20 AUGUST 2021 Science

Table of contents :
CONTENTS
NEWS IN BRIEF
News at a glance
NEWS IN DEPTH
Israel’s grim warning: Delta can overwhelm shots
Nazi massacre unearthed in Poland ‘was really a horror’
Laser-powered fusion effort nears ‘ignition’
Dire warming report triggers calls for more action from China
Antibody acts like short-term malaria vaccine
FEATURES
Evolving threat
Do chronic infections breed dangerous new variants?
POLICY FORUM
Keep climate policy focused on the social cost of carbon
PERSPECTIVES
The fracking concern with water quality
Exploring the path of the variable resistance
Tracking severe malaria disease
Structural hierarchy defeats alloy cracking
Ecology in the age of automation
BOOKS ET AL.
The children’s climate crusade
Before the Big Bang became scientific dogma
LETTERS
Embrace kelp forests in the coming decade
Mexico’s final death blow to the vaquita
Piecing together an African peace park
RESEARCH IN BRIEF
From Science and other journals
REVIEW
Glaciohydrology of the Himalaya-Karakoram
RESEARCH ARTICLES
A DNA repair pathway can regulate transcriptional noise to promote cell fate transitions
Accurate prediction of protein structures and interactions using a three-track neural network
Mechanisms that ensure speed and fidelity in eukaryotic translation termination
Mammalian retrovirus-like protein PEG10 packages its own mRNA and can be pseudotyped for mRNA delivery
Spatiotemporal invasion dynamics of SARS-CoV-2 lineage B.1.1.7 emergence
REPORTS
Large-sample evidence on the impact of unconventional oil and gas development on surface waters
Stabilizing perovskite-substrate interfaces for high-performance perovskite modules
Spatiotemporal characterization of the field-induced insulator-to-metal transition
Hierarchical crack buffering triples ductility in eutectic herringbone high-entropy alloys
Rare variant MX1 alleles increase human susceptibility to zoonotic H7N9 influenza virus
Babbling in a vocal learning bat resembles human infant babbling
Malaria infection and severe disease risks in Africa
Masitinib is a broad coronavirus 3CL inhibitor that blocks replication of SARS-CoV-2
DEPARTMENTS
Editorial
Working Life
Science Staff
Science Careers

Citation preview

SARS-CoV-2 is evolving fast. What will it do next? p. 844

Effects on surface water from hydraulic fracturing pp. 853 & 896

Baby bats babble like human babies p. 923

$15 20 AUGUST 2021 sciencemag.org

PREDICTING

STRUCTURES

Deep learning accurately folds proteins

0820Cover.indd 829

p. 871

8/16/21 12:01 PM

Frontline Fabrics NC State University isn’t just home to the nation’s only dedicated textiles college. We also created the world’s first accredited program for nonwovens, the engineered fabrics used in everything from air conditioning systems to fireproof garments. During the pandemic, our Nonwovens Institute piloted new face masks and filters — and churned out millions of meters of protective material for first responders and the military. Pioneering research. Lifesaving results. results.ncsu.edu

0820Product.indd 830

8/12/21 7:37 AM

Science Webinars help you keep pace with emerging scientific fields! Stay informed about scientific breakthroughs and discoveries. Gain insights into current research from top scientists. Take the opportunity to ask questions during live broadcasts. Get alerts about upcoming free webinars.

Sign up at: webinar.sciencemag.org/stayinformed

0820Product.indd 831

8/12/21 7:37 AM

CONTENTS

2 0 AU G US T 2 0 2 1 • VO LU M E 3 7 3 • I S S U E 6 5 57

850 841 Laser-powered fusion effort nears ‘ignition’

IN BRIEF

By D. Clery

National Ignition Facility’s latest fusion shot records a major jump in energy yield

836 News at a glance

842 Dire warming report triggers calls for more action from China

IN DEPTH

Climate advocates want the world’s largest carbon producer to level off emissions soon and aim for “neutrality” by 2050 By L. Pike

838 Israel’s grim warning: Delta can overwhelm shots With early vaccination and outstanding data, country is the world’s real-life COVID-19 lab

PERSPECTIVES

853 The fracking concern with water quality Tapping into oil and gas reserves comes at the expense of contaminating water By E. Hill and L. Ma REPORT p. 896

854 Exploring the path of the variable resistance Resistive switching studies pave the way to neuromorphic information technologies

By M. Wadman

843 Antibody acts like short-term malaria vaccine

839 Nazi massacre unearthed in Poland ‘was really a horror’

Monoclonal antibodies protected people from infection in a small “challenge” trial By J. Cohen

Excavation finds evidence of both perpetrators and victims of WWII atrocity

FEATURES

By A. Curry

844 Evolving threat

855 Tracking severe malaria disease

New variants have changed the face of the pandemic. What will the virus do next?

Malaria infection prevalence predicts malaria mortality—at least for now

By K. Kupferschmidt

By T. Taylor and L. Slutsker

844 Beta Alpha

By K. Kupferschmidt

Gamma

INSIGHTS

Wuhan strain

Epsilon Delta

Eta

Kappa

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820TOC.indd 832

REPORT p. 907

REPORT p. 926

857 Structural hierarchy defeats alloy cracking Internal herringbone structures in a ductile multicomponent alloy enable crack tolerance By X. An REPORT p. 912

SARS-CoV-2 variants are arrayed on an “antigenic map,” one way of showing how the virus has evolved since since late 2019.

832

848 Do chronic infections breed dangerous new variants?

By H. Hilgenkamp and X. Gao CREDITS: (PHOTO) GORDON WELTERS/LAIF/REDUX; (GRAPHIC) N. DESAI/SCIENCE; (DATA) DEREK SMITH/UNIVERSITY OF CAMBRIDGE; DAVID MONTEFIORI/DUKE UNIVERSITY

NEWS

POLICY FORUM

858 Ecology in the age of automation

850 Keep climate policy focused on the social cost of carbon

Technology is revolutionizing the study of organisms in their natural environment

A proposed shift away from the SCC is ill advised By J. E. Aldy et al.

By T. H. Keitt and E. S. Abelson sciencemag.org SCIENCE

8/17/21 6:19 PM

BOOKS ET AL.

860 The children’s climate crusade Prepared on behalf of young people seeking climate action, a report outlines the US federal government’s failures

853 & 896

By M. B. Gerrard

861 Before the Big Bang became scientific dogma A dual biography traces the entangled efforts of a pair of contentious cosmologists By S. Mitton LETTERS

863 Embrace kelp forests in the coming decade By C. J. Feehan et al.

863 Mexico’s final death blow to the vaquita By C. Sonne et al.

864 Piecing together an African peace park By S. A. Osofsky and R. D. Taylor

RESEARCH

882 RNA delivery

926 Malaria

Mammalian retrovirus-like protein PEG10 packages its own mRNA and can be pseudotyped for mRNA delivery M. Segel et al.

Malaria infection and severe disease risks in Africa R. S. Paton et al.

931 Coronavirus 889 Coronavirus Spatiotemporal invasion dynamics of SARS-CoV-2 lineage B.1.1.7 emergence M. U. G. Kraemer et al. REPORTS

896 Hydraulic fracturing

IN BRIEF

Large-sample evidence on the impact of unconventional oil and gas development on surface waters P. Bonetti et al.

866 From Science and other journals

PERSPECTIVE p. 853

REVIEW

902 Solar cells

869 Glaciers Glaciohydrology of the Himalaya-Karakoram M. F. Azam et al.

Stabilizing perovskite-substrate interfaces for high-performance perovskite modules S. Chen et al.

REVIEW SUMMARY; FOR FULL TEXT: DOI.ORG/10.1126/SCIENCE.ABF3668

907 Oxide electronics

RESEARCH ARTICLES

Spatiotemporal characterization of the field-induced insulator-to-metal transition J. del Valle et al.

870 Systems biology A DNA repair pathway can regulate transcriptional noise to promote cell fate transitions R. V. Desai et al.

PERSPECTIVE p. 854

912 Metallurgy

DOI.ORG/10.1126/SCIENCE.ABC6506

Hierarchical crack buffering triples ductility in eutectic herringbone highentropy alloys P. Shi et al.

871 Protein folding

PERSPECTIVE p. 857

Accurate prediction of protein structures and interactions using a three-track neural network M. Baek et al.

918 Influenza

RESEARCH ARTICLE SUMMARY; FOR FULL TEXT:

EDITORIAL p. 835

Rare variant MX1 alleles increase human susceptibility to zoonotic H7N9 influenza virus Y. Chen et al.

PHOTO: DAVID MCNEW/GETTY IMAGES

876 Translation Mechanisms that ensure speed and fidelity in eukaryotic translation termination M. R. Lawson et al.

PERSPECTIVE p. 855

Masitinib is a broad coronavirus 3CL inhibitor that blocks replication of SARS-CoV-2 N. Drayman et al. DEPARTMENTS

835 Editorial Banking on protein structural data By Jeremy Berg RESEARCH ARTICLE p. 871

938 Working Life Learning to unplug By Eric R. Wengert

ON THE COVER New deep learning software can predict protein structures in minutes by simultaneously considering one-, two-, and three-dimensional information (sequence, distance, and coordinates, respectively), allowing the network to collectively reason about the relationship between a protein’s amino acid sequence and its folded structure. The software can also be used to build models of complex biological assemblies from the sequences of interacting partners in a fraction of the time previously required. See page 871. Illustration: V. Altounian/ Science; Data: Minkyung Baek/University of Washington

923 Vocal development Babbling in a vocal learning bat resembles human infant babbling A. A. Fernandez et al.

Science Staff ............................................. 834 Science Careers .........................................937

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 © 2021 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): $2148; 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.sciencemag.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 sciencemag.org

0820TOC.indd 833

20 AUGUST 2021 • VOL 373 ISSUE 6557

833

8/17/21 6:20 PM

Editor-in-Chief Holden Thorp, [email protected]

BOARD OF REVIEWING EDITORS (Statistics board members indicated with S)

Executive Editor Monica M. Bradford Editors, Research Valda Vinson, Jake S. Yeston Editor, Insights Lisa D. Chong DEPUTY EDITORS Julia Fahrenkamp-Uppenbrink (UK), Stella M. Hurtley (UK), Phillip D. Szuromi, Sacha Vignieri SR. EDITORIAL FELLOW Andrew M. Sugden (UK) SR. EDITORS Gemma Alderton (UK), Caroline Ash (UK), Brent Grocholski, Pamela J. Hines, Di Jiang, Marc S.

Lavine (Canada), Yevgeniya Nusinovich, Ian S. Osborne (UK), Beverly A. Purnell, L. Bryan Ray, H. Jesse Smith, Keith T. Smith (UK), Jelena Stajic, Peter Stern (UK), Valerie B. Thompson, Brad Wible, Yuen Yiu, Laura M. Zahn ASSOCIATE EDITORS Michael A. Funk, Priscilla N. Kelly, Tage S. Rai, Seth Thomas Scanlon (UK), Yury V. Suleymanov LETTERS EDITOR Jennifer Sills LEAD CONTENT PRODUCTION EDITORS Harry Jach, Lauren Kmec CONTENT PRODUCTION EDITORS Amelia Beyna, Jeffrey E. Cook, Chris Filiatreau, Julia Katris, Nida Masiulis, Abigail Shashikanth, Suzanne M. White SR. EDITORIAL COORDINATORS Carolyn Kyle, Beverly Shields EDITORIAL COORDINATORS Aneera Dobbins, Joi S. Granger, Jeffrey Hearn, Lisa Johnson, Maryrose Madrid, Ope Martins, Shannon McMahon, Jerry Richardson, Hilary Stewart (UK), Alana Warnke, Alice Whaley (UK), Anita Wynn PUBLICATIONS ASSISTANTS Jeremy Dow, Alexander Kief, Ronmel Navas, Brian White EXECUTIVE ASSISTANT Jessica Slater ASI DIRECTOR, OPERATIONS Janet Clements (UK) ASI SR. OFFICE ADMINISTRATOR Jessica Waldock (UK)

News Editor Tim Appenzeller NEWS MANAGING EDITOR John Travis INTERNATIONAL EDITOR Martin Enserink DEPUTY NEWS EDITORS Elizabeth Culotta, Lila Guterman, David Grimm, Eric Hand (Europe), David Malakoff SR. CORRESPONDENTS Daniel Clery (UK), Jon Cohen, Jeffrey Mervis, Elizabeth Pennisi ASSOCIATE EDITORS Jeffrey Brainard, Kelly Servick, Catherine Matacic NEWS REPORTERS Adrian Cho, Jennifer Couzin-

Frankel, Jocelyn Kaiser, Rodrigo Pérez Ortega (Mexico City), Robert F. Service, Erik Stokstad, Paul Voosen, Meredith Wadman INTERNS Menna Ibrahim, Sofia Moutinho, Anil Oza, Alex Viveros CONTRIBUTING CORRESPONDENTS Warren Cornwall, Andrew Curry (Berlin), Ann Gibbons, Sam Kean, Eli Kintisch, Kai Kupferschmidt (Berlin), Andrew Lawler, Mitch Leslie, Eliot Marshall, Virginia Morell, Dennis Normile (Tokyo), Elisabeth Pain (Careers), Charles Piller, Michael Price, Tania Rabesandratana (Barcelona), Joshua Sokol, Emily Underwood, Gretchen Vogel (Berlin), Lizzie Wade (Mexico City) CAREERS Rachel Bernstein (Editor), Katie Langin (Associate Editor) COPY EDITORS Julia Cole (Senior Copy Editor), Morgan Everett, Cyra Master (Copy Chief) ADMINISTRATIVE SUPPORT Meagan Weiland

Creative Director Beth Rakouskas DESIGN MANAGING EDITOR Marcy Atarod GRAPHICS MANAGING EDITOR Alberto Cuadra PHOTOGRAPHY MANAGING EDITOR William Douthitt WEB STRATEGY MANAGER Kara Estelle-Powers MULTIMEDIA MANAGING PRODUCER Joel Goldberg DESIGN EDITOR Chrystal Smith DESIGNER Christina Aycock GRAPHICS EDITOR Nirja Desai INTERACTIVE GRAPHICS EDITOR Kelly Franklin SENIOR GRAPHICS SPECIALISTS Holly Bishop, Nathalie Cary SENIOR SCIENTIFIC ILLUSTRATORS Valerie Altounian, Chris Bickel SCIENTIFIC ILLUSTRATOR Ashley Mastin SENIOR PHOTO EDITOR Emily Petersen PHOTO EDITOR Kaitlyn Dolan SOCIAL MEDIA STRATEGIST Jessica Hubbard SOCIAL MEDIA PRODUCER Sabrina Jenkins WEB DESIGNER Jennie Pajerowski SENIOR PODCAST PRODUCER Sarah Crespi VIDEO PRODUCER Meagan Cantwell INTERN Claire Hogan

Chief Executive Officer and Executive Publisher Sudip Parikh Publisher, Science Family of Journals Bill Moran DIRECTOR, BUSINESS SYSTEMS AND FINANCIAL ANALYSIS Randy Yi DIRECTOR, BUSINESS OPERATIONS & ANALYSIS Eric Knott DIRECTOR OF ANALYTICS Enrique Gonzales MANAGER, BUSINESS OPERATIONS Jessica Tierney SENIOR BUSINESS ANALYST Cory Lipman FINANCIAL ANALYST Isacco Fusi ADVERTISING SYSTEM ADMINISTRATOR Tina Burks DIGITAL/PRINT STRATEGY MANAGER Jason Hillman SENIOR MANAGER, PUBLISHING AND CONTENT SYSTEMS Marcus Spiegler ASSISTANT MANAGER DIGITAL/PRINT Rebecca Doshi SENIOR CONTENT AND PUBLISHING SYSTEMS SPECIALIST Jacob Hedrick SENIOR CONTENT SPECIALISTS Steve Forrester, Antoinette Hodal, Lori Murphy PRODUCTION SPECIALIST Kristin Wowk DIGITAL PRODUCTION MANAGER Lisa Stanford CONTENT SPECIALIST Kimberley Oster ADVERTISING PRODUCTION OPERATIONS MANAGER Deborah Tompkins DESIGNER, CUSTOM PUBLISHING Jeremy Huntsinger SR. TRAFFIC ASSOCIATE Christine Hall SPECIAL PROJECTS ASSOCIATE Sarah Dhere ASSOCIATE DIRECTOR, BUSINESS DEVELOPMENT Justin Sawyers GLOBAL MARKETING MANAGER Allison Pritchard DIGITAL MARKETING MANAGER Aimee Aponte JOURNALS MARKETING MANAGER Shawana Arnold MARKETING ASSOCIATES Ashley Hylton, Mike Romano, Tori Velasquez, Jenna Voris, Justin Wood SENIOR DESIGNER Kim Huynh DIRECTOR AND SENIOR EDITOR, CUSTOM PUBLISHING Sean Sanders ASSISTANT EDITOR, CUSTOM PUBLISHING Jackie Oberst DIRECTOR, PRODUCT & PUBLISHING DEVELOPMENT Chris Reid DIRECTOR, BUSINESS STRATEGY AND PORTFOLIO MANAGEMENT Sarah Whalen ASSOCIATE DIRECTOR, PRODUCT MANAGMENT Kris Bishop PRODUCT DEVELOPMENT MANAGER Scott Chernoff PUBLISHING TECHNOLOGY MANAGER Michael Di Natale SR. PRODUCT ASSOCIATE Robert Koepke PRODUCT ASSOCIATE Anne Mason SPJ ASSOCIATE MANAGER Samantha Bruno Fuller SPJ ASSOCIATE Casey Buchta DIRECTOR, INSTITUTIONAL LICENSING Iquo Edim MARKETING MANAGER Kess Knight BUSINESS DEVELOPMENT MANAGER Rasmus Andersen SENIOR INSTITUTIONAL LICENSING MANAGER Ryan Rexroth INSTITUTIONAL LICENSING MANAGER Marco Castellan, Claudia Paulsen-Young CUSTOMER SUCCESS MANAGER Judy Lillibridge SENIOR OPERATIONS ANALYST Lana Guz FULFILLMENT COORDINATOR Melody Stringer DIRECTOR, GLOBAL SALES Tracy Holmes US EAST COAST AND MID WEST SALES Stephanie O'Connor US MID WEST, MID ATLANTIC AND SOUTH EAST SALES Chris Hoag US WEST COAST SALES Lynne Stickrod ASSOCIATE DIRECTOR, ROW Roger Goncalves SALES REP, ROW Sarah Lelarge SALES ADMIN ASSISTANT, ROW Victoria Glasbey DIRECTOR OF GLOBAL COLLABORATION AND ACADEMIC PUBLISHING RELATIONS, ASIA Xiaoying Chu ASSOCIATE DIRECTOR, INTERNATIONAL COLLABORATION Grace Yao SALES MANAGER Danny Zhao MARKETING MANAGER Kilo Lan ASCA CORPORATION, JAPAN Yoshimi Toda (Tokyo), Miyuki Tani (Osaka) DIRECTOR, COPYRIGHT, LICENSING AND SPECIAL PROJECTS Emilie David RIGHTS AND LICENSING COORDINATOR Jessica Adams RIGHTS AND PERMISSIONS ASSOCIATE Elizabeth Sandler LICENSING ASSOCIATE Virginia Warren

MAIN HEADQUARTERS

EDITORIAL

Science/AAAS 1200 New York Ave. NW Washington, DC 20005

[email protected]

SCIENCE INTERNATIONAL

INFORMATION FOR AUTHORS

Clarendon House Clarendon Road Cambridge, CB2 8FH, UK SCIENCE CHINA

Room 1004, Culture Square No. 59 Zhongguancun St. Haidian District, Beijing, 100872 SCIENCE JAPAN

ASCA Corporation Sibaura TY Bldg. 4F, 1-14-5 Shibaura Minato-ku Tokyo, 108-0073 Japan

NEWS

[email protected] sciencemag.org/authors/ science-information-authors REPRINTS AND PERMISSIONS

sciencemag.org/help/ reprints-and-permissions MEDIA CONTACTS

[email protected] MULTIMEDIA CONTACTS

[email protected] [email protected] INSTITUTIONAL SALES AND SITE LICENSES

sciencemag.org/librarian

PRODUCT ADVERTISING & CUSTOM PUBLISHING

advertising.sciencemag.org/ products-services [email protected] CLASSIFIED ADVERTISING

advertising.sciencemag.org/ science-careers [email protected]

AAAS BOARD OF DIRECTORS CHAIR Claire M. Fraser PRESIDENT Susan G. Amara PRESIDENT-ELECT Gilda A.

Barabino TREASURER Carolyn N. Ainslie CHIEF EXECUTIVE OFFICER

Sudip Parikh BOARD Cynthia M. Beall Rosina M. Bierbaum employers.sciencecareers.org Ann Bostrom [email protected] Janine Austin Clayton Laura H. Greene MEMBERSHIP AND INDIVIDUAL SUBSCRIPTIONS Kaye Husbands Fealing sciencemag.org/subscriptions Maria M. Klawe Robert B. Millard MEMBER BENEFITS aaas.org/membership/benefits William D. Provine JOB POSTING CUSTOMER SERVICE

Science serves as a forum for discussion of important issues related to the advancement of science by publishing material on which a consensus has been reached as well as including the presentation of minority or conflicting points of view. Accordingly, all articles published in Science—including editorials, news and comment, and book reviews—are signed and reflect the individual views of the authors and not official points of view adopted by AAAS or the institutions with which the authors are affiliated.

834

20 AUGUST 2021 • VOL 373 ISSUE 6557

Erin Adams, U. of Chicago Takuzo Aida, U. of Tokyo Leslie Aiello, Wenner-Gren Fdn. Deji Akinwande, UT Austin Judith Allen, U. of Manchester Marcella Alsan, Harvard U. Sebastian Amigorena, Inst. Curie James Analytis, UC Berkeley Trevor Archer, NIEHS, NIH Paola Arlotta, Harvard U. David Awschalom, U. of Chicago Clare Baker, U. of Cambridge Delia Baldassarri, NYU Nenad Ban, ETH ZÜrich Nandita Basu, U. of Waterloo Franz Bauer, Pontificia U. Católica de Chile Ray H. Baughman, UT Dallas Carlo Beenakker, Leiden U. Yasmine Belkaid, NIAID, NIH Philip Benfey, Duke U. Kiros T. Berhane, Columbia U. Joseph J. Berry, NREL Alessandra Biffi, Harvard Med. Chris Bowler, École Normale Supérieure Ian Boyd, U. of St. Andrews Emily Brodsky, UC Santa Cruz Ron Brookmeyer, UCLA (S) Christian Büchel, UKE Hamburg Dennis Burton, Scripps Res. Carter Tribley Butts, UC Irvine György Buzsáki, NYU School of Med. Mariana Byndloss, Vanderbilt U. Med. Ctr. Annmarie Carlton, UC Irvine Simon Cauchemez, Inst. Pasteur Ling-Ling Chen, SIBCB, CAS M. Keith Chen, UCLA Zhijian Chen, UT Southwestern Med. Ctr. Ib Chorkendorff, Denmark TU Amander Clark, UCLA James J. Collins, MIT Robert Cook-Deegan, Arizona State U. Virginia Cornish Columbia U. Carolyn Coyne, Duke U. Roberta Croce, VU Amsterdam Ismaila Dabo, Penn State U. Jeff L. Dangl, UNC Chiara Daraio, Caltech Nicolas Dauphas, U. of Chicago Christian Davenport, U. of Michigan Frans de Waal, Emory U. Claude Desplan, NYU Sandra DÍaz, U. Nacional de CÓrdoba Ulrike Diebold, TU Wien Stefanie Dimmeler, Goethe-U. Frankfurt Hong Ding, Inst. of Physics, CAS Dennis Discher, UPenn Jennifer A. Doudna, UC Berkeley Ruth Drdla-Schutting, Med. U. Vienna Raissa M. D'Souza, UC Davis Bruce Dunn, UCLA William Dunphy, Caltech Scott Edwards, Harvard U. Todd Ehlers, U. of TÜbingen Andrea Encalada, U. San Francisco de Quito Nader Engheta, UPenn Karen Ersche, U. of Cambridge Beate Escher, UFZ & U. of Tübingen Barry Everitt, U. of Cambridge Vanessa Ezenwa, U. of Georgia Michael Feuer, GWU Toren Finkel, U. of Pitt. Med. Ctr. Gwenn Flowers, Simon Fraser U. Peter Fratzl, Max Planck Inst. Potsdam Elaine Fuchs, Rockefeller U. Jay Gallagher, U. of Wisconsin Daniel Geschwind, UCLA

Ramon Gonzalez, U. of South Florida Sandra González-Bailón, UPenn Nicolas Gruber, ETH ZÜrich Hua Guo, U. of New Mexico Taekjip Ha, Johns Hopkins U. Sharon Hammes-Schiffer, Yale U. Wolf-Dietrich Hardt, ETH ZÜrich Louise Harra, U. Coll. London Jian He, Clemson U. Carl-Philipp Heisenberg, IST Austria Ykä Helariutta, U. of Cambridge Janet G. Hering, Eawag Heather Hickman, NIAID, NIH Hans Hilgenkamp, U. of Twente Kai-Uwe Hinrichs, U. of Bremen Deirdre Hollingsworth, U. of Oxford Randall Hulet, Rice U. Auke Ijspeert, EPFL Akiko Iwasaki, Yale U. Stephen Jackson, USGS & U. of Arizona Erich Jarvis, Rockefeller U. Peter Jonas, IST Austria Matt Kaeberlein, U. of Wash. William Kaelin Jr., Dana-Farber Cancer Inst. Daniel Kammen, UC Berkeley Kisuk Kang, Seoul Nat. U. Sabine Kastner, Princeton U. V. Narry Kim, Seoul Nat. U. Robert Kingston, Harvard Med. Nancy Knowlton, Smithsonian Institution Etienne Koechlin, École Normale Supérieure Alex L. Kolodkin, Johns Hopkins U. Julija Krupic, U. of Cambridge Paul Kubes, U. of Calgary Gabriel Lander, Scripps Res. (S) Mitchell A. Lazar, UPenn Wendell Lim, UCSF Luis Liz-Marzán, CIC biomaGUNE Omar Lizardo, UCLA Jonathan Losos, Wash. U. in St. Louis Ke Lu, Inst. of Metal Res., CAS Christian Lüscher, U. of Geneva Jean Lynch-Stieglitz, Georgia Inst. of Tech. David Lyons, U. of Edinburgh Fabienne Mackay, QIMR Berghofer Anne Magurran, U. of St. Andrews Asifa Majid, U. of York Oscar Marín, King’s Coll. London Charles Marshall, UC Berkeley Christopher Marx, U. of Idaho David Masopust, U. of Minnesota Geraldine Masson, CNRS Jason Matheny, Georgetown U. Heidi McBride, McGill U. C. Robertson McClung, Dartmouth Rodrigo Medellín, U. Nacional Autónoma de México Jane Memmott, U. of Bristol C. Jessica Metcalf, Princeton U. Baoxia Mi, UC Berkeley Tom Misteli, NCI, NIH Alison Motsinger-Reif, NIEHS, NIH (S) Suresh Naidu, Columbia U. Danielle Navarro, U. of New South Wales Daniel Nettle, Newcastle U. Daniel Neumark, UC Berkeley Beatriz Noheda, U. of Groningen Helga Nowotny, Vienna Sci. & Tech. Fund Rachel O’Reilly, U. of Birmingham Pilar Ossorio, U. of Wisconsin Andrew Oswald, U. of Warwick Isabella Pagano, Istituto Nazionale di Astrofisica Elizabeth Levy Paluck, Princeton U. Jane Parker, Max Planck Inst. Cologne

Giovanni Parmigiani, Dana-Farber Cancer Inst. (S) Daniel Pauly, U. of British Columbia Ana Pêgo, U. do Porto Samuel Pfaff, Salk Inst. Julie Pfeiffer, UT Southwestern Med. Ctr. Philip Phillips, UIUC Matthieu Piel, Inst. Curie Kathrin Plath, UCLA Martin Plenio, Ulm U. Katherine Pollard, UCSF Elvira Poloczanska, Alfred-Wegener-Inst. Julia Pongratz, Ludwig Maximilians U. Philippe Poulin, CNRS Jonathan Pritchard, Stanford U. Lei Stanley Qi, Stanford U. Trevor Robbins, U. of Cambridge Joeri Rogelj, Imperial Coll. London Amy Rosenzweig, Northwestern U. Mike Ryan, UT Austin Miquel Salmeron, Lawrence Berkeley Nat. Lab Nitin Samarth, Penn State U. Erica Ollmann Saphire, La Jolla Inst. Joachim Saur, U. zu Köln Alexander Schier, Harvard U. Wolfram Schlenker, Columbia U. Susannah Scott, UC Santa Barbara Anuj Shah, U. of Chicago Vladimir Shalaev, Purdue U. Jie Shan, Cornell U. Beth Shapiro, UC Santa Cruz Jay Shendure, U. of Wash. Steve Sherwood, U. of New South Wales Brian Shoichet, UCSF Robert Siliciano, JHU School of Med. Lucia Sivilotti, U. Coll. London Alison Smith, John Innes Ctr. Richard Smith, UNC (S) Mark Smyth, QIMR Berghofer John Speakman, U. of Aberdeen Tara Spires-Jones, U. of Edinburgh Allan C. Spradling, Carnegie Institution for Sci. V. S. Subrahmanian, Dartmouth Ira Tabas, Columbia U. Eriko Takano, U. of Manchester Patrick Tan, Duke-NUS Med. School Sarah Teichmann, Wellcome Sanger Inst. Rocio Titiunik, Princeton U. Shubha Tole, Tata Inst. of Fundamental Res. Maria-Elena Torres Padilla, Helmholtz Zentrum München Kimani Toussaint, Brown U. Barbara Treutlein, ETH Zürich Wim van der Putten, Netherlands Inst. of Ecology Henrique Veiga-Fernandes, Champalimaud Fdn. Reinhilde Veugelers, KU Leuven Bert Vogelstein, Johns Hopkins U. David Wallach, Weizmann Inst. Jane-Ling Wang, UC Davis (S) Jessica Ware, Amer. Mus. of Natural Hist. David Waxman, Fudan U. Chris Wikle, U. of Missouri (S) Terrie Williams, UC Santa Cruz Ian A. Wilson, Scripps Res. (S) Hao Wu, Harvard U. Wei Xie, Tsinghua U. Yu Xie, Princeton U. Jan Zaanen, Leiden U. Kenneth Zaret, UPenn School of Med. Bing Zhu, Inst. of Biophysics, CAS Xiaowei Zhuang, Harvard U. Maria Zuber, MIT

sciencemag.org SCIENCE

EDITORIAL

Banking on protein structural data

I

n 1953, the proposed structure of DNA magnificently linked biological function and structure. By contrast, 4 years later, the first elucidation of the structure of a protein—myoglobin, by Kendrew and colleagues—revealed an inelegant shape, described disdainfully as a “visceral knot.” Additional complexity, as well as some general principles, was revealed as more protein structures were solved over the next decade. In 1971, scientists at Brookhaven National Laboratory launched the Protein Data Bank (PDB) as a repository to collect and make available the atomic coordinates of structures (seven at the time) to interested parties. The PDB now includes more than 180,000 structures, and this resource has fueled an incalculable number of advances, including the recent development of powerful structure prediction tools. Biology takes place in three dimensions, yet most biological information is stored in onedimensional sequences of DNA that encode the amino acid sequences of proteins. The transition from one to three dimensions is accomplished through the spontaneous folding of a sequence of amino acids into a folded protein structure. Comparing elucidated structures revealed that proteins that are at least 30% identical in amino acid sequence almost always have the same folded structure; evolutionarily, structure is much more conserved than sequence. Conversely, some short stretches of five or more amino acids can adopt completely different structures; structure is context dependent. Thus, the relationship between sequence and structure is not a simple one. Predicting protein structures from sequences has been a grand challenge for decades. By 1994, fueled by the explosion of sequences, biophysicist John Moult and colleagues organized the first Critical Assessment of Structure Prediction (CASP) meeting. CASP is based on blinded assessments, which are common in clinical trials. Sequences of proteins whose structures had been determined but not publicly shared were made available to would-be predictors to develop and submit structural predictions for subsequent independent assessment. The first CASP meeting was somewhat depressing because the results revealed that predictors were doing substantially worse than they thought. CASP meetings have continued every 2 years and have driven the field forward through feedback and competition. The most recent CASP meeting, in November 2020, was shaken by results

from the company DeepMind. Its AlphaFold program performed substantially better than other programs had in the past, producing many results that are of similar quality to that of experimental structures. The RoseTTAFold program, developed by the laboratory of structural biologist David Baker, builds on this laboratory’s previous work, combined with insights from the DeepMind success (see page 871). The results of both programs are sufficiently good that many are claiming that these represent relatively general (but certainly not perfect, and incomplete) solutions to the structure prediction problem. Notably, both groups have provided their computer code for their methods for others to use, test, and enhance. These programs are based on deep-learning artificial intelligence methods. Such approaches depend on the availability of many thousands of questions with known answers to train the neural networks at their core. Thus, without the sequences with known structures from structural biologists from around the world shared in the PDB, these approaches would not have been feasible. The teams that developed these powerful programs deserve great credit for their accomplishments, but these stand on a foundation of the results from billions of dollars of public fund investments in structural biology and the sustained support of the PDB from around the world (now overseen by the Worldwide PDB). Policies from funders, publishers, and the scientific community have led to requirements that reported structures be promptly deposited in the PDB. As someone who has interacted with the PDB as a consumer, a contributor, a policy-maker, and a funder, I have experienced the power and challenges of trying to optimize such a public resource. The cultural shifts, at the cutting (and often bleeding) edge of open science, were often controversial, but it is hard to argue that they have not both increased the impact of individual determined structures and accelerated scientific progress in many ways. The ever-growing PDB provides researchers with a universe of structures with which to compare their favorite structures. The new structure prediction tools expand this universe further and provide truly compelling evidence of the power of open science. Moreover, these tools bring truth to an old saying in structural biology circles, “The structure prediction problem has been solved; it’s hiding in the PDB.” –Jeremy Berg

Jeremy Berg is professor of computational and systems biology at the University of Pittsburgh School of Medicine, Pittsburgh, PA, USA. [email protected]

PHOTO: WENDIE BERG

“…this resource has fueled an incalculable number of advances…”

10.1126/science.abl8151

SCIENCE sciencemag.org

0820Editorial.indd 835

20 AUGUST 2021 • VOL 373 ISSUE 6557

835

8/17/21 5:44 PM

NEWS IN BRIEF Edited by Jeffrey Brainard

CLIMATE CHANGE

Vast wildfires in Siberia emit a gusher of carbon

COVID-19 taxes ill-prepared Iran | Iran has become one of the largest countries with a high rate of new COVID-19 infections and is struggling to respond. On 16 August, it recorded its highest 7-day daily average of confirmed cases—39,795—since the start of the pandemic. The test positivity rate is a whopping 40%, suggesting a far higher actual tally among Iran’s 83 million people. But the new presidential administration last week rejected calls for a 2-week nationwide lockdown, opting for more limited restrictions. Vaccinations have gone slowly: Iran’s Supreme Leader Ali Khamenei forbade importation of U.S. and U.K. vaccines, prompting officials to scramble to secure doses from China and elsewhere. In June, Iran granted emergency authorization to P U B L I C H E A LT H

836

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820NewsInBrief.indd 836

the size of Texas, NASA estimated, and even traveled 4800 kilometers to the North Pole. Elsewhere, Algeria, Canada, and Turkey have also battled unusually large fires this month. In Greece, a fire that raged across the island of Evia caused the country’s “greatest ecological catastrophe of the last few decades,” Prime Minister Kyriakos Mitsotakis said on 12 August. Even Hawaii suffered its largest ever wildfire this month. The cumulative effect is to put the world on pace for the highest carbon emissions from fires in this century, according to the Global Fire Emissions Database.

a homegrown inactivated-virus preparation, COVIran Barekat; about 1.6 million doses have been administered even though it’s still undergoing a phase 2/3 trial. As of 15 August, 18.4% of Iran’s population has received at least one COVID-19 vaccine dose; 5.2% are fully vaccinated. Economic sanctions and poor planning have resulted in shortages of hospital beds, drugs such as tocilizumab, and even intravenous fluids. Iran’s official count of all COVID-19 deaths was on track this week to surpass 100,000, although the actual toll is thought to be much higher.

Aging expert faces #MeToo claims | The antiaging SENS Research Foundation last week put its most visible researcher and founder, WO R K P L AC E

gerontologist Aubrey de Grey, on administrative leave after two scientistentrepreneurs publicly accused him of sexual harassment. The two women told Endpoint News that they first complained privately to the foundation, which hired a law firm to investigate the allegations, but were disappointed when the foundation allowed de Grey to lead a July fundraiser that netted $25 million. In a blog post, Laura Deming, who heads a venture capital fund focused on longevity, alleged that de Grey, in an email, “hit on me so blatantly” about a decade ago when she was a 17-year-old undergraduate. De Grey, the foundation’s chief science officer, was also accused by Celine Halioua, now CEO of a company developing antiaging treatments for dogs. She says when she was an intern at the foundation in 2016, de Grey

PHOTO: DIMITAR DILKOFF/AFP/GETTY IMAGES

W

racked by heat and drought linked to climate change, Russia’s boreal forest is going up in smoke. Large blazes in the Siberian republic of Sakha have broken Russia’s record for a single summer’s emissions of atmosphere-warming carbon from wildfires. Since June, they have emitted about 505 million tons of carbon, according to the European Union’s Copernicus Atmosphere Monitoring Service—more than Mexico’s total carbon output from all sources in 2018. The smoke from Siberia has covered at least 1.5 million square kilometers, twice

Most of Russia’s wildfires have occurred in its republic of Sakha in Siberia, such as this devastated area west of Yakutsk.

sciencemag.org SCIENCE

8/17/21 6:04 PM

made sexual overtures to her and suggested she sleep with donors. In a Facebook post, de Grey denied the allegations, but wrote that he had written 17-year-old Deming an email “inadvisedly, for sure, and which I unreservedly regret.” He wrote that he would say more after the investigation is finished, adding, “I have nothing to hide.” De Grey has promoted ideas that many aging researchers consider outlandish, such as that the human life span can be extended to 1000 years.

IN FOCUS Parasitic ciliates cover a species of crustacean (Eulimnogammarus verrucosus), which only lives in Russia’s Lake Baikal, with white “fur” in this photo, a runner-up in the BMC Ecology and Evolution Image Competition. The infection was linked to water pollution from industry and tourism around Lake Baikal—the world’s largest freshwater lake and a UNESCO World Heritage Site—that weakened the crustacean’s immune system. Such infections are fatal.

Haiti’s fault zone strikes again | A magnitude 7.2 earthquake that struck Haiti on 14 August killed more than 1400 people. The quake’s source was the Enriquillo-Plantain Garden fault zone, a grinding east-west divide between the North American and Caribbean tectonic plates that also generated Haiti’s 2010 magnitude 7 strike, which killed up to 300,000 people. Changes in stress loads from the 2010 rupture likely helped cause the new quake, and more may be likely in years to come, researchers say; during the 1700s and 1800s, the fault was the likely source of four major earthquakes. Haiti’s limited building codes and lax enforcement left its many masonry buildings especially vulnerable to earthquake damage. N AT U R A L D I S A S T E R S

Academies’s conflicts knocked | The U.S. National Academies of Sciences, Engineering, and Medicine (NASEM) faces criticism after Kaiser Health News revealed last week that the organization had not disclosed drugmaker payments or donations to itself or two authors of its recent report on pharmaceutical waste. In 2019, Congress commissioned NASEM to study the waste created when companies ship products in large vials that can’t be resealed after first use, leading to millions of dollars in medications being thrown away annually. In a report released in February, a 14-member NASEM committee recommended the government not try to recoup the value of those medicines from drug companies. NASEM checks whether its report authors have current conflicts of interest at the time of a study, but the report doesn’t disclose that in recent years one committee member held a paid board position at a pharmaceutical corporation, another received paid consulting income from multiple drug companies, and NASEM itself has accepted millions in donations from drugmakers. NASEM said it is in the process of implementing a new conflict-of-interest policy this fall and might require disclosing such past ties.

PHOTO: KSENIYA VERESHCHAGINA

ETHICS

SCIENCE sciencemag.org

0820NewsInBrief.indd 837

Imposter feelings amid ‘brilliance’ | Scientists and other academics who believe innate talent is a prerequisite for success in their field are more likely to doubt their abilities, and the problem disproportionately affects women, particularly those from underrepresented racial and ethnic groups, a study has found. The research follows up on a 2015 finding, published in Science, that fewer women graduate with Ph.D.s in fields such as math and physics that are seen as valuing sheer brilliance over hard work. In the new study, researchers surveyed nearly 5000 graduate students, postdocs, medical residents, and faculty members in science, engineering, and humanities at nine U.S. universities; they found that women in fields that value brilliance were more likely to report feeling like WO R K P L AC E

imposters than their male peers. The study underscores the need for academic departments to actively welcome scholars from all backgrounds and not send messages that might make some question whether they belong, write the researchers this month in the Journal of Educational Psychology. BY THE NUMBERS

1 in 2700

Risk that the asteroid Bennu will collide with Earth in the year 2181, the highest in the next 300 years, a new NASA-sponsored study found. Among known asteroids, Bennu is considered one of the most likely to strike Earth. 20 AUGUST 2021 • VOL 373 ISSUE 6557

837

8/17/21 6:04 PM

IN DEP TH Medical staff at a COVID-19 isolation unit in Ashkelon, Israel, last week. Officials worry a steep surge in cases will soon fill Israeli hospitals.

COVID-19

Israel’s grim warning: Delta can overwhelm shots With early vaccination and outstanding data, country is the world’s real-life COVID-19 lab By Meredith Wadman

innovation officer at Clalit Health Services (CHS), Israel’s largest health maintenance ow is a critical time,” Israeli Minisorganization (HMO). “If it can happen ter of Health Nitzan Horowitz said here, it can probably happen everywhere.” as the 56-year-old got a COVID-19 Israel is being closely watched now bebooster shot on 13 August, the day cause it was one of the first countries out his country became the first nation of the gate with vaccinations in December to offer a third dose of vaccine to 2020 and quickly achieved a degree of popupeople as young as age 50. “We’re in a race lation coverage that was the envy of other against the pandemic.” nations. The nation of 9.3 million His message was meant for also has a robust public health his fellow Israelis, but it is a infrastructure and a population Sobering setback warning to the world. Israel has wholly enrolled in HMOs that Israel, which has led the world in launching vaccinations and data gathering, among the world’s highest levtrack them closely, allowing it is confronting a surge of cases that officials expect to push hospitals to the brink. els of vaccination for COVID-19, to produce high-quality, realSixty percent of gravely ill patients are fully vaccinated. with 78% of those 12 and older world data on how well vaccines 9000 fully vaccinated, the vast majorare working. 19 December 15 March ity with the Pfizer vaccine. Yet “I watch [Israeli data] very, First 50% the country is now logging one very closely because it is some vaccination vaccinated 7000 of the world’s highest infection of the absolutely best data com8 April rates, with nearly 700 new cases ing out anywhere in the world,” Delta variant daily per million people. More says David O’Connor, a viral identified than half are in fully vaccinated sequencing expert at the Uni5000 people, underscoring the exversity of Wisconsin, Madison. traordinary transmissibility of “Israel is the model,” agrees the Delta variant and stoking Eric Topol, a physician-scientist 3000 concerns that the benefits of at Scripps Research. “It’s pure vaccination ebb over time. mRNA [messenger RNA] vacThe sheer number of vaccines. It’s out there early. It’s 1000 cinated Israelis means some got a very high level population breakthrough infections were [uptake]. It’s a working experi0 March 2020 August January 2021 May August inevitable, and the unvaccinated mental lab for us to learn from.” New daily COVID−19 cases

“N

838

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820NewsInDepth.indd 838

CREDITS: (PHOTO) GIL COHEN MAGEN/XINHUA/GETTY IMAGES; (GRAPHIC) K. FRANKLIN/SCIENCE; (DATA) H. RITCHIE ET AL., OURWORLDINDATA.ORG, 2020

are still far more likely to end up in the hospital or die from COVID-19. But Israel’s experience is forcing the booster issue onto the radar for other nations, including the United States, suggesting as it does that even the best vaccinated countries will face a Delta surge. “This is a very clear warning sign for the rest of world,” says Ran Balicer, chief

sciencemag.org SCIENCE

8/17/21 5:56 PM

NE WS

Israel’s HMOs, led by CHS and Maccabi sah Hospital Ein Kerem and advises the Healthcare Services (MHS), track demogovernment. At his hospital, he is lining graphics, comorbidities, and a trove of coroup anesthesiologists and surgeons to spell navirus metrics on infections, illnesses, and his medical staff in case they become overdeaths. “We have rich individual-level data whelmed by a wave like January’s, when that allows us to provide real-world eviCOVID-19 patients filled 200 beds. “The dence in near–real time,” Balicer says. (The staff is exhausted,” he says, and he has reUnited Kingdom also compiles a wealth of started a weekly support group for them “to data. But its vaccination campaign ramped avoid some kind of PTSD [post-traumatic up later than Israel’s, making its current stress disorder] effect.” situation less reflective of what the future To try to tame the surge, Israel has turned may portend; and it has used three difto booster shots, starting on 30 July with ferent vaccines, making its data harder people 60 and older and, last Friday, expandto parse.) ing to people 50 and older. As of Monday, Now, the effects of waning immunity may nearly 1 million Israelis had received a third be beginning to show in Israelis vaccinated in dose, according to the Ministry of Health. early winter. A preprint published last month Global health leaders including Tedros by physician Tal Patalon and colleagues at Adhanom Ghebreyesus, director-general of KSM, the research arm of MHS, found that the World Health Organization, have pleaded protection from COVID-19 infection during with developed countries not to administer June and July dropped in proboosters given that most of the portion to the length of time world’s population hasn’t resince an individual was vacciceived even a single dose. The nated. People vaccinated in Janwealthy nations pondering or uary had a 2.26 times greater already administering boosters risk of a breakthrough infecso far are mostly targeting spetion than those vaccinated in cial populations such as the imApril. (Potential confounders mune compromised and health include the fact that the oldcare workers, but the United est Israelis, with the weakest States may soon announce a immune systems, were vaccibroader booster plan. nated first.) Studies suggest the extra Ran Balicer, At the same time, cases in the shots could help. Researchers Clalit Health Services country, which were scarcely have shown that boosting inregistering at the start of summer, have douduces a prompt surge in antibodies, which bled every 7 to 10 days since then, with the are needed in the nose and throat as a first Delta variant responsible for most of them. line of defense against infection. The Israeli They have now soared to their highest level government’s decision to start boosting those since mid-February, with hospitalizations 50 and older was driven by preliminary Minand intensive care unit admissions beginistry of Health data indicating people 60 and ning to follow. How much of the current older who have received a third dose were surge is due to waning immunity and how half as likely as their twice-vaccinated peers much to the power of the Delta variant to to be hospitalized in recent days, Mevorach spread like wildfire is uncertain. says. CHS also reported that out of a sample What is clear is that in Israel, “breakof more than 4500 patients who received through” cases are not the rare events the boosters, 88% said any side effects from the term implies. As of 15 August, 525 Israelis third shot were no worse, or milder, than were hospitalized with severe or critical from the second. COVID-19, a 34% increase from just 4 days Yet boosters are unlikely to control a Delta earlier. Of the 525, 60% were fully vaccisurge on their own, says Dvir Aran, a biomednated. Of the vaccinated, 87% were 60 or ical data scientist at Technion. In Israel, cases older. “There are so many breakthrough are rising so steeply that “even if you get infections that they dominate and most of two-thirds of those 60-plus [boosted], it’s just the hospitalized patients are actually vaccigonna give us another week, maybe 2 weeks nated,” says Uri Shalit, a health data expert until our hospitals are flooded.” He says it’s at the Israel Institute of Technology (Techmost critical to vaccinate those who still nion) who has consulted on COVID-19 for haven’t received their first or second doses, the government. “One of the big stories and to return to the masking and social disfrom Israel [is]: ‘Vaccines work, but not tancing Israel thought it had left behind—but well enough.’” has begun to reinstate. “The most frightening thing to the govAran’s message for the United States and ernment and the Ministry of Health is the other wealthier nations considering boostburden on hospitals,” says Dror Mevorach, ers is stark: “Do not think that the boosters who cares for COVID-19 patients at Hadasare the solution.” j

“This is a very clear warning. … If it can happen here, it can probably happen everywhere.”

SCIENCE sciencemag.org

0820NewsInDepth.indd 839

ARCHAEOLOGY

Nazi massacre unearthed in Poland ‘was really a horror’ Excavation finds evidence of both perpetrators and victims of WWII atrocity By Andrew Curry

D

uring World War II, Irena Szydłowska was a courier for the Polish Home Army, an underground force that fiercely resisted Nazi occupiers. On 14 January 1945, records show, she was arrested by the Gestapo. Days later, with defeat looming for Germany and Soviet forces on the horizon, she and hundreds of other Home Army fighters were taken from prison and marched into the forests of northern Poland. Szydłowska, 26, left behind a 4-year-old son. Witness reports collected after the war suggest what happened next: Szydłowska and her fellow prisoners were gunned down by German soldiers who stacked their bodies, doused them in gasoline, and burned them on massive pyres that lit up the forest for 3 days and 3 nights. The ashes were pushed into shallow pits and covered. Then, for 75 years, the site of the mass grave was lost. Now, archaeological excavations near the Polish village of Chojnice have uncovered physical evidence of that massacre and a previous one, recovering victims’ jewelry, bullet casings, burnt human bones, and more. “We knew the victims were buried somewhere, but until our research no one knew where,” says archaeologist Dawid Kobiałka of the Polish Academy of Sciences’s Institute of Archaeology and Ethnology, whose team used everything from archival documents and interviews with survivors to laser scans and excavations. Colleagues say the research, reported this week in the journal Antiquity, is the first to systematically apply archaeological techniques to a World War II–era mass grave outside of concentration camps, where research on human remains is often prohibited by Jewish religious belief. Although there were many similar mas20 AUGUST 2021 • VOL 373 ISSUE 6557

839

8/17/21 5:56 PM

sacres, “There hasn’t been research so far on a World War II site like this,” says University of Vienna archaeologist Claudia Theune. “It adds another category of crime scene.” Legal inquiries trigger most war crime investigations, adds archaeologist Alfredo González-Ruibal of the Institute of Heritage Studies of the Spanish National Research Council. This one was initiated by researchers and is one of the few to be published in a scientific journal, he says. Kobiałka grew up Chojnice, where he heard locals refer to the swampy forest just a few hundred meters from his childhood home as “Death Valley.” The area was a palimpsest of horror: In 1939, advancing German forces rounded up and executed Polish priests and intellectuals, Jewish families, and disabled people, then buried them there in a long line of trenches the retreating Polish army had dug for defense. More than 100 victims of those killings were found after the war and reburied. But hundreds

from an airplane, and spotted a trench line beneath thick vegetation. Then they used ground-penetrating radar and other noninvasive techniques to pinpoint soil disturbances that might indicate burial pits along the trench. All this led them to focus on a wooded area on the edge of town. There, in July 2020, they used metal detectors to uncover a dense collection of bullet shells, buttons, cuff links, a wristwatch stopped a few minutes after 5 o’clock—and Szydłowska’s wedding ring, which a historian identified based on the wedding date and initials engraved inside. The topsoil held pieces of burned human bone. “We used every possible archaeological method,” Kobiałka says. The evidence convinced him they had found the site of the 1945 massacre. Yet the valley’s overlapping atrocities, and the lengths the Nazis went to conceal them, made it hard to be sure. The site is “very complex,” González-Ruibal says. “It was used at

With surveys, excavations, and other archaeological methods, a team led by Dawid Kobiałka (center) located a mass grave in the woods of Poland’s “Death Valley.”

more remained unaccounted for, along with about 500 people killed in January 1945. Kobiałka thought the methods of archaeology might help reveal what happened and where. “I was fully convinced mass killings like that must leave behind a lot of material culture.” His team first dove into archives to find reports of the forced march. They interviewed survivors, including several whose parents were killed in 1939. They also matched aerial photos taken by the Allies in the closing days of the war with laser scans of the modern forest floor taken 840

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820NewsInDepth.indd 840

different times and evidence was destroyed, but they’ve still been able to retrieve a lot of information and even identify people.” With additional funding from the Ministry of Culture, National Heritage and Sport, Kobiałka’s team returned to Chojnice this summer. Over the past few months, the researchers excavated three burial pits filled with ash, bone, and more than 4000 artifacts, including hundreds of shells, all presumably from the 1945 massacre. They found valuables including medallions, cigarette lighters, and another engraved gold wedding ring, suggesting Nazi soldiers were

more interested in covering up their crime quickly than in looting bodies. Researchers also recovered more than 1 ton of human bone. “That amount seems to confirm the historical records that 400 or 500 people were killed and burned” in the 1945 massacre, Kobiałka says. “I’m an experienced archaeologist, but I’ve never experienced anything like this. It was really a horror.” The team plans to analyze the bones before reburying them in Chojnice, hoping DNA might help identify victims and surviving relatives. But exhuming victims can be fraught, cautions Susan Pollock, an archaeologist at the Free University of Berlin who has excavated World War II–era sites in Germany. “The authors imply that recovery and analysis of all remains will serve the interests of families as well as of justice,” she wrote in an email. “I would still ask who sees it this way, and if some of the potential victims and their families do not.” Kobiałka notes the excavations were carried out with the permission of Poland’s Institute of National Remembrance and the support of the local community. He adds that few if any victims of the 1945 massacre were likely to be Jewish, and that local Jewish victims killed in 1939 were exhumed and accounted for after the war. Mixed in with the bone in the woods, the team recovered more than 500 pieces of charcoal and partially burned wood. Analysis showed the fragments were common pine, a species that didn’t grow in the swampy soil of Death Valley during World War II and must have been brought in to make the pyres, Kobiałka says. The team hopes to analyze chemicals preserved in the wood to confirm the use of gasoline or another accelerant to stoke the pyres. A final piece of the puzzle came from analysis of more than 400 recovered bullets and shell casings, which a ballistics expert identified as rounds from pistols commonly used by the Gestapo and German police units; the use of pistols suggests the victims were shot one by one at close range. “It’s important that they found evidence about the people who were killed and the perpetrators,” Theune says. “They found evidence it was a Nazi crime.” The research offers a possible model for other excavations, suggesting the crimes of the past are part of archaeology’s future. “Investigating … crimes against humanity is a huge challenge for excavators,” González-Ruibal says. But, “It’s more important than ever to learn what happened.” Szydłowska’s son survived the war but died in 2004, never knowing his mother’s fate. But prosecutors have now located his daughter, hoping to close a painful chapter in the family’s history. j

PHOTO: D. FRYMARK

NEWS | I N D E P T H

sciencemag.org SCIENCE

8/17/21 5:56 PM

ENERGY

Laser-powered fusion effort nears ‘ignition’ National Ignition Facility’s latest fusion shot records a major jump in energy yield By Daniel Clery

ress was sorely needed, as NIF’s funder, the National Nuclear Security Administration, ore than a decade ago, the world’s was reducing shots devoted to ignition in famost energetic laser started to unvor of using its lasers for other experiments leash its blasts on tiny capsules of simulating the workings of nuclear weapons. hydrogen isotopes, with managers Earlier this year, combining those impromising it would soon demonprovements in various ways, the NIF team strate a route to limitless fusion produced several shots exceeding 100 kJ, inenergy. Now, the National Ignition Facility cluding one of 170 kJ. That result suggested (NIF) has taken a major leap forward. NIF was finally creating a “burning plasma,” Last week, a single laser shot there sparked in which the fusion reactions themselves a fusion explosion from a peppercorn-size provide the heat for more fusion—a runaway fuel capsule that produced eight times more reaction that is key to getting higher yields. energy than the facility had ever achieved: Then, on 8 August, a shot generated the re1.35 megajoules (MJ)—roughly the kinetic markable 1.35 MJ. “It was a surprise to evenergy of a car traveling at 160 kiloeryone,” Herrmann says. “This is a meters per hour. That figure is 70% whole new regime.” of the energy of the laser pulse that Exactly which improvements had triggered it and tantalizingly close the greatest impact and what combito “ignition”: a fusion shot producnation will lead to future gains will ing an excess of energy. “After many take a while to unravel, Herrmann years at 3% of ignition, this is supersays, because several were tweaked exciting,” says Mark Herrmann, head at once in the latest shot. “It’s a very of the fusion program at Lawrence nonlinear process. That’s why it’s Livermore National Laboratory, called ignition: It’s a runaway thing,” which operates NIF. he says. But, “This gives us a lot NIF’s latest shot “proves that a more encouragement that we can go small amount of energy, imploding significantly farther.” a small amount of mass, can get Herrmann’s team is a long way fusion. It’s a wonderful result for from thinking about fusion power the field,” says physicist Michael plants, however. “Getting fusion Campbell, director of the LaboraAn artist’s rendering shows how the National Ignition Facility’s 192 beams in a laboratory is really hard, gettory for Laser Energetics (LLE) at enter an eraser-size cylinder of gold and heat it from the inside to produce ting economic fusion power is even the University of Rochester. x-rays, which then implode the fuel capsule at its center to create fusion. harder,” Campbell says. “So, we all And it is none too soon, as years have to be patient.” NIF’s main task of slow progress have raised questions about a pencil eraser. The gold vaporizes, producing remains “stockpile stewardship”—ensuring whether laser-powered fusion has a practical a pulse of x-rays that implodes the capsule, the United States’s nuclear weapons are safe future. Now, according to LLE Chief Sciendriving the fusion fuel into a tiny ball hot and and reliable. Fusion energy is something of tist Riccardo Betti, researchers need to ask: dense enough to ignite fusion. In theory, if a sideline. But reaching ignition and being “What is the maximum fusion yield you can such tiny fusion blasts could be triggered at able to study and simulate the process will get out of NIF? That’s the real question.” a rate of about 10 per second, a power plant also “open a new window on stewardship,” Fusion, which powers stars, forces small could harvest energy from the high-speed Herrmann says, because uncontrolled fusion atomic nuclei to meld together into larger neutrons they produce to generate electricity. powers nuclear weapons. ones, releasing large amounts of energy. ExWhen NIF launched, computer models NIF’s success is “a remarkable achievetremely hard to achieve on Earth because of predicted quick success, but fusion shots in ment,” says plasma physicist Steven Rose, the heat and pressure required to join nuclei, the early years only generated about 1 kiloco-director of the Centre for Inertial Fusion fusion continues to attract scientific and comjoule (kJ) each. A long effort to better unStudies at Imperial College London. “It’s mercial interest because it promises copious derstand the physics of implosions followed made me feel very cheerful. … It feels like energy, with little environmental impact. and by last year shots were producing 100 kJ. a breakthrough.” Yet among the many approaches being inKey improvements included smoothing out Herrmann admits that, after colleagues vestigated, none has yet generated more enmicroscopic bumps and pits on the fuel captexted last week that they’d gotten an ergy than was needed to cause the reaction sule surface, reducing the size of the hole in “interesting” result from the latest shot, in the first place. Large doughnut-shaped rethe capsule used to inject fuel, shrinking the he became worried something might be actors called tokamaks, which use magnetic holes in the gold cylinder so less energy eswrong with NIF’s instruments. When that fields to cage a superhot plasma for long capes, and extending the laser pulse to keep proved not to be the case, “I did open a enough to achieve fusion, have long been the driving the fuel inward for longer. The progbottle of champagne.” j

IMAGE: LAWRENCE LIVERMORE NATIONAL LABORATORY

M

SCIENCE sciencemag.org

0820NewsInDepth.indd 841

front-runners to achieve a net energy gain. But the giant $25 billion ITER project in France is not expected to get there for more than another decade, although private fusion companies are promising faster progress. NIF’s approach, known as inertial confinement fusion, uses a giant laser housed in a facility the size of several U.S. football fields to produce 192 beams that are focused on a target in a brief, powerful pulse—1.9 MJ over about 20 nanoseconds. The aim is to get as much of that energy as possible into the target capsule, a diminutive sphere filled with the hydrogen isotopes deuterium and tritium mounted inside a cylinder of gold the size of

20 AUGUST 2021 • VOL 373 ISSUE 6557

841

8/17/21 5:56 PM

CLIMATE POLICY

Dire warming report triggers calls for more action from China Climate advocates want the world’s largest carbon producer to level off emissions soon and aim for “neutrality” by 2050 By Lili Pike

T

he sweeping new report documenting the world’s changing climate released on 9 August by the Intergovernmental Panel on Climate Change (IPCC) has put a fresh spotlight on China, a country responsible for more than onequarter of the world’s annual carbon dioxide (CO2) emissions. Climate advocates hope the report will encourage China, which has lagged other big emitters in pledges to reduce CO2 emissions, to take bolder action. Whether it can begin to slash its output significantly in the next 10 years will help determine the magnitude of the global crisis, they say. But whereas many heads of state called for enhanced climate action following the IPCC report, Chinese leaders have stayed quiet. In a statement to Agence France-Presse, China’s Ministry of Foreign Affairs simply reiterated China’s existing climate policies and said the world should have faith in its climate actions. Taking firm action is in not only the planet’s interest, but China’s own. The IPCC report offers a grim synthesis of what China can expect under various scenarios. If temperatures climb 2°C above preindustrial levels, heavy precipitation will become more intense and frequent; drought will become more severe and regular in large parts of China; tropical cyclones will increase in intensity; and, by the end of the century, sea levels will rise 0.3 to 842

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820NewsInDepth.indd 842

0.5 meters and temperatures in some regions could surpass 41°C on 30 days of the year. From its own studies, including the annual “blue book” from the China Meteorological Administration, the Chinese government is well aware of the rising risks of climate change. And the July flood in Zhengzhou, in China’s central Henan province, which killed nearly 300 people and displaced 1.5 million, was a stark reminder of the toll more extreme weather can exact. The storm’s severity surprised even Chinese climate scientists. “Of course we know climate change will bring more and more extreme precipitation and droughts,” says Wang Wen, a hydrologist at Hohai University and one of the lead authors of the report’s chapter on the regional impacts of climate change. Still, “We really didn’t expect such heavy precipitation.” But China’s current climate plans fall short of what IPCC says is needed to stave off the worst climate impacts. In September 2020, President Xi Jinping announced the country will aim to achieve carbon neutrality by 2060. The promise was in part a response to a 2018 IPCC special report that concluded the world will be much better off if it succeeds in limiting temperature rise to 1.5°C, says Jiang Kejun, a senior researcher at the Energy Research Institute, a think tank affiliated with China’s economic planning agency. “IPCC reports really influence our policymaking,” says Jiang, who is also an IPCC lead author. China

also promised to level off its emissions sometime before 2030—a deadline by which the United States and the European Union, the biggest emitters historically, have pledged to cut their emissions by half from 2005 levels. However, the 2018 report showed that sticking to the 1.5°C target requires countries to achieve carbon neutrality by 2050, not 2060. “I think [China is] going to start to get even more pressure to move that 2060 carbon neutrality goal to 2050 because that is really what is in line with the IPCC science,” says Angel Hsu, who studies Chinese climate policy at the University of North Carolina, Chapel Hill. To meet the earlier deadline, China needs to sharply reduce emissions in the coming 5 to 10 years, according to a recent study (Science, 23 April, p. 378). At the moment, carbon emissions are still growing; China was the only major economy where they climbed even amid the pandemic in 2020, according to the International Energy Agency. China’s special envoy for climate change, Xie Zhenhua, recently said developing countries like China should have more time to reach carbon neutrality than nations that industrialized earlier. But at a press conference last week, Inger Andersen, executive director of the United Nations Environment Programme, said China and the other G-20 nations “bear a special responsibility.” She called on them to be “ambitious” in the fresh emissions reduction plans that all nations are expected to submit ahead of the next major international climate negotiations, in Glasgow, U.K., in November. So far, 81 countries have submitted plans, and China has pledged to do so before the meeting begins. European leaders and climate advocates have pushed for China to move its emissions peaking date up from 2030 to 2025, and some have called on the country to stop building coal plants. Last year, China accounted for three-quarters of the new coal power that came online worldwide; more than 200 gigawatts of additional capacity is still planned. But Jiang says the plants are being built to provide energy security and will likely only run at a low capacity. “We can see that coal use will peak soon,” he says. Just how ambitious China will be in tackling its emissions leading up to 2030 may become clearer in the next few months. For now, Chinese climate scientists say IPCC’s message has landed in Beijing. “We cannot wait anymore,” Jiang says. “This is the time we decide our future, not only for China, but also for the world.” j Lili Pike is a freelance journalist in New York City.

PHOTO: AP PHOTO/NG HAN GUAN

Windmills, seen from a high-speed train in China’s Hebei province.

sciencemag.org SCIENCE

8/17/21 5:56 PM

NE WS | I N D E P T H

INFECTIOUS DISEASES

Antibody acts like short-term malaria vaccine Monoclonal antibodies protected people from infection in a small “challenge” trial By Jon Cohen

PHOTO: DENNIS KUNKEL MICROSCOPY/SCIENCE SOURCE

A

powerful new weapon could one day join the global fight against malaria. Drugs and bed nets can already help protect against the disease, which still sickens at least 200 million people a year and kills an estimated 400,000. Vaccines have also shown some promise. But an unusual study reported last week dramatized the potential of monoclonal antibodies, made by genetically engineered cells. Nine volunteers who received the antibodies were deliberately exposed to mosquitoes carrying the parasite that causes malaria. None became infected—and the protection from the antibodies appears to last for more than 6 months. The trial is too small to reach firm conclusions about the monoclonals’ protective efficacy, but other researchers say it is a striking proof of principle. “It’s great,” says Dennis Burton, an immunologist at Scripps Research who has developed monoclonal antibodies to prevent HIV infection, COVID-19, and Zika. “This is a landmark study.” Monoclonal antibodies are costly to make, which could put them out of reach of many developing countries, but the work also informs efforts to improve malaria vaccines. It demonstrates the importance of targeting immune responses to the sporelike stage of Plasmodium falciparum, the protozoan responsible for most of the world’s malaria deaths. The preventive antibody binds to a small portion of the circumsporozoite protein (CSP) that studs the surface of these sporozoites. “It’s the first study that actually assesses the potency of an antibody against the CSP target in humans,” says Hedda Wardemann, an immunologist at the German Cancer Research Center. Several years ago, a research team first isolated a powerful CSP antibody from a person who received an experimental malaria vaccine. Previous studies that have analyzed the genetic makeup of thousands of isolates of P. falciparum showed that 99.9% are identical in the region of the CSP this antibody targets. The “highly conserved” nature of the CSP region means the parasite needs it to survive and thus, the researchers reasoned, it cannot easily mutate in a way that avoids the antibody. The team, led by immunologist Robert SCIENCE sciencemag.org

0820NewsInDepth.indd 843

Seder of the Vaccine Research Center at the National Institute of Allergy and Infectious Diseases, then engineered Chinese hamster ovary cells to churn out mass quantities of a version of the antibody modified to make it last longer in the body. The hope was the antibody would block a key step in the parasite’s complex life cycle, in which sporozoites infect human liver cells. There the parasite matures before emerging to destroy red blood cells and cause disease. In a “challenge” trial, the team gave infusions of the antibody to volunteers and then allowed mosquitoes carrying P. falciparum to feed on their arms. None developed detectable blood levels of the parasite, whereas five of six people in an untreated

Antibodies can prevent this sporelike stage of a malaria parasite from infecting liver cells.

control group did, the research team reported in The New England Journal of Medicine. (The five promptly received treatment, and none became sick.) The group did not challenge two of the treated participants for 36 weeks because the pandemic delayed the research. They, too, were protected, which the investigators say suggests a single monoclonal antibody infusion could shield people for more than 6 months. Seder envisions that travelers, people in the military, or health care workers visiting malarial regions for prolonged periods might one day receive a single treatment with monoclonal antibodies before leaving home. This would be far simpler than taking daily antimalarial pills, which often

have significant side effects and can also fail against resistant strains of P. falciparum. First, though, Seder says the monoclonals need to be tested in the real world. “People said to me when I got this result, ‘Have you broken out the champagne?’” he recounts. “I said, ‘No, I got a beer.’ I’ll only break out the champagne when I have data from Africa.” In the meantime, vaccine developers could benefit. An experimental malaria vaccine called RTS,S, which uses different parts of CSP to stimulate an immune response, had modest success in early clinical trials. Four doses cut infection rates in children by 50% after 1 year, but that dropped to 28% by year four. Although RTS,S has yet to receive regulatory approval, three African countries are part of a pilot program that has given the shots to more than 600,000 children. Wardemann says future studies with monoclonals like Seder’s could help vaccine researchers identify parts of the CSP that stimulate an even more effective or longer lasting immune response. Seder hopes monoclonals could one day supplement vaccines in regions that have a high burden of the disease. The antibodies might prove especially helpful to children, as they have not had time to develop much natural immunity, and to pregnant women, who are at increased risk of severe disease from a P. falciparum infection. But W. Ripley Ballou, who works at the International AIDS Vaccine Initiative and pioneered development of RTS,S, notes that manufacturing monoclonals at the doses used in Seder’s study would cost more than $100 for a 50-kilogram person—too expensive for most countries that have malaria. “This is a great proof of concept, but it’s not yet ready as an intervention,” he says. Seder agrees. His team is developing a more potent monoclonal antibody, which it plans to test next year in a clinical trial in Mali, and he anticipates further improvement. “Suppose my antibody is 90-plus percent effective for 6 months with one subcutaneous shot,” Seder says. “Is that a tool a country could use for elimination of malaria?” Maybe, Wardemann says, if added to bed nets, drugs, and vaccines. “No single measure has done it so far,” she says. “An antibody on top may help.” j 20 AUGUST 2021 • VOL 373 ISSUE 6557

843

8/17/21 5:56 PM

NEWS

FEATURES

Dec. 2019

Wuhan, 26 December 2019 Wuhan, China, 26 December 2019

Mar. 2020

Beta, first seen in South Africa, has shown the strongest evidence of immune escape.

Jun. 2020

Sep. 2020

Dec. 2020

First spotted in India, Delta is rapidly replacing other variants around the globe.

Epsilon

Mar. 2021

Kappa

Iota

Jun. 2021

EVOLVING E

dward Holmes does not like making predictions, but last year he hazarded a few. Again and again, people had asked Holmes, an expert on viral evolution at the University of Sydney, how he expected SARS-CoV-2 to change. In May 2020, 5 months into the pandemic, he started to include a slide with his best guesses in his talks. The virus would probably evolve to avoid at least some human immunity, he 844

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820NewsFeatures.indd 844

suggested. But it would likely make people less sick over time, he said, and there would be little change in its infectivity. In short, it sounded like evolution would not play a major role in the pandemic’s near future. “A year on I’ve been proven pretty much wrong on all of it,” Holmes says. Well, not all: SARS-CoV-2 did evolve to better avoid human antibodies. But it has also become a bit more virulent and a lot more infectious, causing more people to fall

ill. That has had an enormous influence on the course of the pandemic. The Delta strain circulating now—one of four “variants of concern” identified by the World Health Organization, along with four “variants of interest”—is so radically different from the virus that appeared in Wuhan, China, in late 2019 that many countries have been forced to change their pandemic planning. Governments are scrambling to accelerate vaccination

CREDITS: (GRAPHIC) N. DESAI/SCIENCE; (DATA) NEXTSTRAIN; GISAID

Eta

sciencemag.org SCIENCE

8/16/21 5:48 PM

Branching out Each dot represents a virus isolated from a COVID-19 patient in this family tree of SARS-CoV-2, which shows a tiny subset of the more than 2 million viruses sequenced so far. The World Health Organization currently recognizes four variants of concern and four variants of interest. Variants of concern Alpha Beta Gamma Variants of interest Kappa Eta Iota

Gamma was first detected in Brazil and spread widely in South America.

Delta

Lambda

Other Former variants of interest Theta Epsilon

First detected in the United Kingdom, Alpha became the first variant to spread widely.

Lambda

Theta

THREAT

New variants have changed the face of the pandemic. What will the virus do next? By Kai Kupferschmidt

programs while prolonging or even reintroducing mask wearing and other public health measures. As to the goal of reaching herd immunity—vaccinating so many people that the virus simply has nowhere to go—“With the emergence of Delta, I realized that it’s just impossible to reach that,” says Müge Çevik, an infectious disease specialist at the University of St. Andrews. Yet the most tumultuous period in SARSCoV-2’s evolution may still be ahead of us, SCIENCE sciencemag.org

0820NewsFeatures.indd 845

says Aris Katzourakis, an evolutionary biologist at the University of Oxford. There’s now enough immunity in the human population to ratchet up an evolutionary competition, pressuring the virus to adapt further. At the same time, much of the world is still overwhelmed with infections, giving the virus plenty of chances to replicate and throw up new mutations. Predicting where those worrisome factors will lead is just as tricky as it was a year and

a half ago, however. “We’re much better at explaining the past than predicting the future,” says Andrew Read, an evolutionary biologist at Pennsylvania State University, University Park. Evolution, after all, is driven by random mutations, which are impossible to predict. “It’s very, very tricky to know what’s possible, until it happens,” Read says. “It’s not physics. It doesn’t happen on a billiard table.” Still, experience with other viruses gives evolutionary biologists some clues about 20 AUGUST 2021 • VOL 373 ISSUE 6557

845

8/16/21 5:48 PM

NEWS | F E AT U R E S

EXPLAINING THE PAST

Holmes himself uploaded one of the first SARS-CoV-2 genomes to the internet on 10 January 2020. Since then, more than 2 million genomes have been sequenced and published, painting an exquisitely detailed picture of a changing virus. “I don’t think we’ve ever seen that level of precision in watching an evolutionary process,” Holmes says. Making sense of the endless stream of mutations is complicated. Each is just a tiny tweak in the instructions for how to make proteins. Which mutations end up spreading depends on how the viruses carrying those tweaked proteins fare in the real world. The vast majority of mutations give the virus no advantage at all, and identifying the ones that do is difficult. There are obvious candidates, such as mutations that change the part of the spike protein—which sits on the surface of the virus—that binds to human cells. But changes elsewhere in the genome may be just as crucial—yet are harder to interpret. Some genes’ functions aren’t even clear, let alone what a change in their sequence could mean. The impact of any one change on the virus’ fitness also depends on other changes it has already accumulated. That means scientists need realworld data to see which variants appear to be taking off. Only then can they investigate, in cell cultures and animal experiments, what might explain that viral success. The most eye-popping change in SARSCoV-2 so far has been its improved ability to spread between humans. At some point early in the pandemic, SARS-CoV-2 acquired a mutation called D614G that made it a bit more infectious. That version spread around the world; almost all current viruses are descended from it. Then in late 2020, scientists identified a new variant, now called Alpha, in patients in Kent, U.K., that was about 50% more transmissible. Delta, first seen in India and now conquering the 846

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820NewsFeatures.indd 846

world, is another 40% to 60% more transmissible than Alpha. Read says the pattern is no surprise. “The only way you could not get infectiousness rising would be if the virus popped into humans as perfect at infecting humans as it could be, and the chance of that happening is incredibly small,” he says. But Holmes was startled. “This virus has gone up three notches in effectively a year and that, I think, was the biggest surprise to me,” Holmes says. “I didn’t quite appreciate how much further the virus could get.” Bette Korber at Los Alamos National Laboratory and her colleagues first suggested that D614G, the early mutation, was taking over because it made the virus better at spreading. She says skepticism about the virus’ ability to evolve was common in the early days of the pandemic, with some researchers saying D614G’s apparent advantage might be sheer luck. “There was extraordinary resistance in the scientific

from laboratory data alone,” says Christian Drosten, a virologist at the Charité University Hospital in Berlin. He and others are still figuring out what, at the molecular level, gives Alpha and Delta an edge. Alpha seems to bind more strongly to the human ACE2 receptor, the virus’ target on the cell surface, partly because of a mutation in the spike protein called N501Y. It may also be better at countering interferons, molecules that are part of the body’s viral immune defenses. Together those changes may lower the amount of virus needed to infect someone—the infectious dose. In Delta, one of the most important changes may be near the furin cleavage site on spike, where a human enzyme cuts the protein, a key step enabling the virus to invade human cells. A mutation called P681R in that region makes cleavage more efficient, which may allow the virus to enter more cells faster and lead to greater numbers of virus particles in an infected person. In July, Chinese researchers

Hostile takeovers SARS-CoV-2 variants began to emerge in 2020. Alpha surged in many countries in early 2021, then was largely replaced by Delta. Two other variants of concern, Beta and Gamma, account for a smaller number of cases. 100%

Beta

80 Alpha 60 Gamma

40

Delta

20 0 Aug. 2020

Oct. 2020

Dec. 2020

Feb. 2021

community to the idea this virus could evolve as the pandemic grew in seriousness in spring of 2020,” Korber says. Researchers had never watched a completely novel virus spread so widely and evolve in humans, after all. “We’re used to dealing with pathogens that have been in humanity for centuries, and their evolutionary course is set in the context of having been a human pathogen for many, many years,” says Jeremy Farrar, head of the Wellcome Trust. Katzourakis agrees. “This may have affected our priors and conditioned many to think in a particular way,” he says. Another, more practical problem is that real-world advantages for the virus don’t always show up in cell culture or animal models. “There is no way anyone would have noticed anything special about Alpha

Apr. 2021

Jun. 2021

Aug. 2021

posted a preprint showing Delta could lead to virus levels in patient samples 1000 times higher than for previous variants. Evidence is accumulating that infected people not only spread the virus more efficiently, but also faster, allowing the variant to spread even more rapidly. DEADLY TRADE-OFFS

The new variants of SARS-CoV-2 may also cause more severe disease. For example, a study in Scotland found that an infection with Delta was about twice as likely to lead to hospital admission than with Alpha. It wouldn’t be the first time a newly emerging disease quickly became more serious. The 1918–19 influenza pandemic also appears to have caused more serious illness as time went on, says Lone Simonsen, an

CREDITS: (GRAPHIC) N. DESAI/SCIENCE; (DATA) NEXTSTRAIN; GISAID

where SARS-CoV-2 may be headed. The courses of past outbreaks show the coronavirus could well become even more infectious than Delta is now, Read says: “I think there’s every expectation that this virus will continue to adapt to humans and will get better and better at us.” Far from making people less sick, it could also evolve to become even deadlier, as some previous viruses including the 1918 flu have. And although COVID-19 vaccines have held up well so far, history shows the virus could evolve further to elude their protective effect— although a recent study in another coronavirus suggests that could take many years, which would leave more time to adapt vaccines to the changing threat.

sciencemag.org SCIENCE

8/17/21 11:36 AM

CREDITS: (PHOTO) NATIONAL ARCHIVES OF AUSTRALIA; (GRAPHIC) N. DESAI/SCIENCE; (DATA) DEREK SMITH/UNIVERSITY OF CAMBRIDGE; DAVID MONTEFIORI/DUKE UNIVERSITY

epidemiologist at Roskilde University who studies past pandemics. “Our data from Denmark suggests it was six times deadlier in the second wave.” A popular notion holds that viruses tend to evolve over time to become less dangerous, allowing the host to live longer and spread the virus more widely. But that idea is too simplistic, Holmes says. “The evolution of virulence has proven to be quicksand for evolutionary biologists,” he says. “It’s not a simple thing.” Two of the best studied examples of viral evolution are myxoma virus and rabbit hemorrhagic disease virus, which were released in Australia in 1960 and 1996, respectively, to decimate populations of European rabbits that were destroying croplands and wreaking ecological havoc. Myxoma virus initially killed more than 99% of infected rabbits, but then less pathogenic strains evolved, likely because the virus was killing many animals before they had a chance to pass it on. (Rabbits also evolved to be less susceptible.) Rabbit hemorrhagic disease virus, by contrast, got more deadly over time, probably because the virus is spread by blow flies feeding on rabbit carcasses, and quicker death accelerated its spread. Other factors loosen the constraints on deadliness. For example, a virus variant that can outgrow other variants within a host can end up dominating even if it makes the host sicker and reduces the likelihood of transmission. And an assumption about human respiratory diseases may not always hold: that a milder virus—one that doesn’t make you crawl into bed, say—might allow an infected person to spread the virus further. In SARS-CoV-2, most transmission happens early on, when the virus is replicating in the upper airways, whereas serious disease, if it develops, comes later, when the virus infects the lower airways. As a result, a variant that makes the host sicker might spread just as fast as before. EVASIVE MEASURES

From the start of the pandemic, researchers have worried about a third type of viral change, perhaps the most unsettling of all: that SARS-CoV-2 might evolve to evade immunity triggered by natural infections or vaccines. Already, several variants have emerged sporting changes in the surface of the spike protein that make it less easily recognized by antibodies. But although news of these variants has caused widespread fear, their impact has so far been limited. Derek Smith, an evolutionary biologist at the University of Cambridge, has worked for decades on visualizing immune evasion in the influenza virus in so-called antigenic maps. The farther apart two variants are on SCIENCE sciencemag.org

0820NewsFeatures.indd 847

The myxoma virus was released in Australia in 1950 to control rabbits after trials at this test site on Wardang Island. It has evolved to become less virulent over time, but not all viruses do.

Smith’s maps, the less well antibodies against one virus protect against the other. In a recently published preprint, Smith’s group, together with David Montefiori’s group at Duke University, has applied the approach to mapping the most important variants of SARS-CoV-2 (see graphic, below). The new maps place the Alpha variant very close to the original Wuhan virus, which means antibodies against one still neutralize the other. The Delta variant, however, has drifted farther away, even though it doesn’t completely evade immunity. “It’s not an immune escape in the way people think of an escape in slightly cartoonish terms,” Katzourakis says. But Delta is slightly more likely to infect fully vaccinated people than previous variants. “It shows the possible beginning of a trajectory and that’s what worries me,” Katzourakis says.

Viral cartography On this “antigenic map,” produced by Derek Smith, David Montefiori, and colleagues, the distance between two variants indicates how well antibodies against one neutralize the other. Beta has drifted farthest from the strain that emerged in Wuhan, China, in 2019. Alpha

Wuhan strain

Beta

Gamma

Epsilon Delta

Eta

Kappa

Other variants have evolved more antigenic distance from the original virus than Delta. Beta, which first appeared in South Africa, has traveled the farthest on the map, although natural or vaccine-induced immunity still largely protects against it. And Beta’s attempts to get away may come at a price, as Delta has outstripped it worldwide. “It’s probably the case that when a virus changes to escape immunity, it loses other aspects of its fitness,” Smith says. The map shows that for now, the virus is not moving in any particular direction. If the original Wuhan virus is like a town on Smith’s map, the virus has been taking local trains to explore the surrounding area, but it has not traveled to the next city—not yet. PREDICTING THE FUTURE

Although it’s impossible to predict exactly how infectiousness, virulence, and immune evasion will develop in the coming months, some of the factors that will influence the virus’ trajectory are clear. One is the immunity that is now rapidly building in the human population. On one hand, immunity reduces the likelihood of people getting infected, and may hamper viral replication even when they are. “That means there will be fewer mutations emerging if we vaccinate more people,” Çevik says. On the other hand, any immune escape variant now has a huge advantage over other variants. In fact, the world is probably at a tipping point, Holmes says: With more than 2 billion people having received at least one vaccine dose and hundreds of millions more having recovered from COVID-19, variants that evade immunity may now have a bigger leg up than those that are more infectious. 20 AUGUST 2021 • VOL 373 ISSUE 6557

847

8/16/21 5:48 PM

NEWS | F E AT U R E S

Something similar appears to have happened when a new H1N1 influenza strain emerged in 2009 and caused a pandemic, says Katia Kölle, an evolutionary biologist at Emory University. A 2015 paper found that changes in the virus in the first 2 years appeared to make the virus more adept at human-tohuman transmission, whereas changes after 2011 were mostly to avoid human immunity. It may already be getting harder for SARS-CoV-2 to make big gains in infectiousness. “There are some fundamental limits to exactly how good a virus can get at transmitting and at some point SARS-CoV-2 will hit that plateau,” says Jesse Bloom, an evolutionary biologist at the Fred Hutchinson Cancer Research Center. “I think it’s very hard to say if this is already where we are, or is it still going to happen.” Evolutionary virologist Kristian Andersen of Scripps Research guesses the virus still has space to evolve greater transmissibility. “The known limit in the viral universe is measles, which is about three times more transmissible than what we have now with Delta,” he says.

Scratching the surface Researchers trying to understand which genetic changes make SARS-CoV-2 variants more successful have focused on the spike protein, which studs the viral surface and binds to human cells. Alpha, Beta, and Delta have mutations in three key areas of the protein that may affect the virus’ infectiousness and its ability to elude the immune system. SARS-CoV-2

Receptor-binding domain N-terminal domain

Furin cleavage site

Spike protein Mutation ation sites

The limits of immune escape are equally uncertain. Smith’s antigenic maps show the space the virus has explored so far. But can it go much farther? If the variants on the map are like towns, then where are the country’s natural boundaries—where does the ocean start? A crucial clue will be where the next few variants appear on the map, Smith says. Beta evolved in one direction away from the original virus and Delta in another. “It’s too soon to say this now, but we might be head-

Alpha

Beta

Delta

ing for a world where there are two serotypes of this virus that would also both have to be considered in any vaccines,” Drosten says. Immune escape is so worrying because it could force humanity to update its vaccines continually, as happens for flu. Yet the vaccines against many other diseases—measles, polio, and yellow fever, for example—have remained effective for decades without updates, even in the rare cases where immuneevading variants appeared. “There was big

N

ew variants of SARS-CoV-2 have major impacts around the globe, driving up COVID-19 case and mortality numbers (see main story, p. 844). But each of those viruses picks up its crucial changes as it divides in the cells of an infected human being. The nature of those infections—how fast the virus replicates and for how long—may determine the odds that they will give rise to new and more troublesome mutants, researchers say. After someone is infected, the virus starts to multiply at a dizzying rate, producing billions of viral particles within days. Because small copy mistakes happen during every replication cycle, a huge variety of slightly different genomes quickly emerges. With SARS-CoV-2’s genome spanning just 30,000 nucleotides, and only three ways to change any one position, every possible mutation likely arises in an infected individual. The vast majority of those changes offer the virus no benefit, and even those that do only have a small chance of being passed on to the next person. A paper published in 2020 estimated that about 1000 viral particles are transmitted when one person infected another, but a reanalysis by Katia Kölle of Emory University and a colleague, published as a preprint in February, concluded that 99% of all successful transmissions come from three or fewer virus particles. A study published in Science in April by evolutionary biologist Katrina Lythgoe at the University of Oxford put the number of transmitted virus particles at infection between one and eight. This means that, unless a mutation arises early and gives the virus so big an advantage that it quickly becomes dominant in the host, it has a low chance of being transmitted, which puts the brake on virus evolution. “It’s generally thought that when transmission bottlenecks are tight, that slows adaptive evolution at the population level,” Kölle explains.

848

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820NewsFeatures.indd 848

That may sound like good news for humanity, but it is offset by the huge number of SARS-CoV-2 infections globally, says Jesse Bloom, an evolutionary biologist at the Fred Hutchinson Cancer Research Center. Besides, the virus may have a shortcut. In most people, the immune system curbs the infection within days, but a few develop a chronic infection lasting months. That gives time for mutations to accumulate and become dominant, increasing their chances of transmission. In a short-lived acute infection, evolution is “more like roulette,” Kölle says, but in chronic cases, “you have the time needed to adapt to that environment.” Chronic infections may explain why the Alpha variant, first seen in the United Kingdom in late 2020, appeared to emerge with a slew of mutations all at once. In theory, Alpha could have picked up those changes one by one before arriving in the country, says Andrew Rambaut of the University of Edinburgh, but the fact that most of its genome resembles other U.K. viruses at the time suggests instead that a local virus underwent extended evolution in a single patient. “I am still reasonably confident that a chronic infection is the best explanation,” Rambaut says. COVID-19 treatments may accelerate evolution in chronic patients. In July, researchers in Germany published data on six immunocompromised patients treated with a monoclonal antibody that targeted SARS-CoV-2. In five of them, the virus acquired E484K, a mutation known to help it elude the immune system, and the virus rebounded in all five patients. Still, the evidence that chronic patients are the source of new variants is circumstantial, Bloom cautions. People who don’t develop chronic infections but do take longer than average to clear SARS-CoV-2 could also generate and spread mutants, Lythgoe says—and they are more numerous. “Are these the infections that really drive the evolution of acute viruses like SARS-CoV-2? There’s really interesting questions there.” —K.K

CREDITS: (GRAPHIC) N. DESAI/SCIENCE; (DATA) E. WALL ET AL., THE LANCET, 397:10292, 2331 (2021)

Do chronic infections breed dangerous new variants?

sciencemag.org SCIENCE

8/16/21 5:48 PM

PHOTO: SAEED KHAN/AFP/GETTY IMAGES

alarm around 2000 that maybe we’d need to replace the hepatitis B vaccines,” because an escape variant had popped up, Read says. But the variant has not spread around the world: It is able to infect close contacts of an infected person, but then peters out. The virus apparently faces a trade-off between transmissibility and immune escape. Such trade-offs likely exist for SARS-CoV-2 as well. Some clues about SARS-CoV-2’s future path may come from coronaviruses with a much longer history in humans: those that cause common colds. Some are known to reinfect people, but until recently it was unclear whether that’s because immunity in recovered people wanes, or because the virus changes its surface to evade immunity. In a study published in April in PLOS Pathogens, Bloom and other researchers compared the ability of human sera taken at different times in the past decades to block virus isolated at the same time or later. They showed that the samples could neutralize strains of a coronavirus named 229E isolated around the same time, but weren’t always effective against virus from 10 years or more later. The virus had evidently evolved to evade human immunity, but it had taken 10 years or more. “Immune escape conjures this catastrophic failure of immunity when it is really immune erosion,” Bloom says. “Right now it seems like SARS-CoV-2, at least in terms of antibody escape, is actually behaving a lot like coronavirus 229E.” Others are probing SARS-CoV-2 itself. In a preprint published this month, researchers tinkered with the virus to learn how much it has to change to evade the antibodies generated in vaccine recipients and recovered patients. They found that it took 20 changes to the spike protein to escape current antibody responses almost completely. That means the bar for complete escape is high, says one of the authors, virologist Paul Bieniasz of Rockefeller University. “But it’s very difficult to look into a crystal ball and say whether that is going to be easy for the virus to acquire or not,” he says. “It seems plausible that true immune escape is hard,” concludes William Hanage of the Harvard T.H. Chan School of Public Health. “However, the counterargument is that natural selection is a hell of a problem solver and the virus is only beginning to experience real pressure to evade immunity.” And the virus has tricks up its sleeve. Coronaviruses are good at recombining, for instance, which could allow new variants to emerge suddenly by combining the genomes—and the properties—of two different variants. In pigs, recombination of a coronavirus named porcine epidemic diarrhea virus with attenuated vaccine strains of another coronavirus has led to more viruSCIENCE sciencemag.org

0820NewsFeatures.indd 849

Residents line up outside a vaccination center in Sydney, where a rapidly growing outbreak of the highly contagious Delta variant of SARS-CoV-2 led officials to order a new lockdown in June.

lent variants of PEDV. “Given the biology of these viruses, recombination may well factor into the continuing evolution of SARSCoV-2,” Korber says. Given all that uncertainty, it’s worrisome that humanity hasn’t done a great job of limiting the spread of SARS-CoV-2, says Eugene Koonin, a researcher at the U.S. National Center for Biotechnology Information. Some dangerous variants may only be possible if the virus hits on a very rare, winning combination of mutations, he says.

It might have to replicate an astronomical number of times to get there. “But with all these millions of infected people, it may very well find that combination.” Indeed, Katzourakis adds, the past 20 months are a warning to never underestimate viral evolution. “Many still see Alpha and Delta as being as bad as things are ever going to get,” he says. “It would be wise to consider them as steps on a possible trajectory that may challenge our public health response further.” j 20 AUGUST 2021 • VOL 373 ISSUE 6557

849

8/16/21 5:48 PM

INSIGHTS

CLIMATE POLICY

Keep climate policy focused on the social cost of carbon A proposed shift away from the SCC is ill advised By Joseph E. Aldy1,2,3, Matthew J. Kotchen2,4, Robert N. Stavins1,2,3, James H. Stock1,2,3,5

I

n the context of climate change, the application of cost-benefit analysis to inform mitigation policies can help to achieve the best outcomes and avoid the worst: spending trillions of dollars but failing to get the job done (1). The costs of a climate policy are the abatement costs of reducing emissions of carbon dioxide (CO2) (or other greenhouse gases). The standard measure of the benefits of a climate policy is the social cost of carbon (SCC), which measures the avoided economic damages associated with a metric ton of CO2 emissions. Recently, however, there have been calls for an alternative approach to policy evaluation that ignores the benefits of avoided climate damages and instead focuses only on minimizing the 1

John F. Kennedy School of Government, Harvard University, Cambridge, MA, USA. 2National Bureau of Economic Research, Cambridge, MA, USA. 3Resources for the Future, Washington, DC, USA. 4School of the Environment, Yale University, New Haven, CT, USA. 5 Department of Economics, Harvard University, Cambridge, MA, USA. Email: [email protected]

850

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820PolicyForum.indd 850

compliance costs of a given, politically determined climate objective (2, 3). We argue here that a shift from use of the SCC and cost-benefit analysis to an alternative approach for evaluating policy that focuses on costs alone would be misguided. Rather than advocate for alternative approaches, now is the time to support efforts to update the SCC and its application to official climate policy evaluation. THE SCC The economic value of a policy’s climate benefits is the sum of current and future damages that are avoided. The SCC is the monetized value of these benefits per metric ton of CO2 abated. Those avoided damages include impacts on agricultural production, reductions in labor productivity, property damage (particularly along coastal areas), mortality and morbidity impacts, and induced migration, among others. SCC estimations need to integrate climate and economic models, make economic projections into the distant future, and put those damages on a current-day basis by discounting them back to the present (4). And all of this must take place in the context of consid-

erable scientific and economic uncertainty. In the United States, the original federal government estimates of the SCC (5) used the results of scientific and economic research that were then available to develop initial estimates and subsequent updates. Although the Trump administration asserted an alternative set of assumptions to vastly reduce the value of the SCC, it never questioned the use of the SCC for cost-benefit analysis, nor were its alternative set of assumptions adopted more broadly or supported by the economics community. From 2008 to 2019, SCC estimates were used in 60 federal regulatory analyses (see supplementary materials). In addition to such analyses, SCC estimates can be used to inform the design of carbon-pricing climate policy instruments, which use price signals to bring about emissions reductions, typically either as a carbon tax or as a CO2 cap-andtrade system (for example, British Columbia’s carbon tax and the European Union’s emissions trading system). A substantial number of state governments use the federal SCC to evaluate their own energy and climate policies. In New York and Illinois, the SCC serves as the basis for the value of “zero-emission credits” paid to electric utilities under state clean-energy legislation. In Colorado, Minnesota, and Washington, electric utilities are required to use the federal SCC in their resource planning. And in California, legislation requires regulators to incorporate the SCC in policy analysis. Outside of the United States, the Canadian government has adopted the estimation methodology, and several other national governments as well

PHOTO: GORDON WELTERS/LAIF/REDUX

P OLICY FORUM

sciencemag.org SCIENCE

8/13/21 5:56 PM

Cleanup begins after floods in Bad Münstereifel, North Rhine-Westphalia, Germany, 19 July 2021. Estimating impacts of climate change, such as damages from extreme weather events, that could be averted by curbing carbon emissions is critical for policy analysis.

as the International Monetary Fund have developed their own SCC estimates or drawn on the US experience, including France, Germany, Mexico, and Norway (6). On his first day in office, President Biden issued an executive order (7) to reestablish the Interagency Working Group on the Social Cost of Greenhouse Gases and directed it to produce within 30 days an “interim SCC” and a final SCC no later than January 2022. The administration recently issued its interim SCC, with a primary value of $51/ton and ranging from $14 to $152/ton (in 2020 US dollars) (8), which is in line with the SCC used under the Obama administration (5) after adjusting for inflation. These numbers, or any subsequent revisions, are to be used to quantify the climate benefits of federal policies in official cost-benefit analyses (1). In practice, their use can change an overall assessment from negative to positive net benefits (9). President Biden has also tasked agencies to consider applying the SCC to monetize the climate benefits of budget, procurement, and other government actions (7). A TARGET-CONSISTENT PRICE Cost-benefit analysis is not the only way economists evaluate policies. An alternative is cost-effectiveness analysis, which is used to compare policies with the same objective. In the case of climate policy, this involves identifying some target, such as a maximum increase in the global average temperature or a date by which zero net emissions of CO2 is to be achieved, and then comparing the costs of alternative approaches to achieve that target. SCIENCE sciencemag.org

0820PolicyForum.indd 851

With a given target and the cost-effectiveness approach, policy evaluation no longer entails comparing costs to benefits but rather focuses on adopting the lowest cost policy to achieve the target. In practice, the menu of all possible policies is not available, nor are their costs. Still, it is possible to compare the cost of a particular candidate policy with some threshold, which is a modeled estimate of the minimum-cost policy consistent with achieving the given target. In the context of climate policy, the minimum costs are actually a time path of carbon prices equal to the estimated marginal costs of abatement in each future year until the target is met. To be clear, these carbon prices are not equivalent to—and are not intended to reflect—the marginal benefits of lowering emissions, as with the SCC. Instead, they reflect an estimate of the marginal abatement costs of implementing costeffective policies that align with the target, and they are called target-consistent carbon prices. In practice, the SCC is a tool for costbenefit analysis, and the target-consistent price is a tool for cost-effectiveness analysis. Also, these target-consistent prices should not be confused with the specific prices used in carbon-pricing policy instruments such as carbon taxes or cap-and-trade. Although cost-benefit analysis using the SCC has been the dominant approach to climate policy assessment in practice in the United States, there has been some use of cost-effectiveness analysis with targetconsistent pricing. For example, since 2009, the United Kingdom has been using a targetconsistent price to assess climate policies (6). In the US context, this approach has recently been promoted by a pair of prominent economists, Lord Nicholas Stern and Nobel Laureate Joseph Stiglitz (2), who argue that for the purpose of policy evaluation, the target-consistent price should replace the SCC as the benchmark for assessment. The main arguments for relying on costeffectiveness analysis and a target-consistent price are as follows: Climate policy cost-benefit analysis is fine in theory but cannot be implemented in practice because too little is known to estimate the SCC credibly. Beyond that, it may be argued that climate change raises questions of moral responsibility to future generations and Earth’s ecosystem that surpass what economists can monetize, so any SCC will necessarily underestimate damages. Hence, the argument goes, we should be guided by science and moral considerations

to identify some goal, such as keeping warming within 1.5oC, and then translate that into a policy target, such as achieving net zero emissions by 2050. If politicians adopt a given policy target, the task of the policy analyst is then simply to help achieve that target as cost-effectively as possible. Given a target, the technical task is therefore to compute a target-consistent price to use for comparison with the cost of a specific policy. In most climate-economic models, an analyst can estimate a cost-effective emissions time path for achieving a given temperature objective in any future year. Associated with this cost-effective emission time path is an implicit carbon price trajectory, which is the target-consistent price of carbon. PITFALLS OF A TARGET-CONSISTENT PRICE Which to use for evaluating climate policy: an assessment of benefits and costs using the SCC, or cost-effectiveness analysis with a target-consistent price? Some advocates of the target-consistent price call it an alternative estimate of the SCC, which it decidedly is not. Terminology matters. The SCC measures the benefits of reducing CO2 emissions, whereas the target-consistent price is a cost estimate that takes no account whatsoever of benefits. Our view is that a push to replace costbenefit analysis and the SCC with cost-effectiveness analysis and a target-consistent price could set back climate policy, just as the United States is poised to take meaningful climate action. Despite the apparently logical appeal of using some politically defined target as an operational, bright-line rule in the face of uncertainty, we find that calls for the target-consistent approach are unwise for four main reasons. First, the target-consistent approach replaces scientific assessments of damages from storms, floods, fires, and a myriad of other climate impacts with subjective judgments about policy targets and choices. The starting point and necessary condition for the target-consistent price is a political decision about the goal: an emissions objective in a specified future year. Much of the developed world has adopted a target of net zero emissions by 2050, but ultimately that is a political decision. The Trump administration had no such target, so its targetconsistent price would have been zero. A future US administration may have a different view on the target from that of the current administration—perhaps more ambitious, perhaps less—resulting in yet another price. The anchor for the target-consistent price is fundamentally political, not scientific, and therefore subject to arbitrary change. This is not to say that the cost-benefit approach using the SCC is completely immune from political interference in its application to policy 20 AUGUST 2021 • VOL 373 ISSUE 6557

851

8/13/21 5:56 PM

INSIGHTS | P O L I C Y F O RU M

analysis, but that it is more consistent with how scientific assessments are typically undertaken. The cost-effectiveness approach to analyzing policies with the target-consistent price involves scientific estimates in regard to future technological change; the benefitcost approach using the SCC places greater reliance on the science of climate change because the SCC is an estimate of future damages of climate change. Moreover, as a technical matter, calculating the target-consistent price requires making assumptions about a myriad of complementary policies over the next several decades (3). Those assumptions matter tremendously. For example, the UK target-consistent price trajectory for achieving an 80% reduction in emissions by 2050 (a goal set in a 2008 law) assumed that the UK would purchase emission reductions from other countries after 2030 (6). The analysts were thus able to substantially lower the target-consistent carbon price and thus make the approach more appealing politically, by assuming that policies far from certain will unfold. This illustrates another way that the target-consistent price depends on policies and political projections, not science, in sharp contrast with the SCC. Second, the target-consistent price calculation depends not only on assumptions about future politically determined public policies but also on critical assumptions about technologies that are not commercially available today. Many energy-economic models that can solve for ambitious climate goals—such as limiting warming to 1.5°C or net-zero emissions by 2050—do so by assuming aggressive global deployment of bioenergy with carbon capture and storage in the power sector. Uncertainty about this prospect (and uncertainties characterizing other emerging technologies, such as direct air capture) helps explain the large variation in target-consistent prices in the literature for a given goal. The Intergovernmental Panel on Climate Change (IPCC) 2019 special report illustrated that limiting warming to 1.5°C would require carbon prices ranging from $135 to $5500/ton (10). Such uncertainty is also demonstrated by the broad distribution of target-consistent prices for limiting warming to 2°C in the IPCC fifth assessment report’s (AR5) integrated assessment modeling database. A common critique of the SCC is its considerable uncertainty, yet the AR5 target-consistent prices exhibit substantially greater dispersion than the distribution of SCCs estimated through hundreds of thousands of modeling runs in the US Interagency Working Group’s Monte Carlo Simulations of the SCC (11). Third, a target-consistent approach seems unlikely to meet legal requirements, at least in the US context. In 2008, a federal court ruled that fuel economy stan852

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820PolicyForum.indd 852

dards must account for the benefits of reducing CO2 emissions (9). In addition, the Department of Energy must show that the benefits of appliance efficiency standards exceed their economic burdens (12). Under Executive Orders issued by Democratic and Republican presidents since 1981, federal regulatory agencies are required to compare the costs of major regulations with their benefits (1). Because the targetconsistent price is based on costs, with no regard for benefits, it would not fulfill these requirements, posing substantial legal hurdles for officially adopting the approach in the coming years, precisely when meaningful climate action is most needed. Fourth, the target-consistent approach is entirely inward-looking for any country that adopts it because it is essentially a goit-alone approach to evaluating domestic climate policy. By contrast, the SCC inherently builds in the notion of reciprocity among countries because it reflects the global damages of emissions. A future in which all countries seek to guide domestic policy by using the SCC can lead to progress on addressing climate change in a globally efficient and least-cost way. The same cannot be said of the alternative approach. Although it is true that signatories of the Paris Agreement have agreed on a global target for temperature changes, vast uncertainty remains on how emissions will be reduced by each country and on what timetable. THE PATH AHEAD Over the coming months, the Biden administration intends to complete its update to the SCC. We believe that it is critical that the Interagency Working Group should keep its focus on the SCC rather than commencing work on a target-consistent price. This is important not only for the continued use of cost-benefit analysis in the United States but also for maintaining an objective and defensible basis for climate policy around the world. Correctly estimating the SCC is by no means an easy task. In 2017, the National Academy of Sciences (NAS) developed a list of needed improvements (13). Fortunately, research on the NAS list has progressed substantially. In the coming months and beyond, the Biden administration’s Interagency Working Group can pull together the large body of new research on damages, uncertainty, discount rates, socioeconomic projections, and other considerations, including those identified by the NAS and others (13, 14, 15). In particular, estimates of biophysical and monetized damages have continued to improve, and today’s economic environment suggests an update of the discount rate previously used.

Aside from the immediate goal of improving the estimate, a broader goal should be to establish a process by which science drives policy, not the reverse. By establishing a procedural and substantive record, such as periodic peer review through NAS (9, 13), it will be more difficult for a future administration with less ambitious climate goals to undercut the SCC. Although the Trump administration attempted just this (with its inappropriately high discount rate of 7% and no consideration of damages outside the United States), it never gained traction because it was out of step with standard and objective economic analysis, and the process for estimating the SCC was originally intended to be separate from political decision-making. The same cannot be said of the target-consistent approach, which hinges on political decisions. We recognize the value of research that can promote cost-effective ways of achieving climate goals, but this is only one side of the cost-benefit comparison. It is critically important to maintain focus on the benefits of addressing climate change as a means for evaluating and justifying climate policy. With this goal in mind, now is not the time to change lanes and advocate the alternative approach. Instead, we need credible and updated estimates of the SCC. j REF ERENCES AND NOTES

1. K. J. Arrow et al., Science 272, 221 (1996). 2. N. Stern, J. E. Stiglitz,“The Social Cost of Carbon, Risk, Distribution, Market Failures: An Alternative Approach,” working paper 28472 (National Bureau of Economic Research, 2021). 3. N. Kaufman, A. R. Barron, W. Krawczyk, P. Marsters, H. McJeon, Nat. Clim. Chang. 10, 1010 (2020). 4. L. H. Goulder, R. N. Stavins, Nature 419, 673 (2002). 5. Interagency Working Group on Social Cost of Carbon, “Technical support document: Social cost of carbon for regulatory impact analysis under executive order 12866” (plus technical updates in 2013, 2015, and 2016) (US government, 2010). 6. J. Aldy et al.,“Environmental benefit-cost analysis: A comparative analysis between the United States and the United Kingdom,” discussion paper 21-90 (Harvard Environmental Economics Program, January, 2021). 7. J. R. Biden, Executive order 13990 of January 20, 2021: Protecting public health and the environment and restoring science to tackle the climate crisis., Fed. Regist. 86, 7037 (2021). 8. Interagency Working Group on Social Cost of Greenhouse Gases,“Technical support document: Social cost of carbon, methane, and nitrous oxide, interim estimates under executive order 13990” (US government, February 2021). 9. W. Pizer et al., Science 346, 1189 (2014). 10. V. Masson-Delmotte et al., Eds.,“Global warming of 1.5°C,” Special Report (IPCC, 2019). 11. J. Aldy et al., Nat. Clim. Chang. 6, 1000 (2016). 12. 42 U.S. Code § 6295 - Energy Conservation Standards; www.law.cornell.edu/uscode/text/42/6295 13. National Academies of Sciences, Engineering, and Medicine. Valuing Climate Damages: Updating Estimation of the Social Cost of Carbon Dioxide (National Academies, 2017). 14. T. Carleton, M. Greenstone,“Updating the United States government’s social cost of carbon,” working paper no. 2021-04 (Energy Policy Institute, University of Chicago, 2021). 15. G. Wagner et al., Nature 590, 548 (2021). SUPPLEMENTARY MATERIALS

science.sciencemag.org/content/373/6557/850/suppl/DC1 10.1126/science.abi7813

sciencemag.org SCIENCE

8/13/21 5:56 PM

Gas and hydraulic fracturing, or fracking, in the United States, such as this site in California, raise concerns about possible chemical pollution of surface water and groundwater sources.

PERSPECTIVES GEOCHEMISTRY

The fracking concern with water quality Tapping into oil and gas reserves comes at the expense of contaminating water By Elaine Hill1 and Lala Ma2

PHOTO: DAVID MCNEW/GETTY IMAGES

U

nconventional oil and gas development (UOGD) has revolutionized resource extraction over the past two and a half decades. Although these methods to recover oil and gas began in the 1980s, only recently have technological innovations in horizontal drilling and hydraulic fracturing (HF) made it financially feasible to extract resources from difficult-to-access rock formations with low permeability (1). These innovations have massively increased the availability of oil and gas resources for consumption, yielding energy cost savings, employment, and income (2). One estimate finds per household benefits to be about 4.9% of income annually (3). Many have cautioned that human and ecological health may be damaged by the negative environmental impacts that come with these benefits. On page 896 of this issue, Bonetti et al. (4) report that UOGD has increased salt concentrations in surface waters across the United States. The findings have broad implications for research and policy going forward. The investigation of four specific UOGD chemicals was motivated in part by best available data. As Bonetti et al. note, however, there may be other chemicals associated with

1

Department of Public Health Sciences, University of Rochester Medical Center, Rochester, NY, USA. 2 Department of Economics, University of Kentucky, Lexington, KY, USA. Email: [email protected]; [email protected] SCIENCE sciencemag.org

0820Perspectives.indd 853

HF that are potentially more dangerous (5). This highlights the need to expand the set of chemicals measured in the current monitoring system. Moreover, precise measurement of causal effects is dictated by the availability of water monitors, which are spatially sparse. Expanding the geographical scope of water quality surveillance would also improve understanding of the distribution of these effects across regions, different socioeconomic strata, and time. Furthermore, UOGD technology is dynamic. There has been considerable innovation in this industry since UOGD’s mass deployment, which has yielded substantial changes in, for example, the lateral and vertical length of oil and gas wells, the number of wellheads per well pad, whether and how water is reused, and the duration of each stage in the life cycle of a well (6). Notably, these developments affect environmental exposures, and these data are needed to identify the mechanisms of effect. Understanding the exposure pathways at play is necessary for policy to effectively control the environmental damages from these operations. A looming question also remains regarding whether the water impacts from UOGD translate into health damages or damages on other measures of well-being. Evidence of UOGD water impacts (7) along with studies of the health impacts of drinking water (8, 9) provide indirect evidence that UOGD-related water contamination influences health. Direct evidence is needed. More broadly, continued research in this domain would lend

insight into the health benefits of surface water pollution control. A review finds that 67% of US surface water regulations fail a benefit-cost test (10). That these calculations ignore health benefits is among the hypothesized reasons behind the understatement of net benefits. The findings of Bonetti et al. also suggests a need to rethink regulation. Expanding data collection might be achieved by requiring regulatory agencies to collect and report releases of additional chemicals. Because the magnitude of effects reported by Bonetti et al. are below regulatory thresholds that the US Environmental Protection Agency has for drinking water, tightening the stringency of currently regulated chemicals should be considered. Whether it is advantageous to do so depends on whether the associated chemical emissions yield human and ecological health impacts (information that is not yet known). There are also many UOGD chemicals that are unknown to the public. All state regulations allow exemptions for trade secrets to incentivize companies to invest in expensive research and development so that they may recoup the benefits of their investment (11). For regulators, these challenges to setting pollution targets may be overcome by requiring firm disclosure of HF chemicals. Enacting any regulation, however, requires consideration of its cost-effectiveness. At 0.8% of gross domestic product in an average year, water-quality regulation is already among the most expensive environmental policies in 20 AUGUST 2021 • VOL 373 ISSUE 6557

853

8/13/21 5:49 PM

INSIGHTS | P E R S P E C T I V E S

1. US Geological Survey (USGS), “When did hydraulic fracturing become such a popular approach to oil and gas production?”; www.usgs.gov/faqs/when-did-hydraulicfracturing-become-such-a-popular-approach-oil-andgas-production?qt-news_science_products=0#qtnews_science_products. 2. K. Black, A. J. Boslett, E. L. Hill, L. Ma, S. J. McCoy, Annu. Rev. Resour. Econ. 10.1146/annurevresource-110320-092648 (2021). 3. A. Bartik et al., Am. Econ. J. Appl. Econ. 11, 105 (2019). 4. P. Bonetti, C. Leuz, G. Michelon, Science 373, 896 (2021). 5. US Environmental Protection Agency (EPA), “Hydraulic fracturing for oil and gas: Impacts from the hydraulic fracturing water cycle on drinking water resources in the United States” (Report EPA-600-R-16-236F, EPA, 2016). 6. L. Torres, O. P. Yadav, E. Khan, Sci. Total Environ. 539, 478 (2016). 7. E. Hill, L. Ma, Am. Econ. Rev. 107, 522 (2017). 8. A. Ebenstein, Rev. Econ. Stat. 94, 186 (2012). 9. J. Currie, J. Graff Zivin, K. Meckel, M. Neidell, W. Schlenker, Can. J. Econ. 46, 791 (2013). 10. D. Keiser, J. Shapiro, J. Econ. Perspect. 33, 51 (2019). 11. C.B. Johnson, ONEJ. 6, 443 (2021). 12. T. R. Fetter et al., “Learning by viewing? Social learning, regulatory disclosure, and firm productivity in shale gas,” Working paper 25401, National Bureau of Economic Research, Cambridge, MA, December 2018. 13. K. Arrow et al., Science 272, 221 (1996). 10.1126/science.abk3433

854

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820Perspectives.indd 854

Exploring the path of the variable resistance Resistive switching studies pave the way to neuromorphic information technologies By Hans Hilgenkamp and Xing Gao

I

n handling computer hardware, the last thing anyone would like to do is expose electronic components to electrostatic discharges. Nevertheless, this is exactly an approach that researchers are taking toward faster and more energy-efficient computing. Inspired by the functions of neurons and synapses in the brain, resistive switching devices or “memristors” are being explored as building blocks for neuromorphic circuitry. In such devices, the resistance properties are durably altered by applying voltage pulses. On page 907 of this issue, del Valle et al. (1) have imaged the early stages of electric field–induced electronic breakdown and formation of a conducting filament in vanadium oxide. By doing this in a space- and time-resolved manner, the authors provide useful insight

into the characteristic length and time scales involved. Computing systems are commonly based on the Von Neumann architecture, in which the memory is physically separated from the logic circuitry. Data are continuously shuttled between these units. This process is time consuming and presents an important cause of energy dissipation. Both aspects become very noticeable in data-intensive applications, like training deep neural networks. Neural networks are composed of layers of neuron-like devices connected through synapses. The latter comprise weight factors that are adjusted in the training process. In conventional complementary metal-oxide semiconductor (CMOS)–based technology, the weights need to be fetched, adjusted, and put back into the memory in every learning step. In an alternative and ultimately more efficient approach, the weights are embodied in the hardware it-

Moving toward neural networks Memristor crossbar arrays Insulating Intermediate Metallic

Current/Temperature

Building the hardware

Variable electrical resistance Vanadium dioxide (VO2) undergoes a hysteretic insulator-tometal transition (IMT) just above room temperature, where resistivity changes by several orders of magnitude. VO2-based resistive switching device

Creating artificial neurons can be accomplished in single memristive devices or crossbar arrays, providing a pathway for realizing functional network structures for neuromorphic computing.

Fluctuations and nucleation

Stationary filament formation

Tracking the transition between states By applying electric field pulses to the VO2-based device, the IMT takes place through filament formation. The filament forms starting with hotspots that quickly grow and expands to a stationary state as a result of resistive heating. By tailoring current or temperature, many stable states between the insulating and metallic states can be achieved.

GRAPHIC: KELLIE HOLOSKI/SCIENCE

REFERENCES AND NOTES

OXIDE ELECTRONICS

Resistance

the United States (10). Additional regulation may overburden state and local governments. Moreover, firm disclosure of HF fluids may stifle UOGD innovation (12). It is not an overstatement to say that UOGD has affected all dimensions of life for those in exposed communities (3). Many of the impacts have lifelong consequences on individual well-being, including future health, education, and labor market outcomes. The mounting evidence on environmental impacts demonstrates a need to quantify and synthesize the associated health and socioeconomic impacts using a common metric. Benefit-cost analysis is particularly useful to facilitate a comprehensive assessment of the consequences of these innovations (13). This type of analysis also requires the clarification of alternative scenarios for comparison and the time frame of consideration. For example, would increasing UOGD regulation cause companies to revert to coal-based energy production (thereby exacerbating pollution) or would it instead spur the transition to a renewables-based future to aid the longerterm battle with climate change? The counterfactual scenario of comparison changes the net-benefit calculation and the optimal policy choice. It has been more than two decades since the rapid expansion of UOGD, but we are only now beginning to grasp the full scope and extent of the costs associated with these innovations. An understanding of the environmental effects of UOGD is a necessary first step toward a comprehensive assessment of UOGD. Going forward, the mechanisms of impact and their consequences must be clarified to translate this evidence into actionable policy. j

sciencemag.org SCIENCE

8/13/21 5:49 PM

self, and training implies an alteration of the physical properties of the synapse, similar to what happens in the brain. In a fully electronic implementation, this requires the ability to controllably adjust the electrical resistance of a material. This is achieved using the electric field–driven motion of defect states, such as oxygen vacancies and impurity atoms (2), which are resistive switching concepts used also in binary resistive random access memory (ReRAM). Alternatives involve thermally induced alterations of the crystallinity of the material (3) and organic memristors (4). A complication in many techniques is that they involve atomic displacements and reconfigurations, which can lead to a spread in device properties and fatigue. This problem is circumvented by exploiting tunable electronic and/ or magnetic ordering phenomena. The Mott insulator VO2 is an attractive example, exhibiting a hysteretic resistive transition just above room temperature (5). Applying electric field pulses to the material in the highresistive state creates a metallic filament with a conductance that depends on the pulse intensity and duration. Notably, the resistance can be programmed over several orders of magnitude. By studying thin film microdevices with various vanadium oxide stoichiometries, del Valle et al. found that the transition starts with resistance fluctuations and nucleation of the conducting filament in hotspots on a hundreds-of-nanoseconds time scale (see the figure). In an avalanche-like process, the filament subsequently grows, as a result of Joule heating, over a time scale of microseconds. The authors investigated the growth dynamics and the final width of the conducting filament, which depends on both the characteristics of the voltage pulse and the resistivities of the material in the insulating and conducting states. Inhomogeneities play an important role in triggering the transition and in the filament formation by focusing the current. These findings can help to optimize the switching processes—e.g., by deliberately incorporating nanoscopic elements that act as optimized hotspots. The storing of synaptic weights in the neural network hardware is an example of the upcoming in-memory computing paradigm, which aims to circumvent the Von Neumann bottleneck. The practical implementation of this is typically in the form of cross-bar arrays (6), with the current lines acting as the pre- and postsynaptic connections to the neurons. The variable conductance properties of the barrier materials Faculty of Science and Technology and MESA+ Institute for Nanotechnology, University of Twente, Enschede, Netherlands. Email: [email protected] SCIENCE sciencemag.org

0820Perspectives.indd 855

encode for the synaptic weight. Using this setup, Ohm’s law and Kirchhoff’s circuit law are used for matrix-vector multiplications, which are a key processing step in neural network operation. Also, other data-intensive applications can benefit from outsourcing data processing from the logic units to the memory—large-scale database queries being one example (7). In addition to storing information, the switching of VO2 when exceeding a certain threshold voltage can also be used for the realization of the artificial neurons. Using a negative differential resistance that can be invoked in the resistive transition, Yi et al. have even demonstrated 23 different neuronal functionalities with VO2-based memristors (8). Spiking modes of neural network operation are facilitated by this, with further expected enhancements in energy efficiency. The optical reflectivity modulation, as studied by del Valle et al., presents a coupling between the electronic and photonic domains. This allows, for example, for the storing of synaptic weights in a photonic processor—a principle recently used in a photonic tensor core accelerator using phase change materials (9). Future computer systems will likely comprise a heterogeneous mix of electronic, optical, and spintronic components, and efficient coupling between these domains will then be indispensable. The next stage in vanadium oxide memristor research will be to make the step from single resistive switching devices to functional network structures, like multilayer artificial neural networks, and to explore their operation. In this endeavor, other more exotic post–Von Neumann information processing concepts are also of interest (10, 11). The space- and time-resolved optical reflectometry technique as demonstrated by del Valle et al. will enable current pulses and associated resistance modulations passing through such networks to be monitored without interference—tracing, so to say, the path of the variable resistance. j REF ERENCES AND NOTES

1. J. del Valle et al., Science 373, 907 (2021). 2. R. Waser, R. Dittmann, G. Staikov, K. Szot, Adv. Mater. 21, 2632 (2009). 3. I. Boybat et al., Nat. Commun. 9, 2514 (2018). 4. S. Goswami, S. Goswami, T. Venkatesan, Appl. Phys. Rev. 7, 021303 (2020). 5. T. Driscoll, H.-T. Kim, B.-G. Chae, M. Di Ventra, D. N. Basov, Appl. Phys. Lett. 95, 043503 (2009). 6. Q. Xia, J. J. Yang, Nat. Mater. 18, 309 (2019). 7. I. Giannopoulos et al., Adv. Intell. Syst. 2, 2000141 (2020). 8. W. Yi et al., Nat. Commun. 9, 4661 (2018). 9. J. Feldmann et al., Nature 589, 52 (2021). 10. M. Di Ventra, F. L. Traversa, J. Appl. Phys. 123, 180901 (2018). 11. M. A. Nugent, T. W. Molter, PLOS ONE 9, e85175 (2014). 10.1126/science.abh2231

INFECTIOUS DISEASES

Tracking severe malaria disease Malaria infection prevalence predicts malaria mortality— at least for now By Terrie Taylor and Laurence Slutsker

F

emale Anopheles mosquitoes transmit malaria sporozoites to humans in the context of a blood meal. In malariaendemic areas, most of the ensuing infections are asymptomatic. Some, however, progress to an uncomplicated illness (fever, headache, body aches, and pains). Younger individuals, with less clinical immunity to malaria, are at highest risk of developing severe disease (anemia, cerebral malaria, and/or respiratory distress) and of dying (1). Because the relationship between malaria transmission and malaria mortality is so variable, and because both are challenging to measure, it has remained unclear whether decreases in malaria transmission, resulting from control measures, would actually decrease malaria mortality. On page 926 of this issue, Paton et al. (2) find that the higher the prevalence of malaria infection in a given community, the higher the incidence of severe malaria disease. These findings may be useful in tracking the impact of various malaria control measures over time. Measuring malaria transmission is not straightforward. Two of the traditional metrics of exposure to malaria parasites, the entomological inoculation rate and cohort incidence studies, are expensive and difficult to measure. The entomological inoculation rate, or the number of infectious bites per person per unit time (usually per year), is the product of the human biting rate and the sporozoite rate. (Sporozoites are carried in mosquito salivary glands and are injected into skin when female Anopheles mosquitoes take blood meals from humans.) To estimate the human biting rate, volunteers (protected by prophylactic doses of antimalarial drugs) bare their legs and collect mosquitoes as they land. Sporozoite rates can be determined by analyzing the salivary glands of the mosquitoes. Measuring incidence rates for new malaria infections requires treating a study cohort with an effective anDepartment of Osteopathic Medical Specialties, College of Osteopathic Medicine, Michigan State University, 909 Wilson Road, East Lansing, MI 48824, USA. Email: [email protected] 20 AUGUST 2021 • VOL 373 ISSUE 6557

855

8/13/21 5:49 PM

INSIGHTS | P E R S P E C T I V E S

856

It is important to appreciate that the study sites were chosen to represent a range of parasite prevalence rates and are thus a substantial collection of (largely) cross-sectional studies. Longitudinal data are difficult and expensive to collect, but following the effect of changing parasite prevalence rates over time in response to control activities would help to define the kinetics of the change in the incidence of severe malaria. How much time is required for a given community prevalence rate to affect hospitalizations for severe malaria? Over the course of the observational period included in this analysis (2006 to 2020), there was a sea change in recognizing and managing uncomplicated malaria in sub-

Malaria infection and illness A small proportion of Anopheles mosquitoes can transmit malaria parasites to humans. In endemic areas, 25 to 75% of individuals may carry asymptomatic infections (parasite prevalence). A smaller proportion of this group will develop symptomatic malaria illness. Paton et al. showed that parasite prevalence correlates with severe malaria disease. Infected

Asymptomatic

Developed illness

Mosquitos that can transmit malaria: 3%

People infected with malaria: 25 to 75%

Infected people who develop illness: ~10 to 30%

Saharan Africa. Countries rapidly shifted to using highly effective artemisinin combination therapy (ACT) to treat patients with positive results on malaria rapid diagnostic tests (6). Community health workers were empowered to test members of their community with fever and to immediately provide ACTs to anyone with a positive rapid diagnostic test. This public health intervention could stop the progression of a malaria infection to severe disease and diminish the association between malaria prevalence and mortality described by Paton et al., but the design of the study precluded addressing this potential confounder. The findings of Paton et al. are highly relevant to the evaluation and implementation of malaria vaccines. At present, for the most advanced candidate, RTS,S/AS01, a regimen of three monthly doses (with a fourth dose 12 to 18 months later) is envisaged for children between the ages of 5 to 17 months, the age group shown by Paton et al. to have the highest rates of severe malaria. The phase 3 trial results of RTS,S/AS01 showed a 30% reduction in severe malaria in 5 to 17 month olds in settings with good health care (to decrease malaria mortality) and high bed net coverage (to decrease malaria transmission) (7). But would widespread use of an effective malaria vaccine skew the association between community parasite prevalence and severe disease? Community parasite prevalence is unlikely to change in this vaccination scenario because the vaccine targets a small proportion of the population—but an effective vaccine would decrease the rate at which severe disease develops in the vaccinated population. Programmatic evaluation of RTS,S/AS01 is ongoing, with an interim data review and consideration for recommendations from the World Health Organization (WHO) expected late in 2021. The study of Paton et al. is a useful highresolution image of the status quo in three East African settings. The utility of the relatively easily acquired, robust data is clear and emphasizes the value of maintaining health systems that have been strengthened over the course of the COVID-19 pandemic. These systems are the source of the programmatic data needed to devise strategies and to evaluate responses to a variety of public health threats, including, but not limited to, malaria. j REF ERENCES AND NOTES

Children with malaria illness who develop severe disease: 1 to 2%

1. 2. 3. 4. 5.

N. J. White et al., Lancet 383, 723 (2014). R. S. Paton et al., Science 373, 926 (2021). L. Wu et al., Nature 528, S86 (2015). B. Mappin et al., Malar. J. 14, 460 (2015). M. Karra, G. Fink, D. Canning, Int. J. Epidemiol. 46, 817 (2017). 6. WHO, World Malaria Report 2020 (2020); www.who. int/teams/global-malaria-programme/reports/ world-malaria-report-2020/. 7. Wkly. Epidemiol. Rec. 91, 33 (2016). 10.1126/science.abk3443

GRAPHIC: K. FRANKLIN/SCIENCE

timalarial drug, ensuring that the treatment was successful, and then following the cohort longitudinally at frequent intervals to identify when new infections appear. Paton et al. took a different approach and used community prevalence of malaria infection in 26 communities in Uganda, Kenya, and Tanzania, measured directly through surveys of households and in school children or extracted from published literature, as a proxy for malaria transmission. Malaria infection can be measured directly, by microscopy or polymerase chain reaction (PCR) to detect parasite DNA, or indirectly, using rapid diagnostic tests that capture parasite antigens. Of these, PCR is the most sensitive, followed by microscopy and then rapid diagnostic tests (3); the sensitivities of the latter two vary according to age, treatment history, and transmission intensity (4). PCR was not used in any of the sampling sites, and for the sites that used rapid diagnostic tests, the results were converted to the standard microscopy metric, parasites per microliter of blood (4). Measuring malaria-associated mortality is also not straightforward. Many deaths occur outside of hospitals and are not captured systematically. Even within hospitals in malaria-endemic areas, “malaria infection” is not synonymous with “malaria illness.” Paton et al. chose patients who were “sick enough to be admitted to hospital with malaria” as the proxy for malaria-associated mortality and measured three clinical phenotypes (cerebral malaria, severe malarial anemia, and respiratory distress) individually and together. Observations were thus limited to hospitals with the capacity to detect malaria infections, characterize the level of consciousness, measure the degree of anemia, provide blood for transfusion, follow the patient through the entire clinical course, and capture the data reliably. Broadly, Paton et al. found that for every 25% increase in community parasite prevalence (above a baseline of 17.6% and below an upper limit of 75%), annual rates of admission for severe malaria double, and that as prevalence rates rise, the average age of children admitted to the hospital drops (see the figure). A potential, but unavoidable, bias in these results is that these analyses involved populations with easy geographic access to hospitals (5). Whether a patient even presents to a hospital or not depends on a multiplicity of factors: distance between home and the hospital and availability of transportation and community perceptions of the quality of facility-based care (e.g., availability of drugs or competence of health care workers). In difficult to reach rural areas, severe illnesses unfold at home and never touch the health care system. 20 AUGUST 2021 • VOL 373 ISSUE 6557

0820Perspectives.indd 856

8/13/21 5:49 PM

METALLURGY

Structural hierarchy defeats alloy cracking Internal herringbone structures in a ductile multicomponent alloy enable crack tolerance By Xianghai An

GRAPHIC: A. MASTIN/SCIENCE

H

Unlike metals, natural materials, such as bone, shell, and wood, principally consist of hard and soft phases that are hierarchically configured and have favorable combinations of properties (5). For example, wood can be tough and also have substantial tensile and compressive strength. Materials scientists have borrowed these architectural features to engineer nanostructural heterogeneities, such as a spatial gradient structure that has nanoscale crystallites (grains) at the surface but larger internal grains (6), and a heterogeneous structure in which soft lamellae are embedded in a strong lamellae matrix (7). These structures can concurrently activate

opened a vast compositional space for exploration (8). At the atomic level, statistical fluctuations in compositional and packing arrangements of the various elements confer many opportunities for tuning properties and functionalities (8–10). Incorporation of local chemical and structural heterogeneities spanning multiple length scales could greatly enhance materials properties. For example, the combination of crack tolerance and tensile ductility reported by Shi et al. was achieved through their designed DS EHEA (a type of MPEA) that displays multiscale spatial heterogeneities. Chemical complexity occurs at the atomic

igh-performance alloys play a critical role in demanding engineering applications in manufacturing, infrastructure, and transportation (1). In structural applications, they must be strong, ductile, durable, and damage tolerant. However, these characteristics cannot currently be obtained simultaneously (2). Microscale cracks initiated in a tensioned material tend to propagate rapidly and unstably. This process, in turn, can cause catastrophic failure during service or can create strain that becomes highly localized near the crack tip, which makes it difficult to deform the material uniformly durEngineering hierarchical heterogeneity ing processing. On page 912 of Multilevel hierarchy of chemical and nanostructural heterogeneities can enable new combinations of properties and this issue, Shi et al. (3) show functionalities. Shi et al. directionally solidified a multi–principal element alloy of aluminum (Al), iron (Fe), cobalt (Co), and nickel that a directionally solidified (Ni) to create a ductile alloy that resists cracking. Integrating the planar defects could further improve materials properties. (DS) eutectic high-entropy alloy (EHEA) develops a hierarChemical heterogeneity Structural heterogeneity Defect heterogeneity chically organized herringbone fcc bcc TB/SF BEC AEC BEC microstructure that imparts Ni Al Fe Co multiscale crack buffering. This fcc material exhibited exceptional damage tolerance over large tensile deformation, as well as ultrahigh uniform elongation. The plasticity of metals bcc mainly results from the movement of dislocations, so to enhance the load-bearing capacity Microscale structures Atomic-level structures Nanoscale structures of metal, various internal defects A heterogeneous distribution of In these alloys, no element predominates, In the nanodomains of fcc and bcc can be introduced to impede dismicrostructures, such as the and the atomic distribution is phases, spatial planar defects, such location motion (4). However, aligned eutectic colonies (AECs) heterogeneous (left). These elements as twin boundaries (TBs) and increasing the strength of a maand branched eutectic colonies can pack in both face-centered stacking faults (SFs) that are shown terials to a high level invariably (BECs), form that help resist cubic (fcc) and body-centered cubic as dashed lines, are embedded in leads to a drastic loss of toughcrack propagation. (bcc) phases (right). individual nanostructures. ness (the material’s resistance to fracture) and ductility (the material’s various plastic deformation mechanisms to scale, and at the micrometer scale, alternatstretchability without breaking) (2). High increase strength along with ductility and ing soft and hard lamellae form with differlocal strain energy cannot be effectively distoughness. However, if cracking occurs in ent cubic crystal structures. These structures sipated in strong materials when dislocation ductile materials under tension, uniform in turn make up large distinct eutectic coloplasticity is low, so cracks readily initiate and deformation will be prohibited by the high nies aligned along or inclined to the DS direcpropagate. Thus, many new high-strength strain localization around the crack tip and tion that composes a hierarchically arranged alloys cannot be used in safety-critical will undermine ductility. herringbone microstructure (see the figure). structural components, such as aircraft jet Historically, alloys design has been reDeformation initially nucleates microengines, nuclear containment vessels, and stricted to a single primary element, such as cracks in hard lamellae with limited dewind turbines, because they cannot meet iron for steels, copper for bronze, and nickel formability. The abutting microstructural damage-tolerance requirements. for superalloys. Recently, multi–principaltraits with strong strain hardening capabilelement alloys (MPEAs), in which three or ity promote the dissipation of local energy more major elements—such as CoCrNi, and thus blunt crack tips. These cracks are School of Aerospace, Mechanical, and Mechatronic CoCrFeNiMn, and TiZrHfNb—are mixed in arrested and confined within individual Engineering, The University of Sydney, Sydney, NSW 2006, Australia. Email: [email protected] roughly equal amounts (see the figure), have lamellae, which prohibits their unstable 20 AUGUST 2021 • VOL 373 ISSUE 6557

0820Perspectives.indd 857

857

8/13/21 5:49 PM

INSIGHTS | P E R S P E C T I V E S

propagation and catastrophic percolation. The persistent nucleation and growth of these microcracks create extrinsic plasticity that compensates for the low ductility of the brittle phase and enables sustainable uniform deformation. Compared to the conventionally solidified alloy, the self-buffering herringbone EHEA was three times more ductile, accompanied with extraordinary damage tolerance and a simultaneous enhancement of strength and toughness. Shi et al.’s engineering of hierarchical chemical and nanostructural heterogeneities heralds a new approach for developing high-performance alloys. Tuning local compositional fluctuations may energetically alter the nature of a material’s response to external stimuli like brittleness (9, 10). Creation of internal defects within individual nanostructures (see the figure) could activate multiple strengthening and toughening mechanisms (11). The heterogeneous microstructures could be programmed to trigger various intrinsic and extrinsic deformation mechanisms (12). This design concept will require identifying and quantifying which materials parameters endow specific properties to help unravel how these develop in hierarchical structures. An integrated computational and experimental protocol, in conjunction with data science, could accelerate the establishment of a unified design principle and scientific framework for future mechanistic alloy design. Another formidable conundrum is to precisely control and organize spatially local chemical and structural heterogeneities. The advanced additive manufacturing techniques could, through a dedicate multiscale processing control, unlock the full potential of this new alloy design concept to help tackle major economic, energy, and environmental challenges. j REFERENCES AND NOTES

1. 2. 3. 4. 5.

K. Lu, Science 328, 319 (2010). R. O. Ritchie, Nat. Mater. 10, 817 (2011). P. Shi et al., Science 373, 912 (2021). K. Lu, L. Lu, S. Suresh, Science 324, 349 (2009). Z. Q. Liu, M. A. Meyers, Z. Zhang, R. O. Ritchie, Prog. Mater. Sci. 88, 467 (2017). 6. T. H. Fang, W. L. Li, N. R. Tao, K. Lu, Science 331, 1587 (2011). 7. X. Wu, P. Jiang, L. Chen, F. Yuan, Y. T. Zhu, Proc. Natl. Acad. Sci. U.S.A. 111, 7197 (2014). 8. E. P. George, D. Raabe, R. O. Ritchie, Nat. Rev. Mater. 4, 515 (2019). 9. T. Yang et al., Science 362, 933 (2018). 10. Y. J. Chen et al., J. Mater. Sci. Technol. 82, 10 (2021). 11. Z. Cheng, H. Zhou, Q. Lu, H. Gao, L. Lu, Science 362, eaau1925 (2018). 12. X. H. An, S. D. Wu, Z. G. Wang, Z. F. Zhang, Prog. Mater. Sci. 101, 1 (2019). ACKNOWLEDGMENTS

The author acknowledges funding from the Australian Research Council (DE170100053) and The University of Sydney Robinson Fellowship. 10.1126/science. abk1671

858

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820Perspectives.indd 858

ECOLOGY

Ecology in the age of automation Technology is revolutionizing the study of organisms in their natural environment By Timothy H. Keitt and Eric S. Abelson

T

he accelerating pace of global change is driving a biodiversity extinction crisis (1) and is outstripping our ability to track, monitor, and understand ecosystems, which is traditionally the job of ecologists. Ecological research is an intensive, field-based enterprise that relies on the skills of trained observers. This process is both time-consuming and expensive, thus limiting the resolution and extent of our knowledge of the natural world. Although technology will never replace the intuition and breadth of skills of the experienced naturalist (2), ecologists cannot ignore the potential to greatly expand the scale of our studies through automation. The capacity to automate biodiversity sampling is being driven by three ongoing technological developments: the commoditization of small, low-power computing devices; advances in wireless communications; and an explosion in automated data-recognition algorithms in the field of machine learning. Automated data collection and machine learning are set to revolutionize in situ studies of natural systems. Automation has swept across all human endeavors over recent decades, and science is no exception. The extent of ecological observation has traditionally been limited by the costs of manual data collection. We envision a future in which data from field studies are augmented with continuous, fine-scale, remotely sensed data recording the presence, behavior, and other properties of individual organisms. As automation drives down costs of these networks, there will not be a simple expansion of the quantity of data. Rather, the potential high resolution and broad extent of these data will lead to qualitatively new findings and will result in new discoveries about the natural world that will enable ecologists to better predict and manage changing ecosystems (3). This will be es-

pecially true as different types of sensing networks, including mobile elements such as drones, are connected together to provide a rich, multidimensional view of nature. Given the role that biodiversity plays in lending resilience to the ecosystems on which humans depend (4), monitoring the distribution and abundance of species along with climate and other variables is a critical need in developing ecological hypotheses and for adapting to emerging global challenges. Ecosystems are alive with sound and motion that can be captured with audio and video sensors. Rapid advances in audio and video classification algorithms can allow the recognition of species and labeling of complex traits and behaviors, which were traditionally the domain of manual species identification by experts. The major advance has been the discovery of deep convolutional neural networks (5). These algorithms extract fundamental aspects of contrast and shape in a manner analogous to how we and other animals recognize objects in our visual field. Applied to audio signals, these neural networks are highly effective at classifying natural and anthropogenic sounds (6). A canonical example is the classification of bird songs. Other acoustic examples include insects, amphibians, and disturbance indicators such as chainsaws. Naturally, these algorithms also lend themselves to species identification from images and videos. In cases of animals displaying complex color patterns, individuals may be distinguished, allowing minimally invasive mark recapture, an important tool in population studies and conservation (7). Beyond sight and sound, sensors can target a wide range of physical, chemical, and biological phenomena. Particularly intriguing is the possibility for widespread environmental sensing of biomolecular compounds that could, for example, allow quantification of “DNA-scapes” by means of laboratory-on-achip–type sensors (8). Several technological trends are shaping the emergence of large-scale sensor networks. One is the ongoing miniaturiza-

“...monitoring... species along with climate and other variables is a critical need in...adapting to emerging global challenges.”

Department of Integrative Biology, University of Texas at Austin, Austin, TX 78712, USA. Email: [email protected]

sciencemag.org SCIENCE

8/13/21 5:49 PM

PHOTO: SANDRO GRANDA

Small, ruggedized sensors, such as this passive acoustic recorder, enable remote monitoring of biodiversity. New technologies are enabling such devices to process data and transmit information via wireless networks.

tion of technology, allowing deployment of extended arrays of low-power sensor devices across landscapes [for example, (9)]. In many cases, these can be solar-powered in remote locations. The widespread availability of computer-on-a-chip devices along with various attached sensors is enabling the construction of large distributed sensing networks at price points that were formerly unattainable. Similarly, the ubiquitous availability of cloud-based computing and storage for back-end processing is facilitating large-scale deployments. Another trend is advancements in wireless communications. For example, the emerging internet of things (10) enables low-power devices to establish ad hoc mesh networks that can pass information from node to node, eventually reaching points of aggregation and analysis. The same technology used to connect smart doorbells and lightbulbs can be leveraged to move data across sensor networks distributed across a landscape. These protocols are designed for low power consumption but may not have sufficient bandwidth for all applications. An alternative, although more power hungry, is cellular technology, which has increasing coverage globally. In remote locations, where commercial cellular data services may not be available, researchers can consider a private cellular network for on-site telemetry and satellite uplinks for internet streaming. However, in the near term, telecommunications costs and per-device power requirements may nonetheless prove prohibitive in certain highbandwidth applications, such as video and SCIENCE sciencemag.org

0820Perspectives.indd 859

audio streaming. An alternative for sites where communications bandwidth is limited by cost, isolation, or power constraints is edge computing (11). In this design, computation is moved to the sensing devices themselves, which then transmit filtered or classified results for analysis, greatly reducing transmission requirements. One more trend is the advancement of machine-learning methods (12) that can classify and extract patterns from data streams. Much of this technology has been commoditized through intensive development efforts in the technology sector that have resulted in widely available software libraries usable by nonexperts. The aforementioned convolutional neural networks can be coded both to segment data into units and to label these units with appropriate classes. The major bottleneck is in training classifiers because initial training inputs must be labeled manually by experts. Although labeled training sets exist in some domains—most notably, image recognition—future analysts may be able to skip much of the training step as large collections of pretrained networks become available. These pretrained networks can be combined and modified for specific tasks without the requirement of comprehensive training sets. Of particular interest from the standpoint of automation are new developments in continual learning (13), in which networks adjust in response to changing inputs. This holds the promise of automating model adaptation for detecting emerging phenomena, such as species shifting their ranges in response

to climate change or other shifts in ecosystem properties. Ecologists could leverage these developments to create automated sensing networks at scales previously unimaginable. As an example, consider the North American Breeding Bird Survey, a highly successful citizen-science initiative running since the late 1960s with continental-scale coverage. Expert observers conduct point counts of birds along routes, generating data that have proved invaluable in tracking trends in songbird populations (14). Although we hope to see such efforts continue, imagine what could be learned if, instead of sampling these communities once per year, a long-term, continental-scale songbird observatory could be constructed to record and classify bird vocalizations in near–real time along with environmental covariates. Similar networks could use camera traps or video streams to reveal details of diurnal and seasonal variation across diverse floras and faunas. As with all sampling methods, sensing networks will not be without biases in sensitivity and discrimination, yet they hold the extraordinary promise of regional sampling of biodiversity at the organismal scale, something that has proven difficult, for example, by using traditional satellitebased remote sensing. These efforts would complement ongoing development of continental-scale observatories in ecology [for example, (15)] by increasing the spatial and temporal resolution of sampling. j REF ERENCES AND NOTES

1. S. Díaz et al., Science 366, eaax3100 (2019). 2. J. Travis, Am. Nat. 196, 1 (2020). 3. M. C. Dietze et al., Proc. Natl. Acad. Sci. U.S.A. 115, 1424 (2018). 4. B. J. Cardinale et al., Nature 486, 59 (2012). 5. Y. LeCun, Y. Bengio, G. Hinton, Nature 521, 436 (2015). 6. S. S. Sethi et al., Proc. Natl. Acad. Sci. U.S.A. 117, 17049 (2020). 7. R. C. Whytock et al., Methods Ecol. Evol. 12, 1080 (2021). 8. B. C. Dhar, N. Y. Lee, Biochip J. 12, 173 (2018). 9. A. P. Hill et al., Methods Ecol. Evol. 9, 1199 (2018). 10. L. Atzori, A. Iera, G. Morabito, Comput. Netw. 54, 2787 (2010). 11. W. Shi, J. Cao, Q. Zhang, Y. Li, L. Xu, IEEE Internet Things J. 3, 637 (2016). 12. M. I. Jordan, T. M. Mitchell, Science 349, 255 (2015). 13. R. Aljundi, K. Kelchtermans, T. Tuytelaars, Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 11254–11263. 14. J. R. Sauer, W. A. Link, J. E. Fallon, K. L. Pardieck, D. J. Ziolkowski Jr., N. Am. Fauna 79, 1 (2013). 15. M. Keller, D. S. Schimel, W. W. Hargrove, F. M. Hoffman, Front. Ecol. Environ. 6, 282 (2008). ACKNOWL EDGMENTS

Our perspective on autonomous sensing was developed with the support of the Stengl-Wyer Endowment and the Office of the Vice President for Research Bridging Barriers programs at the University of Texas at Austin, and the National Science Foundation (BCS-2009669). Comments from members of the Keitt laboratory, Planet Texas 2050, A. Wolf, and M. Abelson were invaluable in refining our ideas. 10.1126/science.abi4692 20 AUGUST 2021 • VOL 373 ISSUE 6557

859

8/13/21 5:49 PM

INSIGHTS

Kelsey Juliana, a plaintiff in Juliana v. United States, speaks outside the US Supreme Court in 2019.

B O OKS et al .

CLIMATE POLICY

Speth’s volume covers a broader period of time, says more about federal encouragement of fossil fuels, and, as befits a legal filing, is richly documented. While Speth tells readers that key reports on climate change started coming out of Washington in 1965, his detailed recounting begins with the administration in which he was a leading player, that of President Jimmy Carter (1977–1981). We learn that during this period, the scientific findings about the causes of climate change became increasingly confident and the predictions of its impacts became steadily more alarming. Yet under every US president since Carter, fossil fuel extraction and use have continued to grow. Federal “actions on the national energy system over the past several decades are, in my view, the greatest dereliction of civic responsibility in the history of the Republic,” Speth writes. Such a history of accumulating scientific knowledge of the dangers, and failure to act on them, can and has led to massive penalties against private companies—just ask the asbestos and tobacco industries. But the federal government enjoys “sovereign immunity,” meaning that it cannot be sued for monetary damages unless it has consented, and in the case of climate change, it has not. ing the extraction and use of fossil fuels. The Juliana case was ultimately decided To make that case, they brought in James by the Ninth Circuit in January 2020. Speth Gustave Speth. and the plaintiffs’ other experts apparently For half a century, Speth has been one of persuaded the judges of the points they were the nation’s foremost environmental leadtrying to make about the gravity and causes ers. He has been (in chronological order) of climate change, but it was not enough to a cofounder of the Natural Recompel action. To the heartbreak sources Defense Council; chair of many, the majority opinion of the White House Council on ends: “We reluctantly conclude, Environmental Quality under however, that the plaintiffs’ case President Carter; founder and, must be made to the political for a decade, president of the branches or to the electorate World Resources Institute; adat large, the latter of which can ministrator of the United Nachange the composition of the tions Development Programme; political branches through the They Knew: The US and dean of the Yale School of ballot box.” Federal Government’s Forestry and Environmental The current administration Fifty-Year Role in Causing Studies. He is now in a frenetic has made an ambitious pledge the Climate Crisis retirement as a senior fellow at to achieve a net-zero greenhouse James Gustave Speth MIT Press, 2021. 304 pp. Vermont Law School. gas economy by 2050. Whether Working pro bono, Speth prothis will be another example of duced a lengthy report tracing nearly 60 what former director of the Office of Technolyears of federal action on climate change ogy Assessment John Gibbons has referred and fossil fuel development. That report is to as “Disney’s Law” (“wishing will make it the foundation for this book. Parts of this so”) remains to be seen. Ultimately, however, story have been told before, most notably key choices on climate action remain in the by Spencer R. Weart in his 2003 book The hands of voters. j Discovery of Global Warming and Nathaniel Rich in his 2019 book Losing Earth, but 10.1126/science.abj9799

The children’s climate crusade By Michael B. Gerrard

I

n 2010, the group Our Children’s Trust formed to pursue a legal theory developed by University of Oregon law professor Mary Wood, which states that governments owe a duty to young people to protect the atmosphere from climate change. The group brought legal proceedings in all 50 US states, but the cases did not get very far, until one— Juliana v. United States—landed before Judge Ann Aiken of the US District Court in Oregon. Two days after Donald Trump was elected president, Judge Aiken issued a ruling allowing the case to go forward. When Trump took office, his administration tried mightily, but failed, to get the case thrown out. Meanwhile, trial preparations went forward. The plaintiffs needed to prove that the federal government not only knew about the dangers of climate change but also actually helped create them by encouragThe reviewer is a professor and director of the Sabin Center for Climate Change Law, Columbia Law School, New York, NY 10027, USA, and is coeditor of Legal Pathways to Deep Decarbonization in the United States (Environmental Law Institute, 2019). Email: [email protected]

860

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820Books.indd 860

PHOTO: AL DRAGO/BLOOMBERG/GETTY IMAGES

Prepared on behalf of young people seeking climate action, a report outlines the US federal government’s failures

sciencemag.org SCIENCE

8/13/21 2:43 PM

HISTORY OF PHYSICS

Before the Big Bang became scientific dogma A dual biography traces the entangled efforts of a pair of contentious cosmologists By Simon Mitton

PHOTOS (LEFT TO RIGHT): A. BARRINGTON BROWN/SCIENCE SOURCE; GRANGER

T

the atomic nuclei often enough to account for the source of stellar energy. Physicist Hans Bethe made the next advance, finding that proton-proton collisions in the cores of stars like the Sun fuse hydrogen to helium. For more-massive stars, he suggested a cycle of nuclear reactions in which carbon, nitrogen, and oxygen catalyze hydrogen to helium. This scheme left open the question that Gamow and Hoyle would confront head on: How did the elements from carbon to uranium come into existence? Hoyle entered the Cavendish Laboratory at the University of Cambridge in 1936 as a doctoral student supervised by Rudolf Peierls. As academics, including Peierls,

Flashes of Creation: George Gamow, Fred Hoyle, and the Great Big Bang Debate Paul Halpern Basic Books, 2021. 304 pp.

he serendipitous detection of the cosmic microwave background radiation in 1964 changed cosmology forever, settling a long-running debate about the origin of the Universe. The radio hiss hinted that the plode gravitationally, thereby sparking the Universe had arisen from an instantaneous physical conditions conducive to the rapid fiery beginning, a theory championed by assembly of heavier elements. cosmologist George Gamow, who sought The Gamowian school had considered to account for the origin of the chemical the role of neutrinos in core collapse, but elements. His rival, Fred Hoyle, who develHoyle’s powerful rebuttal of their model oped the alternative steady-state theory, in 1946 was vastly more efficient at buildwhich posited an infinite Universe, had ing heavy elements. By 1957, Hoyle’s team long insisted that the chemical elements had completed its brilliant synthesis of formed continuously in the element building via neutron cores of massive stars. Both capture reactions. However, cosmic models were falsifiable steady-state theory came unby solving a simple puzzle: Has der relentless attack as report the Universe evolved? The feeafter report by observational ble whisper detected in 1964 astronomers cemented Big was an undeniable “Yes!” Bang cosmology. In Flashes of Creation, Paul In 1964, Hoyle reluctantly Halpern presents a scintillatconceded that “a small residue ing account of the intellectual of Gamow’s idea”—the syntravails of Gamow and Hoyle, thesis of light elements in the two animated, curious, provocBig Bang—had merit. Within ative, and controversial figures months, news broke of the in 20th-century physics. In this discovery of the cosmic microjoint biography, the reader is inwave background. Hoyle never troduced to the two physicists’ accepted this as evidence that theories and their efforts to ex“the entire cosmos had a start plain the origin of elements. date.” By contrast, Gamow opGamow, we learn, first enFred Hoyle (left) and George Gamow disagreed about the origins of the Universe. portunistically seized the mocountered cosmology in the ment, claiming primary credit early 1920s while studying at the University later fled the Cavendish Laboratory to profor a neglected prediction of the backof Leningrad under Alexander Friedmann, fessorships elsewhere, Hoyle remained at ground temperature made in 1948 by his asthe Russian mathematician who pioneered Cambridge until the war years, working sociates Ralph Alpher and Robert Herman. the idea that the Universe is expanding. In alone on extending Enrico Fermi’s theory In the book’s closing pages, Halpern senGöttingen and Copenhagen, while a docof beta decay. By peacetime, he had develsitively handles with commendable candor toral student in physics, he mingled with oped the steady-state theory and witherthe tragic endgames of these two giants. pioneers who were working on the new ingly dismissed Gamow’s cosmology as a Gamow’s alcoholism, we learn, destroyed quantum theory. These interactions enabled mere “big bang.” him and much of his reputation. And while his breakthrough in 1928, when he showed Hoyle could perceive no merit in GaHoyle commanded great respect after rehow an alpha particle could escape from an mow’s notion that the elements were cresigning from Cambridge in 1972, his little atomic nucleus by quantum tunneling. ated in a flash by the eruption of a primeval tweaks to steady-state cosmology failed to Gamow’s subsequent realization that atom—it violated the conservation laws of find a following. quantum tunneling is reversible spurred physics. His ageless steady-state approach Gamow and Hoyle were friendly rivals two colleagues, Robert Atkinson and Fritz envisaged that new matter trickled conwho seldom interacted in person. Halpern Houtermans, to demonstrate that suffitinuously into the empty space left by the nonetheless renders their contributions ciently energetic protons could penetrate expansion of the Universe. The buildup of and clashes vividly in this expertly crafted chemical elements then arose as a consebiography of two contentious cosmologists quence of the evolution of massive stars, who thrived on ingenious invention. j The reviewer is a Life Fellow of St Edmund’s College, he postulated. When the hydrogen fuel in a University of Cambridge, Cambridge CB3 0BN, UK. Email: [email protected] star’s core became exhausted, it would im10.1126/science.abj9479 SCIENCE sciencemag.org

0820Books.indd 861

20 AUGUST 2021 • VOL 373 ISSUE 6557

861

8/13/21 2:43 PM

Announcing a

Request for Proposal

Operation Payload Delivery Teams will be selected to design and develop a novel, multifaceted, scalable, non-immunogenic strategy to deliver complex macromolecular therapeutic payloads to specific target cells and tissues. The payload must be delivered into the target cells in a biologically active form that can induce phenotypic and/or genotypic changes and enable healing for the patient. Funding as high as $1M per laboratory White papers due September 15, 2021 Full proposals due November 15, 2021 Access the RFP and find more information:

The Ion-ARPA initiative

www.ionispharma.com/ionis-innovation/ion-arpa-initiative/

Launched by Ionis Pharmaceuticals, Ion-ARPA will fund multiple teams to create revolutionary new therapeutic technologies. The Ion-ARPA approach will facilitate innovation of novel cutting-edge technologies and will be funded based on high-risk ideas with high-reward potential for pioneering new markets in health care.

Where Science Gets Social. AAAS.ORG/COMMUNITY

0820Product.indd 862

8/12/21 7:37 AM

INSI GHTS | L E T T E R S

LET TERS

Kelp forests provide a range of ecosystem services.

Edited by Jennifer Sills

PHOTO: REINHARD DIRSCHERL/ALAMY STOCK PHOTO

Embrace kelp forests in the coming decade The United Nations General Assembly recognizes that securing human lives and livelihoods will require putting an end to global habitat degradation and restoring hundreds of millions of hectares of lost habitats. In response, the organization has declared 2021 to 2030 the UN Decade on Ecosystem Restoration (1). Perhaps the greatest challenge for the upcoming decade is upscaling restoration efforts to match the extent of habitat loss (2). Coral reefs and tropical forests have been highlighted as flagships of conservation need and as priority ecosystems in the UN Decade (1). In contrast, kelp forests are conspicuously missing from the UN recommendations (1). Kelp forests provide critical ecosystem services to humans, similar to those provided by coral reefs and tropical forests (3). They also possess a much greater capacity for rapid growth and regeneration than either of these ecosystems (4). The benefits provided by kelp forests span 14 of the 18 categories of nature’s contributions to people identified by the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (5), including key features such as biodiversity provisioning, coastal protection, and carbon dioxide SCIENCE sciencemag.org

0820Letters.indd 863

absorption and storage (3). Recently, the US government included kelp forests as an “essential” component of the federal strategy to address the climate change crisis (6). Kelp forests cover 28% of the world’s coastlines (7) and five times more ocean area than coral reefs (8, 9). They are declining under anthropogenic forces two and four times faster than coral reefs and tropical forests, respectively (10–12). Given the plethora of essential ecosystem services that kelp forests provide to humans, their rapid re-establishment and growth rates, and their past and current rates of decline, we contend that kelp forests hold unprecedented potential for restoration success. Embracing kelp forest restoration will greatly increase our chances of overcoming the upscaling challenge in restoration and delivering effective global action in the UN Decade. Colette J. Feehan1*, Karen Filbee-Dexter2,3,4, Thomas Wernberg2,3 1

Department of Biology, Montclair State University, NJ 07043, USA. 2Institute of Marine Research, His, 4817, Norway. 3School of Biological Sciences and Oceans Institute, University of Western Australia, Perth, WA 6009, Australia. 4Universite Laval, Quebec, QC G1V 0A6, Canada. *Corresponding author. Email: [email protected] REF ERENCES AND NOTES

1. The United Nations Decade on Ecosystem Restoration (2021); www.decadeonrestoration.org/. 2. M. I. Saunders et al., Curr. Biol. 30, R1500 (2020). 3. T. Wernberg, K. Krumhansl, K. Filbee-Dexter, M. F. Pedersen, in World Seas: An Environmental Evaluation (Academic Press, 2019), pp. 57–78. 4. A. M. Eger et al., Front. Mar. Sci. 7, 811 (2020).

5. S. Díaz et al., “Summary for policymakers of the global assessment report on biodiversity and ecosystem services of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services” (IPBES, Bonn, Germany, 2019). 6. US Office of National Environmental Policy Act Policy and Compliance, “Executive Order 14008: Executive order on tackling the climate change crisis at home and abroad” (2021). 7. S. Starko, D. P. Wilkinson, T. T. Bringloe, Biol. Conserv. 257, 109082 (2021). 8. D. R. Jayathilake, M. J. Costello, Biol. Conserv. 252, 108815 (2020). 9. M. Spalding, M. D. Spalding, C. Ravilious, E. P. Green, World Atlas of Coral Reefs (University of California Press, 2001). 10. K. A. Krumhansl et al., Proc. Natl. Acad. Sci. U.S.A. 113, 13785 (2016). 11. National Academies of Sciences, Engineering, and Medicine, “A research review of interventions to increase the persistence and resilience of coral reefs” (The National Academies Press, Washington, DC, 2019). 12. F. Achard et al., Glob. Change Biol. 20, 8 (2014). 10.1126/science.abl3984

Mexico’s final death blow to the vaquita A recent decision by the government of Mexico to reduce enforcement of illegal fishing in the upper Gulf of California is pushing the vaquita (Phocoena sinus)— the world’s smallest cetacean, endemic to Mexico—toward final extinction (1–3). Fewer than 10 individual vaquitas remain after decades of devastating bycatch mortality (1). Yet Mexico will expose the vaquita to continued risks of injury or death by allowing uncontrolled gillnet fishing for the totoaba, an endangered 20 AUGUST 2021 • VOL 373 ISSUE 6557

863

8/16/21 3:47 PM

INSIGHTS | L E T T E R S

fish in the vaquita’s last habitat sanctuary, the upper Gulf of California World Heritage site (1, 3). Mexico’s decision will further facilitate the use of totoaba swim bladders in traditional Chinese medicine (4). Instead, the country should prioritize its responsibility to the Critically Endangered vaquita (1). Previous attempts to protect this small harbor porpoise have failed, including collecting specimens for captive breeding in 2017 (4). Meanwhile, stressors such as vessel strikes, underwater noise, and pollution have increased (4–8). According to the International Union for Conservation of Nature, the vaquita is close to being functionally extinct as a result of its high mortality rates and low reproductive output, coupled with its historically low genetic diversity, all of which jeopardize population health (9, 10). The vaquita’s essential protection measures decrease illegal fishing in these waters, helping to stabilize the ecosystem’s functioning, sustainability, and biodiversity and supporting several of the UN’s Sustainable Development Goals (11). A complex set of problems drives illegal fishing, including local poverty, local organized crime, and international demand for endangered species. Financial incentives attempted by the Mexican government have proven ineffective (9, 12). Mexico should increase enforcement of current regulations that limit fishing in the vaquita’s habitat, which are critical to saving this species from extinction (12). Christian Sonne1*, Pindaro Diaz-Jaimes2, Douglas H. Adams3 1

Aarhus University, Roskilde, Denmark. 2Unidad Académica de Ecologia y Biodiversidad Acuática, Instituto de Ciencias del Mar y Limnologia, Universidad Nacional Autónoma de Mexico, 04510 Mexico City, Mexico. 3Cape Canaveral Scientific, Melbourne Beach, FL 32951, USA, *Corresponding author. Email: [email protected]

Piecing together an African peace park In August 2011, Angola, Botswana, Namibia, Zambia, and Zimbabwe signed a treaty to create the Kavango Zambezi (KAZA) Transfrontier Conservation Area (or Peace Park)—the world’s largest transboundary terrestrial conservation area—to protect the region’s biodiversity and cultural resources and to alleviate poverty (1, 2). Ten years later, the five KAZA countries have made great strides, but the habitat connectivity that KAZA’s wildlife requires for long-term ecological viability (3) remains in question. If key wildlife movement corridors are not reopened and secured, the vision of KAZA’s wildlife providing benefits to the people of the region in perpetuity may not be realized. KAZA is home to the majority (at least 220,000) of what is left of Africa’s elephants (4), and perhaps no other species better demonstrates the need for KAZA to be a truly connected landscape. In northern Botswana’s Ngamiland District, home of the Okavango Delta (a World Heritage site), where the elephant population continues to grow, thousands of elephants are increasingly bottled up between villages and vast livestock disease control fences that prevent them from moving through nearby Namibia into Angola and Zambia (4, 5). Decreasing pressure on Ngamiland’s elephants is crucial to reducing the human-elephant conflict that is unfortunately becoming more and more common (5). Six wildlife dispersal areas, or habitat corridors, have been identified as critical to securing a long-term future for KAZA’s iconic wilderness and species (6). However,

the most important of these corridors, including those that connect Ngamiland to other key parts of KAZA, remain compromised by fences, many of which were put in place decades ago to control animal disease (7) but no longer necessarily serve their original purpose. Today, risks associated with foot and mouth disease can be managed by focusing on biosecurity across the beef production process (8), and contagious bovine pleural pneumonia is no longer the threat that it was in the mid-1990s (9). Neither the livestock nor wildlife sector should dominate the other. Instead, now is the time to make land-use decisions that will be socially, ecologically, and economically sustainable for generations to come (10). Steven A. Osofsky1* and Russell D. Taylor2 1

Cornell University College of Veterinary Medicine, Ithaca, NY 14853, USA. 2World Wildlife Fund Namibia, Windhoek, Namibia. *Corresponding author. Email: [email protected] REF ERENCES AND NOTES

1. Peace Parks Foundation, “KAZA Treaty Signed” (2011); www.peaceparks.org/kaza-treaty-signed/. 2. Kavango Zambezi: Transfrontier Conservation Area (KAZA TFCA) (2019); https://kavangozambezi.org/en/. 3. A. Brennan et al., J. Appl. Ecol. 57, 1700 (2020). 4. KAZA TFCA, “Strategic Planning Framework for the Conservation and Management of Elephants in the Kavango Zambezi Transfrontier Conservation Area” (2019); www.kavangozambezi.org/en/ publications-2019. 5. L. Redmore et al., Ecol. Soc. 25, 27 (2020). 6. Southern African Development Community TFCA Network, “Kavango Zambezi Transfrontier Conservation Area (KAZA TFCA) Master Integrated Development Plan” (2016); https://tfcaportal.org/ kaza-tfca-master-idp. 7. J. E, Mbaiwa and O. Mbaiwa, Int. J. Wildern. 12, 17 (2006). 8. G. Thomson et al., Transbound. Emerg. Dis. 60, 492 (2013). 9. W. Amanfu et al., Vet. Rec. 143, 46 (1998). 10. S. A. Osofsky, J. Wildl. Dis. 55, 755 (2019). 10.1126/science.abl7447

REFERENCES AND NOTES

10.1126/science.abl5834

864

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820Letters.indd 864

PHOTO: GARETH BENTLEY COPYRIGHT WWF

1. A. M. Jaramillo-Legorreta et al., R. Soc. Open Sci. 6, 190598 (2019). 2. L. Rojas-Bracho et al., in Encyclopedia of Marine Mammals, Third Edition, B. Würsig, J. G. M. Thewissen, K. M. Kovacs, Eds. (Academic Press, 2018), pp. 1031–1035. 3. “Mexico gives up on maintaining fishing-free zone to protect vaquita porpoise,” Mexico News Daily (2021). 4. L. Rojas-Bracho et al., Endang. Spec. Res. 38, 11 (2019). 5. R. C. Bishop et al., Science 356, 253 (2017). 6. B. Würsig, in Habitats and Biota of the Gulf of Mexico: Before the Deepwater Horizon Oil Spill, C. H. Ward, Ed. (Springer, New York, vol. 2, 2017), pp. 1489–1587. 7. W. K. Meyer et al., Science 361, 591 (2018). 8. G. Ponce-Vélez, G. de la Lanza-Espino, J. Environ. Protect. 10, 103 (2019). 9. Vaquita (IUCN–SSC Cetacean Specialist Group, 2020); https://iucn-csg.org/vaquita/. 10. P. A. Morin et al., Mol. Ecol. Resour. 21, 1008 (2021). 11. C. C. O’Hara et al., Science 372, 84 (2021). 12. E. C. Alberts, “In the fight to save the vaquita, conservationists take on cartels,” Mongabay (2021).

Unobstructed corridors for wildlife such as elephants are crucial to southern Africa’s conservation efforts. sciencemag.org SCIENCE

8/16/21 3:47 PM

CALL FO R PA PERS

BioDesign Research is a Science Partner Journal published in affiliation with Nanjing Agricultural University (NAU) and distributed by the American Association for the Advancement of Science (AAAS). BioDesign Research publishes high quality breakthrough research, reviews, editorials, and perspectives focusing on in silico biosystems design, genetic or epigenetic modifications, and genome writing or rewriting in any organism.

Submit your research to BioDesign Research today! Learn more at spj.sciencemag.org/bdr The Science Partner Journals (SPJ) program was established by the American Association for the Advancement of Science (AAAS), the non-profit publisher of the Science family of journals. The SPJ program features high quality, online-only, editorially independent openaccess publications produced in collaboration with international research institutions, foundations, funders and societies. Through these collaborations, AAAS expands its efforts to communicate science broadly and for the benefit of all people by providing a top-tier international research organization with the technology, visibility, and publishing expertise that AAAS is uniquely positioned to offer as the world’s largest general science membership society. Learn more at spj.sciencemag.org @SPJournals

@SPJournals

A R T I C L E P R O C E S S I N G C H A R G E S WA I V E D U N T I L 2 0 2 2

0820Product.indd 865

8/12/21 7:37 AM

RESEARCH

VOCAL DEVELOPMENT

Babbling bats

A

notable aspect of language development in humans is the babbling stage. During this time, toddlers make a range of specific sounds as they practice and imitate adult speech. Humans are not the only vocal learners, however, so might we expect such babbling among others? Fernandez et al. recorded the vocalizations of sac-winged bat pups in the wild and found clear evidence of babbling that was consistent with that seen in humans. The shared babbling components suggest that vocal learning may have similar specific mechanisms across a wide array of mammalian species. —SNV

IN S CIENCE JOURNAL S Edited by Stella Hurtley

Science, abf9279, this issue p. 923

As the greater sac-winged bat (Saccopteryx bilineata) learns vocalizations from its elders, it goes through the equivalent of a human infant babbling stage.

Deep learning takes on protein folding In 1972, Anfinsen won a Nobel prize for demonstrating a connection between a protein’s amino acid sequence and its three-dimensional structure. Since 1994, scientists have competed in the biannual Critical Assessment of Structure Prediction (CASP) proteinfolding challenge. Deep learning methods took center stage at CASP14, with DeepMind’s Alphafold2 achieving remarkable accuracy. Baek et al. explored network architectures based on the DeepMind framework. They used a three-track network to process sequence, distance, and coordinate information simultaneously and achieved accuracies 866

approaching those of DeepMind. The method, RoseTTA fold, can solve challenging x-ray crystallography and cryo–electron microscopy modeling problems and generate accurate models of protein-protein complexes. —VV Science, abj8754, this issue p. 871

RNA DELIVERY

Hitching a ride with a retroelement Retroviruses and retroelements have inserted their genetic code into mammalian genomes throughout evolution. Although many of these integrated viruslike sequences pose a threat to genomic integrity, some have been retooled by mammalian cells to perform essential roles in development. Segel et al. found

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820ISIO.indd 866

that one of these retroviral-like proteins, PEG10, directly binds to and secretes its own mRNA in extracellular virus–like capsids. These virus-like particles were then pseudotyped with fusogens to deliver functional mRNA cargos to mammalian cells. This potentially provides an endogenous vector for RNA-based gene therapy. —DJ Science, abg6155, this issue p. 882

HYDRAULIC FRACTURING

Bonetti et al. found a small increase in certain ions associated with hydraulic fracturing across several locations in the United States (see the Perspective by Hill and Ma). These small increases appeared 90 to 180 days after new wells were put in and suggest some surface water contamination. The magnitude appears small but may require that more attention be paid to monitoring near-well surface waters. —BG Science, aaz2185, this issue p. 896; see also abk3433, p. 853

Lightly salted surface waters

SOLAR CELLS

Hydraulic fracturing uses a waterbased mixture to open up tight oil and gas formations. The process is mostly contained, but concerns remain about the potential for surface water contamination.

The buried interfaces of perovskite solar cells are difficult to alter after synthesis. During manufacture, Chen et al. removed perovskite films with dimethyl

Avoiding buried voids

PHOTO: MICHAEL STIFTER

PROTEIN FOLDING

sciencemag.org SCIENCE

8/16/21 5:07 PM

sulfoxide solvent from the holetransfer layer and observed a substantial void fraction that degraded film performance. Replacing most of the dimethyl sulfoxide with carbohydrazide, a lead-coordinating compound with a much higher boiling point, eliminated voids. Such solar cells maintained high power conversion efficiency after 550 hours of operation at 60°C. —PDS Science, abi6323, this issue p. 902

OXIDE ELECTRONICS

Watching a metal filament grow Resistive switching is a process in which the electrical resistance of a sample changes abruptly in response to a voltage pulse, often by orders of magnitude. This process is at the heart of many neuromorphic computing approaches but visualizing it in both space and time is tricky. del Valle et al. monitored the resistive switching in three different vanadium oxide compounds by measuring time- and spaceresolved optical reflectivity (see the Perspective by Hilgenkamp and Gao). A characteristic conducting filament was quickly nucleated on the inhomogeneities in the sample and then propagated due to Joule heating. —JS Science, abd9088, this issue p. 907; see also abh2231, p. 854

MALARIA

PHOTO: BLICKWINKEL/ALAMY STOCK PHOTO

Childhood malaria Understanding how changes in community parasite prevalence alter the rate and age distribution of severe malaria is essential for optimizing control efforts. Paton et al. assessed the incidence of pediatric severe malaria admissions from 13 hospitals in East Africa from 2006 to 2020 (see the Perspective by Taylor and Slutsker). Each 25% increase in community parasite prevalence shifted hospital admissions toward younger children. Low rates of lifetime infections appeared to confer some immunity to severe malaria in very young children. Children under the age of 5 years thus need to SCIENCE sciencemag.org

0820ISIO.indd 867

remain a focus of disease prevention for malaria control. —CA Science, abj0089, this issue p. 926; see also abk3443, p. 855

IN OTHER JOURNALS

Edited by Caroline Ash and Jesse Smith

CANCER

ALT-ering neuroblastoma chemoresistance Neuroblastoma can develop therapeutic resistance by adopting an alternative lengthening of telomeres (ALT) mechanism to continue replicating. Using patient-derived cell lines, Koneru et al. show that this telomere dysfunction promotes constitutive ataxia-telangiectasia mutated (ATM) kinase activation in ALT neuroblastoma. This in turn promotes chemoresistance to combined temozolomide plus irinotecan, a common salvage treatment for ALT neuroblastoma. Adding an ATM inhibitor to the mixture enhanced therapeutic efficacy in in vitro and in patientderived xenografts, suggesting a potential strategy to target ALT neuroblastoma chemoresistance. —CAC Sci. Transl. Med. 13, eabd5750 (2021).

ALLERGY

Understanding eosinophilic esophagitis Eosinophilic esophagitis (EoE) is an allergic disease triggered by exposure to food-derived allergens. Morgan et al. examined tissue-specific immune responses underlying EoE using paired single-cell RNA and T cell receptor sequencing of esophageal, peripheral blood, and duodenal samples collected from patients. Eosinophils enriched for nuclear factor kB signaling pathways and clonally expanded pathogenic effector T helper 2 (peTH2) cells were elevated in the esophagus of patients. In peripheral blood, expression of the chemokine receptor GPR15 enriched for milk-reactive T cells and for peTH2 clonotypes also detected in the esophagus. Thus, certain food antigen–specific T cells are poised for esophageal homing. —CO Sci. Immunol. 6, eabi5586 (2021).

NEURODEVELOPMENT

Hydra recovers from the “Blip”

T

he humble Hydra holds an unusual superpower. If it is disassembled experimentally, its cells will reassemble correctly toadform a complete organism. Three days Da si iust apelectur mintium after disaggregation, the reassembled Hydra recovers conestem sequiberibus doluptatqui synchronized neuronal activity and rebuilds normal neuronal networks. Lovas and Yuste show how the neural net reforms from its individual pieces. The disaggregated neurons first resume firing and then grow neurites as they build small, local connections. These neuronal ensembles then connect into midsized modules as the initially hierarchical network transitions into more distributed structure. Network synchronization emerges as the connections consolidate. Although at first, highly connected nodes carry most of the network traffic, further refinement adds smaller nodes. As the circuits mature, the input of individual neurons shifts and a resilient distributed network reemerges. —PJH Curr. Biol. 10.1016/j.cub.2021.06.047 (2021).

Hydra is a small freshwater organism, related to jellyfish, that is capable of remarkable feats of regeneration.

PARASITOLOGY

Collapse of fluke populations Parasitic flukes have highly complex life cycles circumscribing a definitive host’s ecology. Parasites in wildlife are underappreciated, poorly recorded, and many species are endangered. Sitko and Heneberg collate reports of long-term changes in

fluke communities found in euthanized injured birds over the past 50 years in the Czech Republic. In all, 33 species of fluke were identified and most were species specific. Over the decades, trematode abundance declined and assemblages simplified; for example, since 1980, lapwing specimens have been found to be free of flukes. The authors attribute the parasite declines

20 AUGUST 2021 • VOL 373 ISSUE 6557

867

8/16/21 5:07 PM

RESE ARCH | I N O T H E R J O U R NA L S

Fluid support

T

hree-dimensional printing is a powerful tool for rapidly creating customized shapes, particularly out of soft materials such as polymers. However, it still can be challenging to print very soft materials such as hydrogels, which are of interest for medical devices and cell delivery, because of their low mechanical stiffness. Beh et al. overcame this by patterning a fluid-supported hydrogel precursor that provides structural support for the growing object while rinsing excess precursor from the printed regions. The authors were able to print materials spanning three orders of magnitude in stiffness and combined the printer with an extruder to perform multimaterial printing. —MSL Biomaterials 276, 121034 (2021). A three-dimensional soft hydrogel structure (pictured above) bioprinted with the help of fluid support

not solely to a loss of host abundance but also to changing agricultural practices, landscape simplification, and the broad use of agricultural chemicals such as azole fungicides, a chemical family also used as anthelmintics. This is bad news for the flukes but might not be all bad in the short term for the breeding success of the birds. —CA Parasites Vectors 14, 383 (2021).

largely because of epigenomic effects on chromatin accessibility. Loss of LKB1 activates distinct waves of epigenetic remodeling in primary and metastatic tumor cells. Metastasis was mediated by activation of the transcription factor SOX17. The authors were then able to detect a small population of cells in primary tumors with the epigenetic characteristics of later metastatic cells, revealing a promising therapeutic target. —LBR

CANCER

Nat. Cell Biol. 23, 915 (2021).

Epigenetic control of metastasis

GENOMIC ANALYSIS

Metastasis is when cancer becomes deadly. Unsuccessful searches for driver mutations of metastasis have led some to explore alternative epigenetic causes. Pierce et al. found that the protein kinase LKB1, better known for influencing cancer through regulation of metabolism, had a tumor-suppressive role in one mouse lung cancer model. This tumor develops

Identifying functional genetic variation in humans requires sifting through hundreds of thousands of individual variants and linking them to the trait of interest. We often do not know whether a gene is functional in a tissue or specific cell. Machinelearning models have become valuable for such endeavors. Somepalli et al. developed a

868

PLoS Comput. Biol. 17, e1009194 (2021).

Contrapuntal gene risk

20 AUGUST 2021 • VOL 373 ISSUE 6557

0820ISIO.indd 868

model they call FUGUE, which they used to map the tissuespecific expression of human disease–associated genes and their protein context and interactions. Interestingly, FUGUE revealed that tissue-relevant genes cluster on the genome within topologically associated domains. The authors supply prioritized gene lists for 30 human tissues for genes associated with heart disease, Alzheimer’s disease, cancer, and development. —LMZ

MARTIAN ATMOSPHERE

Methane on Mars is transient and local The Curiosity rover has detected seasonally varying levels of methane in Gale crater on Mars, which could have geological or astrobiological sources. The ExoMars Trace Gas Orbiter (TGO) spacecraft has detected

Astron. Astrophys. 650, A166, A140 (2021).

SCIENTIFIC WORKFORCE

When an internship adds value Graduate schools in the United States are recalibrating the equilibrium between the need to provide adequate professional development and the productivity and length of a doctoral education. Traditionally, more weight was given to laboratory productivity at the expense of professional development. Brandt et al. challenge this belief by analyzing metrics from 10 U.S. institutions participating in the National Institutes of Health’s Broadening Experiences in Scientific Training (BEST) professional development program. Comparing doctoral students who participated in professional development activities with those who did not revealed no differences in time to degree or manuscript output, contradicting the belief that career exploration and professional development programming adversely affect graduate education. These results should inspire and encourage institutions to incorporate more career exploration activities into doctoral training programs. —MMc PLoS Biol. 10.1371/journal. pbio.3000956 (2021).

PHOTO: BIOMATERIALS 276, 121034 (2021)

ADDITIVE MANUFACTURING

no methane on Mars despite having better sensitivity than Curiosity, but had not yet examined the Gale location. Measurements from TGO were always for high altitudes during local dawn or dusk and those from Curiosity for the surface at midnight. Webster et al. used Curiosity to search for methane during the day, finding none and inferring a day–night cycle. Montmessin et al. performed TGO measurements at new locations, including close to Gale, ruling out methane with tight upper limits. Together, these two studies indicate that methane is released and trapped within Gale crater during the night and then rapidly dissipates or is destroyed during the day. —KTS

sciencemag.org SCIENCE

8/16/21 5:07 PM

RE S E ARC H

ALSO IN SCIENCE JOURNALS SYSTEMS BIOLOGY

DNA repair amplifies transcriptional noise The potential role of “noise,” or stochastic variations in rates of gene expression, remains to be elucidated. Desai et al. used screens to identify a compound, 59-iodo-29-deoxyuridine (IdU), that increased gene expression noise in mouse embryonic stem cells in culture without changing the overall rate of transcription of most genes. They propose a model by which the thymidine analog IdU promotes binding of the base excision repair protein AP endonuclease to DNA, thereby inducing helical distortion of DNA and modulating transcriptional bursting. Such modulation of noise enhanced reprogramming of the embryonic stem cells. Thus, variation in gene expression noise could influence developmental or disease processes. —LBR Science, abc6506, this issue p. 870

GLACIERS

Waters of high Asia How the rivers of the HimalayaKarakoram region of Asia respond to climate change is critical for the billion-plus people who depend on the water that they provide. In a Review, Azam et al. discuss recent progress in understanding the importance of glacier and snow melt in the hydrological budget there, which is driven largely by advances in remote sensing and modeling. Observational data remain sparse and challenging to collect. —HJS Science, abf3668, this issue p. 869

TRANSLATION

How translation stops Protein synthesis concludes when a ribosome encounters a stop codon in a transcript, which triggers the recruitment of highly conserved release factors to liberate the protein product. Lawson et al. used traditional SCIENCE sciencemag.org

0820eISIO.indd 868-B

Edited by Stella Hurtley

biochemical methods and single-molecule fluorescence assays to track the interplay of release factors with ribosomes and reveal the molecular choreography of termination. They identified two distinct classes of effectors, small molecules and mRNA sequences, that directly inhibited the release factors and promoted stop codon readthrough. These findings may buttress ongoing efforts to treat diseases caused by premature stop codons, which cause 11% of all heritable human diseases. —DJ Science, abi7801, this issue p. 876

METALLURGY

High-entropy herringbone alloy Eutectic high-entropy alloys have a dual-phase structure that could be useful for optimizing a material’s properties. Shi et al. found that directional solidification of an aluminumiron-cobalt-nickel eutectic high-entropy alloy created a herringbone-patterned microstructure that was extremely resistant to fracture (see the Perspective by An). The structure contained lamellae of hard and soft phases, and the cracks that formed in the hard phase were arrested at the boundary of the soft phase. This, along with stress transfer, allowed a tripling of the maximal elongation while retaining high strength. —BG Science, abf6986, this issue p. 912; see also abk1671, p. 857

3CLpro, plays a key role in these cleavages, making it an important drug target. Drayman et al. identified eight drugs that target 3CLpro from a library of 1900 clinically safe drugs. Because of the challenge of working with SARS-CoV-2, they started by screening for drugs that inhibit the replication of a human coronavirus that causes the common cold. They then evaluated the top hits for inhibiting SARS-CoV-2 replication and for inhibiting 3CLpro. Masitinib, a broad antiviral, inhibited the main proteases of coronaviruses and picornaviruses and was effective in reducing SARSCoV-2 replication in mice. —VV Science, abg5827, this issue p. 931

Science, abg5953, this issue p. 918

PHARMACOLOGY

Precision blockade of inflammatory IL-6

CORONAVIRUS

Fueling outbreaks

The cytokine interleukin-6 (IL-6) has critical functions in various tissues but is also implicated in autoimmune disease. Precisely targeting only the pathway by which IL-6 induces inflammatory immune cell activation would be desirable. Heise et al. developed a chimeric molecule that bound to and trapped an IL-6 trans-signaling protein complex. Compared with a previously developed molecule that targets this complex, the authors’ molecule was more selective for the IL-6 complex than a similar IL-11 complex and more effectively inhibited the IL-6–induced inflammatory activation of cultured T cells. —LKF Sci. Signal. 14, eabc3480 (2021).

CORONAVIRUS

Targeting the main protease of SARS-CoV-2 Inside host cells, the RNA genome of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is translated into two polyproteins that are cleaved to yield the individual viral proteins. The main viral protease, known as Mpro or

excess of that of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Chen et al. used whole-genome sequencing to investigate the contribution of rare mutations among poultry workers, who can be exposed to high levels of H7N9. Multiple defective single-nucleotide variants in the myxovirus resistance Mx1 locus were prevalent in H7N9 patients. In vitro infection experiments and influenza polymerase activity assays showed that 14 of the 17 MxA protein variants had no antiviral activity. Thus, when exposed to high virus loads, individuals with such genetic vulnerabilities may act as crucibles for transmission of virulent new influenza subtypes. —CA

INFLUENZA

Poultry passport to pandemic What conditions are required to nurture the seeds of a pandemic? The avian influenza virus H7N9 rarely spills over into humans, but when it does, mortality exceeds 30%, far in

The B.1.1.7 lineage of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has caused fast-spreading outbreaks globally. Intrinsically, this variant has greater transmissibility than its predecessors, but this capacity has been amplified in some circumstances to tragic effect by a combination of human behavior and local immunity. What are the extrinsic factors that help or hinder the rapid dissemination of variants? Kraemer et al. explored the invasion dynamics of B.1.1.7. in fine detail, from its location of origin in Kent, UK, to its heterogenous spread around the country. A combination of mobile phone and virus data including more than 17,000 genomes shows how distinct phases of dispersal were related to intensity of mobility and the timing of lockdowns. As the local outbreaks grew, importation from the London source area became less important. Had B.1.1.7. emerged at a slightly different time of year, its impact might have been different. —CA Science, abj0113, this issue p. 889

20 AUGUST 2021 • VOL 373 ISSUE 6557

868-B

8/16/21 5:11 PM

ScienceImmunology.org

READY TO PUT THE SPOTLIGHT ON YOUR RESEARCH? Submit your research: cts.ScienceMag.org

Twitter: @SciImmunology Facebook: @ScienceImmunology

RESEAR CH

REVIEW SUMMARY

◥

GLACIERS

Glaciohydrology of the Himalaya-Karakoram Mohd. Farooq Azam*, Jeffrey S. Kargel, Joseph M. Shea, Santosh Nepal, Umesh K. Haritashya, Smriti Srivastava, Fabien Maussion, Nuzhat Qazi, Pierre Chevallier, A. P. Dimri, Anil V. Kulkarni, J. Graham Cogley, Isamohan Bahuguna

BACKGROUND: The Himalayan-Karakoram (HK)

HK rivers. Recent efforts to map and measure glacier extents, mass balance, and velocity, as well as the growth of glacial lakes, have filled several major gaps in glaciohydrology that existed a decade ago. This progress has been achieved primarily through remote sensing and modeling, yet field-based studies remain limited. The combined result of improved models and observations suggests that snow and ice melt are important but spatially variable runoff components in HK rivers. Meltwater contributions are highest closest to the snow and ice sources, and the meltwater contribution in the Indus is greater than in the Ganges and Brahmaputra basins. Meltwater contributions vary widely between catchments as a result of orographic microclimates and the relative proportions of summer and winter precipitation. However, the contributions of runoff components estimated for the same catchments vary between studies, highlighting discrepancies in model approaches and assumptions. Projected 21st-century trends in the seasonality of runoff and the increasing intensity and frequency of extreme runoff events are consist-

region in south Asia is one of the most heavily glacierized and vulnerable mountainous regions on Earth. The Indus, Ganges, and Brahmaputra river systems, which originate from HK glaciers and snowfields, support the water requirements of 1 billion people. HK river basins have the largest irrigated area (~577,000 km2) and the largest installed hydropower capacity (~26,000 MW) worldwide. Optimum planning for management of water demand and supply for agriculture, hydropower, domestic needs, and sanitation requires a consensus on the region’s glaciohydrology. Understanding the uncertainties in glaciohydrological modeling, climate change projections, and their impacts on the availability of water and its transboundary nature in HK rivers is thus critical for sustainable water resource management and regional geopolitics. ADVANCES: Glaciohydrological models have

been used to investigate contributions of glacier and snow melt, impacts of climate change on melt runoff, and future runoff evolution in

IWM

ISM Precipitation

Glacier area/ volume

Debris cover

Black carbon deposition

Permafrost

Snowmelt

Glacier dynamics

Moisture transportation

Sublimation

Permafrost melt

Evapotranspiration

Precipitation

Black carbon Evaporation emission

Ocean

Lake

Groundwater withdrawal

Groundwater flow

ent across a range of climate change scenarios. Total river runoff, glacier melt, and seasonality of flow are projected to increase until the 2050s and then decrease, with some exceptions and large uncertainties. Uncertain future water availability, including glacier and snow melt, hinders policymakers from developing adequate water resource plans that include bilateral cooperation for irrigation, hydropower generation, industrial use, and water-induced hazard mitigation in HK countries. OUTLOOK: We underline the major research

gaps that, if filled, can reduce large uncertainties in glaciohydrological modeling. These research gaps include accurate representations of glacier volumes, precipitation distribution, permafrost, sublimation, and impacts of debris cover, black carbon, dust, and glacier dynamics. Comprehensive field observation–based and remote sensing–based methods and models are needed to fill the knowledge gaps and reduce uncertainties in runoff projections. As a first step (Tier 1), we recommend the development of monitoring networks that measure hydrology and meteorology across the full range of elevations in targeted basins that span a variety of climate regimes. These networks should include fully automatic weather stations situated on selected glaciers, and would provide detailed information for calibration and testing of process-based models, downscaling approaches, and hydrological models. We also recommend developing comparison projects for glacier area and volume, glacier dynamics, permafrost thaw, and snow and ice sublimation studies. Satellite and airborne remote sensing offers a potential for rapid advances in Tier-1 objectives, and includes platforms such as InSAR, GRACE, Icesat-2, high-resolution digital elevation models, and geophysical surveys. Tier-2 recommendations include the development of catchment-wide glaciohydrological models for the selected reference catchments identified in Tier 1 and a strengthened processbased understanding of high-elevation hydrology and meteorology to reduce the uncertainties in projections of runoff components, runoff volumes, and shifts in runoff seasonality. Lastly, the development of collaborative research groups and data-sharing policies among HK countries, combined with integrated and interdisciplinary studies of water access and water vulnerabilities, are strongly recommended to understand the impacts of changing river runoff on economic, agricultural, and human productivity.

▪

The list of author affiliations is available in the full article online. *Corresponding author. Email: [email protected] Cite this article as M. F. Azam et al., Science 373, eabf3668 (2021). DOI: 10.1126/science.abf3668

Simplified hydrological cycle. Representation of major Earth surface system processes, each of which carries a research gap in the glaciohydrology of the Himalaya-Karakoram region. IWM, Indian winter monsoon; ISM, Indian summer monsoon.

READ THE FULL ARTICLE AT https://doi.org/10.1126/science.abf3668

SCIENCE sciencemag.org

20 AUGUST 2021 • VOL 373 ISSUE 6557

869

RES EARCH

REVIEW

◥

GLACIERS

Glaciohydrology of the Himalaya-Karakoram Mohd. Farooq Azam1*, Jeffrey S. Kargel2, Joseph M. Shea3, Santosh Nepal4, Umesh K. Haritashya5, Smriti Srivastava1, Fabien Maussion6, Nuzhat Qazi7, Pierre Chevallier8, A. P. Dimri9, Anil V. Kulkarni10, J. Graham Cogley , Isamohan Bahuguna11 Understanding the response of Himalayan-Karakoram (HK) rivers to climate change is crucial for ~1 billion people who partly depend on these water resources. Policy-makers tasked with sustainable water resources management require an assessment of the rivers’ current status and potential future changes. We show that glacier and snow melt are important components of HK rivers, with greater hydrological importance for the Indus basin than for the Ganges and Brahmaputra basins. Total river runoff, glacier melt, and seasonality of flow are projected to increase until the 2050s, with some exceptions and large uncertainties. Critical knowledge gaps severely affect modeled contributions of different runoff components, future runoff volumes, and seasonality. Therefore, comprehensive field observation–based and remote sensing–based methods and models are needed.

T

he Himalayan-Karakoram (HK) region in South Asia, often called the water tower of Asia or the Third Pole (1, 2), is one of the most heavily glacierized mountain regions on Earth (3). Dynamic storage of water in the HK cryosphere regulates the runoff into regional river systems (primarily the Indus, Ganges, and Brahmaputra basins; Fig. 1, A and B) by releasing water generally in April through October—primarily snow melt in April through June and glacier melt in June through October. Basin runoff is the sum of snow melt, glacier melt, rainfall runoff, and base flow from groundwater recharge (4), minus evaporation and evapotranspiration. In addition, storage and release from glacial lakes and reservoirs can modulate basin flow. Permafrost, which covers a substantial fraction of the HK region (5), controls surface water– groundwater interactions and is important for water and ice storage but is currently not represented in hydrological models. The presence of snow, glacier, and permafrost within each of these basins changes the spatiotemporal river runoff characteristics from what would occur with only rainfall runoff and base flow. Thus, cryospheric change has direct impacts

1

Discipline of Civil Engineering, Indian Institute of Technology Indore, Simrol 453552, India. 2Planetary Science Institute, Tucson, AZ, USA. 3Geography Program, University of Northern British Columbia, Prince George, BC V2N 4Z9, Canada. 4International Centre for Integrated Mountain Development, Kathmandu, Nepal. 5Department of Geology and Environmental Geosciences, University of Dayton, Dayton, OH 45469, USA. 6Department of Atmospheric and Cryospheric Sciences, University of Innsbruck, Innsbruck, Austria. 7National Institute of Hydrology, Roorkee, India. 8 Hydrosciences Laboratory (CNRS, IRD, University of Montpellier), CC 57, 34090 Montpellier, France. 9School of Environmental Sciences, Jawaharlal Nehru University, New Delhi, India. 10Indian Institute of Science, Divecha Center for Climate Change, Bangalore, India. 11Space Application Centre, Ahmadabad, India. *Corresponding author. Email: [email protected] Deceased.

Azam et al., Science 373, eabf3668 (2021)

on runoff (2, 4), and reductions in glacier volumes contribute to sea level rise (6, 7). HK river basins cover an area of 2.75 million km2 and have the largest irrigated area (~577,000 km2), five megacities (Delhi, Dhaka, Karachi, Kolkata, and Lahore) with a total population exceeding 94 million, and the world’s largest installed hydropower capacity (26,432 MW) (Table 1). Meltwater plays an important but variable role in the major rivers and especially their upstream tributaries near the mountain sources (4, 8), with less relative contribution away from the headwaters. Net annual glacier wastage accounts for 0.66% of total HK runoff (Table 1). To evaluate the societal importance of glacier and snow melt runoff, we compute the population impact index (PIX) (8). These basins are home to more than 1 billion people, and the estimated PIX is 34 million (3.34% of total population) (Table 1). We note that PIX is one useful quantifiable metric, but it may not fully characterize vulnerability or complex social, infrastructural, cultural, and economic factors that impinge on water access. Meltwater is most important in highaltitude, proximal reaches of HK watersheds. Meltwater volumes are greatest during the spring, summer, and early autumn when snow and ice melt typically occurs. In summer, when monsoon rains increase river flows, the percentage of glacier and snow meltwater may actually decline. However, in the eastern and central Himalaya, glacier melt may have a greater relative contribution in the post-monsoon months (September through November), which is a dry period but one of active melting. Groundwater also plays an important role when rainfall is scarce downstream (9, 10). Freshwater demand in the region is increasing as a result of the growing population, increased food production, and rising affluence (9, 10).

20 August 2021

HK freshwater supports agriculture-based economies of all the surrounding nations (2, 9–13). However, the water reserves of the HK region are vulnerable because of climate change influences on monsoons, glaciers, and snowpack (2, 8, 14). These vulnerabilities will affect the quantity and seasonality of water discharge, water governance at the national and sub-basin watershed levels, water management mechanisms, and international cooperative and geopolitical relationships regarding transboundary rivers (15, 16). The propagation of an error in the IPCC Fourth Assessment Report (17) published in 2007 highlighted the lack of understanding of glaciers and hydrology of the HK region (18). This error drew the attention of the global scientific community to fill the knowledge gaps (3, 4, 10, 11, 19–27). The IPCC corrected the error, but key issues remain about the reliability of past and future glaciohydrological modeling and confidence in runoff projections. This review is motivated by these issues and by the urgency of regional supply and infrastructure issues that affect water access, water quality, and water vulnerabilities. Our objective is to examine the present and future status of HK glacier water resources and discuss knowledge gaps and future research directions to increase the confidence level of future runoff patterns and trends. HK climate

As a result of their geographical location and high elevations, HK mountain ranges present an orographic barrier to westerly and southerly flows and play a key role in the HK’s climate (25, 28). The climate of the HK is influenced by two synoptic circulation systems: Indian summer monsoon [ISM (28)] and Indian winter monsoon [IWM (25)]. IWM is characterized by Western disturbances (WDs) embedded within large-scale westerly flow (25). During the summer season, the Tibetan Plateau and northern Indian regions heat up and create a large expanse of low atmospheric pressure. The spatial distribution of precipitation associated with ISM and IWM remains a point of discussion. However, in both cases the main precipitation forming mechanism is orographic (25, 28). ISM originates from the Indian Ocean and propagates northward, producing the bulk of annual precipitation over the Indian subcontinent, particularly over the southern rim of the Himalaya. Conversely, during winter, IWM originates over the Mediterranean Sea or West Atlantic Ocean and brings precipitation, mainly in solid form, to the high-altitude mountains. WDs embedded in IWM have a life cycle of 2 to 4 days, but multiple WDs can move through the region in succession over ~10 days (25, 29). The HK mountain ranges act as a barrier to these monsoon systems and lead to orographic 1 of 17

RES EARCH | R E V I E W

Fig. 1. Area of study and glacial coverage. (A) Geographic location of studies discussed in Table 2. Melt is the sum of snow and glacier melts; Other is the sum of rain and base flow (Table 2). The HK region is divided into four subregions: Karakoram (KK), western Himalaya (WH), central Himalaya (CH), and eastern Himalaya (EH) (21). The basin boundaries (IB, GB, and BB) are from Bajracharya and Shrestha (241). (B) Glacier hypsometries for each subregion are extracted from GAMDAM glacier inventory (52), and volumes are extracted from Farinotti et al. (242). Inset in (A) shows the three major glaciohydrological regimes of the HK (55).

precipitation on (i) the southern slopes of the western Himalaya having south-north gradient; and (ii) the eastern and central Himalaya with an east-west gradient (30). In the central Himalaya, the precipitation distribution is characterized by two zones of rainfall maxima elongated along the Himalaya (31, 32). The magnitude of precipitation in the central Himalaya is highly variable, with annual precipitation of ~4000 mm on windward southern slopes and 90% of genes (Fig. 1F). Long noncoding RNAs also exhibited noise enhancement, and weakly expressed genes showed a slightly greater change in Fano (fig. S2). These results of a global increase in transcript noise with little change in mean abundance are in stark contrast to the effects of transcriptional activators or cellular stressors that alter noise in a stereotypic manner together with changes in mean number of transcripts (20, 30). To account for technical noise and to quantify the statistical significance of changes in noise and mean, we used an established Bayesian hierarchical model (BASiCS) (31) to create probabilistic, gene-specific estimates of both mean expression and cell-to-cell transcript variability. Of the 4971 genes analyzed, 945 (~20%) were classified as highly variable, whereas 113 (~2%) showed a significant change in mean expression (Fig. 1, G and H). Bulk RNA-seq measurements of mean abundances confirmed the scRNA-seq findings (fig. S3). Thus, analyses from two methods, Seurat and BASiCS, showed that IdU induced a significant increase in transcript variability (expression noise) but comparatively little change in mean expression. To determine whether certain characteristics could explain a gene’s potential for noise enhancement, we examined gene length, promoter AT content, gene body AT content, number of exons, TATA box inclusion, and strand orientation. None of these characteristics exhibited 1 of 10

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

A

B

Low Noise RNA # RNA #

Fluctuation Amplification

State A

Slc35a2

IdU DMSO

Stx7

Log10(CV2)

1.00 Hadhb

Farsa 0.10

0.01 0.1

0.3

1.0

Desai et al., Science 373, eabc6506 (2021)

D

0.30

Stx7

0.3

0.1 1.0

Hadhb Farsa

0.05

Slc35a2 0.03

3

2

1

0 0

1.00

2% of Genes 0.75

0.50

0.25

1.5

3.0

or G2/M) (33), which showed that Sox2, Oct4, Nanog, and Klf4 were highly variable in each cell cycle phase, indicating that their variability is not cell cycle dependent (fig. S6). Moreover, pseudotime analysis showed no bifurcations, indicating that transcriptional variability was not caused by a differentiation-induced mixture of cell types (fig. S7). Extrinsic variability may also arise from the coordinated propagation of noise through generegulatory networks (34) and can be measured by gene-to-gene correlation matrices (35, 36). If the increase in global transcript noise is extrinsic, then the expression correlation between network partners would increase or remain unchanged. Analysis of gene-to-gene correlation matrices showed that ~80% of genegene pairs lost correlation strength after IdU treatment (Fig. 2A and fig. S8), indicating that enhanced expression noise is uncorrelated and

20 August 2021

Posterior Probability

H

0.0

1

2

3

4

DMSO Fano Factor

Log10(DMSO Mean)

−1.5

2.00

0.60

4

3.5

G

0.00 −3.0

0.20

Log10(Normalized Mean)

IdU Fano Factor

1.0

0.3

IdU DMSO

0.10

F

3.5

0.1

Activator

0.15

0.01 0.07

2.0

Log10(Normalized Mean)

Log10(IdU Mean)

E

?

Mean (RNA/cell)

State B

Log10(Variance)

4.00

Activator

Log(mRNAs/Cell)

Suppressor/ Stressor

Time

C

?

# of Cells

Noise (CV2)

State A

RNA

ON

Suppressor/ Stressor

CV2

Log2(Fold Change in Mean)

predictive power (fig. S4). However, genes susceptible to high noise enhancement were preferentially located within the interior of topologically associated domains (TADs), suggesting that gene topology influences susceptibility to noise enhancement (fig. S4). Ontology analysis of highly variable genes showed enrichment of housekeeping pathways, along with pluripotency maintenance factors, particularly Sox2, Oct4, Nanog, and Klf4 (Fig. 1H and fig. S5). Because these pluripotency maintenance factors are key influencers of cell fate specification, we focused on the molecular mechanisms driving their amplified transcript noise. We investigated whether the enhanced variability arose from extrinsic factors, which included cell cycle phase and cell type identity (32). Cells within the scRNA-seq dataset were computationally assigned a cycle stage (G1, S,

OFF

State B

Posterior Probability

Fig. 1. Genome-wide amplification of cell-to-cell mRNA variability (noise) independently of mean. (A) (Left) Monte Carlo simulations of the two-state random-telegraph model of transcription showing low-noise and higher-noise trajectories with matched mean expression levels. Coefficient of variation (s2/m2, CV2) quantifies magnitude of fluctuations. (Right) Predicted facilitation of state transitions through dithering. (B) (Top) Schematic of two-state random-telegraph model of transcription. (Bottom) Schematic of mean versus CV2 for mRNA abundance. Solid gray line indicates Poisson, inverse scaling of CV2 as a function of mean. The question mark symbolizes unknown noise control mechanisms that amplify fluctuations independently of mean. Histograms depict expected shift in mRNA copy number distributions. (C to F) scRNA-seq of mESCs treated with DMSO (black) or 10 mM IdU (red) for 24 hours. A total of 812 and 744 transcriptomes from DMSO and IdU treatments, respectively, were analyzed. (C) Mean expression versus CV2 and (D) mean versus variance. Four examples of housekeeping genes (purple) demonstrate how IdU increases expression fluctuations with minimal change in mean (white arrows). (E and F) Mean expression (E) and Fano factor (F) (s2/m) of genes in DMSO versus IdU treatments. (G and H) BASiCS analysis of scRNA-seq data. (G) Fold change in mean versus certainty (posterior probability) that a gene is up- or down-regulated. With IdU treatment, 113 genes (red) were classified as differentially expressed (more than a twofold change in mean with >85% probability). (H) Fold change in overdispersion versus certainty (posterior probability) that gene is highly or lowly variable. A total of 945 genes (red) were classified as highly variable (>1.5-fold change in overdispersion with >85% probability).

Klf4 Nanog Sox2 1.00

20% of Genes

Oct4

0.75

0.50

0.25

0.00 −3.0

−1.5

0.0

1.5

3.0

Log2(Fold Change in Over-Dispersion)

not consistent with an extrinsic noise source. Exclusion of these extrinsic noise sources indicates that IdU must amplify intrinsic noise arising from stochastic fluctuations in either transcript production (promoter toggling) or degradation. To test whether a change in promoter toggling could account for IdU-enhanced noise, we used single-molecule RNA FISH (smRNA-FISH) to count both nascent and mature transcripts of Nanog, a master regulator of pluripotency. Transcripts were counted in an mESC line in which both endogenous alleles of Nanog are fused to enhanced green fluorescence protein (eGFP) at the C terminus. This fusion does not alter mRNA or protein half-life nor does it impair differentiation potential (37). smRNAFISH probes to eGFP (the 3′ end of the transcript) were used to count mature transcripts, and probes to the first intron of Nanog (the 5′ 2 of 10

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

Pluripotency Network

Pluripotency Network

961 Genes

Nanog

IdU

C

DMSO

0.015

mature Nanog mRNAs

(n=498)

Transcriptional Center (TC)

(n=567) /µ = 6.8 ±1.2 2 /µ = 11.4 ±0.9 2

*

Density

IdU 0.010

0.005

0.000

80

20

140

200

260

D 0.6

0.4

DMSO

Side Scatter

0 40000

80000

120000

160000

200000

Forward Scatter

DMSO IdU Fold Change in Fano Factor: 3.13 ± 0.14

1.5

3000

7000

1000

500

Intrinsic 0

14000 20000

0

2

/µ 2 = 0.071 ±0.008

500

Fig. 2. Amplification of mRNA noise is not caused by extrinsic sources, results from shorter but more intense transcriptional bursts, and propagates to protein levels. (A) Pearson correlations of expression for gene pairs in scRNA-seq dataset. Hierarchical clustering reveals networks of genes (highlighted in black rectangles) sharing similar correlation patterns. Dashed rectangle highlights network enriched with pluripotency factors such as Nanog. (B to D) Results of smRNAFISH used to count nascent and mature Nanog mRNA in Nanog-GFP mESCs treated with DMSO or 10 mM IdU for 24 hours in 2i/LIF medium. Data are from four biological replicates. (B) (Left) Representative micrograph (maximum intensity projection) of mESCs with DAPI staining in which Nanog transcripts are labeled with probe set for eGFP. Bright foci correspond to TCs as verified by intron probe set. Scale bar, 5 mm. (Right) Distributions of mature Nanog transcripts per cell. Dashed lines represent mean. Averaged Fano factors over all four replicates are reported (±SD), *P = 0.0011, two-tailed, unpaired Student’s t test. (C) Fraction of possible TCs that are active as detected by overlap of signal in exon and intron probe

end of the transcript) were used to identify active transcriptional centers (TCs) and explicitly measure the number of mRNAs actively transcribed at the start of the gene. To minimize extrinsic noise, downstream analyses were limited to cells of similar size (fig. S9A). Consistent with scRNA-seq, smRNA-FISH showed a large increase in cell-to-cell variability of mature Nanog transcript abundance Desai et al., Science 373, eabc6506 (2021)

1000

1340

Time (mins)

Nanog-GFP (a.u.)

30

1000

500

Intrinsic 0

0

2

/µ 2 = 0.154 ±0.011

500

1000

1340

Time (mins)

channels. Each cell is assumed to have two possible TCs. Data represent mean ± SD, **P = 6.9 × 10−5, two-tailed, unpaired Student’s t test. (D) Distributions of nascent Nanog mRNA per TC. Average number of nascent mRNAs over all four replicates are reported, **P = 1.0 × 10−4, two-tailed, unpaired Student’s t test. (E) Representative flow cytometry distribution of Nanog-GFP expression in mESCs treated with DMSO or 10 mM IdU for 24 hours in 2i/LIF medium. Dashed lines represent mean. Fold change in Fano factor (±SD) obtained from three biological replicates. Inset: Representative flow cytometry dot plot showing conservative gating on forward and side scatter to filter extrinsic noise arising from cell size heterogeneity. (F) Time-lapse imaging of Nanog-GFP mESCs treated with either DMSO (n = 1513) or 10 mM IdU (n = 1414) in 2i/LIF medium. Trajectories from two replicates of each condition are pooled, with solid and dashed lines representing mean and SD of trajectories, respectively. Distributions of Nanog-GFP represent expression at the final time point. Intrinsic CV2 of each detrended trajectory was calculated, with the average (±SD) of all trajectories reported.

(about a twofold increase in Fano) with little change in mean abundance (Fig. 2B). Quantification of nascent Nanog transcript abundances using either intron or exon probes showed that fewer cells had active TCs in the presence of IdU (Fig. 2C), whereas the number of nascent (i.e., unspliced) and mature (i.e., spliced) mRNAs at each TC was increased (Fig. 2D and fig. S9, B and C). Fitting of the two-

20 August 2021

20

IdU

2000 1000

10

0

Nascent Nanog mRNA at TC

1500

1500

3000

3.0

0.0 1500

IdU

Nanog−GFP (a.u.)

Density

4.5

0.03 0.00

0.0

µ = 8.3 ±0.3 µ = 12.5 ±0.6

0.06

DMSO

F

Cell Size Filtering 4000

Nanog−GFP (a.u.)

E

0.09

**

0.2

Mature Nanog mRNA/cell

DMSO IdU

0.12

Density

Single Molecule RNA FISH

Fraction of Active TC

B

Gene-Gene Correlation 0.4 0.2 0 -0.2 -0.4

**

961 Genes

DMSO

961 Genes

Nanog

961 Genes

A

state random-telegraph model to smRNA-FISH data revealed that increased variability resulted from shortened burst duration (increased KOFF) and amplified transcription rate (higher KTX) (fig. S9D and table S2). To directly visualize the effect of IdU on burst duration, we performed live-cell imaging of transcription using p21-MS2 reporter cells (38). IdU generated shorter transcriptional bursts (increased KOFF), 3 of 10

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

whereas the total mRNA output remained unchanged (fig. S10), further validating the reciprocal changes in burst kinetics seen with smRNA-FISH data. These results validate previous predictions (24, 29) that enhanced noise could arise from reciprocal changes in transcriptional burst duration (1/KOFF) and intensity (KTX). Although longer polymerase dwell times or slowed polymerase elongation could be alternate hypotheses for the increase in nascent RNA detected by smFISH, these hypotheses are inconsistent with the simultaneous shortening of burst duration and maintenance of transcript output observed with MS2 imaging. The slowed and/or stalled polymerase hypothesis is also not consistent with the equivalent increase in both intron and exon probe intensities at TCs (fig. S9E). These data instead suggest that the IdU-mediated increase in TC intensities results from amplified transcription rates (KTX). To determine whether enhanced transcript variability transmitted to protein abundances, we performed flow-cytometric analysis of Nanog-GFP reporter protein. In IdU-treated cells, the Nanog protein Fano factor increased by about threefold, with little change in mean, indicating that mRNA variability from altered promoter toggling indeed resulted in changes to protein noise (Fig. 2E). The increase in protein noise showed no dependency on cell cycle (fig. S11, C and D) despite G1-to-S cell cycle progression being slightly slowed by IdU treatment (fig. S11, A and B). Consistent with the extrinsic noise analysis above, there was no evidence of aneuploidy after IdU treatment (fig. S11A), precluding the possibility that increased noise results from a subpopulation of cells with nonphysiologic gene copy numbers. Inhibition of transcription with actinomycin D completely abrogated IdU enhancement of Nanog-GFP noise (fig. S12, A and B), indicating that IdU minimally perturbs posttranscriptional sources of gene expression variability (e.g., mRNA degradation, mRNA translation, and protein degradation). When cultured in 2i/LIF medium, Nanog protein expression was unimodal and high, but when cultured in serum/LIF medium, mESCs exhibited bimodal expression with both a highNanog state and a low-Nanog state that predisposed a cell toward differentiation (fig. S12C) (39). Given that IdU-induced amplification of Nanog variability arose from an intrinsic source of noise (i.e., changes in transcriptional bursting), we next tested a previous theoretical prediction that increased transcriptional noise would drive greater excursions from the highNanog state into the low-Nanog state (40). IdU treatment did indeed generate greater excursions into the low-Nanog state for mESCs cultured in serum/LIF (fig. S12C), verifying theoretical predictions. This result demonstrated Desai et al., Science 373, eabc6506 (2021)

how modulation of transcriptional bursting can drive Nanog state switching. To verify that enhanced noise is not a population-level phenomenon brought on by differential responses to IdU in distinct cellular subpopulations (i.e., to verify “ergodicity” and that individual cells exhibit increased fluctuations), we used live-cell time-lapse imaging to quantify both the magnitude (intrinsic-CV2) and frequency content (1/halfautocorrelation time) of Nanog-GFP fluctuations. Single-cell tracking of individual cells showed that IdU induced a twofold increase in the magnitude (intrinsic-CV2) of fluctuations (Fig. 2F and fig. S13A), and autocorrelation analysis of detrended trajectories showed a broadening of the frequency distribution to higher spectra, indicating reduced stability (increased lability) of protein expression levels (fig. S13B). These higher-frequency fluctuations are consistent with amplification of a nongenetic, intrinsic source of noise (41, 42), because genetic sources of cellular heterogeneity, such as promoter mutations, would lead to longer retention of protein states (decreased lability) (43). In silico sorting of cells on the basis of starting Nanog expression verified that noise enhancement was not dependent on the initial state of expression (fig. S14). Fluctuations in promoter toggling therefore drive individual cells to dynamically explore a larger state space of Nanog expression. To further validate that IdU perturbs an intrinsic source of noise, we used an mESC line in which the two endogenous alleles of Sox2 are tagged with distinct fluorophores, which enables quantification of the intrinsic and extrinsic components of noise. Treatment with IdU increased Sox2 intrinsic noise greater than twofold across all expression levels (fig. S15), further validating that IdU enhances intrinsic noise. To pinpoint the molecular mechanism, 14 nucleoside analogs (table S3) were screened for noise enhancement effects. 5′-bromo-2′deoxyuridine (BrdU), 5-hydroxymethylcytosine (hmC), and 5-hydroxymethyluridine (hmU) also increased the Nanog Fano factor to varying degrees (Fig. 3A). hmU and hmC are naturally produced by the Ten-eleven translocation (Tet) family of enzymes during oxidation of thymine and methylated cytosine, respectively (44, 45). Given that these base modifications are removed through base excision repair (BER), we surmised that their incorporation and removal from genomic DNA may cause noise enhancement (Fig. 3B) (46, 47). To test this, we suppressed the expression of 25 genes involved in nucleoside metabolism and DNA repair using CRISPRi [three guide RNAs (gRNAs)/gene; table S4] and quantified how this affected the noise enhancement of IdU. We identified two genes, AP endonuclease 1 (Apex1) and thymidine ki-

20 August 2021

nase 1 (Tk1), the depletion of which abrogated noise enhancement (Fig. 3C). Gene depletion was confirmed by reverse transcription quantitative polymerase chain reaction (fig. S16). Tk1 adds a requisite gamma-phosphate group to diphosphate nucleotides before genomic incorporation (Fig. 3B) (48). Our results indicated that phosphorylation of IdU by Tk1 and subsequent incorporation of phosphorylated IdU into the genome may be necessary for noise enhancement. As validation of this, a combination of 10 mM IdU with increasing amounts of thymidine, a competitive substrate of Tk1, returned Nanog noise to baseline levels (fig. S17A), indicating that noise enhancement is dose dependent on IdU incorporation. scRNA-seq analysis also showed that cells in the S/G2 cell cycle phases, when levels of IdU incorporation are highest, displayed increased levels of transcriptional noise enhancement (fig. S6). The reduction in Nanog noise with the addition of exogenous thymidine indicates that IdU-induced noise amplification is not a generic effect of nucleotide imbalances (i.e., excess pyrimidine bases) within the cell. Apex1 (also known as Ref-1 or Ape1) has a pivotal role in the BER pathway because it incises DNA at apurinic and apyrimidinic (AP) sites through its endonuclease domain, allowing for subsequent removal of the sugar backbone and patching of the gap (49, 50). Chromatin immunoprecipitation confirmed that IdU treatment increased Apex1 recruitment to the Nanog promoter (fig. S17B). To determine whether alternate activators of BER also enhanced noise, we subjected cells to oxidative stress [with hydrogen peroxide (H2O2)] and alkylation damage stress [with methyl methanesulfonate (MMS)]. Similar to IdU, hydrogen peroxide, and methyl methanesulfonate also enhanced gene expression noise without altering the mean level of expression (fig. S18, A and B). By contrast, cells subjected to ultraviolet (UV) radiation (an activator of nucleotide excision repair, which shuts off global transcription) exhibited decreases in both mean and noise (Fano factor) (fig. S18, C and D), markedly differing from BER-mediated noise enhancement. These results further demonstrate the specific ability of BER to modulate gene expression noise. Because BER is initiated by a family of DNA glycosylases that recognize and excise modified bases to create AP sites, we investigated whether perturbation of glycosylases affects gene expression noise. Individual depletion of either uracil-DNA glycosylase or thymine-DNA glycosylase failed to ablate IdU noise enhancement (fig. S19A), presumably because of the overlapping and compensatory action of glycosylase family members in base removal (51, 52). However, overexpression of either uracil-DNA glycosylase or methylpurine-DNA glycosylase alone increased Nanog expression 4 of 10

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

A

B *

*

2.0

Thymidine Kinase 1

P

*

1.5

5’

DNA Pol

1.0

Base Excision Repair Modified base

3’

3’

5’ DNA glycosylase

5’

3’

5’

AP endonuclease (Apex1) 3’

3’

5’

Id C

C

dU

yd

5’

h

Parp + DNA Pol

5’

3’

3’

5’

Non-target gRNA + DMSO Non-target gRNA + IdU

4.5

Non-targeting gRNAs Targeting gRNAs

4

3.0

1.5

3

0.0

1000

100

10000

Apex1 gRNA + DMSO Apex1 gRNA + IdU

4.5

2

Density

Fold Change in Nanog Fano w/ IdU

H

h

C

h

C 5F

M m U

rC 5B

C m H

D

H m U H yd U

0.0

3’

AP site

Repaired site

0.5

M SO B rd U

Fano Fold Change

2.5

Diminished Noise Enhancement of Nanog

Apex1 gRNAs

3.0

1.5

0.0 1000

100

1

10000

Tk1 gRNA + DMSO Tk1 gRNA + IdU

4

Tk1 gRNAs

3 2 1

0 0.0

0.5

1.0

1.5

2.0

0 100

Fold Change in Nanog Fano w/ DMSO Fano Fold Change

Density

4

DMSO IdU + CRT

3

2

1

0 300

E

1000

Fold Change in Noise

3.5

*

4

3

2

1

DMSO

CRT

IdU

2.5 2.0 1.5 1.0 0.5 0.0

IdU scram gRNA

+ +

Apex1 gRNA (endogenous kd) mOrange Ctl (empty vector)

+

IdU + CRT

0.012

*

3.0

5

F

*

4.0

10000

0

10000

Density

D

1000

Nanog-GFP (a.u.)

* *

0.009

+

+

+

+

+

0.003

+

0.000 250

DMSO (n = 742) IdU (n = 875)

* * IdU + CRT (n = 686)

0.006

+

* *

+

Apex1 CI overexpression Apex1 WT overexpression

+

500

750

1000

1200

Mean Psoralen Instensity per Cell (a.u.)

Density

DMSO IdU 200

1000

3000

Fig. 3. Noise amplification independent of mean is caused by Apex1mediated DNA repair. (A) Screening of 14 additional nucleoside analogs. Nanog-GFP mESCs grown in 2i/LIF medium were supplemented with a 10 mM concentration of nucleoside analog for 24 hours. Fano factor for Nanog protein expression was normalized to DMSO. Data represent mean (±SD) of biological replicates, *P < 0.01, Kruskal-Wallis test followed by TukeyÕs multiple comparisons test. (B) Schematic of nucleoside analog incorporation into genomic DNA and removal through the BER pathway. (C) (Left) CRISPRi screening for genetic dependencies of IdU noise enhancement. Nanog-GFP mESCs stably expressing dCas9-KRAB-p2A-mCherry were transduced with a single gRNA expression vector with blue fluorescent protein reporter. A total of 75 gRNAs (25 genes, with three gRNAs/gene) were tested, in addition to three nontargeting control gRNAs. Two days after transduction, each gRNADesai et al., Science 373, eabc6506 (2021)

20 August 2021

expressing population of mESCs was treated with DMSO or 10 mM IdU for 24 hours in 2i/LIF medium. The Nanog Fano factor for DMSO and IdU treatment of each gRNA population was normalized to the Nanog Fano factor for the nontargeting gRNA + DMSO population. Each point represents a gRNA. Dashed horizontal line represents average noise enhancement of Nanog from IdU in the background of nontargeting gRNA expression (black squares). Depletion of Apex1 and Tk1 diminishes noise enhancement of Nanog from IdU. (Right) Representative flow cytometry distributions of Nanog expression for mESCs expressing nontargeting (top right), Apex1 (middle right), or Tk1 (bottom right) gRNAs and treated with DMSO or 10 mM IdU. (D) Combination of IdU and small-molecule inhibitor of the Apex1 endonuclease domain (CRT0044876). (Left) Representative flow cytometry distributions of Nanog expression for mESCs treated with DMSO or 10 mM IdU + 100 mM CRT0044876. (Right) mESCs were treated 5 of 10

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

with DMSO, 100 mM CRT0044876, 10 mM IdU, or 10 mM IdU + 100 mM CRT0044876 for 24 hours in 2i/LIF medium. The Nanog Fano factor for each treatment was normalized to the DMSO control. Data represent mean ± SD of three biological replicates, *P = 0.0028, two-tailed, unpaired Student’s t test. (E) Overexpression of wild-type (WT) or catalytically inactive (CI) Apex1 with simultaneous CRISPRi depletion of endogenous Apex1. (Top) Fold change in Nanog Fano factor for respective treatment condition described in the rectangular grid. An mOrange empty vector was used as a transduction control. The Nanog Fano factor for each treatment was normalized to mOrange

noise in the absence of IdU (fig. S19, B and C). These data suggest that noise-without-mean amplification is an inherent property of BER that occurs for endogenous modifications of both purine and pyrimidine bases. To further confirm that Apex1 is necessary for noise enhancement, we attempted to inactivate (knock out) Apex1 in mESCs but this was lethal. As an alternative, we used a smallmolecule catalytic inhibitor (CRT0044876) specific for the Apex1 endonuclease domain (53). Contrary to the effect of Apex1 depletion, the combination of CRT0044876 with IdU synergistically increased Nanog expression noise without significantly changing the mean (Fig. 3D). The contrasting effects of Apex1 depletion and catalytic inhibition implied that physical binding rather than enzymatic activity of the protein modulates transcriptional bursting. Apex1 is known to induce helical distortions and local supercoiling to identify mismatched bases (54, 55), and catalytically inactive Apex1 mutants bind DNA with higher affinity (56). This suggests that CRT0044876 may lengthen Apex1’s residence time on DNA, thus amplifying topological reformations. We verified that inhibition of Apex1 endonuclease activity with CRT0044876 did not inhibit IdU-mediated enhancement of Apex1 recruitment to the Nanog promoter (fig. S17B). To further test whether Apex1 binding rather than enzymatic activity was responsible for noise enhancement, we expressed a catalytically inactive mutant of Apex1 (56) in cells that had endogenous Apex1 depleted. We found that catalytically inactive Apex1 partially rescued IdU-mediated noise enhancement (Fig. 3E). These data, together with evidence that supercoiling sets mechanical bounds on transcriptional bursting (57, 58), drove us to investigate whether Apex1 recruitment affects supercoiling. To measure supercoiling levels, we used a psoralen–cross-linking assay in which mESCs are incubated with biotinylated-trimethylpsoralen (bTMP), which preferentially intercalates into negatively supercoiled DNA (59). To eliminate DNA replication as a contributor of supercoiling, aphidicolin is added to inhibit DNA polymerases before bTMP incubation (60). IdU treatment significantly increased genomic supercoiling, as demonstrated by an approximately twofold increase in bTMP intercalation (Fig. 3F). Desai et al., Science 373, eabc6506 (2021)

control cells treated with DMSO. Data represent mean ± SD of three biological replicates, *P < 0.005, two-tailed, unpaired Student’s t test. (Bottom) Representative flow cytometry distributions of Nanog expression for each treatment condition. (F) Single-cell quantification of negative supercoiling levels using the psoralen– cross-linking assay. mESCs were treated with DMSO, 10 mM IdU, or 10 mM IdU + 100 mM CRT0044876 for 24 hours in 2i/LIF medium. Distributions for nuclear intensities of bTMP staining are shown. Data are pooled from two biological replicates of each treatment, **P < 0.0001, Kruskal-Wallis test followed by Tukey’s multiple comparisons test.

The combination of IdU and CRT0044876 further increased intercalation, suggesting that supercoiling is correlated with noise enhancement through increased Apex1-DNA interactions (Fig. 3F). IdU treatment followed by a short incubation with bleomycin (which decreases supercoiling through double-stranded breaks) reduced bTMP intercalation below the dimethyl sulfoxide (DMSO) control level, indicating that IdU alone in uncoiled DNA does not increase intercalation (fig. S20). If DNA topology influences transcriptional bursting, then additional modifiers of supercoiling should also affect Nanog noise. Topoisomerase 1 and 2a (Top1 and Top2a, respectively) relax coiled DNA through the introduction of single- and double-stranded breaks, respectively. Depletion of Top1 and Top2a by CRISPRi increased Nanog protein variability (fig. S21A). Inhibition of topoisomerase activity with the small-molecule inhibitors topotecan and etoposide recapitulated these effects (fig. S21B). Furthermore, overexpression of Top1 partially ablated IdUmediated noise enhancement (fig. S21C). However, depletion of chromatin-remodeling proteins known to interact with BER machinery failed to modulate IdU-mediated noise enhancement, suggesting that histone repositioning, a reported modulator of transcriptional noise, is not a major contributor to BER-mediated noise enhancement (fig. S21D). Together with psoralen–cross-linking data, these results indicate that Apex1-induced supercoiling is a significant driver of noisewithout-mean amplification. To understand the mechanism by which Apex1 might increase transcriptional noise without altering mean expression, we developed a series of minimalist computational models to account for the experimental data (supplementary text 2). Monte Carlo simulations of each model using smRNA-FISH data for parameterization (table S2) indicated that a model incorporating transcription-coupled base excision best accounts for noise-withoutmean amplification (Fig. 4, A and B; figs. S22 to S24; and supplementary text 5). In this model, Apex1 binding triggers entry to a negatively supercoiled transcriptionally nonproductive state (ON*), whereas unbinding of Apex1 allows mRNA production to resume with an amplified transcription rate that is proportion-

20 August 2021

al to time spent in the nonproductive state (fig. S25 and supplementary text 6); that is, the longer the residence time in the nonproductive state, the stronger the enhancement of transcription rate once repair is complete (a feedforward loop). This feedforward effect may originate from the increased negative supercoiling during repair, which can facilitate a proportionate increase in upstream binding of transcriptional machinery (61–68). Consistent with this hypothesis, the model accurately predicted noise enhancement mediated by topoisomerase inhibition (fig. S24C and supplementary text 6). The ability to render a gene transcriptionally nonproductive while also stimulating recruitment of transcriptional resources points to a homeostatic mechanism: The BER pathway maintains gene expression homeostasis (i.e., mean) by amplifying transcriptional fluctuations through reciprocal modulation of burst intensity and duration (Fig. 4B). We call this model, and the associated phenomenon, “discordant transcription through repair” (“DiThR,” pronounced “dither”) because of the large discordance in pre-repair versus post-repair transcriptional activity. Sensitivity analysis of the DiThR model revealed that orthogonal modulation of Nanog mean and noise is possible within a large portion of the parameter space (fig. S26, A and B, and supplementary text 7). As validation, we tested the effect of 96 concentration combinations (table S6) of IdU and CRT0044876 to perturb the rates of Apex1 binding and unbinding, respectively. The experimental results confirmed model predictions, showing that Nanog noise could be tuned independently of the mean (Fig. 4C). Testing of BrdU and hmU further validated that there are parameter regimes where noise can be regulated independently of mean (fig. S27). The hmU data in particular showed that the BER pathway can amplify noise while maintaining mean expression when removing a naturally occurring base modification. The different concentration thresholds for noise enhancement among these nucleoside analogs may reflect known differences in incorporation rates (69). Sensitivity analysis indicated that in genes for which KOFF >> KON (i.e., lowly expressed genes), IdU treatment would increase mean abundance (fig. S26E). This prediction was verified experimentally with bulk RNA-seq measurements 6 of 10

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

A OFF

RNA

ON Apex 1

Apex 1

O

Presence of IdU

ON*

B

OFF state

ON* state (undergoing repair)

ON state (Repaired)

Apex1

IdU

+

– – –

+

–

–

Increased supercoiling

+

Increased transcription rate

1

on

S K

on

B

K

of f

0

Experimental Validation of Model

Noise Enhancer 0%

Nanog (# mRNAs)

OSK factors + DMSO

% Oct4-GFP(+) Fold Change (IdU / DMSO)

Day 10

16%

33%

31% 1000

10000 1

10

100

1000

10000

1

10

Nanog-GFP (a.u.)

Fig. 4. Mechanistic model for DiThR and the phenotypic consequence of DiThR on potentiation of cellular reprogramming. (A) Detailed schematic of model 5 (the DiThR model) (see fig. S22 for a schematic of models 1 to 4), which uses transcription-coupled base excision. In the presence of IdU (bottom panel), Apex1 binding occurs when the gene is transcriptionally permissive (i.e., in the ON state). Binding induces negative supercoiling, which lengthens the time that a gene is transcriptionally nonproductive (i.e., in the ON* state) while also facilitating recruitment of transcriptional resources. Upon repair completion, a higher transcriptional rate that is proportional to time spent in ON* state is reached. Mean expression is maintained with larger transcriptional fluctuations. (B) Effective behavior [m, Fano factor (FF), KOFF, fraction of time active (von), burst size (BS), KON] of the DiThR model is compared with experimentally derived values of each Desai et al., Science 373, eabc6506 (2021)

30000

OSK factors + 1uM IdU (48h)

OSK factors + 4uM IdU (48h)

G

OSK factors + 4uM IdU (48h)

1.0

100

10000

Time

1.5

10

3000

28%

OSK factors + 1uM IdU (48h)

OSK factors + DMSO

0.0 1

1000

E

14%

0.5

0 300

Activator + Noise Enhnacer

Time

F

1

Nanog-GFP (a.u.)

< 1%

Activator

2

20 August 2021

100

1000

10000

6

*** ** ***

4

2

0 C trl sc rs hR N A A pe x1 sh R N A A pe 1 x1 sh R N A A pe 2 x1 sh R N A 3

Nanog (# mRNAs)

Untreated Pluripotency

0.6

IdU

DMSO 5.5uM IdU/ 0.6uM CRT 17uM IdU/ 1.9uM CRT 17uM IdU/ 150uM CRT

3

Density

Apex1 inh

Apex1 inh

1.8

Fold change Fano

0.75

IdU

3.0

Fold change Mean

1.25

1

Density

+

10

C

D

ON state

Pol II

–

Experiment Model

FF

Parameter Value

Effective Behavior 100

OFF state

Absence of IdU

Model 5 DiThR Model

parameter (red dots) obtained from smRNA-FISH data. Absolute percentage error was calculated as described in supplementary text 5.2.2. Model 5 (the DiThR model) best matches experimental data. (C) Testing of 96 concentration combinations of IdU and CRT0044876 (Apex1 inh) to validate tunability of Nanog variability. IdU and CRT0044876 were used to increase binding and decrease unbinding of Apex1, respectively. Data represent the average of two biological replicates. Left and center panels are 96-well heatmaps displaying the fold change in Nanog mean and Fano factor for each drug combination compared with DMSO (top left well). An insufficient number of cells for extrinsic noise filtering (> KON are precluded because they will exhibit increased mean). Additionally, genes most susceptible to transcriptional noise enhancement tend to lie far from TAD boundaries. Because TAD boundaries largely overlap with supercoiling domain boundaries (76), and

20 August 2021

transcription-induced supercoiling may directly contribute to the formation of TADs (77), we reasoned that TAD boundaries may maintain a constantly high level of supercoiling, thus offering a narrow dynamic range for noise enhancement. Propagation of transcriptional variability to the protein level likely depends on protein half-lives and thus may not occur for a large swath of proteins. The proteins monitored in this study have either naturally short half-lives (Nanog) or PEST tags (e.g., d2GFP), which minimizes the buffering of transcriptional bursts conferred by longer protein half-lives (78). The ability to independently control the mean and variance of gene expression may indicate that cells can amplify transcriptional noise for fate exploration and specification. Methods summary

Quantification of cell-to-cell variability in gene expression was performed using the following techniques: (i) single-cell RNA-seq, (ii) singlemolecule RNA FISH, (iii) live-cell imaging of RNA transcription with p21-MS2 reporter cell line, (iv) flow cytometry, and (v) live-cell imaging of Nanog-GFP protein expression. For scRNA-seq, mESCs treated with DMSO or 10 mM IdU for 24 hours were prepared for sequencing using 10× genomics specifications. Quality control, normalization, and variability analysis of scRNA-seq data were performed using two packages: Seurat and BASiCS. For smRNAFISH, probes for the first exon, first intron, and 3′ GFP fusion of the Nanog transcript were developed using the designer tool from Stellaris. Nanog-GFP mESCs treated with either DMSO or 10 mM IdU were stained with Nanog mRNA probes and imaged on a Zeiss spinning-disk microscope. RNA spot counting and transcriptional center analysis were performed with FISH-quant. For live-cell imaging of transcription, p21-MS2 reporter U2OS cells were pretreated with Nutlin-3 and either DMSO or 10 mM IdU for 48 hours before imaging. Imaging was performed on a wide-field Olympus microscope for 118 min with 1-min intervals between frames. The cumulative transcription occurring in each cell was calculated based on the normalized transcriptional activity over 118 minutes of imaging. Flow cytometry was used to quantify single-cell variability in protein expression. To filter out gene expression variability arising from cell size heterogeneity, the smallest possible forward and side scatter 8 of 10

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

region containing at least 3000 cells was used to isolate cells of similar size and shape for all analyses. To quantify single-cell fluctuations in Nanog protein expression over time, live-cell time lapse microscopy of Nanog-GFP mESCs was performed using a Zeiss spinning-disk microscope with imaging commencing immediately after addition of either DMSO or 10 mM IdU. Cell segmentation, tracking, and GFP quantification were performed using CellProfiler. Detrended fluorescence trajectories were used for noise autocorrelation and noise magnitude calculations. CRISPRi screening for genetic controllers of transcriptional noise was performed in an arrayed fashion with Nanog-GFP mESCs stably expressing dCas9-Krab::mCherry. A total of 25 genes (three gRNAs/gene) were depleted through individual transduction of cells with gRNA lentiviral constructs harboring a blue fluorescent protein reporter. Forty-eight hours after infection, each population of cells expressing a unique gRNA was treated with either DMSO or 10 mM IdU for 24 hours followed by flow cytometric analysis of NanogGFP expression. Activation of the BER pathway was performed by chemical treatment of mESCs (H2O2 and MMS) and overexpression of DNA glycosylases (Mpg and Ung). Nanog-GFP mESCs were treated with H2O2 and MMS for 1 hour and 24 hours, respectively, before flow cytometric analysis. For overexpression of DNA glycosylases, Nanog-GFP mESCs were transduced with lentiviral constructs harboring doxycycline-inducible cassettes for either Mpg or Ung with an mCherry reporter. Flowcytometric analysis of Nanog-GFP expression was performed on transduced cells after 24 hours of doxycycline induction. Measurement of negative supercoiling in mESCs was performed with bTMP staining. Before bTMP staining, mESCs were treated with 1 mM aphidicolin for 2 hours to remove the confounding effect of DNA replication on genomic supercoiling. After bTMP staining, UV cross-linking was performed for 15 minutes using 365 nm light. Imaging was performed on a Zeiss spinningdisk microscope. Nuclear segmentation using 4′,6-diamidino-2-phenylindole (DAPI) signal and single-cell quantification of bTMP staining intensity were performed using CellProfiler. To elucidate the effect of Apex1 recruitment on transcriptional bursting, five computational models of increasing complexity were constructed based on the random-telegraph model of transcription. For each model, an associated stochastic reaction scheme was numerically solved using Gillespie’s stochastic simulation algorithm to identify the model that best recapitulated experimental data. The effective kinetic rates of Nanog transcription in the control (DMSO) condition from smRNA-FISH data were used as the starting point for all Desai et al., Science 373, eabc6506 (2021)

simulations and were considered constant for all models. For each model, krepair was the single degree of freedom. Identification of the model that best fits experimental data was based on maximum likelihood estimation of krepair for each model, followed by minimization of Akaike information criterion. To assess whether increased fluctuations in gene expression can promote cell fate transitions, three cellular reprogramming assays were tested: (i) Nanog-GFP secondary MEFs harboring stably integrated doxycycline-inducible cassettes for Oct4, Sox2, and Klf4; (ii) Oct4GFP primary MEFs transduced with lentiviral vectors encoding Oct4, Sox2, Klf4, and c-Myc; and (iii) Oct4-GFP MEFs harboring stably integrated doxycycline-inducible cassettes for Oct4, Sox2, Klf4, and c-Myc. For each system, IdU-mediated noise enhancement was implemented for the first 48 hours of reprogramming. Induced pluripotent stem cell formation was assessed using alkaline phosphatase staining and flow-cytometric measurement of fluorescent reporter levels. A detailed account of all methods used in this study is provided in the supplementary materials. RE FERENCES AND NOTES

1. L. Boltzmann, Weitere Studien uber das Wärmegleichgewicht unter Gasmolekulen, Sitzungsber. Kais. Akad. Wiss. Wien Math. Naturwiss 66, 275–370 (1872). 2. S. Arrhenius, Über die Reaktionsgeschwindigkeit bei der Inversion von Rohrzucker durch Säuren. Z. Phys. Chem. 4, 226–248 (1889). doi: 10.1515/zpch-1889-0416 3. L. Roberts, Picture coding using pseudo-random noise. IEEE Trans. Inf. Theory 8, 145–154 (1962). doi: 10.1109/ TIT.1962.1057702 4. P. Fatt, B. Katz, Some observations on biological noise. Nature 166, 597–598 (1950). doi: 10.1038/166597a0; pmid: 14780165 5. A. A. Priplata, J. B. Niemi, J. D. Harry, L. A. Lipsitz, J. J. Collins, Vibrating insoles and balance control in elderly people. Lancet 362, 1123–1124 (2003). doi: 10.1016/S0140-6736(03) 14470-4; pmid: 14550702 6. D. Cohen, Optimizing reproduction in a randomly varying environment. J. Theor. Biol. 12, 119–129 (1966). doi: 10.1016/ 0022-5193(66)90188-3; pmid: 6015423 7. A. Arkin, J. Ross, H. H. McAdams, Stochastic kinetic analysis of developmental pathway bifurcation in phage lambda-infected Escherichia coli cells. Genetics 149, 1633–1648 (1998). doi: 10.1093/genetics/149.4.1633; pmid: 9691025 8. J. L. Spudich, D. E. Koshland Jr., Non-genetic individuality: Chance in the single cell. Nature 262, 467–471 (1976). doi: 10.1038/262467a0; pmid: 958399 9. P. B. Gupta et al., Stochastic state transitions give rise to phenotypic equilibrium in populations of cancer cells. Cell 146, 633–644 (2011). doi: 10.1016/j.cell.2011.07.026; pmid: 21854987 10. L. S. Weinberger, J. C. Burnett, J. E. Toettcher, A. P. Arkin, D. V. Schaffer, Stochastic gene expression in a lentiviral positive-feedback loop: HIV-1 Tat fluctuations drive phenotypic diversity. Cell 122, 169–182 (2005). doi: 10.1016/ j.cell.2005.06.006; pmid: 16051143 11. A. Raj, A. van Oudenaarden, Nature, nurture, or chance: Stochastic gene expression and its consequences. Cell 135, 216–226 (2008). doi: 10.1016/j.cell.2008.09.050; pmid: 18957198 12. J. R. Chubb, T. Trcek, S. M. Shenoy, R. H. Singer, Transcriptional pulsing of a developmental gene. Curr. Biol. 16, 1018–1025 (2006). doi: 10.1016/j.cub.2006.03.092; pmid: 16713960 13. I. Golding, J. Paulsson, S. M. Zawilski, E. C. Cox, Real-time kinetics of gene activity in individual bacteria. Cell 123, 1025–1036 (2005). doi: 10.1016/j.cell.2005.09.031; pmid: 16360033

20 August 2021

14. T. B. Kepler, T. C. Elston, Stochasticity in transcriptional regulation: Origins, consequences, and mathematical representations. Biophys. J. 81, 3116–3136 (2001). doi: 10.1016/S0006-3495(01)75949-8; pmid: 11720979 15. A. Raj, C. S. Peskin, D. Tranchina, D. Y. Vargas, S. Tyagi, Stochastic mRNA synthesis in mammalian cells. PLOS Biol. 4, e309 (2006). doi: 10.1371/journal.pbio.0040309; pmid: 17048983 16. W. J. Blake, M. KAErn, C. R. Cantor, J. J. Collins, Noise in eukaryotic gene expression. Nature 422, 633–637 (2003). doi: 10.1038/nature01546; pmid: 12687005 17. J. Peccoud, B. Ycart, Markovian modeling of gene-product synthesis. Theor. Popul. Biol. 48, 222–234 (1995). doi: 10.1006/tpbi.1995.1027 18. J. Rodriguez, D. R. Larson, Transcription in living cells: Molecular mechanisms of bursting. Annu. Rev. Biochem. 89, 189–212 (2020). doi: 10.1146/annurev-biochem-011520105250; pmid: 32208766 19. D. Nicolas, B. Zoller, D. M. Suter, F. Naef, Modulation of transcriptional burst frequency by histone acetylation. Proc. Natl. Acad. Sci. U.S.A. 115, 7153–7158 (2018). doi: 10.1073/pnas.1722330115; pmid: 29915087 20. A. Bar-Even et al., Noise in protein expression scales with natural protein abundance. Nat. Genet. 38, 636–643 (2006). doi: 10.1038/ng1807; pmid: 16715097 21. R. D. Dar et al., Transcriptional bursting explains the noiseversus-mean relationship in mRNA and protein levels. PLOS ONE 11, e0158298 (2016). doi: 10.1371/journal. pone.0158298; pmid: 27467384 22. J. R. S. Newman et al., Single-cell proteomic analysis of S. cerevisiae reveals the architecture of biological noise. Nature 441, 840–846 (2006). doi: 10.1038/nature04785; pmid: 16699522 23. U. Alon, An Introduction to Systems Biology: Design Principles of Biological Circuits. (Chapman & Hall/CRC, 2007). 24. R. D. Dar, N. N. Hosmane, M. R. Arkin, R. F. Siliciano, L. S. Weinberger, Screening for noise in gene expression identifies drug synergies. Science 344, 1392–1396 (2014). doi: 10.1126/science.1250220; pmid: 24903562 25. M. M. K. Hansen et al., A post-transcriptional feedback mechanism for noise suppression and fate stabilization. Cell 173, 1609–1621.e15 (2018). doi: 10.1016/j.cell.2018.04.005; pmid: 29754821 26. Y. Li et al., Noise-driven cellular heterogeneity in circadian periodicity. Proc. Natl. Acad. Sci. U.S.A. 117, 10350–10356 (2020). doi: 10.1073/pnas.1922388117; pmid: 32358201 27. A. Butler, P. Hoffman, P. Smibert, E. Papalexi, R. Satija, Integrating single-cell transcriptomic data across different conditions, technologies, and species. Nat. Biotechnol. 36, 411–420 (2018). doi: 10.1038/nbt.4096; pmid: 29608179 28. M. M. K. Hansen, R. V. Desai, M. L. Simpson, L. S. Weinberger, Cytoplasmic amplification of transcriptional noise generates substantial cell-to-cell variability. Cell Syst. 7, 384–397.e6 (2018). doi: 10.1016/j.cels.2018.08.002; pmid: 30243562 29. B. Munsky, G. Neuert, A. van Oudenaarden, Using gene expression noise to understand gene regulation. Science 336, 183–187 (2012). doi: 10.1126/science.1216379; pmid: 22499939 30. D. M. Suter et al., Mammalian genes are transcribed with widely different bursting kinetics. Science 332, 472–474 (2011). doi: 10.1126/science.1198817; pmid: 21415320 31. N. Eling, A. C. Richard, S. Richardson, J. C. Marioni, C. A. Vallejos, Correcting the mean-variance dependency for differential variability testing using single-cell RNA sequencing data. Cell Syst. 7, 284–294.e12 (2018). doi: 10.1016/j. cels.2018.06.011; pmid: 30172840 32. M. Kaern, T. C. Elston, W. J. Blake, J. J. Collins, Stochasticity in gene expression: From theories to phenotypes. Nat. Rev. Genet. 6, 451–464 (2005). doi: 10.1038/nrg1615; pmid: 15883588 33. A. Scialdone et al., Computational assignment of cell-cycle stage from single-cell transcriptome data. Methods 85, 54–61 (2015). doi: 10.1016/j.ymeth.2015.06.021; pmid: 26142758 34. J. M. Pedraza, A. van Oudenaarden, Noise propagation in gene networks. Science 307, 1965–1969 (2005). doi: 10.1126/ science.1109090; pmid: 15790857 35. R. Bargaje et al., Cell population structure prior to bifurcation predicts efficiency of directed differentiation in human induced pluripotent cells. Proc. Natl. Acad. Sci. U.S.A. 114, 2271–2276 (2017). doi: 10.1073/pnas.1621412114; pmid: 28167799 36. M. Mojtahedi et al., Cell fate decision as high-dimensional critical state transition. PLOS Biol. 14, e2000640 (2016). doi: 10.1371/journal.pbio.2000640; pmid: 28027308

9 of 10

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

37. C. Sokolik et al., Transcription factor competition allows embryonic stem cells to distinguish authentic signals from noise. Cell Syst. 1, 117–129 (2015). doi: 10.1016/ j.cels.2015.08.001; pmid: 26405695 38. L. A. Carvajal et al., Dual inhibition of MDMX and MDM2 as a therapeutic strategy in leukemia. Sci. Transl. Med. 10, eaao3003 (2018). doi: 10.1126/scitranslmed.aao3003; pmid: 29643228 39. E. Abranches et al., Stochastic NANOG fluctuations allow mouse embryonic stem cells to explore pluripotency. Development 141, 2770–2779 (2014). doi: 10.1242/dev.108910; pmid: 25005472 40. T. Kalmar et al., Regulated fluctuations in nanog expression mediate cell fate decisions in embryonic stem cells. PLOS Biol. 7, e1000149 (2009). doi: 10.1371/journal.pbio.1000149; pmid: 19582141 41. D. W. Austin et al., Gene network shaping of inherent noise spectra. Nature 439, 608–611 (2006). doi: 10.1038/ nature04194; pmid: 16452980 42. N. Rosenfeld, J. W. Young, U. Alon, P. S. Swain, M. B. Elowitz, Gene regulation at the single-cell level. Science 307, 1962–1965 (2005). doi: 10.1126/science.1106914; pmid: 15790856 43. A. Sigal et al., Variability and memory of protein levels in human cells. Nature 444, 643–646 (2006). doi: 10.1038/ nature05316; pmid: 17122776 44. S. Ito et al., Role of Tet proteins in 5mC to 5hmC conversion, ES-cell self-renewal and inner cell mass specification. Nature 466, 1129–1133 (2010). doi: 10.1038/nature09303; pmid: 20639862 45. T. Pfaffeneder et al., Tet oxidizes thymine to 5hydroxymethyluracil in mouse embryonic stem cell DNA. Nat. Chem. Biol. 10, 574–581 (2014). doi: 10.1038/ nchembio.1532; pmid: 24838012 46. Y. F. He et al., Tet-mediated formation of 5-carboxylcytosine and its excision by TDG in mammalian DNA. Science 333, 1303–1307 (2011). doi: 10.1126/science.1210944; pmid: 21817016 47. L. Shen et al., Genome-wide analysis reveals TET- and TDG-dependent 5-methylcytosine oxidation dynamics. Cell 153, 692–706 (2013). doi: 10.1016/j.cell.2013.04.002; pmid: 23602152 48. E. S. J. Arnér, S. Eriksson, Mammalian deoxyribonucleoside kinases. Pharmacol. Ther. 67, 155–186 (1995). doi: 10.1016/ 0163-7258(95)00015-9; pmid: 7494863 49. B. Demple, T. Herman, D. S. Chen, Cloning and expression of APE, the cDNA encoding the major human apurinic endonuclease: Definition of a family of DNA repair enzymes. Proc. Natl. Acad. Sci. U.S.A. 88, 11450–11454 (1991). doi: 10.1073/pnas.88.24.11450; pmid: 1722334 50. T. Lindahl, R. D. Wood, Quality control by DNA repair. Science 286, 1897–1905 (1999). doi: 10.1126/science.286.5446.1897; pmid: 10583946 51. U. Müller, C. Bauer, M. Siegl, A. Rottach, H. Leonhardt, TET-mediated oxidation of methylcytosine causes TDG or NEIL glycosylase dependent gene reactivation. Nucleic Acids Res. 42, 8592–8604 (2014). doi: 10.1093/nar/gku552; pmid: 24948610 52. M. L. Hegde, T. K. Hazra, S. Mitra, Early steps in the DNA base excision/single-strand interruption repair pathway in mammalian cells. Cell Res. 18, 27–47 (2008). doi: 10.1038/ cr.2008.8; pmid: 18166975 53. S. Madhusudan et al., Isolation of a small molecule inhibitor of DNA base excision repair. Nucleic Acids Res. 33, 4711–4724 (2005). doi: 10.1093/nar/gki781; pmid: 16113242 54. O. A. Kladova et al., The role of the N-terminal domain of human apurinic/apyrimidinic endonuclease 1, APE1, in DNA glycosylase stimulation. DNA Repair (Amst.) 64, 10–25 (2018). doi: 10.1016/j.dnarep.2018.02.001; pmid: 29475157 55. C. D. Mol, T. Izumi, S. Mitra, J. A. Tainer, DNA-bound structures and mutants reveal abasic DNA binding by APE1 and DNA repair coordination [corrected]. Nature 403, 451–456 (2000). doi: 10.1038/35000249; pmid: 10667800 56. D. R. McNeill, D. M. Wilson 3rd, A dominant-negative form of the major human abasic endonuclease enhances cellular sensitivity to laboratory and clinical DNA-damaging agents.

Desai et al., Science 373, eabc6506 (2021)

57.

58.

59.

60.

61.

62.

63.

64.

65.

66.

67.

68.

69.

70.

71.

72.

73.

74.

75.

20 August 2021

Mol. Cancer Res. 5, 61–70 (2007). doi: 10.1158/1541-7786. MCR-06-0329; pmid: 17259346 S. Chong, C. Chen, H. Ge, X. S. Xie, Mechanism of transcriptional bursting in bacteria. Cell 158, 314–326 (2014). doi: 10.1016/j.cell.2014.05.038; pmid: 25036631 S. A. Sevier, D. A. Kessler, H. Levine, Mechanical bounds to transcriptional noise. Proc. Natl. Acad. Sci. U.S.A. 113, 13983–13988 (2016). doi: 10.1073/pnas.1612651113; pmid: 27911801 S. Corless, N. Gilbert, Investigating DNA supercoiling in eukaryotic genomes. Brief. Funct. Genomics 16, 379–389 (2017). doi: 10.1093/bfgp/elx007; pmid: 28444308 K. N. Babos et al., Mitigating antagonism between transcription and proliferation allows near-deterministic cellular reprogramming. Cell Stem Cell 25, 486–500.e9 (2019). doi: 10.1016/j.stem.2019.08.005; pmid: 31523028 P. Guptasarma, Cooperative relaxation of supercoils and periodic transcriptional initiation within polymerase batteries. BioEssays 18, 325–332 (1996). doi: 10.1002/bies.950180411; pmid: 8967901 S. Kim, B. Beltran, I. Irnov, C. Jacobs-Wagner, Long-distance cooperative and antagonistic RNA polymerase dynamics via DNA supercoiling. Cell 179, 106–119.e16 (2019). doi: 10.1016/ j.cell.2019.08.033; pmid: 31539491 F. Kouzine, S. Sanford, Z. Elisha-Feil, D. Levens, The functional response of upstream DNA to dynamic supercoiling in vivo. Nat. Struct. Mol. Biol. 15, 146–154 (2008). doi: 10.1038/ nsmb.1372; pmid: 18193062 J. Liu et al., The FUSE/FBP/FIR/TFIIH system is a molecular machine programming a pulse of c-myc expression. EMBO J. 25, 2119–2130 (2006). doi: 10.1038/ sj.emboj.7601101; pmid: 16628215 M. Mizutani, T. Ohta, H. Watanabe, H. Handa, S. Hirose, Negative supercoiling of DNA facilitates an interaction between transcription factor IID and the fibroin gene promoter. Proc. Natl. Acad. Sci. U.S.A. 88, 718–722 (1991). doi: 10.1073/ pnas.88.3.718; pmid: 1992462 S. S. Teves, S. Henikoff, Transcription-generated torsional stress destabilizes nucleosomes. Nat. Struct. Mol. Biol. 21, 88–94 (2014). doi: 10.1038/nsmb.2723; pmid: 24317489 M. Bazlekowa-Karaban et al., Mechanism of stimulation of DNA binding of the transcription factors by human apurinic/ apyrimidinic endonuclease 1, APE1. DNA Repair 82, 102698 (2019). doi: 10.1016/j.dnarep.2019.102698; pmid: 31518879 J. F. Breit, K. Ault-Ziel, A.-B. Al-Mehdi, M. N. Gillespie, Nuclear protein-induced bending and flexing of the hypoxic response element of the rat vascular endothelial growth factor promoter. FASEB J. 22, 19–29 (2008). doi: 10.1096/ fj.07-8102com; pmid: 17766324 B. Leuner, E. R. Glasper, E. Gould, Thymidine analog methods for studies of adult neurogenesis are not equally sensitive. J. Comp. Neurol. 517, 123–133 (2009). doi: 10.1002/cne.22107; pmid: 19731267 M. Delmans, M. Hemberg, Discrete distributional differential expression (D3E)—A tool for gene expression analysis of single-cell RNA-seq data. BMC Bioinformatics 17, 110 (2016). doi: 10.1186/s12859-016-0944-6; pmid: 26927822 W. J. Blake et al., Phenotypic consequences of promotermediated transcriptional noise. Mol. Cell 24, 853–865 (2006). doi: 10.1016/j.molcel.2006.11.003; pmid: 17189188 H. H. Chang, M. Hemberg, M. Barahona, D. E. Ingber, S. Huang, Transcriptome-wide noise controls lineage choice in mammalian progenitor cells. Nature 453, 544–547 (2008). doi: 10.1038/nature06965; pmid: 18497826 G. M. Süel, R. P. Kulkarni, J. Dworkin, J. Garcia-Ojalvo, M. B. Elowitz, Tunability and noise dependence in differentiation dynamics. Science 315, 1716–1719 (2007). doi: 10.1126/science.1137455; pmid: 17379809 J. Pruszak, W. Ludwig, A. Blak, K. Alavian, O. Isacson, CD15, CD24, and CD29 define a surface biomarker code for neural lineage differentiation of stem cells. Stem Cells 27, 2928–2940 (2009). doi: 10.1002/stem.211; pmid: 19725119 S. Semrau et al., Dynamics of lineage commitment revealed by single-cell transcriptomics of differentiating embryonic stem cells. Nat. Commun. 8, 1096 (2017). doi: 10.1038/s41467-01701076-4; pmid: 29061959

76. C. Naughton et al., Transcription forms and remodels supercoiling domains unfolding large-scale chromatin structures. Nat. Struct. Mol. Biol. 20, 387–395 (2013). doi: 10.1038/ nsmb.2509; pmid: 23416946 77. D. Racko, F. Benedetti, J. Dorier, A. Stasiak, Transcriptioninduced supercoiling as the driving force of chromatin loop extrusion during formation of TADs in interphase chromosomes. Nucleic Acids Res. 46, 1648–1660 (2018). doi: 10.1093/nar/gkx1123; pmid: 29140466 78. A. Singh, Transient changes in intercellular protein variability identify sources of noise in gene expression. Biophys. J. 107, 2214–2220 (2014). doi: 10.1016/j.bpj.2014.09.017; pmid: 25418106 79. R. Desai, B. Martin, scRNA-seq and modeling code for: A DNArepair pathway can regulate transcriptional noise to promote cell fate transitions, Zenodo (2021); doi: 10.5281/zenodo.4891977 AC KNOWLED GME NTS

We thank M. Simpson, B. Bruneau, J. Weissman, G. Balazsi, and members of the Weinberger laboratory for thoughtful discussions and suggestions; K. Claiborn for editing; G. Maki for graphics support; N. Raman in the Gladstone Institute Flow Cytometry Facility (NIH S10 RR028962, P30 AI027763, DARPA, and the James B. Pendleton Charitable Trust) for technical assistance; the Gladstone Assay Development and Drug Discovery Core for technical assistance with drug screening; K. Thorn and D. Larson in the UCSF Nikon Imaging Center (NIH S10 1S10OD017993-01A1) for technical assistance with imaging; M. Jost and J. Weissman for CRISPRi reagents; and the Gladstone Institute Genomics Core for technical assistance with single-cell RNA-sequencing. The dual-tagged Sox2 mESCs were a kind donation from B. Bruneau and E. Nora. The Oct4-GFP reprogrammable MEFs (harbor stably integrated OKSM factors) were a kind donation from S. Guo. Funding: R.V.D. is supported by an NIH/NICHD F30 fellowship (HD095614-03). R.A.C. acknowledges support from NIH award 1R01GM126045-05. R.H.S. acknowledges support from NIH awards NS083085 and 1R35GM136296. M.M.K.H. acknowledges support from a Dutch Research Council (NWO) ENW-XS award (OCENW.XS3.055). L.S.W. acknowledges support from a Bowes Distinguished Professorship, Alfred P. Sloan Research Fellowship, Pew Scholars in the Biomedical Sciences Program, NIH award R01AI109593, and the NIH Director’s New Innovator Award (OD006677) and Pioneer Award (OD17181) programs. Author contributions: R.V.D., M.T., and L.S.W. conceived and designed the study. R.V.D., B.M., and M.T. analyzed the sequencing data. R.V.D., X.C., C.U., S.D., and L.S.W conceived and designed the cellular reprogramming experiments. X.C., D.W.H., W.L., R.H.S., R.A.C., and L.S.W conceived and designed the MS2 imaging experiments. R.V.D., X.C., S.C., D.W.H., W.L., and C.U. performed the experiments. R.V.D., X.C., B.M., M.T., R.A.C., M.M.K.H., and L.S.W. analyzed data. R.V.D., M.M.K.H., B.M., and L.S.W. constructed and analyzed the mathematical models. R.V.D., M.M.K.H., and L.S.W. wrote the manuscript. Competing interests: The authors declare no competing interests. Data and materials availability: The raw and processed sequencing data reported herein have been deposited onto the Gene Expression Omnibus under accession number GSE176044. Custom code for analysis of scRNA-seq data and mathematical modeling are available on GitHub at https://github.com/weinbergerlab-ucsf/Code_Desai_et_al and are archived on Zenodo (79). Reagents, including plasmids and cell lines, are available from the corresponding author upon request. SUPPLEMENTARY MATERIALS

science.sciencemag.org/content/373/6557/eabc6506/suppl/DC1 Materials and Methods Supplementary Text Figs. S1 to S32 Tables S1 to S8 References (80–97) MDAR Reproducibility Checklist

31 May 2020; accepted 8 July 2021 Published online 22 July 2021 10.1126/science.abc6506

10 of 10

RESEAR CH

RESEARCH ARTICLES

◥

PROTEIN FOLDING

Accurate prediction of protein structures and interactions using a three-track neural network Minkyung Baek1,2, Frank DiMaio1,2, Ivan Anishchenko1,2, Justas Dauparas1,2, Sergey Ovchinnikov3,4, Gyu Rie Lee1,2, Jue Wang1,2, Qian Cong5,6, Lisa N. Kinch7, R. Dustin Schaeffer6, Claudia Millán8, Hahnbeom Park1,2, Carson Adams1,2, Caleb R. Glassman9,10,11, Andy DeGiovanni12, Jose H. Pereira12, Andria V. Rodrigues12, Alberdina A. van Dijk13, Ana C. Ebrecht13, Diederik J. Opperman14, Theo Sagmeister15, Christoph Buhlheller15,16, Tea Pavkov-Keller15,17, Manoj K. Rathinaswamy18, Udit Dalwadi19, Calvin K. Yip19, John E. Burke18, K. Christopher Garcia9,10,11,20, Nick V. Grishin6,7,21, Paul D. Adams12,22, Randy J. Read8, David Baker1,2,23* DeepMind presented notably accurate predictions at the recent 14th Critical Assessment of Structure Prediction (CASP14) conference. We explored network architectures that incorporate related ideas and obtained the best performance with a three-track network in which information at the one-dimensional (1D) sequence level, the 2D distance map level, and the 3D coordinate level is successively transformed and integrated. The three-track network produces structure predictions with accuracies approaching those of DeepMind in CASP14, enables the rapid solution of challenging x-ray crystallography and cryoÐelectron microscopy structure modeling problems, and provides insights into the functions of proteins of currently unknown structure. The network also enables rapid generation of accurate protein-protein complex models from sequence information alone, short-circuiting traditional approaches that require modeling of individual subunits followed by docking. We make the method available to the scientific community to speed biological research.

T

he prediction of protein structure from amino acid sequence information alone has been a long-standing challenge. The biannual Critical Assessment of Structure Prediction (CASP) meetings have demonstrated that deep-learning methods such as AlphaFold (1, 2) and trRosetta (3), which extract information from the large database of known protein structures in the Protein Data Bank (PDB), outperform more traditional approaches that explicitly model the folding process. The outstanding performance of DeepMind’s AlphaFold2 in the recent 14th CASP (CASP14) meeting (https://predictioncenter.org/casp14/ zscores_final.cgi) left the scientific community eager to learn details beyond the overall framework that was presented and raised the question of whether such accuracy could be achieved outside of a world-leading deep-learning company. As described at the CASP14 conference, the AlphaFold2 methodological advances included (i) starting from multiple sequence alignments (MSAs) rather than from more-processed

features such as inverse covariance matrices derived from MSAs, (ii) replacement of twodimensional (2D) convolution with an attention mechanism that better represents interactions between residues distant along the sequence, (iii) use of a two-track network architecture in which information at the 1D sequence level and the 2D distance map level is iteratively transformed and passed back and forth, (iv) use of an SE(3)-equivariant Transformer network to directly refine atomic coordinates (rather than 2D distance maps as in previous approaches) generated from the twotrack network, and (v) end-to-end learning in which all network parameters are optimized by backpropagation from the final generated 3D coordinates through all network layers back to the input sequence. Network architecture development

Intrigued by the DeepMind results, and with the goal of increasing protein structure prediction accuracy for structural biology research

and advancing protein design (4), we explored network architectures that incorporate different combinations of these five properties. In the absence of a published method, we experimented with a wide variety of approaches for passing information between different parts of the networks, as summarized in the methods and table S1. We succeeded in producing a “two-track” network with information flowing in parallel along a 1D sequence alignment track and a 2D distance matrix track with considerably better performance than trRosetta (BAKER-ROSETTASERVER and BAKER in Fig. 1B), the next-best method after AlphaFold2 in CASP14 (https://predictioncenter.org/casp14/ zscores_final.cgi). We reasoned that better performance could be achieved by extending to a third track operating in 3D coordinate space to provide a tighter connection between sequence, residue-residue distances and orientations, and atomic coordinates. We constructed architectures with the two levels of the two-track model augmented with a third parallel structure track operating on 3D backbone coordinates, as depicted in Fig. 1A (see methods and fig. S1 for details). In this architecture, information flows back and forth between the 1D amino acid sequence information, the 2D distance map, and the 3D coordinates, allowing the network to collectively reason about relationships within and between sequences, distances, and coordinates. By contrast, reasoning about 3D atomic coordinates in the two-track AlphaFold2 architecture happens after processing of the 1D and 2D information is complete (although end-to-end training does link parameters to some extent). Because of computer hardware memory limitations, we could not train models on large proteins directly because the three-track models have many millions of parameters; instead, we presented to the network many discontinuous crops of the input sequence consisting of two discontinuous sequence segments spanning a total of 260 residues. To generate final models, we combined and averaged the 1D features and 2D distance and orientation predictions produced for each of the crops and then used two approaches to generate final 3D structures. In the first, the predicted residueresidue distance and orientation distributions are fed into pyRosetta (5) to generate all-atom models. In the second, the averaged 1D and 2D

1

Department of Biochemistry, University of Washington, Seattle, WA 98195, USA. 2Institute for Protein Design, University of Washington, Seattle, WA 98195, USA. 3Faculty of Arts and Sciences, Division of Science, Harvard University, Cambridge, MA 02138, USA. 4John Harvard Distinguished Science Fellowship Program, Harvard University, Cambridge, MA 02138, USA. 5Eugene McDermott Center for Human Growth and Development, University of Texas Southwestern Medical Center, Dallas, TX, USA. 6Department of Biophysics, University of Texas Southwestern Medical Center, Dallas, TX, USA. 7Howard Hughes Medical Institute, University of Texas Southwestern Medical Center, Dallas, TX, USA. 8Department of Haematology, Cambridge Institute for Medical Research, University of Cambridge, Cambridge, UK. 9Program in Immunology, Stanford University School of Medicine, Stanford, CA 94305, USA. 10Department of Molecular and Cellular Physiology, Stanford University School of Medicine, Stanford, CA 94305, USA. 11Department of Structural Biology, Stanford University School of Medicine, Stanford, CA 94305, USA. 12Molecular Biophysics & Integrated Bioimaging Division, Lawrence Berkeley National Laboratory, Berkeley, CA, USA. 13Department of Biochemistry, Focus Area Human Metabolomics, North-West University, 2531 Potchefstroom, South Africa. 14Department of Biotechnology, University of the Free State, 205 Nelson Mandela Drive, Bloemfontein 9300, South Africa. 15Institute of Molecular Biosciences, University of Graz, Humboldtstrasse 50, 8010 Graz, Austria. 16Medical University of Graz, Graz, Austria. 17BioTechMed-Graz, Graz, Austria. 18Department of Biochemistry and Microbiology, University of Victoria, Victoria, BC, Canada. 19Life Sciences Institute, Department of Biochemistry and Molecular Biology, The University of British Columbia, Vancouver, BC, Canada. 20Howard Hughes Medical Institute, Stanford University School of Medicine, Stanford, CA 94305, USA. 21Department of Biochemistry, University of Texas Southwestern Medical Center, Dallas, TX, USA. 22 Department of Bioengineering, University of California, Berkeley, Berkeley, CA 94720, USA. 23Howard Hughes Medical Institute, University of Washington, Seattle, WA 98195, USA. *Corresponding author. Email: [email protected]

SCIENCE sciencemag.org

20 AUGUST 2021 • VOL 373 ISSUE 6557

87 1

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

Fig. 1. Network architecture and performance. (A) RoseTTAFold architecture with 1D, 2D, and 3D attention tracks. Multiple connections between tracks allow the network to simultaneously learn relationships within and between sequences, distances, and coordinates (see methods and fig. S1 for details). (B) Average TM-score of prediction methods on the CASP14 targets. Zhang-server and BAKER-ROSETTASERVER were the top two server groups, whereas AlphaFold2 and BAKER were the top two human groups in CASP14; BAKER-ROSETTASERVER and BAKER predictions were based on trRosetta. Predictions with the two-track model and RoseTTAFold (both end-to-end and pyRosetta version) were completely automated. (C) Blind benchmark results on CAMEO medium and hard targets; model accuracies are TM-score values from the CAMEO website (https://cameo3d.org/). In (B) and (C), the error bars represent a 95% confidence interval.

features are fed into a final SE(3)-equivariant layer (6), and, after end-to-end training from amino acid sequence to 3D coordinates, backbone coordinates are generated directly by the network (see methods). We refer to these networks, which also generate per-residue accuracy predictions, as RoseTTAFold. The first has the advantages of requiring lower-memory graphics processing units (GPUs) at inference time [for proteins with more than 400 residues, 8 gigabytes (GB) rather than 24 GB] and of producing full side-chain models but requires central processing unit (CPU) time for the pyRosetta structure modeling step. The three-track models with attention operating at the 1D, 2D, and 3D levels and information flowing between the three levels were the best models we tested (Fig. 1B), clearly outperforming the top two server groups (Zhangserver and BAKER-ROSETTASERVER), BAKER human group (ranked second among all groups), and our two-track attention models on CASP14 targets. As in the case of AlphaFold2, the correlation between MSA depth and model accuracy is lower for RoseTTAFold than for trRosetta and other methods tested at CASP14 (fig. S2). The performance of the three-track model on the CASP14 targets was still not as good as AlphaFold2 (Fig. 1B). This could reflect hardware limitations that limited the size of the models we could explore, alternative architectures or loss formulations, or more 872

20 AUGUST 2021 • VOL 373 ISSUE 6557

intensive use of the network for inference. DeepMind reported using several GPUs for days to make individual predictions, whereas our predictions are made in a single pass through the network in the same manner that would be used for a server; after sequence and template search (~1.5 hours), the end-to-end version of RoseTTAFold requires ~10 min on an RTX2080 GPU to generate backbone coordinates for proteins with fewer than 400 residues, and the pyRosetta version requires 5 min for network calculations on a single RTX2080 GPU and an hour for all-atom structure generation with 15 CPU cores. Incomplete optimization due to computer memory limitations and neglect of side-chain information likely explain the poorer performance of the endto-end version compared with the pyRosetta version (Fig. 1B; the latter incorporates sidechain information at the all-atom relaxation stage); because SE(3)-equivariant layers are used in the main body of the three-track model, the added gain from the final SE(3) layer is likely less than that in the AlphaFold2 case. We expect the end-to-end approach to ultimately be at least as accurate once the computer hardware limitations are overcome and side chains are incorporated. The improved performance of the three-track models over the two-track model with identical training sets, similar attention-based architectures for the 1D and 2D tracks, and similar

operations in inference (prediction) mode suggests that simultaneously reasoning at the MSA, distance map, and 3D coordinate representations can more effectively extract sequence-structure relationships than reasoning over only MSA and distance map information. The relatively low computational cost makes it straightforward to incorporate the methods in a public server and predict structures for large sets of proteins, for example, all human G protein–coupled receptors (GPCRs), as described below. Blind structure prediction tests are needed to assess any new protein structure prediction method, but CASP is held only once every 2 years. Fortunately, the Continuous Automated Model Evaluation (CAMEO) experiment (7) tests structure prediction servers blindly on protein structures as they are submitted to the PDB. RoseTTAFold has been evaluated since 15 May 2021 on CAMEO; over the 69 medium and hard targets released during this time (15 May 2021 to 19 June 2021), it outperformed all other servers evaluated in the experiment, including Robetta (3), IntFold6TS (8), BestSingleTemplate (9), and SWISSMODEL (10) (Fig. 1C). We experimented with approaches for further improving accuracy by more intensive use of the network during sampling. Because the network can take templates of known structures as input, we experimented with a sciencemag.org SCIENCE

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

further coupling of 3D structural information and 1D sequence information by iteratively feeding the predicted structures back into the network as templates and random subsampling from the MSAs to sample a broader range of models. These approaches generated ensembles that contained higher-accuracy models, but the accuracy predictor was not able to consistently identify models better than those generated by the rapid single-pass method (fig. S3). Nevertheless, we suspect that these approaches can improve model performance, and we are carrying out further investigations along these lines. In developing RoseTTAFold, we found that combining predictions from multiple discontinuous crops generated more-accurate structures than predicting the entire structure at once (fig. S4A). We hypothesized that this arises from selecting the most relevant sequences for each region from the very large number of aligned sequences that are often available (fig. S4B). To enable the network to focus on the most relevant sequence information for each region while keeping access to the full MSA in a more memory-efficient way, we experimented with the Perceiver architecture (11), updating smaller-seed MSAs (up to 100 sequences) with extra sequences (thousands of sequences) through cross-attention (fig. S4C). As of now, RoseTTAFold only uses the top 1000 sequences because of memory limitations; with this addition, all available sequence information can be used (often more than 10,000 sequences). Initial results are promising (fig. S4D), but more training will be required for rigorous comparison.

a bacterial surface layer protein (SLP) (Fig. 2A), and the secreted protein Lrbp from the fungus Phanerochaete chrysosporium (Fig. 2B and fig. S5C). In all four cases, the predicted models had sufficient structural similarity to the true structures that enabled solution of the structures by MR [see methods for details; the perresidue error estimates by DeepAccNet (12) allowed the more accurate parts to be weighted more heavily]. The increased prediction accuracy was critical for success in all cases; models made with trRosetta did not yield MR solutions. To determine why the RoseTTAFold models were successful where PDB structures had previously failed, we compared the models to the crystal structures we obtained. The images in Fig. 2A and fig. S5 show that in each case, the closest homolog of the known structure was a much poorer model than the RoseTTAFold model; in the case of SLP, only a distant model covering part of the N-terminal domain (38% of the sequence) was available in the PDB, whereas no homologs of the C-terminal domain of SLP or any portion of Lrbp could be detected using HHsearch (13). Building atomic models of protein assemblies from cryo-EM maps can be challenging in the absence of homologs with known structures. We used RoseTTAFold to predict the p101 Gbg binding domain (GBD) structure in a hetero-

dimeric PI3Kg complex. The top HHsearch hit has a statistically insignificant E-value of 40 and only covers 14 out of 167 residues. The predicted structure could readily fit into the electron density map despite the low local resolution [Fig. 2C, top; trRosetta failed to predict the correct fold with the same MSA input (fig. S6)]. The Ca-RMSD (root mean square deviation) between the predicted and the final refined structure is 3.0 Å for the core b sheets (Fig. 2C, bottom). Providing insights into biological function

Experimental structure determination can provide considerable insight into biological function and mechanism. We investigated whether structures generated by RoseTTAFold could similarly provide new insights into function. We focused on two sets of proteins: first, GPCRs of currently unknown structure; and second, a set of human proteins implicated in disease. Benchmark tests on GPCR sequences with determined structures showed that RoseTTAFold models for both active and inactive states can be quite accurate even in the absence of close homologs with known structures [and better than those in current GPCR model databases (14, 15); fig. S7] and that the DeepAccNet model quality predictor (12) provides a good measure of actual model accuracy (fig. S7D). We provide

Enabling experimental protein structure determination

With the recent considerable progress in protein structure prediction, a key question is what accurate protein structure models can be used for. We investigated the utility of the RoseTTAFold to facilitate experimental structure determination by x-ray crystallography and cryo–electron microscopy (cryo-EM) and to build models that provide biological insights for key proteins of currently unknown structures. Solution of x-ray structures by molecular replacement (MR) often requires quite accurate models. The much higher accuracy of the RoseTTAFold method compared with currently available methods prompted us to test whether it could help solve previously unsolved challenging MR problems and improve the solution of borderline cases. Four recent crystallographic datasets (summarized, including resolution limits, in table S2), which had eluded solution by MR using models available in the PDB, were reanalyzed using RoseTTAFold models: glycine N-acyltransferase (GLYAT) from Bos taurus (fig. S5A), a bacterial oxidoreductase (fig. S5B), SCIENCE sciencemag.org

Fig. 2. Enabling experimental structure determination with RoseTTAFold. (A and B) Successful molecular replacement with RoseTTAFold models. SLP is shown in (A). The C-terminal domain is shown at the top, with a comparison of final refined structure (gray) to RoseTTAFold model (blue); there are no homologs with known structure. The N-terminal domain is shown at the bottom; the refined structure is in gray, and the RoseTTAFold model is colored by the estimated root mean square (RMS) error (ranging from blue for 0.67 Å to red for 2 Å or greater). Ninety-five Ca atoms of the RoseTTAFold model can be superimposed within 3 Å of Ca atoms in the final structure, yielding a Ca-RMSD of 0.98 Å. By contrast, only 54 Ca atoms of the closest template (4l3a, brown) can be superimposed (with a Ca-RMSD of 1.69 Å). In (B), the refined structure of Lrbp (gray) with the closest RoseTTAFold model (blue) superimposed is shown; residues having an estimated RMS error greater than 1.3 Å are omitted (full model is in fig. S5C). (C) Cryo-EM structure determination of the p101 GBD in a heterodimeric PI3Kg complex using RoseTTAFold. At the top, RoseTTAFold models colored in a rainbow from the N terminus (blue) to the C terminus (red) have a consistent all-b topology with a clear correspondence to the density map. Shown at the bottom is a comparison of the final refined structure to the RoseTTAFold model colored by predicted RMS error ranging from blue for 1.5 Å or less to red for 3 Å or greater. The actual Ca-RMSD between the predicted structure and final refined structure is 3.0 Å over the b sheets. The figure was prepared with ChimeraX (35). 20 AUGUST 2021 • VOL 373 ISSUE 6557

873

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

RoseTTAFold models and accompanying accuracy predictions for closed and open states of all human GPCRs of currently unknown structure. Protein structures can provide insight into how mutations lead to human disease. We identified human proteins without close homologs of known structure that contain multiple diseasecausing mutations or have been the subject of intensive experimental investigation (see methods). We used RoseTTAFold to generate models for 693 domains from such proteins. More than one-third of these models have a predicted local distance difference test (lDDT) >0.8, which corresponded to an average Ca-RMSD of 2.6 Å on CASP14 targets (fig. S8). Here, we focus on three examples that illustrate the different ways in which structure models can provide insight into the function or mechanisms of diseases. Deficiencies in TANGO2 (transport and Golgi organization protein 2) lead to metabolic disorders, and the protein plays an unknown role in Golgi membrane redistribution into the endoplasmic reticulum (16, 17). The RoseTTAFold model of TANGO2 adopts an N-terminal nucleophile aminohydrolase (Ntn) fold (Fig. 3A) with well-aligned active-site residues that are conserved in TANGO2 orthologs (Fig. 3B). Ntn superfamily members with structures similar to the RoseTTAFold model suggest that TANGO2 may function as an enzyme that hydrolyzes a carbon-nitrogen bond in a membrane component (18). Based on the model, known mutations that cause disease (magenta spheres in Fig. 3A) could act by hindering catalysis [Arg26→Lys (R26K), Arg32→Gln (R32Q), and Leu50→Pro (L50P), near the active site] or produce steric clashes [Gly154→Arg (G154R)] (19) in the hydrophobic core. By comparison, a homology model based on very distant (0.8]. Information on residue-residue coevolution between the paired sequences likely contributes to the accuracy of the rigid-body placement because more-accurate complex structures were generated when more sequences were available (fig. S10). The network was trained on monomeric proteins, not complexes, so there may be some training-set bias in the monomer structures, but there is none for the complexes. To illustrate the application of RoseTTAFold to complexes of unknown structure with more than three chains, we used it to generate models of the complete four-chain human interleukin-12 receptor–interleukin-12 (IL-12R– IL-12) complex (Fig. 4C and fig. S11). A previously published cryo-EM map of the IL-12 receptor complex indicated a similar topology to that of the IL-23 receptor; however, the resolution was not sufficient to observe the SCIENCE sciencemag.org

detailed interaction between IL-12Rb2 and IL-12p35 (34). Such an understanding is important for dissecting the specific actions of IL-12 and IL-23 and generating inhibitors that block IL-12 without affecting IL-23 signaling. The RoseTTAFold model fits the experimental cryo-EM density well and identified a shared interaction between Tyr189 in IL-12p35 and Gly115 in IL-12Rb2 analogous to the packing between Trp156 in IL-23p19 with Gly116 in IL-23R. In addition, the model suggests a role for the IL-12Rb2 N-terminal peptide (residues 24 to 31) in IL-12 binding (IL-12Rb2 Asp26 may interact with nearby Lys190 and Lys194 in IL-12p35), which may provide an avenue to specifically target the IL-12Rb2–IL-12 interaction. Conclusions

RoseTTAFold enables solutions of challenging x-ray crystallography and cryo-EM modeling problems, provides insight into protein function in the absence of experimentally determined structures, and rapidly generates accurate models of protein-protein complexes. Further training on protein-protein complex datasets will likely further improve the modeling of the structures of multiprotein assemblies. The approach can be readily coupled with existing small-molecule and protein binder design methodology to improve computational discovery of new protein and small-molecule ligands for targets of interest. The simultaneous processing

of sequence, distance, and coordinate information by the three-track architecture opens the door to new approaches that incorporate constraints and experimental information at all three levels for problems ranging from cryo-EM structure determination to protein design. REFERENCES AND NOTES

1. A. W. Senior et al., Nature 577, 706–710 (2020). 2. J. Jumper et al., in Fourteenth Critical Assessment of Techniques for Protein Structure Prediction: CASP14 Abstract Book (Protein Structure Prediction Center, 2020), pp. 22–24. 3. J. Yang et al., Proc. Natl. Acad. Sci. U.S.A. 117, 1496–1503 (2020). 4. I. Anishchenko, T. M. Chidyausiku, S. Ovchinnikov, S. J. Pellock, D. Baker, bioRxiv 2020.07.22.211482 [Preprint] (2020); https://doi.org/10.1101/2020.07.22.211482. 5. S. Chaudhury, S. Lyskov, J. J. Gray, Bioinformatics 26, 689–691 (2010). 6. F. B. Fuchs, D. E. Worrall, V. Fischer, M. Welling, arXiv:2006.10503 [cs.LG] (2020). 7. J. Haas et al., Proteins 86, 387–398 (2018). 8. L. J. McGuffin et al., Nucleic Acids Res. 47, W408–W413 (2019). 9. J. Haas et al., Proteins 87, 1378–1387 (2019). 10. A. Waterhouse et al., Nucleic Acids Res. 46, W296–W303 (2018). 11. A. Jaegle et al., arXiv:2103.03206 [cs.CV] (2021). 12. N. Hiranuma et al., Nat. Commun. 12, 1340 (2021). 13. M. Steinegger et al., BMC Bioinformatics 20, 473 (2019). 14. A. J. Kooistra et al., Nucleic Acids Res. 49, D335–D343 (2021). 15. B. J. Bender, B. Marlow, J. Meiler, PLOS Comput. Biol. 16, e1007597 (2020). 16. L. S. Kremer et al., Am. J. Hum. Genet. 98, 358–362 (2016). 17. C. Rabouille, V. Kondylis, Genome Biol. 7, 213 (2006). 18. M. P. Milev et al., J. Inherit. Metab. Dis. 44, 426–437 (2021). 19. S. R. Lalani et al., Am. J. Hum. Genet. 98, 347–357 (2016). 20. T. G. Wolfsberg, P. Primakoff, D. G. Myles, J. M. White, J. Cell Biol. 131, 275–278 (1995). 21. T. Klein, R. Bischoff, J. Proteome Res. 10, 17–33 (2011). 22. S. Zhong, R. A. Khalil, Biochem. Pharmacol. 164, 188–204 (2019). 23. P. Orth et al., J. Mol. Biol. 335, 129–137 (2004).

20 AUGUST 2021 • VOL 373 ISSUE 6557

875

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

24. S. Takeda, T. Igarashi, H. Mori, S. Araki, EMBO J. 25, 2388–2396 (2006). 25. D. R. Flower, A. C. North, C. E. Sansom, Biochim. Biophys. Acta 1482, 9–24 (2000). 26. H. E. Van Wart, H. Birkedal-Hansen, Proc. Natl. Acad. Sci. U.S.A. 87, 5578–5582 (1990). 27. M. Levy, A. H. Futerman, IUBMB Life 62, 347–356 (2010). 28. J. L. Kim, B. Mestre, S.-H. Shin, A. H. Futerman, Cell. Signal. 82, 109958 (2021). 29. E. Winter, C. P. Ponting, Trends Biochem. Sci. 27, 381–383 (2002). 30. S. Spassieva et al., J. Biol. Chem. 281, 33931–33938 (2006). 31. N. Vanni et al., Ann. Neurol. 76, 206–212 (2014). 32. Q. Cong, I. Anishchenko, S. Ovchinnikov, D. Baker, Science 365, 185–189 (2019). 33. Y. Zhang, J. Skolnick, Proteins 57, 702–710 (2004). 34. C. R. Glassman et al., Cell 184, 983–999.e24 (2021). 35. E. F. Pettersen et al., Protein Sci. 30, 70–82 (2021). 36. M. Baek et al., RoseTTAFold: The first release of RoseTTAFold. Zenodo (2021); https://zenodo.org/record/5068265. ACKN OW LEDG MEN TS

We thank E. Horvitz, N. Hiranuma, D. Juergens, S. Mansoor, and D. Tischer for helpful discussions; D. E. Kim for web-server construction; and L. Goldschmidt for computing resource management. T.P.-K. thanks B. Nidetzky and M. Monschein from Graz University of Technology for providing protein samples for crystallization. D.J.O. acknowledges assistance with data collection from scientists of Diamond Light Source beamline I04 under proposal mx20303. T.S., C.B., and T.P.-K. acknowledge the ESRF (ID30-3, Grenoble, France) and DESY (P11, PETRAIII, Hamburg, Germany) for provision of synchrotron-radiation facilities and support during data collection. P.D.A., J.H.P., A.D., and A.V.R. acknowledge support from the Joint BioEnergy Institute, which is supported by the US Department of Energy, Office of Science, Office of Biological and Environmental Research under contract no. DE-AC02-05CH11231 between LBNL and the US Department of Energy. Funding: This work was supported by Microsoft (M.B., D.B., and generous gifts of Azure compute time and expertise); Open Philanthropy (D.B. and G.R.L.); E. and W. Schmidt by recommendation of the Schmidt Futures program (F.D. and H.P.); The Washington Research Foundation (M.B., G.R.L., and J.W.); the National Science Foundation Cyberinfrastructure for Biological Research, award no. DBI 1937533 (I.A.); Wellcome Trust grant number 209407/Z/17/Z (R.J.R.); the National Institutes of Health, grant numbers P01GM063210 (P.D.A. and R.J.R.), DP5OD026389 (S.O.), RO1-AI51321 (K.C.G.), and GM127390 (N.V.G.); the Mathers Foundation (K.C.G.); the Canadian Institute of Health Research (CIHR) Project Grant, grant numbers 168998 (J.E.B.) and 168907 (C.K.Y.); the Welch Foundation I-1505 (N.V.G.); the Global Challenges Research Fund (GCRF) through Science & Technology Facilities Council (STFC), grant number ST/R002754/1: Synchrotron Techniques for African Research and Technology (START) (D.J.O., A.A.v.D., and A.C.E.); and the Austrian Science Fund (FWF), projects P29432 and DOC50 (doc.fund Molecular Metabolism) (T.S., C.B., and T.P.-K.). Author contributions: M.B., F.D., and D.B. designed the research; M.B., F.D., I.A., J.D., S.O., and J.W. developed the deep-learning network; G.R.L. and H.P. analyzed GPCR modeling results; Q.C., L.N.K., R.D.S., and N.V.G. analyzed modeling results for proteins related to the human diseases; C.R.G. and K.C.G. analyzed modeling results for the IL-12R–IL-12 complex; P.D.A., R.J.R., C.A., F.D., and C.M. worked on structure determination; A.A.v.D., A.C.E., D.J.O., T.S., C.B., T.P.-K., M.K.R., U.D., C.K.Y., J.E.B., A.D., J.H.P., and A.V.R. provided experimental data; M.B., F.D., G.R.L., Q.C., L.N.K., H.P., C.R.G., P.D.A., R.J.R., and D.B. wrote the manuscript; and all authors discussed the results and commented on the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The GPCR models of unknown structures have been deposited to http://files.ipd.uw. edu/pub/RoseTTAFold/all_human_GPCR_unknown_models.tar.gz and http://files.ipd.uw.edu/pub/RoseTTAFold/GPCR_benchmark_ one_state_unknown_models.tar.gz. The model structures for structurally uncharacterized human proteins have been deposited to http://files.ipd.uw.edu/pub/RoseTTAFold/human_prot.tar.gz. Coordinates for the full PI3K complex structure determined by cryo-EM are available at the PDB with accession code PDB: 7MEZ. Model structures used for molecular replacement are available at http://files.ipd.uw.edu/pub/RoseTTAFold/MR_models.tar.gz. The refined structures for GLYAT, oxidoreductase, SLP, and Lrbp proteins will be deposited in the PDB when final processing is completed. The method is available as a server at https://robetta. bakerlab.org (RoseTTAFold option), and the source code and model parameters are available at https://github.com/ RosettaCommons/RoseTTAFold or Zenodo (36). This research was funded in whole or in part by Wellcome Trust, grant #209407/

876

20 AUGUST 2021 ¥ VOL 373 ISSUE 6557

Z/17/Z, a cOAlition S organization. The author will make the Author Accepted Manuscript (AAM) version available under a CC BY public copyright license.

Tables S1 to S4 References (37Ð82) MDAR Reproducibility Checklist

SUPPLEMENTARY MATERIALS

science.sciencemag.org/content/373/6557/871/suppl/DC1 Materials and Methods Figs. S1 to S17

7 June 2021; accepted 7 July 2021 Published online 15 July 2021 10.1126/science.abj8754

TRANSLATION

Mechanisms that ensure speed and fidelity in eukaryotic translation termination Michael R. Lawson1†, Laura N. Lessen2,3†, Jinfan Wang1, Arjun Prabhakar,1‡, Nicholas C. Corsepius1§, Rachel Green3,4*, Joseph D. Puglisi1* Translation termination, which liberates a nascent polypeptide from the ribosome specifically at stop codons, must occur accurately and rapidly. We established single-molecule fluorescence assays to track the dynamics of ribosomes and two requisite release factors (eRF1 and eRF3) throughout termination using an in vitroÐreconstituted yeast translation system. We found that the two eukaryotic release factors bound together to recognize stop codons rapidly and elicit termination through a tightly regulated, multistep process that resembles transfer RNA selection during translation elongation. Because the release factors are conserved from yeast to humans, the molecular events that underlie yeast translation termination are likely broadly fundamental to eukaryotic protein synthesis.

P

rotein synthesis concludes when a translating ribosome encounters a stop codon at the end of an open reading frame, triggering recruitment of two factors to liberate the nascent polypeptide: eukaryotic release factor 1 (eRF1), a tRNA-shaped protein that decodes the stop codon in the ribosomal aminoacyl-tRNA site (A site) and cleaves the peptidyl-tRNA bond (1–3), and eukaryotic release factor 3 (eRF3), a GTPase that promotes eRF1 action (4–6). After translation termination, the ribosome, peptidyl-tRNA site (P site) tRNA, and mRNA are released by recycling (4, 7, 8). Despite decades of study, the order and timing of the molecular events that drive translation termination remain unclear because multistep processes are difficult to assess using traditional approaches. A cohesive understanding of translation termination and its underlying steps that are central to normal translation would also support the treatment of diseases caused by premature stop codons, which include cystic fibrosis, muscular dystrophy, and hereditary cancers (9). Because premature stop 1

Department of Structural Biology, Stanford University School of Medicine, Stanford, CA, USA. 2Program in Molecular Biophysics, Johns Hopkins University School of Medicine, Baltimore, MD, USA. 3Department of Molecular Biology and Genetics, Johns Hopkins University School of Medicine, Baltimore, MD, USA. 4Howard Hughes Medical Institute, Johns Hopkins University School of Medicine, Baltimore, MD, USA *Corresponding author. Email: [email protected] (R.G.); [email protected] (J.D.P.) †These authors contributed equally to this work. ‡Present address: Pacific Biosciences Inc., Menlo Park, CA, USA. §Present address: Department of Chemistry, Fresno City College, Fresno, CA, USA.

codons cause 11% of all heritable human diseases (10), stop codon readthrough therapeutics have immense clinical potential (9, 11). Direct tracking of release factor dynamics

Here, we used an in vitro–reconstituted yeast translation system (12) and single-molecule fluorescence spectroscopy to track eukaryotic release factor dynamics and termination directly. We reasoned that ribosomes translating mRNAs with very short open reading frames would provide the simplest system for detailed analysis of the discrete substeps of termination. Ribosome complexes were programmed with Met (M-Stop) or Met-Phe (M-F-Stop) mRNAs, achieved by incubation with purified Met-tRNAMeti, initiation factors, elongation factors, and tRNAs (as appropriate) and then reacted with saturating amounts of eRF1 and eRF3 (13). Peptide release from both M-Stop and M-F-Stop ribosome complexes occurred at similar rates as a longer tetrapeptide (M-F-K-KStop)–programmed ribosome complex (Fig. 1A and fig. S1, A and B) and also matched the rate previously characterized for tripeptideprogrammed ribosome complexes (4, 5). To monitor eRF1 and eRF3 binding to ribosomes in real time, we labeled both proteins specifically with fluorescent dyes (fig. S1, C and D) and established that the labeled proteins exhibited wild-type peptide release activity (Fig. 1B and fig. S1, E and F). Association of eRF1 with the ribosome was monitored by Förster resonance energy transfer (FRET) between 60S subunits labeled with Cy3 (FRET donor) on the sciencemag.org SCIENCE

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

Fig. 1. Bulk biochemical and single-molecule studies of termination. (A) Peptides are liberated at similar rates from ribosomes translating a variety of model mRNAs. (B) Wild-type and labeled release factors liberate peptides from ribosomes. Catalytically dead eRF1 (orange) is inactive. (C) Structural modeling

[Protein Data Bank (PDB) ID: 5LZT (15)] suggests that labeled eRF1 (green, Cy5 labeled at red star) would FRET with ribosomes (red, Cy3 labeled at green star) upon binding to the A site. (D) Example of FRET observed with Cy3-eRF1 and Cy5-60S by total internal reflection fluorescence microscopy.

Fig. 2. eRF3 promotes fast binding of eRF1 to ribosomes halted at stop codons. (A) Experimental setup. (B) Assay schematic. (C) Example of fast binding of eRF1 (Cy3, green) to M-Stop ribosomes (Cy5, red) observed in the presence of eRF3. (D) Binding of eRF1 to M-Stop ribosomes is fast and concentration dependent in the presence of eRF3. Association time distributions were fit to a double-exponential model. (E) Observed rates of eRF1 binding to M-Stop ribosomes (kobs) with and without eRF3. The plateau observed with the eRF1/eRF3/GTP fast phase coincides with the rate of sample mixing in ZMWs (17). (F) Binding of eRF1 to M-Stop ribosomes is slow and eRF1 concentration independent without eRF3. Association time distributions were fit to an exponential model.

C terminus of uL18 (14) and eRF1 labeled with Cy5 (FRET acceptor) on the N terminus. Structural models placed these termini ~50 Å apart when eRF1 was bound in the A site (Fig. 1C) (3, 15). Next, Cy3-labeled ribosomal complexes programmed with Met in the P site and either UAA or UUC in the A site were combined with Cy5-eRF1 and unlabeled eRF3, and FRET was monitored at equilibrium using total internal reflection fluorescence microscopy. eRF160S FRET was only observed when a stop codon was in the A site (Fig. 1D and fig. S1, G and H), demonstrating the specificity of the FRET signal for proper eRF1 association mediated by a stop codon. We leveraged this FRET-based binding signal to determine the roles of eRF1 and eRF3 in translation termination. We first prepared SCIENCE sciencemag.org

80S ribosomal complexes programmed on 5′biotinylated M-Stop mRNAs with Cy5-labeled 60S subunits; these complexes were tethered to neutravidin-coated zero-mode waveguide (ZMW) surfaces (16, 17). Upon start of realtime data acquisition, Cy3-eRF1, excess GTP, and unlabeled eRF3 were added to ZMWs, and Cy3-eRF1 and Cy5-60S fluorescence within individual ZMWs was monitored by excitation with a 532-nm laser (Fig. 2A). Rapid, concentration-dependent eRF1 binding to the ribosomal A site was detected upon delivery of the release factors (Fig. 2, B to D, and fig. S2, A to C). Association kinetics were fit to a double-exponential function with a dominant (56 to 83%) eRF1 concentration–dependent fast phase with a pseudo–second-order rate constant of 6.3 ± 3.9 mM−1 s−1 [95% confidence

interval (CI); Fig. 2, D and E, and fig. S2B]; a minor (17 to 44%) slow phase, which did not vary with eRF1 concentration, was also observed (e.g., kobs = ~0.009 s−1; Fig. 2E and fig. S2B). Conversely, eRF1 bound very slowly to these same complexes in the absence of eRF3 (e.g., kobs = ~0.008 s−1; Fig. 2, E and F, and fig. S2, D and E); the rate constant for this eRF3independent binding was similar to the slow phase observed with eRF3 and also was unaffected by eRF1 concentration. In all cases, eRF1-binding events were long-lived (e.g., t = 227 ± 13 s; fig. S2, B and E), and prolonged detection was likely limited by the dye photobleaching lifetime (fig. S2F). In the presence of eRF3, the rapid eRF1 binding observed here was similar to the rate of Phe-tRNAPhe ternary complex binding to its cognate A-site codon 20 AUGUST 2021 • VOL 373 ISSUE 6557

877

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

Fig. 3. Observing eRF3 dynamics in ZMWs. (A) Assay schematic. (B) Example of eRF3 binding to M-Stop ribosomes. (C) Binding of eRF3 to M-Stop ribosomes is concentration dependent. Association time distributions were fit to an exponential model. (D) GTP hydrolysis by eRF3 is not required for its release from the ribosome in the absence of eRF1.

Fig. 4. eRF3 delivers eRF1 quickly to ribosomes halted at stop codons. (A) Assay schematic. (B) Example of simultaneous binding of eRF1 (Cy5, red) and eRF3 (Cy3.5, yellow) to M-Stop ribosomes (Cy3, green). (C) Postsynchronization plot of fluorescence changes observed upon simultaneous binding of eRF1 and eRF3 (dashed, black vertical line). (D) Simultaneous binding of eRF1 and eRF3 to M-Stop ribosomes is fast and concentration dependent. Association time distributions were fit to a double-exponential model. (E) GTP hydrolysis by eRF3 accelerates its release from the ribosome in the presence of eRF1. 878

20 AUGUST 2021 • VOL 373 ISSUE 6557

under similar conditions (9.0 ± 0.4 mM−1 s−1; fig. S2G). These results indicate that eRF1 binding, which would otherwise be limited by a slow event, is rapid enough to compete with tRNAs for A-site occupancy when assisted by eRF3. We next tracked eRF3 dynamics directly, independently of eRF1, to establish a baseline understanding of its interaction with the ribosome. We used a previously established interribosomal subunit FRET signal to confirm 80S complex formation (14) and monitored dyelabeled eRF3 dynamics by fluorescent bursts that occured upon factor binding to immobilized ribosomes. Ribosomes, Cy3 labeled on uL18, and Cy5 labeled on uS19 (yielding FRET upon 80S formation) were programmed with 5′-biotinylated M-Stop mRNAs and tethered to ZMWs. Next, Cy5-eRF3 and GTP were added to ZMWs and illuminated with 532- and 642-nm lasers. After an initial phase of FRET, typified by rapid Cy5-40S photobleaching, brief bursts of additional Cy5 signal were observed that marked binding and dissociation of eRF3 (Fig. 3, A and B). eRF3 binding was concentration dependent (Fig. 3C), and association kinetics were fit to an exponential function with a pseudo–second-order rate constant of 0.4 ± 0.2 mM−1 s−1 (95% CI; fig. S3, A and B). eRF3 resided briefly on the ribosome (t = 0.15 ± 0.01 s; Fig. 3D), and the dwell times between eRF3-binding events varied with its concentration (fig. S3B), consistent with a bimolecular association reaction. Inclusion of GTP analogs, GDP, or a GTPase-deficient eRF3 mutant [H348E (4)] did not markedly affect the association or dissociation rates of eRF3 (about twofold or less; Fig. 3D and fig. S3C) suggesting that this binding cycle occurs independently of the eRF3 nucleotide–bound state or GTP hydrolysis. Two distinct models could explain how eRF3 promotes the fast association of eRF1 with ribosomes halted at stop codons. eRF3 may first bind to ribosomes, triggering rearrangements that favor subsequent association of eRF1 with ribosomes. Alternatively, eRF3 may act as a chaperone, directly delivering eRF1 to ribosomes (5). To distinguish between these models, we performed single-molecule experiments similar to those described above but now simultaneously tracking fluorescent eRF1 and eRF3. We observed concurrent binding of the two factors to M-Stop ribosomes (Fig. 4, A and B). Although we also observed eRF1 and/or eRF3 binding individually to ribosomes in these experiments (which was unsurprising because the release factors can each bind alone to ribosomes and are at subsaturating concentrations), the likelihood of such independent binding events occurring simultaneously was very low (600 kD), as determined by size exclusion chromatography. (C) Representative negative stain transmission electron micrographs (TEMs) of the Mm orthologs of the CA domain–containing proteins. Scale bar, 100 nm.

SCIENCE sciencemag.org

W.C.

M A PE 6 G M 10 m R TL 1 M

m

A5

PN

M

m

A3 m

PN

M

5

M PN

ZC

m M

m

M

c

C C

H

Ar m

M M

1

M

AP

PN m

M

m

M

O

PR AS

M

m M

A1

0.0

M

HA

50

40

CD81

****

V1

band intensity ratio VLP / (whole cell+VLP)

blot VLP fraction for HA

kDa 160

CNX CNX

filter supernatant with 0.45µm filter and concentrate via ultracentrifugation to obtain VLP fraction

VLP fraction from HEK293FT cells

M

m

AS

HA

PR V1 m AP PN 1 M M m A1 A M rc m Zc M ch c m P 12 M NM m A3 P M NM m A5 P M NM m A6 P M EG m 10 R m TL1 C he rry

overexpress HA-tagged capsid protein in HEK293FT

O

MmPNMA6A

M

MmPNMA5

m

MmPNMA3

E

G

MmMOAP1

MmASPRV1

M

negative stain TEM

MmASPRV1

cryo TEM

C

(D) Representative electron micrographs using cryogenic electron microscopy (cryoTEM) of a selected subset of the identified CA domain–containing proteins. Scale bar, 50 nm. (E) Method for detecting extracellular forms of CA domain– containing homologs. (F) Representative blots of CA domain–containing proteins in the cell-free fraction. CD81 was used as loading control for the ultracentrifuged cell-free fraction. Whole-cell (W.C.) and VLP fraction blots for the endoplasmic reticulum marker CALNEXIN (CNX) ensure equal loading of whole cell protein and the purity of cell-free VLP fraction. (G) Quantification of extracellular CA domain–containing proteins [as in (F)] on the basis of n = 3 replicates. ****P < 0.0001.

20 AUGUST 2021 ¥ VOL 373 ISSUE 6557

883

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

5 significance (-log10 q-value)

NT sgRNA

3

Rn7s2

5

10

ribosome frameshift leads to 2 polypeptides

50

10

CA

PR

NC*

100

50

0

RT*

M

m

differential RNA abundance in VLP fraction (log2 fold change)

150

G

proteolytic processing

5

MmPeg10 mRNA in VLP fraction from N2a cells expressing different MmPeg10 mutants

eCLIP detection of MmPEG10-HA-bound RNAs in N2a cells

H

5' UTR

MmPeg10 3' UTR

1.0

MmPEG10-HA

0.5

MmPEG10(ΔNC)-HA

5' UTR

MmPEG10(ΔRT)-HA 0

5

10

15

enrichment of MmPEG10-bound mRNAs (log2 fold change relative to HA-GFP cells)

HA-GFP

Fig. 2. MmPEG10 protein and mRNA are secreted in vesicles by cells in vitro. (A) Method for identifying nucleic acids that are secreted in the VLP fraction upon gene activation of CA domainÐcontaining proteins. (B) Differential RNA abundance and significance in the VLP fraction from N2a cells after CRISPR activation of endogenous MmPeg10. NT, nontargeting gRNA. (C) Alignment of sequencing reads showing sequencing coverage of the MmPeg10 mRNA from (B). (D) Differential RNA abundance and significance in the VLP fraction from N2a cells after heterologous transfection of MmPeg10. n = 3 replicates. CMV, cytomegalovirus. (E) Four domains of MmPEG10 are translated into two isoforms. These are self-processed by the PEG10 protease into separate 20 AUGUST 2021 • VOL 373 ISSUE 6557

3' UTR

HA-MmPeg10 knock-in mouse

0.5 kbp

2000 reads

significance (-log10 q-value)

eCLIP detection of MmPeg10 mRNA of P30 frontal cortex tissue from HA-MmPeg10 knock-in mouse

MmDdit4 MmDdit4 MmPeg10 MmYwhag

0

884

I

eCLIP detection of MmDdit4 mRNA bound by MmPEG10 mutants in N2a cells

1.5

M

* RNA binding

G

m 10 P (d eg el 10 4 M 16- ΔN m 42 C (d Pe 9) el g 86 10 1- -Δ 10 R 37 T )

*ribosomal frameshift

FP

F

MmPeg10 gene

Pe g

E

MmPeg10

0

VLP

5 kbp

transcription & translation

–5

W.C.

500 reads

0

C57BL/6 wildtype mouse 3 kbp

500 reads

–5

CMV::mCherry CMV::MmPeg10

0 –10

VLP

differential RNA abundance in VLP fraction (log2 fold change)

RNA found in the VLP fraction from N2a cells after heterologous expression of MmPeg10

100

W.C.

1

sequence total RNA

significance (-log10 q-value)

MmPeg10

MmPeg10 sgRNA

2

0 –10

150

RNAseq coverage of MmPeg10 after CRISPR activation in whole-cell and VLP fractions

Rn7s1 MmPeg10

4

filter, concentrate, MNase treat VLP fraction

D

C

RNA found in the VLP fraction from N2a cells after CRISPR activation of endogenous MmPeg10

MmPeg10 sgRNA

B

activate each capsid gene in N2a cells using CRISPR activation

their export and transfer as well. MmArc, by contrast, contains only the CA domain and has also been shown to form capsids and transfer Arc and other mRNAs across synapses (10). To narrow down the scope of our analysis, we focused on CA domain–containing proteins that are conserved between human and

fold MmPeg10 mRNA enrichment

A

that contains the MA, CA, and NC domains. It has been shown to form capsids, bind its own mRNA, and transfer it from motor neurons to muscles at the neuromuscular junction (9). darc1 mRNA binding is dependent on its own 3′untranslated region (3′ UTR), and fusion of this sequence to heterologous mRNAs can initiate

NT sgRNA

of pol, namely a PRO domain and a predicted RT-like domain (Fig. 1A and table S1). Phylogenetic analysis of Peg10 and its homologs supports the origin of this gene from LTR retrotransposons (fig. S1, A and B). Among these genes, Arc is the most well studied. Drosophila Arc1 (darc1) is a gag homolog

domains, of which the NC and RT bind RNA. (F) Fold enrichment of MmPeg10 mRNA compared with GFP in the VLP fraction from N2a cells transfected with wild-type MmPeg10 or deletions of the predicted nucleocapsid (DNC) and reverse transcriptase (DRT) domains. (G) Log2 fold change and significance of bound RNAs from eCLIP data comparing HA-GFP with wild-type MmPEG10-HA. (H) Representative sequencing alignment histogram of the MmDdit4 locus generated from eCLIP of N2a cells transfected with wild-type or mutant MmPeg10. (I) Representative sequencing alignment histogram of the MmPeg10 locus generated from eCLIP data of n = 3 HA-PEG10 and n = 3 untagged animals. sciencemag.org SCIENCE

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

mouse and have detectable levels of mRNA in adult human tissues, reasoning that such proteins were most likely to have been co-opted for important physiological roles in mammals (fig. S2). We produced mouse versions of the selected CA-containing proteins in Escherichia coli and found that a number of these formed higher molecular weight oligomers that were identified by size exclusion (Fig. 1B and fig. S3A), as previously noted for some of these proteins, such as MmArc (10). Electron microscopy of these aggregated proteins showed that MmMOAP1, MmZCCHC12, MmRTL1, MmPNMA3, MmPNMA5, MmPNMA6a, and MmPEG10 selfassemble into capsid-like particles, many of which appear spherical (Fig. 1, C and D, and fig. S3, B and C). MmPEG10 binds and secretes its own mRNA

To determine whether these proteins are secreted within an EV, we overexpressed an epitope-tagged mouse ortholog of each CA-containing gene in human embryonic kidney (HEK) 293 FT cells and harvested both the whole-cell lysate and the viruslike particle (VLP) fraction by clarification and ultracentrifugation of the culture media (Fig. 1E). We found that MmMOAP1, MmArc, MmPEG10, and MmRTL1 were all present in the VLP fraction (Fig. 1F and fig. S4A), but MmPEG10 was the most abundant protein in the VLP fraction (Fig. 1G). Additionally, endogenous MmPEG10, but not MmMOAP1 or MmRTL1, was readily detectable in cell-free adult mouse serum (fig. S4B). We next tested whether any of the capsidlike particles formed by Gag homologs contained specific mRNAs using RNA sequencing. To avoid the possibility of transfected Gag homolog expression plasmids contributing to high background signal during sequencing, we used CRISPR activation (19) to induce expression of endogenous genes in mouse N2a cells (Fig. 2A and fig. S5A). We performed mRNA sequencing on whole-cell lysate and the VLP fraction (after nuclease treatment to remove any residual, unencapsidated RNA) to identify RNA species in the VLP fraction. We found that MmPeg10 transcriptional activation led to the accumulation of appreciable amounts of full-length MmPeg10 mRNA transcripts in the VLP fraction (Fig. 2, B and C). Previous work on MmPEG10 demonstrated that it binds a number of mRNAs inside trophoblast stem cells, including itself (13); however, here we further show that MmPEG10 binds and secretes its own mRNA into the VLP fraction. An important caveat of this experiment is that some of these proteins, particularly MmArc, are subject to regulation at the level of translation, so the lack of enrichment in the VLP fraction could be due to low protein expression (20). To confirm our observation for MmPeg10, we transiently transfected overexpression plasmids of UTR-flanked MmPeg10 into N2a cells SCIENCE sciencemag.org

and found only enrichment for MmPeg10 mRNA in the VLP fraction (Fig. 2D) under this overexpression condition. PEG10 contains two putative nucleic acid-binding domains, namely the NC and RT, which are released from the polypeptide upon PEG10 self-processing (21) (Fig. 2E, supplementary text 1, and fig. S5, B to D). We generated deletions of these domains and found that mRNA export depends on the MmPEG10 NC, as loss of the nucleic acid– binding zinc finger CCHC motif (residues 416 to 429) from the MmPEG10 NC substantially reduced export of its mRNA (Fig. 2F). To better understand the roles of the nucleic acid–binding domains of MmPEG10 in RNA binding, we performed enhanced cross-linking and immunoprecipitation (eCLIP) in N2a cells after transient transfection with hemagglutinin (HA)–tagged MmPeg10 as well as the NC and RT mutants (fig. S6, A and B). Compared with the control, MmPEG10 strongly bound a number of mRNAs in N2a cells, including its own mRNA (Fig. 2G). Notably, both the NC and the RT domains are required for the binding of these mRNAs by MmPEG10 (Fig. 2H and fig. S6C). To confirm MmPEG10’s cellular role in an in vivo context, we generated knockin mice carrying an N-terminal HA tag on the endogenous MmPEG10 protein (fig. S6D). Expression of MmPeg10 in cortical neurons has been demonstrated previously (fig. S6E) (22). Endogenous MmPEG10 was also found to bind its own mRNA as well as other transcripts abundant in neurons (fig. S6, F and G); in contrast to previous datasets, we detected strong MmPEG10 binding in the 5′ UTR, as well as some additional binding near the boundary between the NC and PRO coding sequences and in the beginning of the 3′ UTR (Fig. 2I) (13). Binding of mRNA by MmPEG10 has been reported to increase the cellular abundance of target transcripts (13). To confirm this role of MmPEG10 in its native context in vivo, we perturbed MmPeg10 gene expression in the postnatal mouse brain and assessed the expression changes of MmPEG10-bound transcripts (supplementary text 2). We found that the mRNAs of 49 genes that are down-regulated in the brain upon MmPeg10 knockout are bound to MmPEG10 in the age-matched mouse brain (fig. S7F), suggesting that one of the functions of MmPEG10 is to bind and stabilize mRNAs with fundamental roles in neurodevelopment. Pseudotyped PEG10 VLPs can deliver engineered cargo mRNAs bearing RNA packaging signals from PEG10 UTRs

To reprogram MmPEG10 to bind and package heterologous RNA, we tested whether a cargo mRNA consisting of both the 5′ and 3′ UTR of MmPeg10 flanking a gene of interest would be efficiently packaged, exported, delivered, and

translated in recipient cells (Fig. 3A). This UTR grafting approach has been demonstrated for the Ty3 retroelement and darc1 (9, 23). We first used a Cre-loxP system, a highly sensitive system for tracking RNA exchange that has been used previously with exosomes in vivo (24). We flanked the Cre recombinase coding sequence with the MmPeg10 UTRs and cotransfected it with MmPeg10 with and without a fusogen, the vesicular stomatitis virus envelope protein (VSVg) (Fig. 3A). We found that MmPEG10 VLPs pseudotyped with VSVg are secreted within EVs that mediate transfer of Cre mRNA, not protein, into target loxP– green fluorescent protein (GFP) reporter N2a cells in a VSVg- and MmPeg10 UTR–dependent manner (Fig. 3, B to D; fig. S8; and supplementary text 3). This result suggests that addition of the Peg10 UTRs enables the functional intercellular transfer of an mRNA via VLPs and that these VLPs require a fusogenic protein for cell entry. We next examined whether there is a minimal UTR packaging signal for mediating efficient packaging and functional transfer. The 3′ UTR of MmPeg10 is ~4 kb long, but eCLIP indicates that only portions of the 3′ UTR are bound by MmPEG10 (Fig. 2I). We created constructs that encode the MmPeg10 5′ UTR, Cre, and 500–base pair (bp) segments of the MmPeg10 3′ UTR. We found that the proximal 500 bp of the MmPeg10 3′ UTR are sufficient for efficient functional transfer of Cre mRNA into target reporter cells (Fig. 3E). Notably, no efficient functional mRNA transfer was observed for non–UTR-flanked Cre or for Cre without the proximal 500 bp of the 3′ UTR. Henceforth, we refer to RNA cargo flanked by the MmPeg10 5′ UTR and the proximal 500 bp of the 3′ UTR as Mm.cargo(RNA), where “(RNA)” specifies the cargo being flanked [e.g., Mm.cargo(Cre)]. Like the mouse ortholog, human PEG10 (HsPEG10) is an abundantly secreted protein in the VLP fraction (fig. S10A). Using the same approach that we employed with MmPeg10, we identified that the 5′ UTR and the first 500 bp of the HsPEG10 3′ UTR are sufficient to mediate functional transfer of Cre mRNA, hereafter denoted as Hs.cargo(RNA) (Fig. 3F). Notably, these functional regions of the UTRs are highly conserved across mammals (fig. S10B). Similar to its mouse ortholog, the human system is specific and requires HsPEG10 UTR sequences for functional mRNA transfer, whereas nonflanked Cre produced only minimal reporter cell activity (Fig. 3F). To further boost the packaging of a cargo RNA by PEG10, we explored the impact of removing any additional PEG10 cis binding elements within the MmPeg10/HsPEG10 coding sequence. For both human and mouse orthologs, transfer was increased as a result of recoding the sequence between the NC and the PRO 20 AUGUST 2021 • VOL 373 ISSUE 6557

885

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

A Peg10

transduce target cells with VLPs

Peg10 5' UTR Gene of Interest cargo RNA

Peg10 3' UTR

C

CMV::MmPeg10 + CMV::UTR Cre

Cre lentivirus

CMV::MmPeg10 + CMV::Cre

D

MmPEG10 VLP mediated mRNA transfer and expression 40

GFP

anti-PEG10 immunogold

GFP+ cells (%)

Fusogen

B

UTR-mediated packaging of cargoRNA in PEG10 VLPs

transfect Peg10, cargoRNA & fusogen plasmids

PEG10

DAPI

Negative

PEG10

PEG10

E

cargo mRNA MmPeg10 5' UTR 1

370

0 6667

3390 1 2 3 4 5 6

GFP+ cells (%) 10 20 30 40 50

F

20 10

0 CMV::MmPeg10

+

+

–

CMV::mCherry

–

–

+

–

CMV::VSV-G

+

–

+

+

CMV::UTR Cre

+

+

+

+

cargo mRNA HsPeg10 5' UTR 1

343 Cre

HsPeg10 3' UTR 2605 6618 1 2 3 4 5 6 7 8 9 10

0

GFP+ cells (%) 10 20 30 40 50

0

GFP+ cells (%) 20 40 60 80

**

MmPEG10 +

–

****

Cre

MmPeg10 3' UTR

****

30

HsPEG10 + ****

****

mCherry + untransfected mCherry +

G

0

GFP+ cells (%) 20 40 60 80

SEND(rMmPEG10,VSVg, Cre mRNA)

H SEND(rHsPEG10, VSVg, Cre mRNA)

Fig. 3. Flanking mRNA with MmPeg10 5 and 3 UTRs enables functional intercellular transfer of mRNA into a target cell. (A) Schematic showing reprogramming MmPEG10 for functional delivery of a cargo RNA flanked with the MmPeg10 5′ and 3′ UTRs [hereafter “cargo(RNA)”]. (B) Representative TEMs of VLP fraction immunogold labeled for MmPEG10. Text labels indicate transfection of cells with MmPeg10 or mock (negative). Arrowheads indicate gold labeling. Scale bar, 50 nm. (C) Representative images of loxP-GFP N2a cells treated with VSVg-pseudotyped MmPEG10 VLPs, which were produced by transfecting Mm.cargo(Cre) or Cre mRNA, and a lentivirus encoding Cre. Scale bar, 100 mm. DAPI, 4′,6-diamidino-2-phenylindole. (D) Functional transfer of RNA into loxP-GFP N2a cells mediated by VSVgpseudotyped MmPEG10 VLPs. Data were quantified by flow cytometry 72 hours after VLP addition, n = 3 replicates. (E) Functional transfer of RNA into loxP-GFP N2a cells mediated by VSVg-pseudotyped VLPs that were produced with MmPeg10 or 886

20 AUGUST 2021 ¥ VOL 373 ISSUE 6557

***

***

SEND(rMmPEG10, VSVg, Mm.cargo(Cre))

SEND(rHsPEG10, VSVg, Hs.cargo(Cre))

mCherry and Mm.cargo(Cre) constructs that encoded tiles of the MmPeg10 3′ UTR. Data were quantified by flow cytometry 72 hours after VLP addition, n = 3 replicates. (F) Functional transfer of RNA into loxP-GFP N2a cells mediated by VSVg-pseudotyped VLPs that were produced with HsPEG1010 or mCherry and Hs.cargo(Cre) constructs that encoded tiles of the HsPeg10 3′ UTR. Data were quantified by flow cytometry 72 hours after VLP addition, n = 3 replicates. (G) Functional transfer of RNA into loxP-GFP N2a cells mediated by VSVg-pseudotyped VLPs that were produced with rMmPeg10 and Mm.cargo(Cre) or Cre mRNA. Data were quantified by flow cytometry 72 hours after VLP addition, n = 3 replicates. (H) Functional transfer of RNA into loxP-GFP N2a cells mediated by VSVg-pseudotyped VLPs that were produced with rHsPeg10 and Hs.cargo(Cre) or Cre mRNA. Data were quantified by flow cytometry 72 hours after VLP addition, n = 3 replicates. For all panels, **P < 0.01, ***P < 0.001, ****P < 0.0001, one-way analysis of variance. sciencemag.org SCIENCE

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

domains, which corresponds to the MmPEG10bound region in the eCLIP experiments (Fig. 2I and supplementary text 4).

A

VSVg

Combining these optimizations, we produced VLPs with the recoded mouse and human PEG10 (rMmPEG10 or rHsPEG10), VSVg, and

SynA

B

SynB

rMmPEG10 VLPs pseudotyped with different fusogens on mouse tail-tip fibroblast cells 15

Cre mRNA

tdTomato+ cells (%)

TdTom H2A-Nuclei

Mm.cargo(Cre)

the optimized cargoRNA containing the first 500 bp of the 3′ UTR; we refer to this system as selective endogenous encapsidation for cellular

10 5 0

C

Fusogen V

V

A

A

B

Cre mRNA +

–

+

–

+

–

Mm.cargo(Cre) –

+

–

+

–

+

B

Mm.cargo(SpCas9) Cas9 cargo

VSVg

PEG10

Fusogen

rMmPeg10

****

100

60 40 20 0

G

SEND-mediated sgRNA and cargo(SpCas9) delivery into HEK293FT cells ****

60

30

40

20

20

10

0

0

m

ed ct fe ns tra

g, )) SV s9 V , Ca 10 Sp G o( PE g m .car (rM m D N A, M E S N R sg

H capsid

cargoRNA A

fusogen

PEG10

Cre mCherry

VSVg

SpCas9

SYNA

40 30 20

PNMA

10

healthy gene

SCIENCE sciencemag.org

m

SE N sg D( R rH N s A, P Sp EG C 10 as , V 9 m SV R g, N A)

, Vg 9)) VS as , C 10 Sp EG go( P s ar (rH .c D , Hs N A SE N R sg

k

0

oc

indel at HsVEGFA locus (%)

50

****

40

oc

, , , 10 )) 10 )) iral 10 A) G G G 9 s N s9 iv 9 PE a PE a nt as PE R m pC m pC Le pC m m 9 M S (r (S (rM (S (rM s D a D R D o N pC EN .UT EN arg E c S . S m S m S g, M V ,M g, VS V Vg S V VS

80

R D N A VS (rM m Vg P , s EG gR 1 N 0, A)

80

SEND-mediated sgRNA and cargo(SpCas9) delivery into N2a cell line 50

k

****

sg

100

F

plasmid- or SEND-mediated sgRNA delivery into SpCas9 N2a cell line

SE indel at MmKras locus (%) N sg D(r R Mm N m A, P oc k Sp EG 1 C 0 as , V 9 m SV R g, N A)

E

SE N

indel at MmKras locus (%)

SEND-mediated delivery of SpCas9 into MmKras sgRNA-expressing N2A cell line

indel at MmKras locus (%)

D

apply VLP to MmKras1 sgRNA cell line and quantify indel

produce MmPEG10 VLPs carrying Cas9 cargoRNA

Fig. 4. SEND is a modular system capable of delivering gene editing tools into human and mouse cells. (A) Representative images demonstrating functional transfer of Mm.cargo(Cre) or Cre mRNA in rMmPEG10 VLPs pseudotyped with VSVg (V), MmSYNA (A), or MmSYNB (B) in Ai9 (loxP-tdTomato) tail-tip fibroblasts. Scale bar, 200 mm. (B) Percent of tdTomatopositive cells out of the total number of H2A-stained nuclei from high content imaging of n = 3 replicates of (A). (C) Schematic representing the retooling of SEND for genome engineering. (D) Indels at the MmKras locus in MmKras1-sgRNAN2a cells treated with SEND (VSVgpseudotyped rMmPEG10 VLPs) containing SpCas9 mRNA, Mm.UTR (SpCas9), or Mm.cargo(SpCas9) and a lentivirus encoding SpCas9. Indels were quantified by NGS 72 hours after VLP or lentivirus addition, n = 3 replicates. (E) Indels at the mouse MmKras locus in a constitutively expressing SpCas9 N2a cell line either transfected with a plasmid carrying the MmKras sgRNA or treated with SEND (rMmPEG10, VSVg, or MmKras sgRNA). Indels were quantified by NGS after 72 hours, n = 3 replicates. (F) Indels at the MmKras locus in N2a cells treated with SEND (VSVg-pseudotyped rMmPeg10 SEND VLPs) containing either SpCas9 mRNA or Mm.cargo (SpCas9) and sgRNA. Indels were quantified by NGS 72 hours after VLP addition, n = 3 replicates. (G) Indels at the HsVEGFA locus in HEK293FT cells treated with SEND (VSVg-pseudotyped rHsPEG10 VLPs) containing either SpCas9 mRNA or Hs.cargo(SpCas9) and an unmodified sgRNA. Indels were determined by NGS 72 hours after VLP addition, n = 3 replicates. (H) SEND is a modular delivery platform combining an endogenous Gag homolog, cargo mRNA, and fusogen, which can be tailored for specific contexts.

tailored SEND

20 AUGUST 2021 • VOL 373 ISSUE 6557

887

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

delivery (SEND). With SEND, we detected a substantial (up to 60%) increase in the functional transfer of cargo(Cre) into N2a cells for both human and mouse PEG10 (Fig. 3, G and H). Furthermore, we showed that VLPs produced with rMmPEG10 can mediate the functional transfer of H2B-mCherry (fig. S12, A and B). A comparison of SEND with previously developed delivery vectors showed that SEND is four to five times less potent than an integrating lentiviral vector, as assayed by digital droplet polymerase chain reaction and functional titration (fig. S12, B to E). However, given that SEND delivers mRNA rather than integrating an overexpression cassette, we expect it to perform competitively against other mRNA delivery vehicles. PEG10 is a modular platform for RNA delivery

To generate a fully endogenous SEND system, we tested whether VSVg can be replaced with an endogenous fusogenic transmembrane protein. Given the overlapping tissue expression of MmPeg10/HsPEG10 and syncytin genes (supplementary text 5), we tested the feasibility of pseudotyping the mouse SEND system with MmSYNA or MmSYNB compared with pseudotyping with VSVg. Pseudotyped particles were incubated with tail-tip fibroblasts from loxPtdTomato reporter mice, a cell type that we have found amenable to transduction by these fusogens. Based on previous reports, we added the transduction enhancer vectofusin-1 to the supernatant for MmSYNA and MmSYNB particles to enhance in vitro transduction (25). In these primary cells, both VSVg and MmSYNA enabled SEND-mediated functional transfer of Mm.cargo(Cre), whereas MmSYNB did not (Fig. 4, A and B). Again, this packaging was highly specific, as only UTR-flanked mRNA [i.e., Mm.cargo(Cre)] was functionally transferred. Together with MmSYNA, SEND can be configured as a fully endogenous system for functional gene transfer. Supported by our understanding of the minimal requirements for PEG10-mediated mRNA delivery (i.e., UTRs and an endogenous fusogen), we could begin to probe the endogenous role of MmPEG10-mediated MmPeg10 RNA delivery in neurons. The functional transfer of MmSYNApseudotyped VLPs that carry the native PEG10 transcript into primary mouse cortical neurons led to up-regulation of a number of genes involved in neurodevelopment (supplementary text 6). This finding reinforces the notion that one role of endogenous MmPeg10 delivery is binding and stabilizing specific mRNA transcripts in recipient cells. RNA sequencing of N2a cells receiving Mm.cargo(Peg10) revealed substantial gene expression changes upon MmPeg10 delivery that were largely abrogated with PEG10-mediated Mm.cargo(Cre) delivery (supplementary text 7). This suggests that transferring a reprogrammed cargo does not 888

20 AUGUST 2021 ¥ VOL 373 ISSUE 6557

have the same impact on recipient cells as transferring MmPeg10 and indicates that MmPeg10 transcript delivery rather than the delivery of MmPEG10 protein is responsible for the observed gene expression changes. It remains unclear whether MmPEG10 VLPs are natively pseudotyped by the endogenous fusogen MmSYNA to enable cellular uptake of PEG10 VLPs in the central nervous system. To further characterize the modularity of the components of this system, we tested different cargoRNAs. Using the same pipeline developed for cargo(Cre), we tested whether SEND could mediate the functional transfer of a large ~5-kb Mm.cargo(SpCas9) into N2a cell lines that constitutively express a single guide RNA (sgRNA) against MmKras (Fig. 4C). SEND was able to functionally transfer SpCas9, leading to ~60% insertions and deletions (indels) in recipient cells (Fig. 4D); similar to the results with Cre, SEND is specific and only able to efficiently functionally transfer SpCas9 flanked by either the full-length or optimized Peg10 UTR sequences. To create an all-in-one vector for delivery of sgRNA and SpCas9, we first tested whether an sgRNA can be efficiently delivered by SEND. We independently packaged an sgRNA targeting Kras into rMmPeg10 VLPs by coexpressing rMmPeg10 with VSVg and a U6-driven sgRNA and incubated them with Cas9-expressing N2a cells; we detected very little activity even though direct transfection of the guide showed robust indel formation (Fig. 4E). We found, however, that copackaging the guide alongside Mm.cargo(SpCas9) by coexpressing Mm.cargo (SpCas9) with a U6-driven sgRNA on a separate plasmid was sufficient to mediate 30% indels (Fig. 4F). To determine the reproducibility of this genome-editing approach, we repeated this copackaging strategy with the human SEND system and were able to generate ~40% indels in HEK293FT cells at the HsVEGFA locus (Fig. 4G). The development of SEND (Fig. 4H) from an endogenous retroelement complements existing delivery approaches using lipid nanoparticles (26), VLPs derived from bona fide retroviruses (27–29), and active mRNA-loading approaches in EVs (30, 31). Moreover, SEND may have reduced immunogenicity compared with currently available viral vectors (32) because of its use of endogenous human proteins. Supporting this are gene expression data from the developing human thymus, which demonstrate that HsPEG10 is highly expressed compared with other CA-containing genes in the thymic epithelium (fig. S16) (33), which is responsible for T cell tolerance induction. As a modular, fully endogenous system, SEND has the potential to be extended into a minimally immunogenic delivery platform that can be repeatedly dosed, which greatly expands the applications for nucleic acid therapy.

REFERENCES AND NOTES

1. J. L. Goodier, H. H. Kazazian Jr., Cell 135, 23–35 (2008). 2. A. F. Smit, Curr. Opin. Genet. Dev. 9, 657–663 (1999). 3. M. R. Patel, M. Emerman, H. S. Malik, Curr. Opin. Virol. 1, 304–309 (2011). 4. L. Guio, J. González, Methods Mol. Biol. 1910, 505–530 (2019). 5. C. Feschotte, C. Gilbert, Nat. Rev. Genet. 13, 283–296 (2012). 6. F. J. Kim, J.-L. Battini, N. Manel, M. Sitbon, Virology 318, 183–191 (2004). 7. A. Dupressoir et al., Proc. Natl. Acad. Sci. U.S.A. 106, 12127–12132 (2009). 8. C. Myrum et al., Biochem. J. 468, 145–158 (2015). 9. J. Ashley et al., Cell 172, 262–274.e11 (2018). 10. E. D. Pastuzyn et al., Cell 173, 275–288 (2018). 11. E. Korb, S. Finkbeiner, Trends Neurosci. 34, 591–598 (2011). 12. P. Barragan-Iglesias, J. B. De La Pena, T. F. Lou, S. Loerch, Cell Rep. 10.2139/ssrn.3684856 (2020). doi: 10.2139/ssrn.3684856. 13. M. Abed et al., PLOS ONE 14, e0214110 (2019). 14. R. Ono et al., Nat. Genet. 38, 101–106 (2006). 15. C. Henke et al., Retrovirology 12, 9 (2015). 16. M. Krupovic, E. V. Koonin, Proc. Natl. Acad. Sci. U.S.A. 114, E2401–E2410 (2017). 17. S. O. Dodonova, S. Prinz, V. Bilanchone, S. Sandmeyer, J. A. G. Briggs, Proc. Natl. Acad. Sci. U.S.A. 116, 10048–10057 (2019). 18. M. Campillos, T. Doerks, P. K. Shah, P. Bork, Trends Genet. 22, 585–589 (2006). 19. S. Konermann et al., Nature 517, 583–588 (2015). 20. C. S. Wallace, G. L. Lyford, P. F. Worley, O. Steward, J. Neurosci. 18, 26–35 (1998). 21. M. Golda, J. A. Mótyán, M. Mahdi, J. Tőzsér, Functional Study of the Retrotransposon-Derived Human PEG10 Protease, Int. J. Mol. Sci. 21, 2424 (2020). 22. A. Saunders et al., Cell 174, 1015–1030.e16 (2018). 23. K. Clemens, V. Bilanchone, N. Beliakova-Bethell, Virus Res. 171, 319–331 (2013). 24. K. Ridder et al., OncoImmunology 4, e1008371 (2015). 25. Y. Coquin, M. Ferrand, A. Seye, L. Menu, A. Galy, bioRxiv 816223 [Preprint]. 24 October 2019. 26. P. S. Kowalski, A. Rudra, L. Miao, D. G. Anderson, Mol. Ther. 27, 710–728 (2019). 27. U. Mock et al., Sci. Rep. 4, 6409 (2014). 28. S. J. Kaczmarczyk, K. Sitaraman, H. A. Young, S. H. Hughes, D. K. Chatterjee, Proc. Natl. Acad. Sci. U.S.A. 108, 16998–17003 (2011). 29. P. E. Mangeot et al., Nat. Commun. 10, 45 (2019). 30. R. Kojima et al., Nat. Commun. 9, 1305 (2018). 31. M. E. Hung, J. N. Leonard, J. Extracell. Vesicles 5, 31027 (2016). 32. J. L. Shirley, Y. P. de Jong, C. Terhorst, R. W. Herzog, Mol. Ther. 28, 709–722 (2020). 33. J.-E. Park et al., Science 367, eaay3224 (2020). AC KNOWLED GME NTS

We thank D. S. Yun for electron microscopy assistance, A. Koller for mass spectrometry assistance, L. Wu and the Harvard GMF for the generation of transgenic animals, A. Tang for illustration assistance, and the entire Zhang laboratory for support and advice. Funding: This work was supported by a grant from the Simons Foundation to the Simons Center for the Social Brain at MIT (M.S.); National Institutes of Health Intramural Research Program (E.V.K.); National Institutes of Health grants 1R01-HG009761 and 1DP1HL141201 (F.Z.); Howard Hughes Medical Institute (F.Z.); Open Philanthropy (F.Z.); G. Harold and Leila Y. Mathers Charitable Foundation (F.Z.); Edward Mallinckrodt, Jr. Foundation (F.Z.); Poitras Center for Psychiatric Disorders Research at MIT (F.Z.); Hock E. Tan and K. Lisa Yang Center for Autism Research at MIT (F.Z.); Yang-Tan Center for Molecular Therapeutics at MIT (F.Z.); Lisa Yang (F.Z.); Phillips family (F.Z.); R. Metcalfe (F.Z.); and J. and P. Poitras (F.Z.). Author contributions: M.S. and F.Z. conceived the project. M.S., B.L., and F.Z. designed the experiments. M.S., B.L., J.S., A.L., X.J., and C.C.L. performed the experiments. M.S., B.L., J.S., and F.Z. analyzed the data. M.S., S.L.M., and E.V.K. performed bioinformatics analysis of Gag protein diversity. F.Z. supervised the research and experimental design with support from R.K.M., and M.S., B.L., R.K.M., and F.Z. wrote the manuscript with input from all authors. Competing interests: M.S., B.L., and F.Z. are co-inventors on a US provisional patent application filed by the Broad Institute related to this work. (U.S. Provisional Patent

sciencemag.org SCIENCE

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

Application no. 63/191,067) F.Z. is a cofounder of Editas Medicine, Beam Therapeutics, Pairwise Plants, Arbor Biotechnologies, and Sherlock Biosciences. Data and materials availability: Expression plasmids are available from Addgene under a uniform biological material transfer agreement. Additional information is available through the Zhang Lab website (https://zlab.bio). Next-generation sequencing data generated are available from National Center for Biotechnology Information Sequence Read Archive with

accession number PRJNA743280. All other data are available in the paper and supplementary materials. SUPPLEMENTARY MATERIALS

science.sciencemag.org/content/373/6557/882/suppl/DC1 Materials and Methods Supplementary Text 1 to 7

CORONAVIRUS

Spatiotemporal invasion dynamics of SARS-CoV-2 lineage B.1.1.7 emergence Moritz U. G. Kraemer1*†, Verity Hill2†, Christopher Ruis3†, Simon Dellicour4,5†, Sumali Bajaj1†, John T. McCrone2, Guy Baele5, Kris V. Parag6, Anya Lindström Battle7, Bernardo Gutierrez1, Ben Jackson2, Rachel Colquhoun2, Áine OÕToole2, Brennan Klein8, Alessandro Vespignani8, COVID-19 Genomics UK (COG-UK) Consortium‡, Erik Volz6, Nuno R. Faria1,6,9, David M. Aanensen10,11, Nicholas J. Loman12, Louis du Plessis1, Simon Cauchemez13, Andrew Rambaut2*, Samuel V. Scarpino8,14,15*, Oliver G. Pybus1,16* Understanding the causes and consequences of the emergence of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) variants of concern is crucial to pandemic control yet difficult to achieve because they arise in the context of variable human behavior and immunity. We investigated the spatial invasion dynamics of lineage B.1.1.7 by jointly analyzing UK human mobility, virus genomes, and community-based polymerase chain reaction data. We identified a multistage spatial invasion process in which early B.1.1.7 growth rates were associated with mobility and asymmetric lineage export from a dominant source location, enhancing the effects of B.1.1.7Õs increased intrinsic transmissibility. We further explored how B.1.1.7 spread was shaped by nonpharmaceutical interventions and spatial variation in previous attack rates. Our findings show that careful accounting of the behavioral and epidemiological context within which variants of concern emerge is necessary to interpret correctly their observed relative growth rates.

T

he severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) lineage B.1.1.7 expanded rapidly across the United Kingdom (1, 2) in late 2020 and subsequently spread internationally (3, 4). As of 19 January 2021 (date of the most recent sample in our dataset), B.1.1.7 had reached all but five counties of Wales, Scotland, Northern Ireland, and England, with onward transmission in each. Restrictions on international travel were enacted to contain B.1.1.7’s spread; however, genomic surveillance has since detected the presence and growth of the lineage in many countries worldwide (4, 5). Analyses of genomic, laboratory, secondary contact, and aggregated epidemiological data estimate higher transmissibility of B.1.1.7 compared with previous SARS-CoV-2 lineages (1, 6–9) and potentially a greater risk of hospitalization (10–13). The spatial heterogeneity of SARS-

CoV-2 transmission—and of emerging infectious diseases in general—can have profound effects on the local likelihood and intensity of transmission, final epidemic size, and immunity (14–22). More specifically, estimates of B.1.1.7’s increased relative transmissibility declined during its emergence in the UK (7, 9); understanding why this occurred is necessary if we are to respond effectively to future SARSCoV-2 variants. We reconstructed and quantified the spatial dynamics of B.1.1.7’s emergence and investigated how human mobility and heterogeneity in previous exposure contributed to B.1.1.7’s initial spread and evaluation of higher transmissibility. Spatial expansion and source sink dynamics of B.1.1.7 in the UK

B.1.1.7 can be first detected in COVID-19 Genomics UK Consortium (COG-UK) genome data

Figs. S1 to S16 Tables S1 to S4 References (34Ð56) Data S1 17 January 2021; resubmitted 26 April 2021 Accepted 6 July 2021 10.1126/science.abg6155

in Kent on 20 September 2020 and spread quickly across the UK, with each week adding detections in approximately seven new uppertier local authorities (UTLAs) (Fig. 1, A and B, and table S2). B.1.1.7 was already reported in several UTLAs before the start of the second English lockdown (5 November 2020). By the end of that lockdown (2 December 2020), B.1.1.7 was widespread throughout the UK (Fig. 1, A and B). The spatial expansion of SARS-CoV-2 lineages [for example, (16, 23)] can be tracked by using data from the UK’s national surveillance of SARS-CoV-2 genomes (24). By combining these data with aggregated mobile phone data, we examined the dissemination of B.1.1.7 through human mobility, from its likely location of emergence (Kent and Greater London) to other UK regions (Fig. 1, D and E, and supplementary materials, materials and methods). Human mobility among UK regions increased at the end of the second English lockdown, from 55 million to 75 million weekly movements (Fig. 1E). Because of its centrality, Greater London exhibits an important connective role in the UK human movement network (Fig. 1D; red lines indicate the week the second lockdown was eased). Compared with that of previous weeks, movements out of Greater London were more frequent and reached more destinations (fig. S1). For each UTLA, we found that the date of first detection of B.1.1.7 is predicted well by human mobility from Kent and Greater London to that UTLA [Pearson’s correlation coefficient (r) = –0.73; 95% confidence interval (CI): –0.61, –0.81; Akaike information criteria (AIC) = 734] (Fig. 1C) and similarly well by using movements from Kent and Greater London separately (fig. S2). This correlation strengthens through time as new locations of B.1.1.7 detection are added (fig. S3) and is robust to changes in human mobility through time in among-region human movement (Pearson’s r = –0.44; 95% CI: –0.16, –0.65; P < 0.01;

1

Department of Zoology, University of Oxford, Oxford, UK. 2Institute of Evolutionary Biology, University of Edinburgh, Edinburgh, UK. 3Molecular Immunity Unit, Department of Medicine, Cambridge University, Cambridge, UK. 4Spatial Epidemiology Lab (SpELL), Université Libre de Bruxelles, Bruxelles, Belgium. 5Department of Microbiology, Immunology and Transplantation, Rega Institute, KU Leuven, 3000 Leuven, Belgium. 6MRC Centre for Global Infectious Disease Analysis, Jameel Institute for Disease and Emergency Analytics, Imperial College London, London, UK. 7 Department of Plant Sciences, University of Oxford, Oxford, UK. 8Network Science Institute, Northeastern University, Boston, USA. 9Instituto de Medicina Tropical, Faculdade de Medicina da Universidade de Sao Paulo, Sao Paulo, Brazil. 10Centre for Genomic Pathogen Surveillance, Wellcome Genome Campus, Hinxton, UK. 11Big Data Institute, Li Ka Shing Centre for Health Information and Discovery, Nuffield Department of Medicine, University of Oxford, Oxford, UK. 12Institute of Microbiology and Infection, University of Birmingham, Birmingham, UK. 13 Mathematical Modelling of Infectious Diseases Unit, Institut Pasteur, UMR2000, CNRS, Paris, France. 14Vermont Complex Systems Center, University of Vermont, Burlington, USA. 15Santa Fe Institute, Santa Fe, USA. 16Department of Pathobiology and Population Sciences, Royal Veterinary College London, London, UK. *Corresponding author. Email: [email protected] (A.R.); [email protected] (M.U.G.K.); [email protected] (S.V.S.); [email protected] (O.G.P.) †These authors contributed equally to this work. ‡Consortium members and affiliations are listed in the supplementary materials.

SCIENCE sciencemag.org

20 AUGUST 2021 • VOL 373 ISSUE 6557

889

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

D

A

I

II I

II

III Greater London Kent

2020-09-20

2020-12-29

E

Movements from Kent & Greater London (log2)

10015 90 80 70 60 50 40 30 20 10 0 Nov

15

10

Dec

Week

5 Oct

Nov

Dec

Jan

Earliest B.1.1.7 sequence date

Fig. 1. Human mobility and spatial expansion of B.1.1.7 across the UK. (A) Map at the UTLA level of arrival dates of lineage B.1.1.7. Darker colors indicate earlier dates, and lighter colors indicate later dates. Arrival time is defined as the earliest sampling date of a B.1.1.7 genomic sequence in each UTLA. (B) Cumulative number of UTLAs in which B.1.1.7 has been detected, in 7-day intervals. The blue shaded area indicates the period of the second lockdown in England. (C) Relationship between the arrival time of B.1.1.7 and estimated number of movements from Kent and London during February 2020 for each UTLA (Pearson’s r = –0.73; 95% CI: –0.61, –0.81; P < 0.001) (materials and methods). (D) Human mobility at the UK local

mobility data through 23 January 2021) (materials and methods). Geographic distance from Greater London correlates less strongly with B.1.1.7 arrival times (Pearson’s r = 0.60; 95% CI: 0.44 to 0.71; AIC = 763) (fig. S4). To understand better the spatial dispersal of B.1.1.7 during its emergence, we reconstructed its spread across England using large-scale phylogeographic analysis (25–27). We analyzed 17,716 B.1.1.7 genomes collected between 20 September 2020 and 19 January 2021 (Fig. 2 and fig. S5), collated from polymerase chain reaction (PCR)–positive community samples that represent a random selection of SARSCoV-2–positive samples (28). These genomes 890

20 AUGUST 2021 • VOL 373 ISSUE 6557

Percent of typical mobility

C

Number of weekly movements

Number of UTLAs with at least one B.1.1.7 sequence

B

Oct

III

authority district level (LAD) (table S2) during the epidemiological week 29 November to 5 December 2020. Thicker lines (edges) indicate more movements between regions. Nodes with larger absolute incoming movements are indicated with darker colors. Red lines indicate movements from Greater London. (Insets I, II, and III) Mobility within three UK metropolitan areas. (E) Trends in human mobility across the UK (indicating movements between but not within LADs). The blue shaded areas indicate the period of the first, second, and third lockdown in England. Dark red indicates the timing (20 December 2020) of the Tier 4 restrictions imposed in southeast England, including London (56).

represent ~4% of UK B.1.1.7 cases during the study period [n = 460,510 estimated tests with PCR S-gene target failure (SGTF) between 20 September 2020 and 19 January 2021]. Samples per location (UTLA) and per week in the SGTF and whole-genome datasets are strongly correlated (Pearson’s r = 0.69; 95% CI: 0.63 – 0.73; P < 0.001) (fig. S6) (7), making it feasible to reconstruct B.1.1.7 expansion history by using phylogeographic approaches (29). We identified distinct phases to the emergence of B.1.1.7. Initially, during the second English lockdown, most (71.2%) B.1.1.7 phylogenetic branch movements originated and ended in Greater London or Kent; long-distance

dispersal events were relatively infrequent (Figs. 2 and 3). After the lockdown ended, and new cases in London subsequently rose rapidly, observed virus lineage movements from southeast England to other regions increased, and other large cities started to exhibit local transmission (Figs. 2 and 3). This phase of a growing number of exported B.1.1.7 cases from London and environs stabilized in midDecember and coincided with reduced mobility from Greater London (Tier 4 restrictions were announced on 20 December 2020 and entailed a “Stay at home” order, closure of nonessential shops and hospitality, and strict limitations on household mixing) (Figs. 1E sciencemag.org SCIENCE

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

A

2020-12-01

60000 20000

SGTF cases in London

40000

estimated from mobility and prevalence surveys

0

0

50

100

5000 10000 15000 20000

estimated from continuous phylogeographic analysis

20 2 20 0-0 20 820 -0 03 2 9 20 0-0 -06 2 9 20 0-0 -13 2 9 20 0-0 -20 2 9 20 0-1 -27 2 0 20 0-1 -04 2 0 20 0-1 -11 2 0 20 0-1 -18 2 0 20 0-1 -25 2 1 20 0-1 -01 20 120 -1 08 2 1 20 0-1 -15 2 1 20 0-1 -22 2 1 20 0-1 -29 2 2 20 0-1 -06 2 2 20 0-1 -13 20 220 -1 20 2 2 20 1-0 -27 21 1-0 03 110

C

B

150

2020-12-20

B.1.1.7 lineage exports from London

2020-11-05

D

2020-10-01

2020-11-01

2020-12-01

Fig. 2. Spatial emergence dynamics of SARS-CoV-2 lineage B.1.1.7 in England. (A and B) Continuous phylogeographic reconstruction with phylogeny nodes colored according to their time of occurrence and dispersal direction of phylogeny branches indicated by edge curvature (counterclockwise). From left to right, data to 5 November, 1 December, and 20 December 2020, respectively. (B) Map of the entire reconSCIENCE sciencemag.org

2021-01-01

struction, up to 19 January 2021. (C) Estimated number of weekly exports of lineage B.1.1.7 from the Greater London area, inferred from the continuous phylogeographic analysis (red), and estimated from mobility and prevalence survey data (black). (D) Estimated number of cumulative B.1.1.7 introductions inferred from phylogeographic analysis into each administrative area (UTLA) by 12 December 2020. 20 AUGUST 2021 ¥ VOL 373 ISSUE 6557

891

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

A

10 50 100 200

100km: 72.3 and 27.7 %

50

100km: 72.2 and 27.8 %

30

60

2020-12-20 − 2021-01-12

40

100

2020-12-01 − 2020-12-20

300

400

500

distance (km)

100

200

300

400

500

distance (km)

Fig. 3. Spatial structure of B.1.1.7 lineage dispersal in England from phylogeographic reconstruction. (A) Curved arrows and line thicknesses indicate the direction and intensity of B.1.1.7 lineage flows among regions. Red circles indicate, for a given location, the ratio of inferred local movements to inferred importations into that location. Four time periods are shown (left to

and 2C). However, the total number of B.1.1.7 lineage exports did not immediately decline because the growing number of B.1.1.7 cases in southeast England offset the decline in outward travel (Fig. 2C) (30), indicating a limited effect of delayed action on B.1.1.7 spread from Greater London. Our analysis did not allow us to establish a causal link between nonpharmaceutical interventions (NPIs) and their impact on lineage exportations, so these results should be interpreted with caution. By combining mobility and SGTF data with estimates of the proportion of the population testing SARS-CoV-2–positive (materials and methods), we can estimate the frequency of B.1.1.7 export from Greater London to other English regions (Fig. 2C and fig. S7) and explore its role in accelerating the lineage’s emergence. Using these combined data sources, we estimate that the number of B.1.1.7 case exports from Greater London rose during November (including during lockdown) from 12,000 in early December (Fig. 2C, gray curve), reflecting growth in B.1.1.7 infections in Greater London and an increase in human mobility among UK geographic regions across in late November (Fig. 1E). The esti20 AUGUST 2021 • VOL 373 ISSUE 6557

20 10

0

0 200

0

20

10

4 0 100

892

2020-12-20 − 2021-01-12

40

20

6

100km: 59.6 and 40.4 %

2

Frequency

8

100km: 51.9 and 48.1 %

2020-12-01 − 2020-12-20

2020-11-05 − 2020-12-01 30

2020-09-01 − 2020-11-05

10

B

2020-11-05 − 2020-12-01

80

2020-09-01 − 2020-11-05

100

200

300

400

distance (km)

500

100

200

300

400

500

distance (km)

right) and roughly correspond to (i) before second lockdown, (ii) second lockdown, (iii) after second lockdown, and (iv) implementation of Tier 4 restrictions in southeast England. (B) Distribution of the geographic distances of phylogenetic lineage movement events (>50 km). Those from Greater London are in red, and those from other locations are in gray.

mated intensity of B.1.1.7 case exportation from Greater London remained high in December, peaking in mid-December at ~20,000 weekly exports, before declining in early January after the third national lockdown started on 5 January 2021. These estimates (Fig. 2C, gray curve) closely match the trends in lineage B.1.1.7 movement inferred from phylogeographic analysis (Fig. 2C, red curve), crossvalidating both data sources (exports estimated by using each method are strongly correlated; Pearson’s r = 0.62; 95% CI: 0.61 to 0.64; P < 0.001) (fig. S8). Lineage exportation events estimated from genomic data are lower from late December onward, possibly owing to reporting lags in genomic data generation and/ or delayed care-seeking because of the Christmas holidays (31). Our simple model assumes that nonsymptomatic infectious individuals are equally likely to travel (Fig. 2C, gray line), which may bias our estimates of infectious travellers upward. B.1.1.7 dispersal dynamics shifted in late December to more bidirectional exchange of phylogenetic lineages in and out of Greater London (Fig. 3), coinciding with rapid growth in B.1.1.7 cases across England (9). Throughout, the weekly number of B.1.1.7 cases in a

UTLA was positively associated with the number of B.1.1.7 lineage introductions into that UTLA during that week (Pearson’s r = 0.41, 0.76, 0.91, and 0.73, for October, November, December, and January, respectively; P < 0.001 for all; further analysis is provided in the supplementary materials) (fig. S6). We observed spatial heterogeneity in B.1.1.7 lineage importations; in the phylogeographic analysis, some locations received >500 inferred importations, despite our genomic dataset representing 100 km) and shortest ( 0.75] until mid-December, before declining (the trend remains when accounting for uncertainty in the estimated number of infections across Greater London) (Fig. 4C and fig. S12). This result is robust to the data and methods used to estimate perlocation B.1.1.7 importation rates (figs. S9

and S10). Accounting for continued export of B.1.1.7 from Greater London and Kent can explain in part why estimates of the growth advantage of B.1.1.7 declined during the second half of December 2020, before the implementation of tighter control measures (Tier 4, 20 December) (7, 9). Human mobility and prior outbreaks as predictors of B.1.1.7 growth

The epicenter of SARS-CoV-2 transmission in the UK shifted during the November 2020 lockdown: between 1 September and 1 December 2020, ~80% of reported cases were reported outside London and southeast England, whereas those regions accounted for ~40% of all cases during 1 to 7 December. We sought to understand how, in each location, post-lockdown growth rates related to previous attack rates as well as travel inflow to that location. We investigated predictors of the increase in the relative frequency of B.1.1.7 genomes compared with that of other SARS-CoV-2 lineages (Fig. 5A) (7, 9). In a multivariate model, we found that about half of the variation in the increase in B.1.1.7 relative frequency between 2 and 16 December is associated with human 20 AUGUST 2021 ¥ VOL 373 ISSUE 6557

893

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

A 1

0

0

0.75 0.50 0.25

-2.5 -5 -7.5

Lockdown

Post Lockdown

Tier 4

3rd Lockdown

16 8 4 12 Mobility from Greater London (log)

Sampling point Fig. 5. (A) Frequency of B.1.1.7 at the UTLA level at different sampling times. Pre-lockdown, dates before 5 November; lockdown, 5 to 30 November; post-lockdown, 1 to 15 December; Tier 4, 16 to 31 December; and the most recent sampling point, 1 to 12 January(materials and methods). (B) Increase in the frequency of B.1.1.7 sampled genomes between 2 and 16 December

mobility from Greater London and attack rates before the November lockdown (Fig. 5, B and C). UTLAs with lower previous attack rates tended to have faster-increasing B.1.1.7 frequencies. We repeated this analysis using SGTF case frequency data and obtained similar results (R2 = 0.57, P < 0.001) (fig. S13). However, neither human mobility nor pre-lockdown attack rate were significant predictors of later changes. Instead, change in the relative frequency of B.1.1.7 genomes after 17 December was best predicted simply by its frequency on that date (R2 = 0.13, P < 0.01) (fig. S14), although a model identified through exhaustive search by using Bayesian information criteria (BIC) includes the “frequency of B.1.1.7 on 17 December,” an interaction between arrival time and “frequency of B.1.1.7 on 17 December,” and an interaction between incidence before the November lockdown and mobility from London (BIC 178.467; R2 = 0.68; P < 0.001) (fig. S14). Mobility from Greater London remains a significant predictor of B.1.1.7 growth after controlling for population size by means of both a multivariate regression and modelselection by using exhaustive search with both BIC and AIC. Conclusions, limitations, and future work

We found that the emergence of B.1.1.7 throughout the UK was associated with a high export frequency from a major source location that was identified only retrospectively. This pattern recapitulates at a national scale the role that international mobility played in the early spread of the SARS-CoV-2 pandemic (38–40). We conclude that the exceptionally rapid spatial spread and early growth rates of lineage B.1.1.7 likely reflect the combined effects of its higher intrinsic transmissibility (1, 7, 9) and 894

12

-2

-4

-6

0.00 Pre Lockdown

M. from London (log) 4

Increase in B.1.1.7 frequency (log)

Medway East Sussex Kent Swindon Essex

Increase in B.1.1.7 frequency (log)

Variant frequency

C

B

20 AUGUST 2021 • VOL 373 ISSUE 6557

0.01 0.02 0.03 0.04 Attack rate pre-November 5, 2020

2020 is associated with mobility from Greater London. (C) Increase in the frequency of B.1.1.7 sampled genomes at the UTLA level is associated with previous attack rates in each location. Results for equivalent analyses of SGTF data are similar and are provided in the supplementary materials.

the spatial structure of incidence and mobility before, during, and after the second lockdown in England (41). Understanding what causes a new SARSCoV-2 lineage to grow and replace preexisting lineages is a complex problem. In addition to virus genetic changes to relevant phenotypes (such as per-contact transmissibility, duration of infectiousness, and immune evasion), lineage replacement dynamics are likely affected by spatiotemporal heterogeneity in incidence, NPIs, prior infection, and among-region mobility (42). The role of the latter may be enhanced in the context of low or declining prevalence, as suggested by the frequency growth of lineage B.1.177 in the UK and Europe during summer 2020, which was associated with international travel (43–45). Evidence for the increased intrinsic transmissibility of B.1.1.7 is clear, but estimates have varied considerably [38 to 130% increase (7, 9)]. The growth potential of new SARS-CoV-2 variants will depend also on the average durations of their exposed and infectious phases, as well as their per-contact transmissibility (36). Our results indicate that exportations from a high-incidence epidemic source region raised early locationspecific growth rate estimates across the UK (Fig. 4B), and that this effect declined through time. Similar trends have since been observed for lineage B.1.617.2 into the UK, after its importation from high-incidence regions onto a background of low incidence and lockdown easing. This conclusion is relevant for the interpretation of the current and future estimates of the increased transmissibility of B.1.1.7 (and other variants of concern) in other countries [such as the Untied States and Denmark (3)]. Further epidemiological and experimental work is needed to discriminate transient demograph-

ic factors from the permanent contribution to increased transmissibility conferred by the mutations carried by B.1.1.7. Although B.1.1.7 was first detected in Kent, UK, and is speculated to have accumulated its mutations during a chronic infection (2), because of the strong correlation between human mobility from those areas and date of B.1.1.7 detection elsewhere our results support the hypothesis that B.1.1.7 originated in Kent or Greater London. Further, our phylogeographic reconstruction shows early lineage dissemination from Kent and Greater London, indicating that B.1.1.7 spread through the UK from one dominant UK source region, as opposed to a large undetected epidemic elsewhere, which would likely have resulted in multiple introductions through international travel (16). We demonstrate that large-scale and wellsampled genomic surveillance data can reveal the detailed spatial transmission dynamics of individual SARS-CoV-2 lineages and compensate for their comparatively low genetic diversity (46). To achieve a representative genomic sample, we used only samples from populationlevel testing rather than those from specific outbreak investigations. However, this approach does not fully mitigate reduced representation from populations less likely to seek testing (47), and there is some geographic variation in the proportion of cases sequenced (fig. S15). Greater London consistently has a higher sampling proportion than other regions throughout the study timeframe. Although sampling biases cannot be wholly eliminated, the selection procedure used here, and our cross-validation between independent data sources (human mobility and SGTF datasets), help to ensure that our conclusions are robust. As SARS-CoV-2 genome sequencing efforts are sciencemag.org SCIENCE

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

accelerated worldwide, careful consideration and communication of sampling frameworks are needed to facilitate downstream epidemiological analyses (48). Spatial heterogeneity at the within-city scale was not accounted for in our analysis, consideration of which may further refine our understanding of the mechanisms of lineage emergence and invasion. Coordinated and unified systems of genomic surveillance are needed worldwide to identify, track, and mitigate the transmission of SARS-CoV-2 variants of concern, including mechanisms to pair virus genomic and contact tracing data. Continuing rises in global incidence will increase the rate generation of viral genetic variation, and the accrual of higher levels of population immunity will create new selective pressures (49), the effects of which on virus evolution are difficult to predict (50–52). It is therefore critical to rapidly and accurately disentangle the contributions of genetic and ecological factors to the emergence of new SARS-CoV-2 variants. Geographic variation in vaccine availability, uptake, and delivery is expected to further contribute to variability in COVID-19 burden and the differential risk of disease resurgence (17, 53, 54), which can be mitigated through increased global access to vaccination and continued transmission control measures (52). Importation of SARS-CoV-2 lineages and variants from areas of high incidence will continue to pose a risk to those regions that are reducing NPIs after having controlled infection. RE FE RENCES AND N OT ES

1. M. Chand et al., Investigation of novel SARS-CoV-2 variant Variant of Concern 202012/01 (2021); https://assets. publishing.service.gov.uk/government/uploads/system/ uploads/attachment_data/file/952490/Variant_of_Concern_ VOC_202012_01_Technical_Briefing_4_England.pdf. 2. COVID-19 Genomics Consortium UK (CoG-UK), Preliminary genomic characterisation of an emergent SARS-CoV-2 lineage in the UK defined by a novel set of spike mutations (2020); https://virological.org/t/563. 3. T. Alpert et al., Cell 184, 2595–2604.e13 (2021). 4. Á. O’Toole et al., Wellcome Open Res. 6, 121 (2021). 5. N. L. Washington et al., Cell 184, 2587–2594.e7 (2021). 6. H. Vöhringer et al., on behalf of The COVID-19 Genomics UK (COG-UK) consortium Lineage-specific growth of SARS-CoV-2 B.1.1.7 during the English national lockdown (2020); https://virological.org/t/lineage-specific-growth-of-sars-cov-2b-1-1-7-during-the-english-national-lockdown/575. 7. E. Volz et al., Nature 593, 266–269 (2021). 8. T. C. Jones et al., Science 373, eabi5273 (2021). 9. N. G. Davies et al., Science 372, eabg3055 (2021). 10. R. Challen et al., BMJ 372, n579 (2021). 11. Scientific Advisory Group for Emergencies, NERVTAG: Update note on B.1.1.7 severity, 11 February 2021 (2021); www.gov.uk/ government/publications/nervtag-update-note-on-b117severity-11-february-2021. 12. N. G. Davies et al., Nature 593, 270–274 (2021). 13. D. Frampton et al., Lancet Infect. Dis. S1473-3099(21)00170-5 (2021). 14. A. Rambaut et al., Nature 453, 615–619 (2008). 15. C. Viboud et al., Science 312, 447–451 (2006). 16. L. du Plessis et al., Science 371, 708–712 (2021). 17. R. M. May, R. M. Anderson, Math. Biosci. 72, 83–111 (1984). 18. D. J. Watts, R. Muhamad, D. C. Medina, P. S. Dodds, Proc. Natl. Acad. Sci. U.S.A. 102, 11157–11162 (2005).

SCIENCE sciencemag.org

19. C. J. E. Metcalf, C. V. Munayco, G. Chowell, B. T. Grenfell, O. N. Bjørnstad, J. R. Soc. Interface 8, 369–376 (2011). 20. B. D. Dalziel et al., Science 362, 75–79 (2018). 21. M. S. Y. Lau et al., Nat. Ecol. Evol. 4, 934–939 (2020). 22. G. Chowell, L. Sattenspiel, S. Bansal, C. Viboud, Phys. Life Rev. 18, 66–97 (2016). 23. D. S. Candido et al., Science 369, 1255–1260 (2020). 24. COVID-19 Genomics UK (COG-UK) consortiumcontact@ cogconsortium.uk, Lancet Microbe 1, e99–e100 (2020). 25. P. Lemey, A. Rambaut, J. J. Welch, M. A. Suchard, Mol. Biol. Evol. 27, 1877–1885 (2010). 26. O. G. Pybus et al., Proc. Natl. Acad. Sci. U.S.A. 109, 15066–15071 (2012). 27. S. Dellicour et al., Mol. Biol. Evol. 38, 1608–1613 (2021). 28. UK Department of Health and Social Care, COVID-19 testing data: methodology note; www.gov.uk/government/ publications/coronavirus-covid-19-testing-data-methodology/ covid-19-testing-data-methodology-note. 29. A. Kalkauskas et al., PLOS Comput. Biol. 17, e1008561 (2021). 30. S. Francis, Covid: Christmas Tier-4 heartbreak for Londoners. BBC 19 December 2020; www.bbc.co.uk/news/uk-englandlondon-55380644. 31. Public Health England, Official UK Coronavirus Dashboard; https://coronavirus.data.gov.uk/details/testing?areaType= nation%26areaName=England. 32. J. Bahl et al., Proc. Natl. Acad. Sci. U.S.A. 108, 19359–19364 (2011). 33. G. Dudas et al., Nature 544, 309–315 (2017). 34. Public Health England, Investigation of novel SARS-CoV-2 variants of concern (2020); www.gov.uk/government/ publications/investigation-of-novel-sars-cov-2-variant-variantof-concern-20201201. 35. S. A. Kemp et al., bioRxiv [Preprint] 15 December 2020). doi:10.1101/2020.12.14.422555. 36. S. W. Park et al., bioRxiv [Preprint] 5 May 2021. doi:10.1101/2021.05.03.21256545. 37. W. S. Hart et al., medRxiv [Preprint] 30 May 2021. doi:10.1101/ 2021.05.27.21257936. 38. M. Chinazzi et al., Science 368, 395–400 (2020). 39. M. Worobey et al., Science 370, 564–570 (2020). 40. A. S. Gonzalez-Reiche et al., Science 369, 297–301 (2020). 41. S. Riley et al., bioRxiv [Preprint] 22 January 2021. doi:10.1101/2021.01.20.21250158. 42. N. W. Ruktanonchai et al., Science 369, 1465–1470 (2020). 43. P. Lemey et al., Nature 595, 713–717 (2021). 44. E. Volz et al., Cell 184, 64–75.e11 (2021). 45. E. B. Hodcroft et al., Nature 595, 707–712 (2021). 46. J. Lu et al., Cell 181, 997–1003.e9 (2020). 47. Public Health England, COVID-19: Review of disparities in risks and outcomes (2020); www.gov.uk/government/publications/ covid-19-review-of-disparities-in-risks-and-outcomes. 48. M. U. G. Kraemer et al., Epidemiol. Infect. 147, 1–7 (2018). 49. N. R. Faria et al., Science 372, 815–821 (2021). 50. C. M. Saad-Roy et al., Science 372, 363–370 (2021). 51. B. T. Grenfell et al., Science 303, 327–332 (2004). 52. R. N. Thompson, E. M. Hill, J. R. Gog, Lancet Infect. Dis. 21, 913–914 (2021). 53. B. L. Rice et al., Nat. Med. 27, 447–453 (2021). 54. N. B. Masters et al., Proc. Natl. Acad. Sci. U.S.A. 117, 28506–28514 (2020). 55. V. Hill, COG-UK/B.1.1.7_spatial_analysis_UK: V1.0.0. Zenodo (2021); doi:10.5281/zenodo.5085521. 56. NHS Foundation Trust, Tier 4 Restrictions for London, Bedfordshire and Luton; www.elft.nhs.uk/News/Tier-4Restrictions-for-London-Bedfordshire-and-Luton. ACKN OWLED GMEN TS

We thank all involved in the collection and processing of SARS-CoV-2 testing and genomic data. We also thank Public Health England (PHE) for making anonymized epidemiological data available for this analysis. We thank the Office of National Statistics (ONS) for their effort to publish the Coronavirus (COVID-19) Infection Surveys in real time. Funding: V.H. was supported by the Biotechnology and Biological Sciences Research Council (BBSRC) (grant BB/M010996/1). A.R. acknowledges the support of the Wellcome Trust (Collaborators Award 206298/Z/17/Z–ARTIC network) and the European Research Council (grant agreement 725422–ReservoirDOCS). M.U.G.K. acknowledges support from the Branco Weiss Fellowship. M.U.G.K. and S.D. acknowledge support from the European Union's Horizon 2020 project

MOOD (grant agreement 874850). O.G.P. and M.U.G.K. acknowledge support from the Oxford Martin School. A.L.B., S.V.S., and M.U.G.K. acknowledge support from the Rockefeller Foundation and Google.org. C.R. was supported by a Fondation Botnar Research Award (Programme grant 6063) and UK Cystic Fibrosis Trust (Innovation Hub Award 001). A.L.B. acknowledges support from the Biotechnologyand Biological Sciences Research Council (BBSRC) [grant BB/M011224/1]. S.D. acknowledges support from the Fonds National de la Recherche Scientifique (FNRS; Belgium). G.B. acknowledges support from the Research Foundation–Flanders (Fonds voor Wetenschappelijk Onderzoek–Vlaanderen, G0E1420N and G098321N) and from the Interne Fondsen KU Leuven/Internal Funds KU Leuven under grant agreement C14/18/094. COG-UK is supported by funding from the Medical Research Council (MRC) part of UK Research and Innovation (UKRI), the National Institute of Health Research (NIHR), and Genome Research Limited, operating as the Wellcome Sanger Institute. A.O. is supported by the Wellcome Trust Hosts, Pathogens and Global Health Programme (grant grant.203783/Z/16/Z) and Fast Grants (award 2236). S.B. is supported by the Clarendon Scholarship, University of Oxford and NERC DTP (grant NE/S007474/1). N.R.F. acknowledges support from Wellcome Trust and Royal Society (Sir Henry Dale Fellowship: 204311/Z/16/Z) and Medical Research Council–São Paulo Research Foundation CADDE partnership award (MR/ S0195/1 and FAPESP 18/14389-0). The contents of this publication are the sole responsibility of the authors and do not necessarily reflect the views of the European Commission or any of the other funders. Author contributions: M.U.G.K., A.R., V.H., C.R., S.D., S.V.S., and O.G.P. conceived and planned the research. M.U.G.K., V.H., A.R., C.R., S.D., S.B., G.B., B.K., A.L.B., S.D., S.G., and S.V.S. analyzed the data. M.U.G.K. and O.G.P. wrote the first draft. All authors contributed to writing and interpreting the results. M.U.G.K., A.R., S.V.S., and O.G.P. jointly supervised this work. Competing interests: O.G.P., A.R., and A.O. have undertaken consulting for AstraZeneca relating to the genetic diversity and classification of SARS-CoV-2 lineages. S.V.S. is a paid consultant with Pandefense Advisory and Booz Allen Hamilton; is on the advisory board for BioFire Diagnostics Trend Surveillance, which includes paid consulting; and holds unexercised options in Iliad Biotechnologies. These entities provided no financial support associated with this research; did not have a role in the design of this study; and did not have any role during its execution, analyses, interpretation of the data, and/or decision to submit. Data and materials availability: Aggregated epidemiological data used in this study are available from https://coronavirus.data.gov.uk/details/download. SARS-CoV-2 infection survey data are available from the Office of National Statistics (ONS) and available at www.ons.gov.uk/ peoplepopulationandcommunity/healthandsocialcare/ conditionsanddiseases/datasets/coronaviruscovid19infectionsurveydata. Raw epidemiological SARS-CoV-2 line list data are available from Public Health England (PHE) and aggregated statistics are available via Github. All genomes, phylogenetic trees, and basic metadata are available at the COG-UK Consortium website (www.cogconsortium.uk/ data). The O2 aggregated, anonymized mobile data insights dataset is not publicly available owing to stringent licensing agreements. Information on the process of requesting access to the O2 aggregated mobile data insights dataset is available at [email protected]. The Google COVID-19 Aggregated Mobility Research Dataset is not publicly available owing to stringent licensing agreements. Information on the process of requesting access to the Google mobility data are available from [email protected]. Code and data are available on the following GitHub repository https://github.com/COG-UK/ B.1.1.7_spatial_analysis_UK and permanently on Zenodo (55). This work is licensed under a Creative Commons Attribution 4.0 International (CC BY 4.0) license, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. To view a copy of this license, visit https:// creativecommons.org/licenses/by/4.0/. This license does not apply to figures/photos/artwork or other content included in the article that is credited to a third party; obtain authorization from the rights holder before using such material.

SUPPLEMENTARY MATERIALS

science.sciencemag.org/content/373/6557/889/suppl/DC1 Materials and Methods Figs. S1 to S19 Tables S1 to S3 COVID-19 Genomics UK (CoG-UK) Consortium Author List References (57–72) MDAR Reproducibility Checklist 16 April 2021; accepted 12 July 2021 Published online 22 July 2021 10.1126/science.abj0113

20 AUGUST 2021 • VOL 373 ISSUE 6557

895

RES EARCH

REPORTS

◥

HYDRAULIC FRACTURING

Large-sample evidence on the impact of unconventional oil and gas development on surface waters Pietro Bonetti1, Christian Leuz2*, Giovanna Michelon3 The impact of unconventional oil and gas development on water quality is a major environmental concern. We built a large geocoded database that combines surface water measurements with horizontally drilled wells stimulated by hydraulic fracturing (HF) for several shales to examine whether temporal and spatial well variation is associated with anomalous salt concentrations in United States watersheds. We analyzed four ions that could indicate water impact from unconventional development. We found very small concentration increases associated with new HF wells for barium, chloride, and strontium but not bromide. All ions showed larger, but still small-in-magnitude, increases 91 to 180 days after well spudding. Our estimates were most pronounced for wells with larger amounts of produced water, wells located over high-salinity formations, and wells closer and likely upstream from water monitors.

T

he rise of shale gas and tight oil development has triggered a major public debate about such unconventional development, in which horizontal drilling is combined with hydraulic fracturing (HF). HF is the high-pressure injection of water mixed with chemical additives and propping agents such as sand to create fractures in lowpermeability formations, allowing oil or gas to flow. Additives in the HF fluids vary with the geological characteristics of the formation and by operator, but the fluid mix usually contains friction reducers, surfactants, scale inhibitors, biocides, gelling agents, gel breakers, and inorganic acid (1, 2). HF wells produce large amounts of wastewater, which initially consists of flowback of HF fluids but over time increasingly consists of produced water from deep formations. The latter brine is naturally occurring water, into which organic and inorganic constituents from the formation have dissolved, resulting in high salt concentrations (1, 3–5). Although unconventional oil and gas (O&G) development has been important for energy production (6), we do not fully understand the associated environmental and social risks (7–11). These risks include hydrocarbon emissions, water usage, and pollution, along with potential human and ecological health consequences (7, 10, 12–16). Among these, the impact of unconventional O&G development and HF on water quality remains a key concern 1

IESE Business School, University of Navarra, 21 Avenida Pearson, 08034 Barcelona, Spain. 2Booth School of Business, University of Chicago, and the National Bureau of Economic Research, 5807 South Woodlawn Avenue, Chicago, IL 60637, USA. 3School of Accounting and Finance, University of Bristol, 15-19 TyndallÕs Park Road, Bristol BS8 1PQ, UK. *Corresponding author. Email: [email protected]

896

20 AUGUST 2021 • VOL 373 ISSUE 6557

(2, 8, 9, 17–20). In the US, one reason for this concern is that HF is exempt from the Underground Injection Control provisions of the Safe Drinking Water Act (8). Around the world, unconventional drilling has either just been introduced, or is being considered, by many countries, with uncertain effects on water quality (16). The US Environmental Protection Agency (EPA) reviewed and synthetized scientific evidence concerning the impact of HF on US water resources. The final report concluded that HF activities can affect drinking water resources under some circumstances (17), but the report did not identify widespread evidence of contamination. Groundwater studies primarily examine contamination from stray gas or deep formation brines, which could occur because of cementing or casing failures or because of brine migration to shallow aquifers through faults or other preexisting pathways (2, 21, 22). Instances of stray gas contamination have been found in Pennsylvania in connection with shale gas development of the Marcellus Shale (23–29) but not in Arkansas for the Fayetteville Shale (30). Geochemical evidence of gas contamination has also been documented for the Barnett Shale in Texas (24) and the Denver-Julesburg basin in Colorado (31). Studies of brine migration from deep formations, mostly in northeastern Pennsylvania, have provided mixed evidence (23, 32, 33). No evidence of brine contamination of groundwater has been documented for the Fayetteville Shale (30). For Pennsylvania, increases in shale gas–related contaminants have been documented at groundwater intake locations of community water systems that are in close proximity to shale gas wells (18). For surface water, the evidence is more limited. Instances of contamination have been as-

cribed primarily to discharges of inadequately treated wastewater, HF fluid leaks, and spills and other mishandling of flowback and produced waters (2, 4, 8, 17, 20, 34, 35). Specifically, increased chloride and bromide concentrations downstream of effluents from wastewater treatment plants have been found in western Pennsylvania up to 2011, when the release of wastewaters from unconventional wells into streams through municipal wastewater treatment plants was not prohibited (2, 4, 36). Further, a high frequency of brine spills in North Dakota has resulted in elevated levels of salts and other contaminants in surface waters up to 4 years after the spills occurred (37). A large-sample statistical examination of the effects of shale gas development activities on surface water in Pennsylvania found higher chloride concentrations downstream of wastewater treatment facilities and an association between gas well density in a watershed and increased total suspended solid (but not chloride) concentrations during the period 2000–2011 (38). The authors of this study suggest that both insufficient wastewater treatment and building infrastructure for unconventional O&G extraction could explain the results. Evidence also exists for barium concentrations in Pennsylvania being higher in areas with unconventional wells than in areas without them, but the authors of this study point out that this evidence cannot be solely ascribed to unconventional wells, as it could also reflect the presence of basin brines or a sulfate decrease in acid rains (39). In sum, prior studies document localized instances of surface water contamination related to unconventional O&G development, with spills and leaks being the most common pathway (20). We investigated the potential impact of unconventional O&G development on surface water quality using a large-sample statistical approach. We combined a geocoded database of 46,479 HF wells from 24 shales with 60,783 surface water measurements over 11 years (2006–2016) across 408 watersheds (HUC10s; HUC, hydrologic unit code) with HF activity (Fig. 1, fig. S2, and tables S1 to S3). Our analysis focuses on concentrations of bromide (Br−), chloride (Cl−), barium (Ba), and strontium (Sr) in watersheds exposed to unconventional O&G development. We chose these four ions for the following reasons. First, they are usually found in high concentrations in flowback and produced water from HF wells and hence could indicate surface water impact, if and when it exists (1, 2, 4, 5, 8, 20, 27, 33, 36, 39, 40). Moreover, unlike some organic components of HF fluids, these four ions do not experience biodegradation, and their presence has been measured several years after HF spill events (20, 37). Thus, an analysis of ions is a likely mode of detection (2, 4). Second, many of the water quality concerns associated with HF wastewaters sciencemag.org SCIENCE

RES EARCH | R E P O R T S

Fig. 1. US watersheds with HF well exposure. Location of sample watersheds (HUC10s) with unconventional O&G development (shaded in ocher) and, superimposed, the distribution of HF wells (red triangles). Data on the location of wells come from WellDatabase, Enverus, the Pennsylvania Department of

are related to the chemistry of the deep formation brines as well as the salinity in flowback and produced water (8). Third, the four ions are measured in many watersheds with reasonable frequency, which is not the case for other potential signatures of HF wastewater. As high salt concentrations can also occur in surface waters for many natural and anthropogenic reasons, such as brine migration or road deicing (2, 4, 33, 39), sufficient data is needed to estimate reliable, local, and time-varying baselines for the background ion concentrations. We construct these baselines with regression analysis and then exploit temporal and spatial variation in the spudding of HF wells within and across US watersheds to identify anomalous changes in ion concentrations associated with newly spudded HF wells in the same watersheds. Our regression model includes temperature and precipitation control variables as well as an extensive set of fixed effects that construct local and time-varying baselines for background ion concentrations (41, 42). Specifically, our model allows for arbitrary monthly variation in the average background ion concentrations across subbasins (HUC8) and within a given subbasin over time. This flexible regional baseline controls for subbasin differences in water quality, geochemistry, salinity, climate, water body types, or economic activity and also for over-time changes in subbasin concentrations due to seasons, weather patterns and related road deicing, salinization trends, or economic development. Our model also has a local baseline, using each water monitoring station as SCIENCE sciencemag.org

Environmental Protection, and the Pennsylvania Department of Conservation of Natural Resources. Thin black lines outline HUC10 boundaries; thick black lines depict state boundaries. Maps with a close-up of the main US shales and location of water quality monitoring stations are shown in fig. S2.

its own control, which accounts for arbitrary differences in average local ion concentrations. Our model combines these two baselines and the weather control variables to estimate the association between anomalous concentration changes and new HF wells in the same watersheds (42). Our model explains >80% (in many cases, >90%) of the variation in ion concentrations across watersheds and through time (table S4), suggesting that the model estimates precise baselines for background ion concentrations. We estimated the association between newly spudded HF wells and ion concentrations at the watershed level using a variable that counts the number of HF wells in a watershed at a given point in time (#wellsHUC10). We estimated our regression model for all US watersheds with HF wells and then separately for Pennsylvania, because Pennsylvania accounts for almost 41% of the sample. We found a robust association between new HF wells in a watershed and elevated ion concentrations in its surface waters (Fig. 2 and table S4). For watersheds in Pennsylvania (PA), the coefficients on #wellsHUC10 are positive for all ions and significant for three of them (table S4, column 2, HUC8 model; Br−: 0.00019, P = 0.865; Cl−: 0.00071, P = 0.031; Ba: 0.00038, P = 0.086; Sr: 0.00041, P < 0.001). The lack of significance for Br− could reflect measurement issues (42). For watersheds throughout the US (ALL), the coefficients on #wellsHUC10 are generally comparable to those for Pennsylvania in terms of magnitude and significance, except for Ba, which is rarely measured outside of

Pennsylvania and for which results are weaker (Fig. 2). To gauge the magnitude of the estimated effects, we multiply each coefficient by the respective sample mean ion concentration and the average number of wells per watershed to obtain the ion concentration increase in the average HUC10 implied by our estimation (HUC10 impact in Fig. 2). With this approach, and focusing on coefficients from the HUC8 specification, we estimated an average increase of Cl− by 1322.44 mg/liter for PA and 2232.55 mg/ liter for ALL; Ba by 1.61 mg/liter for PA; and Sr by 5.19 mg/liter for PA and 8.88 mg/liter for ALL (Fig. 2). These magnitudes are very small but need to be interpreted in the context of our analysis. First, the #wellsHUC10 coefficient is by construction a per-well estimate, but this does not imply that each well is associated with a concentration increase, rather it is an average over all wells. Second, by using measurements from all monitors in a given watershed, the estimated well–ion association reflects the average exposure of monitors in the watershed. However, some monitors could be very far away or upstream from wells, in which case they should not be exposed or affected, thus lowering the average. For these reasons, the impact estimates in Fig. 2 are expected to be small; they should increase when the analysis focuses on the most relevant water measurements, as reported below. We found robust results in different sensitivity analyses (table S5). (i) Estimating separate #wellsHUC10 coefficients for watersheds in and outside of Pennsylvania shows that the 20 AUGUST 2021 • VOL 373 ISSUE 6557

897

RES EARCH | R E P O R T S

HUC10 Impact Ions Bromide

Chloride

Barium

Strontium

Mean Concentration

Cum. # Wells

HUC4 Model

HUC8 Model

PA

134.31

112.00

11.14

2.86

ALL

97.38

166.61

5.68

3.25

PA

19,772.05

94.17

1,098.86

1,322.44

ALL

47,522.55

83.87

2,152.80

2,232.55

PA

39.66

106.49

2.15

1.61

ALL

59.37

74.91

0.98

1.02

PA

121.52

104.07

5.31

5.19

ALL

330.17

74.66

10.11

8.88

(HUC4) (HUC8) −1.5 −1 −0.5 0 0.5 1 1.5 Coefficient estimates x 103

Fig. 2. HF wells and water quality. Ordinary least squares (OLS) coefficients and confidence intervals plotted for the associations between ion concentrations and cumulative HF well counts (#wellsHUC10), estimated using eq. S1 and two different model specifications, HUC4 and HUC8 (table S4). We report results for treated watersheds (HUC10s) in Pennsylvania (PA) and for all treated US watersheds (ALL). The last two columns [HUC10 Impact (mg/liter)] report the cumulative impact in the average watershed implied by the coefficient estimates,

findings for Pennsylvania and the other US states are similar. (ii) Estimating separate effects for different time periods shows similar results over time. The coefficients in later periods tend to be smaller but higher in significance, presumably because the frequency of water measurements increases over time. (iii) Our results are similar when the model is estimated over all watersheds within a subregion (including those without HF wells), albeit in some cases slightly weaker, possibly because this specification uses less relevant baselines for the background ion concentrations. (iv) Adding further controls for snow (to account for related discharge of road salts) or for within-watershed seasonality does not alter the findings, suggesting that the monthly subbasin baselines already control for local weather patterns. (v) Our results are also robust to alternative modeling choices, for example, scaling the cumulative number of wells by watershed size, as in (38); estimating the model with weighted least squares (WLS) to give more weight to observations, for which we have more readings to estimate the monthly subbasin baselines; and using alternative transformations to address skewness in ion concentrations. We also investigated whether the results could be driven by HF-related patterns in the frequency of water monitoring (e.g., more measurements shortly after spud dates or closer to wells). However, we found no evidence that water measurement is systematically related 898

g liter

Sample

20 AUGUST 2021 ¥ VOL 373 ISSUE 6557

2

obtained by multiplying the respective coefficient with the sample mean ion concentration and the cumulative number of wells (Cum. # Wells) in the average HUC10 over the sample period. Bold impact numbers are based on significant coefficients. The EPA maximum contaminant level (MCL) is 250,000 mg/liter for Cl− and 2000 mg/liter for Ba. The EPA does not provide a MCL for Br− or Sr. Health advisory levels for 1-day and lifetime exposure to Sr are 25,000 mg/liter and 4000 mg/liter, respectively.

to new wells in a watershed (table S6). We analyzed three other water quality proxies (dissolved oxygen, phosphorus, and fecal coliforms) that are frequently measured but not as indicative of HF-related impacts. Concentration levels of these proxies could be related to other economic activities with water impacts, such as agriculture (43). We used this analysis to gauge how well our model controls for economic activity and other potential confounds. The estimated #wellsHUC10 coefficients for the three analytes are not different from zero (table S7), which contrasts with the results for the ion concentrations we chose for the main analysis (table S4). Up to this point, our analysis estimated the long-run association between HF wells and ion concentrations, because we did not restrict the sample and the estimation to a particular period after well spudding. However, concentration increases could be stronger early on and fade over time. Hence, we estimated the association in specific time windows around a new well spud date, allowing us to map out the estimates through time. For this temporal analysis, we modified eq. S1 (see supplementary materials) by replacing #wellsHUC10 with several time-specific well counts, defined for the following time windows measured in days: [−180, −91], [−90, 0], [1, 90], [91, 180], [181, 360], and >360. The coefficients were estimated relative to measurements collected 180 days or more before a new well spudding (table S8).

We found increases in ion concentrations 91 to 180 days after the spud date, consistently for all four ions, in Pennsylvania and all US watersheds. In Fig. 3, we plotted the coefficients, estimated over all watersheds, for each window together with the 95% confidence interval. For the [91, 180] window, the well count coefficients are significant for all four ions (Fig. 3 and table S8, panel B; Br−: 0.01095, P = 0.036; Cl−: 0.00401, P = 0.022; Ba: 0.00347, P = 0.017; Sr: 0.00289, P = 0.015). These ion concentration increases in the [91, 180] window are at least one order of magnitude larger than the long-run estimates in table S4 (shown as a red dot in Fig. 3 for comparison). The coefficients for the [91, 180] window (Fig. 3) correspond to an average (short-run) increase of 178.64 mg/liter for Br, 16,014.30 mg/liter for Cl−, 15.46 mg/liter for Ba, and 71.34 mg/liter for Sr per watershed with HF wells. Water measurement is often sparse and, as shown earlier, does not increase around the well spud dates. Thus, most measurements naturally fall into the benchmark period of 360] coefficients always have the tightest confidence intervals. These coefficients are, as expected, comparable to the long-run estimates from table S4 (red dots in Fig. 3). For all other coefficients, we have far fewer observations, and hence their confidence intervals are wider. sciencemag.org SCIENCE

RES EARCH | R E P O R T S

SCIENCE sciencemag.org

A 0.02

0.015

Bromide (µg/liter)

0.01

0.005

0

-0.005

-0.01

-0.015 360

Avg. Est.

[181, 360]

>360

Avg. Est.

[181, 360]

>360

Avg. Est.

[181, 360]

>360

Avg. Est.

Time window

B 0.02

0.015

Chloride (µg/liter)

0.01

0.005

0

-0.005

-0.01

-0.015 10 kilohms). Hence, our experiment can be considered voltage controlled before filament formation, and it becomes current limited afterward—a necessary condition to avoid damage. We use an oscilloscope to simultaneously record reflectivity and current. To increase signal-to-noise ratio, averaging over 100 cycles is performed for each reflectivity measurement. The voltage pulse width is a few milliseconds to characterize the device’s response over several decades in time, and the separation between cycles is kept at 1 s to allow for complete cooling down and relaxation to the initial insulating state (11). The top panel in Fig. 1C shows the current versus time when a 24-V step is applied at time t = 0. After an incubation time of ~300 ns, a filament is formed, and the current quickly increases. The bottom panel shows the normalized reflectivity in the center of the gap (red) and 10 mm away (blue). The time axis is on a logarithmic scale. Comparing the curves, we can conclude that metallization happens fast in the center on a time scale of ~10−7 s and then expands at a much slower rate to the final (stationary) filament configuration. This example illustrates how this setup offers a distinct opportunity to observe the resistive switching dynamics, allowing us to form a complete picture of the process. By repeating this measurement every 3 mm, it is possible to capture, with space and time resolution, the dynamical evolution of the fielddriven IMT. This can be seen in Fig. 2A. The color scale indicates the metallic fraction, the sciencemag.org SCIENCE

RES EARCH | R E P O R T S

Fig. 2. Nucleation dynamics of the field-driven IMT. (A) (Top) Current versus time in a VO2 device when the field-driven IMT is triggered by a 12-V step applied at t = 0. Base temperature is T = 335 K. The vertical dashed line shows the moment in which the filament percolates. (Bottom) Space- and time-resolved reflectivity, recorded at the same time as in the top panel. The reflectivity is shown on a logarithmic color scale. The horizontal axis is time in logarithmic scale, and the vertical axis corresponds to the coordinate along which the spot was scanned. The scanning direction is perpendicular to the current direction.

horizontal axis is the time on a logarithmic scale, and the vertical axis is a scan across the direction perpendicular to the current-filament direction. This can be interpreted as a picture of the filament width as a function of time. The dashed white lines mark the width of the electrodes. The current versus time is plotted in the top panel for comparison. There is an ~30-ms incubation time (tinc) between the initial voltage application and the filament formation, clearly identified as a sudden jump in the reflectivity map. tinc is larger than that shown in Fig. 1C because we are applying 12 V instead of 24 V (32). This localized reflectivity jump caused by the emergence of a metallic filament is observed for the three oxides we studied. Notably, subtle changes in rnorm can be observed long before the filament is formed, which points either to a partial metallization of the system or to a temperature increase within SCIENCE sciencemag.org

Dashed white lines indicate the electrode width. The emergence of a filament close to 30 ms is readily visible. (B) Zoomed-in reflectivity maps for VO2, for three different applied voltages: 10 V, 11 V, and 12 V. The time scale has been zoomed into the first 10 ms, and the color scale has been amplified to better appreciate the formation of hotspots before the appearance of the filament. T = 335 K. Full-scale data are shown in fig. S3 (31). (C) Incubation time (tInc) as a function of the applied voltage, at different temperatures. The three panels correspond to the three oxides studied here: V3O5, VO2, and V2O3.

the insulating state. These changes are distributed nonuniformly, indicating the presence of nucleation points that slowly grow in size and eventually trigger the formation of a filament. These results suggest the following qualitative scenario. As the electric field is applied, the current flows inhomogeneously, concentrated in intrinsic defects or inhomogeneities of the film. Defects tend to partially suppress the IMT and lower the film resistivity (33, 34), which helps in focusing the current. As Joule heating concentrates in these hotpots, the temperature increases locally, further metallizing them and concentrating the current even more. A positive feedback loop is established, leading to an instability and ultimately filament formation. For a simple, single-element system, the thermal dynamics will follow @T =@t ¼ E 2 =rðT Þ

k
ðT

T0 Þ

ð1Þ

where E is the electric field, r(T) is the resistivity, T0 is the substrate temperature, and k is the thermal coupling constant between the film and the substrate. Equation 1 highlights the applied voltage and the temperature dependence of the resistivity as the key factors controlling the feedback loop, and changes in any of these parameters are expected to greatly affect the nucleation dynamics. This can be seen experimentally in Fig. 2B, which shows an amplified picture of the nucleation process (before percolation of the filament) for three slightly different applied voltages: 10 V, 11 V, and 12 V. The color scale has been rescaled to better visualize the small reflectivity changes associated with prefilament dynamics [full scale plots shown in fig. S3 (31)]. Comparing the top and bottom panels, it can be observed that even small voltage variations can result in big differences in the nucleation 20 AUGUST 2021 ¥ VOL 373 ISSUE 6557

909

RES EARCH | R E P O R T S

Fig. 3. Dynamics of filament expansion and role of the resistance ratio rIns/rMet. (A) Space- and time-resolved reflectivity during the field-driven IMT. The reflectivity is coded in color scale. The horizontal axis is time on a logarithmic scale, and the y axis corresponds to the coordinate along which the spot was scanned, which is perpendicular to the current direction. Dashed white lines indicate the electrode width. Time is set to zero when the filament percolates, in contrast to Fig. 2, in which it was set to zero when the voltage was applied. The three panels correspond to the three materials used in this work. The measurement conditions were: V = 30 V and T = 328 K for V3O5; V = 20 V and T = 330 K for VO2; and V = 32 V and T = 134 K for V2O3. The inset in the bottom panel shows a zoomed-in picture of the first moments of filament expansion in V2O3, and the time axis is in linear scale. (B) Simulated 2D temperature maps just after filament percolation. The x axis is parallel to the current direction, and the green bars mark the position of the electrodes. Base temperature in the simulation was 0.88 TIMT, where TIMT is the transition temperature. The top panel corresponds to a resistivity ratio rins/rmet = 103, and the bottom panel corresponds to rins/rmet = 2.7 × 105. Because no defects are present in the simulation, the filaments originate at the electrode corners, where the electric field is the largest.

process. Higher voltages enhance inhomogeneous metallization, focusing the current into smaller hotspots. This accelerates the nucleation dynamics, markedly reducing the incubation time (tinc). Figure 2C shows tinc as a function of the applied voltage V, for three oxides: V3O5, VO2, and V2O3. In all cases, subtle voltage variations can change tinc by several orders of magnitude. However, there are very noticeable differences between the three systems. The tinc sensitivity to voltage changes is relatively low for V3O5, higher for VO2, and very high for V2O3. In the V2O3 case, it approaches an all-or-nothing behavior, with tinc decreasing from infinity to a few microseconds with a 105 for V2O3 (fig. S1). To investigate how rins/rmet affects the nucleation and growth dynamics, we modeled our device as a resistor network [see further discussion and fig. S9 in (31)]. Each node in the network can be either insulating or metallic, depending on the local temperature through a Landau-type free energy functional that mimics a first-order phase transition (11, 35), similar to the IMT present in VO2 and V2O3. This functional depends only on the temperature, without any direct contribution from the electric field. This implies that in our simulations, the IMT can only be triggered either by Joule heating or by increasing the overall temperature. At each simulation step, temperature, voltage, and current distributions are updated, providing information about the system dynamics. Figure 3B shows two-dimensional (2D) temperature maps just after the filament percolates for two different cases: rins/rmet = 103 and rins/rmet = 2.7 × 105. Because intrinsic defects are not included in the simulations, the filaments form at the corners of the electrodes, which are the points of maximum electric field. For larger rins/rmet, the current concentrates into a much narrower path, and the nucleation is more heterogeneous. Localized nucleation focuses Joule heating into smaller regions, amplifying the feedback loop and accelerating the nucleation dynamics. This sciencemag.org SCIENCE

RES EARCH | R E P O R T S

explains why tinc is much more sensitive to small voltage increments in V2O3 (rins/rmet ≈ 105) than in VO2 (rins/rmet ≈ 103). Once percolation takes place, the filament starts widening—a feature also captured by the simulations (fig. S10). The initial rate at which it grows is proportional to its temperature. As can be seen in Fig. 3B, the filament temperature is higher for larger rins/rmet, providing a qualitative explanation for the different expansion rates observed in Fig. 3A. The role of system inhomogeneity has been previously explored in (36), where a similar conclusion was reached, with higher rins leading to smaller metallic regions with the subsequent increase of local current density. According to our simulations, for large rins/ rmet, a very small volume of the sample controls the initial nucleation. This renders the metallization of the whole device sensitive to local fluctuations, markedly increasing stochasticity. This picture is supported by tinc error bars in Fig. 2C, which are barely visible for V3O5, are noticeable for VO2, and are very large for V2O3, indicating that stochastic behavior in filament formation increases with an increasing rins/rmet ratio. As a consequence, V2O3 is affected by intense cycle-to-cycle variation [see further discussion and fig. S11 in (31)]. VO2 and V3O5 do not show comparable cycle-to-cycle variations. Although our simulations do not consider defects, our samples feature some device-to-device variability, possibly owing to intrinsic disorder. Even though this leads to noticeable differences in the quasistatic voltage-current characteristics [fig. S5C (31)], reflectivity measurements show that filament nucleation and growth dynamics are very similar for different devices [fig. S12 (31)]. Although the rins/rmet ratio seems to account for most differences between the three oxides, we must also consider other possibilities. The three materials feature very diverse transition temperatures, which could lead to large changes in the parameters that control thermal dynamics. A larger effective thermal conductivity keff would naturally accelerate the switching process. Our films are grown on top of sapphire, which has a large k that increases with lowering temperature and could contribute to the differences between VO2 and V2O3. However, the vanadium films are one-tenth to onehundredth as thermally conductive as sapphire. This makes vertical conduction across the thin film, and not the substrate, the dominant term that determines heat conductance into the environment. According to finite element simulations in (37), in our films, temperature drops almost to the environment value before reaching the sapphire interface. Considering that the thermal conductivities for the three oxides are very similar at our measuring temperatures (0.030 W/K per centimeter at 330 K for V3O5, 0.045 W/K per centimeter at 330 K for VO2, and 0.035 W/K per centimeter at 130 K for SCIENCE sciencemag.org

V2O3) (38), we expect our three devices to have comparable keff values. So, although it is possible that variations in k could enhance the differences between these oxides, we still expect rins/rmet to be the key parameter. This is best appreciated when comparing the switching of VO2 and V3O5 at the same temperature, when the sapphire thermal conductivity is similar [fig. S13 (31)]—its low rins/rmet ratio makes insulating V3O5 very leaky and drastically reduces the sharpness of the transition between low- and high-current states. Our results show that growth and percolation of the metallic phase during the field-driven IMT can be explained just by considering the effect of Joule heating. However, this does not necessarily imply that Joule heating triggers the transition of the first metallic domains. Although this is likely the case for VO2 (39, 40), recent studies in V2O3 have shown that the IMT can be triggered directly by the electric field (12, 37), possibly by carrier injection into the conduction band and destabilization of the insulating phase (36, 41). Whether Joule heating or field effect are responsible for the transition of the first metallic domains does not affect the interpretation of the results shown here. Once a small portion of the sample undergoes the IMT, the current increases and Joule heating takes over, becoming the main driving force governing the nucleation and subsequent growth of the filament. The clear filament thickening after percolation observed for all three materials is a clear hallmark of this process, and it would not be expected if the electric field drove the growth dynamics. Notably, rins/rmet can account on its own for the quantitative differences between V3O5, VO2, and V2O3. Other factors, such as the presence of a coupled structural phase transition or whether the IMT is first or second order, do not seem to play a fundamental role. Our results unveil a complete picture of the field-induced IMT and identify the key parameters that control switching speed, which is vital for proper material selection and device design in emerging information technologies, such as optoelectronics and neuromorphic computing. RE FERENCES AND NOTES

1. D. N. Basov, R. D. Averitt, D. Hsieh, Nat. Mater. 16, 1077–1088 (2017). 2. M. Imada, A. Fujimori, Y. Tokura, Rev. Mod. Phys. 70, 1039–1263 (1998). 3. D. Lee et al., Science 362, 1037–1040 (2018). 4. S. Lupi et al., Nat. Commun. 1, 105 (2010). 5. J. H. Park et al., Nature 500, 431–434 (2013). 6. Y. Kalcheim et al., Phys. Rev. Lett. 122, 057601 (2019). 7. G. Stefanovich, A. Pergament, D. Stefanovich, J. Phys. Condens. Matter 12, 8837–8845 (2000). 8. Y. Zhou et al., IEEE Electron Device Lett. 34, 220–222 (2013). 9. P. Stoliar et al., Adv. Mater. 25, 3222–3226 (2013). 10. J. S. Brockman et al., Nat. Nanotechnol. 9, 453–458 (2014). 11. J. Del Valle et al., Nature 569, 388–392 (2019). 12. Y. Kalcheim et al., Nat. Commun. 11, 2985 (2020). 13. S. Kumar, J. P. Strachan, R. S. Williams, Nature 548, 318–321 (2017).

14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42.

43.

M. Liu et al., Nature 487, 345–348 (2012). N. A. Butakov et al., ACS Photonics 5, 4056–4060 (2018). P. Markov et al., ACS Photonics 2, 1175–1182 (2015). J. del Valle, J. G. Ramírez, M. J. Rozenberg, I. K. Schuller, J. Appl. Phys. 124, 211101 (2018). Y. Zhou, S. Ramanathan, Proc. IEEE 103, 1289–1310 (2015). M. D. Pickett, G. Medeiros-Ribeiro, R. S. Williams, Nat. Mater. 12, 114–117 (2013). J. Del Valle, P. Salev, Y. Kalcheim, I. K. Schuller, Sci. Rep. 10, 4292 (2020). W. Yi et al., Nat. Commun. 9, 4661 (2018). B.-J. Kim, Y. W. Lee, S. Choi, S. J. Yun, H.-T. Kim, IEEE Electron Device Lett. 31, 14–16 (2010). S. Wall et al., Science 362, 572–576 (2018). B. T. O’Callahan et al., Nat. Commun. 6, 6849 (2015). V. R. Morrison et al., Science 346, 445–448 (2014). A. Singer et al., Phys. Rev. Lett. 120, 207601 (2018). S. Guénon et al., EPL 101, 57003 (2013). H. Madan, M. Jerry, A. Pogrebnyakov, T. Mayer, S. Datta, ACS Nano 9, 2009–2017 (2015). S. Kumar et al., Adv. Mater. 25, 6128–6132 (2013). U. Schwingenschlögl, V. Eyert, Ann. Phys. 13, 475–510 (2004). Materials and methods are available as supplementary materials. G. Seo et al., IEEE Electron Device Lett. 32, 1582–1584 (2011). D. Wickramaratne, N. Bernstein, I. I. Mazin, Phys. Rev. B 99, 214103 (2019). Z. Shao, X. Cao, H. Luo, P. Jin, NPG Asia Mater. 10, 581–605 (2018). F. Tesler et al., Phys. Rev. Appl. 10, 054001 (2018). H.-T. Kim et al., New J. Phys. 6, 52 (2004). I. Valmianski et al., Phys. Rev. B 98, 195144 (2018). V. N. Andreev, F. A. Chudnovskii, A. V. Petrov, E. I. Terukov, Phys. Stat. Sol. 48, K153–K156 (1978). A. Zimmers et al., Phys. Rev. Lett. 110, 056601 (2013). D. Li et al., ACS Appl. Mater. Interfaces 8, 12908–12914 (2016). G. Mazza, A. Amaricci, M. Capone, M. Fabrizio, Phys. Rev. Lett. 117, 176401 (2016). J. del Valle et al., Data from Spatiotemporal characterization of the field-induced insulator-to-metal transition, Zenodo (2021); http://doi.org/10.5281/zenodo.4789471. fairfriend92/mrn_simulator_releases: Spatiotemporal characterization of the field-induced insulator-to-metal transition, version 1.1, Zenodo (2021); http://doi.org/10.5281/ zenodo.4813152.

AC KNOWLED GME NTS

The authors thank G. Kassabian and J. Trastoy for helpful discussions. Funding: This work was supported as part of the Quantum Materials for Energy Efficient Neuromorphic Computing (Q-MEEN-C) Energy Frontier Research Center (EFRC), funded by the US Department of Energy, Office of Science, Basic Energy Sciences, under award no. DE-SC0019273. Part of the fabrication process was done at the San Diego Nanotechnology Infrastructure (SDNI) of the University of California San Diego, a member of the National Nanotechnology Coordinated Infrastructure (NNCI), which is supported by the National Science Foundation under grant ECCS-1542148. J.d.V. thanks the Swiss National Science Foundation for an Ambizione Fellowship (no. PZ00P2_185848) that supported him while writing the manuscript. R.R. is supported by the project “MoMA” from the French ANR (no.19-CE30-0020). Y.K. acknowledges funding from the Norman Seiden Fellowship for Nanotechnology and Optoelectronics. Author contributions: J.d.V. and I.K.S. conceived the project; J.d.V., P.S., Y.K., C.A., and M.-H.L. fabricated the samples; J.d.V. and N.M.V. measured the devices with assistance from P.S. and P.N.L.; R.R., P.Y.W., L.F., and M.J.R. performed the numerical simulations. All authors participated in the discussion and interpretation of the results. J.d.V. wrote the manuscript with input and corrections from all authors. Competing interests: The authors declare no competing financial or nonfinancial interests. Data and materials availability: Experimental data (42) and simulation codes (43) are available at Zenodo, a CERN-operated public repository. SUPPLEMENTARY MATERIALS

science.sciencemag.org/content/373/6557/907/suppl/DC1 Materials and Methods Supplementary Text Figs. S1 to S13 References (44–48) 20 July 2020; accepted 8 July 2021 Published online 22 July 2021 10.1126/science.abd9088

20 AUGUST 2021 • VOL 373 ISSUE 6557

911

RES EARCH | R E P O R T S

METALLURGY

Hierarchical crack buffering triples ductility in eutectic herringbone high-entropy alloys Peijian Shi1, Runguang Li2, Yi Li1, Yuebo Wen1, Yunbo Zhong1*, Weili Ren1, Zhe Shen1, Tianxiang Zheng1, Jianchao Peng3, Xue Liang3, Pengfei Hu3, Na Min3, Yong Zhang2, Yang Ren4, Peter K. Liaw5, Dierk Raabe6*, Yan-Dong Wang2,7* In human-made malleable materials, microdamage such as cracking usually limits material lifetime. Some biological composites, such as bone, have hierarchical microstructures that tolerate cracks but cannot withstand high elongation. We demonstrate a directionally solidified eutectic high-entropy alloy (EHEA) that successfully reconciles crack tolerance and high elongation. The solidified alloy has a hierarchically organized herringbone structure that enables bionic-inspired hierarchical crack buffering. This effect guides stable, persistent crystallographic nucleation and growth of multiple microcracks in abundant poor-deformability microstructures. Hierarchical buffering by adjacent dynamic strainÐ hardened features helps the cracks to avoid catastrophic growth and percolation. Our self-buffering herringbone material yields an ultrahigh uniform tensile elongation (~50%), three times that of conventional nonbuffering EHEAs, without sacrificing strength.

C

racks occur in materials if loads cannot be fully dissipated by elastic-plastic work, exposing human lives to risk of failure and integrity loss of safety-critical components (1–6). Some hierarchical composites, such as high-toughness bone (3), feature excellent crack tolerance, but they usually cannot withstand high elongations because there are not enough conventional lattice defects to bear tensile deformation (1–3). In humanengineered formable materials, extensive cracks tend to trigger premature failure (6). These cracks are generally initiated from localized severe plastic deformation, and their propagation cannot be effectively buffered and arrested (5–8). This situation exists because the locally deformed microstructures are usually not characterized by sufficiently sustainable strain-hardening capability, which relieves locally high stresses at the propagating tips of cracks (1–4). So even though some ductile metallic composites exhibit crack tolerance, only limited additional tensile ductility can be achieved (5–11). We show that the conflict between extensive crack generation and high uniform elongation 1

State Key Laboratory of Advanced Special Steel, Shanghai Key Laboratory of Advanced Ferrometallurgy, School of Materials Science and Engineering, Shanghai University, Shanghai, China. 2Beijing Advanced Innovation Center for Materials Genome Engineering, State Key Laboratory for Advanced Metals and Materials, University of Science and Technology Beijing, Beijing, China. 3Laboratory for Microstructures, Shanghai University, Shanghai, China. 4 X-Ray Science Division, Advanced Photon Source, Argonne National Laboratory, Argonne, IL, USA. 5Department of Materials Science and Engineering, University of Tennessee, Knoxville, TN, USA. 6Department Microstructure Physics and Alloy Design, Max-Planck-Institut für Eisenforschung, Düsseldorf, Germany. 7Key Laboratory for Anisotropy and Texture of Materials (Ministry of Education), School of Material Science and Engineering, Northeastern University, Shenyang, China. *Corresponding author. Email: [email protected] (Y.B.Z.); [email protected] (D.R.); [email protected] (Y.-D.W.)

912

20 AUGUST 2021 • VOL 373 ISSUE 6557

can be resolved in eutectic high-entropy alloys (EHEAs). EHEAs are a family of recently developed multi–principal element lamellar composites (12–15). We demonstrate that in EHEAs a herringbone-like hierarchical eutectic microstructure design tends to generate a high density of cracks upon tensile deformation. However, the hierarchical crack buffering prevents the cracks from growing and percolating catastrophically across a huge straining range of ~25%. Consequently, such a high density of cracks is not detrimental to the elongation, but instead can serve as an effective strategy to compensate for the limited tensile ductility of the poor-deformability lamellae. This renders ultrahigh isostrain–forming conditions among adjacent lamellae with different deformabilities, thus achieving a surprisingly high uniform elongation of ~50%. This value is three times that of conventional EHEAs without this type of crack tolerance. We studied two as-cast Al19Fe20Co20Ni41 (at%) EHEAs (15) fabricated by conventional casting and directional solidification (DS), respectively. The conventionally cast EHEA served as the reference material and exhibited a typical lamellar microstructure (Fig. 1A) formed during a eutectic transformation (14). The structure comprises L12 (soft ordered facecentered cubic) and B2 (hard ordered bodycentered cubic) dual-phase lamellae with varying growth directions in different nearequiaxed grains (Fig. 1, B and C). The directionally solidified EHEA displays a directionally grown microstructure (Fig. 1D). It consists of columnar grains aligned along the DS direction (Fig. 1E). They contain aligned (grain center) and branched (rims of the grains) eutectic colonies, both of which comprise soft L12 and hard B2 lamellae with nanoindentation hardness of ~4.2 GPa and ~5.6 GPa, respectively (Fig. 1E). Lamellae consisting of aligned eu-

tectic colonies (AEC, accounting for ~33 vol%) generally grow along the DS direction, whereas lamellae comprising branched eutectic colonies (BEC, 67 vol%) are inclined at 30° to 60° to the DS direction and have a more branched morphology. With these features, the directionally solidified EHEA assumes a new type of hierarchically arranged herringbone microstructure (Fig. 1F). We show that this bone-like structure is formed by the directional growth of cellular solid-liquid interfaces along the DS direction (16–21). The lamellae of both colonies grow perpendicular to the cellular interfaces (Fig. 1I and fig. S1). We did not detect any precipitates or other phases in the dual-phase lamellae by selected-area electron diffraction (SAED) patterns and high-resolution high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM; Fig. 1G and fig. S1). We confirmed this observation with synchrotron high-energy x-ray diffraction (SHE-XRD; Fig. 1H). Our HAADF-STEM energy-dispersive spectroscopy analysis shows that the DS process has a negligible effect on the chemical composition distribution of the eutectic lamellae (fig. S2). We also obtained three-dimensional stereographic microstructure images of the two as-cast EHEAs (fig. S3). The average width of the columnar grains and its L12-phase content are ~54 mm and ~59 vol% in the directionally solidified EHEA, respectively. Both values are slightly larger than in the conventionally cast reference EHEA (grain size ~48 mm, L12phase content ~55 vol%). These two discrepancies are due to the slower cooling rate of the molten EHEA during DS relative to the rate during transient solidification when preparing the reference EHEA. The eutectic lamellar spacing of the reference EHEA is ~2.1 mm, which is smaller than that of the AEC (~2.8 mm) and the BEC (mainly varying within the range 3 to 8 mm). Variable lamellar spacing in the BEC results in significant Vickers hardness fluctuations (192 to 247 HV) compared to the hardness trend in the AEC (263 ± 10 HV). We observed a remarkable improvement in ductility, quantified from engineering stress– strain curves (Fig. 2A), of the directionally solidified material relative to the conventionally cast reference material. The uniform elongation tripled, increasing from ~16% for the reference EHEA to ~50% for the directionally solidified EHEA. Additionally, the directionally solidified EHEA has a ~150-MPa higher yield strength than the conventionally cast EHEA. Although the directionally solidified EHEA has a high content of hard, low-ductility B2 phase (~41 vol%) (15, 22, 23), it nonetheless exhibits a large uniform elongation of ~50%. Its elongation is comparable to that of widely studied, high-ductility, fully homogenized face-centered cubic high-entropy alloys (HEAs) (24–27). The resulting strength-ductility combination, and especially the uniform ductility, in sciencemag.org SCIENCE

RES EARCH | R E P O R T S

Fig. 1. Hierarchically arranged herringbone microstructure. (A to C) Conventionally cast EHEA serving here as reference material. (A) SEM backscattered electron image. (B) Electron backscattering diffraction (EBSD) phase map (left) and inverse pole figure (IPF) map (right). (C) Schematic diagram. (D to I) The directionally solidified EHEA with a hierarchical herringbone microstructure. The black arrows in (D) and (E) indicate the DS direction, and also the tensile loading direction in Fig. 2A. (D) SEM backscatter electron image showing that the microstructure is composed of columnar grains. Grain

the directionally solidified EHEA outperforms that of any other as-cast eutectic and neareutectic HEAs (14, 15, 19, 20, 22, 23, 28–33) (Fig. 2B). We attribute the gain in ductility of the directionally solidified material to its hierarchical herringbone microstructure and the effect that this structure has on crack buffering, as we revealed with detailed characterizations of the deformation and fracture mechanisms (Fig. 3). The post-fractured specimen surface of the EHEA, which we characterized by scanning electron microscopy (SEM), shows a large number of microcracks (Fig. 3H). The microcrack density is as large as ~8 × 104 mm–2 in the AEC, and the microcrack spacing is as SCIENCE sciencemag.org

boundaries are marked by black dashed lines. (E) Enlarged EBSD phase and IPF maps showing the columnar grain consisting of AEC and BEC. Black solid and dashed lines mark grain and colony boundaries, respectively. [(F) and (I)] Schematic diagram of herringbone structure and its formation principle, respectively. (G) HAADF-STEM image and related SAED patterns of B2 and L12 phases. The HAADF-STEM image shows clean dual-phase lamellae without evidence of nanoprecipitates or other phases, as is also indicated in (F). (H) SHE-XRD of B2 and L12 phases.

small as ~0.83 mm in some B2 lamellae of the AEC. We did not find the otherwise typical large secondary cracks that are observed in conventional as-cast materials (5) (fig. S5). The microcracks we observed are mainly distributed in the hard B2 lamellae of the AEC (Fig. 3H), yet they seem to have ultrahigh stability. The cracks remain strictly confined in the individual B2 phase where they formed, and no crack percolation into neighboring B2 lamellae occurs. This applies even in B2 regions with multiple microcracks that are separated by only a single L12 lamella (Fig. 3H). Interestingly, we detected high microcrack populations already at early and medium strains of up to ~25% (Fig. 3J). Considering the ~50% total

ductility, the material features a capability to withstand ~25% more straining after the onset of the first massive crack initiation at modest strains. In this deformation regime, no crack percolation or catastrophic failure event occurred, irrespective of the extreme abundance of microcracks (Fig. 3, E to H). This observation substantiates our claim that this alloy has extreme crack tolerance and a crack-mediated ductility reserve of ~25%. To further verify the exceptional crack tolerance, we evaluated the fracture resistance of the directionally solidified herringbone EHEA by measuring J-integral–based R-curves (where J is a function of the stable crack extension Da) using single-edge bend specimens 20 AUGUST 2021 • VOL 373 ISSUE 6557

913

RES EARCH | R E P O R T S

Fig. 2. Tensile response at ambient temperature. (A) Engineering stress– strain curves of the directionally solidified EHEA compared with the conventionally cast EHEA, displaying a substantial increase in uniform tensile ductility without any strength reduction. The directionally solidified EHEA shows no postuniform ductility, which is also confirmed by the absence of macroscopic necking at the fracture end in the inset of fig. S5A. Tensile loading was performed along the DS direction. Inset shows the corresponding strain-hardening curves.

in accordance with ASTM Standard E1820 (34). The crack-resistance (R-curve) behavior (fig. S6A) displays a surprisingly high crackinitiation fracture toughness JIc of ~318 kJ/m2, determined essentially at Da → 0, and a crackgrowth toughness Jss of ~430 kJ/m2 at a valid Da of ~1 mm. These toughness values are more than double those in the conventionally cast EHEA (fig. S6A). Such fracture properties are comparable with those of high-toughness gradient-structured materials reported recently (35). Thus, these trends demonstrate extreme crack resistance and damage tolerance of the herringbone EHEA. To illuminate the underlying mechanisms responsible for the extraordinary crack tolerance, we characterized the dynamic microcrack evolution in the directionally solidified herringbone structure (Fig. 3, E to H). We revealed an in situ developing hierarchical interplay between crystallographic microcrack guidance and crack blunting. We first observed the evolution of dense slip lines on the prepolished surface of the hard B2 lamellae (Fig. 3E). These slip lines mark regions of strong linear strain localization that arise from prevalent activation of the primary slip system with the highest Schmid factor (35–38). This promotes formation of microcracks in the B2 lamellae and their linear crystallographic propagation along these slip lines (Fig. 3, E and F). Quantitative analysis of these microcracks reveals a rapid increase in crack density in this strain stage (stage I in Fig. 3J), which subsequently increases modestly toward near-saturation. This behavior differs from the 914

20 AUGUST 2021 • VOL 373 ISSUE 6557

MDIH and MBIH refer to multi-slip dislocation–induced hardening and microbandinduced hardening, respectively; eU, uniform strain; sy, yield strength; sUTS, ultimate tensile strength. (B) Yield strength versus uniform strain of the directionally solidified EHEAs compared with those of previously reported as-cast eutectic and near-eutectic HEAs (14, 15, 19, 20, 22, 23, 28–33). (N-)EHEAs refers to eutectic and near-eutectic HEAs. The conventional (N-)EHEAs include directly cast and arc-melting eutectic and near-eutectic HEAs.

gradually increasing crack density observed in most common materials (10). Surprisingly, this rapid increase in microcrack population occurs in a relatively small intermediate-strain range of 25 to 30%. The resulting microcrack density is up to ~5.5 × 104 mm–2, which is more than twice the density increase found in the subsequent large-strain range of 30 to 50% (Fig. 3J). These multiple cracks do not cause specimen rupture; thus, the plastically deformed material reveals impressive tolerance against percolative crack expansion and catastrophic failure. We observed that the microcracks emerging in the B2 phase became arrested at the interfaces to the alternatingly adjacent L12 lamellae of the AEC (Fig. 3F). These softer L12 lamellae serve as soft crack buffers that act on the tips of microcracks, blunt them, and shield the associated high local stresses (10, 35, 39). Without the presence of such alternating soft buffer layers, the cracks would have percolated forward. When the microcracks have penetrated an entire B2-lamellar cross section, the neighboring soft L12 lamellae blunt the crack tips, as evidenced by their rounded shapes (10) (Fig. 3G). Also, for triggering sample fracture, microcracks would not only have to cut through the L12 lamellae and thus cut the whole AEC, but also penetrate into the adjacent BEC zone. This means that the herringbone microstructure has two hierarchical levels that feature crack-arresting properties, namely the alternating hard-soft (B2-L12) phase layers and the changing alignment of the eutectic colonies. Further three-point bending experiments revealed that this hierarchical effect effectively

blunts the crack tip and buffers the crack propagation (fig. S8) (21), thus giving the directionally solidified herringbone EHEA concurrently excellent fracture toughness (fig. S6A) (39). Another important effect is that during tension, new microcracks are frequently generated in the free intact material portions between these fully extended cracks within every single B2 lamella (Fig. 3, G and H). This crack pattern refinement releases stress concentrations around phase interfaces and weakens the stress intensity at the tips of larger microcracks (1–6). Further growth of these fully extended microcracks requires increased mechanical loads for developing them into unstable and critical larger cracks that can cause failure (1–3). However, the overall directional topology of the phases and interfaces means that these microcracks can only grow with an alignment toward the tensile (and not transverse) direction in the B2 lamellae of the AEC. Numerous microcracks develop in such a fashion assuming stable parallelogram-like shapes (Fig. 3H). We confirmed these experimental observations with high-resolution focused ion beam imaging (fig. S5C) and quantitative microcrack studies (stage III in Fig. 3J). This stage features a very slow increase in microcrack density and length. However, these microcracks in the low-ductility B2 lamellae of the AEC exhibit a surprisingly high capacity to accommodate their shapes and thus carry strains, as revealed by their parallelogram-like morphology (Fig. 3H). Even in this high-load regime, no interface delamination cracks are found in the herringbone structure. This effect is attributed to its semisciencemag.org SCIENCE

RES EARCH | R E P O R T S

Fig. 3. Hierarchical crack buffering. (A to C) SEM backscattered electron images showing sequentially activated slip lines from soft BEC to strong AEC. The inset in (B) shows enlarged cross-slip lines. (D) SEM image exhibiting compatible deformation between adjacent columnar grains and no grain-boundary cracks. Black solid and dashed lines mark grain and colony boundaries, respectively. (E to H) Well-controlled microcrack evolution in the AEC. Insets (upper, enlarged; lower, schematic) illustrate dynamic

coherent interface structure, which can bear high shear stresses (17) (fig. S9). By contrast, in the absence of crack tolerance, early failure occurs in the conventionally cast EHEA. In the directionally solidified material, the strain tolerated by the hierarchical herringbone structure, with its high density of cracks in the 25 to 50% tensile strain regime, exceeds the overall elongation (~16%, Fig. 2A) of the conventionally cast EHEA. In general, the cumulative crack damage of the material’s cross section will gradually reduce its effective load-bearing capability per unit area, thus deSCIENCE sciencemag.org

microcrack evolution. (I) EBSD-based image quality (IQ) and IQ with kernel average misorientation (KAM) maps. The KAM is calculated up to the fifth-neighbor shell with a maximum misorientation angle of 5°, which is indicative of the deformation degree. (J) Evolution of microcrack length, microcrack density, and compensated strain in the AEC. Stages I to III correspond to tensile strains of 25 to 30%, 30 to 40%, and 40 to 50%, respectively. Data are means ± SD. (K) Schematic diagram.

creasing the apparent nominal tensile stress (6–11). However, this trend does not appear in the directionally solidified material (Fig. 2A) despite its high crack density. We investigated the microstructure in detail at the nanoscale to better understand the cracktolerance mechanisms. Aberration-corrected transmission electron microscopy (TEM; Fig. 4) shows that the directional movement of massive dislocations in the crystal interior produces the pronounced crystallographic slip-line structures (35–38), which are also visible on the prepolished sample surfaces (Fig. 3). The de-

formation of the B2 lamellae is dominated by planar slip of screw dislocations on {110}h111i slip systems, and the L12 lamellae show planardislocation slip on {111}h011i (Fig. 4, A and B, and fig. S10). As deformation proceeds, the B2 lamellae undergo increasing planar-dislocation shear [characterized by markedly reduced dislocation spacing at tensile strains of 10 to 25%; (fig. S11H)]. This leads to a gradual exhaustion of deformability (fig. S11, E to H). The exhausted node corresponds roughly to ~25% tensile strain, thus triggering extensive microcrack initiation at the low-strain range of 25 to 30% 20 AUGUST 2021 • VOL 373 ISSUE 6557

915

RES EARCH | R E P O R T S

Fig. 4. Microstructural and micromechanical observations of load-bearing response for L12 lamellae. (A and B) Planarslip dislocations in L12 and B2 lamellae, respectively. (C) Deformation-induced microbands in L12 lamellae. The inset (scale bar, 300 mm) exhibits clearer microband structure. (D and E) Dynamic slip band refinement [(D) and (E), top] shown by HAADF-STEM) and fresh slip bands and cross-slip dislocations marked by red lines and yellow arrows, respectively [(E), bottom]. (F) Deformationinduced microbands. The ring-like SAED pattern (inset) suggests that these microbands are similar to low-angle grain boundaries (rather than mechanical twins). The beam directions are [011] in (A), (D), and (E) and [001] in (B). g indicates the direction of the diffraction vector; eE, tensile strain. (G) Real-time stress partitioning of B2 and L12 phases during tensile loading (21). (H) Selected 2D x-ray diffraction images along the full azimuthal angle h (0° to 360°) at tensile strain of ~48%. Note that 90° and 180° correspond to the loading direction (LD) and transverse direction (TD), respectively.

(Fig. 3, E and J) and their crystallographic propagation along these slip lines inside the B2 lamellae (Fig. 3, F and G). In the L12 lamellae, however, we found a dynamic substructure refinement governed by sequentially activated multi-slip dislocation shear and microband formation (Fig. 4). At a tensile strain of ~5%, the deformation of the L12 lamellae is at first dominated by planar-dislocation slip evolving into a banded shear morphology (Fig. 4A). In these slip bands, the spacing among adjacent dislocations gradually decreases from the lamellar interior toward the lamellar interfaces (fig. S11A), thereby establishing a long-range strain gradient (40, 41). These dislocations, piled up against the interfaces, are known as geometrically necessary dislocations (GNDs), creating back-stress hardening (40). At ~15% strain, these pronounced planar dislocations evolve into well-developed slip bands (42) (Fig. 4D). Subsequently, we observed deformation916

20 AUGUST 2021 • VOL 373 ISSUE 6557

driven refinement of slip bands. Fresh slip bands are constantly generated in the free space between initially existing ones (Fig. 4E). The increasing slip-band density supports substantial dislocation storage accompanied by back-stress hardening. By conducting loading-unloadingreloading experiments, we detected a backstress–dominated high kinematic strengthening effect (holding about two-thirds of the applied stress; fig. S12) (40, 43). This quantitatively confirms the leading role of back-stress hardening in the L12 lamellae at this stage. Upon further straining, we observed dislocations in wavyslip patterns between slip bands (Fig. 4E). This suggests that dislocation cross-slip is activated, inducing forest dislocation hardening (43) (fig. S12). At higher strains of 35 to 42%, we observed deformation-induced microbands that subdivide the L12 lamellae into numerous platelike misoriented domains (43–45) (Fig. 4, C and F). Thus, these microbands, analogous to low-

angle grain boundaries (not twins, as evidenced by Fig. 4F, inset), are associated with further lamellar subdivision and refinement. This mechanism promotes a microband-induced hardening effect, as demonstrated by the surprisingly high strain-hardening rate (Fig. 2A, inset). As we further discovered using low-angle annular dark-field STEM (fig. S13), this dynamic substructure refinement favors substantial local strain hardening in the vicinity of the incoming crack tips (3), thus turning the L12 lamellae into very efficient crack buffer regions. This phenomenon endows the herringbone structure with excellent crack tolerance (Fig. 3K). Besides the plastic buffering and crack blunting effect, strong strain hardening of the L12 lamellae (Fig. 4, C and F) contributes to the high load-bearing capability of the material. This counteracts the potential softening effect caused by microcracks in the B2 phase, as revealed by the finding that the material sciencemag.org SCIENCE

RES EARCH | R E P O R T S

does not lose but rather gains tensile strength (Fig. 2A). To provide quantitative insights into the exceptional load-bearing capability, we performed in situ synchrotron experiments. We observed a real-time stress partitioning effect (46–48) between the B2 and L12 phases, as assessed from the SHE-XRD results (Fig. 4G). As we expected, the yielding of the L12 phase leads to stress relaxation and stress transfer to the hard B2 phase. This trend is known from other dualphase composites (46–48). Stress partitioning to hard phases usually increases continuously until fracture (46). However, in the herringbone structure the B2 phase bears decreasing stress after reaching a tensile strain of ~25%, whereas the L12 phase exhibits an opposite trend with remarkable strain-hardening behavior (Fig. 4G). This marks a gradual transfer of the load from the hard but brittle B2 phase to the initially soft but gradually strain-hardened L12 phase. This means that the SHE-XRD probing elucidates that the load-bearing capacity of the L12 phase increases substantially as a result of dynamic substructure refinement, whereas that of the B2 phase decreases gradually because of increasing internal microcracking. To compensate the limited TEM-sampling area, we also conducted 2D SHE-XRD diffraction investigations (48) covering the entire 360° azimuthal range recorded for different planes near the point of fracture (Fig. 4H). No new diffraction lines were detected, which suggests that the high load-bearing capability of the herringbone structure is not caused by conventional phase transformation. However, the diffraction lines of the B2 phase (e.g., marked by red lines) deviate severely from their initial angle. This is caused by the huge elastic strain of the B2 phase (up to ~5.5%; fig. S14). This observation is also revealed by the conventionally cast EHEA, which shows delayed dislocation shear in the B2 lamellae (fig. S15). When further exploring the hierarchical herringbone structure and its microstructural evolution during tensile deformation, we identified a slip line–mediated sequential deformation transfer from the soft BEC to the hard AEC (Fig. 3, A and D) (21). This finding reveals that the BEC features high deformability that allows compatible deformation of adjacent columnar grains (Fig. 3D). At the early deformation stage, dense dislocation pile-up arrays against phase interfaces were observed in the L12 lamellae of the BEC (Figs. 3A and 4A). The associated pileup stresses are accommodated by the elastic deformation of the plastically less compliant B2 lamellae (Fig. 3B), thereby shielding high stress concentrations (48); this mechanism supports compatible co-deformation of adjacent phases and colonies (49). The large local misorientation shown in the kernel average misorientation (KAM) map (Fig. 3I) confirms that the plastic strain incompatibility can be SCIENCE sciencemag.org

well accommodated in the BEC. As the deformation progresses, stable microcracks, as observed in the AEC, can also be generated in the low-ductility B2 lamellae of the BEC, assisting its compliance (fig. S16). These mechanisms of mechanical energy release in the adjacent phases and colonies reduce the chance of grainboundary decohesion or cracking (40, 49) (fig. S5B), which are not accessible to conventionally cast EHEA. The statistically distributed lamellar arrangements in the differently oriented grains cannot sustain a compatible co-deformation among them (21). Consequently, grain-boundary cracking concomitant with mechanical instability triggers premature failure (i.e., deteriorating tensile ductility) (figs. S17 to S19) and also limits the fracture toughness level of the conventionally cast material (40, 49) (figs. S6 and S7). Natural materials with high toughness often also comprise hard and soft components in hierarchical layered architectures (50–52). The lamellar cortical bone, a prime example, consists of mineralized collagen fibrils and a nonfibrillar organic matrix, which acts as a “glue” that holds the mineralized fibrils together (50–52). Healthy lamellar bone resists fracture through complementary intrinsic and extrinsic contributions throughout its hierarchical structure (50, 52). The glue-mediated fibrillar sliding mechanism, analogous to the dislocation-assisted inelastic deformation in our herringbone EHEA, is essential for high plasticity (50). In both materials, plasticity and the resultant ductility provide a major contribution to the intrinsic toughness by dissipating energy and forming plastic zones surrounding incipient cracks, which further serves to blunt crack tips, thereby reducing the driving force for cracking (3, 50). The extrinsic mechanisms, such as collagen-fiber bridging and crack deflection, act principally on the wake of cracks to reduce and shield local stresses/strains experienced at crack tips, thus inhibiting their propagation (50). These effects exhibit a marked similarity to what we identified in our herringbone material (see Fig. 3, E to H, and fig. S8). Of course, a salient characteristic of bone is its ability to remodel itself to heal and repair damage—a trait that is difficult to replicate in our bionic-inspired synthetic materials (50, 51). Our work offers a hierarchical microstructure design approach, realized in a directionally solidified bulk EHEA, that allows reconciliation of crack tolerance and high uniform elongation, features that are usually mutually exclusive in both human-made and biological materials. The crack tolerance is maintained over a huge range of ~25% tensile elongation and enables an increase in the ductility of the material by a factor of 3 relative to conventionally solidified material, without sacrificing strength. These proposed mechanisms exhibit practical merits

in guiding a broader group of eutectic-type cast HEAs and traditional alloy development. The microstructure approach can also be potentially realized in other bulk materials consisting of hard and soft phases that can be rendered into hierarchically organized herringbone microstructures, enabling the design of cracktolerant yet high-deformability materials not by avoiding cracks but by guiding and buffering them. Furthermore, this hierarchical herringbone microstructure design approach and its salient effect on crack buffering show a promising guidance in designing not only new hierarchically structured alloys with high elongation but also new bone-substituting biomaterials with excellent fracture toughness. REFERENCES AND NOTES

1. R. O. Ritchie, Nat. Mater. 10, 817–822 (2011). 2. H. Gao, B. Ji, I. L. Jäger, E. Arzt, P. Fratzl, Proc. Natl. Acad. Sci. U.S.A. 100, 5597–5600 (2003). 3. M. Koyama et al., Science 355, 1055–1057 (2017). 4. L. Liu et al., Science 368, 1347–1352 (2020). 5. P. Shi et al., Nat. Commun. 10, 489 (2019). 6. B. H. Sun et al., Acta Mater. 164, 683–696 (2019). 7. D. Y. Yan, C. C. Tasan, D. Raabe, Acta Mater. 96, 399–409 (2015). 8. A. Alaie, J. Kadkhodapour, S. Z. Rad, M. A. Asadabad, S. Schmauder, Mater. Sci. Eng. A 623, 133–144 (2015). 9. M. Koyama, C. C. Tasan, E. Akiyama, K. Tsuzaki, D. Raabe, Acta Mater. 70, 174–187 (2014). 10. M. M. Wang, C. C. Tasan, D. Ponge, A. C. Dippel, D. Raabe, Acta Mater. 85, 216–228 (2015). 11. Z. Li et al., Nano Lett. 15, 8077–8083 (2015). 12. J. W. Yeh et al., Adv. Eng. Mater. 6, 299–303 (2004). 13. E. P. George, D. Raabe, R. O. Ritchie, Nat. Rev. Mater. 4, 515–534 (2019). 14. Y. Lu et al., Sci. Rep. 4, 6200 (2014). 15. X. Jin, Y. Zhou, L. Zhang, X. Y. Du, B. S. Li, Mater. Lett. 216, 144–146 (2018). 16. Z. Shang et al., Intermetallics 57, 25–33 (2015). 17. H. Bei, E. P. George, Acta Mater. 53, 69–77 (2005). 18. Z. Shang, J. Shen, G. Liu, Y. H. Xu, Mater. Sci. Technol. 34, 1839–1846 (2018). 19. H. T. Zheng et al., Intermetallics 113, 106569 (2019). 20. L. Wang et al., Intermetallics 118, 106681 (2020). 21. See supplementary materials. 22. H. Jiang et al., J. Mater. Sci. Technol. 61, 119–124 (2021). 23. Y. P. Lu et al., Acta Mater. 124, 143–150 (2017). 24. B. Gludovatz et al., Science 345, 1153–1158 (2014). 25. W. Q. Guo et al., Acta Mater. 185, 45–54 (2020). 26. Y. H. Jo et al., Nat. Commun. 8, 15719 (2017). 27. Q. Y. Lin et al., Mater. Res. Lett. 6, 236–243 (2018). 28. X. Jin, Y. Zhou, L. Zhang, X. Y. Du, B. S. Li, Mater. Des. 143, 49–55 (2018). 29. Z. S. Yang et al., Appl. Phys. A 125, 208 (2019). 30. L. L. Ma, J. N. Wang, P. P. Jin, Mater. Res. Express 7, 016566 (2020). 31. X. Jin et al., J. Alloys Compd. 770, 655–661 (2019). 32. Q. F. Wu et al., Acta Mater. 182, 278–286 (2020). 33. X. Jin, Y. X. Liang, J. Bi, B. S. Li, Materialia 10, 100639 (2020). 34. E08 Committee, E1820 Standard Test Method for Measurement of Fracture Toughness (ASTM International, 2017). 35. R. Q. Cao et al., Mater. Today 32, 94–107 (2020). 36. S. Wei et al., Nat. Mater. 19, 1175–1181 (2020). 37. S. Mader, A. Seeger, C. Leitz, J. Appl. Phys. 34, 3368–3375 (1963). 38. Z. Lei et al., Nature 563, 546–550 (2018). 39. Z. Zhang et al., Nat. Commun. 6, 10143 (2015). 40. Y. T. Zhu et al., Mater. Res. Lett. 9, 1–31 (2021). 41. E. Ma, X. Wu, Nat. Commun. 10, 5623 (2019). 42. E. Welsch et al., Acta Mater. 116, 188–199 (2016). 43. T. Yang et al., Science 362, 933–937 (2018). 44. J. D. Yoo, K. T. Park, Mater. Sci. Eng. A 496, 417–424 (2008). 45. P. J. Shi et al., Mater. Today 41, 62–71 (2020). 46. S. Cheng et al., Acta Mater. 58, 2419–2429 (2010). 47. L. L. Ma et al., Acta Mater. 128, 12–21 (2017). 48. Y. J. Shi et al., Mater. Sci. Eng. A 771, 138555 (2020).

20 AUGUST 2021 • VOL 373 ISSUE 6557

917

RES EARCH | R E P O R T S

49. T. Yang et al., Science 369, 427–432 (2020). 50. U. G. K. Wegst, H. Bai, E. Saiz, A. P. Tomsia, R. O. Ritchie, Nat. Mater. 14, 23–36 (2015). 51. G. E. Fantner et al., Nat. Mater. 4, 612–616 (2005). 52. S. Weiner, T. Arad, I. Sabanay, W. Traub, Bone 20, 509–514 (1997). ACKN OW LEDG MEN TS We thank G. Liu, P. M. Cheng, and P. Zhang (Xi’an Jiaotong University) for their help with fracture toughness and nanoindentation tests, and Y. Z. Chen (Northwestern Polytechnical University), J. F. Li (Shanghai Jiaotong University), and X. H. Xiao (Brookhaven National Laboratory) for discussions. Funding: Supported by National Key Research and Development Program of China grants 2018YFF0109404, 2016YFB0300401, and 2016YFB0301401 and National Natural Science Foundation of China grants U1732276 and

U1860202 (Y.B.Z.); National Science Foundation of China (NSFC) grant 51831003 and Funds for Creative Research Groups of China grant 51921001 (Y.-D.W.); National Natural Science Foundation of China grant 51704193 (T.X.Z.); and National Natural Science Foundation of China grant 51904184 (Z.S.). Use of the Advanced Photon Source was supported by the US Department of Energy, Office of Science, Office of Basic Energy Sciences, under contract DE-AC02-06CH11357. Author contributions: Y.B.Z. and P.J.S. designed the study; P.J.S. carried out the main experiments; D.R., Y.-D.W., Y.B.Z., Y.R., W.L.R., Y.Z., P.K.L., R.G.L., and P.J.S. analyzed the data and wrote the main draft of the paper; J.C.P., N.M., and P.F.H. conducted the TEM, LAADF-STEM, and HAADF-STEM characterizations; R.G.L. and Y.R. conducted the high-energy synchrotron XRD; X.L. prepared the TEM specimens with a dual-beam focused ion beam (FIB) instrument; Y.L., Y.B.W., Z.S., and T.X.Z. processed the alloy and tensile samples; all authors contributed to the

INFLUENZA

Rare variant MX1 alleles increase human susceptibility to zoonotic H7N9 influenza virus Yongkun Chen1†, Laura Graf2,3†, Tao Chen4†, Qijun Liao1†, Tian Bai4, Philipp P. Petric2,3,5, Wenfei Zhu4, Lei Yang4, Jie Dong4, Jian Lu4, Ying Chen6, Juan Shen6, Otto Haller2,3,7, Peter Staeheli2,3, Georg Kochs2,3, Dayan Wang4*, Martin Schwemmle2,3*, Yuelong Shu1,4* Zoonotic avian influenza A virus (IAV) infections are rare. Sustained transmission of these IAVs between humans has not been observed, suggesting a role for host genes. We used whole-genome sequencing to compare avian IAV H7N9 patients with healthy controls and observed a strong association between H7N9 infection and rare, heterozygous single-nucleotide variants in the MX1 gene. MX1 codes for myxovirus resistance protein A (MxA), an interferon-induced antiviral guanosine triphosphatase known to control IAV infections in transgenic mice. Most of the MxA variants identified lost the ability to inhibit avian IAVs, including H7N9, in transfected human cell lines. Nearly all of the inactive MxA variants exerted a dominant-negative effect on the antiviral function of wild-type MxA, suggesting an MxA null phenotype in heterozygous carriers. Our study provides genetic evidence for a crucial role of the MX1-based antiviral defense in controlling zoonotic IAV infections in humans.

A

vian influenza A viruses (IAVs) periodically cross the species barrier and infect humans. Although such spillover events are rare, they may represent a source of new pandemic virus strains (1). One of the major avian IAV zoonoses in recent decades was caused by the H7N9 subtype. Fatal cases were first reported in the Yangtze River delta region of China in spring 2013 (2). By the end of 2020, a total of 1568 laboratory-confirmed cases had been reported, with a case fatality

rate of ~39% (3). The zoonotic H7N9 virus has acquired several features necessary for adaptation to mammalian hosts, including altered hemagglutinin (HA) receptor specificity (4) and enhanced viral polymerase activity (5). Still, the molecular mechanisms enabling crossspecies transmission of avian IAVs remain incompletely understood. Exposure to poultry is the main risk factor for productive H7N9 infection of humans (6, 7), yet occupational poultry workers represent only 7% of all cases

discussion of the results and commented on the manuscript. Competing interests: The authors declare no competing interests. Data and materials availability: All data are reported in the main paper and supplementary materials. SUPPLEMENTARY MATERIALS

science.sciencemag.org/content/373/6557/912/suppl/DC1 Materials and Methods Supplementary Text Figs. S1 to S21 Table S1 References (53–63) 13 November 2020; accepted 1 July 2021 10.1126/science.abf6986

(7), indicating that genetic factors may play a role in virus susceptibility. To identify genes that might predispose humans to H7N9 infections, we performed wholegenome sequencing (WGS) of H7N9 patients and healthy controls (fig. S1). We enrolled 220 Han Chinese patients with laboratory-confirmed H7N9 infection between 2013 and 2017 and 121 epidemiologically linked healthy poultry workers as controls (see methods in supplementary materials for more information about study participants). High-depth WGS (average sequencing depth > 30×; table S1) was performed to accurately detect rare variants. After exclusion of samples on the basis of quality control criteria, 217 patients and 116 controls remained for further analysis (see methods in supplementary materials and figs. S1 and S2). In total, we identified 18.5 million highconfidence autosomal variants after strict quality control filtering, including 17.5 million

1

School of Public Health (Shenzhen), Sun Yat-sen University, Shenzhen, China. 2Institute of Virology, Medical Center – University of Freiburg, Freiburg, Germany. 3Faculty of Medicine, University of Freiburg, Freiburg, Germany. 4Chinese National Influenza Center, National Institute for Viral Disease Control and Prevention, Chinese Center for Disease Control and Prevention, Beijing, China. 5Spemann Graduate School of Biology and Medicine, University of Freiburg, Freiburg, Germany. 6BGIShenzhen, Shenzhen, China. 7Department of Molecular Life Sciences, University of Zurich, Zurich, Switzerland. *Corresponding author. Email: [email protected] (Y.S.); [email protected] (M.S.); dayanwang@cnic. org.cn (D.W.) †These authors contributed equally to this work.

Table 1. Carrier frequencies of rare MX1 SNVs in H7N9 patients and controls. N, number of individuals; OR, odds ratio; CI, confidence interval. Sample group

N

Carriers (%)

OR (95% CI)*

P value

H7N9 patients 217 21 (9.68) – – Healthy poultry workers 116 0 (0.00) – 1.67 × 10−4 ............................................................................................................................................................................................................................................................................................................................................ 4,078 72 (1.77) 5.96 (3.40, 10.04) General population I‡ 3.33 × 10−9 ............................................................................................................................................................................................................................................................................................................................................ § 10,588 130 (1.23) 8.61 (5.05, 14.09) 2.29 × 10−12 General population II ............................................................................................................................................................................................................................................................................................................................................ ............................................................................................................................................................................................................................................................................................................................................

*OR for H7N9 patients compared with control groups. Project (ChinaMAP).

918

20 AUGUST 2021 ¥ VOL 373 ISSUE 6557

†P values were calculated using (two-sided) Fisher’s exact test.

‡Published data.

§Data from the China Metabolic Analytics

sciencemag.org SCIENCE

RES EARCH | R E P O R T S

Fig. 1. SNVs detected in H7N9 patients cause loss of MxA antiviral activity. (A) Primary structure of MxA, with arrows indicating positions of MX1 variations. B, bundle signaling element (BSE) consisting of three a helices; L4s, loop. (B) Positions of the rare variations in the MxA structure (Protein Data Bank ID 3SZR). Blue dotted line, unstructured loop L4S; gray dotted line, unstructured N terminus. Single-letter abbreviations for the amino acid residues are as follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr. (C) Antiviral activity of MxA variants in a polymerase reconstitution assay of H7N9. Results are presented relative to activity in the absence of MxA (vector)

single-nucleotide variants (SNVs) and 1 million insertion-deletion polymorphisms (indels). The minor allele frequency (MAF) was