavet, France.
(d), S. Garrappa (d), L. Gerhardt (d), C. Glaser (c, d), T. Glauch on separate individuals (dioecy). The evolu- *Corresponding author. Email: [email protected]
(a, b, c, d, e, f, g, h, i), T. Glüsenkamp (d), A. Goldschmidt (d), tionary pathways from hermaphroditism to †These authors contributed equally to this work.

Fig. 1. Characterization of Lib A B C

mutants. (A) Sexual phenotype
of monoecious melon (WT) and
Little Boy (Lib) androecious
mutant. Ov, ovary; Sg, stigma;
St, stamen. (B) Schematic of
sexual morphs. Female flowers
on first nodes (L1) of lateral
branches became male flowers in
Lib. (C) Cloning of the Lib locus.
DSNP index between monoecious
and androecious bulks on chro-
mosome 10. The highest DSNP
index corresponds to an E-to-K
missense mutation in the gene
CRC. (D) Alignments of CmCRC
with homologous proteins. Cs,
Cucumis sativus; Cp, Cucurbita D
pepo; At, Arabidopsis thaliana; Vv,
Vitis vinifera. Induced mutations
in melon (orange) and cucumber
(green) are shown. (E and
G) Sexual phenotype of monoe-
cious plant (WT), crc mutants
and VIGS-CmCRC (Vi-CRC)
in melon (E) and cucumber (G). E F
(F and H) Statistical analysis
of female flower ratio [females/
(females + males)], in percent, of
L1 nodes of lateral branches in
melon (F) and nodes 1 to 20 on
main stem in cucumber (H).
Bars represent means ± SD
(n = 4 to 5). Different letters
indicate significant differences G H
(P < 0.05, one-way ANOVA with
Tukey’s post hoc test). Scale
bars, 5 mm in (A), (E), and (G).
A typical hermaphroditic flower is built in membrane protein that modulates auxin ho- way in Cucurbits (13). We screened for sex
four concentric whorls: the calyx, the corolla, meostasis (9, 10). transition mutants and analyzed the causative
the androecium, and the gynoecium. Follow- In this study, we investigated how sex de- mutations in relation to the monoecy pathway
ing floral transition, these organs are pro- termination genes lead to unisexual flower in cucumber and melon.
duced from a pool of stem cells within the development in Cucurbitaceae, a large plant
floral meristem. A signaling pathway involving family containing species that display mostly Results
the homeodomain protein WUSCHEL (WUS) unisexual flowers (11). In Cucumis melo (melon) Little Boy sex transition mutant encodes for
manages the maintenance of stem cell activ- most of the flowers are male, and female Crabs Claw
ity in the floral meristem (3–5). Once floral flowers develop only on the first nodes of We conducted a forward genetic screen and
organ primordia are formed, the homeotic branches. At early developmental stages, flow- uncovered a mutant for which female flowers
gene AGAMOUS (AG) orchestrates WUS re- ers are bisexual; sex determination then oc- develop into males (Fig. 1, A and B, and fig.
pression, leading to flower meristem termina- curs through arrest of either the stamen or S1). We named the gene Little Boy (Lib) and
tion (6). This process involves the expression the carpel primordia, resulting in unisexual mapped the mutation to a single gene coding
of KNUCKLES and CRABS CLAW (CRC), down- flowers. This sexual organ arrest is geneti- for a homolog of CRC (Fig. 1C and fig. S2), a
stream genes that act synergistically through cally determined by the interactions of the transcription factor of the YABBY gene family
parallel pathways to repress WUS expression andromonoecious (M), androecious (A), and controlling carpel and nectary development
(7, 8). CRC-control of floral meristem termi- gynoecious (G) genes, which encode for CmACS-7, as well as flower meristem termination (9, 15).
nation involves the transcription of YUCCA CmACS11, and CmWIP1, respectively (12–14). We identified a nonsynonymous E131K muta-
auxin biosynthetic genes and transcriptional Interactions between the M, A, and G genes tion in the highly conserved YABBY domain
repression of TORNADO2 (TRN2), a plasma define the monoecy sex determination path- (Fig. 1, C and D). Because there is no efficient

A B A152V, or Q15* mutations were androecious—

female flowers transformed into male flowers
(Fig. 1, E to H, and fig. S5). Consistent with
this, silent conserved mutations had no ef-
fect on the sexual morph of the plant (fig. S4
and table S2). Based on the high-resolution
genetic map, the VIGS study, and the TILLING
mutants, we concluded that the Lib gene en-
codes for CRC in melon and cucumber.
WIP1 binds to the promoter of CRC to repress

its expression
C D E F The gynoecy gene WIP1 is a master regulator
orchestrating sex determination in cucurbits.
Expression of WIP1 inhibits carpel primordia
and leads to male flowers whereas its non-
expression releases the carpel from this inhi-
bition and leads to female flowers (14). In
unisexual flowers of monoecious plants, the
inappropriate sexual organs stop developing
at stage 6, shortly after the elaboration of
carpel and stamen primordia (fig. S6). Quan-
G H I J
titative real-time PCR (qRT-PCR) analysis in
monoecious melon and cucumber demon-
strates that CRC is highly expressed in female
flower buds at stage 6 and is weak in buds
determined to develop into male flowers (Fig.
2, A and B). Hence, WIP1 function is antag-
onistic to and may be the repressor of CRC
function in male flowers. We examined the
spatial expression patterns of CRC and WIP1
K using in situ hybridization. WIP1 and CRC
showed antagonistic expression patterns in
carpel primordia (Fig. 2, C to J). CRC mRNA
was detected only in carpel primordia of fe-
male flower buds at stage 6 and onward (Fig. 2,
D and F). Independently from its carpel func-
tions, CRC orchestrates nectary development
between stamen and carpel primordia. Con-
sistent with this, CRC expression was not al-
tered in nectary precursor cells (Fig. 2, C to F,
Fig. 2. CRC expression in unisexual flowers. (A and B) qRT-PCR analysis of CRC and WIP1. Bars represent
and fig. S7). To provide genetic support for
means ± SD (n = 3). ***P < 0.001 (Student’s t test). (C to F) CRC in situ expression. In male and female
the functional interaction between CRC and
flowers, CRC transcripts are detected in cells between the stamen and carpel primordia where nectary
WIP1, we generated single and double mutants
primordia arise at stage 8 (white arrows). Black arrows denote CRC transcripts detected in carpel primordia
and analyzed the expression of the two genes.
of female flowers. (G to J) WIP1 in situ expression. Black arrows denote CmWIP1 and CsWIP1 transcripts
We found that CRC is expressed when WIP1
detected in carpel primordia of male flowers. No transcripts were detected in female flowers. C, carpel
function is abolished and Cmcrc/Cmwip1 double
primordia; St, stamen primordia; P, petal; S, sepal. (K) Sexual morphs of melon plants across a combination
mutants are androecious (Fig. 2K and fig. S8).
of alleles at different sex loci. Flower buds at stage 6, from monoecious melon (Cm) and cucumber (Cs) were
Further, in gynoecious plants VIGS suppression
used in (A) and (B) as well as (C) to (J). Scale bars, 100 mm.
of CRC leads to male flower development (fig. S9
and table S1). Together these data indicate
that WIP1 acts upstream and is epistatic to
transformation protocol in cucurbits, we first from monoecious melon and cucumber and CRC in the sex determination pathway.
used virus-induced gene silencing (VIGS) to screened for mutations in Lib. Of the induced To investigate whether WIP1 is a direct re-
knock down the accumulation of Lib tran- mutations identified (35 in total), we observed pressor of CRC expression, we mapped genome-
script in monoecious plants and observed a one nonsense (Q15*) and three missense wide WIP1 DNA binding using ampDAPseq
complete female to male sexual transition in (E131K, K136E, and A152V) mutations alter- (fig. S10). WIP1-binding sites were mostly iden-
vines normally harboring female flowers (Fig. ing highly conserved amino acid positions in tified in promoter and intergenic regions (data
1, E and F, and table S1). This sexual transi- the conserved YABBY domain (Fig. 1D and S1); a consensus motif (Fig. 3A) computed from
tion correlated with the silencing of the Lib fig S4). We backcrossed the mutant lines har- these sequences corresponds to the binding
transcript (fig. S3). To further validate the boring the missense and nonsense mutations site, as previously characterized by Chromatin
role of Lib in unisexual female flower de- to the wild type and followed the segregation Immunoprecipitation sequencing (ChIP-seq) and
velopment, we used TILLING (Targeting In- of the mutations with the phenotypes. As pre- ampDAPseq for three Arabidopsis WIP pro-
duced Local Lesions IN Genomes) collections dicted, plants homozygous for E131K, K136E, teins (16, 17). In particular, we found that the

A B C
D E F
Fig. 3. CmWIP1 binds to the CRC promoter. (A) Distribution of CmWIP1 percentage of the INPUT fraction. Wild-type protoplasts and anti-IgG antibody
DAP-seq peaks and CmWIP1 binding motif. (B) CmWIP1 binding at the CmCRC were used as negative controls. (E) Promoter activity assays. Empty vector
locus. Halo-tag, negative control; -N, N terminus; -C, C terminus; P1 to P8, (35S:none), 35S:WIP1 or variants were cotransformed into melon protoplasts
promoter sequences used in EMSA, ChIP, and promoter activity assays. along with pCmCRC-P2:fLUC. Bars represent means ± SD (n = 4). (F) qRT-
G242R and S306F mutations in WIP1 DNA binding domain. (C) EMSA assay. A PCR analysis of CmCRC. Flower buds at stage 6 were used. wip1-g, a natural
5-mM GST or WIP1-GST protein and a 10-nM DNA probe were used. P1-m1 to epigenetic mutant; G242R and S306F, TILLING mutants (14). In (D) to (F), bars
P1-m4, mutant probes. GST, Glutathione S-transferase. (D) ChIP-qPCR assay in represent mean ± SD (n = 3) and different letters denote significant differences
melon protoplast transformed with 35S:WIP1-GFP. Results are expressed as a (P < 0.05, one-way ANOVA with Tukey’s post hoc test).
CmCRC promoter harbored the consensus tion of CRC through direct binding to the CRC WIP1 and TPL predicted that N1 and N2 mo-
CmWIP1 binding motif and was bound by promoter. tifs likely form a stem-loop structure recruited
CmWIP1 but not by its zinc-finger variants for the interaction with TPL (fig. S14). To test
G242R and S306F (Fig. 3B). The identified WIP1 interacts with the corepressor TPL to whether the predicted interaction involves N1
motif was also conserved across monoecious repress CRC expression and/or N2 motifs we generated Alanine scan-
Cucurbitacea species (fig. S11). To validate the To uncover the molecular mechanisms under- ning mutants. We found that the single n1 and
binding of CmWIP1 to the CmCRC promoter, lying WIP1-mediated transcriptional repression n2 mutant partially affected the interaction
we used Electrophoretic Mobility Shift Assay of CRC, we screened for WIP1 partners using whereas the n1-n2 double mutant completely
(EMSA) and band shifts were obtained only the yeast two-hybrid system (Y2H). Sequencing impaired the interaction (fig. S13C and table
when WT probes were incubated with the and annotation of 125 positive clones revealed S4). This interaction was further verified by
CmWIP1 protein (Fig. 3C and fig. S12). ChIP- that 10 encode a melon protein with 97.8% coimmunoprecipitation (Co-IP) assays using
qPCR analysis further validated the binding amino acid identity to the Arabidopsis TPL WIP-GFP and TPL-Flag constructs transiently
in vivo (Fig. 3D). We next tested the effect of protein, which we designated CmTPL (18). expressed in melon protoplasts (Fig. 4, A and
the binding of CmWIP1 to the CRC promoter Less frequently, WIP1 also interacted with two B). As in Y2H, the interactions were partial for
using a Luciferase reporter assay. We coex- other TPL-related proteins, TPR3 and TPR4 the n1 and n2 mutants and completely absent
pressed 35S:WIP1 variants and pCRC:LUC (fig. S13, A to D, and table S3). Consistent with in the n1 and -n2 double mutants. Next, we
reporter constructs in melon protoplasts. this, we found CmTPL expressed five times carried out a transient expression assay in
Expression of the reporter gene declined con- more than CmTPR3 and CmTPR4 in both male melon protoplasts to examine the effect of
siderably in the samples cotransformed with and female flowers (fig. S13E). To further ana- WIP1-TPL interaction on CRC transcription.
35S:WIP1 but less so in its variants harboring lyze the regions involved in the interaction, we We found the pCRC:LUC reporter activity sub-
mutations within the DNA binding domain tested constructs harboring different CmTPL stantially recovered when melon protoplasts
and in pCRC:LUC promoter mutant forms and CmWIP1 domains. We found CmWIP1-TPL expressed n1 or n2 mutant forms of WIP1 (Fig.
(Fig. 3E and fig. S12), suggesting that WIP1 in- physical interactions involving the N-terminal 4C), indicating that interference with WIP1-
deed represses the transcription of CRC through domain of CmWIP1 and the LisH, CTLH, and TPL interaction leads to CRC expression.
binding to its promoter. Consistent with this, CRA domains of TPL (fig. S13B). The WIP N-
WIP1-mediated repression of CRC expression is terminal domain harbors two conserved mo- Mutations in WIP1-TPL interacting domains lead
released in gynoecious lines harboring CmWIP1- tifs, N1 and N2. N1 is a Leucine-rich motif and to expression of CRC and femaleness
G242R or -S306F missense alleles (Fig. 3F), N2 fits with the canonical EAR motif (Fig. 4H). To verify the biological relevance of WIP1-TPL
validating that WIP1 represses the transcrip- A model of the three-dimensional structure of interaction in CRC repression, we first silenced

A B C
D E F G
H I J K
Fig. 4. WIP1 recruits Topless to repress CRC. (A) Co-IP assays. GFP-tagged (H) Sequence alignment of WIP1 and homologous proteins showing N1 and N2
WIP1−N or variants (asterisk) were cotransformed into melon protoplasts along motifs. Initials of species are indicated (fig. S4). LxLxL, EAR motif. Induced
with FLAG-tagged TPL-N. (B) Quantification of FLAG-tagged TPL-N from Co-IP mutations in melon (orange) and cucumber (green) are shown. (I) The female
assays in (A). (C) Promoter activity assay. Empty vector (35S:none), 35S:WIP1 or ratio [females/(females + males)], in percent, of nodes 2 to 10 of lateral
variants were cotransformed into melon protoplasts along with pCmCRC-P2: branches (melon) and nodes 1 to 20 of main stem (cucumber). qRT-PCR analysis
fLUC. Statistical analysis of female flower ratio in monoecious (WT), VIGS- of CmCRC (J) and ChIP-qPCR analysis of H3K9ac (K) at stage 6 flower buds.
CmTPL (Vi-CmTPL) and tpl mutants in melon (D) and cucumber (F). Flower CmACTIN1, control. Bars represent means ± SD (n = 3 to 5); different letters
phenotypes of tpl mutants in melon (E) and cucumber (G); white and red denote significant differences (P < 0.05, one-way ANOVA with Tukey’s post
triangles denote male and female flowers, respectively; scale bars, 5 mm. hoc test). *P < 0.05, **P < 0.01 (Student’s t test).
the expression of TPL, TPR3, and TPR4. VIGS typed 10 missense mutants. We found the four the N1 motif (L77F and L81F) and one in the
of TPR3 and TPR4 did not show any pheno- mutations (E110K, G585R, R674W, and P771S)— N2 motif (P124S) (Fig. 4H). We observed for
type alteration (table S5). By contrast, TPL- all predicted to be deleterious—to lead to female both n1 and n2 TILLING mutants the devel-
silenced lines showed flowers ranging from flower development on vines determined to de- opment of female features, such as an ovary with
partial feminization to complete male to fe- velop male flowers (Fig. 4, D to G). This sex- ovules, style, and stigma in flowers determined
male sex transition, both in melon and cu- ual transition affected 30 to 40% of total male to be male (Fig. 4I and fig. S17). We next ana-
cumber (Fig. 4, D and F, and fig. S15). This flowers, both in melon and cucumber. This lyzed the expression of CRC in WIP1 n1 and n2
sexual transition was stronger in plants in sexual transition is likely caused by gene dosage mutants. Consistent with the feminization of
which the VIGS vector targeted both TPL and as heterozygous TPL splicing mutants in cu- male flowers, CRC expression was derepressed
TPR3 (table S5). Consistent with the VIGS cumber showed enhanced femaleness (Fig. 4F). (Fig. 4J). Because TPL recruits histone deacet-
data, induced mutations in TPR3 and TPR4 The WIP1-TPL interaction implicates N1 ylase to repress the expression of target genes,
did not show any alteration of either vege- and N2 motifs of WIP1 (fig. S13C and Fig. 4, we analyzed the histone acetylation levels of
tative or flower development (table S6). By A to C). To further characterize in planta CRC in male and female flowers. ChIP assays
contrast, TPL loss-of-function mutants were their roles in sex determination, we screened with anti-H3K9ac antibodies in both male and
unable to develop flowers (fig. S16). To circum- for induced mutations in the two motifs. We female flowers revealed a considerable decrease
vent the nonflowering phenotype, we pheno- identified three missense mutations—two in in H3K9ac in the core of the gene in male

A B C
Fig. 5. WIP1 delays floral stem cell termination. (A) CmWUS in situ CmYUC4, CmYUC6, CmYUC8, and CmTRN2 in laser microdissected carpel
expression in shoot apical meristem (SAM) and flower buds at stage 6 in primordia from male or female flower buds at stage 6. Bars represent
melon. Stages 2, 4, and 7 are shown in fig. S18. Scale bars, 100 mm. CmWUS means ± SD (n = 3). *P < 0.05, **P < 0.01, ***P < 0.001 (Student’s t test).
transcripts were detected in carpel primordia of male (but not female) flowers (C) Model explaining how WIP1-TPL controls the expression of CRC to lead to
at stage 6. (B) qRT-PCR analysis of CmWUS, CmAG, CmWIP1, CmCRC, male flower development.
flowers compared with female flowers (Fig. through interference with CRC function in floral chantia, MpWIP is necessary for air pore mor-
4K). Collectively, these results support the state- meristem determinacy in the carpel primordia. phogenesis. In flowering plants, WIPs are
ment that the WIP1-TPL complex binds to the implicated in transmitting tracts, primary root
promoter of CRC to inhibit its expression, Discussion formation, and sex determination (22–25).
thereby promoting the development of male During flower development, proliferation of Based on their sequence homology and com-
flowers. stem cells is terminated in a timely manner plementation experiments, it has been sug-
to allow for carpel development. This process gested that WIP genes might show conserved
Repression of CRC by WIP1 delays floral stem involves the expression of CRC in the carpel functions across species (16). Our work reveals
cell termination in male flowers primordia, a mechanism conserved across angio- the mode of action of WIP proteins. The WIP
Flower stem cell termination is a prerequisite sperms (8). In unisexual flowers, both the male- zinc finger domain binds to a conserved cis-
for carpel development. We monitored the ex- and female-promoting genes (WIP1 and ACS7, regulatory element (Fig. 3), and the N1 and N2
pression of WUS in male and female flowers respectively), are also expressed in the carpel protein motifs cooperate to recruit the TPL
during sex transition stages through in situ hy- primordia (12–14). The integration of WIP1, family corepressors to repress gene transcrip-
bridization. Although no WUS mRNA was de- ACS7, and CRC into a genetic model of sex de- tion (Fig. 4). In conclusion, our findings reveal
tected at stage 6 in female flowers, WUS mRNA termination reveals a mechanism in which WIP1 the molecular mechanisms driving the devel-
levels were very high in carpel primordia of male interferes with carpel determinacy to lead to opment of male and female unisexual flowers
flowers at the same stage (Fig. 5A and fig. S18). male flower development whereas ACS7 is ex- in cucurbits and help us to better understand
To quantify the differential expression, we laser pressed following carpel determinacy to lead how WIP proteins regulate growth and devel-
microdissected carpel primordia of male and to female flower development (Fig. 5C). WIP1- opment in land plants.
female flowers at stage 6 and analyzed the mediated transcription repression of CRC im-
expression of WUS by qRT-PCR (fig. S19). plicates the recruitment of TPL. However, how REFERENCES AND NOTES
Consistent with the results of the in situ hybrid- ACS7 expression in the carpel is controlled is 1. D. Charlesworth, Annu. Rev. Plant Biol. 67, 397–420 (2016).
ization, WUS was expressed three times more in unclear. As in Arabidopsis, we found CRC expres- 2. G. Feng et al., Curr. Opin. Plant Biol. 54, 61–68 (2020).
3. U. Brand, J. C. Fletcher, M. Hobe, E. M. Meyerowitz, R. Simon,
that of male versus female carpel primordia sion leading to down regulation of TRN2 and
Science 289, 617–619 (2000).
(Fig. 5B). In Arabidopsis, CRC suppresses WUS the up-regulation of at least 3 YUC genes (9, 10). 4. H. Schoof et al., Cell 100, 635–644 (2000).
expression through the up-regulation of YUC4 We suggest that the generated auxin maxima in 5. T. Laux, K. F. Mayer, J. Berger, G. Jürgens, Development 122,
and the down-regulation of TRN2 (9, 10). YUC4 the carpel primordia promote the expression 87–96 (1996).
6. X. Liu et al., Plant Cell 23, 3654–3670 (2011).
belongs to a multigene protein family that en- of ACS7. Consistent with this, cucumber plants 7. T. Payne, S. D. Johnson, A. M. Koltunow, Development 131,
codes flavin-containing monooxygenases that treated with auxins show enhanced femaleness 3737–3749 (2004).
catalyze the rate-limiting step in auxin biosyn- through the ethylene pathway and this effect 8. B. Sun, Y. Xu, K.-H. Ng, T. Ito, Genes Dev. 23, 1791–1804 (2009).
9. N. Yamaguchi, J. Huang, Y. Xu, K. Tanoi, T. Ito, Nat. Commun.
thesis (9, 10). As above, we analyzed the ex- is blocked by ethylene inhibitors (19–21). 8, 1125 (2017).
pression of TRN2 and different YUCCA genes, CRC is also required for nectary develop- 10. N. Yamaguchi et al., Nat. Commun. 9, 5290 (2018).
expressed in melon flowers compared with leaf ment (15), yet its expression is not affected in male 11. A. Kocyan, L.-B. Zhang, H. Schaefer, S. S. Renner,
Mol. Phylogenet. Evol. 44, 553–577 (2007).
and root (fig. S20) in laser microdissected car- flowers. We found WIP1 and CRC expression
12. A. Boualem et al., Science 321, 836–838 (2008).
pel primordia by qRT-PCR. We found three YUC patterns overlapping in the carpel primordia but 13. A. Boualem et al., Science 350, 688–691 (2015).
genes up-regulated and TRN2 down-regulated, not in the nectary, leading to normal develop- 14. A. Martin et al., Nature 461, 1135–1138 (2009).
in female versus male flowers, respectively (Fig. ment of the nectary in male flowers (fig. S21). 15. J. L. Bowman, D. R. Smyth, Development 126, 2387–2396 (1999).
16. M. V. G. Roldan et al., Commun. Biol. 3, 239 (2020).
5B). Taken together, these results suggest that WIP proteins are limited to land plants and 17. R. C. O’Malley et al., Cell 165, 1280–1292 (2016).
CmWIP1 promotes male flower development control key developmental functions. In Mar- 18. H. Szemenyei, M. Hannon, J. A. Long, Science 319, 1384–1386 (2008).

RESEAR CH
19. C. Shen, G. Li, L. Dreni, D. Zhang, Front. Plant Sci. 12, 635500 (2021). Recherche, grant EPISEX, ANR-17-CE20-0019; Agence National de la reserved; exclusive licensee American Association for the Advancement
20. S. Shannon, M. D. de La Guardia, Nature 223, 186 (1969). Recherche, grant NECTAR, ANR-19-CE20-0023; Laboratoire d'Excellence of Science. No claim to original US government works. https://www.
21. T. Trebitsh, J. Rudich, J. Riov, Plant Growth Regul. 5, 105–113 (1987). Sciences des Plantes de Saclay, grant SPS, ANR-10-LABX-40-SPS; and sciencemag.org/about/science-licenses-journal-article-reuse
22. B. C. Crawford et al., Science 347, 655–659 (2015). the Inititiative d’Excellence Paris-Saclay, grant Lidex-3P, ANR-11-IDEX-
23. V. A. Jones, L. Dolan, Development 144, 1472–1476 (2017). 0003-02. Author contributions: Conceptualization: A.Be. and A.Bo. SUPPLEMENTARY MATERIALS
24. M. Sagasser, G.-H. Lu, K. Hahlbrock, B. Weisshaar, Genes Dev. Methodology: S.Z., F.-Q.T., C.T., C.-H.C., F.S., H.M., and C.C. Investigation:
16, 138–149 (2002). science.org/doi/10.1126/science.add4250
A.Be., S.Z., F.-Q.T., A.Bo., and C.D. Visualization: S.Z. and F.-Q.T. Funding
25. B. C. Crawford, G. Ditta, M. F. Yanofsky, Curr. Biol. 17, 1101–1108 Materials and Methods
acquisition: A.Be. Project administration: A.Be. Supervision: A.Be., A.Bo.,
(2007). Figs. S1 to S21
and C.D. Writing – original draft: A.Be., S.Z., and F.-Q.T. Writing – review
Tables S1 to S7
and editing: A.Be., S.Z., F.-Q.T., A.Bo., and C.D. Competing interests:
ACKN OW LEDG MEN TS References (26–50)
Authors declare that they have no competing interests. Data and
The authors thank S. Grant, J. Szecsi, R. Sablowski, and O. Martin for materials availability: The raw NGS data used and generated in this
Data S1
critical reading of the manuscript. We thank V. Rittener for technical study have been deposited in the SRA database under accession
assistance and IPS2 glasshouse staff and the research facilities number PRJNA847166. All other data are available in the main text or
provided by GAFL, as well as the experimental unit for plant handling. supplementary materials. The described biological material can be
Funding: This work was funded by the following: European Research obtained from A.Be. under a material transfer agreement with INRAE. Submitted 10 June 2022; accepted 5 October 2022
Council, grant ERC-SEXYPARTH, 341076; Agence National de la License information: Copyright © 2022 the authors, some rights 10.1126/science.add4250
◥ CRBN is a 50-kDa protein containing three

REPORTS folded domains: an N-terminal Lon protease-
like domain (hereafter, Lon domain), an inter-
STRUCTURAL BIOLOGY medial helical bundle (HB), and a C-terminal
thalidomide-binding domain (TBD) that har-
Molecular glue CELMoD compounds are regulators bors the drug-binding pocket. CRBN is recruited
to the Cullin-4–ROC1 ligase module by the adap-
of cereblon conformation tor protein DDB1, which consists of three WD-40
beta propeller domains (BPA, BPB, and BPC).
Edmond R. Watson1, Scott Novick2,3, Mary E. Matyskiela4†, Philip P. Chamberlain4†, The HB of CRBN docks into a central hydro-
Andres H. de la Peña4†, Jinyi Zhu4, Eileen Tran4, Patrick R. Griffin2,3, Ingrid E. Wertz4, Gabriel C. Lander1* phobic cleft formed at the interface of BPA and
BPC of DDB1, while the mobile BPB domain in-
Cereblon (CRBN) is a ubiquitin ligase (E3) substrate receptor protein co-opted by CRBN E3 ligase modulatory teracts with Cullin-4, positioning CRBN-bound
drug (CELMoD) agents that target therapeutically relevant proteins for degradation. Prior crystallographic substrates for ubiquitination. Over a dozen
studies defined the drug-binding site within CRBN’s thalidomide-binding domain (TBD), but the allostery of crystal structures of the CRBN-DDB1 complex
drug-induced neosubstrate binding remains unclear. We performed cryo–electron microscopy analyses show the Lon domain and TBD of CRBN tightly
of the DNA damage-binding protein 1 (DDB1)–CRBN apo complex and compared these structures with interacting with one another in what can be
DDB1-CRBN in the presence of CELMoD compounds alone and complexed with neosubstrates. Association described as a closed conformation (CRBNclosed)
of CELMoD compounds to the TBD is necessary and sufficient for triggering CRBN allosteric rearrangement (2, 4–10). These structures of the closed con-
from an open conformation to the canonical closed conformation. The neosubstrate Ikaros only stably former bound to small molecules have provided
associates with the closed CRBN conformation, illustrating the importance of allostery for CELMoD compound the basis for structure-guided drug design of
efficacy and informing structure-guided design strategies to improve therapeutic efficacy. next-generation CELMoD agents with en-
hanced neosubstrate specificity and potency,
such as CC-92480 (mezigdomide). Although it
ukaryotic proteins are targeted for degra- finger protein ROC1 (fig. S1A). CRBN is also the is thought that association of the TBD and Lon
E dation through covalent attachment of

ubiquitin moieties to specific residues, pri-
marily lysine side chains (1). Ubiquitination
is facilitated by ubiquitin ligases (E3s),
which coordinate and position targeted sub-
strates for ubiquitin ligation, often through
cellular receptor for a class of drugs known as
CELMoD (CRBN E3 ligase modulatory drug)
agents, which bind to a conserved hydrophobic
pocket in CRBN to create a molecular surface
capable of recruiting neosubstrates, that is, cel-
lular proteins that are targeted only in response
domains is concomitant with drug and neo-
substrate binding, this assumption is chal-
lenged by structures of the isolated TBD bound
to ligands, demonstrating that the TBD is com-
petent for drug binding in the absence of the
Lon domain (11).
adaptor proteins and interchangeable recep- to drug. Such ligand-induced CRL4-CRBN re- Two crystal structures have captured the
tor modules. Substrate specificity of one such cruitment facilitates neosubstrate ubiquitination Lon domain and TBD in an open conformation
E3 complex, the Cullin-4 RING ligase (CRL4), and subsequent proteasomal degradation, (CRBNopen) where the two domains are sep-
is mediated by the cereblon (CRBN) substrate thereby dictating therapeutic efficacy (2). The arated and positioned at a ~45° angle relative
receptor. CRBN facilitates the transfer of ubiquitin CELMoD agent lenalidomide (Revlimid) has to one another. These reports raise questions
to substrate proteins by reversibly interacting been used as a first-line therapy for multiple about the assumed constraints associated with
with the CRL4 core complex, which consists of myeloma and other hematological malignancies drug accessibility and binding, as well as the
the Cullin-4 scaffold, the DNA damage-binding for more than a decade, and next-generation mechanisms by which ligands modify the sur-
protein 1 (DDB1) adaptor protein, and RING CELMoD compounds markedly improve patient face chemistry of CRBN to enable neosubstrate
outcomes in clinical trials (3). Although crystal- recruitment and subsequent ubiquitination and
lographic structural snapshots of CRBN-DDB1 degradation (4, 5). If CRBNopen represents a
1
Department of Integrative Structural and Computational bound to a variety of drugs and neosubstrates physiological conformer (i.e., not an artifact
Biology, Scripps Research, La Jolla, CA 92037, USA.
2
Department of Molecular Medicine, Scripps Research, have facilitated a molecular description of the of crystallization), we must reconsider re-
Jupiter, FL 33458, USA. 3UF Scripps Biomedical Research, stable quaternary complex, the allostery asso- cruitment of drug and neosubstrates in the
University of Florida, Jupiter, FL 33458, USA. 4Bristol-Myers ciated with substrate targeting remains un- context of these two conformers. Additionally,
Squibb, San Diego, CA 92121, USA.
†Present address: Neomorph, Inc., San Diego, CA 92121, USA. known, and the structural features of the the dynamics of interconversion between the
*Corresponding author. Email: [email protected] unliganded complex have yet to be described. open and closed conformers, the effect that

RES EARCH | R E P O R T S
neosubstrates in prior structures (2, 12), at its

A Apo CRBN-DDB1 TBD B
HB
canonical position adjacent to the thalidomide-
TBD
Lon binding pocket. Instead, we observe strong,
CRBN
Sensor
low-resolution density corresponding to the
loop sensor loop engaged in previously unobserved
interactions with a helix in CRBN’s HB (res-
idues ~210 to 220) and a loop of BPC in DDB1
(residues ~776 to 780) (Fig. 1B). These regions
BPC
HB
of the HB and DDB1 are usually disordered in
DDB1
crystal structures (5, 10, 12, 13), consistent with
BPA Sensor loop the flexibility of this region observed in our
cryo-EM density. However, the clearly observed
DDB1 association between the sensor loop and this
BPB HB-DDB1 interface in our cryo-EM structures
suggests that interaction between these flexi-
ble loops plays a role in maintaining the TBD
C CRBN-DDB1∆BPB + Pomalidomide D in the CRBNopen conformer.
Sensor loop Pomalidomide
To better understand how CELMoD agents
Sensor loop
Lon
TBD interact with and influence this open CRBN
CRBN Pomalidomide
conformation, we added saturating amounts
N-terminal N-terminal of different compounds to CRBN-DDB1 for
belt
belt cryo-EM analyses. Addition of pomalidomide
HB was sufficient to induce conformational rear-
rangement within CRBN for only ~20% of the
DDB1 BPC particles. The ~3.9-Å-resolution structure of
∆BPB BPA CRBNclosed arising from pomalidomide-bound
particles shows the ~15-kDa TBD positioned
adjacent to the Lon domain and unambiguous
density within the ligand-binding pocket con-
sistent with pomalidomide association (Fig. 1C).
Necessarily, the sensor loop in CRBNclosed is
Fig. 1. CRBNopen is allosterically modulated to CRBNclosed by pomalidomide. (A) ~3.5-Å-resolution
observed in the canonical b-hairpin fold inter-
cryo-EM reconstruction of CRBN-DDB1 in the unliganded apo form. The Lon domain (tan) is separated from
facing the TBD and Lon domain, and a portion
the TBD (red), while the helical bundle (HB, light red) mediates interaction with DDB1, composed of beta
of the CRBN N terminus (residues 48 to 63)
propeller A (BPA, light blue), beta propeller B (BPB, medium blue), and beta propeller C (BPC, steel blue).
that is disordered in the CRBNopen becomes
CRBN and DDB1 are thresholded independently to illustrate important features. To the right, a ribbon
ordered, extending from the Lon domain, truss-
representation of the CRBN-DDB1 complex modeled from density is shown (same color scheme as on the
ing the TBD, and supporting the closed con-
left). The sensor loop is denoted as a dotted magenta line extending away from the TBD domain. (B) The
formation (Fig. 1C). To better define the role of
unsharpened cryo-EM map shows the path of the sensor loop, which interacts with the HB and BPA of
these N-terminal residues, hereafter referred to
DDB1. (C) Surface representation of the ~3.9-Å-resolution cryo-EM reconstruction of CRBN-DDB1 in the
as the N-terminal belt, in CRBN closure, we
closed form in complex with pomalidomide. The N-terminal belt (orange) wraps around the TBD domain.
truncated 63 residues from the CRBN N terminus
To the right, a ribbon representation of CRBN-DDB1DBPB in complex with pomalidomide is shown, with the
(CRBNDNTD) and incubated the CRBNDNTD-DDB1
sensor loop (magenta) adopting a b-hairpin organization that is tightly packed between the TBD and
with saturating pomalidomide concentrations.
Lon domains. (D) A detailed view of the density corresponding to pomalidomide (green) and the sensor
Deletion of the N terminus all but abolishes
loop (magenta) in CRBNclosed.
CRBN closure, with less than ~2% of the com-
plexes adopting a closed conformation with a
ligands and neosubstrates have on this process, and distribution in ice (see methods in the sup- poorly ordered sensor loop, underscoring the
and how these dynamics affect neosubstrate plementary materials). Unexpectedly, whether importance of the CRBN N-terminal belt in
positioning and subsequent degradation copurified with either full-length DDB1 or stabilizing the CRBNclosed conformer (fig. S2A).
may have profound effects on the design, and DDB1 lacking BPB (CRBN-DDB1DBPB, a con- We expected to observe the pomalidomide li-
ultimately the therapeutic impact, of next- struct commonly used for x-ray crystallogra- gand in the binding pocket of only the CRBNclosed
generation CELMoD compounds. phy), apo-CRBN exclusively adopts an open conformer, but we were surprised to find strong
To better understand the spectrum of CRBN conformation, with the TBD and Lon do- density corresponding to pomalidomide with-
conformations, we used cryo–electron micros- mains separated from one another (Fig. 1A in the binding pocket of CRBNopen as well
copy (cryo-EM) to examine the population en- and fig. S1B). (fig. S3A). Even with the ligand bound—and
semble of CRBN-DDB1 conformers present Our ~3.5-Å-resolution cryo-EM structure con- in stark contrast to prior crystal structures of
in solution under various conditions. To over- firms the physiological existence of the open the CRBNopen conformer—the sensor loop of
come preferred orientation and partial dena- conformer and provides the first opportunity our cryo-EM CRBNopen conformer remains en-
turation or dissociation of the complex during to examine this CRBN conformation in the gaged in interactions with the CRBN HB and
cryo-EM grid preparation, a combination of absence of ligand or substrate. Notably, we do DDB1 as we observe in the apo-CRBN conformer
grid pretreatment with a CRBN-agnostic pro- not observe density for the “sensor loop,” a (fig. S3B). These results indicate that limita-
tein, incorporation of mild amphiphilic deter- beta-insert hairpin within CRBN’s TBD (resi- tions in neosubstrate recruitment, which di-
gent, and sample dilution immediately before dues ~341 to 361) that has been shown to directly translate to drug efficacy, may not solely
plunging improved particle homogeneity rectly bind immunomodulatory drugs and correlate with CRBN-binding kinetics. Rather,

A Pomalidomide C CRBN-DDB1 + Iberdomide

80% open 20% closed
+ Ikaros ZF 1-2-3
+ Pomalidomide Sensor
Loop
PDB:4ci3
Iberdomide
+ Iberdomide
50% open 50% closed Sensor

Loop
PDB:5v3o
B HDX solvation difference map upon Iberdomide addition
Lon TBD Lon
TBD
HB
DDB1 DDB1
open
CRBN CRBNclosed
Fig. 2. Improved CELMoD compounds drive CRBNclosed and recruit Ikaros. (B) Colored space-filling representation of CRBN models with residue-specific
(A) The CRBNopen transition to CRBNclosed is differentially regulated between coloring according to changes in solvency upon addition of iberdomide as
pomalidomide (~20% of particles adopt CRBNclosed) and the CELMoD agent detected by HDX-MS. Orange residues correspond to peptides with mild HDX
iberdomide (~50% of particles adopt CRBNclosed). (Top right) The ~3.9-Å-resolution differential, and red residues correspond to peptides with large HDX differential.
cryo-EM structure of pomalidomide-induced CRBNclosed. (Bottom right) The Mixed conformations are expected during the experiment, so CRBNopen and
~3.7-Å-resolution reconstruction of iberdomide-induced CRBNclosed. The inset CRBNclosed are both shown. (C) The ~3.6-Å-resolution reconstruction of
panels on the far right depict surface representations of published crystal iberdomide-induced CRBNclosed in complex with Ikaros ZF1-ZF2-ZF3 (gold). DDB1
structures illustrating the ligand (green) interactions with sensor loop (magenta). and CRBN are thresholded independently to illustrate important features.
conversion of the CRBNopen to the CRBNclosed molecule can arch over the top of the sensor mation, consistent with transition from CRBNopen
conformation may be a key mechanistic bottle- loop b-hairpin (13, 14), a conformation that may to CRBNclosed (Fig. 2B and fig. S4A), whereas
neck in the allosteric transformation of apo- aid in the stabilization of this structural motif. the DDB1 surfaces were unchanged (fig. S4B).
CRBN to a neosubstrate-binding conformation These data support our proposed allosteric As expected, the most extreme change is ob-
after ligand association. model whereby enhanced ligand interaction served for the sensor loop itself, consistent with
We were intrigued by the observation that with the sensor loop stabilizes the b-hairpin conformation of the buried hairpin we observe in
saturating concentrations of pomalidomide formation, which promotes the closure of the CRBNclosed (Fig. 1C). These findings offer a po-
did not induce widespread closure of the CRBN TBD and Lon domains and ultimately trans- tential new modality for drug development,
domains, despite high occupancy. We rational- lates to improved neosubstrate ubiquitination wherein properties influencing not only bind-
ized that next-generation CELMoD agents with and subsequent degradation efficacy. We note, ing kinetics for CRBN and neosubstrates (15)
improved binding affinity may also have a however, that these extended moieties are lim- but also the capacity to stabilize the b-hairpin
more pronounced allosteric effect through en- ited in resolution in our cryo-EM structure conformation of the sensor loop as a means of
hanced interactions with the sensor loop, con- (fig. S3C) and have high B-factors in published both initiating and maintaining the closed con-
tributing to adoption of the sensor loop to the x-ray structures. This observation, coupled with formation should be considered. The CRBNclosed
CRBNclosed–competent b-hairpin conformation. recently described alternative positions of these conformation may in turn be a prerequisite for
Indeed, an analysis of CRBN-DDB1 in the pres- extended ligand groups (14), suggests that substrate recruitment.
ence of CC-220 (iberdomide), a molecule with iberdomide’s interaction with the sensor loop Indeed, the substantial population of ligand-
~20-fold improved affinity and ~24-fold enhanced may be dynamic. It is likely that a more stable bound CRBNopen conformers in our datasets
Ikaros degradation (13), shows that nearly 50% interaction with the sensor loop could be at- indicates that ligand binding is not concurrent
of particles have shifted to the CRBNclosed tained with next-generation CELMoD agents. with adoption of the CRBNclosed conformation
conformer (Fig. 2A). As with pomalidomide, To further investigate the allosteric influ- and that CRBN closure is dependent on sub-
in the presence of excess ligand, we observe ence of iberdomide on CRBN closure, we used sequent allosteric triggers. We posit that ligand
density consistent with iberdomide within hydrogen-deuterium exchange mass spectrom- binding is involved in inducing the sensor
the glutarimide-binding pocket of both the etry (HDX-MS) to profile changes in solvent loop’s adoption of the b-hairpin arrangement,
CRBNclosed and CRBNopen conformers, suggest- accessibility of CRBN-DDB1 peptides in the thereby untethering the TBD from the CRBN
ing that drug occupancy within CRBN is not re- presence of iberdomide. Compared with unli- HB to enable the CRBNclosed conformation for
sponsible for the increased allosteric transition. ganded CRBN-DDB1, addition of drug sub- substrate interaction. Despite extensive focused
Two prior crystal structures of the iberdomide- stantially reduced solvation for regions within classification of the tether region, we were
bound CRBN show that the additional phenyl the Lon, TBD, and HB domains found at the unable to identify a population of liganded
and morpholino moieties of the iberdomide domain-domain interface of the closed confor- CRBNopen conformers where the sensor loop is

Fig. 3. DDB1 and next-generation

A
CELMoD agents further poise
CRBN substrates for ubiquitination
in disease contexts. (A) Three-
dimensional classification of
unliganded CRBN-DDB1 yields
three discrete positions of DDB1’s 4CI1 5HXB 2HYE
mobile BPB propeller (colored Linear Hinged Twisted Linear Hinged Twisted
blue). (Left) ~3.5-Å-resolution Linear Hinged Twisted
cryo-EM reconstructions of C
B Ikaros ZF1-2-3
CRBN-DDB1 with BPB in a linear
(left), hinged (middle), or twisted CC-92480
(right) position. (Right) Structural
models in similar orientations for
reference, adopting a linear (PDB
ID 4CI1, left), hinged (PDB ID
5HXB, middle), or twisted (PDB ID 180° 180° 180°
2HYE, right) position. (B) The

~3.2-Å-resolution cryo-EM recon- CC-92480 bridges TBD~Lon
structions of CRBN-DDB1 F150
complexed with mezigdomide with CC-92480
a linear (left), hinged (middle), or

F102
twisted (right) position of BPB.
(Inset) ~3.1-Å-resolution focused
refinement of particles from all three
orientations reveal a connection
between mezigdomide and the Lon D
domain, “stapling” the CRBNclosed 1 2 3 4 1 2 3 4
conformation. (C) The ~3.1-Å- 1) DMSO alone (control)
resolution reconstruction of MBP-Ikaros Ubiquitination 2) Pomalidomide
mezigdomide-induced CRBNclosed 3) CC-220 Iberdomide
bound to Ikaros ZF1-ZF2-ZF3. The MBP-Ikaros
4) CC-92480 Mezigdomide
inset shows a focused view of 30 min 60 min
connection between mezigdomide
E
and the Lon domain, stapling CELMoD Ikaros ZF1-2-3
closed TBD interacts with Lon
CRBN in the presence of Ikaros. Lon
TBD
(D) In vitro ubiquitination assay
tracking MBP-Ikaros ubiquitination in
Sensor
the presence of different molecules Sensor loop
loop up
down
used in this study (labeled on the
right). (E) Mechanistic model
illustrating pathway of assembly.
First, unliganded CRBN is open with
the sensor loop attached to HB.
Second, ligand is added and asso-
ciates with TBD in the hydrophobic OPEN OPEN OPEN CLOSED CLOSED
pocket. Third, binding of ligand in Sensor loop Drug binding Sensor loop adopts TBD & Lon close, Substrate binding
the tri-Trp pocket stabilizes the contacts HB and BPC in tri-Trp pocket β-hairpin conformer stabilized by
N-terminal belt
sensor loop as a b-hairpin detached
from the HB. Fourth, sensor loop
refolding promotes transition of CRBNopen to CRBNclosed without an observed intermediate, and the N-terminal belt is seated (orange). Fifth, substrate is recruited to the
CRBNclosed conformation for subsequent ubiquitination by CRL4.
untethered from the HB in any of our datasets. bound to neosubstrates (4, 9), although it was CRBN-DDB1 in the presence of pomalidomide
This finding indicates that ligand-mediated speculated that these conformations may have or iberdomide. Regardless of drug identity or
detachment of the sensor loop from the HB is resulted from crystallization conditions and Ikaros ZF construct design, our single-particle
concomitant with adoption of the structured lattice contacts. To investigate drug-mediated analyses reveal density consistent with the lo-
b-hairpin and triggers immediate closure of the CRBN-DDB1 recruitment of neosubstrates, we cation and positioning of a single Ikaros ZF
CRBN TBD and Lon domains. used the zinc finger (ZF) transcription factor motif associated with the TBD domain of the
We next aimed to probe the role of drug bind- Ikaros, a cellular target of pomalidomide and CRBNclosed conformer (Fig. 2C and fig. S5B),
ing in neosubstrate recruitment and positioning iberdomide. We generated different constructs again confirmed in HDX-MS analyses showing
in the context of the CRBNopen and CRBNclosed containing tandem Ikaros ZF1, ZF2, and ZF3 do- even more pronounced changes for residues
conformers. Prior crystallographic studies mains (fig. S5A) for increased recruitment effi- within the sensor loop that contact the known
yielded structures of CRBNopen conformers ciency (4) and incubated each individually with Ikaros recruitment motif (fig. S4, C and D). We

were unable to detect any Ikaros association molecules. Mezigdomide, for example, was re- directed molecular glue therapeutics (Fig. 3E).
with the CRBNopen conformer, consistent with a cently shown to display efficient, rapid neosub- By characterizing the conformational rearrange-
mechanism whereby drug-induced CRBN clos- strate degradation kinetics and is considerably ments inherent to the CRBN-DDB1 system,
ing occurs before substrate recruitment. This more efficacious in the treatment of relapse and showing how three distinct degrader mol-
result further demonstrates the importance of or refractory patients who no longer respond ecules affect allostery and neosubstrate-binding
an allosteric control mechanism, whereby drug- to primary treatment with lenalidomide and/or capacity, we reveal how conformational con-
induced CRBN closing occurs prior to neosub- pomalidomide (18). To investigate the structural trol of the mobile drug-binding TBD within
strate recruitment. features responsible for the enhanced efficacy CRBN has cryptically driven the therapeutic
The Cullin-4–ROC1 ligase core recruits the of this compound, we incubated mezigdomide success of neosubstrate-targeting agents.
DDB1-CRBNclosed-Ikaros complex by associat- with CRBN-DDB1 for cryo-EM analyses. One
ing with the second beta propeller of DDB1 hundred percent of the complexes adopted the
1. K. N. Swatek, D. Komander, Cell Res. 26, 399–422 (2016).
(called BPB), which is located distally to CRBN. CRBNclosed conformer. Density correspond- 2. P. P. Chamberlain et al., Nat. Struct. Mol. Biol. 21, 803–809 (2014).
BPB extends from the DDB1 core as a mobile ing to mezigdomide is clearly observed in the 3. P. G. Richardson et al., Blood 138 (suppl. 1), 2731 (2021).
element, and this flexibility has been implicated canonical binding site in the ~3.2-Å-resolution 4. Q. L. Sievers et al., Science 362, eaat0572 (2018).
5. G. Petzold, E. S. Fischer, N. H. Thomä, Nature 532, 127–130 (2016).
in the mechanism of DDB1 as a promiscuous structures of CRBN-DDB1 with BPB in the 6. M. E. Matyskiela et al., Nat. Struct. Mol. Biol. 27, 319–322 (2020).
adaptor that bridges Cullin-4 with structural- three discrete orientations (Fig. 3B). We also 7. C. Surka et al., Blood 137, 661–677 (2021).
ly diverse substrate-binding modules. Corre- observe protruding drug density that extends 8. E. S. Fischer et al., Nature 512, 49–53 (2014).
9. R. P. Nowak et al., Nat. Chem. Biol. 14, 706–714 (2018).
spondingly, BPB has been captured in a variety to the distal Lon domain, a feature that had 10. E. S. Wang et al., Nat. Chem. Biol. 17, 711–717 (2021).
of conformations in several prior studies (16). not been observed in the structures of any other 11. M. D. Hartmann, I. Boichenko, M. Coles, A. N. Lupas,
We sought to describe the motion of BPB and liganded CRBN (Fig. 3B, right). We were able to B. Hernandez Alvarez, PLOS ONE 10, e0128342 (2015).
12. M. E. Matyskiela et al., Nature 535, 252–257 (2016).
its potential relevance to Ikaros positioning for unambiguously fit mezigdomide into the den- 13. M. E. Matyskiela et al., J. Med. Chem. 61, 535–542 (2018).
ubiquitination in the context of the CRL4- sity to reveal that the benzonitrile moiety is 14. C. Heim, M. D. Hartmann, Acta Crystallogr. D Struct. Biol. 78,
CRBN complex assembly. Our image analy- likely stabilized by aromatic or cation-pi inter- 290–298 (2022).
ses of the BPB domain within our apo CRBN actions with CRBN Phe102 and Phe150 of the 15. N. R. Kong, H. Liu, J. Che, L. H. Jones, ACS Med. Chem. Lett.
12, 1861–1865 (2021).
data reveal that, rather than demonstrating Lon domain. Association of the glutarimide 16. N. Bai et al., J. Biol. Chem. 298, 101653 (2022).
a full continuum of motion, the BPB position and the benzonitrile moieties to distinct do- 17. S. Banchenko et al., PLOS Pathog. 17, e1009775 (2021).
is limited to three metastable positions. Ap- mains within CRBN, that is, the TBD and Lon 18. J. D. Hansen et al., J. Med. Chem. 63, 6648–6676 (2020).
proximately 65% of the particles adopt a “linear” domains, respectively, both provides avidity AC KNOWLED GME NTS
orientation, in which the broad face of BPB is for improved ligand affinity and serves to sta- We thank J. C. Ducom at Scripps Research High Performance
in a similar orientation as BPC (Fig. 3A, left). A ple the domains together in the closed confor- Computing and C. Bowman at Scripps Research for computational
~70° rotated “hinged” orientation, where the mation, which is traditionally accomplished support, as well as B. Anderson at the Scripps Research Electron
Microscopy Facility for microscopy support. Computational
BPB face more closely matches BPA, is the sec- by the CRBN N-terminal belt (Fig. 1C). Ikaros is analyses of EM data were performed using shared instrumentation
ond major class observed, with ~20 to 30% of recruited to mezigdomide-bound CRBN-DDB1 funded by NIH S10OD021634 to G.C.L. Funding: This work was
particles (Fig. 3A, middle), and only a minor (Fig. 3C), and its ubiquitination efficiency is supported by National Institutes of Health grant S10OD021634
(G.C.L.). Author contributions: Conceptualization: E.R.W., M.E.M.,
population contain BPB in the third, “twisted” enhanced in assays with Cullin-4–ROC1 (Fig. 3D). P.P.C., and G.C.L. Methodology: E.R.W., S.N., M.E.M., P.P.C.,
orientation, rotated ~140° from the linear ori- Within the mezigdomide-bound dataset, we A.H.d.l.P., J.Z., E.T., P.R.G., I.E.W., and G.C.L. Investigation: E.R.W.,
entation (Fig. 3A, right). Further classification observe a substantial population of CRBNclosed J.Z., and G.C.L. Visualization: E.R.W., S.N., and J.Z. Project
administration: P.P.C., I.E.W., and G.C.L. Supervision: P.P.C., I.E.W.,
of particles contributing to each of these states particles that lack discernible density for the and G.C.L. Writing – original draft: E.R.W. and G.C.L. Writing –
reveals only minor positional variability for BPB, stabilizing N-terminal belt (fig. S2B). We thus review & editing: E.R.W., I.E.W., and G.C.L. Competing interests:
as has been previously described (16), centered propose that there is functional redundancy M.E.M., P.P.C., A.H.d.l.P., J.Z., E.T., and I.E.W. are or were
employees of Bristol-Myers Squibb and have received Bristol-
around these discrete conformations. in securing the closed conformer and that Myers Squibb stock. G.C.L. has received research funding from
These three principal orientations were ob- mezigdomide further promotes neosubstrate Bristol-Myers Squibb. A provisional patent regarding aspects of this
served in recent low-resolution cryo-EM structures recruitment and subsequent ubiquitination work has been filed under US Provisional Patent Application no.
63/293,402. E.R.W. has received consulting fees for service rendered
of DDB1-DCAF1 (DDB1 and Cullin-4 associated by prolonging the mechanistic cycle of closure,
in cryo-EM method applications. P.P.C. is a founder, employee, and
factor 1) bound to Cullin-4 (17), and our struc- neosubstrate binding, N-terminal engagement, equity holder of Neomorph Inc. M.E.M. and A.H.d.l.P. are employees
tures demonstrate that the BPB likely switches and stepwise reversal to the open conforma- and stockholders of Neomorph Inc. Data and materials availability:
between these three energetic minima irrespec- tion. These data elucidate a distinctive, previ- Atomic models have been deposited to the Protein Data Bank
(PDB) under IDs 8CVP, 8D7U, 8D7V, 8D7W, 8D7X, 8D7Y, 8D7Z, 8D80,
tive of the substrate adaptor module or interac- ously unexplored compound design modality and 8D81, as described in table S2. Electron microscopy reconstructions
tion with Cullin. Although we were unable to and describe the structural mechanism for en- have been deposited to the Electron Microscopy Databank (EMDB)
discern allosteric coordination upon ligand or hanced therapeutic benefit of mezigdomide in under accession codes EMDB-27012, EMDB-27234, EMDB-27235,
EMDB-27236, EMDB-27237, EMDB-27238, EMDB-27239, EMDB-27240,
Ikaros binding, the population distribution of patients unresponsive to pomalidomide, fur- EMDB-27241, and EMDB-27242, as described in table S1. License
the states varies greatly compared with those ther illustrating the therapeutic value of allo- information: Copyright © 2022 the authors, some rights reserved;
observed for the Cullin-bound DDB1 (17). Nota- steric control. Although not tested directly, we exclusive licensee American Association for the Advancement of
Science. No claim to original US government works. https://www.
bly, the twisted orientation that we observe in speculate that incorporation of these improved science.org/about/science-licenses-journal-article-reuse
<10% of our particles is the majority conformer design elements to other molecular glues and
when bound to Cullin (17), suggesting that as- heterobifunctional molecules would similarly SUPPLEMENTARY MATERIALS
sociation with Cullin may substantially modu- improve their efficacy. science.org/doi/10.1126/science.add7574
late BPB positioning between these three states, Our results provide mechanistic insights into Figs. S1 to S9
thereby influencing proximity and positioning the therapeutic efficacy of CELMoD compounds, Tables S1 to S3
of neosubstrates for ubiquitination. a class of molecules that epitomize molecu- References (19–32)
Given the potential clinical relevance of the lar glues and are central players in the field of
observed CRBN allostery, we considered the targeted protein degradation. In so doing, we
mechanisms of action responsible for the im- highlight previously unappreciated allosteric Submitted 1 July 2022; accepted 5 October 2022
proved efficacy of next-generation CELMoD effects for consideration in the design of CRBN- 10.1126/science.add7574

VOLCANIC PLUMES 20 December 2021 to 15 January 2022 (17). In

the lead-up to the 15 January 2022 eruption,
The January 2022 eruption of Hunga Tonga-Hunga described later in the text, a large phreatomag-
matic eruption, similar to the December 2019
Ha’apai volcano reached the mesosphere eruption of Anak Krakatau, Indonesia (18),
was observed on 13 January in geostationary
Simon R. Proud1,2*, Andrew T. Prata2, Simeon Schmauß3 satellite data. The volcanological setting of
HTHH is very different from those of the
Explosive volcanic eruptions can loft ash, gases, and water into the stratosphere, which affects both Mount Pinatubo and El Chichón volcanoes
human activities and the climate. Using geostationary satellite images of the 15 January 2022 because of the abundance of seawater avail-
eruption of the Hunga Tonga-Hunga Ha’apai volcano, we find that the volcanic plume produced by able for magma interaction. Initial estimates
this volcano reached an altitude of 57 kilometers at its highest extent. This places the plume in the of the total mass of SO 2 released by the
lower mesosphere and provides observational evidence of a volcanic eruption injecting material 15 January eruption from hyperspectral
through the stratosphere and directly into the mesosphere. We then discuss potential implications sounders and ultraviolet instruments were
of this injection and suggest that the altitude reached by plumes from previous eruptions, such as 0.2 to 0.4 Tg SO2 (19, 20). This amount of
the eruption of Mount Pinatubo in 1991, may have been underestimated because of a lack of SO2 is low and suggests that large amounts
observational data. of SO2 may have been scavenged through
wet deposition (18, 21).
On 15 January 2022, at about 04:06 UTC
arge explosive volcanic eruptions are im- troposphere–lower stratosphere (11); however, (17:06 local time), the HTHH volcano violently
L portant because they can affect climate

(1), disrupt aviation (2), and pose numer-
ous hazards to communities living near-
by (3, 4). The degree to which explosive
volcanic eruptions affect climate is largely de-
pendent on the volcano’s latitude, the plume
few have been observed to reach higher than
30 km and affect the climate (12). Notable ex-
amples that have been observed and quanti-
fied using satellite observations include the
plumes of Mount Pinatubo (Philippines) in 1991,
which injected ∼20 Tg of SO2 (13) and reached
erupted, producing the large volcanic plume
shown in Fig. 1. A second smaller eruption
occurred at 08:00 UTC, with no further large
eruptions occurring thereafter. This eruption
was one of the most powerful in recent years,
triggering a tsunami that was felt across the
height, and the amount of SO2 gas that is re- 40 km at its highest point (14), and El Chichón, Pacific and atmospheric waves (22) that cir-
leased (5). When a substantial amount of SO2 which released ∼7.5 Tg of SO2 into the atmo- cled Earth multiple times, which were seen
is released into the atmosphere, it is converted sphere (15) in 1982, reaching 31 km (16). Based as fluctuations in global pressure sensor read-
to sulfate aerosols that can, in the most exon the satellite data record, in general, indi- ings for several days. The volcano’s location is
treme cases, persist for several years (6), and vidual explosive volcanic eruptions are not ex- well covered by satellite sensors, with three geo-
when injected at low latitudes, the aerosol pected to have a measurable climate impact stationary weather satellite platforms provid-
disperses into both hemispheres (1, 7). The unless >1 Tg SO2 is released into the strato- ing imagery of the area at visible and infrared
persistent aerosol veil that is produced affects sphere (12).
climate because of the reflection of incoming The Hunga Tonga-Hunga Ha’apai (HTHH) 1
National Centre for Earth Observation, RAL Space, STFC
visible radiation and absorption of near-infrared volcano (20.536°S, 175.382°W) is an under- Rutherford Appleton Laboratory, Harwell OX11, UK.
2
radiation (8), which results in a cooling of the water caldera volcano located ~70 km north- Atmospheric, Oceanic and Planetary Physics, University of
troposphere (9) and heating of the stratosphere northwest of Tonga’s capital, Nukualofa. Oxford, Parks Road, Oxford OX1 3PU, UK. 3Munich University
of Applied Sciences, Lothstraße. 34, 80335 Munich,
(10). Within the satellite era, there are numerous Recent Surtseyan-style eruptive activity was Germany.
examples of volcanic plumes reaching the upper observed in 2009, from 2014 to 2015, and from *Corresponding author. Email: [email protected]
Fig. 1. The HTHH eruption as viewed from the Himawari-8 weather cloud at 34 km. (C) At 05:40 UTC, as the volcanic umbrella spreads
satellite. (A) At 04:10 UTC on 15 January 2022, ~10 min after the eruption south-westward and the Sun begins to set, emphasizing the shadows that
began. (B) At 04:50 UTC, after the initial dome collapsed, leaving remnants we used to calculate plume altitude and highlighting wave structure in
at 55- to 58-km altitude that cast a shadow (to the right) onto the umbrella the umbrella top.

by satellite and compared with a penetrated the stratosphere but also reached
vertical profile of temperature from the lower mesosphere.
radiosonde measurements or weather When a satellite views a high-altitude cloud,
model outputs such as those produced the cloud location within the satellite image
by the European Centre for Medium- will be incorrect because of a parallax shift, the
Range Weather Forecasts (ECMWF) magnitude of which depends on the viewing
(23). This technique makes use of the angle between satellite and cloud and upon
relationship between decreasing at- the cloud altitude. If two or more satellites view
mospheric temperature—and hence a cloud from differing locations, then its actual
cloud temperature for a cloud in position and altitude are found by iteratively
thermodynamic equilibrium with its adjusting estimated cloud altitude to minimize
surroundings—and altitude in the the difference in the parallax-corrected posi-
troposphere (24). However, for large tion between satellites (25). We have previous-
volcanic eruptions, the cloud pene- ly used this approach to successfully estimate
trates the tropopause and enters the the altitude and trajectory of the Chelyabinsk
stratosphere, where temperature in- meteor (26), demonstrating that the technique
creases with altitude, thus render- is appropriate for high-altitude clouds. Here,
ing temperature-based techniques we apply the approach to a series of manually
Fig. 2. Parallax-based retrievals of plume altitude at
inaccurate: In such cases, the plume selected cloud positions observed by Himawari-8
04:30 UTC on 15 January 2022 overlaid on Himawari-8
in the stratosphere will initially be and GEO-KOMPSAT-2A (GK-2A) to the west
high-resolution data for the same time frame.
cooler than the ambient air but will of the eruption and the Geostationary Oper-
Colored circles represent Advanced Himawari Imager (AHI)–
warm as the plume enters thermo- ational Environmental Satellite (GOES-17) to
Advanced Meteorological Imager (AMI) retrievals, and the
dynamic equilibrium with its sur- the east across images taken between 04:16
two crosses represent AHI-AMI–Advanced Baseline Imager
roundings, which presents several and 06:36 UTC, after which the Sun set. In
(ABI) retrievals.
possible altitude solutions for a single addition, we perform a more detailed anal-
cloud temperature. This is further ysis of the 04:36 UTC imagery from Himawari-8
complicated because the eruption and GK-2A to map variations in altitude across
wavelengths every 10 min (summarized in itself is likely to affect local vertical temper- the plume, which is shown in Fig. 2. Because
table S1), and at about 07:05 UTC, the National ature profiles. In this report, we make use of these methods rely on manual point selection
Oceanic and Atmospheric Administration the multiple satellite sensors that viewed the and analysis, they cannot map altitude across
(NOAA) enabled a fast-scanning mesoscale eruption from very different viewing geome- the whole plume. To gain a broad perspective of
sector over the volcano, producing imagery tries to compute plume altitude based on the plume altitude, we therefore also applied a
every minute. parallax effect, which does not suffer from stereoscopic vision tool to estimate altitude, which
Typically, the altitude of a volcanic plume is the limitations described earlier in the text, is less accurate but can provide spatial infor-
estimated from its top temperature measured and find that the HTHH eruption not only mation, and the standard temperature-based
Fig. 3. Temporal evolution of volcanic plume altitude derived from visible data across the entire volcanic plume, and the green markers are parallax
and infrared data. Infrared (IR) heights are derived from Himawari-8 brightness heights derived from a manual analysis of data from Himawari-8, GK-2A, and
temperature measurements and the ECMWF temperature profile. The gray- GOES-17. Error bars show the spread of estimated altitudes due to simulated
shaded area represents one standard deviation from the mean. The blue lines geolocation uncertainty. Because of sunset, no visible altitudes are available after
indicate altitudes estimated by the stereoscopic method applied to VIS (visible) 06:46 UTC. All times given are for 15 January 2022.

eruption may have been responsible for subse-

quent apparent increases in mesospheric clouds.
However, there is no agreement as to why some
volcanic eruptions but not others are asso-
ciated with mesospheric cloud increases (30).
Our observations of the HTHH plume provide
direct evidence that volcanic eruptions can in-
ject material into the mesosphere and will en-
able a more detailed analysis of mesospheric
chemistry and transport.
However, our work also raises questions:
What mechanisms contributed to this erup-
tion reaching such high altitude but with little
SO2 being observed? What is the composition
of the hazy substance that is visible atop the
highest portions of the plume, and how long
will it persist in the mesosphere? In addition,
we show that the mesospheric altitudes achieved
by this plume were only visible in satellite im-
ages taken at two times (those starting at 04:30
and 04:50 UTC), highlighting the importance
of frequent, every 10 min in this case, satellite
observations. Previous eruptions were observed
much less frequently by satellite, once every
hour by only one satellite (GMS-4) in the case
of Mount Pinatubo, meaning that the peak
altitude reached by the volcanic plume is likely
to have fallen between observations and that
parallax-based height estimates were not pos-
Fig. 4. Vertical structure of the volcanic plume. Represented are a lower near-tropopause layer, the sible. If the 2022 HTHH eruption had been ob-
altitude of which cannot be accurately determined; the main umbrella at higher altitude; and the dome-like served by the same satellite as that in the case of
protrusion seen in the 04:30 UTC satellite images from 15 January 2022. Layer thickness is computed Mount Pinatubo, the peak altitudes observed
from the spread of parallax-based altitude retrievals for each layer and, when optically thick, by using the infrared would have been substantially lower than those
brightness temperature approach. Overlaid are the 00:00 UTC Global Forecast System (GFS) and ECMWF described here: 32 km per infrared data and
temperature profiles that show the tropopause near 100 hPa and the stratopause between 1 and 1.5 hPa. 39 km using the shadow technique used for
Mount Pinatubo (14).
retrieval described earlier in the text. The is too optically thin, though, to support alti-
results of all three techniques are shown tude retrieval. By 04:56 UTC, the eruption
1. A. Robock, Rev. Geophys. 38, 191–219 (2000).
in Fig. 3. Further information on the meth- had produced a vertical column stretching 2. T. J. Casadevall, Ed., “Volcanic ash and aviation safety:
ods used here, including tables of parallax- from the surface to the mesosphere, includ- Proceedings of the first international symposium on volcanic
retrieved altitudes, is given in the supplementary ing two tendrils that reached 58 km (0.28 hPa). ash and aviation safety” (US Geological Survey Bulletin 2047,
US Government Printing Office, 1994).
materials. These tendrils cast a shadow onto the remain- 3. S. C. Loughlin, S. Sparks, S. K. Brown, S. F. Jenkins, C. Vye-Brown,
Altitude retrievals from the early phase of ing volcanic cloud at 35-km altitude, and cal- Eds., Global Volcanic Hazards and Risk (Cambridge Univ. Press,
the eruption show a rapidly ascending plume culating tendril altitude from the shadow 2015).
4. T. M. Wilson, S. Jenkins, C. Stewart, in Volcanic Hazards, Risks
that reached 25 km about 15 min after the erup- length gives an altitude of 57.5 km, which and Disasters, J. F. Shroder, P. Papale, Eds. (Elsevier, 2015),
tion began and 40 km, in the upper strato- closely matches the parallax-derived alti- pp. 47–86.
5. C. Schnetzler, G. Bluth, A. Krueger, L. Walter, J. Geophys. Res.
sphere, after 25 min. By 04:36 UTC, a half hour tude. Thereafter, as shown in Fig. 3, peak alti- 102, 20087–20091 (1997).
after the eruption began, a dome, ~90 km in tude decreased, and no further mesospheric 6. S. Kremser et al., Rev. Geophys. 54, 278–335 (2016).
diameter, was visible that extended from 34 km intrusions were detected. Analysis of the in- 7. J. Hansen, A. Lacis, R. Ruedy, M. Sato, Geophys. Res. Lett. 19,
215–218 (1992).
up to 57 km altitude (~0.3 hPa in the ECMWF frared imagery shows two clouds over the next 8. G. L. Stenchikov et al., J. Geophys. Res. 103, 13837–13857 (1998).
analysis), which is 13 km above the local strato- 12 hours: One moving south-west in the pre- 9. E. G. Dutton, J. R. Christy, Geophys. Res. Lett. 19, 2313–2316
(1992).
pause (1.5 hPa)—well within the mesosphere. vailing winds at 30 to 35 km altitude and one 10. K. Labitzke, M. P. McCormick, Geophys. Res. Lett. 19, 207–210
We estimate the uncertainty on this altitude to moving east at the tropopause. (1992).
be less than 1.5 km. Encircling this dome was a The influence of volcanoes on the mesosphere 11. A. Bernard, W. I. Rose Jr., Nat. Hazards 3, 59–67 (1990).
12. S. Carn, L. Clarisse, A. Prata, J. Volcanol. Geotherm. Res. 311,
donut-shaped structure with altitudes peak- is unclear. Previous eruptions into the strato- 99–134 (2016).
ing at 41 km (2.2 hPa) and a secondary layer sphere have indirectly affected the mesosphere 13. S. Guo, G. J. Bluth, W. I. Rose, I. M. Watson, A. Prata, Geochem.
Geophys. Geosyst. 5, Q04001 (2004).
just above the ECMWF tropopause at 17.5 km through the upward-ascent transport of vol-
14. R. E. Holasek, S. Self, A. W. Woods, J. Geophys. Res. 101,
(91 hPa). A schematic of the eruption at this canic aerosols after an eruption (27), although 27635–27655 (1996).
point is shown in Fig. 4. Ten minutes later, the chemical processes acting on mesospheric 15. A. Krueger, N. Krotkov, S. Carn, J. Volcanol. Geotherm. Res.
175, 408–414 (2008).
the dome collapsed to leave the expanding volcanic aerosols are the subject of debate (28). 16. M. Matson, J. Volcanol. Geotherm. Res. 23, 1–10 (1984).
donut cloud, although the haziness visible Vertical propagation of gravity waves from 17. M. Brenna et al., Lithos 412-413, 106614 (2022).
in the central region above the volcano indi- volcanoes can affect the mesospheric temper- 18. A. T. Prata et al., Sci. Rep. 10, 3584 (2020).
19. Global Volcanism Program, in vol. 47 of the Bulletin of the
cates that high-altitude aerosols such as sul- ature (29), whereas the aerosols and water vapor Global Volcanism Network, A. E. Crafford, E. Venzke, Eds.
fates or ash are likely to be present. This haze injected to high altitude by the 1883 Krakatoa (Smithsonian Institution, 2022).

20. M. Zuo et al., Adv. Atmos. Sci. 10.1007/s00376-022-2034-1 (2022). E. Carboni and R. Grainger regarding the HTHH eruption and drafts https://datasvc.nmsc.kma.go.kr/. ECMWF and GFS profile data
21. W. I. Rose et al., Nature 375, 477–479 (1995). of this paper. We also acknowledge the NOAA’s Big Data Program (filename TProfs.csv) and the Python code used to perform
22. D. A. Yuen et al., Earthquake Research Advances 2, 100134 for making GFS, GOES-17, and Himawari-8 data openly accessible; this analysis are available on GitHub: https://github.com/
(2022). the Japan Meteorological Agency for its Himawari-8 data access; simonrp84/Tonga_Volcano_Code. The lists of points used for
23. R. G. Owens, T. Hewson, “ECMWF forecast user guide” and the Korea Meteorological Administration (KMA) for making tri- and dual-satellite parallax calculations are also on the same
(European Centre for Medium-Range Weather Forecasts, GEO-KOMPSAT-2A data available via its open application GitHub page (filenames Points_Dual.csv and Points_Tri.csv).
2018). programming interface (API). We also thank the ECMWF for The infrared and stereoscopic retrieval outputs are in the
24. C. Oppenheimer, Int. J. Remote Sens. 19, 2829–2864 (1998). provision of the temperature vertical profile analysis used in this Height_Out and Heightmap subdirectories, respectively. License
25. L. Merucci, K. Zakšek, E. Carboni, S. Corradini, Remote Sens. 8, study. Funding: S.R.P.’s work on this study was funded as part of information: Copyright © 2022 the authors, some rights
206 (2016). NERC’s support of the National Centre for Earth Observation, reserved; exclusive licensee American Association for the
26. S. Proud, Geophys. Res. Lett. 40, 3351–3355 (2013). award ref. NE/R016518/1; and by a NERC Innovation fellowship, Advancement of Science. No claim to original US government
27. M. J. Mills, O. B. Toon, G. E. Thomas, J. Geophys. Res. D award ref. NE/R013144/1. A.T.P. was funded by NERC’s support of works. https://www.science.org/about/science-licenses-journal-
Atmospheres 110, D24208 (2005). the Radar-supported Next-Generation Forecasting of Volcanic article-reuse
28. C. George et al., in Atmospheric and Aerosol Chemistry, Ash Hazard project, award ref. NE/S003843/1. Author
V. McNeill, P. Ariya, Eds. (Topics in Current Chemistry 339, contributions: S.R.P. formulated the research, wrote the parallax
Springer, 2012), pp. 1–53. and infrared retrieval code, and analyzed the results. A.T.P. SUPPLEMENTARY MATERIALS
29. C. von Savigny et al., Meteorol. Z. 29, 3–18 (2020). also analyzed the results, and both S.R.P. and A.T.P. contributed to science.org/doi/10.1126/science.abo4076
30. C. Bertolin, F. Domınguez-Castro, Holocene 30, 682–690 (2020). drafting the manuscript. S.S. devised and performed the Materials and Methods
stereoscopic height retrievals. Competing interests: The Table S1
authors declare no competing interests. Data and materials References (31–38)
ACKN OW LEDG MEN TS availability: GOES-16 and Himawari-8 satellite data are available
We thank J. Amos for the initial recognition that this eruption through NOAA’s Big Data Program: https://www.noaa.gov/nodd/ Submitted 1 February 2022; accepted 8 October 2022
reached the mesosphere and acknowledge helpful discussions with datasets. GK-2A data are available via KMA’s open data API: 10.1126/science.abo4076
QUANTUM TECHNOLOGY ency to photons makes scalability challenging.

Although other promising platforms, such as
Robust multi-qubit quantum network node with trapped ions (16) and neutral atoms (17) in
optical cavities, achieved efficient coupling to
integrated error detection photons, scalable implementation of quantum
repeaters requires access to auxiliary memory
P.-J. Stas1*, Y. Q. Huan1*, B. Machielse1,2*, E. N. Knall3, A. Suleymanzade1, B. Pingault3,4,5, M. Sutula1, qubits to perform entanglement swapping,
S. W. Ding3, C. M. Knaut1, D. R. Assumpcao3, Y.-C. Wei1, M. K. Bhaskar1,2, R. Riedinger1,6,7, purification, and error detection (8, 9), which
D. D. Sukachev1,2, H. Park1,8, M. Lončar3, D. S. Levonian1,2, M. D. Lukin1† thus far has been challenging to achieve in a
single setting. For SiVs, the technical complex-
Long-distance quantum communication and networking require quantum memory nodes with efficient optical ity associated with the need for operation at
interfaces and long memory times. We report the realization of an integrated two-qubit network node based dilution refrigerator temperatures [required
on silicon-vacancy centers (SiVs) in diamond nanophotonic cavities. Our qubit register consists of the SiV to avoid phonon-induced dephasing (18)] has
electron spin acting as a communication qubit and the strongly coupled silicon-29 nuclear spin acting as a thus far impeded the prospects for scalability.
memory qubit with a quantum memory time exceeding 2 seconds. By using a highly strained SiV, we realize Here, we address this long-standing challenge
electron-photon entangling gates at temperatures up to 1.5 kelvin and nucleus-photon entangling gates up to by using a highly strained SiV (18, 19) featur-
4.3 kelvin. We also demonstrate efficient error detection in nuclear spin–photon gates by using the electron ing the 29Si isotope, with its nuclear spin (20)
spin as a flag qubit, making this platform a promising candidate for scalable quantum repeaters. serving as a deterministic long-lived memory
qubit. We demonstrate full control of this in-
tegrated two-qubit register at increased tem-
he ability to distribute quantum infor- over extended distances. These require net- peratures, including selective qubit readout
T
mation over extended distances (1, 2) work nodes containing multiple qubits that can via a phase-based readout method, enabling
constitutes an important enabling tech- collect, store, and process information com- multiple electron state resets before deteri-
nology in quantum information science, municated through photonic channels (9). Al- oration of the nuclear spin memory.
with applications in quantum key dis- though theoretical proposals for all-photonic The 29SiV in an external magnetic field con-
tribution (3, 4), nonlocal sensing (5), and dis- repeater schemes (10) potentially circumvent stitutes a two-qubit system of four spin states
tributed quantum computation (6, 7). A key this requirement, they involve efficient sources with nondegenerate transition frequencies
requirement for the realization of such long- of large-scale photonic states, which are very (Fig. 1A). The electron and nuclear spin qubits
distance quantum networking involves the challenging to realize. are coherently controlled by using microwave
development of quantum repeaters (8) to miti- Color centers in diamond nanophotonic struc- (MW) and radiofrequency (RF) pulses, re-
gate photonic qubit loss during transmission tures (11–14) have recently emerged as lead- spectively, which are delivered through gold
ing candidates for realizing such nodes owing coplanar waveguides (Fig. 1B). The SiV is em-
1
Department of Physics, Harvard University, Cambridge, to their long coherence times, high-fidelity bedded in a nanophotonic cavity (Fig. 1C), which
MA 02138, USA. 2AWS Center for Quantum Networking, Boston, single-qubit gates, efficient qubit-photon in- enhances optical transitions to the excited-
MA 02210, USA. 3John A. Paulson School of Engineering and
Applied Sciences, Harvard University, Cambridge, MA 02138,
terfaces, and high experimental repetition rates. state manifold at 737 nm (Fig. 1A, wavy arrows)
USA. 4QuTech, Delft University of Technology, 2600 GA The integration of all these features into a single that is used for state readout and spin-photon
Delft, Netherlands. 5Kavli Institute of Nanoscience Delft, Delft device has led to the demonstration of memory- entanglement. The SiV-cavity system exhibits
University of Technology, 2600 GA Delft, Netherlands.
6 enhanced quantum communication (11) with high-contrast spin-dependent reflection spectra
Institut für Laserphysik und Zentrum für Optische
Quantentechnologien, Universität Hamburg, 22761 Hamburg, the silicon-vacancy (SiV) center. Important (Fig. 1D) enabled by the strong cavity coupling
Germany. 7The Hamburg Centre for Ultrafast Imaging, steps toward quantum networking have been [cooperativity C ¼ 1:6 (21)].
22761 Hamburg, Germany. 8Department of Chemistry and taken by using multi-qubit registers based on High-fidelity resonant readout of the elec-
Chemical Biology, Harvard University, Cambridge, MA 02138, USA.
*These authors contributed equally to this work. nitrogen-vacancy (NV) centers and nearby nu- tron state is achieved by measuring the re-
†Corresponding author. Email: [email protected] clear spins (15), but the low coupling effici- flected intensity of a laser at the frequency of

A B D F
Count intensity (a.u.)

Refl. intensity
Optical (~406 THz) e
excited state
(norm.)
manifold e MW (~12 GHz)
RF (~30 MHz)
Refl. phase (rad)

3µm e 29
Si e
29
e Si
qubit manifold
n C e
e
29
29
Si
Si Photon detection time (ns)
e RF2 1µm
n
G
Si state fidelity
MW1
C Laser detuning (GHz)
MW2 E
n 29 EOM
e RF1
e Si
29
n
# of 95% fidelity electron readouts
Fig. 1. Quantum network node based on 29SiV. (A) Energy-level diagram of reflection intensity (top) and phase (bottom) as a function of laser frequency
the 29SiV. Allowed magnetic dipole transitions are shown by arrows in the qubit with a cavity-SiV detuning of D ¼ 69 GHz chosen to maximize optical
manifold for the case when the B-field is oriented along the symmetry axis. contrast. (E) Experimental setup for phase readout of the electron spin state.
(B) False-color scanning electron microscope (SEM) image of the device, (F) Measured spin-dependent phase shift of the beating pattern between
showing the gold coplanar waveguides in yellow. (C) Zoomed-in SEM image sidebands. (G) Coherence of a nuclear superposition state as a function of
of the nanophotonic cavity in which the 29SiV is located. (D) Spin-dependent the number of electron readouts for resonant and phase readouts.
maximum reflection intensity contrast (Fig.

1D, black line) (11). However, the nuclear qubit
experiences dephasing from the laser during
readout, with the highest decoherence rate
occurring when the laser is near the resonant
readout frequency (21). To enable selective
readout of the electronic spin qubit, we probe
the system at laser frequencies at which min-
imal nuclear dephasing occurs and make use
of the electron spin–dependent phase—instead
of intensity—of reflected photons for readout.
Using an electro-optic modulator (EOM) to
generate sidebands (Fig. 1E), we send two tones
(blue and magenta lines in Fig. 1D) along the
same path and measure the spin-dependent
reflected phase difference via the resulting beat-
ing pattern to determine the electron state (Fig.
1F) (21). With this phase-based readout, the
electron can be read 14 times at 95% fidelity
before causing a 1=e loss of nuclear coherence.
This constitutes an eightfold improvement over
the resonant readout method (Fig. 1G), with
further improvements expected through cavity
design and magnetic field optimization (21).
The two-qubit 29SiV system is fully controlled
by selectively driving the four single-spin–
flipping transitions (Fig. 1A) to implement
Fig. 2. Long-lived quantum memory based on 29Si nuclear spin. (A) Pulse sequence for a decoupled
the four possible controlled-NOT (CNOT) gates:
Ce NOTn gate. (B) Sequence for swapping the electron state onto the nucleus, after which a nuclear XY8
two electron-flipping gates Cn NOTe (MW1) and
decoupling sequence with n ¼ 1; 64; 256; 1024 pulses is applied and the nuclear state is measured through
Cn NOTe (MW2), and two nucleus-flipping gates
resonant readout of the electron. (C) Measured nuclear state fidelity after sequence (B). Inset: T2;n as a
Ce NOTn (RF1) and Ce NOTn (RF2), where the
function of the number of decoupling pulses n fitted to T2;n ºna where a ¼ 0:53.
absence (presence) of the overbar indicates con-
ditioning on the control spin being in the down
(up) state. We measure a fidelity of 99:9 T 0:1% ence time using a dynamical decoupling se- tation conditional on the same initial electron
for the Cn NOTe gate with a gate time of 30.0 ns. quence and interleave it with stepwise nuclear state, achieving a gate fidelity of 93:7 T 0:7%
As the time to drive a nuclear p rotation (23 ms) rotations (22, 23) to implement a decoupled (21) within a gate time of 29.2 ms. With the
is longer than the electron dephasing time T2;e
Ce NOTn (Fig. 2A). Specifically, we apply RF1 in decoupled Ce NOTn , we can swap a superpo-
(5 ms) (21), transfer of electron superposition the first window, and alternate it with apply- sition state from the electron onto the nuclear
states onto the nucleus is not possible with di- ing RF2 after every subsequent unconditional spin state (Fig. 2B), where it can be stored
rect driving for the Ce NOTn gate. To circumvent electron p pulse to account for the flipping using an XY8 decoupling sequence. We find
this problem, we increase the electron coher- electron state while keeping the nuclear ro- that the nuclear coherence time scales with n,

Fig. 4. Spin-photon entanglement with integrated

error detection. (A) Gate sequence for entanglement
between a photon and nucleus and subsequent
state storage with error detection (ED) based on
resonant readout of the electronic spin. (B) Bell
state fidelity improvement with ED at 0.1 K
and 4.3 K. (C) Bell state fidelity as a function of
Fig. 3. Spin-photon entanglement at elevated temperatures. (A) Coherence time of the electron (top) and storage time for the sequence shown in (A) before
nucleus (bottom) obtained using 8-XY8 sequences as a function of temperature. Initial increase of the nuclear and after error detection. Fits are exponential curves
T2;n is due to motional averaging of the noise bath. Solid lines are fits to decoherence models of the spin decaying to the maximally mixed-state fidelity
environment described in (21). (B) Electron spin–photon gate implementation with (C) reconstructed density Fþ rmm Fþ ¼ 0:25 as the nuclear lifetime T1;n is
matrix at 1.5 K using resonant electron readout. (D) Circuit diagram for the photon-nucleus entangling (PHONE) comparable to the coherence time; fitted decay
gate. CNOT gates with black and white control dots refer to Cn NOTe and Cn NOTe gates, respectively. constants are 4.5 (3.8) ms with (without) ED.
Reconstructed density matrices for photon-nuclear spin entanglement at (E) 100 mK and (F) 4.3 K.
the number of decoupling pulses applied, as gate that directly entangles the 29Si nuclear error correction protocols (24), PHONE gate
T2;n ºn0:53 with a maximum measured nuclear spin with a photonic qubit using only fast errors can be reduced at the cost of some gate
memory time of T2;n ¼ 2:1 T 0:1 s (Fig. 2C) MW gates and allows operating tempera- failure probability (25) by measuring the state
when using a 128-XY8 sequence. tures up to 4.3 K. In this scheme, the photon- of the electron spin qubit (Fig. 4A). Specifi-
To enable robust spin-photon entanglement, nucleus-electron system is initialized in the cally, if the electron spin is measured to be in
we use a SiV with large residual strain [ground ðe þ lÞð↑n þ ↓n Þ↓e =2 state, and we apply a se- the ↑e state after a PHONE gate application,
state splitting DGS ¼ 554 GHz (21)], which quence of CNOTs as shown in Fig. 3D such the prepared photon–nuclear spin entangled
greatly suppresses the rate of thermal deco- that the electron state—and as a result, the SiV state is discarded. As shown in Fig. 4B, the use
herence processes (19), enabling operation at reflectivity—in each time bin is conditional on of the procedure results in a Bell state fidel-
1.5 K without an appreciable reduction in T2;e the nuclear state; the finalpffiffi entangled state is ity increase of 2% (7%) to Fnγ ¼ 0:87 T 0:02
(Fig. 3A). We implement high-temperature given by ðe↑n þ l↓n Þ↓e = 2. We measure the re- (0:71 T 0:02) with an error detection rate of
spin-photon entangling gates between the sulting Bell state fidelity to be Fng ¼ 0:85 T 0:02 8.4% (13.9%) at 0.1 K (4.3 K). In Fig. 4C, we
electron spin and time-bin qubits in the fe; lg (0:66 T 0:02) at 0.1 K (4.3 K) (Fig. 3, E and F), combine all these components to implement a
basis, corresponding to the presence of a pho- with dominant infidelities due to MW gate PHONE gate at 4.3 K with error detection and
ton in either the early or late time bin. The errors and the nuclear state dependence of the store the spin-photon entangled state in the
entangling gate sequence (11) shown in Fig.p 3Bffiffi optical transition frequency, which prevents nuclear spin memory using an echo sequence
generates the entangled state ðe↓e þ l↑e Þ= 2 the photon frequency from being tuned to (Fig. 4A) for more than 2.5 ms above the
conditioned on the detection of a single re- maximize contrast for both nuclear spin states threshold value of 50%, as compared to 1.5 ms
flected photon. We find a Bell state fidelity of simultaneously (Fig. 1D) (21). without error detection. The error detection
Feγ ¼ 0:91 T 0:02 (0:90 T 0:01) at 0.1 K (1.5 K) A key feature of the PHONE gate is that the protocol detects MW gate errors and T1;e- and
(Fig. 3C), which is primarily limited by resid- electron should always remain in the ↓e state T2;e -limited depolarization and dephasing,
ual reflections from the ↓e state and imperfect after a successful gate application. Because the and the large gain in fidelity at 4.3 K is due to
photonic state measurements. electron spin mediates the interface between the greater contribution of the detectable errors
To further extend the entanglement capa- the photon and the nucleus, electron spin flips from short T1;e and T2;e times at this tem-
bilities of our spin-photon interface, we intro- can be used as an integrated error witness to perature (21). Conversely, the improvement
duce a photon-nucleus entangling (PHONE) detect gate errors. Similar to flag qubits in from error detection at 0.1 K is limited as the

from photon state measurement and SiV opti- 7. D. Cuomo, M. Caleffi, A. S. Cacciapuoti, IET Quantum Commun. the NSF, CUA, DoD/ARO DURIP, AFOSR MURI, and DOE (award no.
cal contrast. 1, 3–8 (2020). DE-SC0020115). Devices were fabricated at Harvard CNS, NSF
8. W. Dür, H.-J. Briegel, J. I. Cirac, P. Zoller, Phys. Rev. A 59, award no. 1541959. Y.Q.H. acknowledges support from the A*STAR
These observations open up several new ave- 169–181 (1999). National Science Scholarship. D.A., E.N.K., and B.M. acknowledge
nues for realizing quantum networks and ex- 9. L. Childress, J. M. Taylor, A. S. Sørensen, M. D. Lukin, support from an NSF GRFP No. DGE1745303. B.P. acknowledges
Phys. Rev. Lett. 96, 070504 (2006). support through a Marie Skłodowska-Curie fellowship from the
ploring their applications. The access to an 10. K. Azuma, K. Tamaki, H.-K. Lo, Nat. Commun. 6, 6787 (2015). European Union’s Horizon 2020 research and innovation
additional memory qubit directly enables im- 11. M. K. Bhaskar et al., Nature 580, 60–64 (2020).
programme under the Grant Agreement No. 840968 (COHESiV).
12. E. R. Schmidgall et al., Nano Lett. 18, 1175–1179 (2018).
proved memory-enhanced quantum key dis- 13. A. E. Rugar et al., ACS Photonics 7, 2356–2361 (2020). M.S. acknowledges funding from the NASA Space Technology
tribution (11) by extending the time window 14. X. Guo et al., Nano Lett. 21, 10392–10399 (2021). Graduate Research Fellowship Program. R.R. acknowledges
15. M. Pompili et al., Science 372, 259–264 (2021). support from the Alexander von Humboldt Foundation and the
for photons arriving from different commu- Cluster of Excellence ‘Advanced Imaging of Matter’ of the Deutsche
16. J. Schupp et al., PRX Quantum 2, 020331 (2021).
nicating parties to the memory node for asyn- 17. A. Reiserer, N. Kalb, G. Rempe, S. Ritter, Nature 508, 237–240 Forschungsgemeinschaft (DFG) - EXC 2056 - project ID
chronous Bell measurement, whereas the phase (2014). 390715994. Author contributions: P.-J.S., Y.Q.H., B.M., and
18. K. D. Jahnke et al., New J. Phys. 17, 043011 (2015). D.S.L. performed the measurements and analyzed data. B.M. and
readout protocol facilitates electron resets 19. S. Meesala et al., Phys. Rev. B 97, 205444 (2018). E.N.K. fabricated the devices. A.S., B.P., M.S., S.D., C.M.K., D.R.A.,
between entanglement attempts while infor- 20. B. Pingault et al., Nat. Commun. 8, 15579 (2017). Y.-C.W., M.B., R.R., and D.S. assisted with experiments. All work
21. Further details on theoretical analysis and additional supporting
mation is stored on the nucleus. A nuclear mem- experimental results are provided in the supplementary materials.
was supervised by H.P., M.L., and M.D.L. All authors discussed
ory can additionally expand the capabilities the results and contributed to the manuscript. Competing
22. C. Bradley et al., Phys. Rev. X 9, 031045 (2019).
23. T. van der Sar et al., Nature 484, 82–86 (2012). interests: The authors declare that there are no competing
of the SiV as a single photon source for the 24. R. Chao, B. W. Reichardt, Phys. Rev. Lett. 121, 050502 (2018). financial interests. Data and materials availability: Data and code
creation of photonic cluster states (26, 27). 25. J. Borregaard, P. Kómár, E. M. Kessler, A. S. Sørensen, can be found on Zenodo (32). License information: Copyright ©
The methods demonstrated here can also M. D. Lukin, Phys. Rev. Lett. 114, 110502 (2015). 2022 the authors, some rights reserved; exclusive licensee
26. A. Russo, E. Barnes, S. E. Economou, Phys. Rev. B 98, 085303 American Association for the Advancement of Science. No claim to
enable the deployment of scalable SiV-based (2018). original US government works. https://www.science.org/about/
quantum repeater networks. Based on a large- 27. C. P. Michaels et al., Quantum 5, 565 (2021). science-licenses-journal-article-reuse
28. B. Machielse et al., Phys. Rev. X 9, 031022 (2019).
scale survey of a second chip fabricated using 29. S. Maity et al., Phys. Rev. X 12, 011056 (2022). SUPPLEMENTARY MATERIALS
the current process (21), we find that more 30. E. N. Knall et al., Phys. Rev. Lett. 129, 053603 (2022).
31. J. Borregaard et al., Phys. Rev. X 10, 021071 (2020). science.org/doi/10.1126/science.add9771
than 11% of all SiVs are sufficiently strained Supplementary Text
32. P.-J. Stas et al., Zenodo (2022). https://doi.org/10.5281/
for operation at 1.5 K, indicating the possibility zenodo.7101712. Figs. S1 to S15
for a sizable number of highly strained devices Tables S1 to S6
ACKN OWLED GMEN TS References (33–56)
in nanophotonic cavities that can operate at
elevated temperatures. Moreover, recently dem- We thank J. Borregaard for discussions and J. MacArthur for Submitted 17 July 2022; accepted 27 September 2022
assistance with electronics. Funding: This work was supported by 10.1126/science.add9771
onstrated techniques for strain tuning of
nanocavities (28) can enable fully determinis-
tic access to high-strain operation. The fidelity
gain from our error detection could be larger MONKEYPOX
for more complex spin-photon entangling se-
quences, such as PHONE-type gates entangling Multiple lineages of monkeypox virus detected
successive photons with the nucleus or entan-
gling a photon with multiple nuclei strongly in the United States, 2021–2022
coupled to the SiV. It also opens opportunities
to operate more effectively in regimes where Crystal M. Gigante1, Bette Korber2, Matthew H. Seabolt1,3, Kimberly Wilkins1, Whitni Davidson1,
the SiV electron coherence properties deterio- Agam K. Rao1, Hui Zhao1, Todd G. Smith1, Christine M. Hughes1, Faisal Minhaj1,
rate, including at higher temperatures as dem- Michelle A. Waltenburg1, James Theiler4, Sandra Smole5, Glen R. Gallagher5, David Blythe6,
onstrated here, and in a misaligned magnetic Robert Myers6, Joann Schulte7, Joey Stringer7, Philip Lee8, Rafael M. Mendoza9,
field as is required for acoustic spin control LaToya A. Griffin-Thomas10, Jenny Crain11, Jade Murray12, Annette Atkinson12, Anthony H. Gonzalez13,
(29) and single-photon generation (30). The June Nash13, Dhwani Batra1, Inger Damon1, Jennifer McQuiston1, Christina L. Hutson1,
use of cavities with higher cooperativity as Andrea M. McCollum1, Yu Li1*
demonstrated previously (11) should allow
for higher fidelity and efficiency of electron- Monkeypox is a viral zoonotic disease endemic in Central and West Africa. In May 2022, dozens of non-
spin photon and PHONE gates, as well as im- endemic countries reported hundreds of monkeypox cases, most with no epidemiological link to
proved 29Si state preservation during electron Africa. We identified two lineages of monkeypox virus (MPXV) among two 2021 and seven 2022 US
readout (21). Finally, nearby 13C spins, such as monkeypox cases: the major 2022 outbreak variant called B.1 and a minor contemporaneously sampled
one associated with the SiV used in this work variant called A.2. Analyses of mutations among these two variants revealed an extreme preference
(21), can be used as additional memory re- for GA-to-AA mutations indicative of human APOBEC3 cytosine deaminase activity among Clade IIb
sources (15, 22). Apart from realizing multi- MPXV (previously West African, Nigeria) sampled since 2017. Such mutations were not enriched
node quantum network protocols (8, 9), these within other MPXV clades. These findings suggest that APOBEC3 editing may be a recurrent and a
systems can also allow for the generation of dominant driver of MPXV evolution within the current outbreak.
complex photonic tree cluster states that en-
able robust one-way long-distance quantum
communication (31). onkeypox is a viral zoonotic disease outbreak in western Africa occurred in Nigeria
M
caused by monkeypox virus (MPXV) after decades of no identified cases (2), and
RE FE RENCES AND N OT ES that is endemic in West and Central during 2018 to 2021, eight cases were exported
1. H. J. Kimble, Nature 453, 1023–1030 (2008).
Africa. There have been several reported from Nigeria to non-endemic countries (2–5).
2. S. Wehner, D. Elkouss, R. Hanson, Science 362, eaam9288 (2018).
3. A. K. Ekert, Phys. Rev. Lett. 67, 661–663 (1991). cases of travel-associated monkeypox In 2021, there were two US monkeypox cases
4. H.-K. Lo, M. Curty, K. Tamaki, Nat. Photonics 8, 595–604 (2014). in non-African countries in recent years. In in travelers from Nigeria, one in Maryland and
5. E. T. Khabiboulline, J. Borregaard, K. De Greve, M. D. Lukin, 2003, an outbreak of monkeypox in the United one in Texas (4, 5). In May of 2022, this pattern
Phys. Rev. Lett. 123, 070504 (2019).
6. H. Buhrman, H. Röhrig, Mathematical Foundations of Computer States was linked to imported African small of monkeypox cases in travelers from Nigeria
Science 2003 (Springer Berlin Heidelberg, 2003), pp. 1–20. mammals (1). In 2017, the largest monkeypox shifted. As of 28 September 2022, 67,602 cases

of monkeypox were reported in 99 non-endemic and table S3). Each case also reported travel to cestor of Lineage A to the estimated ancestors
countries, most with no epidemiological link different countries in the Middle East and West of both variant B.1 and variant A.2, including
to Africa (6), and 25,509 cases have been con- Africa (table S4). This suggests that although the each internal branch, revealed significantly in-
firmed in the United States (7). three MPXVs share a common ancestor, they creased APOBEC3 motif G-to-A changes, indi-
Comparison of MPXV sequences from nine likely represent separate introductions to the cating that sequential acquisition of mutations
US monkeypox cases from 2021 and 2022 United States. in distinct individuals sampled historically since
(ON563414.3, ON674051, ON675438, ON676703, Real-time polymerase chain reaction (PCR) 2017 have contributed to the high proportions
ON676704, ON676705, ON676706, ON676707, testing of USA_2021_MD and the five 2022 of GA-to-AA APOBEC3 (fig. S2 and table S6).
and ON676708) revealed two distinct lineages MPXV variant B.1 samples revealed a lower sen- Given the limited variation found in 2022 out-
(Fig. 1) within MPXV Clade IIb (formerly named sitivity (average Ct delay of 6.88, ranging from break sequences in variant B.1, there were only a
West African MPXV found east of the Dahomey 5.3 to 8.3) in the Orthopoxvirus generic OPX3 small number of SNPs relative to the ancestor
Gap). Neutral nomenclature was chosen to re- real-time PCR assay (14) compared with the for any one sequence. When considered together,
duce the stigmatization that can be associated Monkeypox Clade II–specific real-time PCR however, all eight SNPs observed among the
with naming that is based on locations (8). assay (15) (table S5, when performed as de- 12 variant B.1 genomes available in early May
We maintain the designation of two clades scribed in Methods). By contrast, similar Ct 2022 were GA-to-AA APOBEC3 context muta-
of MPXV, Clade I: formerly Congo Basin/Central values were observed for USA_2022_FL001, tions (Fig. 2, fig. S2, and table S6). Outbreak-
African and Clade II: formerly West African, which USA_2022_VA001, or USA_2021_TX samples related sequence data have increased very
are based on genetic distance and evidence- (average difference of –0.78 Ct, ranging from rapidly, enabling the analysis of 397 outbreak
based clinical differences (9–12). Five of the –1.4 to 0.5). Careful comparison of Ct values genomes sampled between 1 May and 15 July
seven May 2022 US sequences formed a mono- between the two assays can, in theory, differ- 2022, available through the GISAID data shar-
phyletic clade with 2022 MPXV sequences from entiate between cases belonging to variant B.1 ing initiative (21). This greatly expanded data-
Europe (Fig. 1), with most genomes within this and cases from other Clade IIb MPXV without set confirmed the pattern observed among
clade containing zero to two nucleotide changes sequencing. Sequence examination revealed a the first 12 sequences: 275/308 (89%) of ob-
in nonrepeat regions (fig. S1 and table S1). This single nucleotide polymorphism (SNP) in the served unique G-to-A mutations occurred in
clade will be referred to as the current pre- reverse primer-binding site for the OPX3 real- an APOBEC3 context, and 261/275 (95%) of
dominant 2022 MPXV outbreak variant, B.1 time PCR assay (DNA polymerase gene VACV- these were specifically 5′ GA-to-AA (Fig. 2,
[which is based on proposed MPXV naming Cop E9L, C322T) in USA_2021_MD and variant fig. S3, and table S7). Beyond variant B.1, by
from Nextstrain (13)]. MPXV from a 2021 travel- B.1 sequences that was absent from other Clade partitioning Lineage A into sublineages (A.2,
associated case from Nigeria to Maryland IIb MPXV sequences. The decreased sensi- Nigeria 2017, and Nigeria 2017–2019; Fig. 2)
(USA_2021_MD) displayed high similarity tivity in the OPX3 assay is unrelated to the we found that G-to-A mutations in variant
to variant B.1 sequences, with ~13 nucleotide US Food and Drug Administration–cleared A.2 were in an APOBEC3 context and were
differences (Fig. 1, fig. S1, and Table 1). The VAC1 assay (16) used for MPXV screening. Use enriched throughout Lineage A (Fig. 2, fig.
USA_2021_MD and 2022 outbreak sequences of different commercial reagents, run param- S2, and table S6). Four additional MPXV se-
shared many mutations that separated them eters, or PCR platforms may result in differ- quences belonging to variant A.2 have now been
from MPXV sequences from Nigeria and travel- ent results. detected in India and Thailand (22); 45/47 (96%)
associated cases from 2017–2019 (table S2). When we compared the 2022 outbreak MPXV G-to-A mutations were in an APOBEC3 context,
Two 2022 US MPXV sequences, USA_2022_ sequences in Fig. 1 with other MPVX sequences and all 45 of these were specifically 5′ GA-to-AA
FL001 and USA_2022_VA001, and one 2021 in Lineage A, we noticed a marked prepon- (table S7).
US sequence, USA_2021_TX, formed a mono- derance of G-to-A mutations, specifically 5′ By contrast, APOBEC3 context G-to-A changes
phyletic clade (variant A.2) that was polyphy- GA-to-AA or 5′ GG-to-AG, indicative of host were lower than would be expected by chance
letic to other 2022 MPXV sequences from the Apolipoprotein B mRNA Editing Catalytic relative to other G-to-A changes across Clade I
United States and Europe (Fig. 1). Each genome Polypeptide-like3 (APOBEC3) activity (17, 18) and Clade IIa MPXV (Fig. 2, fig. S2, and table
contained ~80 nucleotide changes relative to (Fig. 2 and Table 2). Looking at this systemat- S8). To better resolve where within the phylog-
the variant B.1 MPXV sequences (fig. S1). The ically, we found that a significant enrichment eny the switch in mutational patterns arose,
three variant A.2 genomes displayed ~30 unique of APOBEC3 context G-to-A mutations was we evaluated the mutational frequencies along
nucleotide differences from each other (fig. S1 evident throughout Lineage A, among Clade IIb the internal branches in Clade IIb leading into
MPXV sequences sampled from 2017 to 2022 Lineage A (Fig. 2, fig. S2, and table S8), and
1
National Center for Emerging and Zoonotic Infectious (Fig. 2 and Table 2). Overall, among unique mu- these were not found to be statistically enriched
Diseases, Centers for Disease Control and Prevention, tations within Lineage A, 167 G-to-A mutations for APOBEC3 context G-to-A changes.
Atlanta, GA, USA. 2T-6: Theoretical Biology and Biophysics,
Los Alamos National Laboratory, Los Alamos, New Mexico, were in an APOBEC3 context, whereas nine In the 15 July 2022 GISAID sample, there
USA; New Mexico Consortium, Los Alamos, NM, USA. G-to-A mutations were not in an APOBEC3 were three highly related sequences, one each
3
Leidos Inc., Reston, VA 20190, USA. 4ISR-3: Space Data motif, and only 14 mutations were not G-to- from Italy, Spain, and France. These were chi-
Science and Systems, Los Alamos National Laboratory, Los
Alamos, NM, USA. 5Massachusetts Department of Public
A (Fig. 2 and Table 2). The vast majority of meric in that most of the genome was typical
Health, Jamaica Plain, MA, USA. 6Infectious Disease the APOBEC3 context G-to-A mutations were of variant B.1, but each inverted terminal re-
Epidemiology and Outbreak Response Bureau, Maryland specifically GA-to-AA (156 of 167, or 95%; Fig. 2), peat region carried five consecutive bases that
Department of Health, Baltimore, MD, USA. 7Dallas County
Health and Human Services Public Health Laboratory, Dallas,
indicating that if these mutations were gen- were typically found among other Lineage A
Texas, USA. 8Florida Department of Health Bureau of Public erated by APOBEC3 editing, then APOBEC3G and Clade IIb MPXV, but not among B.1
Health Laboratories-Jacksonville, Jacksonville, FL, USA. was not the major form because it produces sequences. A detailed analysis of these se-
9
Florida Department of Health in Broward County, Hollywood,
GG-to-AG changes (19, 20). quences is provided in fig. S4, and they were
FL, USA. 10Virginia Department of General Services, Division
of Consolidated Laboratory Services, Richmond, VA, USA. We explored the abundance of APOBEC3 excluded from the analysis shown in fig. S3.
11
Virginia Department of Health, Richmond, VA, USA. 12Utah motif G-to-A mutations along the evolution- This pattern is potentially indicative of re-
Department of Health and Human Services, Salt Lake City, ary trajectory from the common ancestor of combination, an important aspect of poxvirus
UT, USA. 13Sacramento County Public Health, Sacramento,
CA, USA. Lineage A to the variants A.2 and B.1 (Fig. 2 evolution (23); however, assembly errors
*Corresponding author. Email: [email protected] and fig. S2). The path from the estimated an- caused by reference calling in low coverage

Fig. 1. Phylogenetic analysis of Clade IIb MPXV genome sequences. Variant B.1 sequences are shown in blue text, variant A.2 sequences in orange text, and green
branches indicate Clade IIb. Phylogenetic analysis was performed in BEAST v1.8.3 using the HKY+G model and constant coalescent prior on complete genome
alignments after removing all sites containing gaps. Scale bar is in substitutions per site; posterior support values are shown at branch points.
regions could not be ruled out as a possible

explanation. Table 1. Unique coding nucleotide changes in 2022 predominant MPXV outbreak variant B.1.
Given the clear transition in the mutational Mutations listed (all SNPs) were shared among all MPXV Clade IIb variant B.1 sequences examined in
pattern in Lineage A relative to Clades I and IIa Fig. 1 and were not present in USA_2021_MD or UK-P2 (MT903344.1). Gene homologs in vaccinia virus
(Fig. 2), we explored the hypothesis that there Copenhagen (VACV-Cop) are given for each gene. Additional information can be found in table S2.
was a concomitant change in the evolutionary
rate in Lineage A. First, we used the fixed lo-
cal clock model as implemented in BEAST to MD 2022 Amino acid
Position* Effect Note
2021 Outbreak change
compare the estimated mean evolutionary rate
of Lineage A (7.2 × 10−6 ± 8.9 × 10−7 SD) with 3120, 194114 G A Synonymous VACV-Cop C19L ankyrin-like
.....................................................................................................................................................................................................................
those of Clade I (1.9 × 10−6 ± 3.1 × 10−7 SD) VACV-Cop F13L
39148 C T Missense Glu353Lys
and Clade IIa (3.9 × 10−6 ± 8.9 × 10−7 SD) (table major envelope antigen of EEV
.....................................................................................................................................................................................................................
S9). We also separated Lineage A, retaining VACV-Cop G8R VLTF-1
73248 G A Missense Asp88Asn
Nigerian-SE-1971 as an outgroup, from Clade I (late transcription factor 1)
.....................................................................................................................................................................................................................
and IIa and repeated the analysis using two VACV-Cop G9R entry/fusion
74214 G A Missense Met142Ile
trees. We felt this to be the more appropriate complex component
.....................................................................................................................................................................................................................
alternative given the transition in the under- VACV-Cop L4R ss/dsDNA
77392 G A Missense Glu162Lys
lying evolutionary model. In this analysis, we binding protein
.....................................................................................................................................................................................................................
found that Lineage A had an even higher evo- VACV-Cop J6R RNA polymerase
84596 C T Synonymous
lutionary rate estimate of 2.8 × 10−5 ± 8.8 × subunit (RPO147)
.....................................................................................................................................................................................................................
10−6 SD, whereas estimates for Clades I and VACV-Cop A46R IL-1/TLR
150480 C T Missense His221Tyr
IIa were essentially unchanged (table S9). Thus, signaling inhibitor
.....................................................................................................................................................................................................................
using either approach, we observed higher evo- 170273 G A Synonymous VACV-Cop B12R Ser/Thr kinase
.....................................................................................................................................................................................................................
lutionary rates within Lineage A, consistent VACV-Cop B21R membrane-associated
183534 C T Missense Pro722Ser
with mutations being driven by host editing glycoprotein
.....................................................................................................................................................................................................................
mediated through APOBEC3 over what is
*Position in MT903344.1.
expected by errors in the viral polymerase.

An enriched evolutionary rate for Clade B.1

outbreak viruses has been similarly reported
by others (18, 22). A cautionary note regarding
the robustness of estimates of rates of evo-
lutionary change is that molecular clock as-
sumptions may be violated in a scenario where
mutations are dominated by host-mediated
APOBEC3 deaminase activity, as different peo-
ple may have different levels of APOBEC3 activ-
ity, and this may vary during the course of an
infection. For example, transient increases in
G-to-A ABOBEC3 motif–enriched mutations in
HIV genomes in single infected hosts have been
observed both in HIV-1 infection in humans and
simian-HIV (SHIV) infections in monkeys (24, 25)
All publicly available MPXV genomes from
the 2022 monkeypox outbreak to date belong to
Clade II, which may cause less severe disease
and have a lower case fatality rate than Clade I
MPXV (10–13). Genomes published during the
2022 outbreak share a common ancestor with
MPXV sequences from Nigeria (Clade IIb); how-
ever, sequences from surrounding countries are
limited, and most of our understanding of these
relationships comes from viruses linked to or
identified in Nigeria. The high similarity among
the current predominant 2022 MPXV outbreak
variant B.1 sequences analyzed here is similar to
the 0.4 to 1.5 SNPs reported between genomes
from epidemiologically linked samples from the
same transmission chain (3). Among Clade IIb
sequences, many unique mutations were shared
between USA_2021_MD and variant B.1 genomes,
further indicating that any evolutionary force
leading to these changes most likely preceded the
2022 outbreak and were found in a common
ancestor shared between the variant B.1 and
USA_2021_MD. By contrast, there were suffi-
cient differences (>30 SNPs) among the three
variant A.2 sequences to suggest that they
likely represent separate introductions into
the United States.
After the reemergence of MPXV in Nigeria in
2017 and before 2022, there had only been two
reported events of person-to-person transmission
Fig. 2. Analysis of APOBEC3 motif mutations in MPXV. (A) Maximum likelihood phylogenetic tree using
of MPXV outside of Africa (3, 26). Several factors
IQ-TREE. Enrichment for G-to-A APOBEC context mutations (blue line) was found in Lineage A, within Clade IIb (green
may be contributing to this recent increased
branches). A detailed version of this tree with taxa names is included in fig. S2. The scale bar indicates the number
spread between humans, including behaviors
of substitutions per sequence site. (B) Mutational patterns found among Clade I, Clade IIa, and Lineage A in the
involving close contact and failure to recognize
phylogeny above. All unique mutations in a clade relative to the most recent common ancestor of that clade are shown
or diagnose monkeypox to prevent spread. It is
in a single panel, and the class of each mutation is shown on either the forward or reverse complement strand.
currently unclear whether the APOBEC3 motif or
The increased number of APOBEC3 context mutations, indicated by blue ticks, in Lineage A, versus red and gray
other mutations observed in the 2022 outbreak
captures the dominance of GA-to-AA APOBEC3 context mutations in this lineage and contrasts with Clades I and IIa.
MPXV variants have affected or will affect viral
The exact numbers and statistics are provided in Table 2. The A.2 and B.1 variants within Lineage A in the tree and an
transmissibility or pathogenicity. An E353K mu-
updated analysis of 397 variant B.1 sequences from GISAID demonstrate the continuing dominance of GA-to-AA
tation in F13L (which codes for the target of
APOBEC3 context mutations among currently sampled outbreak sequences. (C) Bar charts showing the total
tecovirimat, an antiviral agent indicated for the
number of each class of unique mutations within each clade shown in (B) for Clade I, Clade IIa, and lineage A, as well
treatment of smallpox) was shared by all se-
as for 397 sequences from lineage B.1 from fig. S3A.
quences in the predominant 2022 outbreak clus-
ter; functional studies revealed no effect on the
efficacy of tecovirimat (the median effective and restrict the replication of exogenous viruses but can also act on DNA viruses (28, 29). It is
concentration was 0.007731 ± 0.002 mM for 353E through cytosine-to-uracil deaminase activ- unlikely that the GA-to-AA substitutions we
and 0.002860 ± 0.001 mM for 353K) (fig. S5). ity (19, 27). APOBEC3 proteins act mainly on report here were caused by sequencing error
APOBEC3 proteins are an important compo- single-stranded DNA and have been extensively because the sequences used in the analyses were
nent of the vertebrate innate immune system studied in RNA viruses, including HIV (20), produced across many different laboratories

Table 2. Summary of the counts of different classes of all unique mutations observed in the different clades shown in Fig. 2. C-to-T mutations in the forward
strand are included as G-to-A in the reverse complement as a simplified way of tallying all mutations that occurred in the context of an APOBEC3 motif in the viral
genome relative to those that did not. P values are derived from a Fisher’s exact test of a contingency table based on columns 4 to 7. First, we identified all G’s either
in an APOBEC context (GA or GG) or not in an APOBEC context (GC or GT), and then we determined the number that had undergone G-to-A substitutions and the
number where the G was unchanged within each of the two contexts. An ABOBEC motif odds ratio >1 indicates APOBEC context enrichment. The final column shows the
number of GA-to-AA and GG-to-AG mutations that constitute the total G-to-A APOBEC context mutations in column 4.
G in
G-to-A G-to-A G not in
All other APOBEC P value GA-to-AA +
Clade APOBEC non APOBEC APOBEC context, q value Odds ratio
mutations context, (two-sided) GG-to-AG
context context no change
no change
Non-APOBEC I 277 72 36552 176 30580 <0.000001 0.000006 0.3423 45 + 27
.....................................................................................................................................................................................................................................................................................................
context IIa 225 65 36548 132 30578 <0.000001 0.000006 0.412 35 + 30
.....................................................................................................................................................................................................................................................................................................
Lineage A 14 167 36415 9 30669 <0.000001 0.000006 15.63 159 + 8
APOBEC context .....................................................................................................................................................................................................................................................................................................
B.1 (n = 12) 0 8 36517 0 30675 0.0095 0.022 Infinity 8+0
enriched .....................................................................................................................................................................................................................................................................................................
B.1 (n = 397) 115 275 35064 33 29705 <0.000001 0.000006 7.06 261 + 14
............................................................................................................................................................................................................................................................................................................................................
using multiple methods, including both short- 2. A. Yinka-Ogunleye et al., Lancet Infect. Dis. 19, 872–879 C. M. Hughes, F. Minhaj, M. A. Waltenburg, J. Theiler, S. Smole,
and long-read approaches and with and with- (2019). G. R. Gallagher, D. Blythe, R. Myers, J. Schulte, J. Stringer, P. Lee,
3. M. R. Mauldin et al., J. Infect. Dis. 225, 1367–1376 (2022). R. M. Mendoza, L. A. Griffin-Thomas, J. Crain, J. Murray,
out PCR. The presence of mutations consistent 4. V. Costello et al., Emerg. Infect. Dis. 28, 1002–1005 A. Atkinson, A. H. Gonzalez, J. Nash, D. Batra, I. Damon,
with APOBEC3 editing in several branches of (2022). J. McQuiston, C. L. Hutson, A. M. McCollum, Y. Li, Multiple
the Clade IIb MPXV will cause discordance in 5. A. K. Rao et al., MMWR Morb. Mortal. Wkly. Rep. 71, 509–516 lineages of monkeypox virus detected in the United States,
(2022). 2021- 2022, Dryad (2022); https://doi.org/10.5061/dryad.
estimates of evolutionary rate and divergence 6. Centers for Disease Control and Prevention, “2022 global map jq2bvq8cd.
time using many standard methods. & case count” (CDC, 2022); https://www.cdc.gov/poxvirus/ 35. GISAID sequence alignment for: C. M. Gigante, B. Korber,
Specific enrichment of the APOBEC3 motif monkeypox/response/2022/world-map.html. M. H. Seabolt, K. Wilkins, W. Davidson, A. K. Rao, H. Zhao,
7. Centers for Disease Control and Prevention, “2022 U.S. map & T. G. Smith, C. M. Hughes, F. Minhaj, M. A. Waltenburg, J. Theiler,
mutations in Clade IIb MPXV since 2017 may case count” (CDC, (2022); https://www.cdc.gov/poxvirus/ S. Smole, G. R. Gallagher, D. Blythe, R. Myers, J. Schulte,
suggest something different in virus-host in- monkeypox/response/2022/us-map.html. J. Stringer, P. Lee, R. M. Mendoza, L. A. Griffin-Thomas, J. Crain,
teractions, facilitating this mutational pattern. 8. C. Happi et al., PLOS Biol. 20, e3001769 (2022). J. Murray, A. Atkinson, A. H. Gonzalez, J. Nash, D. Batra, I. Damon,
9. A. M. Likos et al., J. Gen. Virol. 86, 2661–2672 (2005). J. McQuiston, C. L. Hutson, A. M. McCollum, Y. Li, Multiple
Such a pattern might be caused by sustained
10. K. Brown, P. A. Leggat, Trop. Med. Infect. Dis. 1, 8 (2016). lineages of monkeypox virus detected in the United States,
transmission in a new host or a new route of 11. M. G. Reynolds et al., Emerg. Infect. Dis. 13, 1332–1339 2021- 2022, GISAID EpiPox Downloads (2022); https://doi.org/
infection. The APOBEC3 (A-to-G) locus carries (2007). 10.55876/gis8.220715bk.
seven genes or pseudogenes and is a primate- 12. C. L. Hutson et al., BioMed Res. Int. 2015, 965710 (2015). 36. Code used for APOBEC3 analysis for: C. M. Gigante, B. Korber,
13. J. Hadfield et al., Bioinformatics 34, 4121–4123 (2018). M. H. Seabolt, K. Wilkins, W. Davidson, A. K. Rao, H. Zhao,
specific expansion of the APOBEC family of 14. M. G. Reynolds et al., Am. J. Trop. Med. Hyg. 82, 746–754 T. G. Smith, C. M. Hughes, F. Minhaj, M. A. Waltenburg, J. Theiler,
genes; APOBEC3 is not found in rodents (30), (2010). S. Smole, G. R. Gallagher, D. Blythe, R. Myers, J. Schulte,
which are thought to be the primary reservoir 15. Y. Li, H. Zhao, K. Wilkins, C. Hughes, I. K. Damon,J. Virol. Methods J. Stringer, P. Lee, R. M. Mendoza, L. A. Griffin-Thomas, J. Crain,
169, 223–227 (2010). J. Murray, A. Atkinson, A. H. Gonzalez, J. Nash, D. Batra,
of monkeypox virus in Africa. Additionally, 16. Y. Li, V. A. Olson, T. Laue, M. T. Laker, I. K. Damon, J. Clin. Virol. I. Damon, J. McQuiston, C. L. Hutson, A. M. McCollum, Y. Li,
APOBEC3 has been found at higher levels in 36, 194–203 (2006). Multiple lineages of monkeypox virus detected in the United
mucosal tissues than in skin in humans (31), 17. A. R. O’Toole, A. Rambaut, “Initial observations about putative States, 2021- 2022, Zenodo (2022); https://doi.org/10.5281/
APOBEC3 deaminase editing driving short-term evolution of zenodo.6970457.
so permucosal transmission may provide in- MPXV since 2017” (Virological.org, 2022); https://virological.org/
creased opportunity for APOBEC3 editing of t/initial-observations-about-putative-apobec3-deaminase-editing-
MPXV. Before this outbreak, APOBEC3 editing driving-short-term-evolution-of-mpxv-since-2017/830. AC KNOWLED GME NTS
18. J. Isidro et al., Nat. Med. 28, 1569–1572 (2022). We thank B. Foley for providing a draft alignment of MPXV;
was not recognized or appreciated as a mech- 19. R. S. Harris, J. P. Dudley, Virology 479-480, 131–145 (2015). D. MacCannell, K. Patel, Y. Tao, S. Tong, B. Cherney, and J. Barnes
anism of poxvirus mutation. Two studies in 20. M. T. Liddament, W. L. Brown, A. J. Schumacher, R. S. Harris, from the US Centers for Disease Control and Prevention; A. Kim
vaccinia virus found that the overexpression of Curr. Biol. 14, 1385–1391 (2004). and S. Lukula from the Maryland Department of Health Laboratory;
21. GISAID, www.gisaid.org B. Downes from Fairfax County, Virginia Department of Health;
APOBEC3 had no effect on viral replication and
22. B. Jolly, V. Scaria, J. Infect. 22, 0.1016/j.jinf.2022.08.013 C. Brown, L. Madoff, M. DeMartino, E. Buzby, S. Ash, and J. Hall
that endogenous or overexpressed APOBEC3 (2022). from the Massachusetts Department of Health; J. Jaramillo
was not degraded by vaccinia virus, as it is by 23. J. J. Esposito et al., Science 313, 807–812 (2006). from Dallas County LRN; D. Hairfield and K. Stratton from the
other viruses (32, 33). However, neither of these 24. E. E. Giorgi, H. Li, T. Bhattacharya, G. M. Shaw, B. Korber, mBio Virginia Department of General Services, Division of Consolidated
11, e00324-e20 (2020). Laboratory Services; and the authors from the originating
studies performed sequencing. The fact that the 25. B. F. Keele et al., Proc. Natl. Acad. Sci. U.S.A. 105, 7552–7557 laboratories and the submitting laboratories who generated and
enrichment can be observed at essentially all (2008). shared the GISAID genetic sequence data on which this research is
levels within Clade IIb MPXV since 2017 suggests 26. G. Hobson et al., Euro Surveill. 26, 2100745 (2021). based (GISAID acknowledgments tables S10 and S11 are provided
27. J. D. Salter, R. P. Bennett, H. C. Smith, Trends Biochem. Sci. 41, in the supplementary materials). Funding: No external funding
that it is a recurrent and dominant mutational 578–594 (2016). was used for the Centers for Disease Control and Prevention (CDC)
effect in recent MPXV evolution. Furthermore, 28. J. P. Vartanian, D. Guétard, M. Henry, S. Wain-Hobson, Science effort in this investigation. B.K. was supported by the Laboratory
the majority of the mutations detected among 320, 230–233 (2008). Directed Research and Development program of Los Alamos
29. J. P. Vartanian et al., PLOS Pathog. 6, e1000928 (2010). National Laboratory under project number 20220660ER. Author
expanding outbreak samples were 5′ GA-to-AA 30. A. Jarmuz et al., Genomics 79, 285–296 (2002). contributions: Conceptualization: Y.L., C.M.G., B.K., J.M., I.D.;
(Fig. 2 and table S7), indicating that this mu- 31. C. J. Warren, K. Van Doorslaer, A. Pandey, J. M. Espinosa, Funding acquisition: Y.L., C.M.G., B.K.; Investigation: M.H.S., K.W.,
tational bias is continuing. D. Pyeon, Virus Evol. 1, vev015 (2015). W.D., A.K.R., H.Z., C.M.H., F.M., M.A.W., J.T., S.S., G.R.G., D.B.,
32. M. Kremer et al., Virol. J. 3, 86 (2006). R.M., J.S., P.L., R.M.M., L.A.G., J.C., J.M., A.A., A.H.G., J.N., D.B.,
RE FE RENCES AND N OT ES 33. L. Soday et al., Cell Rep. 27, 1920–1933.e7 (2019). A.M.M., T.G.S.; Methodology: Y.L., B.K., J.T., C.M.G., M.H.S.;
1. Centers for Disease Control and Prevention (CDC), MMWR Morb. 34. Sequence alignments for: C. M. Gigante, B. Korber, M. H. Seabolt, Visualization: C.M.G., B.K., J.T., Y.L.; Writing – original draft: C.M.G.,
Mortal. Wkly. Rep. 52, 561–564 (2003). K. Wilkins, W. Davidson, A. K. Rao, H. Zhao, T. G. Smith, B.K.; Writing – review and editing: C.M.G., B.K., Y.L., I.D., A.M.M.,

C.L.H., J.M. Competing interests: The authors declare no science.org/about/science-licenses-journal-article-reuse. In the Tables S1 to S9
competing interests. Data and materials availability: All data are interest of rapid dissemination of results with immediate public health References (37–43)
available in the main text or public databases and GISAID. MPXV relevance, the author will make the Author Accepted Manuscript MDAR Reproducibility Checklist
genomes were deposited to GenBank under accession numbers (AAM) version available under a CC BY public copyright license. GISAID Acknowledgment Tables S10 and S11
ON563414.3, ON674051, ON675438, ON676703, ON676704,
ON676705, ON676706, ON676707, and ON676708. Alignments
and code have been made available online (34–36). License SUPPLEMENTARY MATERIALS
information: Copyright © 2022 the authors, some rights reserved; science.org/doi/10.1126/science.add4153 Submitted 12 June 2022; accepted 10 October 2022
exclusive licensee American Association for the Advancement of Materials and Methods Published online 20 October 2022
Science. No claim to original US government works. https://www. Figs. S1 to S5 10.1126/science.add4153

online @sciencecareers.org THE TIME IS NOW
EXECUTIVE DIRECTOR,
INTEGRATIVE LIFE SCIENCES INSTITUTE (ILSI)
UNIVERSITY OF OKLAHOMA (NORMAN, OK)
We strive for excellence in research, teaching, and service, while
fostering a robust exchange of ideas on local and global health–related issues.
TO APPLY: APPLY.INTERFOLIO.COM/113559
The Indiana University School of Public Health
– Bloomington invites applications for multiple The University of Oklahoma is seeking nominations and applications for
openings in administrative faculty, tenured/ the position of inaugural Executive Director of the ILSI and at the rank of
tenure-track faculty, and post-doctoral fellows in all
departments. SPH-B seeks individuals passionate
Full Professor. The academic home department will be aligned with the
about our mission to promote health for all members successful candidate’s disciplinary interests.
of society through rigorous and impactful research,
excellence in teaching and mentoring, and service to The Executive Director, who will report directly to OU’s Vice President for
the university, profession, and the community.
Research and Partnerships, will provide strong leadership in guiding the
SPH-B is committed to growing and sustaining a mission of ILSI, which is to advance positive health outcomes through
multi-culturally diverse, equitable, and inclusive convergence research to create new diagnostics, therapies, policies, and
living, learning, and working environment. Of practices that will enable recognition of and response to continuing and
the current 100 T/TT faculty, 35% are from emerging disease threats and address social and environmental risk factors
underrepresented racial and ethnic groups as
defined by the National Science Foundation (NSF). impacting health equity.
We are currently accepting applications for the The successful candidate will be an experienced and dynamic leader with
following administrative faculty positions: exceptional scholarly credentials in life sciences research. The candidate will
- Chairperson of the Department of Environmental promote and strengthen collaborative efforts between the faculty, students, and
and Occupational Health
- Associate Dean for Faculty Affairs
staff across our three campuses, to incentivize and develop transdisciplinary
- Associate or Executive Assistant Dean for teams in areas of strategic importance to the ILSI. Establishing strong
Academic Affairs and Enrollment partnerships with OU Health Sciences Center and the Oklahoma Medical
Research Foundation is an important objective for ILSI. In addition, the
We are currently accepting applications for
multiple tenured/tenure-track positions in the
successful candidate will build a robust and sustainable financial portfolio
job focus: diversity
departments of: of dedicated internal and extramural funding for the operational costs of the
- Applied Health Science institute and in support of its convergence research mission. Applicants must
- Environmental and Occupational Health hold a Ph.D. and/or M.D. in life sciences research or related area, have at least
- Epidemiology and Biostatistics two years of relevant academic or corporate leadership experience, and show
- Kinesiology
- Health & Wellness Design
evidence of a nationally recognized and well-funded research program.
Learn more about For more information contact: Ann West ([email protected])
these and other #YearOfLeverage The University of Oklahoma is an equal opportunity institution.
openings at
go.iu.edu/4BrJ
For more information, please visit http://www.ou.edu/eoo/
TENURE TRACK ASSISTANT PROFESSOR POSITION

IN POPULATION BIOLOGY 2023 Schaefer Research Scholars Program Awards
The Department of Ecology and Evolutionary Biology (EEB) at UCLA is The Vagelos College of Physicians and Surgeons (VP&S) is pleased
searching for a tenure track Assistant Professor in population biology. We seek to announce the 2023 Schaefer Research Scholars Program Awards.
candidates who apply quantitative approaches to test theory-driven questions
on the ecology or evolution of populations. The successful candidate will be The Awards, made possible through a bequest from Dr. Ludwig
expected to develop a rigorous, externally funded research program and teach Schaefer, are made annually to four research scientists who have dis-
at both the undergraduate and graduate levels with innovative pedagogical
approaches. We are especially interested in candidates who can develop courses
tinguished themselves in human physiology, as broadly defined, and
that satisfy the UCLA diversity requirement or include topics of diversity, whose current work is of outstanding merit. The proposed research
equity and inclusion and how they relate to our field. We welcome candidates must have the potential to illuminate the field. Two awards are made
who demonstrate a commitment to equity and inclusion through research,
teaching, mentoring, community engagement and/or service and who apply to research scientists residing or working in North or South America,
inclusive mentoring practices. Necessary qualifications include a PhD degree in and two awards are made to research scientists residing or working
a relevant discipline and a strong background in quantitative methods. Evidence
of sustained research productivity through journal publications is required.
outside North or South America. Each award consists of a $50,000
cash prize and up to $200,000 in direct research support.
Submit application packages online through https://recruit.apo.ucla.edu/
JPF07931 and include the following: 1) cover letter; 2) curriculum vita;
3) statement of research interests; 4) statement of teaching expertise; Applications must include a cover sheet; a research proposal (one
5) statement of formal and informal activities to promote equity, diversity and page); a research budget (not to exceed $200,000 in total direct
inclusion; 6) names and contact info of three referees (reference letters will only
be requested of select candidates), and 7) reference check authorization release
costs) delineated by cost category (salary, fringe, supplies, etc.) for
form. The University of California seeks to recruit and retain a diverse workforce one year (7/1/2023–6/30/2024); a curriculum vitae (not to exceed 10
as a reflection of our commitment to serve the people of California, to maintain pages); and a page summarizing applicant’s research support. Inter-
the excellence of the University, and to offer our students richly varied
disciplines, perspectives, and ways of knowing and learning. Complete nal candidates must obtain a nomination letter from their Department
applications must be submitted by November 16, 2022. For full information Chair. External candidates must present letters from the Dean or
about this position, please visit: https://recruit.apo.ucla.edu/JPF07931.
equivalent in their home institution as well as from the Columbia Uni-
The University of California is an Equal Opportunity/Affirmative Action versity Irving Medical Center collaborator, if applicable.
Employer. All qualified applicants will receive consideration for employment
without regard to race, color, religion, sex, sexual orientation, gender identity,
national origin, disability, age or protected veteran status. For the complete Nomination deadline: December 12, 2022, at 5:00 p.m. (EST)
University of California nondiscrimination and affirmative action policy, see: To apply visit https://www.ps.columbia.edu/schaefer
UC Nondiscrimination & Affirmative Action Policy.
(http://policy.ucop.edu/doc/4000376/NondiscrimAffirmAct)
and submit all required materials.
Awardees will be notified in February 2023.
online @sciencecareers.org
Vice Chair & Director of the Radiation and
Molecular Oncology Section
The Department of Radiation Oncology at the University of Texas Southwestern Medical
Center is recruiting a senior biology faculty member and/or physician-scientist in
biology to join, build, and lead the department’s Radiation and Molecular Oncology
section. The section has existing strengths within mechanistic studies of DNA damage/
repair and radiotherapy translation in combination with novel systemic therapies, and there
is a strong desire to expand/enhance research into realms of immune and metabolic
modulation of cancer behavior and therapy responsiveness.
The director will provide section leadership and oversee research investigators and day-to-day operations of a large research infrastructure, including existing
core facilities and 25,000 sq. ft. of laboratory space assigned to multiple researchers in the department. Candidate will report directly to the department Chair.
The section currently has 18 faculty, including 10 principal investigators. Successful candidate will be provided resources to develop integrated programs
with both the Clinical and Medical Physics and Engineering divisions of the department as well as the ability to pursue novel investigations and treatments.
Candidates should have an interest in initiating new translational projects and collaborative research both within and outside the department, as well as those
associated with the Harold C. Simmons Comprehensive Cancer Center. If close to promotion timeline, associate professor candidates may be taken.
Skills & Qualifications:

• Required record of excellence in research accompanied by a strong publication and funding history.
• Interest in some of the following topics that include but are not limited to radiation biology/oncology, cancer biology, immunology, metabolic study,
DNA repair, combination therapies, and the development of biomarkers.
• Open to initiating translational projects and collaborative research with clinicians and scientists across disciplines.
• Strong interpersonal skills and motivation for program building, as well as interests in basic and translational science training.
• Must have a Ph.D., M.D., or M.D./Ph.D. degree or equivalent. Candidates who intend to maintain a clinical practice must have a medical degree from an
job focus: diversity

ACGME-approved medical school or equivalent and have completed a radiation oncology residency program from an approved institution.
To learn more about

this position or to apply:
myIDP:
A career plan customized
for you, by you.
CALL FOR NOMINATIONS: SCOLNICK PRIZE IN NEUROSCIENCE

The McGovern Institute for Brain Research is accepting nominations for the 19th annual Edward M.
Scolnick Prize in Neuroscience. The Prize recognizes an outstanding discovery or significant advance in
the field of neuroscience. The prize is $200,000. The recipient presents a public lecture at MIT, hosted by the
McGovern Institute and followed by a dinner in Spring 2023.
There’s only one
Nomination Deadline: December 15, 2022
Nomination procedures:
Features in myIDP include: Candidates for the award must be nominated by individuals affiliated with universities, hospitals, medicals
 Exercises to help you examine your schools, or research institutes, with a background in neuroscience. Self-nomination is not permitted. Each
skills, interests, and values. nomination should include:
• A biosketch or CV of the nominee;
 A list of 20 scientific career paths • A letter of nomination with a summary and analysis of the major contributions of the nominee to the field
with a prediction of which ones best of neuroscience.
fit your skills and interests. • Up to two representative reprints will be accepted.
Selection Procedure:
Visit the website and start • Members of the selection committee and faculty affiliated with MIT are not eligible.
planning today! • Announcement of the award recipient will be made in January 2023.
myIDP.sciencecareers.org • Recipient must attend all events to be awarded the prize.
Past Scolnick Prize Recipients:

2022: David Ginty, Harvard University; 2020: Joshua Sanes, Harvard University; 2019: Richard Huganir,
In partnership with: Johns Hopkins University; 2018: David J. Anderson, HHMI, Caltech; 2017: Catherine Dulac, HHMI,
Harvard University; 2016: Cornelia Bargmann, HHMI, The Rockefeller University; 2015: Charles Gilbert,
The Rockefeller University; 2014: Huda Zoghbi, HHMI, Baylor University; 2013: Thomas Jessell, HHMI,
Columbia University; 2012: Roger Nicoll, UCSF; 2011: Bruce McEwen, The Rockefeller University; 2010:
Lily and Yuh-Nung Jan, UCSF; 2009: Jeremy Nathans, Johns Hopkins University; 2008: Michael Davis,
Emory University; 2007: David Julius, UCSF; 2006: Michael Greenberg, Children’s Hospital/HMS; 2005:
Judith Rapoport, NIH; 2004: Masakazu Konishi, CalTech
Send nomination packet to: [email protected]

JOB FOCUS: DIVERSITY
ASSISTANT PROFESSOR - BACTERIOLOGIST
Department of Immunobiology
College of Medicine, University of Arizona Health Sciences
What's Your
The Department of Immunobiology at the University of Arizona (UA) College of Medicine-Tucson (COM-T)
is seeking applicants for a tenure-track Assistant Professor position in bacteriology. We recognize the power
Next Career
of a diverse community and strongly encourage qualified individuals with varied experiences, perspectives,
and backgrounds to apply.
Move?
Applicants must have demonstrated the ability to establish an independent research program and obtain
extramural funding. Individuals whose research advances the understanding of host-microbe interactions,
including but not limited to respiratory diseases with a bacterial etiology, bacterial metabolism, regulation of
host responses, microbiomes and/or mucosal immunology during dysbiosis and/or pathogenesis are strongly
encouraged to apply. Applicants must have postdoctoral experience and a PhD, MD, or equivalent degree
in their field of study. The successful applicant is expected to obtain extramural funding for their research
program and participate in graduate student mentoring and graduate and medical student teaching. They are
From networking
also expected to interact with other faculty on larger collaborative scientific projects. Along with submitting
your CV and Cover Letter, the applicant should arrange for 3 letters of recommendation to be sent separately to mentoring to
by senior faculty with extensive knowledge of the candidate. These letters should be sent to Tonya
Fotheringham ([email protected]). Salary and start-up funds will be commensurate with the applicant’s
qualifications and the nature of the applicant’s project.
evaluating your
Information on the highly collaborative nature of the Department of Immunobiology and the
skills, find answers
interdepartmental/interdisciplinary ABBS graduate (PhD) program to which it belongs can be obtained
through the links on the department website (immunobiology.arizona.edu). Additional information on to your career
UA Centers, Institutes, and core facilities can be found at the UA COM-T website (medicine.arizona.
edu/about-college/centers). UA is in the midst of the 4th Industrial Revolution strategic plan designed to
significantly increase its research capacity over the next 10 years.
questions on
The University of Arizona has been recognized on Forbes list of America’s Best Employers in the United Science Careers
States and has been awarded the Work-Life Seal of Distinction by World@Work! For more information about
working at the University of Arizona and relocations services, please go to talent.arizona.edu.
For full details and qualifications, and to complete an on-line application, see req12237 at To view the complete
talent.arizona.edu.
collection, visit
Outstanding UA Benefits! The University of Arizona is an Equal Opportunity Employer Minorities/Women/
Vets/Disabled. ScienceCareers.org/
booklets
YOUR NEXT
BIG SCIENTIFIC
DISCOVERY:
A NEW JOB.
Find your next job at ScienceCareers.org
Faculty position
in Hydraulic Engineering &
Fluid-Structure Interactions
at the Ecole polytechnique fédérale
de Lausanne (EPFL)
The EPFL School of Architecture, Civil and Environmental Engineering (ENAC)
invites applications for an Assistant (tenure-track) or Associate (tenured)
Professor of Hydraulic Engineering & Fluid-Structure Interactions in the
Institute of Civil Engineering.
The Institute of Civil Engineering in ENAC carries out basic and translational
research spanning fundamental understanding of Hydraulic Engineering,
Structures and Materials, Transportation and Networks, and Geosystems and
Geomechanics. It covers a diverse portfolio in research, teaching and innova-
tive technology development across transversal themes: understanding and
adapting to climate change, natural hazards, infrastructure resilience and sus-
tainability, design for re-use, geo-energy and geosystems.
Areas of interest include but are not limited to: fundamental research in the
development of renewable energy systems such as dams and hydropower
systems, advanced modelling of hydraulic infrastructure, fluid-structure
interactions, resource conservation, and risk analysis for extremes. Additional
areas include experimental methods as well as sensing data for the develop-
ment of techniques that leverage digital twins for reliable maintenance and
safety assessment of hydraulic infrastructure, as well as analysis of extreme
natural hazards and adaptation to climate change.
The successful candidate will have a strong research background in Hydrau-
lic Engineering and Fluid-Structure Interactions as well as the potential to
develop an innovative and internationally recognized research program.
As a faculty member in the Civil Engineering Section, the appointee will excel
in undergraduate and graduate teaching and in guiding research students.
EUROSCIENCE POSTDOCTORAL PROGRAMME
EPFL, with its main campus located in Lausanne on the shores of Lake Geneva,
is a dynamically growing and well-funded institution fostering excellence
Linköping University is one of Sweden’s six large universities,
and diversity. It is well equipped with experimental and computational infra-
currently enrolling 35 900 students. The Centre for Systems
structure and offers a fertile environment for research collaboration between
Neurobiology involves some 50 independent research groups,
various disciplines. The EPFL environment is multilingual and multicultural,
from the Faculty of Medicine and the Faculty of Science and
with English serving as a common interface. EPFL offers significant start-up
Engineering, as well as the University Hospital.
resources and internationally competitive salaries and benefits as well as
excellent experimental and computational facilities.
The Centre for Systems Neurobiology is now seeking
Postdoctoral Fellows within several neuroscience research The following documents are requested in PDF format: cover letter including
areas: Animal behavior/Behavioral genetics/ Evolution (Urban a statement of motivation, curriculum vitae, publication list, concise state-
Friberg, Per Jensen), Clinical neurophysiology and ments of research vision and teaching interests (up to five pages for each),
neuromodulation (Magnus Thordstein), Ion channels (Antonios and the names and addresses (including e-mail) of at least three (for the rank
Pantazis), Neurodegenerative disease (Peter Nilsson), of Assistant Professor) or five (for the rank of Associate Professor) references
Neuroendocrinology/neuroimmunology (Anders Blomqvist, who have already agreed to supply a letter upon request. Applications should
David Engblom), Psychiatric neuroscience and addiction (Eric be uploaded to the EPFL recruitment web site:
Augier), and Sensory and cognitive systems and pain (Sarah
McIntyre, Saad Nagi, Håkan Olausson, Marcin Szczot). https://facultyrecruiting.epfl.ch/position/40599565
Formal evaluation of the applications will begin on December 1st, 2022.
For full consideration please apply as soon as possible, and no
later than November 30th. Further enquiries should be made to:
Prof. Dimitrios Lignos
For more details regarding the Centre, the different research Chair of the Search Committee
labs involved in the programme, and to submit a letter-of-intent e-mail: [email protected]
please go to: https://liu.se/en/research/center-for-systems- For additional information on EPFL, please consult the websites: www.epfl.ch,
neurobiology. enac.epfl.ch, iic.epfl.ch, sgc.epfl.ch.
For information regarding the university and the region, please go EPFL is an equal opportunity employer and a family-friendly university that
to: liu.se, eastsweden.com. is committed to increasing the diversity of its faculty. It strongly encourages
women to apply.
WO RKING LIFE
By Isabella Bennett
Our labor isn’t free
I
saw the bright yellow envelope tucked beneath my windshield wiper and my heart sank. Another
parking ticket. Without looking at the amount, I slid the ticket into my glovebox next to its twin,
opened the trunk, and handed the faculty candidate beside me the suitcase I had stored for
them. I was in the second year of my Ph.D. and the fifth month of serving on a faculty search
committee. When I was invited to join the committee, the faculty members in charge told me it
would be “a lot of reading but a good experience and some fancy dinners.” I felt honored to be
included. Little did I know the costs my volunteer commitment would entail.
It started off great. I read about versity coffee shop. I appreciated

40 applications and the associ- the recognition and figured the
ated novella-length CVs, evaluat- caffeine would help as I tried to
ing each using an extensive rubric. catch up on my research. I thought
But my stress level rose when two that was that.
grants I’d submitted—the first I’d A couple weeks after the search
ever written, with little expecta- concluded, however, I reconnected
tion of success—were funded. I with a mentor at another institu-
worried the faculty search would tion. As we caught up, she asked
mean I couldn’t give my newly how the university had compen-
funded project the time and en- sated me for such a significant
ergy it deserved, but I felt I owed it commitment. When I said some-
to the candidates and my mentors thing about food and experience,
on the committee to continue. So, I she told me how she arranges for
stretched myself to the limit. I told students in her department to
my partner not to expect to see be paid for any service work, at
much of me, much less have any well above minimum wage. The
help around the house. I stopped thought had never occurred to me.
responding to calls from friends
and family.
“The expectation should never After that call, I couldn’t stop
thinking about why I wasn’t paid
After several weeks, my stress be that experience for those 5 months of work. A park-
started to manifest physically. An ing pass and gas reimbursement
old back injury flared up and I and a line on your resume … would have been helpful, at least.
found myself lying on the floor,
unable to walk. With the help of
are fair compensation.” But as an early-career grad student
still finding my footing, it didn’t
muscle relaxers and treatment, I occur to me to advocate for myself.
managed to attend the virtual candidate interviews from I don’t think anyone was maliciously trying to take advan-
my couch, and eventually got back on my feet. But there tage of me; it’s just so ingrained in academic culture that
was no question I had too many responsibilities, too lit- graduate students will perform uncompensated labor that
tle time, and now a handful of hospital bills to top it off. no one—including me—gave it a second thought.
Still, I’d come this far, and I was committed to seeing the Countless other students have served their universities
search through. essentially for free, too. This needs to change. I spent more
Months later, when we brought the top candidates to than 200 hours on this search, and although it was a good
campus for in-person interviews, I shared meals with learning experience, it cost me a lot more than a pair of
them, introduced them before their lectures, and drove $25 parking tickets. My research, health, and personal life
ILLUSTRATION: ROBERT NEUBECKER

them around the campus, accumulating those yellow uni- suffered. The expectation should never be that experience
versity parking tickets. I didn’t have the time or energy to and a line on your resume—and perhaps a gift card—are
dispute them. I was behind on my research, drained, and fair compensation. j
socially starved.
After the last interview, the department sent an email Isabella Bennett is a graduate student at the University of Vermont.
thanking me for my work with a $25 gift card to the uni- Send your career story to [email protected].

An estate gift
to AAAS
Going all the way back to 1848, our founding year, the American “As a teacher and instructor, I bear responsibility for the
Association for the Advancement of Science (AAAS) has been younger generations. If you have extra resources, concentrate
them on organizations, like AAAS, that are doing work for all.”
deeply committed to advancing science, engineering and
—Prof. Elisabeth Ervin-Blankenheim, 1848 Society member
innovation around the world for the benefit of all people.
Today, we are dedicated to advocating for science and scientific

If you intend to include AAAS in your estate plans, provide this
evidence to be fully and positively integrated into public policy and information to your lawyer or financial adviser:
for the community to speak with one voice to advance science and
Legal Name: American Association for the Advancement of Science
engineering in the United States and around the world.
Federal Tax ID Number: 53-0196568
By making AAAS a beneficiary of your will, trust, retirement plan Address: 1200 New York Avenue, NW, Washington, DC 20005
or life insurance policy, you will become a member of our 1848 If you would like more information on making an estate gift to
Society and will help fuel our work on behalf of science and society AAAS, cut out and return the form below or send an email to
– including publishing the world’s most promising, innovative [email protected]. Additional details are also available
research in the Science family of journals and engaging in the online at www.aaas.org/1848Society.
issues that matter locally, nationally and around the world.
cut here
Yes, I would like more information about joining the AAAS 1848 Society.
PLEASE CONTACT ME AT:
Name:
Address:
City: State: Zip code: Country:
Email: Phone:
RETURN THIS FORM TO:

AAAS Office of Philanthropy and Strategic Partnerships • 1200 New York Avenue, NW • Washington, DC 20005 USA
science.org/journal/signaling
PUT YOUR
RESEARCH
OUT IN
FRONT
Submit your research:
cts.ScienceMag.org
Twitter: @SciSignal
Facebook: @ScienceSignaling

Science - 4 11 2022

Uploaded by

Copyright:

Available Formats

Science - 4 11 2022

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Science - 4 11 2022

Uploaded by

Copyright:

Available Formats

Berlin shows how biodiversity Neutrinos from an active Eppendorf & Science Prize for neural

Register today and join us March 2-5, 2023!

BE A PART OF THE CONVERSATION

Diversity, Equity and Inclusion Research Funding and Infrastructure

Science Communication Artificial Intelligence

STEM Education Information Technology and Security

International S&T Collaboration Science and Values

Social and Behavioral Sciences Global and Public Health

Science Policy Climate Change

(Left) Junsuk Rho, (Far right) Soojin Park

The transformation of theory into practice

Wild boars have adapted

459 Teeth record pneumonia—and racism 470 Arctic wildfires at a warming

PHOTO: GREGOR FISCHER/PICTURE ALLIANCE/GETTY IMAGES

450 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE

543 Plant science

Molecular glue CELMoD compounds are

474 & 538

SCIENCE science.org 4 NOVEMEBER 2022 • VOL 378 ISSUE 6619 451

452 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE

Measure emissions to manage emissions

to 2021’s staggering 414.72 ppm global

SCIENCE science.org 4 NOVEMEBER 2022 • VOL 378 ISSUE 6619 455

Science society lists Earth’s top ‘geoheritage’ sites

PHOTO: PETER THOMPSON/HERITAGE IMAGES/GETTY IMAGES

456 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE

Academy probe clears biologist

SCIENCE science.org 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 457

Should Webb telescope’s data be open to all?

458 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE

the Terry collection ethically,” says Carlina

SCIENCE science.org 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 459

In 1923, City Hospital #2 in St. Louis, where the

complex history of racism, she and her col-

PHOTO: ST. LOUIS POST-DISPATCH/POLARIS

460 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE

A cleaner way to mine

SCIENCE science.org 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 461

Prominent astronomer barred after complaints

462 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE

Black institution in Washington, D.C., who

SCIENCE science.org 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 463

for research at Northern Illinois University SCIENCE AND ENVIRONMENT

464 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE

You could be next to win this prize.

Learn more at: eppendorf.com/prize

Meet the Winner 2022

Learn more at: eppendorf.com/prize

466 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE

SCIENCE science.org 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 467

THE BERLIN SCHOOL OF ECOLOGY, as it came

468 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE

SCIENCE science.org 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 469

Wildfires are becoming bigger in

Arctic wildfires at a warming threshold

470 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE

Peat neurotransmitter release by the neuron. An

SCIENCE science.org 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 47 1

472 4 NOVEMBER 2022 • VOL 378 ISSUE 6619 science.org SCIENCE