136 103 45MB
English Pages [158] Year 2023
Prominent neuroscientist faces suspicions of data doctoring p. 754
International law and reliance on CO2 removal p. 772
Molecular glue degrader design wins Science & SciLife Lab Prize p. 779
$15 17 NOVEMBER 2023 science.org
HEAPS OF
WARMING Municipal solid waste emits large amounts of greenhouse gases pp. 762 & 797
Publish your research in the Science family of journals The Science family of journals (Science, Science Advances, Science Immunology, Science Robotics, Science Signaling, and Science Translational Medicine) are among the most highlyregarded journals in the world for quality and selectivity. Our peer-reviewed journals are committed to publishing cutting-edge research, incisive scientific commentary, and insights on what’s important to the scientific world at the highest standards.
Submit your research today! Learn more at Science.org/journals
1117Product.indd 738
11/9/23 8:21 AM
CONTENTS 17 N O V E M B E R 2 0 2 3 VO LU M E 3 8 2 ISSUE 6672
754
NEWS IN BRIEF
746 News at a glance IN DEPTH
748 AI is set to revolutionize weather forecasts Cheap and fast algorithms are matching— and surpassing—the world’s top models By P. Voosen CREDITS: (TOP TO BOTTOM) TAVO MONTAÑEZ; MARTIN SURBECK/KOKOLOPORI BONOBO RESEARCH PROJECT
RESEARCH ARTICLE BY R. LAM ET AL. 10.1126/SCIENCE.ADI2336; PODCAST
749 Deal to build pint-size nuclear reactors is canceled NuScale Power’s small modular reactors promised cheaper nuclear power, but costs soared and utilities balked By A. Cho
750 Ousted biologist starts over Fired for sexual misconduct, biologist David Sabatini lands new job in Prague. Reactions are mixed
753 Australian science agency faces scrutiny over industry influence
764 Grabbing neuropeptide signals in the brain
Lawsuit related to 2010 Deepwater Horizon oil spill reveals documents that suggest oil firm BP reviewed CSIRO studies By R. Kurmelovs
Bioengineered sensors resolve the dynamics of neuropeptide action RESEARCH ARTICLE p. 786
FEATURES
754 Brain games? Whistleblowers and former lab members suggest a star neuroscientist routinely manipulated data, compromising a planned NIH stroke trial and key Alzheimer’s research
By E. A. Groisman et al.
INSIGHTS
The efficiency of targeted DNA insertion by CRISPR transposons is improved By Y. Dhingra and D. G. Sashital RESEARCH ARTICLE p. 784
PERSPECTIVES
760 Between-group cooperation in bonobos Bonobos provide insight into the origins of partner-specific cooperation in human groups By J. B. Silk
769 A highly efficient solid-state heat pump The high efficiency of a newly developed electrocaloric device brings theory closer to reality By J. Tušek RESEARCH ARTICLE p. 801
RESEARCH ARTICLE p. 805
751 Rewriting DNA in the body lowers cholesterol
Reducing methane emissions from solid waste is already technically possible
Verve Therapeutics says its base-editing approach may help prevent heart disease in many people By J. Kaiser
By M. E. Webber and Y. R. Glazer RESEARCH ARTICLE p. 797
763 A dynamic biointerface controls mussel adhesion
As nations push for green hydrogen and ammonia, researchers warn of side effects
The mussel-adherent secreta interface reveals how nonliving material can be compatible with tissue By G. Pan and B. Li
By K. Bourzac
RESEARCH ARTICLE p. 829
1117TOC_18262750.indd 739
Nutrient starvation of beneficial bacteria helps them colonize the human gut
768 A tool for more specific DNA integration
762 Solid waste, a lever for decarbonization
SCIENCE science.org
766 Advancing the fitness of gut commensal bacteria
By C. Piller
By M. Wadman
752 Carbon-free fuels could have a climatic dark side
By R. A. Romanov and T. Harkany
760 & 805 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
739
11/14/23 5:22 PM
CONTENTS
771 C. R. Rao (1920–2023)
815 Catalysis
Pioneering statistician and father of information geometry By D. Banks and J. L. Clarke
Nickel-catalyzed ester carbonylation promoted by imidazole-derived carbenes and salts C. Yoo et al.
POLICY FORUM
820 Metabolism
772 Legal limits to the use of CO2 removal
Autoregulatory control of mitochondrial glutathione homeostasis Y. Liu et al.
Climate targets that depend heavily on CO2 removal may contravene international law
829 Biomaterials
By R. F. Stuart-Smith et al.
A strong quick-release biointerface in mussels mediated by serotonergic ciliabased adhesion J. Sivasundarampillai et al.
BOOKS ET AL.
775 The responsibility turn
PERSPECTIVE p. 763
Lessons from the COVID-19 pandemic inspire a guide to recognizing the politics of modeling By E. Nabavi and S. Razavi
834 Ocean heat
An amateur archaeologist’s exploits highlight the damage wrought to Indigenous sites at the turn of the 20th century By M. M. Martin LETTERS
777 Vietnam’s vital role in primate conservation By A. Maheshwari et al. 777 Learn from tobacco to reduce betel nut use By S. Chen et al. 778 Polio eradication efforts: Above all, do no harm By S. A. Plotkin and K. Chumakov
778 Errata
RESEARCH
763 & 829 Reconstructed features of a small region in the mussel byssus stem root, which strongly anchors and rapidly detaches from surfaces
786 Neuroscience A tool kit of highly selective and sensitive genetically encoded neuropeptide sensors H. Wang et al. RESEARCH ARTICLE SUMMARY; FOR FULL TEXT: DOI.ORG/10.1126/SCIENCE.ABQ8173 PERSPECTIVE p. 764
Particle-phase accretion forms dimer esters in pinene secondary organic aerosol
Gluing the pieces together By Z. Kozicka
Credentials aren’t everything
Emergent symmetry in a low-dimensional superconductor on the edge of Mottness
By K. Suleta
P. Chudzinski et al. ON THE COVER
801 Electrocalorics
784 CRISPR
High cooling performance in a double-loop electrocaloric heat pump J. Li et al. PERSPECTIVE p. 769
805 Anthropology Cooperation across social borders in bonobos L. Samuni and M. Surbeck PERSPECTIVE p. 760
RESEARCH ARTICLE SUMMARY; FOR FULL TEXT: DOI.ORG/10.1126/SCIENCE.ADG3053
779 Prize Essay 846 Working Life
RESEARCH ARTICLES
Design principles of 3D epigenetic memory systems J. A. Owen et al.
743 Editorial
792 Solid-state physics
PERSPECTIVE p. 762; PODCAST
785 Epigenetics
Science, justice, and evidence By J. Mnookin
C. M. Kenseth et al.
781 From Science and other journals
PERSPECTIVE p. 768
741 Editorial
Correction is courageous By H. H. Thorp
Curbing global solid waste emissions toward net-zero warming futures Z. X. Hoy et al.
RESEARCH ARTICLE SUMMARY; FOR FULL TEXT: DOI.ORG/10.1126/SCIENCE.ADJ8543
DEPARTMENTS
787 Atmospheric aerosols
IN BRIEF
J. T. George et al.
An all-metal fullerene: [K@Au12Sb20]5− Y.-H. Xu et al.
797 Global warming
Mechanism of target site selection by type V-K CRISPR-associated transposases
840 Inorganic chemistry
810 Solar cells Bimolecularly passivated interface enables efficient and stable inverted perovskite solar cells C. Liu et al.
A person recovering materials amid a massive fire at the Bhalswa landfill site in India symbolizes the challenges of solid waste mismanagement. Urgent action is vital as temperatures approach the Paris Agreement limit and jeopardize the Global Methane Pledge goal. Solutions demand a critical shift in technical and behavioral practices to prevent an impending environmental catastrophe and unsustainable use of resources. See pages 762 and 797. Image: Sanchit Khanna/Hindustan Times via Getty Images Science Staff .............................................. 742 Science Careers ........................................ 844
SCIENCE (ISSN 0036-8075) is published weekly on Friday, except last week in December, by the American Association for the Advancement of Science, 1200 New York Avenue, NW, Washington, DC 20005. Periodicals mail postage (publication No. 484460) paid at Washington, DC, and additional mailing offices. Copyright © 2023 by the American Association for the Advancement of Science. The title SCIENCE is a registered trademark of the AAAS. Domestic individual membership, including subscription (12 months): $165 ($74 allocated to subscription). Domestic institutional subscription (51 issues): $2411; Foreign postage extra: Air assist delivery: $107. First class, airmail, student, and emeritus rates on request. Canadian rates with GST available upon request, GST #125488122. Publications Mail Agreement Number 1069624. Printed in the U.S.A. Change of address: Allow 4 weeks, giving old and new addresses and 8-digit account number. Postmaster: Send change of address to AAAS, P.O. Box 96178, Washington, DC 20090–6178. Single-copy sales: $15 each plus shipping and handling available from backissues.science.org; bulk rate on request. Authorization to reproduce material for internal or personal use under circumstances not falling within the fair use provisions of the Copyright Act can be obtained through the Copyright Clearance Center (CCC), www.copyright.com. The identification code for Science is 0036-8075. Science is indexed in the Reader’s Guide to Periodical Literature and in several specialized indexes.
740
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
science.org SCIENCE
CREDIT: JENAES SIVASUNDARAMPILLAI
776 When collectors came for the American West
Surface climate signals transmitted rapidly to deep North Atlantic throughout last millennium W. Lu et al.
EDITORIAL
Science, justice, and evidence
C
ourts in the United States have increasingly relied on scientific evidence and expert testimony to help resolve questions of fact. On 1 December 2023, amendments to Federal Rule of Evidence 702 will take effect, further clarifying the court’s responsibilities as a gatekeeper for expert evidence. This update comes just a few months after the 30-year anniversary of the Supreme Court’s landmark decision on how federal judges should evaluate scientific evidence. Daubert v. Merrell Dow was hailed as a victory for the use of scientific information in the legal system and certainly cast a much-needed spotlight on scientific evidence in the courtroom. But the nuanced and flexible nature of the “Daubert standard” has since led to substantial inconsistencies in its application. Most strikingly, it has had far more impact in civil cases than criminal cases. Daubert’s core tenet—that scientific evidence introduced in court should be adequately valid and reliable—needs to be taken just as seriously in the criminal justice system and for forensic science as it has been in civil cases. Daubert instructs judges to be “gatekeepers” responsible for assessing the validity of the science brought to court. Previously, courts often asked only whether the science was “generally accepted” by the relevant scientific community. Because judges often treated an expert witness’s own assertion of general acceptance as adequate, the rule did not present much of a bar for admissibility. With Daubert, the onus is more squarely on the judge to assess validity. The opinion details numerous possible factors to consider (including testing, error rate, peer review, and general acceptance), but gives little truly concrete guidance and allows the court great flexibility in weighing these factors. When the Daubert decision was handed down, many legal analysts and scientists agreed that the use of experts in court had long been a mess. Some critics lambasted judges for too often permitting expert testimony that wasn’t scientifically credible; others worried that juries were fundamentally incapable of making reasoned decisions when competing experts offered wildly different, contradictory testimony. Daubert shined a much-needed spotlight on expert evidence—but 30 years later, what has been its evidentiary impact? In civil cases, Daubert has, as a general matter, increased judicial scrutiny of experts such as epidemi-
ologists, economists, physicians, and financial experts, among others. Parties now challenge the admissibility of experts far more regularly, and both empirical analysis and lawyers’ reported experiences suggest that more expert testimony is excluded because of Daubert’s judicial gatekeeping regime. To be sure, whether courts are making accurate decisions about validity is a harder question to answer, and judges retain enormous discretion when ruling on admissibility of experts. But on the civil side, Daubert has generally raised the bar. Unfortunately, there has been far less real change in criminal cases. Many kinds of forensic evidence, from fingerprints to bloodstain pattern analysis to firearms identification, continue to enter court with remarkably little scientific scrutiny or proof of accuracy and validity. The National Academy of Sciences in 2009 and the President’s Council of Advisors on Science and Technology in 2016 sounded alarms about the inadequately developed scientific foundation of forensic pattern evidence. These reports articulated the acute need for sound scientific research on forensic techniques and cast substantial doubts on the adequacy of current evidence of validity. These techniques generally lack a well-grounded statistical foundation, and knowledge of actual error rates is scant. Moreover, most crime labs are not independent from law enforcement. Legal actions based on Daubert have been less frequent in criminal cases (partly because public defenders are less likely to have resources to mount effective challenges), and even when raised, have led to little concrete change. Because Daubert permits so much discretion, judges have often taken a lightweight approach to assessing forensic science, emphasizing “general acceptance” of a technique by the practicing community or the experience of the testifying examiner, instead of seriously addressing scientific validity. Other judges rely on outdated precedent rather than taking a clear-eyed look at the evidence. The time is right for a national commission that includes leading judges, scientists, legal academics, and forensic practitioners to jointly develop a framework to ensure that forensic science used in the courtroom is valid and reliable. Science and the legal system must expect much more from evidence if Daubert’s gatekeeping mandate is to truly transform justice. –Jennifer Mnookin
Jennifer Mnookin is chancellor and professor of law at the University of Wisconsin–Madison, Madison, WI, USA. [email protected]
PHOTO: UNIVERSITY OF WISCONSIN–MADISON
“Many kinds of forensic evidence… enter court with remarkably little scientific scrutiny…”
10.1126/science.adm8834
SCIENCE science.org
1117Editorial_18234182.indd 741
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
741
11/14/23 5:22 PM
Editor-in-Chief Holden Thorp, [email protected]
BOARD OF REVIEWING EDITORS (Statistics board members indicated with S)
Executive Editor Valda Vinson Editor, Research Jake S. Yeston Editor, Insights Lisa D. Chong Managing Editor Lauren Kmec DEPUTY EDITORS Gemma Alderton (UK),Stella M. Hurtley (UK), Phillip D. Szuromi, Sacha Vignieri SR. EDITORS Caroline Ash (UK),
Michael A. Funk, Brent Grocholski, Di Jiang, Priscilla N. Kelly, Marc S. Lavine (Canada), Sarah Lempriere (UK), Mattia Maroso, Yevgeniya Nusinovich, Ian S. Osborne (UK), L. Bryan Ray, H. Jesse Smith, Keith T. Smith (UK), Jelena Stajic, Peter Stern (UK), Valerie B. Thompson, Brad Wible ASSOCIATE EDITORS Bianca Lopez, Sarah Ross (UK), Madeleine Seale (UK), Corinne Simonti, Yury V. Suleymanov, Ekeoma Uzogara SENIOR LETTERS EDITOR Jennifer Sills NEWSLETTER EDITOR Christie Wilcox RESEARCH & DATA ANALYST Jessica L. Slater LEAD CONTENT PRODUCTION EDITORS Chris Filiatreau, Harry Jach SR. CONTENT PRODUCTION EDITOR Amelia Beyna CONTENT PRODUCTION EDITORS Anne Abraham, Robert French, Julia Haber-Katris, Nida Masiulis, Abigail Shashikanth, Suzanne M. White SR. EDITORIAL MANAGERS Carolyn Kyle, Beverly Shields SR. PROGRAM ASSOCIATE Maryrose Madrid EDITORIAL ASSOCIATES Aneera Dobbins, Joi S. Granger, Lisa Johnson SR. EDITORIAL COORDINATORS Jeffrey Hearn, Alexander Kief, Ronmel Navas, Jerry Richardson, Alice Whaley (UK), Anita Wynn EDITORIAL COORDINATORS Maura Byrne, Clair Goodhead (UK), Isabel Schnaidt, Qiyam Stewart, Brian White ADMINISTRATIVE COORDINATOR Karalee P. Rogers ASI DIRECTOR, OPERATIONS Janet Clements (UK) ASI OFFICE MANAGER Victoria Smith ASI SR. OFFICE ADMINISTRATORS Dawn Titheridge (UK), Jessica Waldock (UK) COMMUNICATIONS DIRECTOR Meagan Phelan DEPUTY DIRECTOR Matthew Wright SENIOR WRITER Walter Beckwith WRITERS Joseph Cariz, Abigail Eisenstadt, Nyla Husain SENIOR COMMUNICATIONS ASSOCIATE Zachary Graber COMMUNICATIONS ASSOCIATES Kiara Brooks, Haley Riley, Mackenzie Williams, Sarah Woods
News Editor Tim Appenzeller NEWS MANAGING EDITOR John Travis INTERNATIONAL EDITOR David Malakoff DEPUTY NEWS EDITORS Rachel Bernstein, Shraddha Chakradhar, Elizabeth Culotta, Martin Enserink, David Grimm, Eric Hand, Kelly Servick, Matt Warren (Europe) SR. CORRESPONDENTS Daniel Clery (UK), Jon Cohen, Jeffrey Mervis, Elizabeth Pennisi ASSOCIATE EDITORS Jeffrey Brainard, Michael Price NEWS REPORTERS Adrian Cho,
Jennifer Couzin-Frankel, Jocelyn Kaiser, Rodrigo Pérez Ortega (Mexico City), Robert F. Service, Erik Stokstad, Paul Voosen, Meredith Wadman CONTRIBUTING CORRESPONDENTS Warren Cornwall, Andrew Curry (Berlin), Ann Gibbons, Sam Kean, Kai Kupferschmidt (Berlin), Andrew Lawler, Mitch Leslie, Virginia Morell, Dennis Normile (Tokyo), Elisabeth Pain (Careers), Charles Piller, Gabriel Popkin, Joshua Sokol, Richard Stone, Emily Underwood, Gretchen Vogel (Berlin), Lizzie Wade (Mexico City) CAREERS Katie Langin (Associate Editor) INTERN Phie Jacobs COPY EDITORS Julia Cole (Senior Copy Editor), Hannah Knighton, Cyra Master (Copy Chief) ADMINISTRATIVE SUPPORT Meagan Weiland
Creative Director Beth Rakouskas DESIGN MANAGING EDITOR Chrystal Smith GRAPHICS MANAGING EDITOR Chris Bickel PHOTOGRAPHY MANAGING EDITOR Emily Petersen MULTIMEDIA MANAGING PRODUCER Kevin McLean WEB CONTENT STRATEGY MANAGER Kara Estelle-Powers DESIGN EDITOR Marcy Atarod DESIGNER Noelle Jessup SENIOR SCIENTIFIC ILLUSTRATOR Noelle Burgess SCIENTIFIC ILLUSTRATORS Austin Fisher, Kellie Holoski, Ashley Mastin SENIOR GRAPHICS EDITOR Monica Hersher GRAPHICS EDITOR Drew An-Pham SENIOR GRAPHICS SPECIALISTS Holly Bishop, Nathalie Cary SENIOR PHOTO EDITOR Charles Borst PHOTO EDITOR Elizabeth Billman SENIOR PODCAST PRODUCER Sarah Crespi SENIOR VIDEO PRODUCER Meagan Cantwell SOCIAL MEDIA STRATEGIST Jessica Hubbard SOCIAL MEDIA PRODUCER Sabrina Jenkins WEB DESIGNER Jennie Pajerowski
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 MANAGER, BUSINESS OPERATIONS Jessica Tierney MANAGER, BUSINESS ANALYSIS Cory Lipman MANAGER, WEB ANALYTICS Samantha Cressman BUSINESS ANALYSTS Kurt Ennis, Maggie Clark FINANCIAL ANALYST Isacco Fusi BUSINESS OPERATIONS ADMINISTRATOR Taylor Fisher SENIOR PRODUCTION MANAGER Jason Hillman SENIOR MANAGER, PUBLISHING AND CONTENT SYSTEMS Marcus Spiegler CONTENT OPERATIONS MANAGER Rebecca Doshi SENIOR CONTENT & PUBLISHING SYSTEMS SPECIALIST Jacob Hedrick SENIOR PRODUCTION SPECIALIST Kristin Wowk PRODUCTION SPECIALISTS Kelsey Cartelli, Audrey Diggs DIGITAL PRODUCTION MANAGER Lisa Stanford SENIOR DIGITAL ADVERTISING SPECIALIST Kimberley Oster ADVERTISING PRODUCTION OPERATIONS MANAGER Deborah Tompkins DESIGNER, CUSTOM PUBLISHING Jeremy Huntsinger SR. TRAFFIC ASSOCIATE Christine Hall SPECIAL PROJECTS ASSOCIATE Shantel Agnew 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, Lorena Chirinos Rodriguez, Jenna Voris SENIOR DESIGNER Kim Huynh DIRECTOR AND SENIOR EDITOR, CUSTOM PUBLISHING Erika Gebel Berg ASSISTANT EDITOR, CUSTOM PUBLISHING Jackie Oberst PROJECT MANAGER Melissa Collins DIRECTOR, PRODUCT & PUBLISHING DEVELOPMENT Chris Reid DIRECTOR, BUSINESS STRATEGY AND PORTFOLIO MANAGEMENT Sarah Whalen DIRECTOR, PRODUCT MANAGEMENT Kris Bishop PRODUCT DEVELOPMENT MANAGER Scott Chernoff PUBLISHING PLATFORM MANAGER Jessica Loayza SR. PRODUCT ASSOCIATE Robert Koepke PRODUCT ASSOCIATES Caroline Breul, Anne Mason ASSOCIATE DIRECTOR, INSTITUTIONAL LICENSING MARKETING Kess Knight BUSINESS DEVELOPMENT MANAGER Rasmus Andersen ASSOCIATE DIRECTOR, INSTITUTIONAL LICENSING SALES Ryan Rexroth INSTITUTIONAL LICENSING MANAGER Marco Castellan, Claudia Paulsen-Young SENIOR MANAGER, INSTITUTIONAL LICENSING OPERATIONS Judy Lillibridge SENIOR OPERATIONS ANALYST Lana Guz SYSTEMS & OPERATIONS ANALYST Ben Teincuff FULFILLMENT ANALYST Aminta Reyes ASSOCIATE DIRECTOR, US ADVERTISING 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
Rie Rambelli (Tokyo), Miyuki Tani (Osaka) DIRECTOR, COPYRIGHT, LICENSING AND SPECIAL PROJECTS Emilie David RIGHTS AND PERMISSIONS ASSOCIATE Elizabeth Sandler LICENSING ASSOCIATE Virginia Warren RIGHTS AND LICENSING COORDINATOR Dana James CONTRACT SUPPORT SPECIALIST Michael Wheeler EDITORIAL
MEDIA CONTACTS
[email protected]
[email protected]
NEWS
PRODUCT ADVERTISING & CUSTOM PUBLISHING
[email protected] INFORMATION FOR AUTHORS
science.org/authors/ science-information-authors
science.org/subscriptions MEMBER BENEFITS
advertising.science.org/ aaas.org/membership/ products-services benefits [email protected] CLASSIFIED ADVERTISING
REPRINTS AND PERMISSIONS
MEMBERSHIP AND INDIVIDUAL SUBSCRIPTIONS
INSTITUTIONAL SALES AND SITE LICENSES
science.org/help/ reprints-and-permissions
advertising.science.org/ science.org/librarian science-careers [email protected] AAAS BOARD OF DIRECTORS
MULTIMEDIA CONTACTS
JOB POSTING CUSTOMER SERVICE
[email protected] [email protected]
employers.sciencecareers.org [email protected]
CHAIR Gilda A. Barabino PRESIDENT Keith R. Yamamoto PRESIDENT-ELECT Willie E. May
TREASURER Carolyn N. Ainslie CHIEF EXECUTIVE OFFICER
Sudip Parikh BOARD Cynthia M. Beall Janine Austin Clayton Kaye Husbands Fealing Kathleen Hall Jamieson Jane Maienschein Robert B. Millard Babak Parviz William D. Provine Juan S. Ramírez Lugo Susan M. Rosenberg Vassiliki Betty Smocovitis
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.
742
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
Erin Adams, U. of Chicago Takuzo Aida, U. of Tokyo Leslie Aiello, Wenner-Gren Fdn. Deji Akinwande, UT Austin James Analytis, UC Berkeley Paola Arlotta, Harvard U. Delia Baldassarri, NYU Nenad Ban, ETH Zürich Christopher Barratt, U. of Dundee Franz Bauer, Pontificia U. Católica de Chile Ray H. Baughman, UT Dallas Carlo Beenakker, Leiden U. Yasmine Belkaid, NIAID, NIH Kiros T. Berhane, Columbia U. Joseph J. Berry, NREL Alessandra Biffi, Harvard Med. Chris Bowler, École Normale Supérieure Ian Boyd, U. of St. Andrews Malcolm Brenner, Baylor Coll. of Med. Emily Brodsky, UC Santa Cruz Ron Brookmeyer, UCLA (S) Christian Büchel, UKE Hamburg Johannes Buchner, TUM 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 Wendy Cho, UIUC Ib Chorkendorff, Denmark TU Chunaram Choudhary, Københavns U. Karlene Cimprich, Stanford U. Laura Colgin, UT Austin James J. Collins, MIT Robert Cook-Deegan, Arizona State U. Virginia Cornish, Columbia U. Carolyn Coyne, Duke U. Roberta Croce, VU Amsterdam Molly Crocket, Princeton U. Christina Curtis, Stanford U. Ismaila Dabo, Penn State U. Jeff L. Dangl, UNC Nicolas Dauphas, U. of Chicago Frans de Waal, Emory U. Claude Desplan, NYU Sandra DÍaz, U. Nacional de CÓrdoba Samuel Díaz-Muñoz, UC Davis 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 A. Ehlers, U. of Glasgow Nader Engheta, UPenn Tobias Erb, MPS, MPI Terrestrial Microbiology Karen Ersche, U. of Cambridge Beate Escher, UFZ & U. of Tübingen Barry Everitt, U. of Cambridge Vanessa Ezenwa, U. of Georgia Toren Finkel, U. of Pitt. Med. Ctr. Natascha Förster Schreiber, MPI Extraterrestrial Phys. Peter Fratzl, MPI Potsdam Elaine Fuchs, Rockefeller U. Caixia Gao, Inst. of Genetics and Developmental Bio., CAS Daniel Geschwind, UCLA Lindsey Gillson, U. of Cape Town Gillian Griffiths, U. of Cambridge Simon Greenhill, U. of Auckland Nicolas Gruber, ETH Zürich Hua Guo, U. of New Mexico
Taekjip Ha, Johns Hopkins U. Daniel Haber, Mass. General Hos. Sharon Hammes-Schiffer, Yale U. Wolf-Dietrich Hardt, ETH Zürich Louise Harra, UCL Kelley Harris, U. of Wash Carl-Philipp Heisenberg, IST Austria Christoph Hess, U. of Basel & U. of Cambridge Heather Hickman, NIAID, NIH Hans Hilgenkamp, U. of Twente Janneke Hille Ris Lambers, ETH Zürich Kai-Uwe Hinrichs, U. of Bremen Deirdre Hollingsworth, U. of Oxford Christina Hulbe, U. of Otago, New Zealand Randall Hulet, Rice U. Auke Ijspeert, EPFL Gwyneth Ingram, ENS Lyon Darrell Irvine, MIT Akiko Iwasaki, Yale U. Erich Jarvis, Rockefeller U. Peter Jonas, IST Austria Matt Kaeberlein, U. of Wash. Daniel Kammen, UC Berkeley Kisuk Kang, Seoul Nat. U. V. Narry Kim, Seoul Nat. U. Nancy Knowlton, Smithsonian Etienne Koechlin, École Normale Supérieure Alex L. Kolodkin, Johns Hopkins U. LaShanda Korley, U. of Delaware Paul Kubes, U. of Calgary Chris Kuzawa, Northwestern U. Laura Lackner, Northwestern U. Gabriel Lander, Scripps Res. (S) Mitchell A. Lazar, UPenn Hedwig Lee, Duke U. Fei Li, Xi'an Jiaotong U. Ryan Lively, Georgia Tech Luis Liz-Marzán, CIC biomaGUNE Omar Lizardo, UCLA Jonathan Losos, WUSTL Ke Lu, Inst. of Metal Res., CAS Christian Lüscher, U. of Geneva Jean Lynch-Stieglitz, Georgia Tech David Lyons, U. of Edinburgh Fabienne Mackay, QIMR Berghofer Zeynep Madak-Erdogan, UIUC Vidya Madhavan, UIUC Anne Magurran, U. of St. Andrews Ari Pekka Mähönen, U. of Helsinki Asifa Majid, U. of Oxford Oscar Marín, King’s Coll. London Charles Marshall, UC Berkeley Christopher Marx, U. of Idaho David Masopust, U. of Minnesota Geraldine Masson, CNRS Jennifer McElwain, Trinity College Dublin Rodrigo Medellín, U. Nacional Autónoma de México C. Jessica Metcalf, Princeton U. Tom Misteli, NCI, NIH Jeffery Molkentin, Cincinnati Children's Hospital Medical Center Alison Motsinger-Reif, NIEHS, NIH (S) Danielle Navarro, U. of New South Wales Daniel Neumark, UC Berkeley Thi Hoang Duong Nguyen, MRC LMB Beatriz Noheda, U. of Groningen Helga Nowotny, Vienna Sci. & Tech. Fund Pilar Ossorio, U. of Wisconsin Andrew Oswald, U. of Warwick Isabella Pagano, Istituto Nazionale di Astrofisica Giovanni Parmigiani, Dana-Farber (S) Sergiu Pasca, Standford U. Daniel Pauly, U. of British Columbia Ana Pêgo, U. do Porto 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 Lei Stanley Qi, Stanford U. Simona Radutoiu, Aarhus U. Trevor Robbins, U. of Cambridge Joeri Rogelj, Imperial Coll. London John Rubenstein, SickKids 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, UCL Emma Slack, ETH Zürich & U. of Oxford Richard Smith, UNC (S) John Speakman, U. of Aberdeen Allan C. Spradling, Carnegie Institution for Sci. V. S. Subrahmanian, Northwestern U. Sandip Sukhtankar, U. of Virginia Naomi Tague, UC Santa Barbara Eriko Takano, U. of Manchester A. Alec Talin, Sandia Natl. Labs 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 Li-Huei Tsai, MIT Jason Tylianakis, U. of Canterbury Matthew Vander Heiden, MIT Wim van der Putten, Netherlands Inst. of Ecology Ivo Vankelecom, KU Leuven Judith Varner, UC San Diego Henrique Veiga-Fernandes, Champalimaud Fdn. Reinhilde Veugelers, KU Leuven Bert Vogelstein, Johns Hopkins U. Julia Von Blume, Yale School of Med. David Wallach, Weizmann Inst. Jane-Ling Wang, UC Davis (S) Jessica Ware, Amer. Mus. of Natural Hist. David Waxman, Fudan U. Alex Webb, U. of Cambridge Chris Wikle, U. of Missouri (S) Terrie Williams, UC Santa Cruz Ian A. Wilson, Scripps Res. (S) Hao Wu, Harvard U. Li Wu, Tsinghua U. Amir Yacoby, Harvard U. Benjamin Youngblood, St. Jude Yu Xie, Princeton U. Jan Zaanen, Leiden U. Kenneth Zaret, UPenn School of Med. Lidong Zhao, Beihang U. Bing Zhu, Inst. of Biophysics, CAS Xiaowei Zhuang, Harvard U. Maria Zuber, MIT
science.org SCIENCE
EDITORIAL
Correction is courageous
I
n a year when disagreements over scientific matters like COVID-19 continue to occupy political discourse, the surfacing of a spate of high-profile research errors is regrettable. It’s crucial that the public trusts science at a time when so many topics—artificial intelligence, climate change, and pandemics—cast shadows of uncertainty on the future. Errors, intentional or not, erode confidence in science. It’s not surprising that science integrity has become a focal point for major institutions in the United States, from the White House to the National Institutes of Health. Evaluating policies on misconduct is essential, but the idea of a scientific ecosystem that is free of errors is an unattainable utopia. However, evolving a more responsive ecosystem is entirely possible, and scientific journals, institutions, and researchers must together move more intentionally in this direction. Efforts to minimize external attacks on science have included getting scientists to refrain from talking about divisive politics and strengthening processes for catching research errors. Neither has worked well. Researchers are human and therefore have opinions. They also are fallible. The same holds true for journal editors and peer reviewers of research papers. A better approach is to strengthen the process for correcting errors in the scientific record with expedience and transparency. A recent study by Kathleen Hall Jamieson and colleagues suggests that public trust of such a system would go a long way in building trust across ideologies. Unfortunately, the sluggishness of journals to correct errors, the silence of research institutions regarding alleged errors, and the defensiveness of researchers have created an environment that is difficult to change. The conflation of correcting the scientific record and assigning responsibility for misconduct is a barrier to a more responsive ecosystem. Most journal policies are explicit that error corrections do not necessarily reflect misconduct. Nonetheless, the stigma that researchers associate with corrections and retractions, regardless of whether misconduct is acknowledged, is an obstacle to getting researchers, institutions, and journals to collaborate on better handling errors. One solution is to increase the use of a mechanism that most journals have of issuing a notice of concern that alerts readers to a possible correction or retraction. This can ameliorate
criticisms about the sluggishness of existing processes. In the past four years, Science has doubled the rate at which it issues an Editorial Expression of Concern. Unfortunately, authors and institutions often contest such notices, even though they can be removed if concerns are resolved or a correction is posted. The journal PLOS ONE was recently sued over a Notice of Concern. In this case, an author of a published paper contacted the journal to request a correction. PLOS ONE decided to post a notice while the correction was being investigated and finalized. PLOS ONE, like Science, is clear in its policies that an alert does not assert misconduct. Nevertheless, another author of the paper sued to prohibit the notice because she claimed it would damage her reputation. The lawsuit argued that a Notice of Concern implies intentional misconduct or that something “otherwise nefarious” took place. This supposed stigma is not supported by studies on the effects of retractions on future citations. As Ivan Oransky of Retraction Watch told me, “The data are clear: When retraction notices describe exactly why papers are retracted, researchers whose work is retracted for honest error do not see a reputation hit, as measured by citations to their work.” The Jamieson study tested several variables associated with support for science funding in the United States among liberals and conservatives. The strongest association with ideology was for a parameter called “Unbiased,” which does not require that scientists be free of expressing political views. The authors noted that scientists are passionate humans and subject to confirmation biases. Therefore, their analysis asked whether respondents perceive that scientists overcome their human biases and offer unbiased findings. They found that support for funding science would likely increase if the public believes that science has protections in place to reduce the impact of human bias on findings. The more that conflicts within the scientific community are associated with correcting the scientific record, the less trust the public will have in the scientific enterprise. Editorial notices of possible errors are by no means the total solution to strengthening integrity in science, but they represent an important way to help uphold the self-correcting nature of science. –H. Holden Thorp
H. Holden Thorp Editor-in-Chief, Science journals. [email protected]
PHOTO: CAMERON DAVIDSON
“…a scientific ecosystem that is free of errors is an unattainable utopia.”
Published online 9 November 2023; 10.1126/science.adm8205
SCIENCE science.org
1117Editorial_18234182.indd 743
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
743
11/14/23 5:22 PM
Advertorial
The worldís first AI university pairs machine learningís great power with great responsibility Professor Eric Xing, MBZUAI president
As the first institution exclusively focused on graduate-level training in artificial
“Public perception of AI has been mostly driven by tools that not only represent
intelligence (AI), the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)
very limited examples of the capabilities of generative AI models but also afford little
aims to contribute to the rapidly developing field of AI. Its goals for the effective and
insight into how they work or why they often do not,” Xing explains. Unfamiliarity
responsible advancement of AI perfectly position MBZUAI at an inflection point in
with AI’s subtle limitations and incomplete understanding of how it generates
human history.
outputs represents a “black box,” forming the basis of much anxiety and fear.
In 2023, Morgan Stanley analysts described the potential impact of AI on the
As both a clinician and computer scientist, Dr. Michael Matheny, MD, MS, MPH,
world economy as a $6 trillion USD opportunity. AI’s disruption of the status quo
director of Vanderbilt University’s Center for Improving the Public’s Health through
has been well-documented. However, realizing its true potential will require minds
Informatics, is acutely aware of that tension. He says it stems from embracing AI
specifically trained in AI development and deployment, and appreciative of its power
without a rudimentary understanding of how it works and—importantly—an ability
and possibility.
to recognize when it doesn’t.
“The magnitude of AI-driven disruption mirrors that proceeding from the Age of
“Failure to give adequate weight to the word ‘artificial’ while focusing too much
Enlightenment,” says Professor Eric Xing, MBZUAI president. “This technology has
on a poorly defined and inaccurate representation of ‘intelligence’ is a common trap
initiated paradigm shifts.”
in how the public views AI,” he says.
An AI-driven renaissance
supervised manner. Bias stems from biased training data, exacerbated through user
Xing predicts the future impact of AI to parallel the impact of Newton’s Principia
interactions, and supported by reinforcement learning.
Incredibly subtle errors can arise in the output of models trained in a specifically
Mathematica and the Scientific Revolution thereafter. Xing predicts the future impact of AI to parallel the impact of Newton’s Principia Mathematica and the Scientific
Answering a call for responsible innovation
Revolution thereafter. In this regard, machine learning and deep neural networks
Matheny’s experience with designing AI healthcare demonstrated how much
(DNNs) have driven discoveries and established new capabilities across a broad
more complicated the AI development process needs to be compared to the
range of tasks. DNNs can process data and offer predictions in minutes that would
typical software development lifecycle. Beyond simply determining whether an
take humans years. For example, the DNN AlphaFold’s breakthroughs in predicting
AI-based solution is warranted, there are ethical questions regarding responsibility,
protein structures are widely recognized as solving the protein folding problem—a
accountability, and transparency.
Xing suggests that AI presents humans with another opportunity to approach
“Building these tools requires a holistic understanding of the questions that need to be addressed before starting,” advises Matheny. “In a world where AI
problems in ways not yet fully formalized. However, he compares the rapid pace of
has garnered labels as both the solution to and potential cause of many human
progress to the idea of building the plane while flying it.
problems, training developers to be both technically proficient and cognizant of the social impact of their creation takes on an unprecedented level of importance.”
Disruption at the speed of computation
To this end, Xing explains that MBZUAI’s focus on furthering AI research and
The speed and power of AI have fostered varying degrees of amazement, confusion,
education attempts to balance a desire to optimize its utilization with healthy
and fear. Large language models (LLMs) like ChatGPT and commercial tools
discretion. “A central tenet of our mission is to train future leaders in AI to
like chatbots do not provide sufficient context to understand their benefits and
understand and accept the responsibility that comes with stewardship of something
limitations.
this powerful and transformative.”
1117Product.indd 744
PHOTO: PROVIDED BY MBZUAI
“grand challenge” of biochemistry that mystified scientists for more than 50 year.
11/9/23 12:32 PM
Produced by the Science
Fighting fake news
Increasing the efficiency of AI tools
Weaponizing language to (mis)shape public perceptions is an old story with
address the environmental impact associated with training and updating LLMs.
new chapters. Unfortunately, AI is contributing to that story at a speed and
Their models (e.g., Vicuna, Jais, and foundation models like GET for biology problems)
scale that has proven difficult to manage.
employ a type of parallelism that efficiently shares the training workload between
Healthy skepticism is especially important in our current age of misinformation. However, the advent of LLMs offers new challenges for
MBZUAI faculty and their stateside collaborators have developed methods to
multiple devices through adaptively optimized communications between them, resulting in up to 50% greater computational efficiency.
fact-checkers. Preslav Nakov, department chair of Natural Language Processing at MBZUAI, studies how AI can help identify linguistic patterns
Optimizing energy grids: MBZUAI is partnering with the Global Energy
that raise red flags for fake news.
Interconnection Research Institute to optimize the integration of renewable energy
“The way that LLMs are trained does not optimize them for factuality,” he advises. To generate text, LLMs perform next-word prediction by sampling among the top choices available. Given the rise of machine-generated content online, there is also a danger that LLMs use such data in their training, which will only make the problem worse. “Because malicious users can impact how LLMs answer specific questions, fixing the factuality problem is a very active research area.”
resources into existing power grids. They aim to decrease current grids’ carbon footprint while effectively balancing the use of both conventional and green energy. Transforming global health and biomedical research Climate change and disease eradication: Partnerships with Malaria No More and Reaching the Last Mile are helping MBZUAI forecast climate trends. The predictive capacity of AI allows healthcare workers to intervene sooner during disease outbreaks, thereby preventing or limiting the spread of contagions.
Training systems to detect stylistic anomalies that separate real news from fake news or even satire is one area of Nakov’s research that has
Revolutionizing personalized medicine: Researchers are applying AI for a Human
shown significant promise. “Although both fake news and satire are
Phenotype Project and the Emirati Genome Program to improve patient health
technically false, the former specifically intends to deceive,” he explains.
according to their specific phenotype and molecular profiles.
Moreover, the imagination required to generate ‘good’ fake news and satire remains uniquely human, thereby certainly making it easier to recognize
Adapting education to a new age
machine-generated content.
AI has the potential to radically alter education methods. Xing explains that using AI
Notably, LLMs also lack linguistic nuance. “The intent of fake news is not just to lie but also to persuade,” says Nakov. “Although LLMs can produce false information, they do not understand how to persuade.” In recent work, Nakov and his colleagues inserted persuasion techniques into the text-generation process in an LLM. The output was more persuasive fake news, which was then used as synthetic training data to help detect human-generated fake news. Nakov’s group is also developing tools to perform real-time fact-checking of text generated by LLMs that will trigger edit operations to fix the output before its delivery. Although the specter of AI-generated fake news may seem insurmountable,
to solve problems doesn’t necessarily equate with understanding how the problem was solved. “Calculators did not remove the requirement to learn mathematical techniques, and the printing press did not eliminate the need for writing skills,” he says. “AI may force educators to reconsider the best ways to introduce or reinforce concepts, but I see this as beneficial for both the teachers and those being taught.” Importantly, AI-powered learning can improve education access in remote areas and offer a more personalized learning approach for those in need. MBZUAI is also engaged in driving innovation in this area, which might ultimately have the greatest impact on how the world adapts to AI. “Humans excel at adaptation, and creating a symbiotic relationship with AI may
AI techniques can also play a powerful role in correcting misinformation.
be the ultimate goal during this evolutionary period,” explains Xing. “At MBZUAI,
For Nakov, this magnifies the importance of MBZUAI and its mission. “The
this starts with attracting and educating brilliant students eager to tackle difficult
roles played by LLMs continue to expand,” he says. “This emphasizes the
problems.”
critical nature of our ambitious research and training goals in optimizing the accuracy of the information that they provide.”
Sponsored by
Harnessing AI to address global challenges Establishing worldwide collaborations was a major goal for Xing when planning to bring MBZUAI onto the global stage of AI-driven scientific research. “Our focus on open collaboration with other institutes and researchers has attracted like-minded scientists and technologists that are the faculty of MBZUAI,” he says. “Our size allows flexibility, which has translated into amazing opportunities with global partners.”
1117Product.indd 745
11/9/23 12:32 PM
NEWS
“
I ran … to the living room and hugged the guys.
”
Biologist James Kempton, in The Guardian, after a camera trap set up by his team revealed an image of an Attenborough’s long-beaked echidna, named for naturalist David Attenborough and not confirmed in the wild since 1961. The image g was taken this summer duringg the team’s expedition p in Papua p New Guinea.
IN BRIEF Edited by Jeffrey Brainard
The Allen Telescope Array in California is the first radio telescope designed for the search for extraterrestrial intelligence.
ASTRONOMY
Extraterrestrial intelligence hunt gets big financial boost
Young shooting survivors suffer H E A LT H | Firearm shootings are the leading cause of death of children and adolescents in the United States, and survivors face a lasting burden. A study has found that even after their gunshot injuries heal, they endure health complications and
746
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117NIB_18270015.indd 746
search at optical wavelengths. His bequest to the SETI Institute far exceeds the funding of Breakthrough Listen, a 10-year SETI project financed with $100 million from tech investor Yuri Milner. Antonio’s gift will allow the SETI Institute, which had income of $28 million last year, to branch out from core programs, according to an institute statement. The gift is one of the largest ever to a scientific organization unaffiliated with a university, according to a list compiled by The Chronicle of Philanthropy.
financial costs much higher than people their age who are not shot. In one of the first such analyses, researchers used 14 years’ worth of health insurance claims data to compare the shooting survivors, ages 19 and younger, with a control population. They found survivors have on average more than twice as many pain
disorders, a 68% increase in psychiatric disorders, and a 144% increase in substance use disorders by 1 year after the incident. The cost of their health care ballooned by $34,884 on average, 17 times more than for controls. Their parents also had a 30% increase in psychiatric disorders, the researchers reported last week in
PHOTO: BEN MARGOT/AP
T
he search for extraterrestrial intelligence, or SETI, received a huge gain last week: a gift of $200 million from the estate of Franklin Antonio, co-founder of Qualcomm, which makes semiconductors and software supporting wireless technology. Antonio, who died last year, was a longtime supporter of the field. He served as the primary benefactor of SETI research on the Allen Telescope Array, a set of radio dishes that listen for alien activity, and also backed PANOSETI, an all-sky
science.org SCIENCE
11/14/23 5:21 PM
Health Affairs. They recommend expanding screening for mental health conditions among those affected by the shootings.
Budget woe slows Mars mission | NASA has slowed development of its Mars Sample Return (MSR) mission, citing uncertainty over its funding and design, agency officials said this week. The move comes after a recent independent review found the bid to retrieve Mars rocks could cost $8 billion to $11 billion—more than the U.S. Senate appears willing to provide. NASA says it will pause work on a spacecraft element meant to capture the rock samples once they reach orbit around Mars, before they are returned to Earth. Instead, the agency will focus on the project’s first leg, a lander to collect the samples and rocket them to the rendezvous. The agency says it expects to finalize revisions to its MSR plans by the spring of 2024; work on the return system could resume once the mission’s future is clear. P L A N E TA RY S C I E N C E
A research watchdog for Australia | Australia should create an independent, government-funded body to investigate research misconduct, a new report recommends. “Australia is one of the few countries with a developed research sector that does not have a research integrity watchdog,” instead relying on institutions to investigate their own scientists, the Australia Institute, a think tank, noted in its 12 November report. Those probes are typically kept secret, it notes, enabling institutions to “sweep matters of research integrity under the rug.” (Institutions in the United States and some other countries conduct their own investigations but must report results to independent oversight bodies.) The report says the watchdog should make its findings public and institutions should be bound by those findings. Australia’s new body should also establish a “clear and enforceable” definition of research misconduct. Australian universities have been skeptical of the need for an independent investigative body, but The Sydney Morning Herald reports they have signaled support for the new approach.
PHOTOS: (LEFT TO RIGHT) YAD VASHEM, THE LANCET; GHETTO FIGHTERS ARCHIVE, THE LANCET
RESEARCH INTEGRITY
China avoids methane promise C L I M AT E P O L I C Y
| Disappointing climate
specialists, China released a plan last week for reducing methane emissions that lacks targets, such as the 30% drop by 2030 promised by the United States and 150 other countries. The blueprint from SCIENCE science.org
1117NIB_18270015.indd 747
Physicians Anna Braude Heller (left) and Israel Milejkowski resisted the Nazi regime in the Warsaw Ghetto.
BIOETHICS
Teaching lessons from the Nazi medical era
A
ll health care students worldwide should learn the history of medicine during the Nazi regime and the Holocaust, according to an expert commission sponsored by The Lancet. In a report published last week, the commission also recommended an international organization focused on the topic and a digital library accessible in multiple languages to health care students around the world. Although the report highlights some doctors, nurses, and midwives who worked against the regime’s murderous practices, the medical profession had one of the highest rates of Nazi party membership; more than half of Germany’s non-Jewish doctors joined the party. Doctors participated in human experimentation in concentration camps and in “euthanasia” programs that murdered more than 200,000 people deemed mentally unfit. Including the topic in medical education could “counteract an ever-present risk of medical injustices” and “the tendency to objectify patients and research participants,” the commission’s cochairs told Science. The full interview is available at https://scim.ag/NaziMed.
China’s environment ministry did include intentions to capture and reuse more of the short-lived greenhouse gas, which traps heat far more effectively than carbon dioxide. China is the world’s largest emitter of methane, in part because it mines large amounts of coal, releasing methane in the process. The country had promised at a 2021 climate summit in Glasgow, Scotland, to cooperate on reducing methane. The issue is expected to arise at a climate summit next month in the United Arab Emirates.
U.S. reports climate risks, gains | Global warming is changing life for everyone in the United States, but marginalized communities are feeling the worst of it, according to the latest U.S. National Climate Assessment, released on 14 November. The congressionally mandated report, published
C L I M AT E P O L I C Y
every 4 years, lays out how the country is warming faster than the global average, with a familiar litany of impacts: extreme rainfall battering the Northeast, floods intruding on freshwater aquifers in the South, acidifying oceans threatening fisheries off Alaska’s coast. Many climate impacts, such as wildfire smoke and inland flooding, have the largest effect on those least able to avoid air pollution or afford alternative housing. The U.S. is responsible for 17% of current global warming, the report finds, and is still not on track to cut its greenhouse gas emissions enough to meet its international commitments. But progress on climate adaptation is being made, the assessment adds. Pittsburgh is adjusting construction codes so new buildings better handle stormwaters, for example. Northwest tribes are managing forests for carbon retention. And renewable power stations are sprouting across the country. 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
747
11/14/23 5:21 PM
NEWS
IN DEP TH
Ten-day artificial intelligence forecasts can match official predictions in a fraction of the time.
ARTIFICIAL INTELLIGENCE
AI is set to revolutionize weather forecasts By Paul Voosen
M
eteorologists call it the “quiet revolution”: a gradual but steady improvement in weather forecasting. Today, the 6-day forecast is about as good as the 3-day forecast from 30 years ago. Rarely do severe storms or heat waves catch people unaware. This revolution has saved lives and money, but it also comes with a cost: billions of dollars’ worth of energy-hungry supercomputers that must run 24/7 just to produce a few forecasts a day. Artificial intelligence (AI) is now spurring another revolution within numerical weather prediction, as the field is known. In mere minutes on cheap desktop computers, trained AI systems can now make 10-day forecasts that are as good as the best traditional models— and in some cases even better. The world’s top weather agency, the European Centre for Medium-Range Weather Forecasts (ECMWF), has embraced the technology: Last month it began to generate its own experimental AI forecasts. The algorithms could enable more frequent forecasts and free up computing resources for other thorny problems. “It’s very, very exciting to know we can generate global predictions that are skillful, really cheaply,” says Maria Molina, an AI-focused research meteorologist at the University of Maryland. Some of the world’s biggest tech giants are jockeying to claim the most skillful model, 748
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117NewsInDepth_18261920.indd 748
including Google DeepMind, which describes its GraphCast model in Science this week, and Huawei, which published a similar model, called Pangu-Weather, in Nature earlier this year. Google also has a short-term AI weather model that makes rolling 24-hour predictions that are more accurate than nearly any weather agency’s. It’s incredible progress on a task that was thought infeasible just a few years ago, says Aditya Grover, an AI researcher at the University of California, Los Angeles. “From a technology standpoint, we have all the ingredients in place.” Traditional weather models start by feeding a snapshot of current conditions, based on observations from satellites, weather stations, balloons, and buoys, into a gridlike computer model that divides the atmosphere into millions of boxes. The snapshot is run forward in time by applying the physical laws of fluid dynamics to each box—at great computational expense. The models can take several hours to run on supercomputers with 1 million processors, and weather agencies typically produce updates just four times a day. The new AI models skip the expense of solving equations in favor of “deep learning.” They identify patterns in the way the atmosphere naturally evolves, after training on 40 years of ECMWF “reanalysis” data—a combination of observations and short-term model forecasts that represents modelers’ best and most complete picture of past weather. When fed
a starting snapshot of the atmosphere based on the same combination of observations and modeling, GraphCast can outperform the ECMWF forecast out to 10 days on 90% of its verification targets, including hurricane tracks and extreme temperatures. Although it took 32 computers 4 weeks to train the AI model, the resulting algorithm is lightweight enough to work in less than 1 minute on a single desktop computer, says Rémi Lam, lead author of the GraphCast paper. “It is fast, accurate, and useful.” These benefits seem to hold even in more realistic settings. Earlier this year, ECMWF researchers ran Pangu, feeding it only the observations that go into its operational weather model. Those observations offer a more limited picture of the atmosphere than the reanalysis snapshots used to test GraphCast. The skill of Pangu’s forecast was similar to ECMWF’s main model, although its predictions of rainfall and other fine-scale features were slightly fuzzier. “It was an even playing field,” says Zied Ben Bouallègue, who led the analysis, released as an arXiv preprint in July. “We were surprised to see the good results.” These advances came startlingly fast. A key step came in 2020, when a group led by Stephan Rasp, now also at Google, created WeatherBench, which made the ECMWF reanalysis data easy to digest and also, to provoke competition, provided a benchmark for measuring forecast skill. In 2022, after a few
PHOTO: NATIONAL OCEANIC AND ATMOSPHERIC ADMINISTRATION VIA AP
Cheap and fast algorithms are matching—and surpassing—the world’s top models
science.org SCIENCE
11/14/23 5:26 PM
PHOTO: A. CHO/SCIENCE
NE WS
months of work during a sabbatical, Ryan Keisler, a physicist now at KoBold Metals, a mineral exploration company, published a preprint describing a simple model with considerable skill in 6-day forecasts. “Given how much historical data there was to learn from, it just had to work at some level,” Keisler says. A next step will be to produce ensemble results, a forecasting innovation that helps capture uncertainty by running a model multiple times to create a range of possible outcomes. AI researchers could follow the traditional technique of tweaking initial weather conditions just slightly before each model run, or they could adapt the AI generative techniques making waves in text and image generation to create tweaked conditions on the fly. “I’m pretty sure every group is working on that,” Rasp says. Such ensemble forecasts could help the AI models better predict extreme events, such as strong hurricanes, that they currently underestimate in intensity. To improve further, the AI models could be weaned off the reanalysis data, which carry the biases of traditional models. Instead, they could learn directly from the petabytes of raw observation data held by weather agencies, Keisler says. Google’s short-term weather model already does so, training itself on data from weather stations, radar, and satellites. The potential for these models doesn’t stop at weather prediction, says Christopher Bretherton, an atmospheric scientist at the Allen Institute for AI. They cannot project climate on their own, because the 40-year training data sets are not long enough to capture global warming trends, which are subject to complex feedbacks from clouds, gases, and aerosols that can accelerate or slow climate change. But they could assist a new generation of high-resolution climate models being developed to run on exascale computers, the latest ultrafast machines. Once those models produce enough output for the AIs to be trained on, the AIs could take over. “We can make emulators of these models and then run them 100 times faster,” Bretherton says. Few expect traditional forecasts to disappear anytime soon, but AI is “rapidly approaching the point where it could be a useful complement,” says Matthew Chantry, who coordinates ECMWF’s AI work. Adoption might be slowed by unease about the black-box nature of the AI: Researchers often can’t say how such systems reach their conclusions. But that concern can be overstated, says Chantry, who notes that traditional models are also so complicated that “there’s a degree of opaqueness already built into them.” Ultimately, it will come down to users, Grover says. “If you’re a farmer in the field, would you care about the more accurate forecast, or the one you can write down with physical equations?” j SCIENCE science.org
1117NewsInDepth_18261920.indd 749
ENERGY
Deal to build pint-size nuclear reactors is canceled NuScale Power’s small modular reactors promised cheaper nuclear power, but costs soared and utilities balked By Adrian Cho
A
plan to build a novel nuclear power plant comprising six small modular reactors (SMRs) fell apart last week when prospective customers for its electricity backed out. Utah Associated Municipal Power Systems (UAMPS), a coalition of community-owned power systems in seven western states, withdrew from a deal to build the plant, designed by NuScale Power, because too few members agreed to buy into it. The project, subsidized by the U.S. Department of Energy (DOE), sought to revive the moribund U.S. nuclear industry, but its cost had more than doubled to $9.3 billion. “We still see a future for new nuclear,” says Mason Baker, CEO and general manager of UAMPS, which planned to build the plant in Idaho. “But in the near term, we’re going to focus on … expanding our wind capacity, doing more utility-scale solar, [and] batteries.” NuScale, which was spun out of Oregon State University in 2007, declined to make anyone available for an interview. But David Schlissel of the Institute for Energy Economics and Financial Analysis says, “The communities and their ratepayers have avoided a giant financial debacle.” To some observers, the plan’s collapse also raises questions about the feasibility of other planned advanced reactors, meant to provide clean energy with fewer drawbacks than existing reactors. NuScale’s was the most conventional of the designs, and the closest to construction. “There’s plenty of reasons to think [the other projects] are going to be even more difficult and expensive,” says Edwin Lyman, a physicist and director of nuclear power safety at the Union of Concerned Scientists. The U.S. nuclear industry has brought just two new power reactors online in the past quarter-century. In a deregulated power market, developers have struggled with the enormous capital expense of building a power reactor. Two new reactors at Plant Vogtle in Georgia, one of which came online in May, cost more than $30 billion. To whack down cost, engineers at NuScale decided to think small. Each NuScale
SMR would produce just a fraction of the 1.1 gigawatts generated by one of the new Vogtle reactors. As originally conceived in 2014, the plant would contain 12 SMRs, each producing 60 megawatts of electricity, and would cost $4.2 billion. Small reactors are not an obvious winner. Basic physics dictates that a bigger nuclear reactor will be more fuel efficient than a smaller one. And a big nuclear plant can benefit from economies of scale. However, a small reactor can be simpler. For example, NuScale engineers rely on convection to drive cooling water through the core of each SMR, obviating the need for expensive pumps. SMRs also can be massproduced in a factory and shipped whole to a site, reducing costs.
A mock-up of part of NuScale’s reactor enabled engineers to study what it would be like to work inside.
Size aside, NuScale’s SMR is relatively conventional. Whereas other advanced reactor designs rely on exotic coolants, NuScale’s sticks to water. It also uses the same low-enriched uranium fuel as existing power reactors. Those features helped the NuScale design win approval from the Nuclear Regulatory Commission (NRC) in September 2020—the only advanced reactor to have done so. DOE agreed to host the plant at its Idaho National Laboratory, avoiding the state and local permitting processes commercial reactors ordinarily face. Still, by the time NRC 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
749
11/14/23 5:26 PM
NEWS | I N D E P T H
approved the design, the cost for the project has risen to $6.1 billion. That led DOE to pledge $1.4 billion to the project and developers to scale back to six modules, each pumping out 77 megawatts. In January, an analysis revealed that the cost had increased by another $3 billion and suggested power from the plant would cost $89 per megawatt-hour, roughly three times as much as power from wind or utility-scale solar. Why the costs sky-rocketed remains unclear. Lyman notes that NuScale’s first plant was always going to be expensive, as the company still needed to optimize its production lines. Even so, he says, NuScale designers overestimated how much they could save with a simpler design. “They never demonstrated that you could compensate for that penalty in economies of scale with these other factors.” Jacopo Buongiorno, a nuclear engineer at the Massachusetts Institute of Technology, says the NuScale design has an Achilles’ heel. Each reactor’s core resides within a double-walled steel cylinder, with a vacuum between the walls to keep heat from leaking out. The reactor modules sit in a big pool of water, which in an emergency can flood into the vacuum space around a reactor to prevent it overheating. Compared with a conventional reactor’s building, the pool requires more reinforced concrete, the price of which has soared, Buongiorno says. “In terms of tons of reinforced concrete per megawatt of power, NuScale’s design is off the chart.” UAMPS’s members balked at the cost of that power. UAMPS had the right to break the deal if by early next year members didn’t agree to buy 80% of the plant’s 462 megawatt output, Baker says. The agency had commitments for just 26%. On 7 November the 26 of the 50 UAMPS members that had signed up for the project voted to terminate it, Baker says. Other, more ambitious nuclear projects are in the works. DOE has agreed to help a company called Terrapower develop a reactor that will use molten sodium as a coolant and another company, X-energy, develop an SMR cooled by helium gas. Both plants would use novel fuel enriched to 20% uranium-235. That fuel is not yet commercially available, and it could make those designs even more expensive, Lyman says. Buongiorno says he wouldn’t read NuScale’s failure as a verdict on all advanced reactor designs. “I would steer clear of broadstroke comments in terms of cost,” he says. Baker says he has no doubt that the country needs new nuclear plants to supplement the fluctuating supply of power from wind and solar. “To achieve the nation’s decarbonization goals, it’s got to happen.” j 750
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117NewsInDepth_18261920.indd 750
COMMUNITY
Ousted biologist starts over Fired for sexual misconduct, biologist David Sabatini lands new job in Prague. Reactions are mixed By Meredith Wadman
The same year, he dropped a job offer at New York University after Science made the offer public and student protests erupted. avid Sabatini, the high-flying bioSome of his new colleagues welcomed logist who lost positions at three Sabatini’s hiring. Zuzana Kecˇkéšová, a molecprominent U.S. institutions after breaching sexual misconduct poliular biologist at IOCB, knew Sabatini when cies, began a new job on 1 October as she was a postdoc at the Whitehead from a senior scientist at the prestigious 2008–17. She wrote: “I do not believe that and wealthy Institute of Organic Chemistry striking Dr. Sabatini from the list of people and Biochemistry Prague (IOCB), an arm of who can ever hold a job again helps solve the the Czech Academy of Sciences (CAS). The structural problems of women in science. … hire has divided Czech scientists and igWe welcome him to our midst.” (Kecˇkéšová nited new debate about second chances for is also one of two “ethical proxies” at IOCB.) those who commit sexual misconduct. Several scientists noted that IOCB has “I am very honored to join IOCB,” Sabatini been a leader in promoting workplace equity said by email. He added that he spent much in a country whose proportion of women of the past 2 years “of deep sorrow” in rescientists—27%—is among the lowest in the flection. “In my new lab I will European Union. Given IOCB’s be extra vigilant to make sure public profile, they expect it to that all lab members feel welbe vigilant. come. … I will try my best to Other Czech scientists were not cause offense.” upset. “He’s going to lead people? Jan Konvalinka, director of Oh my goodness,” says Vladimíra the 940-person institute, said Petráková, a biophysicist and Zuzana Kečkéšová, in a statement: “We believe that group leader at CAS’s J. Heyrovsky IOCB [Sabatini] has been punished Institute of Physical Chemistry. enough for his previous actions and that the “An independent [probe] concluded he is a research community will be served best if sexual harasser. That sends a very bad mesthis brilliant scientist returns to research.” sage … that we don’t want to create a safe In 2021, the Howard Hughes Medical Instispace for our employees and students.” tute fired Sabatini, and the Whitehead InstiSabatini “is dismissive and filed a defamatute for Biomedical Research forced him out tion suit,” added Marcela Linková, a socioafter an investigation found he violated the logist who heads the National Contact Centre institute’s sexual harassment and relationfor Gender and Science, part of CAS’s Instiship policies. That investigation found that tute of Sociology. “Any person in a position Sabatini conducted a clandestine sexual reof power over junior colleagues who does not lationship with a woman scientist whom he acknowledge the amounts of power they have was mentoring while she launched a lab at and how that limits a junior person’s maneuthe Whitehead. The investigation also found, vering space—and uses that power against among other behavior, that he created a lab that colleague—is not trustworthy for a suculture that rewarded sexualized banter and pervisory position.” created a “pervasive” fear of retaliation. Sabatini began as a senior group leader Although Sabatini has admitted mistakes, at IOCB on 1 October. A co-discoverer of he has maintained that the relationship was mTOR, a protein that regulates growth and consensual, the probe was unfair, and his aging, Sabatini says he will continue to focus punishment disproportionate. Soon after he on growth regulation in animals. He will relost his Whitehead position, Sabatini sued ceive startup funds from IOCB and intends the institute, its director, and the woman scito apply for grants from the Czech governentist for defamation and workplace discrimment and the European Research Council. A ination. The woman scientist countersued. $25 million, 5-year pledge from New York The litigation is ongoing. City hedge fund manager Bill Ackman and an In 2022, Sabatini resigned a separate, tenanonymous donor is not involved in his fundured professorship at the Massachusetts Ining, he said. Sabatini says he expects his lab stitute of Technology (MIT), which had found will “probably” include 12 to 15 people when he violated its rules on sexual relationships. fully staffed. j
D
“We welcome him to our midst.”
science.org SCIENCE
11/14/23 5:26 PM
Base editors change a gene by nicking one strand of double-stranded DNA; CRISPR cuts both.
BIOMEDICINE
Rewriting DNA in the body lowers cholesterol Verve Therapeutics says its base-editing approach may help prevent heart disease in many people By Jocelyn Kaiser
PHOTO: RAMON ANDRADE 3DCIENCIA/SCIENCE PHOTO LIBRARY
A
technique for precisely rewriting the genetic code directly in the body has slashed “bad” cholesterol levels— possibly for life—in three people prone to dangerously high levels of the artery-clogging fat. The feat relied on a blood infusion of a so-called base editor, designed to disable a liver protein, PCSK9, that regulates cholesterol. “It is a breakthrough to have shown in humans that in vivo base editing works efficiently in the liver,” says Gerald Schwank, a gene-editing researcher at the University of Zurich. The approach, developed by the biotech Verve Therapeutics, is more precise, and possibly safer, than disrupting a gene with CRISPR, the gene-editing tool from which base editing is derived. Reported this week at the American Heart Association meeting in Philadelphia, the base editor results mark the first time this CRISPR variant has been infused into people to treat a disease. The success is also a proof of principle for using gene editing for a common health problem like high cholesterol rather than a rare disease. Some clinicians worry, however, that the treatment’s still undisclosed cost could be exorbitant. And the risks of base editing remain unclear. One of the trial’s 10 participants, nearly all of whom have various gene mutations resulting in high cholesterol levels, had a heart attack possibly related to the treatment. “It worked. But we won’t know for years how safe this is,” says SCIENCE science.org
1117NewsInDepth_18261920.indd 751
cardiologist Karol Watson of the University of California, Los Angeles. CRISPR has recently chalked up several clinical victories. U.S. regulators are poised to approve the gene editor for the blood disorder sickle cell disease. And in small studies, an infusion of CRISPR has been used to shut off a liver protein to treat two genetic diseases. But CRISPR severs both strands of DNA, letting cells imperfectly repair the break. This can result in potentially harmful DNA rearrangements that could flip on a cancer gene. Base editors, a twist on CRISPR invented in 2016 by David Liu’s lab at Harvard University and the Broad Institute, nick just one DNA strand and swap out pairs of the four DNA bases. In base editing’s first clinical test, researchers last year engineered donated immune cells in a dish to target a teenager’s leukemia, then infused them to put her disease into remission so she could get a stem cell transplant. In the Verve trial, however, the editing took place directly in the liver, an easy organ to target because it sucks up foreign particles. The trial subjects have a disease called heterozygous familial hypercholesterolemia (FH), usually caused by a defect in one copy of a gene that encodes a cell surface protein needed by the liver to clear the blood of low-density lipoproteins (LDLs), the “bad” cholesterol. People with FH must take daily statins and other drugs to control their cholesterol levels, but many struggle to keep to the lifelong regimen. Without any treatment, many would suffer heart attacks or strokes by age 50.
FH patients still make some LDL receptors, and Verve’s strategy is to keep those molecules around for longer by eliminating PCSK9, an enzyme that normally removes the receptors from cells. Its treatment consists of messenger RNA (mRNA) that instructs cells to manufacture the gene editor’s protein components. Packaged in tiny balls of fat called lipid nanoparticles, it travels to the liver, where an additional RNA strand guides the base editor to the gene for PCSK9. The combo makes a one–base pair change so that cells can produce only shortened, nonfunctional enzyme. In three patients receiving the highest doses of the base editor, blood levels of functioning PCSK9 protein dropped between 47% and 84% and LDL levels have fallen between 39% and 55% for as long at 6 months. This is roughly comparable to drops in LDL from the relatively new, injected PCSK9-blocking drugs, which some people now take with or instead of statins. However, two patients who already had severely blocked arteries had heart problems after the base editor infusion. One died from cardiac arrest, a case that a safety board found was unrelated to the infusion, Verve says. The other person survived a heart attack, but it came just a day after treatment and could have been related. The man, however, had chest pains prior to the trial that he didn’t mention to investigators. Had he done so, “he would not have been enrolled,” Verve CEO Sek Kathiresan says. Some clinicians are concerned about a different potential risk: Like standard CRISPR, base editing could make changes to other, nontargeted genes. Endocrinologist Anne Goldberg of the Washington University School of Medicine in St. Louis, who treats FH patients, says that although the Verve treatment “could be a game changer,” she wants to see more safety data. “CRISPR in humans makes me a bit nervous,” she says. Verve plans to test its treatment in a total of about 40 FH patients and then compare the approach with a placebo in a larger trial of such patients. Kathiresan has said the treatment will be more affordable than some of the gene therapies with milliondollar price tags. He says the PCSK9 base editor could one day be used to treat people who don’t have FH but have early heart disease. It could even be given widely to older adults to ward off disease. “Down the road, maybe you turn 50, and this is what you get and it prolongs your life,” he says. “That’s the ultimate vision.” The wait, however, could be long. j 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
751
11/14/23 5:26 PM
NEWS | I N D E P T H
ATMOSPHERIC CHEMISTRY
Carbon-free fuels could have a climatic dark side By Katherine Bourzac
A
s climate-friendly fuels, hydrogen (H2) and ammonia (NH3) are enticing. Because they lack carbon, they can be burned to produce nothing but environmentally benign water and nitrogen (N2). But if producers do not take care to prevent leaks or incomplete combustion, researchers are now warning, the fuels could generate pollutants that could harm human health and shrink or reverse the climate benefits. For example, one analysis finds that, under a worst case scenario, using ammonia as a fuel could have a greenhouse gas footprint as bad as burning an equivalent amount of coal. “We can’t just be hoping these things work,” says Amilcare Porporato, an environmental engineer at Princeton University and a co-author of the study, which was published last week in the Proceedings of the National Academy of Sciences. “We need to do due diligence.” These potential side effects are too often overlooked, says Paul Wolfram, a researcher at the Joint Global Change Research Institute. “The focus is almost solely on [carbon dioxide] emissions,” he says. Today, hydrogen and ammonia mainly come from energy-intensive, polluting processes. But they can also be made cleanly, with renewable electricity, resulting in a green fuel. Green hydrogen got a boost last month, when the U.S. Department of Energy
announced $7 billion in funding to support several hubs to make it (Science, 20 October, p. 253). Ammonia has an additional advantage: Unlike hydrogen, it can be liquefied at mild pressures and transported relatively easily (Science, 13 July 2018, p. 120). When ammonia leaks or isn’t burned completely, however, the nitrogen it contains can give rise to reactive nitrogen species. These compounds include nitrous oxide (N2O), a greenhouse gas about 273 times more potent than carbon dioxide, and other nitrogen oxides, collectively called NOx, which are notorious air pollutants. For instance, NO2 is a key ingredient in smog and acid rain and leads to the formation of particles that can cause asthma. Porporato and his colleagues modeled best and worst case scenarios in a society that could be producing some 1600 million tons of ammonia per year by 2060–70. (Currently, the annual ammonia market is about 180 million tons per year, and most is used for fertilizer.) They looked at scenarios in which between 0.5% and 5% of the nitrogen in ammonia was lost as reactive nitrogen compounds instead of being converted back into harmless atmospheric nitrogen. They found that emission rates of nitrous oxide in particular controlled whether ammonia would help control global warming or harm the climate as much as burning coal. Hideaki Kobayashi, a combustion engineer developing ammonia turbines at Tohoku University, says the group’s projections are
CF Industries is building one of the world’s largest green ammonia plants in Donaldsonville, Louisiana.
752
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117NewsInDepth_18261920.indd 752
too pessimistic. His turbines, he says, burn ammonia efficiently and produce little to no nitrous oxide, while catalytic converters in the exhaust systems get rid of any nitrogen oxides. And he says regulations will help limit emissions in Japan and elsewhere. But Wolfram worries about the potential for rogue emissions in the marine shipping industry, which is now rolling out ammonia-powered ships. In a 2022 study of the potential impacts, he and his colleagues found that switching the entire maritime shipping industry to ammonia would take about four times as much of the chemical as the market currently produces. If just 0.4% of the nitrogen in this fuel were converted to nitrous oxide, they found it would completely zero out the benefits of switching from carbon-based fuels. The shipping industry is regulated by the International Marine Organization, but it may be hard to spot leaks from individual ships, Wolfram says. “To me it seems hard to monitor and control all these emissions.” Hydrogen also comes with problems. Leaks from pipelines and other infrastructure could indirectly lead to rising levels of methane, a strong greenhouse gas emitted by natural sources and fossil fuel production. That’s because hydrogen reacts with and depletes hydroxyl radicals, chemical species in the atmosphere that play a key role in breaking down methane. A 2022 analysis of green hydrogen found that if leak rates are as high as 10%, its climate impact would be about half that of an equivalent amount of fossil fuels—still an improvement, but not quite living up to hydrogen’s promise. Keeping the leakage to just 1% or so could preserve the climate benefits, the study found. To know what to expect, researchers need more data about real-world leak rates, says Ilissa Ocko, a climate scientist at the Environmental Defense Fund (EDF) who led the 2022 study. Next year, EDF will launch a monitoring campaign in Europe and North America to provide what Ocko says are the first measurements of leak rates in the field. The ammonia and hydrogen economies are in their infancy but could soon grow up, she adds. “We need to make sure we address these issues before it becomes a problem.” j Katherine Bourzac is a science journalist in San Francisco.
PHOTO: EMILY KASK FOR THE WASHINGTON POST VIA GETTY IMAGES
As nations push for green hydrogen and ammonia, researchers warn of side effects
science.org SCIENCE
11/14/23 5:26 PM
A lawsuit over the Deepwater Horizon oil spill has raised questions about BP’s influence over research.
SCIENTIFIC INTEGRITY
Australian science agency faces scrutiny over industry influence Lawsuit related to 2010 Deepwater Horizon oil spill reveals documents that suggest oil firm BP reviewed CSIRO studies By Royce Kurmelovs
PHOTO: U.S. COAST GUARD VIA AP
A
ustralia’s leading research agency is facing questions about possible ethical lapses after a U.S. law firm released documents suggesting some of its scientists did not disclose that they had allowed oil giant BP to review studies prior to publication in a journal or presentation at a conference. “It’s a mystery why BP’s legal team would be reviewing independent scientific publications” by researchers at the Commonwealth Scientific and Industrial Research Organisation (CSIRO), says attorney Jason Clark of the Downs Law Group, which last week released the documents. On 8 November, Clark sent CSIRO a letter asking the agency to explain color-coded spreadsheets it obtained from BP as part of a lawsuit brought against the company by workers and others claiming they were harmed by the 2010 Deepwater Horizon oil spill in the Gulf of Mexico. The spreadsheets track the status of numerous scientific manuscripts and presentations, many apparently funded at least in part by BP. The forms indicate BP lawyers reviewed nine studies that listed CSIRO scientists as a lead author or as co-authors. CSIRO is Australia’s peak research body, SCIENCE science.org
1117NewsInDepth_18261920.indd 753
employing thousands of researchers across fields ranging from agriculture to robotics. It is known to many Australians for helping invent Wi-Fi. BP’s tracking forms record company lawyers monitoring and making notes on the studies. In each case, the firm alleges BP’s role in “ghost managing” the paper or presentation was “either undisclosed or insufficiently disclosed.” The letter asks CSIRO to make public any communications its researchers had with BP about the studies and drafts of the studies, as well as any contractual obligations they had to BP. In a response sent on 9 November, CSIRO said it was “considering the various matters raised in your letter and will respond as soon as possible.” A CSIRO spokesperson told Science the agency “stands by its research” and that “BP did not have final approval or veto rights in respect of CSIRO’s research presentations and publications relating to the 2010 Deepwater Horizon oil spill.” It also rejected “the assertion that BP was ghost writing” the studies. The manuscripts in question originated with a team led by CSIRO geoscientist Andrew Ross of the agency’s oil and gas research division, which helped track the impact of the 3.2 million barrels of oil that leaked from the Deepwater Horizon rig,
which was operated by BP. The team was involved in developing a geochemical sensor, known as a “sniffer,” that could detect oil in water. The papers cover topics including the distribution of natural hydrocarbon seeps on the gulf floor, a description of the hydrocarbon monitoring tool, and hydrocarbon concentrations in various parts of the gulf. CSIRO promoted this work in a publicity video posted online. The BP spreadsheets include information on the journals targeted for submissions. One note, for example, says the authors of a study on hydrocarbon concentrations planned to submit it to Science, but a second comment suggests getting the paper accepted by a different journal— Environmental Science & Technology—was “more likely in my opinion.” It’s not clear whether the paper was ever submitted to either journal. The law firm says it could not find the paper in either journal, although it did find that researchers later gave a talk with the same title at an annual Australian oil and gas industry conference. The firm also says that talk—and other studies— included no disclosure of BP’s review. Ross and other CSIRO scientists named as authors of the nine papers did not respond to requests for comment. BP said it doesn’t comment on ongoing legal matters. Companies often ask to review research they help fund, although many research institutions have strict rules against allowing funders to approve manuscripts for publication. Clark suspects the reviews were part of a BP effort to reduce its legal liability for the Deepwater Horizon spill. The spreadsheets and other company documents, he wrote, “point to BP manipulating science to promote the false premise in the scientific literature that the [spill] and BP’s response were less harmful to people and the environment than independent science provides.” There is currently no evidence BP did more than review the studies. Clark wants to have the draft studies made public to understand the full extent of any influence. The controversy has stirred concerns about CSIRO’s independence and transparency. “If this behavior occurred … it damages the organization very badly,” says climate scientist John Church of the University of New South Wales, a former CSIRO scientist. “Science is meant to be independent of individual stakeholders. … For some external person or group to come along and impose … their desires for telling the story in a particular way, it should not be allowed.” j Royce Kurmelovs is a journalist in Adelaide, Australia. 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
753
11/14/23 5:26 PM
NEWS
FEATURES
BRAIN GAMES?
I
n 2022, the U.S. National Institutes of Health (NIH) placed a large bet on an experimental drug developed to limit brain damage after strokes. The agency committed up to $30 million to administer a compound called 3K3A-APC in a study of 1400 people shortly after they experience an acute ischemic stroke, a perilous condition in which a clot blocks blood flow to part of the brain. 754
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117NewsFeature_18243948.indd 754
By Charles Piller The gamble seemed warranted. Lab studies, most by a longtime grantee, prominent University of Southern California (USC) neuroscientist Berislav Zlokovic, had generated promising data. A small safety study of the drug, sponsored by a company Zlokovic co-founded called ZZ Biotech, was also encouraging. Analyses of data from the phase 2 trial hinted that the treatment reduced
the number of tiny, asymptomatic brain hemorrhages after stroke patients received either surgery to remove the clot, the clotbusting drug tissue plasminogen activator (tPA), or both. For many years, scientists have tried to reduce the brain cell death, bleeding, and inflammation that can follow a stroke, some of which results from disruption of the blood-brain barrier—a system of tiny blood vessels that delivers oxygen and nutrients
ILLUSTRATION: TAVO MONTAÑEZ
Whistleblowers and former lab members suggest a star neuroscientist routinely manipulated data, compromising a planned NIH stroke trial and key Alzheimer’s research
science.org SCIENCE
11/14/23 9:11 AM
NE WS
but shields the brain from toxic substances. tPA, the only approved stroke drug in the United States and Europe, can vastly reduce death and disability by clearing a stroke’s blockage, but the drug, too, can cause dangerous brain bleeding. 3K3A-APC could help mitigate such damage and prevent brain cells from dying, ZZ Biotech said. Because of its potential to address an unmet medical need, the U.S. Food and Drug Administration (FDA) gave the compound “fast track” status, with the prospect of “accelerated approval and priority review.” ZZ Biotech says the new trial should start within a few months. But a 113-page dossier obtained by Science from a small group of whistleblowers paints a less encouraging picture. The dossier, which they submitted to NIH, highlights evidence from the phase 2 trial that the experimental remedy might have actually increased deaths in the first week after treatment: Six of the 66 stroke patients who received 3K3A-APC died within that period, SCIENCE science.org
1117NewsFeature_18243948.indd 755
compared with one among 44 in the placebo group, although the death rate evened out after a month. Patients who received the drug also trended toward greater disability and dependency at the end of the trial, 90 days after treatment. Deepening the concern, the dossier also highlights evidence that dozens of papers from Zlokovic’s lab—including many supporting the idea that the compound was ready for human testing—contain seemingly doctored data that suggest scientific misconduct. The whistleblowers say apparent changes to images used for protein identification and other purposes seem to skew results in favor of the scientist’s hypotheses, which include influential ideas about the blood-brain barrier and its role in stroke and Alzheimer’s disease, as well as how 3K3A-APC supposedly affects it. Zlokovic’s institution, USC, will confidentially review the content of the dossier, a spokesperson said, adding, “USC takes any allegations relating to research integrity seriously.” Zlokovic declined requests for an interview about the whistleblowers’ findings. But an attorney representing him told Science in a statement that Zlokovic “is committed to fully cooperating” with the USC inquiry. Without providing specifics, the statement noted that some elements of the dossier are “based on information and premises Professor Zlokovic knows to be completely incorrect,” or pertain to experiments not completed in his lab. But speaking to Science anonymously, four former members of Zlokovic’s lab say the anomalies the whistleblowers found are no accident. They describe a culture of intimidation, in which he regularly pushed them and others in the lab to adjust data. Two of them said he sometimes had people change lab notebooks after experiments were completed to ensure they only contained the desired results. “There were clear examples of him instructing people to manipulate data to fit the hypothesis,” one of the lab members says. Given the dossier findings, its authors want all clinical testing of 3K3A-APC halted for now. Multiple neurologists and neuroscientists who reviewed the dossier for Science agree. “To have a fourfold increase in mortality in the first few days of giving the drug really gives me pause,” says Wade Smith, a neurologist at the University of California, San Francisco. Smith found the whistleblower report so disturbing that he couldn’t sleep the night after he read it. Because the drug has to be given soon after a stroke, Smith and others point out, hundreds of patients—or their family members by proxy—might have just hours or even minutes to decide whether to join the
trial. Given what Smith calls possible “scientific fraud” in the preclinical research supporting 3K3A-APC’s supposed protective effects, he thinks the trial should not go forward until NIH, USC, and other organizations can address the whistleblowers’ allegations. “If we’re wrong about the trial, and we really upset some people, well, then I’m sorry,” Smith says. “But the opposite is unfathomable.” All told, the whistleblowers raise concerns about images from 35 basic research studies Zlokovic’s team has published, as well as data from two reports on the phase 2 trial of 3K3A-APC. The publications have a single common author: Zlokovic. In 29 of them—including the main report on the phase 2 trial—he occupied the last author slot, denoting his senior role. No other author is on even half of the dossier’s papers. Some of Zlokovic’s collaborators argue that extensive work outside his lab supports the promise of the potential drug enough to press on with the phase 3 trial. But the lead whistleblower, Vanderbilt University neuroscientist Matthew Schrag, hopes NIH will delay it after seeing the dossier and initiate a sweeping examination of the challenged papers. “Numerous articles appear to warrant retraction and it is likely that this has involved a range of grants as well,” the document’s introduction notes. NIH told Science it takes research integrity concerns very seriously, but otherwise declined to comment. Much of the data described in the dossier “is clearly, undeniably, the result of misrepresentation,” says a leading neuroscientist who studies some of the same topics as Zlokovic. “That saddens me, because he’s a very respected member of this community.” The researcher insisted on anonymity, concerned about becoming embroiled in a controversy that he predicts will threaten Zlokovic’s career. A CELEBRATED AND ECLECTIC neuroscientist
and biotech entrepreneur, Zlokovic has rarely hit a false note—as an M.D.-Ph.D. researcher, institute leader, entrepreneur, and even a talented amateur opera singer. Throughout his steady climb into the academic stratosphere, the charismatic physician—“Betza” to friends and close colleagues—always found time to maintain his vocal gifts. “Science requires clear and perfect language, while music is a universal language,” he told the Cure Alzheimer’s Fund, one of his patrons. He proved that maxim a few years ago at a neuroscience conference reception, belting out a credible version of O Sole Mio to the evident delight of his youthful audience. 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
755
11/14/23 9:11 AM
NEWS | F E AT U R E S
Zlokovic even credits opera as opening a door to his lifelong research interests. During a fellowship at Queen Elizabeth College in London, his department chair asked him to attend a dinner party. “He invited me because he knew I could sing,” Zlokovic told AAAS, publisher of Science, in 2014 when the organization made him one its prestigious Fellows. He apparently charmed the dinner guests, including eminent physiologist Hugh Davson, an expert in the blood-brain barrier. Davson inspired Zlokovic to study the role of the barrier in neurological problems of aging, including Alzheimer’s and an oftenrelated condition, cerebral amyloid angiopathy (CAA), in which protein deposits replace the smooth muscle fibers of blood vessel walls and weaken them. Trained as a doctor at the University of Belgrade in 1978, Zlokovic stayed on there to complete a Ph.D. in physiology in 1983. He later joined the faculty at USC, then spent more than a decade at the University of Rochester before returning to USC in 2012 to direct the Zilkha Neurogenetic Institute, created in 2002 with $20 million from the W. M. Keck Foundation and a matching grant from married philanthropists Selim Zilkha and Mary Hayley. (Zilkha inspired the other “Z” in ZZ Biotech.) Under Zlokovic’s leadership, the USC institute has expanded to more than 30 labs and grown its annual funding more than 10-fold, exceeding $39 million in 2022. NIH grants to Zlokovic have totaled about $93 million. A prodigious fundraiser, in the past decade alone he has added at least $28 million from private sources, according to USC. Hard-driving and prolific, Zlokovic has pioneered work on pericytes, cells that surround the brain’s capillaries and help maintain the blood-brain barrier. He has also linked the blood-brain barrier to Alzheimer’s disease, partly by showing it helps move beta-amyloid proteins, widely viewed as a cause of the disease, out of the brain. That work won him a share of the $100,000 Potamkin Prize in 2009 from the American Academy of Neurology. Last year, the Journal of Molecular Neuroscience published an analysis of key “influencers” who have explored the role of the blood-brain barrier in mild cognitive impairment—early symptoms of dementia. Zlokovic, it concluded, dominated that field of research. Science also used Dimensions Analytics, a database of scholarly research from the U.K. company Digital Science, to examine Zlokovic’s influence in related research categories. The data show that for 756
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117NewsFeature_18243948.indd 756
decades, he has led the world in citations to studies of the influence of the blood-brain barrier and pericytes on stroke or Alzheimer’s (see chart, p. 757). Zlokovic and a collaborator, Scripps Research biochemist John Griffin, also dominate studies of an enzyme called activated protein C (APC). It acts as an anticoagulant in the body, and they argue that the molecule protects against blood clots and inflammation in the brain’s blood vessels, suggesting it could lead to a treatment for stroke. In 2007, Zlokovic helped launch ZZ Biotech to pursue that hope, and according to the company, now owns a roughly 3% stake in it. The firm has worked to turn a
“I immediately felt nauseous. … The integrity of the scientific record is so fundamental to what we do that seeing this kind of data anomaly is distressing.” Chris Schaffer, Cornell University
safer and more effective version of APC, created in Griffin’s lab, into a drug. 3K3A-APC, a form of APC in which three amino acids have been changed, is the result. Company CEO Kent Pryor said last year that for people who experience acute ischemic stroke, “3K3A-APC is a potential game changer.” ON PUBPEER, a website where scientists and
data sleuths publish concerns about possible image doctoring and other forms of scientific misconduct—often anonymously— doubts about Zlokovic’s research began to surface in 2017. Schrag, who previously uncovered possible image manipulation in other lines of Alzheimer’s research (Science, 22 July 2022, p. 358), agreed to help Science scrutinize some of the claims and other work by Zlokovic. He concluded that work from the lab warranted a close look. Over a few weeks Schrag examined Zlokovic’s publications, many in leading journals, and the 3K3AAPC clinical trial reports. He also recruited
Kevin Patrick, a forensic image analyst who is not a scientist and uses the pseudonym “Cheshire” on social media. Patrick, who agreed to have his identity revealed publicly for the first time in this article, made additional findings. Mu Yang, a neurobiologist at Columbia University, also contributed. Schrag and Yang worked independently from their respective universities. The dossier they compiled also folds in comments about Zlokovic’s research posted to PubPeer by Patrick and microbiologist and forensic image analyst Elisabeth Bik, among others. Science did not pay the dossier authors or anyone else for scrutinizing Zlokovic’s work. According to Schrag, he, Patrick, and Bik might file a federal whistleblower lawsuit to receive a portion of any NIH funds the government claws back from USC if federal authorities deem Zlokovic’s work fraudulent. Molecular biologist Mike Rossner—president of Image Data Integrity, a former Journal of Cell Biology editor, and a consultant on image manipulation—also evaluated the Zlokovic dossier. And Bik reviewed images in it that she had not personally posted to PubPeer. Both agree the dossier shows strong evidence of errors or misconduct in many of Zlokovic’s papers. Some images, including in studies about 3K3A-APC and APC, appear doctored in ways that could affect the interpretation of the data, the two say. For example, a 2013 study in The Journal of Neuroscience suggests that 3K3A-APC confers a range of protections for brain cells. According to the whistleblowers, key Western blots—which use antibodies to visualize specific proteins within a tissue sample—seem to have been improperly copied and flipped horizontally. And a 2022 mouse study in Frontiers in Neuroscience showing that 3K3A-APC protects brain cells and the blood-brain barrier from stroke damage includes a crucial image that appears to have been duplicated, in altered form, from a 2019 Nature Neuroscience paper on a different topic. The dossier authors and others who reviewed it for Science note that some apparently duplicated images could be simple mistakes. Other anomalies might be innocent digital artifacts. For example, Western blots sometimes gain unnatural-looking qualities during the publication process. Doubts about an image in a paper can often be resolved only by comparing the original, uncropped, high-resolution version against the published example, and Zlokovic did not respond to Science’s request for original images. But everyone who has seen the findscience.org SCIENCE
11/14/23 9:11 AM
Multiple pericytes on a capillary
A giant in brain science Database searches covering 1990 to the present show neuroscientist Berislav Zlokovic’s outsize influence in multiple areas of research related to the blood-brain barrier, created by brain capillaries and cells called pericytes (above). The topics include mechanisms and possible treatments for stroke and Alzheimer’s disease. But many key papers are now under scrutiny. SUBFIELD Search terms in title or abstract
TOTAL ZLOKOVIC PAPERS
WORLD RANK Paper tally
Pericyte + Alzheimer’s
25*
1
14,341*
1**
Blood-brain barrier + Alzheimer’s
91
1
28,623*
1**
7
4
1675*
Activated protein C + Alzheimer’s***
55
3
17,681*
1
Activated protein C + stroke
45
2
3499
2
Blood-brain barrier + stroke
23
27
3001
7
8
14
1431
13
44
2
4283
Cerebral amyloid angiopathy + blood-brain barrier
Pericyte + stroke 3K3A-APC***
CITATIONS TO ZLOKOVIC PAPERS
WORLD RANK
1
1**
*More than double the next highest **Zlokovic trainees or close collaborators occupy at least six of the top 10 rankings. ***Search in full article
we do that seeing this kind of data anomaly is distressing.” Schaffer, an expert in optical imaging for neuroscience, took the dossier’s analysis of the two papers even further. When he adjusted the image contrast, details missed by the whistleblowers swam into view: In the 2004 paper, superimposed square boxes cover the nuclei of some brain cells. The boxes may mask signs of nuclear fragmentation indicating that the supposedly protected cells were dying, he suggests. “It’s hard to imagine those boxes emerging as a digital artifact,” Schaffer says. “They’re perfect squares.” The Cornell scientist says whoever apparently manipulated the image might have wanted to show “cleaner,” more consistent data or, in a less charitable interpretation, tried to obscure signs that APC and 3K3AAPC didn’t actually protect brain cells. Another neuroscientist who reviewed the dossier but declined to be named for fear of courting a legal dispute was shocked to see apparently manipulated images in two papers he had peer reviewed. “One of them is so clear in retrospect that I should have spotted it,” he says. “But we are not trained as referees to spend much time looking for such things.” The scientist adds: “Overall, I broadly agree with the conclusions [in the dossier]. My main residual question is, why? Why would one bother to go to these lengths to change images, when the guy has the resources to generate loads of great papers without doing this?” THE FOUR FORMER lab members who allege
CREDITS: (PHOTO) DON W. FAWCETT/SCIENCE SOURCE; (DATA) DIGITAL SCIENCE DIMENSIONS
ings says they raise serious and far-reaching questions about his lab practices, research results, and the pending clinical trial. ONE OF SEVEN neuroscientists who reviewed
the dossier is Stanford University’s Thomas Südhof, a Nobel laureate who has seen some of his own papers criticized on PubPeer. (He conceded some errors and rejected other critiques as unfounded.) He cautions against uncritically accepting every apparent image anomaly as evidence of misconduct. “I’m not implying that some of the key papers underlying the clinical trials have fraudulent elements,” Südhof says. Even if some duplicated images are innocent errors, he says, they suggest a worrisome carelessness by Zlokovic and his co-authors. Others are “very hard to explain” as accidental, Südhof adds. He was particularly struck by a 2004 Nature Medicine paper that appears to show a single blood vessel cross section copied and pasted digitally in two other places within an image (see image, p. 758). SCIENCE science.org
1117NewsFeature_18243948.indd 757
Südhof calls it “almost impossible” to explain as unintended. Chris Schaffer, a Cornell University biomedical engineer, says he was most taken aback by a pair of papers published 5 years apart that seem to use the same image to represent different results (see image, p. 759). In a 2004 paper the image purportedly shows how natural APC prevents brain cells from dying. But a 2009 paper includes what appears to be the same image as evidence that the ZZ Biotech compound also protects the brain but without causing hazardous bleeding, a drawback of natural APC. Data from that paper helped set the dose of 3K3A-APC initially tested in people. The dossier suggests that cellular features had been removed from a raw image before its use in the 2004 paper, and that the original image was used in the later paper. The whistleblower analysis was so persuasive that it left Schaffer shaken: “I immediately felt nauseous,” he says. “The integrity of the scientific record is so fundamental to what
Zlokovic pushed them and others to manipulate data paint a picture of a pressure cooker environment in which their boss expected new data almost every week, always in line with his hypotheses. All worked with Zlokovic for years and published with him, and they gave similar descriptions of the research environment. Science also spoke briefly to a fifth former lab member, Angeliki Nikolakopoulou, the first author on three papers in the dossier and now principal scientist at Bionaut Labs, a Los Angeles biotech company. “The only thing I can tell you after being a member of his lab for 8 years is that there is no misconduct,” she said, then hung up. In lab meetings, the accusers say, researchers were discouraged from speaking up and contributing intellectually to the lab’s work, which was tightly controlled by Zlokovic. “It’s science. So normally you would express your opinions,” one notes. Instead, that person says, newcomers soon learned that speaking up meant facing “humiliation”—a term three of the insiders used—and dismissal of their comments. 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
757
11/14/23 9:11 AM
Except to answer questions, one researcher AFTER A LONG HISTORY of failed stroke drugs, Science provided the dossier to USC neurosays, “we were all silent.” a panel of researchers and physicians logist Patrick Lyden, principal investigaAll four say Zlokovic routinely castigated in 1999 established rigorous criteria for tor for the 3K3A-APC phase 2 trial and junior scientists when his desired experimenclinical trials of stroke treatments—known head of the planned phase 3 study. Lyden, tal results were not obtained. “If you are not by the acronym STAIR, for the Stroke who works at the Zilkha institute and was in agreement with him, you will lose the lead Treatment Academic Industry Roundan author of the Stroke paper certifying authorship on a paper or on a project,” one of table that established them. The panel 3K3A-APC as ready for clinical trials, said the scientists says. “Of course, this is imporwanted to help ensure that only clearly in a statement that the alleged problems in tant for your career.” Another says, “If the data promising treatments would be tested in Zlokovic’s papers were outweighed by the does not look like the hypothesis, we were stroke victims. support for the drug from other sources. afraid to even bring it to the lab meeting.” In a 2013 paper in the journal Stroke, But even if 3K3A-APC truly meets the One researcher described how a group of Zlokovic and colleagues assessed 3K3A-APC STAIR criteria, the drug’s phase 2 trial, lab members approached Zilkha’s human research to see whether it met the 10 STAIR known as RHAPSODY, was problematic, resources department about the according to the whistleblowers and “toxic environment.” The complaint others. They say the trial might have was rejected because they insisted unintentionally favored 3K3A-APC. Copy and paste? on remaining anonymous for fear Stroke patients were given the ZZ In a 2004 Nature Medicine paper, Berislav Zlokovic and colleagues of retaliation. Biotech drug or a placebo after stanconcluded that activated protein C (APC), a natural protein and the Several former lab members prodard care—tPA, the surgical removal basis of a potential stroke drug, sharply reduces injuries to blood vided details of experimental data of the clot, or both. But according to vessels and neurons in mice. But whistleblowers have questioned the from Zlokovic’s lab that they say the final report on the RHAPSODY paper’s supporting data. One image purports to show neurons (green) were falsified. These included extrial, among patients who got both and cross sections of blood vessels (red). Yet the three vessels appear to be identical, although rotated, suggesting “fabrication,” according periments referenced in the whistlestandard treatments, the placebo to the whistleblowers. blower dossier. In some cases, they group received tPA, on average, said, data points that would have more than 2 hours later than those 0° invalidated the desired results were given ZZ Biotech’s experimental Original removed. “It was not real science. drug. A table in the published final orientation He already knew what he wanted trial report seems to indicate that 1 to say” before the experiment was some of the placebo patients even completed, one says. “I started hatgot tPA outside of the American ing science. … It made me sick.” Stroke Association–approved win90° Two of the insiders also say dow of no more than 4.5 hours after Zlokovic sometimes had his team acute ischemic stroke. 2 improperly alter existing note“Even minutes [of delay before books. Normally these notebooks— getting tPA] are considered a signifiin which scientists record details of cant difference,” says neuroscientist 3 their work as it proceeds—provide Andreas Charidimou of Boston Unia ground truth for an experiment’s versity, who reviewed the dossier. That methods and results. As a result, disparity “pushed the data to show 180° they’re also often central to misbenefits in the experimental drug.” conduct investigations. In response to that concern, But two of the former lab memLyden gave Science a different verbers say that after an experiment sion of the phase 2 trial data, showwas completed and its results pubing no delays in the placebo patients lished, Zlokovic sometimes admonreceiving tPA. But in the revised ished his scientists to make sure data table, patients who had clots the notebooks were “clean.” That removed surgically followed by a was understood to mean pasting placebo waited, on average, more into them printouts of the pubthan 2 hours longer than the surgilished results and methodology or cal patients later treated with 3K3Aomitting contrary details that chalAPC. Some of the placebo patients lenged the paper’s conclusions. had surgery beyond the study’s preZlokovic explained that those changes were criteria. For example, a potential stroke scribed 6-hour limit. Although there was needed in case of an “audit,” according to drug had to show promise in both sexes of no indication that care was purposely dethe two scientists. two animal species—in that case, rats and layed, the trial’s prespecified criteria indiTwo of the former lab members say they mice. And both animal behavior and tissue cate those patients should not have been have wrestled with whether to speak out for samples had to validate the drug’s efficacy eligible for the study. years, knowing it might damage their own after a stroke. 3K3A-APC easily met all the Charidimou, a veteran stroke investigacareers. “This is a moment in life when I STAIR criteria, they concluded. Among tor, says the longer the delay before the must choose between what is right and other evidence cited, it reduced the volume surgery, the less successfully it prevents what is easy,” one says. “The easy option of brain tissue damaged by the stroke and brain damage. So Lyden’s new data table, would be not speaking with you. I decided stopped some bleeding caused by tPA. like the published version, showed that the that when I lay on my deathbed one day, I Much of the evidence that 2013 paper reexperiment favored patients treated with might regret not doing what is right.” lied on is now in question, however. 3K3A-APC, Charidimou contends. 758
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117NewsFeature_18243948.indd 758
CREDITS: (GRAPHIC) C. BICKEL/SCIENCE; (IMAGES) D. LIU ET AL., NATURE MEDICINE 10, 12 (2004)
NEWS | F E AT U R E S
science.org SCIENCE
11/14/23 12:18 PM
CREDITS: (GRAPHIC) C. BICKEL/SCIENCE; (IMAGES, LEFT TO RIGHT) H. GUO ET AL., EUROPEAN JOURNAL OF NEUROSCIENCE 29, 6 (2009); H. GUO ET AL., NEURON 41, 4 (2004)
Despite either possible advantage, 3K3AAPC didn’t show a statistically significant advantage over the placebo in brainhemorrhage volume. Its superiority in the rate of hemorrhages was barely significant only for the tiniest, asymptomatic bleeds, detected using brain scans. And in addition to the six deaths soon after the compound’s use, some neurologists are concerned that more patients in the treatment group than placebo group experienced potentially damaging brain swelling in the days after the intervention—the opposite of the drug’s intended effect. Lyden challenged those negative interpretations, saying that in part because of the trial’s small size there were no “statistically significant differences in safety outcomes,” and the phase 3 trial would better evaluate any drug side effects. Science also shared the dossier with Pryor, ZZ Biotech’s CEO, and Griffin, 3K3AAPC’s co-developer. Both defended the drug. “While Prof. Zlokovic’s lab has run the majority of the stroke animal models using 3K3A-APC, results from other laboratories have shown similar findings and everything we have seen is consistent with the externally established mechanism of the drug,” Pryor said in his statement. He added that the evidence shows it’s safe enough for further testing. “If anything is uncovered between now and then that would change our feelings, we have time to halt the study commencement.” In Griffin’s statement to Science, he said ample evidence from several labs supports the promise of 3K3A-APC. “There is no basis for implying there is any need to delay continuation of the RHAPSODY trials because of any important deficiency in current fundamental knowledge about 3K3A-APC,” he said, calling the dossier’s claims “unjustified assertions.” Griffin is a co-author on 11 of the questioned Zlokovic papers, published over 20 years. Telling Science he stands by the “fundamental principal conclusions” of those involving APC, Griffin says he can’t vouch for specific images flagged as potentially altered because he lacks access to the originals for comparison. Griffin calls Zlokovic “a brilliant scientist … of unimpeachable integrity.” SEVERAL SUBFIELDS of neurological research
could face a reckoning if Zlokovic’s work comes unraveled for sloppiness or misconduct. All of the papers challenged in the dossier pertain to some degree to Zlokovic’s 4-decade interest in the blood-brain barrier, which has become increasingly important in research on CAA, Alzheimer’s, stroke, and other neurological conditions. SCIENCE science.org
1117NewsFeature_18243948.indd 759
For example, NIH funding for studies concerning both “blood-brain barrier” and “Alzheimer’s” shot up from $13 million in 2006 to $241 million last year. The agency’s funding also rose dramatically for work examining the blood-brain barrier and stroke or pericytes. The basic-science papers described in the whistleblower dossier, excluding a recently posted preprint, have been cited more than 8400 times. On average, those papers have been 27 times as influential as comparable work in the same fields pub-
saying corrections were in process. It remains to be seen just how damaging an invalidation of Zlokovic’s work would be to other research on Alzheimer’s and CAA, and to drug development for stroke. That’s “one of the million-dollar questions,” Charidimou says. “We need to clarify which of these findings are replicable and correct, and which are completely off.” Any impacts on research might take years to play out. But in the coming weeks NIH and FDA will face a pressing issue: whether to postpone or halt human testing
Image déjà vu The whistleblower dossier on Berislav Zlokovic analyzed two images published 5 years apart, concluding one is a manipulated version of the other, even though the papers say they represent different experiments— one on activated protein C (APC) alone and the second comparing it with ZZ Biotech’s potential stroke drug. 2009 image
2004 image
Enhanced 2004 image
These fluorescence-stained brain cells were said to show that the compound 3K3A-APC better limits brain cell death than its inspiration, APC. (European Journal of Neuroscience)
This appears to be the same as the 2009 image but with certain cells removed or changed— possibly to remove undesired evidence about APC. (Neuron)
Imaging software reveals additional signs of manipulation— including how some cells were “erased,” and the nuclei (lighter sections) of others obscured.
Removed cells
lished during the same years, according to Dimensions. They have been cited in 49 patents by 30 companies, universities, and foundations—indicating broad interest in commercializing discoveries that are now in question. To Schaffer, the dossier findings “unquestionably trigger the need for a robust investigation [of all the questioned papers] that goes all the way back to raw data and includes interviews of the scientists who conducted this work.” The goal would be to see which of Zlokovic’s contributions rest on solid data from his own lab or others, Schaffer adds. “Until that investigation has been conducted, the scientific community should use caution in building on these results.” Prior to the dossier’s creation, Patrick and Bik had conveyed concerns to several journals that published some of the Zlokovic papers, requesting an examination. In six cases, corrections were made. After Science provided the dossier to Zlokovic, he or colleagues also responded on PubPeer to comments on several other papers,
Evident blue squares on top of nuclei
Erasure artifacts
of 3K3A-APC. Like NIH, FDA declined to comment on the matter. If the trial goes forward, NIH should force Zlokovic and his close collaborators “to recuse themselves from the conduct of the trial,” says University of Calgary neurologist Eric Smith. “If any unconflicted investigators are left,” he adds, they should “justify to an independent oversight committee that there is sufficient support for the scientific rationale, independent of the Zlokovic work, to continue.” The evidence would need to be compelling, he says. “As a site investigator for stroke trials, I would not agree to participate in the phase 3 trial based on what I know now.” j With reporting by Madeleine Sherer of the Investigative Reporting Workshop at American University and Science Contributing Correspondent Cathleen O’Grady. Aaron Sorensen of Digital Science provided technical assistance. This story was supported by the Science Fund for Investigative Reporting. 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
759
11/14/23 9:11 AM
INSIGHTS
PERSPECTIVES ANTHROPOLOGY
Bonobos provide insight into the origins of partner-specific cooperation in human groups By Joan B. Silk
S
cientific reconstructions of the behavior of the ancestors of modern humans are informed by evidence from fossils, ancient artifacts, genomes of ancient and modern peoples, and the behavior of living nonhuman primates. Chimpanzees (Pan troglodytes) and bonobos (Pan paniscus) play an especially important role in these efforts because they are the closest living relatives of modern humans. However, it is becoming increas760
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117Perspectives_18183742.indd 760
ingly clear that reconstructions based on studies of chimpanzees and bonobos look very different. On page 805 of this issue, Samuni and Surbeck (1) present data about one important element of the divergent story lines—the propensity of bonobos to engage in friendly interactions with members of neighboring groups. In most species of nonhuman primates, interactions between groups range from passive avoidance to active aggression. In chimpanzees, intergroup interactions are uniformly hostile, and there is little over-
lap in the home ranges of neighboring groups (2). Chimpanzee males collectively patrol the borders of their territories and sometimes launch lethal attacks when they come upon isolated individuals from other groups. Bonobos, by contrast, are tolerant of other groups, and the ranges of neighboring groups overlap. Intergroup interactions are generally relaxed, and members of different groups mingle. While they are together, individuals from different groups may engage in friendly and cooperative interactions. To enhance understanding of bonobos’
PHOTO: MARTIN SURBECK/KOKOLOPORI BONOBO RESEARCH PROJECT
Between-group cooperation in bonobos
science.org SCIENCE
11/10/23 4:41 PM
Pairs of bonobos (Pan paniscus) from different groups engage in cooperative interactions such as grooming each other.
behavior in intergroup interactions, Samuni and Surbeck studied the interactions within and between bonobo communities. They focused on two small groups of bonobos in the Kokolopori Bonobo Reserve of the Democratic Republic of Congo that spent about 20% of their time together. During these encounters, pairs of bonobos from different groups groomed each other, formed coalitions, and shared food; they also engaged in aggressive interactions. The relative frequency of these kinds of interactions within and between groups was very similar. Samuni and Surbeck conducted a series of analyses to assess the processes that shape cooperation within and between the groups. Kinship is an important element of School of Human Evolution and Social Change, Institute of Human Origins, Arizona State University, Tempe, AZ, USA. Email: [email protected] SCIENCE science.org
1117Perspectives_18183742.indd 761
cooperative behavior in many nonhuman primate species (3). For example, female baboons (Papio species) selectively groom their mothers, daughters, and sisters. However, kinship seems to account for relatively little cooperation in bonobos and chimpanzees. In these species, females usually leave their birth groups when they reach sexual maturity and move into other groups. This means that adult males may live in the same groups as their mothers, fathers, adult sons, and brothers. Adult females may live in the same groups as their sons and grandchildren but will not have other close relatives in their groups. Genetic analyses indicated that there were very few pairs of close kin within each of the two Kokolopori communities and few pairs of close kin living in different groups. This indicates that although kinship may bias patterns of interaction in bonobos, it is not the primary foundation for cooperation within or between bonobo groups. Evolution of cooperation among nonkin can occur through processes of assortment, such as reciprocal altruism and partner choice, which enable cooperators to selectively interact with other cooperators (4). In models of reciprocal altruism, actors selectively help partners that have helped them in the past and terminate relationships with individuals that do not reciprocate. Reciprocal altruism requires individuals to interact repeatedly and keep track of benefits given and received from particular individuals. In models of partner choice, individuals are expected to leave unrewarding relationships and seek out new partners with more-cooperative tendencies. For partner choice to favor cooperation, individuals must have stable cooperative dispositions and some knowledge of the cooperative dispositions of others. It is difficult to distinguish between these two processes in naturalistic studies, but it is possible to examine some of the basic predictions that are derived from models of these processes. For example, both models predict that cooperation will be selectively directed to specific partners, not randomly distributed among potential partners. Using simulation methods, which accounted for variation in opportunities to interact, Samuni and Surbeck show that bonobo cooperation is nonrandomly distributed among potential partners both within and between groups. Samuni and Surbeck also report evidence that individuals vary in their propensities to participate in cooperative interactions. They found that bonobos that were most likely to form coalitions and share food with members of their own groups were also more likely to engage in these kinds of interactions with members of the other group. If cooperation is sustained by the exchange of ben-
efits, highly cooperative individuals would be expected to interact selectively with other highly cooperative individuals in betweengroup interactions. The bonobos showed this pattern, suggesting that they may have some way of recognizing the cooperative tendencies of individuals in other groups. The authors also show that bonobos that groom and share food with more individuals also receive grooming and food donations from more individuals. This holds for grooming and food sharing within groups as well as between groups. Taken together, these analyses suggest that bonobos treat individuals in other groups in much the same way that they treat individuals in their own groups. The same assortative mechanisms that underlie cooperation within groups, which may include reciprocal altruism and partner choice, also seem to shape cooperative relationships between groups. If this conclusion is correct, then it might provide insight about the origins of one unusual feature of modern human societies. Humans form multilevel societies, in which residential groups (or bands) are aggregated into larger ethnolinguistic units. Social identity is based on membership in the ethnolinguistic unit, and this enables people to orchestrate large-scale cooperative endeavors, including warfare (5). If the ancestors of modern humans treated members of other groups the same way that bonobos do, it might be the first step toward the evolution of multilevel societies. This does not mean that reconstructions of ancestral hominin species should be based only on bonobos. There are other ways in which chimpanzees seem more similar to humans than do bonobos (6). For example, hunting animal prey and tool use are more common in chimpanzees than bonobos, and both of these behaviors are likely to have played important roles in the lives of ancestral hominins. There are also differences in their social relationships. Male chimpanzees form strong bonds with one another and support preferred partners in agonistic interactions, whereas bonobo males form stronger ties to females than to males and rarely form coalitions with other males. Understanding the selective forces that created these differences may help to elucidate how and why humans became such an unusual ape. j REF ERENCES AND NOTES
1. L. Samuni, M. Surbeck, Science 382, 805 (2023). 2. T. Furuichi, Int. J. Primatol. 41, 203 (2020). 3. K. E. Langergraber, in The Evolution of Primate Societies (Univ. Chicago Press, 2012). 4. R. McElreath, R. Boyd, Mathematical Models of Social Evolution: A Guide for the Perplexed (Univ. Chicago Press, 2007). 5. C. Handley, S. Mathew, Nat. Commun. 11, 702 (2020). 6. T. Gruber, Z. Clay, Evol. Anthropol. 25, 239 (216). 10.1126/science.adl1813 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
761
11/10/23 4:41 PM
CLIMATOLOGY
Solid waste, a lever for decarbonization Reducing methane emissions from solid waste is already technically possible By Michael E. Webber1 and Yael R. Glazer2
O
n 20 December 2015, a mountain of urban refuse collapsed in Shenzhen, China, killing at least 69 people and destroying dozens of buildings (1). The disaster exposed the horrible yet real idea that society’s wastes could pile up uncontrollably, directly threatening our lives. But there is another looming threat from solid waste beyond its sheer volumes and mass: the destabilizing impacts of the greenhouse gases it emits. On page 797 of this issue, Hoy et al. (2) report that rapid and large reductions of methane emissions from the world’s solid waste sector are needed to meet the global warming limit set by the Paris Agreement. The good news is that this can be achieved with existing technologies and modified behaviors. Staving off the worst effects of climate change is urgent, and as such, large-scale solutions tend to grab headlines, such as decarbonizing the power sector, electrifying transportation, and tightening efficiency standards. Unfortunately, another critical lever—municipal solid waste management— has been overlooked despite its potential for rapid impact on reducing carbon emissions and its ability to give more room in the global 1
Walker Department of Mechanical Engineering, The University of Texas at Austin, Austin, TX, USA. 2Cockrell School of Engineering, The University of Texas at Austin, Austin, TX, USA. Email: [email protected]
762
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117Perspectives_18183742.indd 762
carbon budget to hard-to-abate sectors, such as long-haul aviation and manufacturing. Municipal solid waste—the garbage that ends up in landfills, recycling centers, compost sites, and ecosystems—is particularly relevant to global warming because solid waste is a major source of atmospheric methane [carbon dioxide (CO2), methane, and nitrous oxide are the primary greenhouse gases]. Methane’s molecular structure also traps more heat than does CO2 and is responsible for approximately one-third of global warming (3). Because methane is both short-lived (it has a half life of 10.5 days in the atmosphere; CO2 has a half life of 120 years) and is a very potent greenhouse gas, its reduction is doubly impactful in terms of the global economy’s ability to rapidly restrict climate warming (4). More than 100 countries have committed to the Global Methane Pledge, with a goal of reducing methane emissions by 30% from 2020 to 2030. It is hard to imagine meeting this aggressive target without tackling the problem of solid waste as soon as possible. Consequently, rapid action on solid waste is a critical precondition for getting the global economy to net zero (at which carbon emissions are balanced by carbon removal from the atmosphere) by 2050. Lack of action will worsen the problem because emissions related to solid waste otherwise would likely increase by more than 60% between 2016 and 2050 (5).
The challenge that municipal solid waste poses to sustainability and climate change has been known for a long time. Despite decades of calls for sustainable waste management from analysts and experts, and a call to rapid action by the International Energy Agency 20 years ago (6), too little has happened. Some population-dense rich countries that bear high fees for dumping waste in landfills and expensive energy— for example, Switzerland, Japan, and South Korea—have prioritized reducing the use of landfills by converting waste to energy. This would also offset the use of fuels such as natural gas and coal (7). Although there are many benefits from this approach, the prime motivations have been space constraints and energy costs rather than emissions reduction. Hoy et al. robustly tackle the question of how managing methane emissions from the global solid waste sector contributes to limiting global warming. The authors modeled greenhouse gas emissions from 2020 to 2050 using longitudinal data from the 43 highest municipal solid waste–producing countries (representing ~86% of global municipal solid waste generation in 2016) as well as panel regression modeling of waste generation per gross domestic product (with projected population) to fill in remaining data gaps. Methane emissions from disposal and treatment plants were calculated according to guidelines of the Intergovernmental Panel on Climate Change. Overall, the study found that municipal solid waste under a business-as-usual management pathway will be responsible for 32 to 35 GtCO-we (billion tonnes of CO2 warming-equivalents) of emissions between 2020 and 2050, exceeding the sector’s emissions budget of 11 to 27 GtCO-we to stay within 1.5° or 2.0°C of the warming limit of the Paris Agreement. Thankfully, Hoy et al. provide some cause for optimism by noting that nearly 90% of the solid waste industry’s methane emissions can be avoided with technologies that already exist. This is in contrast to the International Energy Agency’s claim that as of 2021, half the technologies needed to eliminate CO2 emissions had not yet been invented (8). Thus, mitigation by the solid waste sector can skip an expensive and slow innovation stage and instead focus on the complicated, but necessary, behavioral changes (including shifts in diet and consumption patterns) and updated waste-handling process such as separation, collection, and treatment. Existing solid waste–handling pathways include landfilling, recycling, combustion,
PHOTO: BASRI MARZUKI/NURPHOTO VIA GETTY IMAGES
Solid waste sites provide a considerable amount to global methane emissions and are an important target for reductions.
science.org SCIENCE
11/10/23 4:41 PM
INSIGHTS | P E R S P E C T I V E S
composting, and conversion into biogas (6, 7). Hoy et al. assessed four mitigation pathways across low-, middle-, and high-income countries and the cumulative impacts of these pathways on reducing greenhouse gas emissions between 2020 and 2050. The pathways included anaerobicially digesting organic waste and using the biomethane it produces (70% emissions reduction); halving waste generation (63% reduction); composting organic waste (57% reduction); and retrofitting landfills to add biogas-capture systems (27% reduction). The good news is that each pathway makes a nontrivial contribution to achieving net-zero goals for the sector. The bad news is that none of them can reduce emissions 100%, so multiple solutions must be implemented in parallel. High-income countries, which generate a lot of waste from processed and packaged goods, should focus on halving waste generation. Middle- and lower-income countries, which have higher proportions of organic matter in their waste streams, should focus on anaerobic digestion. Ultimately, Hoy et al. conclude that the key levers for minimizing greenhouse gas emissions are reducing the volume of municipal solid waste and managing organic waste sustainably. However, quickly overcoming behavioral and organizational barriers to implement these solutions will require educating and incentivizing communities, industries, and governments from ethical, environmental, and financial perspectives, for example. It turns out that the wealth of solid garbage that societies generate is a valuable part of the global climate change solution. We all should be as abundantly motivated to deal with it. j REFERENCES AND NOTES
1. H. Yang, X. Huang, J. R. Thompson, R. M. Bright, R. Astrup, Science 351, 674 (2016). 2. Z. X. Hoy et al., Science 382, 797 (2023). 3. IPCC, 2023: Summary for policymakers, in Climate change 2023: Synthesis report, H. Lee, J. Romero, Eds. (IPCC, 2023), pp. 1–34; doi:10.59327/IPCC/ AR6-9789291691647.001 4. Nature 596, 461 (2021). 5. S. Kaza, L. Yao, P. Bhada-Tata, F. Van Woerden, What a Waste 2.0: A Global Snapshot of Solid Waste Management to 2050 (World Bank Publications, 2018). 6. International Energy Agency (IEA), “Municipal solid waste and its role in sustainability: A position paper prepared by IEA Bioenergy” (IEA, 2003 https://www.ieabioenergy.com/blog/publications/ position-paper-municipal-solid-waste-and-its-role-insustainability-2/ 7. US Energy Information Administration, “Biomass explained: Waste-to-energy (municipal solid waste),” 31 October 2022; https://www.eia.gov/energyexplained/ biomass/waste-to-energy.php. 8. IEA, “The path to limiting global warming to 1.5 C has narrowed, but clean energy growth is keeping it open,” 26 September 2023; https://www.iea.org/news/ the-path-to-limiting-global-warming-to-1-5-c-hasnarrowed-but-clean-energy-growth-is-keeping-it-open 10.1126/science.adl0557 SCIENCE science.org
1117Perspectives_18183742.indd 763
MATERIALS SCIENCE
A dynamic biointerface controls mussel adhesion The mussel-adherent secreta interface reveals how nonliving material can be compatible with tissue By Guoqing Pan1 and Bin Li2
M
arine mussel adhesion to surfaces has been extensively studied owing to its potential as an adhesive in wet conditions (1-3). However, most studies focus on the chemical binding mechanism of mussel byssus, an adhesive secreta, to surfaces (4), whereas the connections between living tissue and the byssus have rarely been explored. Although strong adherence of mussels on rocky reefs is necessary for survival, these sessile organisms can liberate themselves from anchored substrates to regain mobility when encountering predators or harsh environments (5). How do mussels ensure strong and compact connection to byssus yet quickly release it when needed? On page 829 of this issue, Sivasundarampillai et al. (6) reveal that the dynamic biointerface between mussel tissue and byssus plays an important role in Mytilus mussels. Their finding could be informative about how nonliving materials can be dynamically interfaced with living tissue, as in the case of detachable biosensors and medical implants. The byssus is a nonliving proteinaceous biopolymer secreted by mussels. The proximal end of the byssus stem transitions into multiple flattened and tapered roots that are used for interfacing with living mussel tissue. The detachable nature of the byssus indicated that strong interfacial interactions between living tissue and the byssus stem roots mainly involve noncovalent interactions, a class of reversible intermolecular bindings. However, noncovalent bindings are generally weak, so how can they support the strong connection between the nonliving biopolymer and living tissue in mussels? Previous studies indicated that enhanced noncovalent interfacial bindings can be constructed on surfaces using either complementary chemical groups or 1
Institute for Advanced Materials, School of Materials Science and Engineering, Jiangsu University, Zhenjiang, Jiangsu, China. 2 Medical 3D Printing Center, Orthopedic Institute, Department of Orthopaedic Surgery, The First Affiliated Hospital, School of Biology and Basic Medical Sciences, Suzhou Medical College, Soochow University, Suzhou, Jiangsu, China. Email: [email protected]; [email protected]
distinct geometric structures (7). For example, a chemically designed hydrogenbonded supramolecular system enabled strong substrate interaction (8). Similarly, macrocycle host-guest chemistry with complementary noncovalent bindings was used to enhance interfacial adhesion (9). However, these sophisticated noncovalent surfaces are rarely found in living systems probably owing to their potential toxicity. Instead, the natural biomaterial interface between proteinaceous byssus and living tissue does not involve special chemistries. Thus, its distinct structure may provide clues to unveil the mechanical stability of the biointerface. Sivasundarampillai et al. investigated whether the strong byssus-tissue interface in Mytilus mussels may arise owing to a multiscale hierarchical structure. They found that the byssus stem root features many individual lamellae sheets that can interdigitate between the living tissue that is carpeted with billions of soft motile cilia. The branched lamellar byssus stem root and the countless cilia greatly increase the adhesive contact at interfaces. In addition, the viscoelasticity of cilia can provide a buffering effect to counteract the mechanical mismatch between the soft living tissue and relatively stiff nonliving byssus. The interdigitated structure between byssus stem root and the living tissue facilitates the enhancement of interfacial junctions. Different from the specific adhesive proteins at the distal end of byssus, the proximal byssus-tissue interfaces are composed of several common proteins, such as collagens. This implies that the strong byssustissue connection relies mainly on a multiscale hierarchical structure rather than special chemical interactions. Sivasundarampillai et al. also uncovered its release mechanism, which involves a neurochemical-regulated microscale contact. They found that the oscillating motion of cilia, triggered by neurochemical factors, correlates with byssus detachment. When cilia motility is activated—for example, through addition of serotonin—decreased adhesive contact between cilia and lamellae occurs, resulting in clean byssus release. 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
763
11/10/23 4:41 PM
INSIGHTS | P E R S P E C T I V E S
synthetic dynamic biointerfaces based on surface introduction of dynamic chemistries have been widely explored for manipulating cell behaviors or modulating tissue adhesion (12, 13). However, few recapitulate biological feedback behaviors, such as mussel byssus release and lizard tail autotomy. Thus, the essential step to approach natural biointerfaces is mimicking their distinct structures. Studies on other similar dynamic biointerfaces in nature are also necessary to verify the conclusion that synthetic biointerfaces should mimic natural biointerfaces to improve performance. The injury-triggered claw self-fracture in lobsters and crabs could offer clues. Further, more efforts are needed to translate these conceptualized biointerfaces onto medical implants and devices (14). Can dynamic biointerfaces be upgraded with reversibility for advanced human-machine systems? There is more to come. j REF ERENCES AND NOTES
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14.
H. Lee et al., Science 318, 426 (2007). B. K. Ahn, J. Am. Chem. Soc. 139, 10166 (2017). J. Bai et al., Research 2022, 9823784 (2022). C. Zhang et al., Chem. Sci. 13, 1698 (2022). E. Carrington et al., Annu. Rev. Mar. Sci. 7, 443 (2015). J. Sivasundarampillai et al., Science 382, 829 (2023). H. Fan, J. Gong, Adv. Mater. 33, 2102983 (2021). S. Chen et al., Angew. Chem. Int. Ed. 61, e202203876 (2022). J. Liu et al., Angew. Chem. Int. Ed. 57, 8854 (2018). N. S. Baban et al., Science 375, 770 (2022). P. A. Fleming et al., Biol. Rev. Camb. Philos. Soc. 82, 481 (2007). Y. Ma et al., Acc. Chem. Res. 52, 1611 (2019). W. He et al., Exploration 2, 20210093 (2022). Q. Zhao, X. Du, Smart Mater. Med. 3, 37 (2022).
ACKNOWL EDGMENTS
The authors receive support from the National Natural Science Foundation of China (32222041, 32130059, and 81925027), the National Natural Science Foundation of Jiangsu Province (BK20220059), and the Jiangsu Specially Appointed Professor Program. 10.1126/science.adl2002
Dynamic biointerfaces in natural systems The interfaces that facilitate Mytilus mussel byssus release and lizard tail fracture exhibit similar multiscale hierarchical structures. The multipoint microcontacting mechanism based on noncovalent interactions endows stable yet detachable interfacial adhesions and connections in mussels and lizards. This provides a paradigm for the design of detachable implant biomaterials and could be informative about other autotomy behaviors. Multipoint microcontacting mechanism
Multiscale hierarchical structure Generator
Lamella
Neurochemical-induced adhesion loosening + Dopamine + Serotonin Force
Byssus
Stem
Stem root Distal
Bending-induced propagation Complementary grooves
Proximal Nanoporous microstructures
764
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117Perspectives_18183742.indd 764
Lamella
Cilia
Force
NEUROSCIENCE
Grabbing neuropeptide signals in the brain Bioengineered sensors resolve the dynamics of neuropeptide action By Roman A. Romanov1 and Tibor Harkany1,2
C
ommunicatIon between neurons in the brain involves the release of either fast (for example, glutamate, g-aminobutyric acid) or slow neurotransmitters (for example, catecholamines, histamine) from the presynaptic terminal, alone or together with neuromodulators. Fast neurotransmitters induce ion currents in the postsynaptic neuron. By contrast, slow neurotransmitters and neuromodulators act on metabotropic G protein–coupled receptors (GPCRs) in the postsynaptic membrane to trigger intracellular second messenger cascades. Neuropeptides are a superfamily of neuromodulators—more than 100 have been identified. However, studying neuropeptides is challenging owing to the limitations of available tools for their detection. Furthermore, despite decades of drug development aimed at neuropeptide-GPCRs, neither their localization nor the dynamics of ligand-induced activation is sufficiently understood. On page 786 of this issue, Wang et al. (1) describe GPCR-activation– based sensors (GRABs) that can track neuropeptide action in vivo. Such GRABs have the potential to provide new information on physiological processes (2) and the role of GPCRs in brain diseases. In synaptic neurotransmission, an action potential provokes the release of a chemical messenger from the presynaptic terminal(s) of a neuron to affect a second, subordinate neuron. It was initially suggested that one neuron releases the same neurotransmitter at all of its synapses (3). Accordingly, the type of neurotransmitter used would deter1
Department of Molecular Neurosciences, Center for Brain Research, Medical University of Vienna, Vienna, Austria. 2Department of Neuroscience, Biomedicum 7D, Karolinska Institutet, Solna, Sweden. E-mail: [email protected]
GRAPHIC: N. BURGESS/SCIENCE
This may enable mussels to translate external environmental factors into the release of specific neurochemicals in vivo, demonstrating the dynamic nature of the musselbyssus interface. The microscale contact between cilia and lamellae may represent a general strategy for achieving strong and controlled interfacial interactions. Multiscale hierarchical structures also exist at the lizard tail interfaces between two segments that allow the tail to be cast off when under threat, called caudal autotomy-induced fracture (10). Analogous to the interdigitation of byssus stem root in mussel tissue, the lizard tail interface displays a “plug-and-socket” structure consisting of multiple muscle bundles with dense adhesive nanopores on one side and complementary grooves on the other side (see the figure). This structure can provide discontinuous microscale bindings for stability and noncovalent-dominated wet adhesion that ensures the stable connection of tail to the lizard body. Despite the stability enhancement under tension, the hierarchical multiscale contacts at the autotomized interface show high vulnerability under bending. Wiggle-caused failure of local microscale contacts can trigger entire tail fracture through a crack propagation process. The most interesting point that deserves in-depth inquiry is the controllable interfacial adhesions in these natural examples, which allow autotomy on demand (byssus release can be broadly considered a kind of autotomy of nonliving tissue) (11). In the clinic, temporary implanted materials or devices, especially neural implants, are expected to be removed or exchanged without damaging tissues. Thus, the autotomized interfaces in mussels and lizards may provide a paradigm for implantable biomaterials. Currently,
science.org SCIENCE
11/10/23 4:41 PM
GRAPHIC: N. CARY/SCIENCE
Postsynapse
Presynapse
mine whether that neuron inhibited neuronal cells, including endoSensors for detecting neuropeptides or activated its partner postsynaptic crine cells of the gastrointestinal Neurotransmitters are secreted from small clear vesicles in the neuron or neurons. Fast neurotranstract and pancreas, for long-range presynapse and induce ion flux in the postsynapse. By contrast, mitters are limited in number and signaling. In the pancreas, SST neuropeptides are released from large dense-core vesicles and bind to G protein–coupled receptors (GPCRs). Depending on the coupled produce almost binary (“yes” or is a marker of d cells (14), whose G protein subunit, GPCRs act through different second messengers “no”) codes for neurotransmission. activity modulates the release of [including cyclic adenosine monophosphate (cAMP) and Ca2+] to either This communication needs conboth insulin and glucagon. It is augment or inhibit the action of neurotransmitters. GPCR-activation– tinuous recalibration to optimize its therefore notable that Wang et al. based sensors (GRABs) are engineered by inserting a fluorophore strength. Neuropeptides are ideal detected glucose-induced, pulsainto the third intracellular loop of a GPCR. Upon neuropeptide binding, for this role. This view was recogtile release of SST when expressing the fluorophore substantially increases the amount of light it emits. nized by the co-release hypothesis neuropeptide-GRABs in isolated for neuropeptides (4) and other pancreatic islets of mice. This sugFast neurotransmitters Neuropeptides messengers (5). gests that GRABs could be used to Small clear vesicles Large dense-core vesicles Most, if not all, synapses contain study organism-wide neuropeptide rapid recycling slow recycling both a neurotransmitter and a neusignaling. ropeptide, at least in some brain The introduction of GRABs for regions (6). Neuropeptide precurneuropeptides could provide opsors (prepropeptides) can often portunities to establish why some Neuropeptide be cleaved into alternative short cells are enriched in many neurofragments, each with different (or peptides and to study the biophysiIonotropic GPCR overlapping) action. Therefore, cal properties of their release—for GRAB receptor neuropeptides could produce comexample, the release threshold and binatorial codes to modulate the dependence on repetitive firing of G i G o G s G q action or prime the sensitivity of the presynaptic neuron or neurons Fluorophore Ion neurons to neurotransmitters. This (15). Other unanswered questions flux cAMP cAMP Ca2+ notion is supported by the obserinclude whether one neuropeptide vation that neuropeptides comcan functionally substitute another monly have multiple cognate GPCRs—some third intracellular loop of the GPCR of in(redundancy), and whether individual stimulating (through Gaq and Gas protein terest. These sensors substantially increase synapses of a neuron differ in their neucascades) and some inhibiting (through their fluorescence when the receptor adopts ropeptide load and signaling properties. Gai and Gao protein signaling) the postits ligand-bound conformation. Several genThe future use of GRABs emitting different synaptic neuron. Thus, GPCRs augment erations of these sensors have been engicolors of light could yield information on or dampen the efficacy of neurotransmitneered for small-molecule transmitters, and GPCR heteromerization and cooperativity. ters, respectively. This concept was later even lipids (11). GRABs have been develOrgan-specific expression of GRABs could expanded to include neuropeptide release oped for neuropeptides, too (12). Recently, a produce insights into long-range (bodyfrom the dendrites and soma of neurons (7) GRAB sensor was used to detect the release wide) signaling by neuropeptides, which for self-modulation or “retrograde commuof the neuropeptide oxytocin from not only will be important if these neuromodulanication,” a form of fine-tuning presynaptic the axons but also soma and dendrites of tors are to retake their place among targets neurotransmitter release. hypothalamic neurons (13). for drug development. j In vivo experimental evidence, together Wang et al. generated GRABs for several R EF ERENCES AND NOTES with observations in humans, suggests that neuropeptides (see the figure), including 1. H. Wang et al., Science 382, eabq8173 (2023). the activation of neuropeptide-GPCRs might somatostatin (SST) and corticotropin-re2. A. Alpár et al., EMBO J. 37, e100087 (2018). prevent excess excitation in epilepsy (8), leasing hormone (CRH), but their strategy 3. J. C. Eccles, P. Fatt, K. Koketsu, J. Physiol. 126, 524 (1954). normalize mood (9) and eating disorders, is applicable to almost any neuropeptide4. T. Hökfelt et al., Nature 284, 515 (1980). alleviate pain, and protect neurons against GPCR (12). The authors studied sensor 5. G. Burnstock, Exp. Physiol. 94, 20 (2009). age-dependent degeneration. Unfortunately, function in cultured cells and used virus6. C. Decavel, A. N. Van den Pol, J. Comp. Neurol. 302, 1019 the experimental tools available to precisely mediated delivery to express them in spe(1990). follow neuropeptide synthesis and action in cific brain areas and cell types in mice. 7. M. Ludwig, G. Leng, Nat. Rev. Neurosci. 7, 126 (2006). 8. C. G. Wasterlain et al., Epilepsia 43 (Suppl 5), 20 (2002). space and time are limited, chiefly because They established that GRABs could detect 9. W. Zhong et al., Proc. Natl. Acad. Sci. U.S.A. 119, the inhibition of axonal transport is necesendogenous neuropeptide release. When e2123146119 (2022). sary to identify the neurons that produce the authors used GRABs to investigate 10. F. Sun et al., Cell 174, 481 (2018). these small peptides. Current approaches the role of SST in associative learning and 11. S. Singh et al., bioRxiv 2023.03.03.531053 (2023). 12. T. Qian, H. Wang, X. Xia, Y. Li, Curr. Opin. Neurobiol. 81, rely on either indirect read-outs, including that of CRH in acute stress, they observed 102751 (2023). changes in membrane biophysics, enzyme neuropeptide action during discrete be13. T. Qian et al., Nat. Biotechnol. 41, 944 (2023). activity, or gene expression, or pharmacohavioral sequences. Particularly impres14. R. Luft et al., Med. Biol. 52, 428 (1974). logical tools. Furthermore, the subcellular sive was the activation of CRH-sensing 15. J. M. Lundberg et al., Proc. Natl. Acad. Sci. U.S.A. 78, 5255 (1981). location of neuropeptide-GPCRs remains GRABs within the paraventricular nucleus mostly unresolved because producing antiof the hypothalamus, which suggests that ACKNOWL EDGMENTS bodies against transmembrane proteins is a CRH has effects on local circuits at the The authors were supported by the European Research major challenge. same time as it is being released into the Council (2020-AdG-101021016, T.H.), the Swedish Research Council (2018-02838, T.H.), Hjärnfonden (FO2022-300, T.H.), GRABs are a growing family of biosenhypophyseal portal system to activate the and the Novo Nordisk Foundation (NNF23OC0084476, T.H). sors that can detect ligand binding at high hypothalamic-pituitary-adrenal axis for spatiotemporal resolution (10) and are encorticosterone production. gineered by inserting a fluorophore into the Neuropeptides are also released by non10.1126/science.adl1788 SCIENCE science.org
1117Perspectives_18183742.indd 765
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
765
11/10/23 4:41 PM
INSIGHTS | P E R S P E C T I V E S
HYPOTHESIS
Advancing the fitness of gut commensal bacteria Nutrient starvation of beneficial bacteria helps them colonize the human gut By Eduardo A. Groisman, Weiwei Han, and Emilia Krypotou
I
mproving the colonization, survival, and persistence of beneficial microbes in the human gut holds great therapeutic potential because they play critical roles in health and disease. In particular, gut commensal bacteria harbor many genes that have no homologs in the hosts that they inhabit, and so they enable various functions that are not encoded in host genomes. For mammals, these functions include energy extraction from otherwise indigestible dietary fibers, vitamin production, and resistance to pathogens. Disruption of the human gut microbiota is associated with metabolic disorders, immune deficiencies, altered susceptibility to pharmacological agents, mental health problems, and some types of cancer. To harness the benefits of gut commensals as probiotics, it is critical to identify bacterial determinants and host conditions, such as diet and eating patterns (e.g., fasting), that advance their fitness. Most microbes in the gastrointestinal tract of healthy adults belong to two bacterial phyla: Bacillota (formerly known as Firmicutes) and Bacteroidota (formerly known as Bacteroidetes). Within Bacteroidota, species of the genus Bacteroides, which are Gram-negative anaerobic bacteria, are widespread among humans and provide important health benefits, such as the ability to break down complex polysaccharides in nutrient-rich vegetables. Therefore, revealing the mechanisms responsible for the fitness of beneficial commensal bacteria in the gut will help the engineering of probiotics with desirable properties, such as increased persistence in this environment. The abundance of individual microbial species within the gut is largely ascribed to host genetic factors and environmental perturbations—notably diet. The impact of diet reflects that gut microbial species differ in their ability to take up and break down specific nutrients. For example, the commensal bacterium Bacteroides thetaiotaomicron has an extraordinary capacity to use a wide variety of complex carbohydrates, including dietary plant polysaccharides, host glycans, and milk oligosaccharides (1). This capac-
ity forms the foundation of food webs that provide nutrition and vitamins to intestinal microbial residents and their host. Moreover, this breadth of activity reflects that nearly 18% of B. thetaiotaomicron genes are devoted to carbohydrate utilization (2). Abundant in lean and healthy humans, B. thetaiotaomicron is being evaluated as a probiotic for gastrointestinal disorders, including ulcerative colitis and Crohn’s disease (3). Certain genes advance bacterial fitness in the gut in a host diet–dependent manner because they enable the use of dietary nutrients (4, 5). However, gut microbes face limited nutrients daily. For example, humans do not normally ingest nutrients at night, nocturnal animals such as mice do not eat during the day, and reptiles can survive for weeks without eating. As a result, gut microbes manifest a starvation response even though many can consume the oligosac-
utilization regulator (Cur, also designated BT4338)—is essential for the fitness of B. thetaiotaomicron and other Bacteroides species in the mouse gut (7). Cur alters the transcription of hundreds of genes within 10 minutes of B. thetaiotaomicron undergoing carbon starvation in laboratory media (8). For example, the expression of several genes responsible for the utilization of monosaccharides and polysaccharides are up-regulated in a Cur-dependent manner upon carbon starvation. Paradoxically, the B. thetaiotaomicron gene most highly induced upon carbon starvation in a Cur-dependent manner encodes a paralog of the essential elongation factor G (EF-G) (8), which is the only protein synthesis factor that participates in two distinct steps of translation: It accelerates by 10,000-fold the movement of mRNAs and transfer RNAS (tRNAs) through the ribosome after incorporation of each amino acid into a growing polypeptide chain, and it helps recycle ribosomes once they reach a stop codon. Canonical EF-G proteins hydrolyze one guanosine triphosphate (GTP) molecule at each ribosome translocation step and one GTP molecule during ribosome recycling. Therefore, elongation is the most energydemanding step of protein synthesis, which is the most energy-consuming cellular process. The EF-G paralog—called EF-G2—has a distinct property that is absent from all EF-G proteins described to date: It catalyzes ribosome translocation without hydrolyzing GTP, albeit at a slower rate than EF-G (9). Moreover, this EF-G2–mediated energy-saving protein synthesis is critical for B. thetaiotaomicron fitness in the mouse gut (9). EF-G2 is >10-fold more abundant than canonical EF-G in B. thetaiotaomicron harvested from mice (9), suggesting that this bacterium experiences carbon starvation in the mouse gut. By contrast, EF-G is much more abundant than EF-G2 during fast growth of bacteria in laboratory media (9). EF-G2 deployment during carbon starvation provides distinct benefits to B. thetaiotaomicron because, unlike canonical EF-G, EF-G2 does not hydrolyze GTP when binding to vacant (nontranslating) ribosomes, which is more likely to occur under nutrient-limiting conditions because the frac-
“The combined effects of engineered probiotics and dietary interventions may expand therapeutic options…”
Department of Microbial Pathogenesis, Yale School of Medicine, New Haven, CT, USA. Email: [email protected]
766
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117Perspectives_18183742.indd 766
charides lining the intestinal epithelium and access by-products generated by other gut microbes. Having coevolved with their hosts, microbes may use starvation or the limitation of certain nutrients as a cue that aids their persistence and survival in the nutritionally fluctuating gut environment. Moreover, if nutrient limitation itself promotes the fitness of beneficial gut bacteria, does it also account for at least some of the health benefits that humans seem to receive from caloric restriction or intermittent fasting diets (6)? Depriving Bacteroides of a carbohydrate source (i.e., carbon starvation) for a short time during growth in laboratory media increases the abundance or activity of determinants of gut fitness. These determinants control hundreds of genes, proteins, and metabolites, unlike factors produced in response to a particular diet-derived nutrient that typically regulate far fewer genes, proteins, and metabolites. How do carbon starvation–induced determinants promote bacterial fitness in the gut? And could these contribute to human health? The master transcriptional regulator of carbohydrate utilization—carbohydrate
science.org SCIENCE
11/10/23 4:41 PM
tion of nontranslating ribosomes increases and the pool of charged tRNAs decreases (see the figure). However, EF-G2 cannot fully replace the essential EF-G because ribosome recycling is strictly dependent on GTP hydrolysis, and EF-G2 lacks guanosine triphosphatase (GTPase) activity. EF-G2 is well conserved across the Bacteroides genus, sharing 75 to 100% identity with the B. thetaiotaomicron EF-G2 protein (9), including in species that require EF-G2 for fitness in the mouse gut (7). The Bacteroides EFG2 protein harbors a 26–amino acid stretch that is essential for protein synthesis despite being absent from canonical EF-G proteins
(p)ppGpp. These molecules are required for survival against carbon starvation and for bacterial fitness in the mouse gut (10). Thus, B. thetaiotaomicron likely synthesizes (p)ppGpp in the gut specifically during carbon limitation. (p)ppGpp advances B. thetaiotaomicron fitness in two major ways: by altering the mRNA abundance of ~60% of genes when the bacterium experiences carbon starvation in the mouse gut and by modifying the abundance of dozens of metabolites (10). For example, concentrations of the tricarboxylic acid cycle metabolites succinate and a-ketoglutarate (aKG) increase during carbon starvation in a
Fitness factors produced in beneficial gut bacteria In beneficial commensal bacteria, such as Bacteroides thetaiotaomicron, nutrient abundance results in energy-demanding protein synthesis driven by EF-G. This results in the expression of fitness factors that are nutrient specific. Conversely, shortterm carbon starvation induces fitness factors that support energy-saving protein synthesis. Additionally, increased Rho activity in phase-separated condensates promotes transcription termination events when the RNA polymerase becomes uncoupled from ribosomes and (p)ppGpp alters mRNA and metabolite abundance. Together, this reduces bacterial growth in low-nutrient conditions, which increases survival and may contribute to the health benefits associated with intermittent fasting diets. Nutrient abundance
Carbon starvation
RNA polymerase DNA rut site (exposed)
Ribosomes rut site (covered)
Rho mRNA (p)ppGpp
EF-G GTP
GDP
EF-G2 GTP
Phase separation GDP
Fast protein synthesis
Slow protein synthesis
Nutrient-specific fitness factors
Carbon starvation fitness factors
GRAPHIC: A. FISHER/SCIENCE
EF-G, elongation factor G; GDP, guanosine diphosphate; GTP, guanosine triphosphate; (p)ppGpp, guanosinetetraphosphate and guanosinepentaphosphate; rut, Rho utilization.
or EF-G2 proteins outside the Bacteroidota phylum (9). Transcriptional activation of the EF-G2– encoding fusA2 gene by Cur is required for B. thetaiotaomicron fitness in the mouse gut and is exhibited by other Bacteroides species (8). By promoting fusA2 transcription, Cur enables bacteria to carry out energyefficient protein synthesis when carbon starvation compromises energy availability. During carbon starvation, B. thetaiotaomicron accumulates guanosinetetraphosphate (ppGpp) and guanosinepentaphosphate (pppGpp) (10), together called SCIENCE science.org
1117Perspectives_18183742.indd 767
(p)ppGpp-dependent manner. aKG accumulation is critical for survival against carbon starvation because supplementation with an aKG precursor is sufficient to restore survival in a B. thetaiotaomicron mutant lacking (p)ppGpp (10). The (p)ppGpp-dependent changes in metabolites may result from changes in gene expression from promoters that are sensitive to (p)ppGpp-bound forms of RNA polymerase and from changes in the activity of enzymes allosterically regulated by (p)ppGpp. A (p)ppGpp regulatory property that B. thetaiotaomicron shares with dis-
tant relatives is the ability to decrease the expression of components of the protein synthesis machinery and associated factors (10). Together, lowering total protein synthesis by (p)ppGpp and the use of EF-G2 likely cause a concomitant decrease in bacterial growth rate, which may favor survival of gut bacteria by conferring resistance to host-derived antimicrobial agents that kill actively growing microbes. In Gram-negative bacteria, transcription and translation are coupled so that they occur at the same rate. Therefore, a slowdown in translation in B. thetaiotaomicron facing carbon starvation may result in the uncoupling of the two processes. This uncoupling would increase the likelihood of the transcription termination factor Rho gaining access to Rho utilization (rut) sites in mRNAs and terminating transcription before the end of a gene or operon. rut sites that are usually protected by the ribosome closely trailing RNA polymerase become unmasked when transcription is uncoupled from translation. How does B. thetaiotaomicron cope with the anticipated demand for increased Rho activity when facing carbon starvation? Carbon starvation promotes liquidliquid phase separation (LLPS) of Rho in B. thetaiotaomicron (11), whereby Rho is sequestered into membraneless compartments. This sequestration increases the ability of Rho to terminate transcription at certain sites and changes the mRNA abundance of hundreds of genes (11). LLPS of Rho is controlled by an intrinsically disordered region in the Rho protein that is also present in Bacteroides species other than B. thetaiotaomicron but is absent from Rho in distantly related bacteria, such as Escherichia coli (11). This LLPS enables Rho to terminate transcription at more sites, some of which correspond to genes that control bacterial fitness in the gut (11). The increased Rho activity may also affect other processes, including suppression of pervasive transcription, repression of horizontally acquired genes, maintenance of chromosome integrity, and prevention of RNA-DNA hybrid formation. That short-term carbon starvation promotes the accumulation of fitness determinants in B. thetaiotaomicron may contribute to the reported improvement in human health ascribed to diets that entail restricted calories or intermittent fasting (12). The limitation of certain nutrients may favor colonization, survival, and persistence of some gut species over others. For example, calorie restriction diets are associated with increased Bacteroidota and decreased Bacillota abundance in the human gut (13), and intermittent fasting is associated with 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
767
11/10/23 4:41 PM
INSIGHTS | P E R S P E C T I V E S
REFERENCES AND NOTES
1. N. D. Schwalm III, E. A. Groisman, Trends Microbiol. 25, 1005 (2017). 2. J. Xu et al., Science 299, 2074 (2003). 3. R. Hansen et al., Clin. Transl. Gastroenterol. 12, e00287 (2021). 4. R. W. P. Glowacki et al., Cell Host Microbe 27, 79 (2020). 5. E. D. Sonnenburg et al., Proc. Natl. Acad. Sci. U.S.A. 103, 8834 (2006). 6. M. M. Mihaylova et al., Cell Metab. 35, 1114 (2023). 7. M. Wu et al., Science 350, eaac5992 (2015). 8. G. E. Townsend II et al., mBio 11, 10.1128/mbio.03221-19 (2020). 9. W. Han et al., EMBO J. 42, e112372 (2023). 10. W. B. Schofield, M. Zimmermann-Kogadeeva, M. Zimmermann, N. A. Barry, A. L. Goodman, Cell Host Microbe 24, 120 (2018). 11. E. Krypotou et al., Science 379, 1149 (2023). 12. L. Kern, D. Kviatcovsky, Y. He, E. Elinav, Curr. Opin. Microbiol. 73, 102287 (2023). 13. S. Pisanu et al., Nutrients 12, 2707 (2020). 14. X. Hu et al., NPJ Biofilms Microbiomes 9, 19 (2023). 15. G. E. Townsend II et al., Proc. Natl. Acad. Sci. U.S.A. 116, 233 (2019).
GENE EDITING
A tool for more specific DNA integration The efficiency of targeted DNA insertion by CRISPR transposons is improved By Yukti Dhingra and Dipali G. Sashital
C
RISPR-associated transposases (CASTs) enable programmable DNA integration by combining DNA-binding Cas effector proteins with DNA-inserting enzymes called transposases (1, 2). CASTs can introduce large DNA cargoes for genome engineering in bacterial and human cells (1–5). Type V-K CASTs consist of the Cas effector Cas12k, the transposase TnsB, the adenosine triphosphatase (ATPase) TnsC, and the adaptor protein TniQ. These CASTs are particularly attractive genome engineering tools owing to their minimal size and number of subunits and their ability to integrate cargo DNA unidirectionally (2). However, a limitation of type V-K CASTs is their high propensity for genome-wide off-target integration (2, 3, 6, 7). On page 784 of this issue, George et al. (8) report that off-target integration by type V-K CASTs occurs independently of Cas12k, revealing a mechanism of untargeted transposition. This untargeted transposition was dependent on TnsC concentration, which
allowed the authors to engineer a CAST system with much improved specificity. CAST systems are classified as type I or type V, depending on the composition of their Cas effector, and are natively expressed in bacteria (1, 2). Canonically, transposition of a DNA cargo by CAST systems is directed to a specific genomic location by a Cas effector (1, 2). Similar to other Cas effectors, Cas12k proteins can be programmed by the user with a guide RNA (gRNA) that allows Cas12k to bind complementary DNA sequences, resulting in the formation of an RNA-DNA hybrid that is further stabilized by binding of the ribosomal protein S15 (9, 10). The binding of TniQ to Cas12k nucleates TnsC filamentation along the DNA, exposing sites for transposition by TnsB. TnsC filaments are a defined length, which ensures that integration of DNA cargo by TnsB is directed at sites that are approximately 60 base pairs from the Cas12k target sequence. Integration by type V-K CASTs has also been observed at untargeted locations across the genome (2, 3, 6, 7), although the mechanism of this off-target integration was
CRISPR-associated transposases perform two types of transposition In the RNA-dependent transposition pathway, the Cas12k–guide RNA (gRNA) complex recognizes and unwinds a target sequence adjacent to the protospacer-adjacent motif (PAM), forming an R-loop that is stabilized upon recruitment of S15 and TniQ. TniQ primes the formation of TnsC filaments adjacent to the target site (TS). These filaments provide a platform for the recruitment of TnsB, which inserts the cargo DNA at the TS. During RNA-independent transposition, TnsC forms filaments at AT-rich sites adjacent to a consensus TS, which is recognized by TnsB. This TnsC filamentation drives the insertion of cargo at untargeted sites in an RNA-independent manner. RNA-dependent transposition
RNA-independent transposition AT-rich sequence TS
PAM Target
Cargo gRNA
S15
TnsB
TniQ
Cas12k
GRAPHIC: K. HOLOSKI/SCIENCE
increased B. thetaiotaomicron abundance in the human gut (14). It is possible, however, that these changes in Bacteroidota and Bacillota abundance (13) reflect changes in dietary composition, rather than caloric restriction per se, that may be directly responsible for the improvement in human health. It is presently unknown whether the B. thetaiotaomicron determinants that promote fitness in the mouse gut also do so in the human gut. How can the increasing knowledge of the interplay between diet and bacterial fitness determinants be used to develop probiotics that promote human health? A proof of principle was established when a B. thetaiotaomicron strain engineered to resist silencing of the fitness-promoting transcription factor Roc by simple sugars outcompeted wild-type B. thetaiotaomicron in mice fed a simple sugar diet (15). This experiment demonstrated how engineering B. thetaiotaomicron in which the production of a fitness determinant is impervious to dietary signals confers a fitness advantage in the mouse gut in a diet-dependent manner. Notably, exploiting the properties of Bacteroides that are associated with human health may also entail inactivating some genes to prevent pathogenic outcomes that can result from escaping the gastrointestinal tract or from aiding gut colonization by enteric pathogens. Additional genes specifying desirable traits could also be incorporated. The combined effects of engineered probiotics and dietary interventions may expand therapeutic options, such as changing the abundance of engineered probiotics only in response to specific dietary components. j
TS
ACKNOWLEDGMENTS
Thanks to J. Aronson for comments on the manuscript. The authors are supported by National Institutes of Health grant GM123798. All authors contributed equally. 10.1126/science.adh9165
768
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117Perspectives_18183742.indd 768
TnsC
Cargo
Cargo science.org SCIENCE
11/10/23 4:41 PM
previously unknown. George et al. assessed genome-wide integrations by a heterologous type V-K CAST in the bacterium Escherichia coli and found that untargeted integration requires only TnsB, TnsC, and TniQ and occurs both in the presence and absence of Cas12k or gRNA. These findings led the authors to propose that type V-K CASTs use an RNA-independent pathway to catalyze untargeted insertions alongside the RNAdependent transposition that occurs at the Cas12k-defined target sites. George et al. used cryo–electron microscopy to establish the structure of an RNAindependent transposition complex composed of TnsB, TnsC, and TniQ bound to DNA substrates (see the figure). The overall architecture is similar to that of the Cas12kcontaining RNA-dependent transposition complex (9, 10), with two helical turns of a TnsC filament flanked on either end by TniQ and TnsB. However, TnsC filaments in the RNA-independent complex interacted with the opposite DNA strand than the RNA-dependent complex, resulting in opposing polarity of the two complexes with respect to the DNA. Using a reconstituted CAST system, George et al. found that at low TnsC concentrations, transposition mainly occurred through the RNA-guided pathway, resulting in >99% on-target integration events. The concentration of TnsC directly correlated with the frequency of untargeted insertions, which strongly suggests that TnsC filamentation promotes RNA-independent transposition and that controlled stoichiometry of this component can limit off-target insertions. George et al. also found that RNAindependent insertion does not happen at random sites. TnsC binding strongly correlated with AT-rich regions of the genome, and analysis of the locations of untargeted insertions revealed long AT-rich sequences upstream of the preferred TnsB target sequence (2, 11). The ~25–base pair length of the ATrich motif is consistent with the length of the TnsC filament in the RNA-dependent transposition complex (9, 10, 12). The presence of an AT-rich motif on only one side of the target site suggests directional recruitment of TnsB upon TnsC filamentation, leading to the same orientation of insertion in both RNA-dependent and RNA-independent transposition pathways. Leveraging these observations, George et al. were able to direct RNA-independent DNA insertion either downstream or upstream of AT-rich sites. These findings provide opportunities to engineer type V-K CAST systems to reduce untargeted integration. A particularly inRoy J. Carver Department of Biochemistry, Biophysics, and Molecular Biology, Iowa State University, Ames, IA, USA. Email: [email protected] SCIENCE science.org
1117Perspectives_18183742.indd 769
triguing possibility is that modulating the levels of TnsC in cells could reduce off-target integration. Indeed, George et al. found that in E. coli reduction of TnsC expression by use of a weaker promoter greatly increased ontarget insertion frequency. Improving targeted insertions would overcome a major limitation in the practical application of type V-K CASTs. However, other limitations of these systems remain. DNA cargoes are typically introduced into cells as a segment of a larger plasmid. Although some transposases use cut-and-paste DNA insertion, in which the DNA cargo is cut out of the plasmid and inserted into the target DNA, type V-K CASTs use a copy-and-paste mode of transposition that initially results in the insertion of the entire plasmid that bears the DNA cargo (3, 6, 7, 13). Strategies to overcome this cointegration of the plasmid with the cargo would enhance the utility of type V-K CASTs for programmable large-scale DNA insertions. The discovery of untargeted integration by type V-K CASTs raises several questions. For example, do type I CAST systems also have an RNA-independent pathway that functions alongside the RNA-dependent pathway? Some type I-F CAST systems are also associated with an additional component, TnsD, which is a member of the same family of proteins as TniQ (14, 15). In these systems, TnsD drives an RNA-independent insertion pathway, co-opting the transposition components of the CAST system but acting independently of the Cas effector (14). Future structural studies will be necessary to delineate the mechanism of the two pathways in TnsDencoding CAST systems. RNA-independent transposition might be selectively functional, thereby allowing transposon mobilization when existing CRISPR RNAs do not support such movement. Nevertheless, how the targeted and untargeted pathways are regulated remains unclear. j REF ERENCES AND NOTES
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15.
S. E. Klompe et al. Nature 571, 219 (2019). J. Strecker et al., Science 365, 48 (2019). P. L. H. Vo et al., Nat. Biotechnol. (2023). G. D. Lampe et al., Nat. Biotechnol. (2023). C. J. Tou, B. Orr, B. P. Kleinstiver, Nat. Biotechnol. 41, 968 (2023). P. L. H. Vo et al., Mob. DNA 12, 13 (2021). B. E. Rubin et al., Nat. Microbiol. 7, 34 (2022). J. T. George et al., Science 382, eadj8543 (2023). M. Schmitz et al., Cell 185, 4999 (2022). J. U. Park et al., Nature 613, 775 (2023). J. U. Park et al., Science 373, 768 (2021). M. W. G. Walker et al., Nucleic Acids Res. 51, 4519 (2023). J. U. Park et al., Proc. Natl. Acad. Sci. U.S.A. 119, e2202590119 (2022). S. E. Klompe et al., Mol. Cell 82, 616 (2022). M. Saito et al., Cell 184, 2441 (2021).
ACKNOWL EDGMENTS
D.G.S. is supported by a National Institutes of Health R35 grant (GM140876). 10.1126/science.adl0863
APPLIED PHYSICS
A highly efficient solid-state heat pump The high efficiency of a newly developed electrocaloric device brings theory closer to reality By Jaka Tušek
E
lectrocaloric heat pumping is a caloric solid-state technology that, along with magnetocaloric and mechanocaloric counterparts, represents promising alternatives to the traditional, relatively inefficient, and environmentally harmful vapor-compression systems. Electrocaloric technologies are based on ferroelectric materials exposed to an electric field, which triggers changes in the material’s polarization, so that applying an electric field increases the material’s temperature, whereas its removal induces cooling. A thermodynamic cycle can be built around this effect, offering environmentally friendly and, theoretically, highly efficient cooling and heat-pump systems. However, prototypes of caloric devices have so far failed to achieve 40% of maximum efficiency (Carnot efficiency) (1). On page 801 of this issue, Li et al. (2) demonstrate a regenerative electrocaloric device that could reach 67% of Carnot efficiency, a maximum temperature span of >20 K, and a maximum cooling power of ~4 W. This high efficiency is an important improvement because it finally confirms numerous theoretical studies that predict high efficiency of caloric technologies. Cooling and air conditioning are essential to modern society but have a high environmental cost. Nearly all cooling and heat-pumping devices in use today rely on vapor-compression technology, which works by circulating a refrigerant through a closed system, repeatedly compressing and expanding it, which through the liquid-gas phase change of the refrigerant allows heat to be pumped from low to high temperatures. Presently, these systems account for ~20% Faculty of Mechanical Engineering, University of Ljubljana, Askerceva, Ljubljana, Slovenia. Email: [email protected] 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
769
11/10/23 4:41 PM
INSIGHTS | P E R S P E C T I V E S
770
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117Perspectives_18183742.indd 770
An efficient regenerative electrocaloric device The highly efficient electrocaloric device prototype involves electrocaloric materials arranged as thin films in multilayer capacitors. These are stacked into efficient heat transfer geometries. Including excellent work recovery, the device could reach 67% of Carnot efficiency, a temperature (T) span of >20 K, and a cooling power of ~4 W. Material Thot Electric field on
Tcold Electric field off
Geometry DT
Oscillating heat transfer medium
Electrocaloric material Electrodes
Efficient electrocaloric device with work recovery
20 K, a threshold that current electrocaloric materials cannot achieve. This limitation has prompted the adoption of multistage thermodynamic cycles to increase the temperature span above that of the electrocaloric material. These cycles can cascade (9) or be regenerative (10, 11), and to date, the regenerative approach, originally developed in magnetocaloric technology (12), has demonstrated superior cooling performance. An electrocaloric regenerator is a porous structure made of electrocaloric material through which a heat transfer medium is oscillating, which determines the temperature span between the heat sink and heat source. However, an efficient and powerful electrocaloric regenerator requires excellent heat transfer geometry with a large specific heattransfer area, thin walls, and small hydraulic diameters to facilitate substantial coolingheating performance.
The third critical component for realizing an efficient electrocaloric device is efficient work recovery. Namely, the polarization of electrocaloric materials necessitates electrical work, whereas during depolarization (when the electric field is removed), electric energy is released and can be stored to facilitate polarization in subsequent cycles. By using an efficient capacitor-charging converter, an astonishing 99.7% of the energy can be recovered (13). The regenerative electrocaloric device presented by Li et al. is an improved iteration of a similar device (11) and successfully integrates all key components of an efficient caloric device. It combines an electrocaloric material with large reversible temperature changes and small hysteresis (PST-MLC) as well as an efficient regenerator geometry with an effective system architecture based on a double heat-transfer loop, which divides the inlet and outlet fluid flow on each side of the electrocaloric regenerator and thus reduces dead volume and mixing losses. Although the performance of this system may fall short of the requirements of many practical applications, which often require cooling capacities of at least several hundred watts at temperature spans exceeding 20 K, the work of Li et al. underscores the immense future potential of electrocaloric technology. Further research and scaling are needed to increase the cooling power of electrocaloric devices. However, a key advantage of electrocaloric technology over other caloric approaches is its potential for miniaturization and compactness. Electrocaloric devices can therefore be used for microcooling applications, such as thermal management of electronic devices. In addition, the prototype of Li et al. shows considerable potential for scaling up to larger cooling and heating powers. j REF ERENCES AND NOTES
1. A. Greco, C. Aprea, A. Maiorino, C. Masselli, Int. J. Refrig. 106, 66 (2019). 2. J. Li et al., Science 382, 801 (2023). 3. D. Coulomb et al., “35th Informatory note on Refrigeration Technologies. The impact of the refrigeration sector on climate change” (International Institute of Refrigeration, 2017). 4. International Energy Agency (IEA), “The Future of Cooling” (IEA, 2018); https://www.iea.org/reports/ the-future-of-cooling. 5. A. S. Mischenko, Q. Zhang, J. F. Scott, R. W. Whatmore, N. D. Mathur, Science 311, 1270 (2006). 6. B. Neese et al., Science 321, 821 (2008). 7. S. Kar-Narayan, N. D. Mathur, Appl. Phys. Lett. 95, 242903 (2009). 8. B. Nair et al., Nature 575, 468 (2019). 9. Y. Meng et al., Nat. Energy 5, 996 (2020). 10. U. Plaznik et al., Int. J. Refrig. 98, 139 (2019). 11. A. Torelló et al., Science 370, 125 (2020). 12. J. A. Barclay, W. A. Steyert, Active magnetic regenerator US patent US4332135A (1982). 13. S. Mönch et al., IEEE J. Emerg. Sel. Top. Power Electron. 11, 4491 (2023). 10.1126/science.adl0804
GRAPHIC: A. MASTIN/SCIENCE
of global electricity consumption and 7.8% of greenhouse gas emissions (3). The demand for cooling and air conditioning are predicted to triple by 2050 (4). Concurrently, heat pumps—refrigerators in heating mode— are an essential part of decarbonizing the heating sector. About 190 million heat pump units were in operation globally in 2021, but they meet only 10% of global heating demand in buildings, falling short of the net-zero emissions targets set for 2050. Although some large-scale vapor-compression systems can achieve Carnot efficiencies that exceed 50%, smaller systems, typically in the kilowatt range (such as single-room airconditioning devices), usually operate below 30% of Carnot efficiency. In addition, these systems still mostly rely on environmentally harmful refrigerants such as hydrofluorocarbons, which have high global warming potential. The Kigali Amendment to the Montreal Protocol calls for synthetic refrigerants with high global warming potential to be phased out over the next decade and replaced with natural refrigerants such as ammonia, carbon dioxide, or hydrocarbons. But these natural refrigerants have limitations, such as flammability, toxicity, or relatively low efficiency. Developing efficient electrocaloric devices requires consideration of the electrocaloric materials, their stacking into efficient heat transfer geometries, and excellent work recovery (the utilization of the energy released when the electric field is removed) (see the figure). Large electrocaloric responses, with temperature changes of up to 12 K, have been observed in thin films of some inorganic and organic materials (5, 6). Thin films are crucial to achieve temperature changes above 10 K owing to their greater electric breakdown strength (the maximum electric field strength that an electric material can withstand without undergoing electrical breakdown) compared with that of bulk materials. Nonetheless, thin films face challenges related to their brittleness and small mass of electrocaloric material, which hinder their practical application in heat-pumping systems. To overcome these challenges, multilayer capacitors (MLCs) have been explored that combine the robust electric breakdown strength of thin films with the mechanical properties and greater active electrocaloric mass of bulk materials (7). Notably, highquality MLCs based on PbSc0.5Ta0.5O3 (PSTMLC) have demonstrated large and highly reversible electrocaloric effects (resulting from small hysteresis losses that represent an unrecoverable part of the input work), achieving temperature changes of ~5.5 K at 29 V/µm over a wide temperature range around room temperature (8). Most cooling and heat-pumping applications require a temperature span of at least
science.org SCIENCE
11/10/23 4:41 PM
RETROSPECTIVE
C. R. Rao (1920–2023) Pioneering statistician and father of information geometry By David Banks1 and Jennifer L. Clarke2
PHOTO: PENN STATE
C
alyampudi Radhakrishna (“C. R.”) Rao, a giant in the field of statistics, died on 22 August at the age of 102. Rao invented information geometry and other fundamental tools in statistics, and he contributed to geology, biometry, demography, econometrics, genetics, anthropology, medicine, and national planning through foundational statistical theorems, scientific collaborations, and decades of advocacy of statistical practices. Generations of statisticians benefited from his mentorship. Rao was born on 10 September 1920, the eighth of his parents’ 10 children, in what was then Madras Presidency in Britishruled India and is now Vijayanagara, Karnataka. He obtained a bachelor’s degree and an MA in mathematics from Andhra University in Visakhapatnam, India, in 1939 and 1940, respectively, as well as an MA in statistics from the University of Calcutta in 1943. He began training as a statistical apprentice at the Indian Statistical Institute (ISI) in Calcutta in 1941 and joined full time 3 years later, working with Prasanta Chandra Mahalanobis, one of the founders of the organization. Rao soon became the de facto editor of Sankhya, a prominent early research journal in statistics that exists to this day. In 1945, Rao published a paper that contained three results, each of which transformed the field of statistics. His work focused on the statistical estimator, a rule for calculating an estimate of a given quantity. (The sample mean, for example, is a routinely used estimator of the population mean.) The Cramér-Rao lower bound derives the minimum possible variance of an unbiased estimator in finite samples— estimators that achieve this bound cannot be improved. The Rao-Blackwell theorem provides a blueprint for automatically improving the efficiency of an estimator, often up to the lower bound. The third result in Rao’s 1945 paper quantifies the distance between statistical distributions. This key tool, which delineates the structure of many probabilities, launched the field of in-
1
Department of Statistical Science, Duke University, Durham, NC, USA. 2Department of Statistics, University of Nebraska–Lincoln, Lincoln, NE, USA. Email: david.banks@ duke.edu; [email protected] SCIENCE science.org
1117Perspectives_18183742.indd 771
formation geometry, which has since made advances possible in artificial intelligence, signal processing, physics, and many other fields. In 1946, Mahalanobis sent Rao to work as a research scholar at King’s College, University of Cambridge, and at the University’s Museum of Archaeology and Anthropology. There, he used discriminant analysis to study biometric differences in the museum’s collection of prehuman skeletal remains. At Cambridge, Ronald A. Fisher supervised Rao’s work and invited him to join his genetics laboratory, where they studied genetic linkage (the tendency of DNA sequences located close together to be
inherited together) in mice. This experience contributed to Rao’s 1948 doctoral thesis on the statistical challenges that arise in classifying biological species. In 1965, Rao received a DSc degree, also from Cambridge. Upon completion of his PhD, Rao returned to ISI. For more than 40 years, he developed programs that shaped the future of Indian statistics, serving as head and later director of the ISI’s Research and Training School. He and Mahalanobis set up statistical bureaus in the states of India and built a network of statistical agencies at the district level for data collection. As a result of their efforts, India has one of the world’s best national statistical systems. Initially aimed at collecting biometric data, Rao expanded the mission of these agencies to include collection of public health and economic data. Rao also founded the Indian Econometric Society, which promotes quan-
titative studies in economics for planning purposes. Before reaching India’s mandatory retirement age of 60, Rao moved to the United States, where his children resided, and accepted a position as a professor of statistics at the University of Pittsburgh in Pennsylvania. In 1988, he transferred to the Pennsylvania State University to lead the Center for Multivariate Statistics, and in 2010, he moved in with his daughter in New York and became a professor in the Department of Biostatistics at the University of Buffalo’s School of Public Health and Health Professions, a position he held until his death. Rao wrote books on topics as diverse as matrix algebra, signal detection, neural networks, and multivariate analysis. Especially influential were Advanced Statistical Methods in Biometric Research (1952) and Linear Statistical Inference and Its Applications (1965). His work won many honors, including the International Prize in Statistics in 2023, which is regarded as the Nobel Prize in the field of statistics, the US National Medal of Science in 2002, and the Padma Vibhushan in 2001, India’s secondhighest public award. Author J.L.C., one of Rao’s last PhD students, remembers how Rao inspired students by serving as an example of accomplishment in the face of adversity. At a holiday dinner J.L.C. hosted in 1998, Rao told her and her family how his seminal paper nearly went unpublished amid the unrest in Calcutta after World War II. Rao had to dodge across town, avoiding civil authorities, including the police, to deliver a hard copy of the manuscript to the Calcutta Mathematical Society. Rao’s support enabled many students and faculty from India to study and visit the United States, fostering a mutually beneficial exchange of statistical knowledge and mentorship. Author D.B. met Rao in 1983 when he, a graduate student, was assigned to be Rao’s driver during his visit to Virginia Tech. An enduring friendship began, and over the years, Rao shared warm stories of his time working with Fisher and Mahalanobis. He had a gentle sense of humor and joked that he had averted a famine in India by correcting weight measurements after realizing that some had been made in British stones and others in British pounds. Rao was a man of vision, kindness, and simplicity, an eminent genius who was willing to talk with graduate students, and a guide to generations of mentees. He was both loved and celebrated by the statistical community, and his contributions have forever changed the field. j 10.1126/science.adl1762 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
77 1
11/10/23 4:41 PM
INSIGHTS
P OLICY FORUM CLIMATE POLICY
Legal limits to the use of CO2 removal Climate targets that depend heavily on CO2 removal may contravene international law By Rupert F. Stuart-Smith1, Lavanya Rajamani2, Joeri Rogelj3,4,5, Thom Wetzer1,2
T
he Intergovernmental Panel on Climate Change (IPCC) has indicated that to hold global warming to 1.5°C, consistent with the goals of the 2015 Paris Agreement, global carbon dioxide (CO2) emissions need to be reduced to net zero by around mid-century (1). This global goal can be achieved by following various technologically feasible emissions pathways (1), but the range of possible strategies create legal and policy uncertainty regarding the emissions reductions required by states. Pathways differ in their rates of gross and net CO2 emission reductions, their corresponding dependence on CO2 removal (CDR) to stay within the cumulative emissions limit imposed by the global temperature goal (2), and the type of CDR they intend to deploy. In the lead up to this year’s United Nations (UN) Climate Conference (COP28) in Dubai, we present scientific and legal bases for our argument that emission-reduction pathways that depend heavily on CDR may contravene norms and principles of international law. CDR IN PARIS-ALIGNED MITIGATION PATHWAYS Nearly all pathways that limit warming well below 2°C require some CDR (1), but a wide range of mitigation strategies exist that entail similar climate outcomes with radically differing CDR reliance. Some involve deep, immediate cuts in gross CO2 emissions (see the figure, scenario 1); others would scale up CDR while deemphasizing gross CO2 emissions cuts (see the figure, scenario 2). In either case, net-zero CO2 emissions is achieved when CDR offsets remaining gross emissions (see the figure). A recent analysis of emission reduction contributions of the world’s wealthiest nations found that they anticipate CDR of ~2.2 bil1
Oxford Sustainable Law Programme, Smith School of Enterprise and the Environment, University of Oxford, Oxford, UK. 2Faculty of Law, University of Oxford, Oxford, UK. 3Grantham Institute for Climate Change and the Environment, Imperial College London, London, UK. 4 Centre for Environmental Policy, Imperial College London, London, UK. 5Energy, Climate, and Environment Program, International Institute for Applied Systems Analysis, Laxenburg, Austria. Email: [email protected]
772
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117_PolicyForum_18187382.indd 772
lion tonnes of CO2 per year (18% of their present emissions) to reach net-zero CO2 emissions (3). Without sufficiently deep near-term emissions cuts, temperatures are projected to rise beyond acceptable levels long term. Under “peak-and-decline” pathways, 1.5°C will be temporarily overshot as cumulative net-CO2 emissions exceed a 1.5°C–consistent carbon budget (1). Net-negative emissions are then needed to withdraw excess emissions from the atmosphere and return warming to 1.5°C (see the figure, sce-
“...lawyers and policy advocates lack tools to hold states accountable for excessive CDR reliance...” nario 3). Almost all pathways assessed in the IPCC’s Sixth Assessment Report that return warming to 1.5°C by the end of this century involve some overshoot and net-negative CO2 emissions after 2050 (1). CDR therefore serves two main purposes in achieving climate goals: offsetting gross CO2 emissions to reach net zero and recapturing CO2 emitted in excess of a carbon budget. Because current emissions are large relative to the remaining carbon budget, emission reductions in the coming decade substantially determine the magnitude of removals required for both uses in subsequent decades. The extent of CDR dependence is consequently a corollary of the rate of near-term emissions reduction. Cumulatively, countries’ policies are incompatible with the Paris Agreement’s temperature goal and will result in permanently exceeding 2°C of warming unless vast quantities of CO2 are removed from the atmosphere (4). However, individual states’ dependence on CDR to meet climate targets, and the consistency of that dependence with applicable international legal norms and principles, is currently unknown and unaddressed in literature. Consequently, lawyers and policy advocates lack tools to hold states accountable for excessive CDR reliance because of in-
adequate near-term emissions cuts. This has posed challenges in judicial evaluation of states’ mitigation policy. For example, in Urgenda Foundation v. State of the Netherlands, the Hague Court of Appeal and the Supreme Court of the Netherlands relied on older emissions pathways premised on lower CDR use to minimize risk. Although understandable in the context, relying on outdated modeling is legally suboptimal. We identify illustrative norms and principles of international law that offer a framework for assessing the legality of emissions pathways. RISKS OF HIGH CDR DEPENDENCE Ahead of COP28, the promotion of CDR to accommodate continued fossil fuel use by the United Arab Emirates’ COP presidency has dominated public discourse and been characterized as “dangerous” by Christiana Figueres, the former Executive Secretary of the UN Framework Convention on Climate Change (UNFCCC) (5). Excessive CDR reliance carries risks that jeopardize the Paris Agreement’s temperature goal and may cause harmful impacts, including those associated with overshooting 1.5°C. These risks and impacts are bases for legal scrutiny of heavily CDR-dependent pathways. First, technological, legal, social, and economic uncertainty regarding increasing the rate of CDR in coming decades risks nondeployment of CDR. This risk is amplified by the lack of legally binding commitments to scale up CDR to necessary levels. Many of the long-term low-emission development strategies submitted to the UNFCCC acknowledge this uncertainty: 27% rely on bioenergy with carbon capture and storage (BECCS) but note that it is not immediately deployable, and 13% characterize direct air carbon capture and storage (DACCS) as a future option “should its cost be significantly reduced” (6). Second, CO2 removed by means of CDR may not be stored permanently, which is a particularly acute risk for terrestrial carbon sinks enhanced by afforestation or reforestation and soil carbon storage (7). Third, CDR deployment may cause adverse social, economic, and environmental impacts, including competition with agriculture for land science.org SCIENCE
11/10/23 4:45 PM
that may be increased by following higher-overshoot pathways. In addition to these treaty norms, the customary international law norm of harm prevention is engaged in relation to states’ actions THE LEGAL IMPLICATIONS on climate change. This norm reOF CDR DEPENDENCE TO MEET quires a high standard of due diliCLIMATE GOALS gence from states to prevent transEmission-reduction pathways that boundary environmental harm depend heavily on CDR, with all (12). Among the factors influencing Scenario approach Scenario outcomes these incumbent risks, may conflict the standard of due diligence in Scenarios 1 2 Scenarios 1 and 2 ( ) / Scenario 3 ( ) with norms and principles of interrelation to climate harms are the Gross C02 emissions Net CO2 national law, spanning treaty and consequences of failing to exercise Gross C02 removal Temperature custom. The Paris Agreement sets such due diligence. Given the cata direction of travel by identifying astrophic impacts of continuing Same climate outcomes, different risks a long-term temperature goal (Arclimate change, due diligence reScenarios 1 and 2 ticle 2), and imposing binding obliquires states to take urgent, transgations on Parties to submit every formative action that has realistic, 5 years nationally determined conscientifically backed prospects of tributions (NDCs) in line with the stabilizing global temperatures. global temperature goal [Article 4 CDR-dependent pathways that (2)], each reflecting a progression involve substantial risks are not on the previous [Article 4 (3)]. Parin keeping with norms and printies are encouraged to explain how ciples of international law, only 2020 2040 2080 2100 2120 2060 their NDCs are fair, ambitious, and a few of which are detailed here. 2140 contribute to the global temperaThese norms and principles can be Scenario 2's reliance on greater CO2 removal to reach the same outcome as Scenario 1 poses increased risks. ture goal (8). Parties are also urged used to identify limits to states’ reto submit long-term low-emission liance on CDR in climate strategies, development strategies [Article 4 providing a basis for assessing the (19)] “in line with the best availadequacy of near-term greenhouse able science” and for their NDCs gas mitigation ambition. Further to be aligned to such long-term analysis of the quantitative limits 2020 2040 2060 2080 2100 2120 2140 low-emission development stratimplied by these principles could egies (9). These provisions, and provide a basis for litigation chalassociated COP decisions that prolenging states’ net-zero targets as Different climate outcomes, different risks vide interpretative context, create a unimplementable and unreliable Scenarios 1, 2, and 3 normative pull (a legal direction of and current NDCs as inadequate. travel) toward aligning short-term These implied limits under interDelayed action of Scenario 3 allows for more NDCs with long-term strategies national law could complement net C02 emissions than in Scenarios 1 and 2. and encouraging states to lay out relevant legal rules in a given jupathways to the global temperarisdiction to challenge net-zero tarture goal that are rooted in sciengets and associated policy packages Scenario 3 requires more CO2 removal in the long tific evidence. Emission-reduction in domestic courts. run to return temperatures to the target level. pathways that depend heavily on Past litigation demonstrates the CDR, given their corresponding effectiveness of such legal strate2020 2040 2060 2080 2100 2120 2140 risks and uncertainties, go against gies. National courts, famously in Scenario 3 overshoots the the grain of these provisions. the 2019 judgment of the Supreme targeted temperature. Heavily CDR-reliant pathways Court of the Netherlands in are also incompatible with a huUrgenda Foundation v. State of the man rights approach to achieving Netherlands, and international fora, climate goals. Indeed, the impacts including in the 2022 decision of of climate change on human rights the UN Human Rights Committee 2020 2040 2060 2080 2100 2120 2140 are subject to growing litigation in in the Torres Islands case against national and regional courts (10). Australia (Billy et al. v. Australia), Even if the preambular reference to human culture, and health, among others, under found violations of the rights to private life rights in the Paris Agreement (Preambular a range of human rights instruments. As and to culture, among others, due to inadeRecital number 11) does not render it a have other UN bodies, the UN Committee quate mitigation and/or adaptation action “human rights treaty,” it signals that states on Economic, Social and Cultural Rights by states. In Urgenda, the court’s ruling comneed to consider how climate change noted that failure to prevent foreseeable pelled the Netherlands to reduce emissions threatens their ability to meet their obligahuman rights harm caused by climate by 25% by 2020. Increasingly, courts are also tions under multilateral human rights treachange could constitute a breach of states’ recognizing that choosing pathways that ties. Most states have obligations in relaobligations to respect, protect, and fulfill postpone stringent mitigation action and/or tion to rights to life, privacy and home life, all human rights for all people (11)—harms rely on potentially costly CDR to be deployed
CDR risks and climate outcomes
A given net-emissions pathway (and climate outcome) can result from gross emissions of different levels, balanced by corresponding dependence on different levels of CO2 removal (CDR). For a given climate outcome, greater dependence on CDR to balance larger gross emissions entails additional risks (top two panels). Delayed reductions in net emissions create increased reliance on CDR to remove excess emissions produced in peak-and-decline pathways, resulting in increased climate change impacts during (and for many impacts, after) the temperature overshoot period (bottom two panels).
GRAPHIC: M. HERSHER/SCIENCE
Temperature
C02 Emissions
Temperature
C02 Emissions
(1). Last, peak-and-decline pathways that temporarily exceed temperature limits result in elevated climate change impacts during and after the overshoot period.
SCIENCE science.org
1117_PolicyForum_18187382.indd 773
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
773
11/10/23 4:45 PM
INSIGHTS | P O L I C Y F O RU M
later threaten intergenerational rights. This was recognized in the 2021 order of the German Constitutional Court in Neubauer v. Germany, which declared part of the Federal Climate Protection Act unconstitutional and prompted the German government to enhance its mitigation target from a 55% to a 65% reduction in greenhouse gas emissions by 2030. A human-rights approach, focused on immediate harms to individuals, demands urgent mitigation and adaptation action and militates against risk taking. This includes overshoot pathways that will occasion irreversible and irreparable harm to people and planet. As in Neubauer, states’ net-zero targets and NDCs can be challenged as unfair, inter alia, because of the distribution of mitigation action and costs between generations. Such arguments may be used to challenge plans that require substantial net-negative emissions later in the century and leave future generations to retrieve excess emissions. Interdisciplinary research similar in
tries agreed to maximize collective efforts to limit warming to 1.5°C. Emission-reduction targets should “reflect [a Party’s] highest possible ambition, reflecting its common but differentiated responsibilities and respective capabilities” [(14), p. 23]. This implies that emissions should stay within a country’s fair share of the global emissions budget. Although there is no multilaterally agreed framework for assessing fair shares of states, previous literature, cited in some cases by litigants, offers an approach for quantifying states’ fair-share budgets on the basis of a total carbon budget, historical emissions, sustainable development needs, capability to decarbonize, population, and international environmental law principles that pertain to these issues (13). Meeting fair-share contributions to the Paris Agreement goals requires some CDR to offset (i) gross emissions that remain at the point of net zero and (ii) cumulative net emissions that exceed countries’ fair carbon budgets. However, most states’ submissions to the UNFCCC do not quantify planned gross emissions at net zero, preventing estimation of CDR reliance (6). Moreover, states’ (implicit) CDR dependence derived from emissions produced before reaching net zero in excess of their Paris-aligned carbon budget is neither found in states’ international disclosures (6) nor in national policies. Accordingly, states’ exact dependence on CDR is unknown yet may be crucial for meeting climate targets and present substantial risks.
“…commitments..consistent with some 1.5°C–aligned emissionreduction pathways may still be inconsistent with international law…” approach to ours synthesized legal expertise and social-science modeling to identify “national fair share” ranges compatible with international law norms and principles and the Paris Agreement’s temperature goal (13) and is being used as a framework by claimants in pending cases before the European Court of Human Rights (such as Duarte Agostinho v. Portugal and KlimaSeniorinnen v. Switzerland) and in assessments of countries’ climate policies (https://climateactiontracker.org). Such litigation could, if successful, compel more ambitious near-term targets, as it has in the Netherlands as a consequence of Urgenda. Even the act of filing cases can lead to reevaluation of near-term target setting in policy circles. Possible causes of action and prospects of success vary between jurisdictions. Nevertheless, norms and principles of international law can provide interpretational context for domestic legal provisions, and interdisciplinary research of the sort we describe can form an evidence base for such cases. COUNTRIES’ MITIGATION RESPONSIBILITIES In adopting the Paris Agreement, and through subsequent COP decisions, coun774
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117_PolicyForum_18187382.indd 774
MOVING FORWARD The unquantified extent and geographical and technological makeup of states’ CDR dependence limits legal scrutiny of climate targets and should advance calls for enhanced disclosure (7) in countries’ reporting of their emissions mitigation action domestically and to the UNFCCC, including through disaggregating targets for removals and net-emission reductions. Even climate commitments that are conceivably consistent with some 1.5°C–aligned emission-reduction pathways may still be inconsistent with international law norms because all but the most ambitious cuts in gross emissions create high CDR dependence. The estimated extent of states’ dependence on CDR does not conform to international law norms and principles. Doing so would require far steeper nearterm emissions cuts than are planned under most national policies. Our analysis also demonstrates the need for interdisciplinary scientific and legal re-
search that clarifies the appropriateness of and risks associated with specific emission-reduction pathways. Such research would provide bases for legal scrutiny, including by quantifying implied state and corporate CDR dependence. Moreover, legal analysis could identify a range of states’ CDR reliance that is consistent with interpretation of international and domestic laws, given the associated risks. Improved disclosure, coupled with further scientific and legal research, will clarify how states need to accelerate emission reductions. Otherwise, given the recent growth in climate-related legal action (15), states’ CDR dependence may be the next aspect of climate (in)action to be challenged in court. j REF ERENCES AND NOTES
1. M. Pathak et al., in Climate Change 2022: Mitigation of Climate Change. Contribution of Working Group III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, P. R. Shukla et al., Eds. (Cambridge Univ. Press, 2022). 2. J. Rogelj et al., Nature 573, 357 (2019). 3. H. J. Buck, W. Carton, J. F. Lund, N. Markusson, Nat. Clim. Chang. 13, 351 (2023). 4. United Nations Environment Programme, “Emissions Gap Report 2022: The Closing Window - Climate crisis calls for rapid transformation of societies” (United Nations, 2022). 5. D. Carrington, The Guardian 16 May 2023. 6. UNFCCC, “Long-term low-emission development strategies. Synthesis report by the secretariat” (United Nations, 2022). 7. M. J. Mace, C. L. Fyson, M. Schaeffer, W. L. Hare, Glob. Policy 12, 67 (2021). 8. UNFCCC, “Decision 4/CMA.1,” in Report of the Conference of the Parties serving as the meeting of the Parties to the Paris Agreement on the third part of its first session, held in Katowice from 2 to 15 December 2018 (United Nations, 2018). 9. UNFCCC, “Decision 1/CMA.3 Glasgow Climate Pact,” in Report of the Conference of the Parties serving as the meeting of the Parties to the Paris Agreement on its third session, held in Glasgow from 31 October to 13 November 2021 (United Nations, 2021). 10. A. Savaresi, J. Setzer, J. Hum. Rights Environ. 13, 7 (2022). 11. UN Committee on Economic Social and Cultural Rights, “Climate change and the International Covenant on Economic, Social and Cultural Rights: Statement of the Committee on Economic, Social and Cultural Rights” (United Nations, 2018). 12. L. Rajamani, in Due Diligence in the International Legal Order (Oxford Univ. Press, 2020), pp. 163–180. 13. L. Rajamani et al., Clim. Policy 21, 983 (2021). 14. UNFCCC, “Adoption of the Paris Agreement,” in Report of the Conference of the Parties on its twenty-first session, held in Paris from 30 November to 13 December 2015 (United Nations, 2015). 15. J. Setzer, C. Higham, “Global trends in climate change litigation: 2022 snapshot,” policy report (Grantham Research Institute on Climate Change and the Environment and the Centre for Climate Change Economics and Policy, 2022). ACKNOWL EDGMENTS
This work was supported by the European Climate Foundation (R.F.S.-S., L.R., J.R., and T.W.) and the Foundation for International Law for the Environment (R.F.S.-S.). 10.1126/science.adi9332
science.org SCIENCE
11/10/23 4:45 PM
“War against the virus” metaphors may have negatively impacted COVID-19 modeling efforts.
B O OKS et al . MODELING
policies to illustrate how the exaggerated precision of a model can be conflated with truth. He explains how regulatory measures introduced into legislation are expressed in very precise terms, whereas estimates of carbon costs differ by several orders of magnitude, depending on the model applied. In the book’s final section, two chapters delve specifically into COVID-19 modeling, calling for a new legal and philosophical framework for future modeling that centers the issue of responsibility. Here, Paolo Vineis and Luca Savarino highlight the challenge of balancing individual and collective responsibility in modeling. They argue that the “war against the virus” metaphor employed during the pandemic influenced COVID-19 modeling, leading to a narrow focus on immediate causes and an overemphasis on individual responsibility. This resulted in undesirable outcomes at the societal level, a phenomenon they call “de-responsibilization.” The breadth and depth of The Politics of Modelling, rooted in its multidisciplinary approach, also reveal weaknesses, which are particularly notable as the logical flow falters ematical modeling: assumptions, hubris, in applying the five principles. It thus falls framing, consequences, and unknowns. short of offering a comprehensive frameThe Politics of Modelling serves as an apwork for responsible modeling, necessitatpendix to this short manifesto that is strucing further research within the modeling tured into three main sections: “Meeting community to develop and test responsible Models,” “The Rules,” and “The Rules in mechanisms in various contexts. The book Practice.” The chapter authors come from also fails to provide a detailed account of how diverse disciplinary backgrounds, providresponsible modeling relates to the growing ing a kaleidoscopic view of current modelbody of research on “good modeling pracing practices, enriched by varitices,” in which modelers seek to ous anecdotes. establish trustworthy modeling In the book’s first section, workflows (2), and would have Philip Stark describes what is benefited from a more comprehidden behind a model, as well hensive discussion of the philoas the issues that can make it sophical underpinning of the wrong. Contrary to common responsibility paradigm. intuition, he explains, what apNevertheless, this book pears to be “science” is often a stands as a pioneering work in a mathematical amplification of The Politics of Modelling: field that has historically focused Numbers Between judgments built into a model on statistical and algorithmic Science and Policy through assumptions, parammethods for generating numAndrea Saltelli and eters, statistics, probabilities, bers. Along with the manifesto Monica Di Fiore, Eds. and many more “degrees of and recent research that draws Oxford University Press, 2023. 272 pp. freedom.” Using an example lessons from COVID-19 modeltreatment of uncertainty in the ing (3), the book represents a Intergovernmental Panel on Climate Change nascent endeavor within the modeling comreport, he explains how models can produce munity to embrace responsible modeling. j hard-to-falsify explanations grounded in falREF ERENCES AND NOTES lacious reasoning. 1. A. Saltelli et al., Nature 582, 482 (2020). In “The Rules,” various authors provide de2. S. H. Hamilton, C. A. Pollino, D. S. Stratford, B. Fu, tailed discussions for each of the five themes A. J. Jakeman, Environ. Model. Softw. 148, 105278 (2022). of responsible modeling. The section includes 3. E. Nabavi, IEEE Trans. Technol. Soc. 3, 252 (2022). examples of modeling failures and a list of best practices that can be adopted to mitigate them. Here, Andy Stirling uses carbon tax 10.1126/science.adl3473
The responsibility turn
Lessons from the COVID-19 pandemic inspire a guide to recognizing the politics of modeling By Ehsan Nabavi1 and Saman Razavi2,3
M
PHOTO: VCG VIA GETTY IMAGES
athematical modelers have traditionally considered politics irrelevant to their practice, arguing that their work represents objective science, the implications of which are determined by others’ use of it. However, this perspective has evolved as researchers have incorporated insights from science and technology studies to explore modeling beyond its technical aspects. The recent book The Politics of Modelling, edited by Andrea Saltelli and Monica Di Fiore, represents such an endeavor. This book follows a manifesto—titled “Five ways to ensure that models serve society”—written by Saltelli and a group of concerned scientists at the start of the COVID-19 pandemic to raise awareness of the politics inherent in modeling practices, with the goal of establishing new social norms among modelers, decision-makers, and citizens (1). As the title suggests, the manifesto outlines five key factors that must be considered for responsible mathThe reviewers are at the 1Responsible Innovation Lab, Australian National Centre for the Public Awareness of Science, The Australian National University, Canberra, Australia; 2Institute for Water Futures and Mathematical Sciences Institute, The Australian National University, Canberra, Australia; and 3School of Environment and Sustainability, Department of Civil, Geological and Environmental Engineering, and Global Institute for Water Security, University of Saskatchewan, Saskatoon, Canada. Email: [email protected]; [email protected]
SCIENCE science.org
1117Books_18197332.indd 775
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
775
11/10/23 4:50 PM
INSIGHTS | B O O K S
When collectors came for the American West An amateur archaeologist’s exploits highlight the damage wrought to Indigenous sites at the turn of the 20th century By Michelle M. Martin
I
n her new book, Sins of the Shovel, archaeologist Rachel Morgan takes readers into the chaotic, controversial, and sometimes dark world of American archaeology in the late 19th and early 20th centuries. The quasi-scientific work of amateur archaeologist Richard Wetherill, who spent his career excavating sites in the American Southwest, forms the core of her narrative, highlighting the lack of oversight and respect shown to Indigenous archaeological sites in this region. Wetherill and his family excavated numerous ancestral Indigenous locations, including Cliff Palace, Colorado; Chaco Canyon, New Mexico; and Grand Gulch, Arizona, at a time when America’s Indigenous population reached its lowest demographic count of roughly 290,000 persons. Anthropologists and archaeologists during this period firmly (and falsely) believed that Indigenous Americans were a vanishing race, leading to intense competition for artifacts among scholars, amateur archaeologists, and collectors as they worked to acquire as much Indigenous cultural material as possible under the premise of saving it for future generations. Collecting practices included not only the looting of historic sites but also the theft of human remains, funerary goods, and sacred objects from Indigenous communities across the American West. These materials were displayed in museums, used for scientific study, and coveted by private collectors. Morgan recounts the Wetherills’ unregulated excavation practices and how they hoped to profit from the sale of artifacts. The family’s first major archaeological excursion took them to the magnificent cliff dwellings at Mesa Verde in Colorado, which were located near their cattle ranch. From 1888 to 1890, they excavated, cataloged, and photographed their finds. For the financially strapped Wetherill, the money that could be gained from the sale of artifacts trumped scientific study, and he sold some The reviewer is at the Department of History, Northeastern State University, Tahlequah, OK 74464, USA. Email: [email protected]
776
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117Books_18197332.indd 776
of the cultural materials to the Colorado Historical Society. Wetherill’s discoveries at Mesa Verde drew the attention of photographers and adventurers, who sought out his services as a guide in the region, and his newfound celebrity opened doors. In 1892, he met New York physician Frederick E. Hyde, who was also drawn to the Southwest and its cultural treasures. Wetherill, his father and sons, and
Sins of the Shovel: Looting, Murder, and the Evolution of American Archaeology Rachel Morgan University of Chicago Press, 2023. 328 pp.
materials—a tactic that drew the ire of professional archaeologists. As the 20th century dawned, Wetherill and his extended family established trading posts, ran archaeological digs, and collected cultural materials, some of which were displayed at the 1904 World’s Fair. Throughout the book, Morgan capably links the Wetherills to other archaeologists and collectors working in the Southwest during this era, creating a complex portrait of the curiosity and greed that underpinned the work of many archaeologists, both pro-
Richard Wetherill (far right) poses on horseback alongside his brothers.
Hyde formed the Hyde Exploring Expedition (HEE) that year, and in 1895 Wetherill and HEE traveled to the Keet Seel site in Arizona’s Tsegi Canyon, where they unearthed some of the most extensive collections of pottery in the region. Despite his successes, however, Wetherill was continually strapped for cash. In 1896, HEE began extensive excavations in Chaco Canyon, New Mexico, of one of the largest, most well-preserved Puebloan sites in the Southwest under the supervision of archaeologist George Pepper. From 1897 to 1898, Wetherill worked incessantly at Chaco Canyon and was always on the lookout for ways to provide a stable income for his family. He scavenged wooden beams from the region’s ancient dwellings and used them to build a three-story trading post with rooms that served as storehouses for excavated
fessional and amateur. She also succeeds in revealing the unglamorous side of cultural collecting, weaving the financial woes of the Wetherill family into her retelling of their archaeological successes and failures. As Morgan brings her work to a close, she provides the reader with a detailed portrait of the various US laws enacted to curb the unscrupulous collecting of Indigenous cultural materials, sacred items, and human remains. Her tour of the legal side of archaeology is informative and provides a good counterbalance to the drama of the Wetherill saga. This section also includes an excellent discussion of Indigenous archaeologists and the critical perspectives they bring to the field to help ensure that the days of wanton looting and pothunting never return. j 10.1126/science.adk5061
PHOTO: BLM CANYONS OF THE ANCIENTS VISITOR CENTER AND MUSEUM, WETHERILL ARCHIVES, 2000.19.P.559.O
ARCHAEOLOGY
science.org SCIENCE
11/10/23 4:50 PM
LET TERS
The endangered eastern black crested (cao vit) gibbon lives near the border of China and Vietnam.
Edited by Jennifer Sills
PHOTO: ZHAOCHAO FOR GIBBON JOURNAL
Vietnam’s vital role in primate conservation More than 50% of all primate species are at risk of global extinction (1). About 90% of primate species are concentrated within tropical rain forests (2). Vietnam is home to 25 primate species, 10 of which are listed as Critically Endangered on the International Union for Conservation of Nature Red List, and 5 of which are endemic to the country (3). Although sustained conservation initiatives, including community-based conservation, law enforcement, and ecotourism, have had considerable success in Vietnam (3, 4), primates are still at risk. By better conserving its forests, Vietnam can protect crucial primate habitat as well as human well-being. Primate populations in Vietnam have faced threats for decades. Extensive use of herbicides and military shelling during the 1960s and 1970s caused substantial destruction and fragmentation of forests in Vietnam (5). The displacement and migration of people due to warfare led to the establishment of settlements and the expansion of agriculture, further exacerbating the loss of natural habitats (5, 6). Because arboreal primates are not well-equipped to traverse substantial distances on the ground between fragmented forest areas, their populations became isolated (4, 5). Such populations—some consisting of just a single group or population—have experienced a decline in genetic diversity, making them vulnerable to inbreeding, susceptibility to disease, and reduced fertility (3–5). Today, primate species continue to suffer from habitat loss and fragmentation, hunting, SCIENCE science.org
1117Letters_18235392.indd 777
poaching, illegal trade, and the impacts of climate change (3, 4). Primates are a critical component of Vietnam’s biodiversity and economic stability. Their vital role as seed dispersers and pollinators facilitates forest regeneration, maintains plant diversity, and supports the overall health of forests and the ecosystem services they provide, including carbon storage and water purification (7). Because Vietnam’s primates attract nature enthusiasts, conservation efforts can boost ecotourism (8), creating economic opportunities for local communities. Primate research in Vietnam contributes to understanding of primates worldwide, and primates hold cultural significance in many societies and Indigenous communities in Vietnam (8, 9). To protect the country’s primate species, Vietnam should take steps to conserve the forests that serve as their habitats. Conserving forests will reap benefits not only for endangered primate species but for other biodiversity and humans as well. Forests play a crucial role in sequestering carbon dioxide and mitigating climate change (10). Healthy ecosystems also help regulate disease vectors and provide natural resources such as clean water and food (10), which are essential for human wellbeing. Protecting Vietnam’s primates and their habitats is essential for the long-term well-being of both the natural world and human society. Aishwarya Maheshwari*, Estella Snowden, Lam Van Hoang Fauna & Flora–Vietnam, Hanoi, Vietnam. *Corresponding author. Email: [email protected] REF ERENCES AND NOTES
1. A. Estrada et al., Sci. Adv. 3, e1600946 (2017). 2. R. A. Mittermeier, D. L. Cheney, in Primate Societies,
4. 5. 6. 7. 8. 9. 10.
B. Smuts, D. Cheney, R. Seyfarth, R. Wrangham, T. Struhsaker, Eds. (University of Chicago Press, 1987), pp. 477–490. H. H. Covert et al., Anthropol. Now 9, 27 (2017). T. Nadler, T. N. Vu, U. Streicher, Vietnamese J. Primatol. 1, 7 (2007). A. H. Westing, Nat. Resour. J. 23, 365 (1983). J. E. Lambert, P. A. Garber, Am. J. Primatol. 45, 9 (1998). T. K. P. Dang, Sustainability 15, 4601 (2023). T. M. Hoang, Am. Anthropol. 118, 130 (2016). Millennium Ecosystems Assessment, Ecosystems and Human Well-being: Synthesis (Island Press, Washington, DC, 2005). 10.1126/science.adl3062
Learn from tobacco to reduce betel nut use About 600 million people worldwide chew betel nut, making it the fourth most common addictive substance in the world, next to tobacco, alcohol, and caffeine (1, 2). In 2003, the International Agency for Research on Cancer, a subsidiary of the World Health Organization (WHO), classified betel nut as a Group 1 carcinogen, drawing global attention to the issue of betel nut consumption (3). The WHO Framework Convention on Tobacco Control (FCTC), a global convention addressing the control of tobacco and smokeless tobacco (including mixtures of betel nut and tobacco) (4), initiated measures to limit the use of tobacco in 2003 and smokeless tobacco in 2005. According to a July WHO report, the number of people who smoke tobacco is decreasing (5). However, the number of people who consume betel nut has increased over the past 2 decades (6). Heightened efforts are required to curb betel nut consumption. The regulation of tobacco has been more widespread and effective than that 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
777
11/14/23 9:11 AM
INSIGHTS | L E T T E R S
Subscribe to News from Science for unlimited access to authoritative, up-to-the-minute news on research and science policy.
of betel nut. Tobacco control measures are implemented in 183 FCTC nations, covering 5.6 billion people under at least one tobacco control measure (5). In contrast, only 57 countries have control policies for smokeless tobacco (including betel nut) (7). The strict implementation of tobacco control policies in many countries reflects the global consensus that smoking is harmful to health. Comparatively, the culture and tradition of consuming smokeless tobacco (including betel nut) are deeply ingrained in many countries (8), whereas the health risks are not as well known. To control global betel nut consumption effectively, regulations should be updated to align more closely with regulations for tobacco (7, 9, 10). Each country should establish a national betel nut control program, similar to the National Tobacco Control Programme (5), to coordinate efforts. Betel nut sales, like tobacco sales (5), should be prohibited to minors. Tobacco efforts can also serve as the model for betel nut public health campaigns. Educational organizations should better inform students about the health risks associated with betel nut (1), and information campaigns should be designed to increase understanding among the public. Accurate, prominent, and strict warnings on betel nut packaging could help to increase users’ awareness of the risks (11). Furthermore, national governments should increase trade tariffs and raise consumption taxes on betel nut purchases to match or exceed tax rates on tobacco (12). Finally, the scientific community should strengthen the monitoring of betel nut production and use. Shigao Chen1, Shengpei Dai2, Yuanyuan Hou1* 1
Institute of Scientific and Technical Information, Chinese Academy of Tropical Agricultural Sciences, Haikou 571101, China. 2College of Geography and Environmental Science, Hainan Normal University, Haikou 571158, China. *Corresponding author. Email: [email protected] REF ERENCES AND NOTES
bit.ly/NewsFromScience
1. S. Gunjal et al., Subst. Use Misuse 55, 1533 (2020). 2. P. K. Singh, A. Yadav, L. Singh et al., BMJ Open 11, e043987 (2021). 3. International Agency for Research on Cancer Working Group on the Evaluation of Carcinogenic Risks to Humans, IARC Monographs on the Evaluation of Carcinogenic Risks to Humans, Vol. 85: Betel-quid and areca-nut chewing and some areca-nut derived nitrosamines (2004). 4. WHO, WHO Framework Convention on Tobacco Control (2003); https://iris.who.int/bitstream/ handle/10665/42811/9241591013.pdf. 5. WHO, “WHO Report on the Global Tobacco Epidemic, 2023: Protect people from tobacco smoke” (2023); https://iris.who.int/bitstream/han dle/10665/372043/9789240077164-eng.pdf. 6. W. J. Moss, New Engl. J. Med. 387, 1059 (2022). 7. A. Chugh et al., Lancet Glob. Health 11, e953 (2023). 8. J. Kaur, T. Thamarangsi, A. V. Rinkoo, Indian J. Pub. Health 61, 3 (2017). 9. R. Mehrotra et al., Lancet Oncol. 20, e208 (2019). 10. M. C. Mitchell, “The political economy of tobacco in
Indonesia: How ‘Two Fires Fell Upon the Earth,’” Master’s Thesis, George Mason University (2013). 11. M. Arora, R. Madhu, Indian J. Cancer 49, 336 (2012). 12. T. Sein, T. Swe, M. M. Toe, K. K. Zaw, T. O. Sein, Indian J. Cancer 51, S3 (2014). 10.1126/science.adk7903
Polio eradication efforts: Above all, do no harm In the News story “Global polio eradication effort struggles with the end game” (22 September, p. 1271), J. Cohen interviews Tom Frieden, who deplores the slow eradication of wild polio virus and recommends the use of further attenuated oral polio vaccine strains to reduce the numbers of vaccine-caused paralytic polio. Although vaccine campaigns that use the improved vaccine virus would prevent polio caused by wild virus, they would still cause cases of polio. This plan is unacceptable when a safer, more effective vaccine protocol exists. The best global vaccination strategy would prevent vaccine-induced paralysis by routinely immunizing with inactivated polio vaccine during the first months of life, in addition to the later use of live polio vaccine (1). This protocol has already been successfully used in some countries (2) but has not been applied worldwide. Although the use of both polio vaccines will be more costly in the short run, it would be more ethical to prevent both wild and vaccine-caused polio. Moreover, such a strategy likely would eliminate paralytic polio faster than the current strategy and would greatly assist the ultimate withdrawal of the oral polio vaccine after the eradication is complete. Stanley A. Plotkin1* and Konstantin Chumakov2 1
University of Pennsylvania, Philadelphia, PA, USA. The George Washington University, Washington, DC, USA. *Corresponding author. Email: [email protected] 2
REF ERENCES AND NOTES
1. A. M. McBean, J. F. Modlin, Pediatr. Infect. Dis. J. 6, 881 (1987). 2. L. N. Alexander et al., JAMA 292, 1696 (2004). COMPETING INTERESTS
K.C. provides paid consulting services to PATH, a nonprofit organization. 10.1126/science.adl2371
ERRATA Erratum for the Research Article “The molecular epidemiology of multiple zoonotic origins of SARS-CoV-2” by J. E. Pekar et al., Science 382, eadl0585 (2023). Published online 13 October 2023 10.1126/science.adl0585 science.org SCIENCE
1117Letters_18235392.indd 778
11/14/23 9:11 AM
PRIZE ES SAY GRAND PRIZE WINNER
Zuzanna Kozicka Zuzanna Kozicka received undergraduate degrees from the University of Edinburgh and a PhD from the University of Basel and Friedrich Miescher Institute for Biomedical Research. She started her postdoctoral fellowship at the Dana Farber Cancer Institute in November 2023. Her research focuses on modulating protein-protein interactions with small molecules for therapeutic purposes. www.science.org/ doi/10.1126/science.adl4288
MOLECULAR MEDICINE
Gluing the pieces together Illuminating the path to degrading troublesome proteins By Zuzanna Kozicka1,2
M
ost biological processes are driven by proximity—our cells rely on the right proteins meeting at the correct time and place. With the help of small molecules, we can interfere with these encounters and block or force certain partnerships. A particularly promising strategy, targeted protein degradation, involves bringing together the degradation machinery, typically E3 ligase proteins, and a cellular “offender” (1). E3 ligases act as garbage patrols, adding ubiquitin chains that serve as disposal tags to problematic proteins to direct them for proteasomal destruction. Traditional approaches to inactivate disease-causing proteins (e.g., putting a small-molecule wrench in the active site) cannot effectively disarm proteins with no defined binding pockets or those with nonenzymatic functions. Hijacking degradation machinery to break down an offender of interest is such an exciting strategy because it circumvents these limi-
1
Dana-Farber Cancer Institute, Boston, MA, USA. 2Friedrich Miescher Institute for Biomedical Research, University of Basel, Basel, Switzerland. Email: [email protected]
tations and makes the entire target protein disappear from the cell. Nevertheless, bringing two unrelated proteins together with a small, “drug-like” molecule is not easy. We know of only a few compounds capable of such E3 ligase–offender matchmaking, most of which were discovered by chance. These so-called molecular glue degraders leverage the everso-slight natural affinity two proteins may have for each other to bring them together in a tight, cooperative interaction. The flagship molecular glue degrader is thalidomide, a drug that was infamous for causing fetal limb deformities but was later redeemed as a treatment for multiple myeloma, a type of blood cancer. Years later, the drug was found to bind cereblon (CRBN), an E3 ligase receptor, and redirect the cellular garbage patrol to non-native and, notably, otherwise undruggable substrates, for example, IKAROS family zinc finger protein 1 (IKZF1) and IKZF3 transcription factors, which multiple myeloma cells need to survive (2–4). The astonishing clinical success of thalidomide analogs (e.g., Revlimid) sparked extensive research efforts in the field. The serendipitous discovery of another class of
Diversity in molecular matchmaking
CREDITS: (PHOTO) COURTESY OF ZUZANNA KOZICKA; (GRAPHIC) ADAPTED FROM Z. KOZICKA
The crystal structure of DNA damage–binding protein 1 (DDB1) and cyclin-dependent kinase 12 (CDK12)– cyclin K brought together by CR8 is shown. Surprisingly, many other diverse compounds (six examples are shown) promote the formation of an analogous complex, leading to cyclin K degradation.
Cyclin K
CDK12
DDB1
SCIENCE science.org
1117PrizeEssay_18204064.indd 779
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
779
11/10/23 5:11 PM
780
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117PrizeEssay_18204064.indd 780
CATEGORY WINNER: CELL & MOLECULAR BIOLOGY
Rachel Kratofil Rachel Kratofil received an undergraduate degree from the University of Victoria and a PhD in immunology from the University of Calgary. She is presently a postdoctoral fellow at New York University Grossman School of Medicine. Her research aims to uncover how barrier tissues adapt to microbial cues in the context of injury. www.science.org/doi/10.1126/ science.adl4292 CATEGORY WINNER: GENOMICS, PROTEOMICS, & SYSTEMS BIOLOGY
Yodai Takei Yodai Takei received undergraduate and master’s degrees from the University of Tokyo and a PhD from the California Institute of Technology, where he is presently a postdoctoral scholar in the Division of Biology and Biological Engineering. His research focuses on understanding spatiotemporal regulation of chromatin organization and gene expression. www.science. org/doi/10.1126/science.adl4460 CATEGORY WINNER: ECOLOGY & ENVIRONMENT
Jessica Kendall-Bar Jessica Kendall-Bar received undergraduate degrees from the University of California, Berkeley, and a PhD in Ecology and Environmental Biology from the University of California, Santa Cruz. She is presently a Schmidt AI in Science postdoctoral fellow at Scripps Institution of Oceanography at the University of California, San Diego. Her research investigates the resilience and precarity of ocean ecosystems through neurophysiology, signal processing, and advanced data visualization. www.science.org/ doi/10.1126/science.adl4885
cluding published inhibitors with cryptic degrader activity. Remarkably, through solving 28 additional crystal structures, we found that they all share a common mechanism. Despite their diversity, all compounds engage the kinase pocket of CDK12
and reach over to a particular residue on DDB1 (Arg928), albeit in different ways. This wealth of structural data allowed us to define the structural fingerprint of cyclin K degraders, design more potent and selective compounds, and tune the balance between degradation and kinase inhibition. Furthermore, our study yielded previously uncharacterized scaffolds for the selective inactivation of CDK12-cyclin K, which are emerging targets in oncology (11), and revealed transcriptional signatures for cyclin K degraders that are suggestive of their distinctive therapeutic utility. My thesis work has substantial implications for molecular medicine. It defined a functionally distinct class of molecular glue degraders and has provided the first large glue structure-activity relationship study, which has advanced our molecular understanding of how these compounds achieve their remarkable matchmaking. We also learned that the biochemical affinity of the ternary complex predicts the extent of target degradation observed in cells, a reassuring finding that can streamline initial degrader screening. Conceptually, our findings suggest that low-affinity protein-protein interfaces that feature a defined cavity provide exceptionally attractive opportunities for prospective glue design. Further, our work emphasizes that molecular glue degraders come in many different flavors. If one compares cyclin K degraders to thalidomide analogs, we propose that interface size and a compound’s relative contribution drive both their structure-activity relationship and substrate specificity. More broadly, these findings highlight the prodigious ability of bound compounds to modify protein surfaces and thereby induce new protein-protein interactions, an aspect largely overlooked until recently (12–14). Now that we are on high alert for such activities, the capacity to both harness serendipitous gluing and design drugs that forge new (or strengthen existing) matches will have a tremendous impact on the future of medicine. j REF ERENCES AND NOTES
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14.
A. D. Cowan, A. Ciulli, Biochem 91, 295 (2022). T. Ito et al., Science 327, 1345 (2010). J. Krönke et al., Science 343, 301 (2014). G. Lu et al., Science 343, 305 (2014). T. Han et al., Science 356, eaal3755 (2017). M. Słabicki et al., Nature 585, 293 (2020). Z. Kozicka, N. H. Thomä, Cell Chem. Biol. 28, 1032 (2021). C. Mayor-Ruiz et al., Nat. Chem. Biol. 16, 1199 (2020). T. M. Leissing, L. M. Luh, P. M. Cromm, Drug Discov. Today Technol. 37, 73 (2020). Z. Kozicka et al., Nat. Chem. Biol. (2023). P. Lei et al., Eur. J. Med. Chem. 240, 114603 (2022). L. H. Jones, Cell Chem. Biol. 25, 30 (2018). M. Słabicki et al., Nature 588, 164 (2020). E. J. Hanan et al., J. Med. Chem. 63, 11330 (2020). 10.1126/science.adl4288
PHOTOS: (T0P TO BOTTOM) COURTESY OF RACHEL KRATOFIL; COURTESY OF YODAI TAKEI; JEFF DILLON
matchmakers, aryl sulphonamides, ignited further enthusiasm and hinted that such compounds are not the rare, isolated marvels they were thought to be (5). Could we perhaps find more molecular glue degraders or even learn to prospectively design them and unleash their full potential in molecular medicine? Determined to address these questions, I began my doctoral work with Nicolas Thomä. Together with Benjamin Ebert’s group, we embarked on a systematic search for molecular glue degraders. We analyzed datasets of drug cytotoxicity and correlated them with E3 ligase expression levels across hundreds of cancer cell lines, reasoning that ligase-dependent toxicity could imply compound-induced degradation of essential proteins (6). This search led to the identification of CR8, a preclinical cyclin-dependent kinase (CDK) inhibitor, as a molecular glue degrader of the kinase coactivator cyclin K. We found that CR8 brings together CDK12–cyclin K and DNA damage–binding protein 1 (DDB1), leading to cyclin K ubiquitination. Our crystal structure revealed a distinctive arrangement, in which CDK12 unexpectedly replaced the ligase substrate receptor module (see the figure). This finding demonstrated that the prime consideration for productive complex formation is the right geometry for ubiquitination, an observation with important implications for degrader design. As we inspected the CDK12-DDB1 interface, we found that CR8 bound in the active site of CDK12 and noticed that its solvent-exposed pyridine ring, which is absent in related inhibitors, protrudes from the pocket and engages DDB1. This finding suggests more broadly that modification of surface-exposed moieties can confer new activities to inhibitors, a notion that provides a semirational approach for molecular glue discovery (7). We conducted another search for such matchmakers, led by Georg Winter’s laboratory, this time by comparing compound toxicity in cells with intact versus impaired degradation machinery (8). This search yielded a set of cyclin K degraders called dCeMM2 to 4, which, perplexingly, looked very different to CR8. Could these degraders even work through the same mechanism? After all, the glue acts as the pivotal puzzle piece that connects both interfaces, and even small chemical changes are expected to have major consequences, an observation that is well established for thalidomide analogs (9). To understand this chemical diversity, we systematically evaluated the cyclin K degrader structureactivity relationship, analyzing nearly 100 compounds (10). We identified more than 40 different molecular glue degraders, in-
science.org SCIENCE
11/10/23 5:11 PM
PRIZE ES SAY CATEGORY WINNER: CELL & MOLECULAR BIOLOGY
Rachel Kratofil
PHOTO: COURTESY OF RACHEL KRATOFIL
Rachel Kratofil received an undergraduate degree from the University of Victoria and a PhD in immunology from the University of Calgary. She is presently a postdoctoral fellow at New York University Grossman School of Medicine. Her research aims to uncover how barrier tissues adapt to microbial cues in the context of injury. www.science. org/doi/10.1126/science.adl4292
CELL & MOLECULAR BIOLOGY
Working up an appetite to promote repair Immune-derived hunger hormones restore tissue after infection By Rachel M. Kratofil
N
eutrophils, monocytes, and macrophages are among the first immune responders to an infectious agent. Although a strong inflammatory response is essential to clear bacteria, the subsequent tissue repair responses that restore homeostasis after infection remain less clear. Neutrophils, which have remarkable phagocytic abilities and antipathogen defenses (1), are recruited en masse from the blood and bone marrow to sites of infection, where they function to capture and kill bacteria. Monocytes are also recruited to infection sites, and it has been presumed that these cells mature into macrophages and aid neutrophils in bacterial clearance. However, the exact functional role of recruited monocytes during bacterial infection was unknown. Monocytes are incredibly plastic and can functionally adapt during inflammation and injury to promote tissue repair (2–4). Although monocytes can phagocytose bacteria, is this their bona fide in vivo function, or can recruited monocytes acquire reparative roles? I joined the University of Calgary laboratory of Paul Kubes, who specializes in intravital microscopy (IVM), a powerful imaging technique with which to visualize immune cell function in live mice. Using IVM, we can study immune cell recruitment, migration, spatial and temporal localization, and cell-cell interactions during infection and injury (3, 5–8). For my PhD, I developed a foreign-body skin infection model by infecting mice with a low inoculum of Staphylococcus aureus on a bead. The low inoculum and foreign body make for a physiologically relevant model compared with high doses of planktonic (free) bacteria injected into the skin (the latter causing substantial tissue damage). After infection, I found that neutrophils and monocytes were recruited in similar numbers but formed a specific localization pattern, with neutrophils in close contact with bacteria and monocytes sur-
Department of Pathology, New York University Grossman School of Medicine, New York, NY, USA. Email: [email protected] SCIENCE science.org
1117PrizeEssay_18204064.indd 779-B
rounding the neutrophils within the tissue (see the figure). IVM revealed that neutrophils were highly motile and phagocytosing bacteria, but monocytes were nonmotile and further away from bacteria, which suggested that monocytes were not in search of bacteria. These behavioral observations were bolstered by follow-up functional analyses of neutrophil and monocyte depletion. The former resulted in uncontrolled bacterial growth and dissemination to internal organs; by contrast, monocyte depletion had no effect on bacterial clearance or dissemination—a surprising observation that left us bemused. What were monocytes doing during infection? Monocytes were recruited to and actively migrated into a skin infection yet were not interacting with bacteria or participating in bacterial clearance. This puzzling observation indicated that monocytes had other unappreciated non– pathogen-killing functions that may limit infectious pathology and promote repair. The C-C chemokine receptor type 2 (CCR2) is required for monocyte egress from the bone marrow into blood (9). Therefore, CCR2-deficient mice have no circulating monocytes in naïve (uninfected) mice and no monocyte recruitment to the skin infection. Despite no defects in bacterial clearance, CCR2-deficient mice unexpectedly had delayed wound healing accompanied by a significant overgrowth of dysfunctional blood vasculature, which could be rescued with adoptive transfer of monocytes into CCR2-deficient mice after infection (8). My next challenge was to identify the angiogenic mediator that drove the increased vasculature response. I harvested skin infections from wild-type and CCR2-deficient mice and measured several inflammatory, repair, and angiogenic mediators, which led us to discover that leptin was substantially up-regulated in the wounds of CCR2-deficient mice after infection. Leptin is a metabolic hormone produced by adipocytes (fat cells) and is normally associated with the feeling of satiety after having eaten a meal (10). However, leptin has previously been described as an angiogenic factor, acting directly on Leptin receptor (LepR)–positive endothelial cells (11, 12). A prolonged ex17 NOVEMBER 2023 • VOL 382 ISSUE 6672
780-A
11/10/23 5:19 PM
Multiphoton intravital microscopy of infection and tissue repair mechanisms in skin Following skin infection with Staphylococcus aureus bead, neutrophils and monocytes are recruited to the infection site but have a unique spatial localization. Neutrophils migrate into the infection site and function to phagocytose and kill bacteria while monocytes remain at the periphery of the infection and are critical for tissue repair. In the absence of monocytes, with either CCR2-deficient mice or anti-CCR2 antibody depletion, bacterial clearance remained unchanged but there was a significant overgrowth of dysfunctional blood vessels in the wounds from elevated leptin. Leptin, produced by an expanded population of hypodermal adipocytes that accumulated at the wound, acted on LepR+ endothelial cells to promote angiogenesis. Recruited monocytes were an important source of the hunger hormone, ghrelin, which regulated angiogenesis and counteracted leptin’s pathologic effect to promote healing.
Monocyte
S. aureus
Neutrophil
Blood vasculature
Ghrelin
Collagen
Leptin
Adipocytes
topoietic cells to regulate the vasculature response after ghrelin-deficient bone marrow transfer. However, we cannot exclude the possibility of other hematopoietic cells providing this hormone and would require further experimentation to delineate the cellular source. My dissertation research unearthed monocytes and the hunger hormones leptin and ghrelin as key regulators of angiogenesis during infection and tissue repair. The combination of cutting-edge IVM and functional analyses in vivo allowed me to observe some unexpected biology under the microscope that ultimately changed the current dogma of how immune cells function during inflammation. This fundamental discovery of hormonal regulation of revascularization may extend beyond infection and apply to other diseases, such as cancer. j REF ERENCES AND NOTES
Neutrophils clear bacteria
Monocytes facilitate tissue repair Wildtype
Leukocyte recruitment to skin infection
Spatial segregation of monocytes and neutrophils
Immune-derived ghrelin regulates vasculature
Monocyte-deficient c
Leptin-driven pathologic angiogenesis
1. K. Ley et al., Sci. Immunol. 3, eaat4579 (2018). 2. R. M. Kratofil, P. Kubes, J. F. Deniset, Arterioscler. Thromb. Vasc. Biol. 37, 35 (2017). 3. D. Dal-Secco et al., J. Exp. Med. 212, 447 (2015). 4. K. Rahman et al., J. Clin. Invest. 127, 2904 (2017). 5. A. S. Neupane et al., Cell 183, 110 (2020). 6. B. G. Surewaard et al., J. Exp. Med. 213, 1141 (2016). 7. R. M. Kratofil et al., Nature 609, 166 (2022). 8. J. Wang et al., Science 358, 111 (2017). 9. C.-L. Tsou et al., J. Clin. Invest. 117, 902 (2007). 10. M. D. Klok, S. Jakobsdottir, M. L. Drent, Obes. Rev. 8, 21 (2007). 11. M. R. Sierra-Honigmann et al., Science 281, 1683 (1998). 12. A. Bouloumié, H. C. A. Drexler, M. Lafontan, R. Busse, Circ. Res. 83, 1059 (1998). 10.1126/science.adl4292
CCR2, C-C chemokine receptor type 2; LepR+, Leptin receptor-positive.
780-B
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117PrizeEssay_18204064.indd 779-C
gests that adipocyte-derived leptin drives pathologic angiogenesis. Once we parsed out this pathological role for leptin in delaying repair, I wondered precisely how monocytes were able to restrain this pathology. One factor known to counteract leptin’s action is the hunger hormone ghrelin, which is produced abundantly by the stomach when one feels hungry (10). Together, leptin and ghrelin maintain energy balance throughout the body, raising the tantalizing possibility that this axis may be co-opted in tissue angiogenic repones as well. To test whether ghrelin could inhibit leptin-driven angiogenesis, I injected recombinant ghrelin into the skin of CCR2-deficient mice after infection. This intervention led to a notable reduction in vasculature uncovering the involvement of hunger hormones in tissue repair. Could monocytes be a source of ghrelin? My research points to monocyte-derived ghrelin being a critical regulator on the basis of (i) increased Ghrl mRNA expression in infection-associated monocytes and (ii) the inability for ghrelin-deficient hema-
GRAPHIC: ADAPTED FROM R. M. KRATOFIL
pansion of hypodermal adipocytes and accumulation of adipocytes in the wounds of CCR2-deficient mice led us to hypothesize that adipocyte-derived leptin may be derailing the angiogenic response in the absence of monocytes. If so, how exactly did leptin promote angiogenesis? To investigate leptin’s role in angiogenesis, I injected recombinant leptin into the skin surrounding the wounds of wild-type mice and observed that there was increased angiogenesis compared with that in control-treated wounds. Consistently blocking leptin signaling with super-mouse leptin antagonist (SMLA) in monocyte-deficient mice ameliorated the hyper-angiogenic phenotype. Because endothelial cells robustly express LepR, we postulated that LepR signaling directly into endothelia fuels their pathology. Conditionally deleting LepR in endothelial cells confirmed that leptin drives angiogenesis through endothelial LepR. Adipocytes isolated from the wounds of CCR2-deficient mice expressed higher leptin as compared with that in adipocytes from wild-type mice, which strongly sug-
science.org SCIENCE
11/10/23 5:19 PM
PRIZE ES SAY CATEGORY WINNER: GENOMICS, PROTEOMICS, & SYSTEMS BIOLOGY
Yodai Takei
PHOTO: COURTESY OF Y. TAKEI
Yodai Takei received undergraduate and master’s degrees from the University of Tokyo and a PhD from the California Institute of Technology, where he is presently a postdoctoral scholar in the Division of Biology and Biological Engineering. His research focuses on understanding spatiotemporal regulation of chromatin organization and gene expression. www.science.org/doi/10.1126/ science.adl4460
GENOMICS, PROTEOMICS, & SYSTEMS BIOLOGY
Imaging nuclear architecture in single cells Multiplexed imaging uncovers precise three-dimensional maps of single nuclei By Yodai Takei
T
he smallest functional unit in our bodies, the cell, is spatially organized with various molecules, including DNA, RNA, and proteins. Studying the subcellular organization of molecules in individual cells is fundamental for understanding diverse cell types and states, from neurons to cancer cells. The cell’s nucleus—a highly organized membrane-bound organelle that contains genomic DNA in eukaryotes—is closely linked with complex gene regulation in higher organisms. Revealing its three-dimensional (3D) organization is therefore of considerable interest to a wide range of fields, including developmental biology, neuroscience, and human disease. Great advances have been made, from the groundbreaking observations of the nuclear structures in mammalian neurons under light microscopy by Santiago Ramón y Cajal in 1910 to modern genomics and imaging technologies (1, 2). However, technical limitations in the direct measurement of spatial organization of molecules in the nucleus still limit what insights can be gained into 3D nuclear architecture and gene regulation. When I started my doctoral program, I was interested in the challenge of visualizing the mammalian nucleus with its molecular components and understanding its organizational principles. Fluorescence imaging was particularly powerful for this purpose because it can directly visualize individual molecules in single cells. However, conventional fluorescence microscopy can only distinguish several target species at one time owing to the limited number of orthogonal fluorescent colors. This limitation made it extremely challenging to image global nuclear architecture, which required the measurement of many targets. During my doctoral studies in Long Cai’s laboratory at the California Institute of Technology, I contributed to
Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, CA, USA. Email: [email protected] SCIENCE science.org
1117PrizeEssay_18204064.indd 779-D
substantially improving the multiplex capabilities of fluorescence imaging with the development of transcriptome-scale (i.e., >10,000 RNA species) and genome-scale (i.e., >3000 DNA loci) imaging technologies (3–5). These new technologies enabled us to directly map 3D nuclear architecture across diverse mouse cell types and to address fundamental questions in nuclear organization and gene regulation at the single-cell level. The ability to connect 3D chromosome organization and transcriptomic information within individual cells would greatly advance our understanding of gene regulation in the nucleus. To this end, I codeveloped intron sequential fluorescence in situ hybridization (seqFISH), an imaging-based technology to profile nascent transcripts at the transcriptome scale in single cells (3). SeqFISH works by constructing temporal barcodes on single RNA molecules through sequential rounds of hybridization and imaging of fluorescently labeled probes (6). Because the nascent transcripts typically appear near their genomic loci, we can simultaneously profile transcriptional activities and chromosome organization in each cell. Through extensive optimization of the original implementations of seqFISH, we were able to scale up in situ transcriptomic profiling to the transcriptome scale (3, 6) (see the figure) and open the door to discovery-driven single-cell spatial transcriptomics. This technology can now provide direct views of the distribution of nascent transcriptomes in the nuclei. We revealed that in single mouse embryonic stem cells (mESCs), nascent RNA synthesis tends to occur at the surfaces of individual chromosome territories, whereas relative positions of chromosomes in the nucleus are variable. We also demonstrated that imagingbased nascent transcriptome profiles can accurately capture differences in cell states (e.g., cell cycles and pluripotency states) across hundreds of mESCs. To understand nuclear architecture, it is critical to record the organization of chromosomes, including noncoding regions, and then simultaneously compare 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
780-C
11/10/23 5:21 PM
Intron sequential fluorescence in situ hybridization (seqFISH) (top left) can visualize the nascent transcriptome (>10,000 genes) with single RNA molecules in the nucleus of a mouse embryonic stem cell (mESC) (3). Nascent intron spots are colored by 12 pseudocolors. Integrated spatial genomics can capture a wide range of nuclear features (top right), including chromosome structures at 1-megabase resolution by DNA seqFISH+, subnuclear structures, and 12 nuclear zones by sequential immunofluorescence in mESCs (4). Integrated spatial genomics can capture nuclear architecture across diverse cell types in the mouse brain cortex (bottom) (5). Cell types are transcriptionally identified by RNA seqFISH. In each cell type, specific sets of genomic loci (colored spots) are organized with their corresponding subnuclear structures, such as nuclear speckles (pink) and heterochromatic foci (green). Technology Intron seqFISH
Technology Integrated spatial genomics (RNA seqFISH, DNA
Modality RNA
Modality
seqFISH+, sequential immunofluorescence) DNA, RNA, Protein
20 chromosomes 15 subnuclear structures
Cells
Chromosome 1 3.1 Mb
Chromosome 3
195.2 Mb 3.6 Mb
159.0 Mb
Nucleus (mESC) Nucleus (mESC) 5 µm
Nucleolus (Fibrillarin)
RNA polymerase II (Ser5-P)
Nuclear zones
Single RNA molecules 500 nm
5 µm
Sequential rounds of imaging
Diverse cell types in the mouse brain Cell types
10 µm
Inhibitory neuron
Excitatory neuron
Astrocyte
Nucleus
2 µm
this with the higher-order spatial relationships between chromosome structures, nuclear bodies, chromatin states, and transcriptional states within single cells. I developed integrated spatial genomics to achieve this goal and was able to image chromosome structures along with mature and nascent RNAs and subnuclear structures in single cells (4) (see the figure). With this comprehensive technology, we created direct 3D maps of single-cell nuclear architecture—including DNA, RNA, and protein molecules—across hundreds of mESCs and revealed organizational principles of the nuclei in unprecedented 780-D
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117PrizeEssay_18204064.indd 779-E
detail. We found that specific DNA-protein spatial proximities occur consistently across individual cells, although the exact 3D organization of chromosomes and subnuclear structures is variable in single cells. In addition, we uncovered combinatorial chromatin states that separate individual nuclei into different nuclear zones with distinct transcriptional activities. Our single-cell multimodal imaging provides a key foundation for directly understanding the 3D nuclear architecture of the cell. In complex mammalian tissues, diverse cell types operate different gene expression programs to achieve cell type–specific
cellular functions. I aimed to understand how single-cell nuclear architectures relate to cell type–specific gene expression programs in mammalian tissues. Therefore, I applied the integrated spatial genomics approach to tissue sections of the mouse brain cortex (5) (see the figure). We profiled thousands of single cells from diverse cell types of neuronal and glial cells and were able to observe a range of cell type– specific nuclear features—e.g., nuclear and subnuclear morphologies, interchromosomal interactions, and physical scaling of chromosomes. Although we revealed that specific genome loci straddle specific subnuclear structures in each cell type, the degree of similarity across cell types was subnuclear marker specific. Notably, the association of genomic loci with nuclear speckles was similar among cell types and correlated with underlying genomic features (e.g., gene density). However, the subtle cell type–specific changes were well correlated with cell type–specific transcriptional profiles, which suggests that seemingly minor changes in the nuclear architecture can substantially affect cell type–specific gene expression programs. My thesis work demonstrated the ability of imaging-based single-cell genomics and transcriptomics to precisely map 3D nuclear architecture and elucidate underlying organizational principles. We can further integrate imaging-based multimodal profiling with additional spatiotemporal information, such as chromatin dynamics in live cells and cellular-level spatial organization within tissues, to understand the spatiotemporal nature of diverse biological systems across different scales (7, 8). We are at the very beginning of the imaging-based single-cell multiomics era. I anticipate an explosion of imaging-based single-cell multiomics in the coming years in combination with advanced computational and modeling approaches that will accelerate the discoveries of fundamental principles in single-cell biology and beyond (9, 10). j REF ERENCES AND NOTES
1. M. Lafarga, I. Casafont, R. Bengoechea, O. Tapia, M. T. Berciano, Chromosoma 118, 437 (2009). 2. J. Dekker et al., Nature 549, 219 (2017). 3. S. Shah et al., Cell 174, 363 (2018). 4. Y. Takei et al., Nature 590, 344 (2021). 5. Y. Takei et al., Science 374, 586 (2021). 6. E. Lubeck, A. F. Coskun, T. Zhiyentayev, M. Ahmad, L. Cai, Nat. Methods 11, 360 (2014). 7. Y. Takei, S. Shah, S. Harvey, L. S. Qi, L. Cai, Biophys. J. 112, 1773 (2017). 8. C. L. Eng et al., Nature 568, 235 (2019). 9. T. Zhou, R. Zhang, J. Ma, Annu. Rev. Biomed. Data Sci. 4, 21 (2021). 10. Y. Takei et al., bioRxiv 10.1101/2023.05.07.539762 (2023). 10.1126/science.adl4460
GRAPHIC: ADAPTED FROM Y. TAKEI
Single-cell nuclear architecture captured by multiplexed genomic and transcriptomic imaging
science.org SCIENCE
11/10/23 5:21 PM
PRIZE ES SAY CATEGORY WINNER: ECOLOGY & ENVIRONMENT
Jessica Kendall-Bar
PHOTO: JEFF DILLON
Jessica KendallBar received undergraduate degrees from the University of California, Berkeley, and a PhD in Ecology and Environmental Biology from the University of California, Santa Cruz. She is presently a Schmidt AI in Science postdoctoral fellow at Scripps Institution of Oceanography at the University of California, San Diego. Her research investigates the resilience and precarity of ocean ecosystems through neurophysiology, signal processing, and advanced data visualization. www.science.org/doi/10.1126/ science.adl4885
ECOLOGY & ENVIRONMENT
Lessons from sleep in the deep Records of seal sleep at sea reveal extreme sleep duration flexibility By Jessica Kendall-Bar
H
ave you experienced sleep deprivation after a fitful 2-hour slumber? Now imagine sleeping 2 hours a day for 7 months straight, like Mirounga angustirostris, the northern elephant seal. My PhD research unraveled the extreme sleep patterns of these massive deep-diving pinnipeds and explored the limits of mammalian sleep (1). Sleep deprivation in mammals impairs immune function, memory, and learning (2, 3). Humans require a remarkably consistent 6 to 9 hours of sleep per night, which varies by less than an hour across seasons (4). Consistent sleep time is common among mammals: Large herbivores such as African elephants sleep a low, consistent 2 hours per day, whereas most carnivorous mammals sleep around 12 hours per day (3). However, my research uncovered that the elephant seal switches between sleep extremes from season to season (1). In their breeding season on land, they sleep 10 hours per day, yet during months-long foraging trips to sea, they subsist on 2 hours of sleep, all while they navigate thousands of kilometers into the open ocean and back. Diverse sleep patterns provide new perspectives on the role of slow-wave sleep (SWS) and rapid-eye movement (REM) sleep. These two sleep modes convergently evolved in mammals and birds and are characterized by distinct electroencephalographic (EEG) signatures. Atypical sleep patterns, such as those in pinnipeds, fuel the ongoing debate over the function of REM sleep. Inconsistent evidence for REM sleep’s homeostatic regulation challenges its proposed critical role for memory and learning—for example, fur seals forgo REM sleep in water, with no apparent rebound (5). Instead, they sleep unihemispherically, with half of their brain awake and one eye open to monitor their surroundings, presumably watching for predators (6). However, elephant seals and other phocid seals sleep bilaterally (both hemispheres asleep), just like humans (1, 7). As bilat-
Center for Marine Biotechnology and Biomedicine, Scripps Institution of Oceanography, University of California, San Diego, San Diego, CA, USA. Email: [email protected] SCIENCE science.org
1117PrizeEssay_18204064.indd 779-F
eral sleepers, how do elephant seals evade killer whales and white sharks while sleeping at sea? So far, all marine mammal sleep studies had been in captivity (7, 8). To characterize sleep in a relevant ecological context, I became convinced we had to know where, when, and how seals sleep in the wild. The first challenge was to monitor sleep signals noninvasively, with surfacemounted EEG sensors, something never done previously with wild, free-ranging animals (7). The sensors were successful, and we were able to observe large slow waves through the seals’ thick (~2 to 3 cm) blubber layer. The next challenge was to create a seaworthy EEG device to follow the seals into the mesopelagic zone. I designed a neoprene headcap with gold EEG sensors and modified a datalogger to withstand the pressures of the deep sea, down to 2000 m (1, 7, 9) (see the figure, panel D). Would this new “sleep cap” work underwater during deep dives? With a research team from the Williams and Costa laboratories, we equipped five seals at University of California, Santa Cruz’s Long Marine laboratory and eight seals at Año Nuevo State Park with the marine sleep caps (1). In Año Nuevo’s shallow lagoons, we recorded large slow waves during long breath-holds (~10 min), which demonstrated that the sleep cap worked in harsh saltwater. Finally, we were ready to tackle a decades-old question: When, where, and how do marine mammals sleep at sea? We translocated two seals from Año Nuevo to Monterey and monitored their diving behavior as they swam back to the colony, 60 km across the deep Monterey canyon. Another seal instrumented on the beach startled us when she took an impromptu 2-day trip to sea. I was relieved when she returned but faced a new challenge to interpret gigabytes of at-sea sleep data: how to retrace these drifts in and out of consciousness in the pitch blackness of the Monterey submarine canyon. The EEG data showed that elephant seals packed naps into 10-min segments within 25-min dives to the seafloor (~200 m), or drifting through water as deep as 377 m (1). To connect the seals’ movements with their neurological state, I created 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
780-E
11/10/23 5:21 PM
Insights from sleep in seals at sea
REF ERENCES AND NOTES
(A) Bar plot of hours of sleep per day comparing sleep on land, on the continental shelf (< 200 m), and in the open ocean (> 200 m) above a schematic of risk in the ocean. Schematic reflects prevailing hypotheses of increased probability of attack by white sharks or killer whales near the surface. (B) Three-dimensional reconstruction of sleeping dive showing transition from wakefulness to slow-wave sleep (SWS), followed by the onset of spiraling during the transition to rapid-eye-movement (REM) sleep. (C) Sleep spiral showing body posture every 20 seconds and raw electroencephalogram (EEG) traces in the background. (D) Sleep monitor with EEG headcap, electrocardiogram (ECG) sensors, and data logger. (E) Range-wide “sleepscape” showing hours of sleep per day and quantitative sleep-estimate results for 323 adult female seals (334 records). (F) Three-dimensional tracks from one of three EEG-instrumented seals at sea with color representing sleep (SWS in blue and REM sleep in yellow). Land Hours of 10.8 h sleep per day
Continental shelf (< 200 m)
Open ocean (> 200 m)
2.5 h
1.7 h
B
C
Sleep spiral
23-minute dive ACKNOWL EDGMENTS Active waking Quiet waking
D
Light SWS Deep SWS
Logger ECG Headcap ECG
E
lower predation risk
REM
Deep SWS
Daily Sleep Estimates Long post-molt trip (N=140 seals): 2.2 ± 1.6 h / day Short post-breeding trip (N=183 seals): 1.2 ± 1.2 h / day
2 2000 km 0
three-dimensional visualizations of these first recordings of sleep at sea (1, 10, 11). This visualization revealed something unexpected: REM “sleep spirals” (see the figure, panels B and C). The sleep paralysis typically associated with REM sleep seemed to prevent the stabilization and upright posture that are possible during SWS. Unlike the sleep of other pinnipeds, a large proportion of these seals’ sleep at sea was this vulnerable, paralytic REM sleep (1, 5, 8). Their sleeping dives took them below sunlit surface waters into dark depths, where their predators were less likely to see them. The seals’ long, efficient breath holds liberated a safe sleeping niche deep below the waves. Using the Costa laboratory’s 20-year dataset of time-depth records from >300 adult female seals, I leveraged my EEG data to build a sleep-identification algorithm, to detect sleep across 3 million dives, and derive a population-level sleep estimate (1). During 7-month trips across the northern Pacific, seals slept only 2.2 hours per day and 1.2 hours for shorter 2-month trips (see the 780-F
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117PrizeEssay_18204064.indd 779-G
This work was supported by the National Ocean Partnership Program, National Science Foundation, Strategic Environmental Research and Development Program, Office of Naval Research, Joint Industry Project of the International Association of Oil and Gas Producers, a National Geographic Early Career Grant, a Steve & Rebecca Sooy Graduate Research Fellowship, Achievement Rewards for College Scientists, and a Special Research Grant from University of California, Santa Cruz. Data were collected under National Marine Fisheries Service Permit 23188. 10.1126/science.adl4885
F
Daily sleep estimate (hours per day) 8
4
Active waking
300 m
North Pacific sleepscape
6
J. M. Kendall-Bar et al., Science 380, 260 (2023). J. S. Durmer, D. F. Dinges, Semin. Neurol. 25, 117 (2005). J. M. Siegel, Nature 437, 1264 (2005). G. Yetish et al., Curr. Biol. 25, 2862 (2015). O. I. Lyamin et al., Curr. Biol. 28, 2000 (2018). J. M. Kendall-Bar, A. L. Vyssotski, L. M. Mukhametov, J. M. Siegel, O. I. Lyamin, PLOS ONE 14, e0217025 (2019). 7. J. M. Kendall-Bar et al., Anim. Biotelem. 10, 16 (2022). 8. O. I. Lyamin, J. M. Siegel, Handbook of Behavioral Neuroscience (Elsevier, 2019), vol. 30, pp. 375–393. 9. N. C. Rattenborg et al., Nat. Commun. 7, 12468 (2016). 10. J . Kendall-Bar et al., 2021 IEEE VIS Arts Program (VISAP), New Orleans, LA, 2021, pp. 1–10. 11. J. Kendall-Bar, Visualizing Life in the Deep: Code repository for visualizing marine mammal tag data (2021); https://github.com/jmkendallbar/ VisualizingLifeintheDeep. 12. R. S. Beltran et al., Sci. Adv. 7, eabd9818 (2021).
3 EEG records at sea (juveniles) 263 m 334 time-depth records at sea (adult females)
figure, panel E), which rivaled the African elephant’s current record of only 2 hours per day. Elephant seals exhibit unparalleled sleep flexibility among mammals, which challenges assumptions about baseline sleep requirements and chronic sleep deprivation in mammals. Northern elephant seals rely on REM sleep across habitats, but their access to REM sleep at sea is inextricably linked to their diving capacity (1). I mapped sleep time across the North Pacific, creating a range-wide “sleepscape,” which can help identify, manage, and protect critical resting habitats (see the figure, panel E). In years to come, sea-level rise may disrupt seals’ sleep on land, and shifting food availability with climate change may affect sleep at sea (1, 12). For endangered phocid seals with lowlying terrestrial habitats, such as Hawaiian monk seals, it is vital to consider sleep alongside conservation efforts. As fellow bilateral sleepers, these “extreme nappers” may also hold clues to our own reliance on sleep, its varied forms, and its elusive, yet critical role. j
GRAPHIC: ADAPTED FROM J. KENDALL-BAR
A
1. 2. 3. 4. 5. 6.
science.org SCIENCE
11/10/23 5:21 PM
RESEARCH IN S CIENCE JOURNAL S Edited by Michael Funk
ATMOSPHERIC AEROSOLS
Forming dimer esters
T
he physical and chemical properties of secondary organic aerosols (SOAs) depend on the species that they contain, but it is far from clear what those species are and how they were formed. Kenseth et al. report the structures of several types of dimer esters formed from the ozonolysis of a-pinene and b-pinene, major global sources of SOAs. The chemistry they describe likely represents a general pathway to dimeric compounds in ambient SOAs. —HJS Science, adi0857, this issue p. 787
Pinene, which contributes to the distinctive smell of conifer forests, reacts with ozone to form low-volatility products in secondary aerosols.
METABOLISM
PHOTO: JUNIORS BILDARCHIV GMBH/ALAMY STOCK PHOTO
Tuning mitochondrial glutathione levels Glutathione (GSH) has important roles as an antioxidant and in iron homeostasis and other cellular functions. It is transported into mitochondria by the transporter protein SLC25A39, and the abundance of SLC25A39 increases if concentrations of GSH in mitochondria are low, providing a feedback control to maintain GSH concentrations. Liu et al. report that the abundance of SLC25A39 is regulated by its association with a protease, AFG3L2. They propose that when mitochondrial GSH concentrations are low, binding of AFG3L2 to SLC25A3p (and thus its degradation) is SCIENCE science.org
1117ISIO_18230344.indd 781
prevented because an iron–sulfur cluster competitively binds to SLC25A30. Such a mechanism could help to coordinate iron homeostasis and maintenance of GSH concentrations in the mitochondria. —LBR Science, adf4154, this issue p. 820
INORGANIC CHEMISTRY
Gold and antimony have a ball The C60 fullerene molecule has attracted widespread attention for its high-symmetry structure resembling a soccer ball. Xu et al. now report a geometrically similar, fivefold symmetric metal cluster. The compound comprises 12 gold and 20 antimony atoms
on the surface that collectively share six delocalized negative charges and enclose a single positive potassium cation in the center. It was prepared by reaction of K8SnSb4 with a gold(I) phosphine complex in ethylenediamine solution and characterized crystallographically as a salt with external cryptand-sequestered potassium ions. —JSY
catalysts for the carbon monoxide insertion chemistry. Yoo et al. now report promising laboratory-scale results using more Earth-abundant nickel as the catalyst. Key to the robustness of the nickel under the reaction conditions is complexation with N-heterocycle carbene ligands in place of more conventional phosphines. —JSY
Science, adj6491, this issue p. 840
Science, ade3179, this issue p. 815
CATALYSIS
Carbonylation for a nickel The conversion of methanol to acetic acid and related ester carbonylation reactions are cornerstones of modern industrial chemistry. Current practice relies on precious rhodium or iridium
OCEAN HEAT
Dampening temperature change Approximately 90% of the excess heat that Earth has accumulated due to anthropogenic climate change has been absorbed by
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
781
11/14/23 9:08 AM
RESEARCH | I N S C I E N C E J O U R NA L S
the oceans, thereby moderating atmospheric temperature increases. Much of that heat has been transferred from the surface to the deep ocean by large-scale ocean circulation patterns. Was this mechanism as active before humans started warming the planet? Lu et al. report observations from subpolar North Atlantic sediments showing that the Atlantic Meridional Ocean Circulation has dampened atmospheric and upper ocean temperature changes for at least the past 1200 years by rapidly transferring heat from the surface to the deep ocean. —HJS
achieves strong performance in both perfect and imperfect information games, including chess, Go, Texas Hold’em Poker, and a game called Scotland Yard. This advance is an important step toward truly general algorithms for learning in arbitrary environments with minimal domain knowledge. —AJC
Science, adf1646, this issue p. 834
Although inverted perovskite solar cells minimize losses at hole-transport layers, recombination-induced losses occur at top electron-transport layers. Liu et al. used two different passivation molecules to tackle this problem. A sulfur-modified methylthio molecule provided chemical passivation, and a diammonium molecule repelled minority charge carriers and reduced contact-induced recombination. These cells had a certified quasi– steady-state power conversion efficiency and operated stably at 65°C for more than 2000 hours in ambient air. —PDS
Decitabine in the driver’s seat DNA methyltransferases (DNMTs) are increased in neuroendocrine prostate cancer (NEPC), an aggressive cancer associated with frequent metastases and poor clinical outcomes. Yamada et al. deleted DNMT genes in NEPC, leading to reduced tumor development and metastases in mice. The DNMT inhibitor decitabine also attenuated tumor growth in two preclinical cancer models. Because decitabine treatment increases the expression of the cell surface protein B7-H3, the authors combined decitabine treatment with an antibody-drug conjugate in B7-H3-low prostate cancers, leading to a synergistic response. —MLN Sci. Transl. Med. (2023) 10.1126/scitranslmed.adf6732
COMPUTER SCIENCE
Generalized game playing with AI Online games such as chess, Go, and Texas Hold’em Poker have long been used as proving grounds for artificial intelligence (AI) software, but AI strategies typically work on one game at a time. Schmid et al. developed a new general purpose AI program they call “Student of Games” that unifies multiple gameplaying approaches. The program 782
Sci. Adv. (2023) 10.1126/sciadv.adg3256
SOLAR CELLS
Doubling down on passivation
Science, adk1633, this issue p. 810
NEURODEGENERATION SOLID-STATE PHYSICS
Looking for higher symmetry The concept of emergent symmetry applies to systems in which a more symmetric state emerges at low temperatures from an initial high-temperature state. Demonstrating this emergence in an experiment is very challenging. In search of experimental signatures, Chudzinski et al. studied the one-dimensional material, Li0.9Mo6O17, which lies close to the Mott transition but can also host superconductivity. The researchers measured a surprisingly isotropic magnetoresistance, indicating that rather than being a Mott insulator or a superconductor, the system has fallen into a state with higher, emergent symmetry. —JS Science, abp8948, this issue p. 792
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117ISIO_18230344.indd 782
Edited by Caroline Ash and Jesse Smith
Gut microbes in Alzheimer’s disease Alterations to the community of microbes living in the gut have been observed in individuals with Alzheimer’s disease, a neurodegenerative disease that causes memory impairment. However, whether these changes have a role in the disease is not clear. Grabrucker et al. found that administering gut microbiome samples from individuals with Alzheimer’s disease to young rats lacking microbiota of their own resulted in impaired performance in memory tests. The impairment correlated with the severity of the disease in the individual from whom the sample was obtained and was associated with reduced survival of newborn neurons in the hippocampus (a brain region involved in memory
function). The findings indicate a role for the gut microbiome in Alzheimer’s disease, making it a potential therapeutic target. —SAL Brain (2023) 10.1093/brain/awad303
SIGNALING
b-arrestin signaling doesn’t stop here b-arrestins act as scaffolds for assembling various components of cell-signaling pathways, such as active G protein–coupled receptors (GPCRs) in the fundamental extracellular signalregulated kinase 1/2 (ERK1/2) cascade. It was assumed that b-arrestins acted simply by localizing the necessary components of a pathway. However, Kahsai et al. show that b-arrestins not only assemble signaling proteins
PHOTO: EDWIN GIESBERS/MINDEN
PROSTATE CANCER
IN OTHER JOURNALS
science.org SCIENCE
11/14/23 9:08 AM
AESTHETICS
reaction mechanisms of chemical processes at the solid−liquid interface. —YS
Watch that bird!
B
irds are one of the most visible components of wildlife, and people everywhere notice and usually appreciate them. Understanding what humans find attractive about birds not only reveals aspects of human cognition but may also help us improve conservation messaging and the perceived value of nonhuman species. Santangeli et al. used an online app to collect self-reported citizen science data on the attractiveness of various traits across nearly all bird species, including color, ornamentation, and distribution. They found that brightly colored birds, particularly those clad in blue and red, were ranked as the most attractive. However, more widely distributed birds were found to be more attractive than scarcer species, suggesting that people may be drawn to the more familiar than the rare. —SNV
J. Phys. Chem. Lett. (2023) 10.1021/acs.jpclett.3c02233
NANOMATERIALS
Mimicking technetium Technetium has no stable isotopes, but nanoparticles of a molybdenum−ruthenium− carbon solid solution appear to mimic its electronic properties. Okazoe et al. report that the nanoparticles, which were made by reducing the respective metal carbonyls and annealing, showed type II superconductivity for cubic nanoparticles containing at least 30% Mo. The transition temperature changed continuously with metal composition in this range and can be thought of as a continuous change in the electronic structure. Density function theory suggests that Mo0.53Ru0.47C0.41 has a similar electronic structure to TcC0.41 because both have a transition temperature of 3.8 kelvin. —PDS
npj Biodiversity (2023) 10.1038/s44185-023-00026-2
Rainbow lorikeets are familiar birds now found widely across Australia in part because of human interventions.
J. Am. Chem. Soc. (2023) 10.1021/jacs.3c06594
INTERSTELLAR MEDIUM but also allosterically modulate the kinase activity of ERK1/2 in cells for autophosphorylation and downstream substrate phosphorylation. Different arrestin isoforms have different phosphorylation capacity, thus allowing fine-tuning of signaling. —LBR Proc. Natl. Acad. Sci. U.S.A. (2023) 10.1073/pnas.2303794120
NEUROSCIENCE
The limitations of DCNNs Can deep convolutional neural networks (DCNNs) learn internal representations of faces like humans do? To test the utility of DCNNs as models of human cognitive and neural representations of naturalistic faces, Jiahui et al. developed a dynamic stimulus set comprising 707 naturalistic short video clips of faces. They then compared this stimulus set in terms of representational SCIENCE science.org
1117ISIO_18230344.indd 783
geometries produced by DCNNs with the behavioral responses of perceived similarity and categorical attributes by human raters and their brain activation measured by functional MRI. Although currently used DCNNs can successfully model representations of categorical face attributes, their utility for modeling human cognitive and neural representations of dynamic, naturalistic faces is limited and does not extend to human processes for face individuation. —PRS Proc. Natl. Acad. Sci. U.S.A. (2023) 10.1073/pnas.2304085120
SURFACE CHEMISTRY
Modeling chemistry at interfaces Surface chemistry at the solidliquid interface plays a key role in many important processes.
However, atomic-scale mechanistic understanding of it typically is absent because of the complexity of interfacial chemistry coupled with numerous external degrees of freedom. Hasegawa et al. proposed to combine a single-component artificial force induced reaction (SC-AFIR) method and an effective screening medium combined with the reference interaction site model (ESMRISM) for systematic studies of interfacial processes. As an example, SC-AFIR+ESM-RISM successfully explored the dissociation pathways of a water molecule at the Cu(111)−water interface and their energy dependence on the interfacial solvation environment modulated by NaCl. The present work is an important step in the development of computational tools for revealing and analyzing
3D maps of diffuse interstellar bands Diffuse interstellar bands (DIBs) are hundreds of broad optical and near-infrared absorption lines that appear ubiquitously in the interstellar medium. Their carriers are mostly unidentified but are thought to be large organic molecules. The Gaia Collaboration used spectra and parallaxes of over 6 million stars to map the three-dimensional (3D) distribution of two near-infrared DIBs using empirical templates and stacking. They found that these two DIBs had spatial distributions that were similar (but not identical) to solid dust grains. The DIBs had a low (but nonzero) abundance inside the Local Bubble, a region of low-density interstellar medium through which the Sun is currently passing. —KTS Astron. Astrophys. (2023) 10.1051/0004-6361/202347103
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
783
11/14/23 9:08 AM
RESEARCH
ALSO IN SCIENCE JOURNALS MICROBIOLOGY
Starving the gut microbiota The commensal bacteria in the gut provide many benefits to the host, such as extraction of nutrients and resistance to pathogens. Indeed, certain species are also associated with health and are being investigated as potential probiotics. To ensure that commensal bacteria survive when administered as probiotics and/or are encouraged to flourish through dietary intervention, it is important to understand what helps them to live in the gastrointestinal tract. In a Perspective, Groisman et al. discuss the importance of short-term nutrient starvation on commensal bacteria fitness, which has implications for the potential health benefits of dietary interventions such as fasting. The adaptations that allow these bacteria to survive nutrient deprivation could inform probiotic engineering and dietary interventions to increase the abundance of beneficial bacteria in the gut. —GKA Science, adh9165, this issue p. 766
CRISPR
How to choose a target Type V-K CRISPR-associated transposases (CASTs) catalyze the insertion of large genetic payloads into the genome with high efficiency and easy programmability. However, the accuracy of these multicomponent enzymes is currently limited, and the root cause of this shortcoming has been unknown. George et al. now show that type V-K CASTs exhibit an RNA-independent integration pathway primarily driven by the AAA+ ATPase TnsC (see the Perspective by Dhingra and Sashital). Untargeted transposition events occur preferentially at A/T-rich sites, with further context effects imposed by sequence motifs recognized by the transposase TnsB. Using mechanistic insights gained from biochemical and genetic experiments, the 783-B
authors tuned TnsC availability in cells and substantially improved overall integration specificity, enabling downstream applications for precision genome engineering. —DJ Science, adf8543, this issue p. 784; see also adl0863, p. 768
EPIGENETICS
How cells remember their identities Our cells remember their differentiated identities, including nerve, muscle, blood, etc., using chemical modifications placed along the genome known as epigenetic marks. However, it is not clear how epigenetic memory can be stable, because individual marks are constantly lost and rewritten. Owen et al. developed a model built on prior experimental findings, which reveals that the three-dimensional folding of the genome can help cells remember as long as the systems that write the marks satisfy certain design principles that they discovered. The model unites many classic observations and makes predictions that emerging experimental techniques can test. It also hints at a surprising analogy between how cells remember and how memories can be stored in neural networks. —DJ Science, adg3053, this issue p. 785
the Perspective by Romanov and Harkany). These sensors had a highly sensitive, specific, and robust response to their respective ligands in both cell lines and primary neurons without affecting endogenous signaling pathways. These new tools provide the opportunity to address key questions regarding neuropeptides, their function, and their role in both health and disease. —PRS Science, abq8173, this issue p. 786; see also adl1788, p. 764
ELECTROCALORICS
Cooling with electric fields Electrocaloric materials pump heat out of a system through a phase transition driven by changing an electric field. The strategy is an attractive alternative to vapor compression cooling because it is scalable and potentially more efficient. Li et al. developed a device that can generate a temperature difference of 20 kelvin or 4.2 watts of cooling power (see the Perspective by Tušek). The material does not break down under repeated field cycling, and the devices can be further optimized to compete with other cooling strategies. —BG Science, adi5477, this issue p. 801; see also adl0804, p. 769
NEUROSCIENCE
Sensors for neuropeptides Neuropeptides and their receptors are ancient, potent, and ubiquitous signaling molecules that can exert persistent control of physiology and behavior. However, despite the high functional importance of neuropeptides, when, where, and how they exert their effects in complex brain systems is poorly understood. Wang et al. developed and characterized a series of genetically encoded sensors for detecting neuropeptides (see
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117ISIO_18230344.indd 781-B
Edited by Michael Funk
BIOMATERIALS
Robust adhesion, rapid detachment There are many strategies in nature and biomedicine for establishing strong connections between living tissue and nonliving surfaces, but the mechanisms for separating these biointerfaces quickly and on demand are less well understood. Mytilus mussels can strongly adhere to inorganic surfaces, but they can also rapidly detach when threatened. Sivasundarampillai et al. used
advanced imaging and spectroscopy methods to study the detachment process (see the Perspective by Pan and Li). They found that the responsiveness of this quick release relies on the oscillating motion of cilia and subsequently the change of mechanical interaction between the byssus stem and mussel foot tissues. The beating movement can be influenced by the application of serotonin and dopamine, thus implicating neurotransmitters in controlling the mechanical interaction between living and nonliving tissues. —MSL Science, adi7401, this issue p. 829; see also adl2002, p. 763
GLOBAL WARMING
Slowing warming from waste Any chance that we may have to limit anthropogenic global warming to 1.5°C or 2.0°C, as stipulated by the Paris Accord, will require large and rapid decreases in greenhouse gas emissions across multiple sectors and species. Methane emissions from solid waste sites provide a considerable fraction of the global methane budget and are an important target for reductions. Hoy et al. report that the global solid waste sector is not currently on track to meet Paris targets unless abrupt interventions are made (see the Perspective by Webber and Glazer). With appropriate action, however, a net-zero warming contribution from the solid waste sector relative to 2020 can be realized. —HJS Science, adg3177, this issue p. 797; see also adl0557, p. 762
ANTHROPOLOGY
Making bonobo friends Humans are very good at cooperating with others outside of our family, kin, and cultural groups. Although such cooperation among individuals within science.org SCIENCE
11/14/23 9:09 AM
groups is also common in other animals, doing so outside of groups has rarely been observed. Samuni and Surbeck looked at cooperative behaviors such as grooming and food sharing in bonobos and found that individuals that cooperated more within their own group were also more likely to cooperate with those in other groups (see the Perspective by Silk). Furthermore, such cooperation was neither rare nor opportunistic. Social openness in one of our closest relatives suggests that our cooperativeness may be older than we thought. —SNV Science, adg0844, this issue p. 805; see also adl1813, p. 760
PHYSIOLOGY
Cell type matters for blood pressure
expression program remain poorly defined. Chopp et al. demonstrated that the paralog transcription factors Zfp148 and Zfp281 promote CD4+ T cell differentiation in mice by inducing the expression of key CD4+ lineage genes, including Thpok. In addition to their effects in DP thymocytes, loss of Zfp148 and Zfp281 in postthymic T cells impaired T helper 2 (TH2) cell effector functions in response to airway allergens. Zfp281 interacted with the transcription factor Gata3 and was recruited to Gata3-binding sites within genes encoding Thpok and type 2 cytokines. These findings identify Zfp148 and Zfp281 as key factors that cooperate with Gata3 to promote CD4+ lineage commitment and support TH2 responses. —CO Sci. Immunol. (2023) 10.1126/sciimmunol.adi9066
Ion channel activity in the endothelial cells that line blood vessels can alter blood pressure by changing the contractility of the surrounding smooth muscle. Mata-Daboin et al. found that the calcium-activated chloride channel TMEM16A in endothelial cells induced arteries to relax, in contrast to its vasoconstrictive effect in smooth muscle cells (see the Focus by Tammaro). Various vasodilators activated TMEM16A by stimulating calcium ion influx into endothelial cells. This effect was lost in mice with an endothelial cell–specific deficiency in TMEM16A, and these mice also had higher systemic blood pressure. —WW Sci. Signal. (2023) 10.1126/scisignal.adh9399; see also 10.1126/scisignal.adk5661
T CELLS
Nudging thymic T cell fate T helper cells develop in the thymus from double-positive (DP) precursors that express both CD4 and CD8 cell surface coreceptors. The transcription factors that control the induction of the CD4+ lineage gene SCIENCE science.org
1117ISIO_18230344.indd 781-C
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
783-C
11/14/23 9:09 AM
RES EARCH ◥
RESEARCH ARTICLE SUMMARY CRISPR
Mechanism of target site selection by type V-K CRISPR-associated transposases Jerrin Thomas George, Christopher Acree, Jung-Un Park, Muwen Kong, Tanner Wiegand, Yanis Luca Pignot, Elizabeth H. Kellogg, Eric C. Greene, Samuel H. Sternberg*
INTRODUCTION: Targeted insertion of large genetic payloads without DNA double-strand breaks remains a major challenge for genome engineering. CRISPR-associated transposases (CASTs) represent a promising alternative to nuclease- and prime editing–based approaches and involve the repurposing of nuclease-deficient CRISPR effectors to facilitate RNA-guided transposition. Type V-K CASTs offer several potential upsides compared with other homologous systems because of their compact size, easy programmability, and unidirectional integration behavior.
ualize single transposase molecules using fluorescence and electron microscopy. We reasoned that a deeper understanding of target site selection and transpososome assembly would reveal new opportunities for technology engineering and improvement. RESULTS: Using biochemical and cellular trans-
position experiments, we found that a representative CAST system from Scytonema hofmannii (ShCAST) was highly prone to catalyzing untargeted transposition in a reaction that proceeded independently of Cas12k and the guide RNA. Gene deletion experiments identified the minimal necessary machinery as TnsB, TnsC, and TniQ, and a cryo–electron microscopy (cryo-EM) structure revealed a BCQ transpososome complex that resembled the Cas12k-containing transpososome, with TnsC playing a major role in defining the overall architecture. Additional biochemical experiments identified TnsC as the primary driver of untargeted integration but also showed that TnsB exhibits an integration preference for RNA-targeted sites over untargeted sites. Next, using single-molecule experiments and meta-analyses of genome-wide integration data, we discovered that AT-rich regions are
RATIONALE: Despite these desirable properties,
type V-K CASTs exhibit poor fidelity compared with type I-F CASTs, and the molecular basis for this lack of specificity has remained elusive. We rationalized that determining the relative involvement of each CAST component during on- versus off-target insertion, including the guide RNA and Cas effector itself, would enable us to unravel the basis for this decreased specificity. To achieve this, we sought to monitor transposition using a combination of biochemistry and high-throughput sequencing, together with biophysical approaches, to vis-
TSD
associated transposases can exhibit both RNAguided and RNA-independent pathways and that the TnsC ATPase plays a major role in dictating target site selection. Whether both pathways are active in a native microbial context remains unknown, although we speculate that untargeted transposition likely represents the relic of an earlier, more primitive transposon lifestyle before CRISPR-Cas–targeting systems were acquired. This work highlights the importance of determining molecular mechanisms as an entry point to enable new opportunities for leveraging CASTs as an accurate, kilobasescale genome engineering tool.
▪
The list of author affiliations is available in the full article online. *Corresponding author. Email: [email protected] Cite this article as J. T. George et al., Science 382, eadj8543 (2023). DOI: 10.1126/science.adj8543
READ THE FULL ARTICLE AT https://doi.org/10.1126/science.adj8543
Bits
TnsC IIβ
0.1
TnsB consensus
Capping
15
5
10
0
-5
-15
-10
-25
-20
-35
-30
-45
-40
-55
-50
-60
-65
-70
TniQ Filament dis-/assembly
IIβ
TnsC
poly A/T
short poly A/T
K290
0.2
0.0
CONCLUSION: Our results reveal that CRISPR-
Untargeted (BCQ) DNA integration pathway
0.3 K290
preferred hotspots for untargeted transposition because of the binding specificity imparted by TnsC, and that TnsB also imposes local sequence bias to determine the precise insertion site. Knowledge of these motifs allowed us to direct untargeted transposition events to userdefined regions of a plasmid, and we confirmed the role of TnsC in mediating AT-rich preference by mutating a key DNA strand–contacting residue, K103. Finally, we harnessed knowledge of the role played by TnsC in directing untargeted transposition to design improved ShCAST vectors that suppressed RNA-independent transposition events and increased type V-K CAST specificity up to 98.1% in Escherichia coli without compromising the efficiency of ontarget integration.
TnsB-mediated disassembly
TnsB
Donor
TnsB
Transposition -25
-30
-35
-40
-45
TSD L
Tn
R
Mechanism of RNA-independent, untargeted integration by type V-K CASTs. We interrogated high-throughput sequencing datasets to reveal a consensus motif at untargeted integration events (left) characterized by TnsC- and TnsB-specific footprints. Together with single-molecule data and cryo-EM structures, these results revealed a BCQ transposition pathway (right) in which AT-rich sites are preferentially bound by TnsC filaments and capped by TniQ, leading to recruitment of TnsB-donor DNA complexes for downstream integration. 784
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
science.org SCIENCE
RES EARCH
RESEARCH ARTICLE
◥
CRISPR
Mechanism of target site selection by type V-K CRISPR-associated transposases Jerrin Thomas George1, Christopher Acree1†, Jung-Un Park2‡, Muwen Kong1, Tanner Wiegand1, Yanis Luca Pignot1§, Elizabeth H. Kellogg2‡, Eric C. Greene1, Samuel H. Sternberg1* CRISPR-associated transposases (CASTs) repurpose nuclease-deficient CRISPR effectors to catalyze RNA-guided transposition of large genetic payloads. Type V-K CASTs offer potential technology advantages but lack accuracy, and the molecular basis for this drawback has remained elusive. Here, we reveal that type V-K CASTs maintain an RNA-independent, “untargeted” transposition pathway alongside RNA-dependent integration, driven by the local availability of TnsC filaments. Using cryo– electron microscopy, single-molecule experiments, and high-throughput sequencing, we found that a minimal, CRISPR-less transpososome preferentially directs untargeted integration at AT-rich sites, with additional local specificity imparted by TnsB. By exploiting this knowledge, we suppressed untargeted transposition and increased type V-K CAST specificity up to 98.1% in cells without compromising on-target integration efficiency. These findings will inform further engineering of CAST systems for accurate, kilobase-scale genome engineering applications.
B
acteria encode diverse mobile genetic elements that exhibit a wide spectrum of transposition behaviors ranging from selective targeting of fixed attachment sites to promiscuous insertion into degenerate sequence motifs (1). Although insertion specificity is often dictated by a single recombinase enzyme (2, 3), some transposons encode heteromeric transposase complexes that distribute DNA target and integration activities across multiple distinct molecular components (4, 5). Tn7-like transposons are unique in this regard, in that they have evolved to exploit diverse molecular pathways for target site selection, including site-specific DNAbinding proteins (6), replication fork–specific DNA-binding proteins (7–9), CRISPR RNA– guided DNA binding complexes (10–12), and additional DNA targeting pathways that have yet to be characterized (13). CRISPR-associated transposases (CASTs), in particular, represent both a fascinating example of CRISPR-Cas exaptation and an opportune starting point for the development of next-generation tools for programmable, large-scale DNA insertion (14, 15). CAST systems characterized to date fall within either type I or type V classes, which differ in their reliance on either Cascade or Cas12k effector complexes, respectively (10–12, 16, 17). 1
Department of Biochemistry and Molecular Biophysics, Columbia University, New York, NY 10032, USA. 2Department of Molecular Biology and Genetics, Cornell University, Ithaca, NY 14853, USA.
*Corresponding author. Email: [email protected] †Present address: Department of Molecular Physiology and Biophysics, Vanderbilt University, Nashville, TN 37212, USA. ‡Present address: Department of Structural Biology, St. Jude Children’s Research Hospital, Memphis, TN 38105, USA. §Present address: Department of Biochemistry, Ludwig-Maximilians-University Munich, 81377 Munich, Germany.
George et al., Science 382, eadj8543 (2023)
Although the core transposition machinery is conserved across CAST families and includes a DDE-family transposase for integration (TnsB), an AAA+ ATPase for target site selection (TnsC), and an adaptor protein for CRISPR-transposition coupling (TniQ), key molecular features distinguish the integration behaviors of archetypal type I-F and type V-K systems. Whereas second-strand cleavage is catalyzed by the TnsA endonuclease in type I-F CASTs, leading to cut-and-paste transposition products, type V-K CASTs lack TnsA and instead mobilize through a copy-and-paste process, yielding cointegrate products (18–20). Type I-F CASTs achieve single-digit genomic integration efficiencies when expressed in mammalian cells, as opposed to low but detectable activity only on ectopic plasmid targets for an improved type V-K CAST homolog (20, 21). Additionally, heterologous expression of the CAST machinery from both systems yields vastly different integration specificities, with VchCAST (I-F) exhibiting mostly on-target activity in bacterial cells compared with an abundance of off-target insertions catalyzed by a representative V-K CAST system from Scytonema hofmannii (ShCAST) (11, 14, 15). Despite these differences, type V-K CASTs have a compact coding sequence composed of four components compared with type I-F CASTs (1666 versus 2748 amino acids) and integrate predominantly in a unidirectional orientation (11). The molecular basis underlying these distinguishing properties remains unexplored, particularly for type V-K CAST systems, limiting their practical application. Recent structural studies have provided new insights into the overall architecture of RNAguided, ShCAST transpososome complexes
17 November 2023
(22, 23). Target sites are marked by Cas12k binding (24, 25), in conjunction with TniQ and ribosomal protein S15, which engages the tracrRNA component (22), leading to stable R-loop formation reminiscent of other CRISPR effectors. In a key next step that is still poorly understood, TnsC assembles into filaments around double-stranded DNA, which can form adjacent to bound Cas12k-TniQ complexes (22) or on naked DNA (24, 26), acting as a platform for the subsequent recruitment of the TnsB transposase that is scaffolded along conserved binding sites in the transposon left and right ends. DNA integration then occurs through a concerted transesterification reaction at sites exposed by the TnsC filament, leading to transposons inserted at a fixed spacing downstream of the Cas12k-bound target site (11, 23). Whether a similar assembly pathway is operational at the many off-target integration events observed with ShCAST expression in cells, or if these represent an alternative transposition pathway, has not been systematically explored (Fig. 1A). Here, we set out to investigate the mechanism of target site selection for the archetypal type V-K CAST system from S. hofmannii, focusing special attention on the role of TnsC in regulating fidelity. We found that ShCAST is prone to extensive, RNA-independent transposition through a pathway that requires only TnsB, TnsC, and TniQ. Although these untargeted integration events initially appear random, analysis of high-throughput sequencing data revealed a bias for AT-rich sites, which was corroborated by single-molecule biophysical studies of TnsC DNA-binding behavior. By modulating DNA substrates in biochemical transposition assays, we demonstrated that the preference for AT-rich sequences could lead to predictable reaction outcomes. Furthermore, we found that transposition specificity could be substantially improved by limiting cytoplasmic TnsC levels, further highlighting the role of TnsC filament formation in pathway choice between RNA-dependent and RNAindependent transposition. Collectively, our results underscore the value of mechanistic studies in revealing new opportunities to engineer and leverage CAST systems as a potent DNA insertion technology. Results Type V-K CASTs perform RNA-dependent and RNA-independent transposition
Previous studies of the type V-K ShCAST system from S. hofmannii revealed that a considerable proportion of genomic integration events occurs at sites distant from the target site dictated by the guide RNA (11, 14, 15). To understand the molecular basis of these events, we applied a high-throughput sequencing approach to unbiasedly capture genomewide integration events upon ShCAST expression with various genetic perturbations (Fig. 1B 1 of 11
RES EARCH | R E S E A R C H A R T I C L E
TniQ
?
sgRNA
Donor
?
TnsB
L
Tn
RNA-dependent integration
D
Add NGS adaptors NGS & map to genome
R
RNA-independent integration
E
10 8
1.0
6 4 2
tnsB
100
Integration reads (%)
Normalized reads (X 105)
Tn
L
R
0
On-target
Untargeted
100
Enrich Tn-end sequences
? L
C
CmR R
Genomic DNA/plasmid Tn5-based tagmentation
% on-target
B
TnsC
50
0 100 50 0
sg RN A sg -1 RN A sg -2 RN A sg -3 RN A sg -4 RN A5
S15 Cas12k
On-target efficiency (%)
A
tnsC tniQ
L CmR R
pHelper 0
pDonor
4.5
0
F
0.5
1.0
1.5 2.0 2.5 3.0 E. coli genomic coordinate (Mb)
TnsC K103 & T121 3' Target DNA
TnsC ~25 bp
5' TnsB STC
Transposon DNA
tnsB
G
TniQ
TniQ
~13 bp
3.5
tnsC tniQ
pHelper
25
Normalized reads (X 104)
∆C
as ∆C 12 as k, 12 ∆s k, gR ∆s WT N gR A, N ∆T A n ∆T iQ ns B ∆T ns C
0.0
4.0
cas12k
+ pDonor
4.5
sgRNA
pCas12k
On-target Untargeted
20 15 10 5 0 Lac
J23114
J23105
J23119
Promoter strength
Fig. 1. Type V-K CASTs direct frequent Cas12k- and RNA-independent transposition events. (A) Schematic of type V-K CAST transposition occurring at on-target sites (RNA-dependent) and untargeted sites (RNAindependent). (B) Experimental TagTn-seq pipeline used for in vitro and genomic samples. (C) Fraction of total genome-mapping integration reads detected at on-target and untargeted sites for the WT pHelper expression plasmid across multiple sgRNAs (top), plotted above on-target transposition efficiencies for the same sgRNAs as measured by Taqman qPCR (bottom). (D) Total genome-mapping reads detected for WT pHelper or pHelper with the indicated deletions, normalized and scaled. (E) Magnified view of integration reads comprising ≤1% of E. coli genome-mapping reads in an experiment performed without Cas12k and guide RNA. (F) Cryo-EM reconstruction of the untargeted transpososome revealing the assembly of TniQ (orange), TnsC George et al., Science 382, eadj8543 (2023)
17 November 2023
(green), and TnsB (purple) in a strand-transfer complex (STC). The target DNA and transposon DNA are represented in light blue and dark blue, respectively. For visualization, a composite map was generated using two local resolution-filtered reconstructions from the focused refinements. Magnified and cutaway views show TnsC forming a helical assembly on the target DNA, positioning residues K103 and T121 (pink) adjacent to one strand of the target DNA (dark blue). The 5′ and 3′ ends of the TnsC-interacting DNA strand are indicated. Two turns of TnsC and TnsB footprint on DNA until TSD cover ~25 and 13 bp, respectively. Only selected TnsC monomers are represented in the cutaway for clarity. (G) Cas12k and the sgRNA were cloned onto a separate vector, and the promoter driving Cas12k expression was varied. Reads detected at on-target and untargeted sites during transposition assays were normalized and scaled. For (C), (D), (E), and (G), the mean is shown from N = 2 independent biological replicates. 2 of 11
RES EARCH | R E S E A R C H A R T I C L E
and materials and methods). After testing five distinct single-guide RNAs (sgRNAs), we found that the fraction of on-target integration events ranged from 12 to 76%, and that most events occurred elsewhere, with DNA insertions seemingly randomly distributed across the genome at low individual frequencies (Fig. 1C and fig. S1, A and B). We analyzed their proximal genetic neighborhood and failed to detect enriched sequence similarity to the guide RNA (fig. S1, C to E), suggesting that these events were not mismatched off-targets aberrantly targeted by RNA-guided Cas12k, but rather were the consequence of RNA-independent transposition; therefore, we tentatively referred to these as untargeted integration events (Fig. 1C). When we deleted cas12k and the sgRNA from the original pHelper expression plasmid, the CRISPRlacking ShCAST system still produced efficient genome-wide transposition products (Fig. 1, D and E) (11). These results establish that type V-K CAST systems are capable of both RNAdependent targeted DNA integration and RNAindependent untargeted DNA integration. We performed additional control experiments and confirmed that TnsC, an AAA+ regulator, and TnsB, the DDE-family transposase, are essential for both RNA-dependent and RNAindependent transposition, because their deletion completely abrogated integration (Fig. 1D). We initially hypothesized that TnsB and TnsC would comprise the minimum necessary protein components for RNA-independent transposition, similar to the reliance of phage Mu transposition on two homologous gene products, MuA and MuB (27). However, tniQ deletion had a severe effect on untargeted transposition (Fig. 1D), suggesting a crucial role in stabilizing and/ or interacting with the TnsBC transpososome. Recent structures revealed that DNA-bound TnsC oligomers are capped on the N-terminal face by one or more TniQ protomers (22, 26), and our biochemical experiments similarly demonstrated that TniQ only stably associated with DNA in the presence of TnsC, as reflected by fluorescence polarization experiments (fig. S1F). Thus, much like the requirement for TnsB, TnsC, and TniQ in transposition by Tn5053 (28), we conclude that ShCAST—and perhaps type V-K CAST systems more generally— maintain a prominent BCQ pathway that facilitates CRISPR RNA–independent, untargeted transposition. To capture the architecture of components contributing to untargeted integration, we used cryo–electron microscopy (cryo-EM) to visualize TnsB, TnsC, and TniQ in a strand-transfer complex (STC). Although the cryo-EM density of TniQ was less well resolved (~8 Å) compared with other subunits (fig. S2 and table S1), likely due to heterogeneity of binding configurations, we were able to unambiguously dock atomic models of all protein components and DNA into the map (Fig. 1F). The overall structure of George et al., Science 382, eadj8543 (2023)
the BCQ transpososome resembled the Cas12kcontaining transpososome (fig. S3A), with two turns of TnsC filaments preferentially selected even with free DNA available at the TniQ end. Further, DNA-interacting residues of TnsC (K103 and T121) were positioned to follow the helical symmetry of duplex DNA, as in the structure of helical TnsC filaments (22–24, 26). However, in contrast to the Cas12k-containing transpososome (23), the polarity of the interacting DNA strand in the BCQ transpososome was 3′ to 5′, following the direction of TniQ- to TnsB-binding face of TnsC, which was also noted in the structure of TniQ-TnsC (22, 26) (fig. S3B). Therefore, the BCQ transpososome structure, which represents a low-energy configuration of TnsC, reveals that DNA contacts in TnsC filaments are maintained differently at on-target and untargeted sites. Yet, the overall architecture of the BCQ transpososome, comprising the TnsB STC, two turns of a TnsC minifilament [spanning a DNA-binding footprint of 25 base pairs (bp)] and TniQ, resembles the on-target Cas12k-bound transpososome. We next sought to determine whether the presence of Cas12k and an appropriate sgRNA would reduce the frequency of untargeted transposition events by sequestering protein components at the on-target site. After cloning cas12k onto a separate expression plasmid and systematically varying its promoter strength, we found that on-target integration events were proportionally increased, although without a reduction in the frequency of untargeted integration (Fig. 1G and fig. S1G). These results indicate that, at least under these expression conditions, the availability of Cas12k-sgRNA complexes limits RNA-guided DNA integration efficiency but does not directly affect the BCQ pathway. Type V-K CAST systems often encode a MerR-family transcriptional regulator adjacent to the Cas12k gene (12, 29), and a recent study demonstrated that these Cas V-K repressor (CvkR) proteins down-regulate both Cas12k and TnsB expression, although with distinct effects (30). Thus, although our present knowledge about CAST activity is largely limited to comparative genomics and artificial heterologous overexpression, it appears likely that CAST transposition in native contexts is regulated to modulate the frequency of RNA-dependent and RNA-independent target pathways. Relative TnsB and TnsC stoichiometry determines the transposition pathway choice
Many other bacterial transposons encode transposition proteins homologous to ShCAST, including type I CASTs, Tn7, Tn5053, IS21, and Mu (4, 8, 10, 27, 28). The TnsBC module is common to all, and in the case of Mu, the AAA+ ATPase component known as MuB plays a dominant role in directing untargeted, genomewide transposition by recruiting the MuA trans-
17 November 2023
posase to potential integration sites (5). Moreover, structural studies have demonstrated that MuB and ShTnsC both form continuous, nonspecific filaments on double-stranded DNA (dsDNA) (5, 24, 26), in contrast to the discrete closed or semiclosed rings formed by TnsC from VchCAST (Tn6677) and Escherichia coli Tn7 (31, 32). Therefore, we set out to experimentally investigate the role of TnsC in target site selection and the effect of variable stoichiometries of TnsB, TnsC, and TniQ on untargeted integration. However, one of the major hindrances that we encountered while trying to vary the expression of transposon components in cells was the associated toxicity, particularly with the overexpression of TnsC (Fig. 2A and fig. S4, A and B). When we inoculated liquid cultures with a strain expressing tnsC alone from a strong promoter and induced overexpression in the lag phase, we observed a complete growth arrest for most of the clones, with only a few strains undergoing delayed growth, likely due to suppressor mutations in the plasmid or genome (fig. S4A). This cellular toxicity was completely rescued with a mutation to the arginine finger motif, which abrogates TnsC filamentation (33) and transposition, or was partially rescued by coexpression of TnsC and TnsB (Fig. 2A and fig. S4C). These results implicate nonspecific DNA filamentation as a likely source of cellular toxicity, which can be relieved in part by the ability of TnsB to disassemble TnsC filaments, as demonstrated from in vitro experiments (24, 26, 34). To modulate the stoichiometries of CAST components contributing to untargeted integration, while avoiding confounding factors such as toxicity, we adopted a biochemical approach. After recombinantly expressing and purifying ShCAST components and testing the activity of TnsC and TnsB in vitro (fig. S4, D to G), we established a plasmid-to-plasmid (pDonor-to-pTarget) transposition assay (Fig. 2B). In initial experiments, we amplified ontarget products by targeted polymerase chain reaction (PCR), thereby revealing molecular requirements for each of the transpososome components and the expected distance separating the target and integration site (fig. S5, A to C). Next, we coupled our biochemical experiments with tagmentation-based highthroughput sequencing to unbiasedly map DNA transposition events regardless of their insertion site (Fig. 2B and materials and methods). We found that, at low (0.1 mM) concentrations of TnsC, transposition was highly accurate, with >99% of reads representing on-target integration events, defined as occurring within a 100-bp window downstream of the target site (Fig. 2, C and D). However, when we systematically increased the concentration of TnsC while keeping all other components constant, the frequency of untargeted integration events increased, approaching levels similar to those 3 of 11
RES EARCH | R E S E A R C H A R T I C L E
0.2
//
NGS & map to plasmid
On-target
0
200
400 600 Time (min)
pTarget coordinate
800
pDonor coordinate
E 20 10 0
On-target specificity (%)
On-target region
Untargeted region
[0 - 75000] [TnsB]: 1 µM
1,000
100
50
0
F
2 µM
[0 - 170]
Normalized reads – 1 µM TnsC
Total reads (X 106)
D
1 µM
On-target 1,000 1,500 pTarget coordinate (bp)
500
0.0
0.5 µM
0.1
0.25 0.5 1 [TnsC] (µM)
2
r = 0.84
100
10
5 µM
1 1 10 100 1,000 Normalized reads – 0.5 µM TnsC
Fig. 2. Biochemical reconstitution of transposition reveals distinct efficiencies at on-target and untargeted sites. (A) Growth curves upon induction of WT or mutant TnsC with or without TnsB. Data are shown as mean ± SD for N = 2 independent biological replicates inoculated from individual colonies. (B) Assay schematic for probing in vitro plasmid-to-plasmid transposition events using recombinantly expressed CAST components. (C) In vitro integration reads mapping to pTarget from experiments in which TnsC was titrated from 0.1 to 2 mM. Data were normalized and scaled to highlight untargeted integration events relative to on-target insertions. (D) On-target specificity from biochemical transposition assays at varying TnsC concentrations, calculated as the fraction of on-target reads divided by total plasmid-mapping reads (bottom). Total
observed in cellular transposition assays (Fig. 2D and fig. S5E). Substantial untargeted integration events also occurred in the absence of Cas12k and sgRNA under these conditions, in agreement with in vivo experiments (fig. S5D). TnsC concentrations of 1 mM or higher resulted in a decrease in both on-target and untargeted integration, which may be have been caused by the prohibitive coating of DNA by TnsC filaments (see below). When interpreted together with structural data, these results suggest that RNA-independent transposition is likely initiated by the formation of dsDNA-bound TnsC filaments. Untargeted integration events were not randomly distributed across pTarget but instead were clustered into specific and reproducible hotspot regions (Fig. 2E), suggesting a selectivity for certain, as-yet-undetermined sequence features (see below). George et al., Science 382, eadj8543 (2023)
2 µM
1,500
2,000
pTarget coordinate (bp)
2,500
G 15 10 5 0
100
50
0
1 2 5 [TnsB] (µM)
integration activity also decreased as a function of TnsC concentration, as seen by the normalized plasmid-mapping reads (top). (E) Scatter plot showing reproducibility between untargeted integration reads observed in vitro at two high TnsC concentrations; each data point represents transposition events mapping to a single base-pair position within pTarget. The Pearson linear correlation coefficient is shown (two-tailed P < 0.0001); on-target events were masked. (F) Normalized integration reads detected at a representative untargeted site (left) and at the on-target site (right), with 1 mM TnsC and the indicated TnsB concentration. Note the differing y-axis ranges. (G) On-target specificity from biochemical transposition assays at 1 mM TnsC and the indicated TnsB concentration, shown as in (D).
We next tested the impact of other protein components on in vitro transposition activity. Ribosomal protein S15, a recently described host factor that stimulates ShCAST transposition by binding the sgRNA (22), substantially increased the frequency of on-target integration events, as measured both by deep sequencing and quantitative PCR (qPCR), but had no discernible effect on untargeted integration events (figs. S5, G and H). However, increasing the concentration of TniQ led to a monotonic increase in the frequency of untargeted integration events without a major effect on ontarget integration (fig. S5F), suggesting that the RNA-independent pathway may be more sensitive to limited TniQ availability. The TnsB transposase has been previously shown to disassemble TnsC filaments from dsDNA (24, 26, 34), so it is possible that tit-
17 November 2023
1,000 1,100
2,000
Total reads (X 104)
TagTn-seq library preparation
0.25 µM
Normalized reads
OD600
0.6
[TnsC]: 0.1 µM
On-target specificity (%)
Target
Full range not shown
//
CmR R
37 °C, 2 h
0.4
[0 - 350]
//
L
pTarget pDonor
//
Cas12k + sgRNA
C
//
0.8
TniQ S15 TnsC TnsB
//
B
TnsC TnsC + TnsB TnsC R189E TnsC R189E + TnsB Empty vector Empty vector + TnsB
Normalized reads
A
rating excess amounts of TnsB would lead to partial or full disassembly of TnsC filaments necessary for transposition, regardless of their molecular context. However, when we varied the amount of the TnsB transposase, we observed distinct effects at on-target and untargeted sites (Fig. 2F and fig. S5, I to K). Increasing TnsB led to a notable increase in RNA-guided integration but resulted in a slight decrease in untargeted events (fig. S5, I to K), leading to an overall rescue of specificity at on-target sites with high TnsC concentrations. This observation suggests that TnsC filaments at targeted versus untargeted sites are differentially susceptible to TnsB-induced disassembly and/or react to undergo strand transfer with distinct kinetics. Transpososome structures reveal that TnsB interacts with TnsC filaments on only one face (23, 34), and no major structural changes are associated 4 of 11
RES EARCH | R E S E A R C H A R T I C L E
with TnsC filaments at on-target and untargeted sites (Fig. 1F). Therefore, we suggest that the distinct nature of DNA interactions made by TnsC at both of these sites determines filament stabilization versus disassembly. Collectively, these results suggest that the natural propensity of TnsC to form long filaments on dsDNA exerts a fitness cost on cells in the absence of accessory transposase machinery and is a driver of RNA-independent, untargeted transposition. We next sought to investigate whether TnsC exhibits any bias when selecting RNA-independent sites for transposition. TnsC preferentially targets AT-rich DNA during RNA-independent transposition
We developed a single-molecule approach to visualize DNA binding by TnsC using DNA curtains (Fig. 3A), in which l-phage genomic DNA molecules are tethered between chrome patterns on a quartz slide and imaged by total internal reflection fluorescence microscopy (35). Fluorescently labeled TnsC remained fully active for RNA-guided transposition, albeit with slightly increased specificity relative to wild-type (WT) (fig. S6A), suggesting that the N-terminal appendage may subtly affect DNA binding and/ or transpososome assembly. In DNA curtains experiments, TnsC exhibited stable and highaffinity binding in the presence of ATP, and the data furthermore revealed a marked preference for the 3′ half of the genome (Fig. 3B). The l-phage genome is known to be divided into a GC-rich half and an AT-rich half (36), and our analyses revealed a significant correlation between AT content and TnsC localization (Fig. 3C), indicating that TnsC filaments preferentially accumulate on the AT-rich half of the l-phage genome. Time-course experiments further revealed that TnsC binds to AT-rich regions at a faster rate before saturating the entire l-DNA substrate within 5 to 10 min of incubation (Fig. 3D and fig. S6B and movie S1). A preference for AT-rich regions has been previously observed for MuB in both single-molecule microscopy experiments and in vivo transposition studies (37, 38), supporting the idea that this property is likely to be broadly conserved across AAA+ regulators from other transposon families. Incubation of DNA curtains with high TnsC concentrations resulted in complete coating of the l-DNA substrate (fig. S6C and movie S2), which could explain the decrease in both on-target and untargeted integration observed in biochemical transposition assays at similarly high TnsC concentrations (Fig. 2C and fig. S5E). Given our observation that untargeted transposition events in biochemical assays preferred certain hotspot regions of pTarget and were reproducible between independent experiments (Fig. 2E), we hypothesized that AT content might be an underlying feature explaining George et al., Science 382, eadj8543 (2023)
these data. We analyzed the nucleotide composition surrounding all unique integration events on pTarget and found that they were indeed skewed toward more AT-rich DNA (fig. S6, D and E, and materials and methods). Direct visual superposition of AT content and DNA integration data further revealed that hotspot regions for untargeted transposition in pTarget generally correlated with regions of higher AT content (fig. S6F). We observed the same phenomenon after performing transposition assays with a l-DNA substrate and repeating similar analyses to assess AT bias in the location of untargeted integration sites (fig. S6, G to I). Finally, we analyzed untargeted integration events in the E. coli genome from experiments performed without Cas12k and sgRNA and found that these were also highly enriched at local regions of high AT content (Fig. 3E and fig. S6J). These results provide evidence that ATrich sites on DNA are preferentially bound by TnsC, and thus are preferentially “targeted” for RNA-independent transposition. We next sought to uncover additional sequence features common to CRISPR-independent ShCAST transposition products. We performed a meta-analysis of all genome-wide integration sites after orienting the flanking sequences based on the asymmetric transposon ends, and then generated a consensus sequence logo of the resulting alignment (Fig. 3G and materials and methods). This analysis revealed two notable clusters of sequence features: nucleotide preferences directly within and surrounding the target-site duplication (TSD), and an AT-rich nucleotide cluster located upstream of the integration site (Fig. 3H). The AT-rich region spans ~25 bp and could thus accommodate two turns of a dsDNA-bound TnsC filament, similar to the TnsC architecture and footprint observed within the context of Cas12kcontaining and Cas12k-lacking transpososomes (Fig. 1F) (23). The observation that this region is located on only one side of all integration sites suggests that RNA-independent integration events result from binding of TnsC filaments to AT-rich DNA, followed by directional recruitment of TnsB to define downstream sites for transposon insertion in the same leftright (L-R) orientation as occurs at RNA-guided target sites (11) (Fig. 3I). In general, we refer to this orientation as TnsC-LR, which could be applicable to other systems using a AAA+ ATPase for integration. We found that higher sequence conservation was located farthest from the site of integration, proximal to the presumed region where TnsC filaments are capped by TniQ (Fig. 1F), and the observed dinucleotide periodic trend is reminiscent of structures demonstrating that TnsC monomers contact every two bases of DNA (24, 26). The greatest conservation in the sequence logo corresponds to sequences contacted by TnsB within the BCQ transpososome. As with
17 November 2023
prior library-based experiments for both type I-F and V-K CAST systems, our results indicate that TnsB preferentially integrates into sites containing GCWGC within the TSD (Fig. 3H) (11, 39). However, we also uncovered a bias for (A/T) at symmetric positions located ±5 bp from the TSD center, which is contacted by residue K290 of two TnsB monomers within the transpososome (Fig. 3J). Nucleotide preferences ±12 bp from the TSD could also be explained by the proximity of these residues with the TnsB IIb DNA-binding domain (R416, T417, Q425, and N428), which also makes similar sequence contacts to the penultimate TnsBbinding sites located within the transposon left and right ends (23, 34, 40). When we analyzed untargeted integration events from our previously published ShCAST data (14), in which NGS libraries were generated and sequenced using an alternative strategy, we observed the same sequence features, confirming the robustness of this observation (fig. S7A). The absence of any conserved sequence features upstream of the AT-rich region, where the target site would normally be located during Cas12k-mediated transposition, corroborated our earlier interpretation that most of the cellular transposition events were RNA-independent. Previously, it was shown that a K103A point mutation, one of the two TnsC residues that contact DNA, increased the number of untargeted events without compromising the ability of TnsC to bind DNA (26). In agreement with these results, when we tested the same mutant in cellular transposition assays, we observed a severe loss of on-target events but a preservation of untargeted events, which were enriched near the E. coli origin of replication (figs. S4C and S6K). When we performed a meta-analysis of untargeted integration events, we found that the TnsC K103A mutant no longer exhibited an A/T preference, in contrast to WT TnsC (Fig. 3F and fig. S7B). This observation, together with the loss of on-target integration (fig. S4C), suggests that the K103A mutation results in a more promiscuous mode of DNA binding that supports integration anywhere in the genome without specific sequence requirements. We previously reported transposition activity for a type V-K CAST homolog also found in S. hofmannii, ShoCAST (previously referred to as ShoINT), which is diverged from ShCAST and more similar to AcCAST (11, 14). We were curious as to whether untargeted ShoCAST transposition events would exhibit similar sequence preferences as ShCAST, so we performed meta-analyses on published transposition data (14). Highly similar motifs emerged in the resulting sequence logo, but with a major difference in the window of AT-rich DNA located upstream of the integration site, which spanned only ~10 bp compared with ~25 bp observed with ShCAST (fig. S7C). This difference is consistent with the finding that ShoCAST and 5 of 11
RES EARCH | R E S E A R C H A R T I C L E
B
C
P
Pedestal (P)
λ-DNA ATP
mNG-TnsC
B
P
r = 0.887 25
60 50
20
40 15
Fluorescence intensity
AT conrtent (%)
70
D
AT content Fluorescence intensity
30 0
20
F
1.0
0.5 Wild-type TnsC 0.0 20
40 60 AT content (%)
AT distribution in E. coli genome
1.0
Tn
R
Tn
R
TSD
R
0
Coordinates
K103A TnsC
Untargeted L
Sequence conservation
80
TniQ
J
H TSD
TnsC
IIβ
K290
ShCAST ΔCas12k ΔsgRNA
0.3 K290
K290
TSD
Bits
0.2 IIβ
0.1
TnsC
70
65
55
60
50
45
35
40
25
3
IIβ
TnsB
Fig. 3. RNA-independent integration events occur at preferred sequence motifs. (A) Schematic for single-molecule DNA curtains assay to visualize TnsC binding. l-phage DNA substrates are double-tethered between chrome pedestals and visualized used total internal reflection fluorescence microscopy. (B) mNGlabeled TnsC preferentially binds AT-rich sequences on the l-DNA substrate near the 3′ (pedestal) end (movie S1). (C) Correlation between AT content and mNGTnsC fluorescence intensity visualized along the length of l-DNA. The Pearson linear correlation coefficient is shown (two-tailed P < 0.0001). Data are shown as mean ± SD for N = 66 molecules. (D) Binding kinetics for mNG-TnsC at AT-rich and AT-poor regions of the l-DNA substrate. Apparent kobs at AT-rich sites ≈0.37 min−1, 95% confidence interval (CI) = 0.35 to 0.39, and at AT-poor sites ≈0.28 min−1, 95% CI = 0.27 to 0.30. Data are shown as mean ± SD for N = 87 molecules (thick line, shaded region). Binding kinetics for AT- and GC-rich sites when compared gave a P value of 0.017 upon bootstrapping. (E) Cumulative frequency distributions for the AT content within a 100-bp window flanking integration events using ShCAST with WT TnsC and sgRNA-1 (N = 5505 unique integration events), compared with random sampling of the E. coli genome (N = 50,000 counts). The distributions were significantly different on the basis of results of a Mann-Whitney U test (P = 1.48 × 10−135). (F) Cumulative frequency distribution
AcCAST integrate ~10 bp closer to the target site than ShCAST (11, 14), suggesting that the transpososomes from this subfamily of CAST systems, for both RNA-dependent and RNAindependent transposition pathways, may comprise a shorter TnsC filament spanning only one turn of DNA. George et al., Science 382, eadj8543 (2023)
30
15
K290 20
5
10
0
-5
-10
-15
-20
-25
-35
-30
-40
-45
-55
-50
-60
-65
-70
IIβ 0.0 5
Tn
L
On-target
70 bp L
40 60 AT content (%)
30
TnsB
70 bp
0.5
10 20 Time (min)
I Genomic coordinate
0.0 20
80
G
Unique integration events
App. kobs AT-rich region ≈ 0.37 min-1 AT-poor region ≈ 0.28 min-1
-45
AT distribution in E. coli genome
Cumulative frequency
Cumulative frequency
Unique integration events
AT-rich AT-poor
0.4
0
Reads
E
0.8
0.0
40
λ-DNA coordinate (kb)
AT-poor AT-rich 10 µm
1.2
-20
B
Barrier (B)
Normalized intensity
A
comparison as in (E) but with a K103A TnsC mutant (N = 1932 unique integration events), which revealed a loss of AT bias (P = 0.1349). (G) Meta-analysis of untargeted transposition specificity was performed by extracting sequences from a 140-bp window flanking the integration site and generating a consensus logo. (H) WebLogo from a meta-analysis of untargeted genomic transposition (N = 5855 unique integration events) with a modified pHelper lacking Cas12k and sgRNA. The site of integration is noted with a maroon triangle. An AT-rich sequence spanning ~25 bp likely reflects the footprint of two turns of a TnsC filament (black), whereas motifs within/near the TSD represent TnsB-specific sequence motifs (green). Specific TnsB residues and domains contacting the indicated nucleotides are shown. The magnified inset highlights periodicity in the sequence bound by TnsC. (I) Schematic showing the relative spacing of sequence features bound by Cas12k, TnsC, and TnsB in both on-target (RNA-dependent) and untargeted (RNA-independent) DNA transposition. In both cases, the TnsC footprint covers ~25 bp of DNA and directs polarized, unidirectional integration downstream in a L-R orientation. (J) Magnified view of the ShCAST transpososome structure highlighting sequence-specific contacts between TnsB and the target DNA observed in (H). The Protein Data Bank identification number is 8EA3 (23).
Altogether, these observations demonstrate how subtle sequence motifs at RNA-independent integration sites can be gleaned from metaanalyses of genome-wide integration data. They furthermore reveal that RNA-independent transposition is not random, but rather, that the BCQ transposition pathway pref-
17 November 2023
erentially selects certain genomic regions over others. Preferred sequence motifs lead to semi-targeted, RNA-independent transposition
To test the importance of TnsBC-specific sequence motifs more directly for CRISPR-independent 6 of 11
RES EARCH | R E S E A R C H A R T I C L E
integration, we designed biochemical transposition assays using a series of isogenic pTarget substrates that differed only in the sequence content of a select region that was poorly targeted in previous experiments (Fig. 4A, substrate pT-1). We hypothesized that we could generate targeted, RNA-independent insertions within this region if an optimal sequence were designed to include both the poly-A stretch and flanking TnsB consensus motifs observed in the sequence logo described above (Fig. 3H, substrate pT-2). As further controls, we substituted the poly-A stretch with either poly-AT or poly-GC, mutagenized the TnsB consensus, or replaced both motifs (Fig. 4A, substrates pT-3 through pT-6). We then tested each substrate in biochemical transposition assays and plotted the normalized integration frequency within this window of interest. The resulting data demonstrate that RNAindependent integration events occur in predictable ways depending on the sequence features uncovered through our analyses (Fig. 4B). Substrate pT-2 exhibited a predominant integration product precisely at the engineered site and in the expected T-LR orientation, whereas this integration product was entirely absent when the poly-A was replaced with poly-GC, strengthening our conclusion that favorable TnsC filamentation is important for RNA-independent integration (Fig. 4B, substrates pT-4 and pT-6). When we retained the poly-A stretch but mutated the consensus motif favored by TnsB, integration products were more heterogeneously positioned (Fig. 4B, substrate pT-5), suggesting
A
that preferred TnsB-DNA interactions play an important role in dictating the exact insertion site, as similarly concluded by our recent study on the type I-F VchCAST system (39). When we replaced the poly-A sequence with poly-AT (thus increasing the A content on the opposite strand), the intended integration event was diminished in frequency and accompanied by an increase in upstream integration events on the opposite strand (Fig. 4B, substrate pT-3), demonstrating that nucleotide composition can modulate the preferred directionality of TnsC filament formation and thus integration. These experiments reveal that TnsC prefers to filament unidirectionally on A-rich DNA stretches, leading to downstream integration in the T-LR orientation. The efficiency and exact site of integration is thus a combination of TnsC filament formation propensity and local TnsB sequence preferences. TnsC availability controls the specificity of cellular ShCAST transposition activity
Beyond highlighting the role of TnsC in biasing untargeted integration events to occur at select hotspot regions of the genome, our results more generally implicate TnsC filament formation as a major driver of RNA-independent transposition activity. Because our in vitro results suggest that TnsB differentially selects TnsC filaments at on-target versus untargeted sites, we hypothesized that this difference could be exploited to increase the overall on-target integration accuracy. To test this hypothesis, we designed perturbations intended to repress TnsC
B pTarget
Intended integration site
[0-120]
Modified region
rev
Modified region
filament formation at nonÐCas12k-bound target sites either by fusing TnsC directly to CRISPR effector proteins, or by lowering overall TnsC expression levels. When we fused Cas12k and TnsC, we observed an increase in on-target accuracy (fig. S8A), as was recently reported by Tou et al. (20). We initially hypothesized that this effect might result from local seeding of TnsC filaments upon Cas12k target binding, but coexpression of unfused Cas12k had no adverse effect on specificity, suggesting instead that TnsC filamentation may be partially impaired with an N-terminal adduct. We also replaced Cas12k with dCas9 and generated dCas9-TnsC fusions, hoping to similarly seed TnsC filaments at target sites bound by dCas9-sgRNA complexes. However, we noted no observable on-target integration and severely diminished untargeted integration events (fig. S8B), suggesting that these designs were nonfunctional. These experiments suggested that fusion strategies may be poorly suited to increase the probability of TnsC filament formation at RNA-dependent target sites without extensive further engineering and mutagenesis. Next, we pursued an alternative strategy, motivated by our biochemical observation that increasing TnsC concentration tilted the balance between RNA-dependent (on-target) and RNAindependent (untargeted) transposition toward the latter pathway (Fig. 2, C and D). To determine whether the same feature was applicable in cellular experiments, we relocated tnsC from the original high-copy pHelper plasmid to a
pT-1
fwd pT-1 Original
poly-A
Consensus motif
poly-AT
Consensus motif
poly-GC
Consensus motif
poly-A
Mutated motif
poly-GC
Mutated motif
pT-2 pT-3 pT-4
pT-5
pT-2
Normalized integration reads
Original
pT-3
pT-4
pT-5
pT-6
pT-6
650
Fig. 4. Artificial induction of semi-targeted RNA-independent transposition at preferred motifs. (A) A region on pTarget exhibiting low integration activity (original, blue) was substituted with rationally engineered sequences (colored lines) based on TnsC- and TnsB-binding preferences, generating the indicated pTarget variants (pT-1 to pT6). (B) After performing biochemical transposition assays with the indicated pTarget substrates, integration reads George et al., Science 382, eadj8543 (2023)
17 November 2023
700 750 pTarget coordinate (bp)
850
900
were normalized and mapped to either the forward strand (fwd, red) or reverse strand (rev, black). The intended untargeted integration site based on optimized poly-A and TnsB consensus motifs is marked with a maroon triangle and dotted line; the representative region at right (850 to 900 bp) is shown to highlight consistency in integration events observed elsewhere on pTarget. 7 of 11
RES EARCH | R E S E A R C H A R T I C L E
A
C and D). Cells expressing TnsC under control of a T7 promoter also showed a significant enrichment for insertion events across the T7 RNAP gene, suggesting that these clones were likely enriched within the population as a way of escaping TnsC-induced toxicity (fig. S8, C and E). To determine whether this increased specificity effect was generalizable, we tested a range of guides previously shown to exhibit low ontarget accuracy when tested with pHelper and pDonor. In all cases, we observed a substantial increase in the relative frequency of on-target
C
B sgRNA P* tnsB
tnsC
cas12k
tniQ
On-target Untargeted
pHelper
pTnsC
+ 100
pDonor
T7
La
c
Promoter (P*)
TnsC (Anti-Flag)
0
GAPDH
T7
0 0.0
Original
0 0.0
0.5
1.0
98.1%
1.5
1.0
1.5
2.0
2.5
3.0
3.5
2.0 2.5 3.0 E. coli genomic coordinate (Mb)
4.0
3.5
4.5 4.0
4.5
1.0 Low TnsC ShCAST (Lac)
0 0.0 0.5
1.0
0.5 1.5
short poly A/T
1.0
1.5
2.0
2.5
3.0
4.0
3.5
2.0 2.5 3.0 E.coli genomic coordinate (Mb)
3.5
4.5 4.0
4.5
On-target (RNA-dependent pathway)
poly A/T
Target
sgRNA
S15 TniQ
TnsC
Cas12k
TniQ
Filament dis-/assembly
Total reads (%)
0.5
Untargeted (RNA-independent, BCQ pathway)
Low TnsC
Low TnsC
Low TnsC
Original
Low TnsC
Original
Low TnsC
Original
100
0 0.0
E
Untargeted
On-target
Original
D
Lac
Original ShCAST
39.9%
100
50
integration events with the modified CAST construct (Fig. 5D). This effect did not come at the expense of efficiency, as qPCR measurements revealed that on-target integration occurred with an equal or higher efficiency under low-TnsC conditions compared with our original pHelper design (fig. S8F). It is possible that at low TnsC conditions, the decreased availability of TnsC at untargeted sites presumably titrates fewer TnsB-DNA complexes away from on-target sites and might be the cause of the increased on-target efficiency.
1.0
Integration reads (%)
Total reads (%)
L CmR R
100 Integration reads (%)
separate, medium-copy plasmid, where it was controlled by its own promoter (Fig. 5A). When we tested genomic integration activity under various promoter strengths, we observed considerable differences in on-target specificity (Fig. 5, A and B). Consistent with our in vitro results, low TnsC expression from a lac promoter resulted in 98% of integration events occurring on-target, whereas high TnsC expression with a T7 promoter resulted in considerably lower accuracy (57%), akin to the original pHelper vector (Fig. 5, A to C, and fig. S8,
Capping
TnsB-mediated disassembly
Donor
Donor
TnsB
50
TnsC
TnsB consensus
TnsB
Transposition TSD L
Tn
R
L
Tn
R
0 sgRNA-1
sgRNA-2
sgRNA-3
sgRNA-4
sgRNA-5
Fig. 5. The fidelity of RNA-guided DNA integration is controlled by TnsC concentration. (A) Schematic of alternative ShCAST expression strategy in which TnsC was encoded on a separate plasmid (pTnsC) driven by a Lac or T7 promoter. Distinct cellular expression levels were confirmed by Western blot against a 3xFLAG epitope tag fused to TnsC (bottom). (B) Fraction of total genome-mapping integration reads detected at on-target and untargeted sites upon TnsC expression with a Lac or T7 promoter. (C) Genome-wide view of E. coli genome-mapping reads for the original WT ShCAST system compared with a modified ShCAST system with low TnsC expression. The magnified view visualizes reads comprising ≤1% of genome-mapping reads. The target site is marked with a green triangle. (D) Fraction of total genome-mapping integration reads detected at on-target and untargeted sites, with the original ShCAST system or modified ShCAST system with low TnsC expression. Data for five George et al., Science 382, eadj8543 (2023)
17 November 2023
Integration efficiency
sgRNAs are shown. For (B) and (D), the mean is shown from N = 2 independent biological replicates. (E) Model for target-site selection and transpososome assembly during on-target, RNA-dependent transposition (right) or untargeted, RNA-independent transposition (left) by type V-K CAST systems. Within the untargeted pathway, TnsC preferentially forms filaments at AT-rich regions and is capped by TniQ, leading to the downstream site being selected by TnsB for integration. Cas12k-bound targets may better nucleate TnsC filament formation, and we hypothesize that TnsC filaments loaded at Cas12k-bound targets serve as better substrates for DNA integration, compared with untargeted sites. All structures of TnsC filaments representing untargeted sites (22Ð24, 26), including the BCQ transpososome, reveal K103 residues of the TnsC monomers forming the filament proximal to TnsB, contacting DNA with opposite strand polarity compared with on-target structures (fig. S3B) (22, 23). This could be decisive for the distinct efficiencies observed at these sites. 8 of 11
RES EARCH | R E S E A R C H A R T I C L E
Collectively, these results highlight the importance of relative expression levels for distinct components when delivering CAST machineries into target cells of interest and further confirm the key role of TnsC in driving RNAindependent transposition events. Discussion
Recent structures have shed light on the assembly of transpososome components for RNAguided integration by CAST systems (22, 23). However, a major proportion of integration events for type V-K CASTs occurs at untargeted sites across the genome, for which there was no known mechanistic basis. Combining structural and functional evidence, we establish here that type V-K CASTs maintain a distinct RNA-independent pathway facilitated by TnsB, TnsC, and TniQ (Fig. 5E). Our experiments revealed that the ability of TnsC to promiscuously form filaments on AT-rich DNA is a major driver of untargeted insertions. The role of TnsB in transposition, particularly at untargeted sites, is somewhat paradoxical, given its ability to disassemble TnsC and simultaneously facilitate integration. We speculate that stochastic TniQ binding might stabilize a specific configuration of TnsC filaments at untargeted sites, making them resistant to TnsB-mediated dissociation and instead promoting strand transfer. TnsC disassembly may therefore be less efficient in cellular contexts than originally observed in vitro at high protein concentrations (24, 26, 34). Our results highlight the competition between TnsB recruitment at TnsC-bound, RNA-guided target sites versus AT-rich untargeted sites, and indicate that TnsB preferentially reacts with Cas12k-bound on-target sites compared with untargeted sites (Fig. 5E). Future work will be necessary to resolve more precise kinetics of TnsC filamentation/disassembly in the presence of TniQ and TnsB and differential TnsB transposition kinetics as a function of TnsC assembly state. The structure of the BCQ strand-transfer complex reveals two turns of a TnsC filament, an overall architecture reminiscent of the Cas12kcontaining on-target transpososome (23), with no major structural differences associated with TnsC in either of these assemblies. However, TnsC residues K103 and T121 proximal to TnsB in the BCQ transpososome contact the DNA in 3′ to 5′ strand polarity, following the direction of TniQ- to TnsB-binding face of TnsC (Fig. 1F). The same strand is contacted in the case of random TnsC-DNA filaments (24, 26) and the nonproductive Cas12k transpososome (22), whereas in the productive on-target Cas12k transpososome, TnsC monomers proximal to TnsB contact the opposite strand (5′ to 3′). This interaction is thought to be a consequence of TnsC filament nucleation by Cas12k-TniQ and stabilization by additional DNA interacGeorge et al., Science 382, eadj8543 (2023)
tions (R182 and K119) (22, 23). TnsC variants with mutations to the DNA-contacting residues retain the ability to filament on DNA (24, 26) and maintain a substantial proportion of integration at untargeted sites accompanied by a drop in on-target integration (fig. S4C). This suggests that these residues may not be a prerequisite for TnsC-DNA binding. Rather, we propose that these residues may serve as an intrinsic regulatory feature to ensure that random TnsC filaments default to contacting in the 3′ to 5′ strand polarity. This interaction mode could represent an energetically less favored or passive TnsC configuration for TnsBmediated integration, thereby permitting only a subset of sites scanned by TnsC in the genome to be licensed for untargeted transposition. It is well known that poly-A tracts in the genome represent regions of altered DNA curvature (41), and our single-molecule experiments reveal that TnsC filamentation exhibits inherent affinity for AT-rich locations. Further meta-analyses of integration data revealed a preference for AT-rich sequences across a ~25-bp window spanning about two turns of a TnsC filament upstream of features recognized by TnsB. We suggest that AT-rich genomic regions with altered DNA curvature may resemble the bending of DNA observed between unproductive and productive Cas12k transpososomes (22), leading to preferential TnsC recruitment and a more favorable DNA-binding mode that promotes TnsB-based DNA integration. To our surprise, the TnsC K103A mutation led to a complete loss of AT preference in the integration profile, suggesting that this mutant may achieve an energetically more favorable filamentation state regardless of nucleotide composition. The loss of on-target integration for K103A may result from mutant TnsC filaments titrating TnsB-donor DNA complexes to untargeted sites in the genome more effectively than WT TnsC filaments. The type V-K BCQ pathway, although not exactly similar, resembles the TnsE-mediated pathway in Tn7-like transposable elements, in which structural features associated with DNA replication are recognized to primarily drive widespread mobilization into plasmids (7, 8). A gain-of-function TnsC mutant (A225V) was also identified for E. coli Tn7, and it was capable of transposition in the absence of either of the two targeting factors, TnsD and TnsE (42), and facilitated integration at AT-rich sequences (43). It is possible that in a native setting, type V-K CASTs exhibit low-frequency insertion at AT-rich sites that might be triggered by certain stimuli specific to cyanobacteria. Such a model would imply a transient selfish behavior by CASTs, possibly when the availability of plasmid-targeting guide RNAs are limiting for its proliferation or in situations when mobilization to a new AT-rich site is beneficial for propagation of the element. Although there is
17 November 2023
currently no direct evidence to support this hypothesis, recent experiments with a native type V-K CAST system in cyanobacteria indicate that expression of Cas12k and TnsB are regulated by a CvkR transcriptional repressor (30), a feature that could be important in modulating the choice between targeted and untargeted transposition pathways. Considering the highly conserved operonic nature of TnsB, TnsC, and TniQ in Tn7-like elements, Tn5053, and type V-K CASTs, tight regulation in the stoichiometry of these proteins could be important for accessing an RNA-independent untargeted pathway (13). Future studies will be necessary to investigate this hypothesis further by deep sequencing bacteria with native type V-K elements to ensure that these rare events are not missed. Alternatively, the BCQ pathway may be an evolutionary relic of a primitive selfish pathway before these transposons acquired CRISPR-Cas–based targeting modules. From a technology perspective, type V-K CASTs are among the most compact type of CRISPR-associated transposases, in terms of coding size, and thus offer a major potential opportunity relative to type I CAST systems. However, two key properties that limit their use for genome engineering applications are low on-target specificity and the generation of cointegrate transposition products due to lack of TnsA (14, 15, 18, 44). Recent efforts substantially decreased cointegrate formation by fusing an endonuclease to TnsB and improved specificity using chimeric fusion proteins or supplementing additional components such as pir (20). However, these strategies also compromise on-target integration efficiency and do not address the root cause of promiscuity. Our results provide a deeper molecular understanding of how type V-K CAST components undergo both RNA-guided and RNA-independent transposition. We identified TnsC filamentation on AT-rich DNA sequences as being the primary driver of untargeted integration and showed that under limiting TnsC concentrations, RNA-guided transposition becomes the primary pathway of choice biochemically and in cells. This rescue in specificity was generalizable for all the guides that we tested and resulted in equal or higher ontarget integration efficiency compared with the original ShCAST pHelper. Our combined use of biochemical, structural, and single-molecule experiments reveals the mechanistic intricacies associated with target site selection by type V-K CAST and offers new opportunities for targeted DNA integration applications. Methods summary
All plasmid constructs used for this study were cloned using a combination of Gibson assembly, inverse (around-the-horn) PCR, restriction digestion, and ligation. Transposition assays 9 of 11
RES EARCH | R E S E A R C H A R T I C L E
in E. coli were performed in BL21 (DE3) cells based on a previously described method (14). For biochemical reconstitution of CAST transposition, proteins were expressed as N-terminal His6-SUMO-TEV fusions and purified according to previous protocols (11, 24). Cryo-EM structure determination of the BCQ transpososome was performed by incubating purified TnsB, TnsC, and TniQ proteins with a synthetic DNA substrate as previously described (26), preparing and imaging grids using a 200-kV Talos Arctica (Thermo Fisher), and performing downstream image analysis. TnsCbinding activity was tested using a fluorescence polarization assay with a 5′ fluorescein-tagged DNA substrate, and ATP hydrolysis was measured using a Malachite green phosphate assay. Biochemical plasmid-to-plasmid transposition reactions were incubated for 2 hours at 37°C and then quenched by flash-freezing in liquid nitrogen. On-target transposition efficiency for biochemical and E. coli transposition assays was measured using qPCR, and specificity measurements were made using tagmentation-based transposon insertion sequencing (TagTn-seq), with library preparation performed using Nextera XT DNA Library Preparation Kit (Illumina). Next-generation sequencing was performed on an Illumina NextSeq platform with a NextSeq high-output kit. Custom Python scripts were used for mapping transposon end–containing reads to the E. coli genome or plasmid substrates used in biochemical transposition. Genetic neighborhood and AT-enrichment analyses at transposon insertion sites were performed using various custom Python scripts available online. Motifs for untargeted transposition were determined by extracting and comparing a window of sequences adjacent to each integration site, and sequence consensus logos showing per residue conservation were plotted. Single-molecule double-tethered dsDNA curtain experiments were performed as previously described (35) using a fluorescent mNeonGreen-TnsC and were analyzed to examine TnsC binding to AT- and GC-rich regions. Western blots were performed with FLAG epitope-tagged TnsC variants expressed with a T7 or lac promoter and anti-FLAG antibody. A detailed materials and methods section for this study is provided in the supplementary materials. RE FE RENCES AND N OT ES
1. N. L. Craig, Target site selection in transposition. Annu. Rev. Biochem. 66, 437–474 (1997). doi: 10.1146/annurev. biochem.66.1.437; pmid: 9242914 2. P. Siguier, E. Gourbeyre, M. Chandler, Bacterial insertion sequences: Their genomic impact and diversity. FEMS Microbiol. Rev. 38, 865–891 (2014). doi: 10.1111/1574-6976.12067; pmid: 24499397 3. P. Siguier, E. Gourbeyre, A. Varani, B. Ton-Hoang, M. Chandler, Everyman’s guide to bacterial insertion sequences. Microbiol. Spectr. 3, A3–A0030, 2014 (2015). doi: 10.1128/microbiolspec. MDNA3-0030-2014; pmid: 26104715 4. E. Arias-Palomo, J. M. Berger, An atypical AAA+ ATPase assembly controls efficient transposition through DNA remodeling and transposase recruitment. Cell 162, 860–871 (2015). doi: 10.1016/j.cell.2015.07.037; pmid: 26276634
George et al., Science 382, eadj8543 (2023)
5. N. Mizuno et al., MuB is an AAA+ ATPase that forms helical filaments to control target selection for DNA transposition. Proc. Natl. Acad. Sci. U.S.A. 110, E2441–E2450 (2013). doi: 10.1073/pnas.1309499110; pmid: 23776210 6. K. Y. Choi, J. M. Spencer, N. L. Craig, The Tn7 transposition regulator TnsC interacts with the transposase subunit TnsB and target selector TnsD. Proc. Natl. Acad. Sci. U.S.A. 111, E2858–E2865 (2014). doi: 10.1073/pnas.1409869111; pmid: 24982178 7. J. E. Peters, N. L. Craig, Tn7 recognizes transposition target structures associated with DNA replication using the DNA-binding protein TnsE. Genes Dev. 15, 737–747 (2001). doi: 10.1101/gad.870201; pmid: 11274058 8. J. E. Peters, N. L. Craig, Tn7: smarter than we thought. Nat. Rev. Mol. Cell Biol. 2, 806–814 (2001). doi: 10.1038/35099006; pmid: 11715047 9. J. A. Finn, A. R. Parks, J. E. Peters, Transposon Tn7 directs transposition into the genome of filamentous bacteriophage M13 using the element-encoded TnsE protein. J. Bacteriol. 189, 9122–9125 (2007). doi: 10.1128/JB.01451-07; pmid: 17921297 10. J. E. Peters, K. S. Makarova, S. Shmakov, E. V. Koonin, Recruitment of CRISPR-Cas systems by Tn7-like transposons. Proc. Natl. Acad. Sci. U.S.A. 114, E7358–E7366 (2017). doi: 10.1073/pnas.1709035114; pmid: 28811374 11. J. Strecker et al., RNA-guided DNA insertion with CRISPRassociated transposases. Science 365, 48–53 (2019). doi: 10.1126/science.aax9181; pmid: 31171706 12. S. E. Klompe, P. L. H. Vo, T. S. Halpin-Healy, S. H. Sternberg, Transposon-encoded CRISPR-Cas systems direct RNA-guided DNA integration. Nature 571, 219–225 (2019). doi: 10.1038/ s41586-019-1323-z; pmid: 31189177 13. G. Faure et al., Modularity and diversity of target selectors in Tn7 transposons. Mol. Cell 83, 2122–2136.e10 (2023). doi: 10.1016/j.molcel.2023.05.013; pmid: 37267947 14. P. L. H. Vo et al., CRISPR RNA-guided integrases for highefficiency, multiplexed bacterial genome engineering. Nat. Biotechnol. 39, 480–489 (2021). doi: 10.1038/s41587-02000745-y; pmid: 33230293 15. B. E. Rubin et al., Species- and site-specific genome editing in complex bacterial communities. Nat. Microbiol. 7, 34–47 (2022). doi: 10.1038/s41564-021-01014-7; pmid: 34873292 16. S.-C. Hsieh, J. E. Peters, Discovery and characterization of novel type I-D CRISPR-guided transposons identified among diverse Tn7-like elements in cyanobacteria. Nucleic Acids Res. 51, 765–782 (2023). doi: 10.1093/nar/gkac1216; pmid: 36537206 17. M. Saito et al., Dual modes of CRISPR-associated transposon homing. Cell 184, 2441–2453.e18 (2021). doi: 10.1016/ j.cell.2021.03.006; pmid: 33770501 18. P. L. H. Vo, C. Acree, M. L. Smith, S. H. Sternberg, Unbiased profiling of CRISPR RNA-guided transposition products by long-read sequencing. Mob. DNA 12, 13 (2021). doi: 10.1186/ s13100-021-00242-2; pmid: 34103093 19. P. A. Rice, N. L. Craig, F. Dyda, Comment on “RNA-guided DNA insertion with CRISPR-associated transposases”. Science 368, eabb2022 (2020). doi: 10.1126/science.abb2022; pmid: 32499410 20. C. J. Tou, B. Orr, B. P. Kleinstiver, Precise cut-and-paste DNA insertion using engineered type V-K CRISPR-associated transposases. Nat. Biotechnol. 41, 968–979 (2023). doi: 10.1038/s41587-022-01574-x; pmid: 36593413 21. G. D. Lampe et al., Targeted DNA integration in human cells without double-strand breaks using CRISPR-associated transposases. Nat. Biotechnol., 1–12 (2023). doi: 10.1038/ s41587-023-01748-1; pmid: 36991112 22. M. Schmitz, I. Querques, S. Oberli, C. Chanez, M. Jinek, Structural basis for the assembly of the type V CRISPRassociated transposon complex. Cell 185, 4999–5010.e17 (2022). doi: 10.1016/j.cell.2022.11.009; pmid: 36435179 23. J.-U. Park et al., Structures of the holo CRISPR RNA-guided transposon integration complex. Nature 613, 775–782 (2023). doi: 10.1038/s41586-022-05573-5; pmid: 36442503 24. I. Querques, M. Schmitz, S. Oberli, C. Chanez, M. Jinek, Target site selection and remodelling by type V CRISPR-transposon systems. Nature 599, 497–502 (2021). doi: 10.1038/s41586021-04030-z; pmid: 34759315 25. R. Xiao et al., Structural basis of target DNA recognition by CRISPR-Cas12k for RNA-guided DNA transposition. Mol. Cell 81, 4457–4466.e5 (2021). doi: 10.1016/j.molcel.2021.07.043; pmid: 34450043 26. J.-U. Park et al., Structural basis for target site selection in RNA-guided DNA transposition systems. Science 373, 768–774 (2021). doi: 10.1126/science.abi8976; pmid: 34385391
17 November 2023
27. R. M. Harshey, Transposable phage mu. Microbiol. Spectr. 2, 2.5.31 (2014). doi: 10.1128/microbiolspec.MDNA3-0007-2014; pmid: 26104374 28. G. Ya. Kholodii et al., Four genes, two ends, and a res region are involved in transposition of Tn5053: A paradigm for a novel family of transposons carrying either a mer operon or an integron. Mol. Microbiol. 17, 1189–1200 (1995). doi: 10.1111/ j.1365-2958.1995.mmi_17061189.x; pmid: 8594337 29. S. Hou et al., CRISPR-Cas systems in multicellular cyanobacteria. RNA Biol. 16, 518–529 (2019). doi: 10.1080/ 15476286.2018.1493330; pmid: 29995583 30. M. Ziemann et al., CvkR is a MerR-type transcriptional repressor of class 2 type V-K CRISPR-associated transposase systems. Nat. Commun. 14, 924 (2023). doi: 10.1038/s41467023-36542-9; pmid: 36801863 31. Y. Shen et al., Structural basis for DNA targeting by the Tn7 transposon. Nat. Struct. Mol. Biol. 29, 143–151 (2022). doi: 10.1038/s41594-022-00724-8; pmid: 35173349 32. F. T. Hoffmann et al., Selective TnsC recruitment enhances the fidelity of RNA-guided transposition. Nature 609, 384–393 (2022). doi: 10.1038/s41586-022-05059-4; pmid: 36002573 33. P. I. Hanson, S. W. Whiteheart, AAA+ proteins: Have engine, will work. Nat. Rev. Mol. Cell Biol. 6, 519–529 (2005). doi: 10.1038/ nrm1684; pmid: 16072036 34. J.-U. Park, A. W.-L. Tsai, T. H. Chen, J. E. Peters, E. H. Kellogg, Mechanistic details of CRISPR-associated transposon recruitment and integration revealed by cryo-EM. Proc. Natl. Acad. Sci. U.S.A. 119, e2202590119 (2022). doi: 10.1073/ pnas.2202590119; pmid: 35914146 35. A. Meir, M. Kong, C. Xue, E. C. Greene, DNA llight on complex molecular systems during homologous recombination. J. Vis. Exp. (160): (2020). doi: 10.3791/61320 36. A. Skalka, E. Burgi, A. D. Hershey, Segmental distribution of nucleotides in the DNA of bacteriophage lambda. J. Mol. Biol. 34, 1–16 (1968). doi: 10.1016/0022-2836(68)90230-1; pmid: 4999721 37. E. C. Greene, K. Mizuuchi, Direct observation of single MuB polymers: Evidence for a DNA-dependent conformational change for generating an active target complex. Mol. Cell 9, 1079–1089 (2002). doi: 10.1016/S1097-2765(02)00514-2; pmid: 12049743 38. J. Ge, Z. Lou, H. Cui, L. Shang, R. M. Harshey, Analysis of phage Mu DNA transposition by whole-genome Escherichia coli tiling arrays reveals a complex relationship to distribution of target selection protein B, transcription and chromosome architectural elements. J. Biosci. 36, 587–601 (2011). doi: 10.1007/s12038-011-9108-z; pmid: 21857106 39. M. W. G. Walker, S. E. Klompe, D. J. Zhang, S. H. Sternberg, Novel molecular requirements for CRISPR RNA-guided transposition. Nucleic Acids Res. 51, 4519–4535 (2023). doi: 10.1093/nar/gkad270; pmid: 37078593 40. F. Tenjo-Castaño et al., Structure of the TnsB transposase-DNA complex of type V-K CRISPR-associated transposon. Nat. Commun. 13, 5792 (2022). doi: 10.1038/s41467-022-33504-5; pmid: 36184667 41. T. E. Haran, U. Mohanty, The unique structure of A-tracts and intrinsic DNA bending. Q. Rev. Biophys. 42, 41–81 (2009). doi: 10.1017/S0033583509004752; pmid: 19508739 42. A. E. Stellwagen, N. L. Craig, Gain-of-function mutations in TnsC, an ATP-dependent transposition protein that activates the bacterial transposon Tn7. Genetics 145, 573–585 (1997). doi: 10.1093/genetics/145.3.573; pmid: 9055068 43. M. C. Biery, F. J. Stewart, A. E. Stellwagen, E. A. Raleigh, N. L. Craig, A simple in vitro Tn7-based transposition system with low target site selectivity for genome and gene analysis. Nucleic Acids Res. 28, 1067–1077 (2000). doi: 10.1093/nar/ 28.5.1067; pmid: 10666445 44. W. Chen et al., Targeted genetic screening in bacteria with a Cas12k-guided transposase. Cell Rep. 36, 109635 (2021). doi: 10.1016/j.celrep.2021.109635; pmid: 34469724 45. Custom Python scripts for: J. T. George et al., Mechanism of target site selection by type V-K CRISPR-associated transposases, Zenodo (2023); doi: 10.1101/2023.07.14.548620 AC KNOWLED GME NTS
We thank N. Jaber and S. R. Pesari for laboratory support; P.A. Sims for helpful discussions on deep sequencing; M. Jovanovic for help with mass spectrometry; R. T. King for help with Taqman qPCR; D. R. Gelsinger for help with TagTn-seq; G. D. Lampe, F. T. Hoffmann, and S. Tang for critical feedback on the manuscript; J. E. Peters for useful discussions; L. F. Landweber for qPCR
10 of 11
RES EARCH | R E S E A R C H A R T I C L E
instrument access; the J. P. Sulzberger Columbia Genome Center for NGS support; and the Cornell Center for Materials Research facility, K. Spoth, and M. Silvestry-Ramos for maintenance of the electron microscopes used for this research (NSF MRSEC program, DMR-1719875). Funding: J.T.G. is supported by International Human Frontier Science Program postdoctoral fellowship LT001117/2021-C. E.C.G. is supported by National Institutes of Health (NIH) grant R35GM118026. E.H.K. is supported by NIH grant R01GM144566 and a Pew Biomedical Scholarship. S.H.S. is supported by NIH grants DP2HG011650, R21AI68976, and R01EB031935; a Pew Biomedical Scholarship; a Sloan Research Fellowship; an Irma T. Hirschl Career Scientist Award; and a generous startup package from the Columbia University Irving Medical Center Dean’s Office and the Vagelos Precision Medicine Fund. Author contributions: J.T.G. and S.H.S. conceived of and designed the project. J.T.G. performed most experiments. C.A. assisted in the analyses of high-throughput sequencing data and contributed computational support. J.P. and
George et al., Science 382, eadj8543 (2023)
E.H.K. performed cryo-EM experiments and data analysis. M.K. and E.C.G. assisted with single-molecule biophysics experiments. T.W. contributed bioinformatics and structural analyses. Y.L.P. assisted with protein biochemistry. J.T.G. and S.H.S. discussed the data and wrote the manuscript with input from all authors. Competing interests: Columbia University has filed a patent application related to this work for which J.T.G. and S.H.S. are inventors. S.H.S. is a cofounder and scientific adviser to Dahlia Biosciences, a scientific adviser to CrisprBits and Prime Medicine, and an equity holder in Dahlia Biosciences and CrisprBits. The remaining authors declare no competing interests. Data and materials availability: Cryo-EM reconstructions of the BCQ transpososome are available through the Electron Microscopy Data Bank with accession code EMD-41280. Next-generation sequencing data are available in the National Center for Biotechnology Information (NCBI) Sequence Read Archive with BioProject accession code PRJNA1010381. Custom Python scripts used for computational analysis of next-generation sequencing data
17 November 2023
are available at Zenodo (45). License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/about/science-licensesjournal-article-reuse SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adj8543 Materials and Methods Figs. S1 to S8 Tables S1 to S6 References (46–62) Movies S1 to S2 MDAR Reproducibility Checklist Submitted 19 July 2023; accepted 23 September 2023 10.1126/science.adj8543
11 of 11
Where Science Gets Social.
AAAS.ORG/COMMUNITY
AAAS’ Member Community is a one-stop destination for scientists and STEM enthusiasts alike. It’s “Where Science Gets Social”: a community where facts matter, ideas are big and there’s always a reason to come hang out, share, discuss and explore.
RESEAR CH
RESEARCH ARTICLE SUMMARY
◥
EPIGENETICS
Design principles of 3D epigenetic memory systems Jeremy A. Owen, Dino Osmanović, Leonid Mirny*
INTRODUCTION: During development, our cells establish different identities—nerve, muscle, blood, and so on—which can remain stable even as the cells grow and divide. This stability is a form of epigenetic memory. Evidence suggests that this memory is partly held within chromatin itself, probably in patterns of chemical modifications of DNA and histones known as “epigenetic marks.” But individual marks are highly dynamic, being placed and removed by many mechanisms, and are partly lost when DNA is replicated. How can patterns of dynamic marks be a seat of stable memory? RATIONALE: In the transcriptionally silent heterochromatin, the loss of histone marks is counterbalanced by “reader-writer” enzymes that effectively spread marks between neighboring nucleosomes. However, in simple models, this mechanism cannot maintain mark patterns; if mark spreading is strong enough to restore a partially erased pattern, marks tend to spread uncontrollably. An account of epigenetic memory must invoke ingredients beyond simple mark spreading. A tantalizing possibility is that this missing element is the three-dimensional (3D) folding of the genome. Recent experiments suggest that marks might spread not just along the chromatin polymer but also in three dimensions between nucleosomes near each other in space. Additionally, regions carrying heterochromatic marks are known to attract each other, making heterochromatin segregate from other regions
1 Compartmentalized nucleus
and forming a denser nuclear compartment. Mechanisms for epigenetic memory could rely on this interplay between 3D compartmentalization and mark dynamics. RESULTS: We developed a simple theoretical model of the dynamics of chromatin and its marks through the cell cycle. Over a single cell generation, we found that if marks spread in 3D, they sharply localize to dense chromatin regions. In fact, our model of mark dynamics is precisely analogous to a susceptible-infectedsusceptible (SIS) epidemic model on a network; mark localization to the dense compartment is akin to an epidemic becoming endemic in densely populated areas while vanishing elsewhere. However, over successive cell generations, as chromatin refolds according to the dynamic marks, we still see an all-or-none tendency for marks to spread everywhere or to be lost globally, destroying any epigenetic memory. We found that one more ingredient is needed: the limitation of reader-writer enzymes relative to their histone substrates. This very plausible but often neglected element completely changes model behavior, yielding a memory of mark patterns stable for hundreds of cell generations. Our findings are insensitive to variations in model assumptions—for example, to the precise way loss of marks occurs—as long as three ingredients are present. First, there must be a considerable density difference between chromatin compartments. Second, a reader-writer enzyme must be able to spread marks in 3D.
2 Marks spread in 3D
And third, critically, those enzymes must be limited in abundance. We propose that the presence of these elements amounts to a basic design principle for epigenetic memory systems that exploit 3D genome structure for their function. We find that our model provides a unified account of many observations, ranging from classic phenomena such as position-effect variegation to recent studies of the dynamics of mark recovery after replication. We can also make a number of predictions, particularly about single-cell epigenetic heterogeneity, that emerging techniques may be able to test. CONCLUSION: Our work reveals a mechanism by which a coupling between 3D folding of the genome and mark dynamics could help cells remember their identities. Intuitively, the mechanism relies on the encoding of memory in different forms in different phases of the cell cycle. In interphase, memory is held in the 3D structure of the genome, whereas in mitosis, when the 3D structure is being totally reorganized, memory is held in the sequence of marks. The encoding of a mark pattern could also be thought of as the “learning rule” of an associative memory in a Hopfield network, where connections between marked regions are established by folding them together (“mark together, park together”). In this analogy, the mark dynamics are like the “update rule” that allows recovery of a stored memory. We should keep in mind the possibility that epigenetic systems are capable not just of memory but also of more sophisticated information processing.
▪
The list of author affiliations is available in the full article online. *Corresponding author. Email: [email protected] Cite this article as J. A. Owen et al., Science 382, eadg3053 (2023). DOI: 10.1126/science.adg3053
READ THE FULL ARTICLE AT https://doi.org/10.1126/science.adg3053
3 Limited “reader-writer” enzyme
Epigenetic memory of mark patterns
[E] ≪ [ ]
B
t (generations)
0
Enzyme E
A B
A
200
Design principles of 3D epigenetic memory. (Left) Our model of chromatin and its dynamics through the cell cycle reveals three key elements for stable epigenetic memory (right). SCIENCE science.org
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
785
RES EARCH
RESEARCH ARTICLE
◥
EPIGENETICS
Design principles of 3D epigenetic memory systems Jeremy A. Owen1 , Dino Osmanović2, Leonid Mirny1* Cells remember their identities, in part, by using epigenetic marks—chemical modifications placed along the genome. How can mark patterns remain stable over cell generations despite their constant erosion by replication and other processes? We developed a theoretical model that reveals that threedimensional (3D) genome organization can stabilize epigenetic memory as long as (i) there is a large density difference between chromatin compartments, (ii) modifying “reader-writer” enzymes spread marks in three dimensions, and (iii) the enzymes are limited in abundance relative to their histone substrates. Analogous to an associative memory that encodes memory in neuronal connectivity, mark patterns are encoded in a 3D network of chromosomal contacts. Our model provides a unified account of diverse observations and reveals a key role of 3D genome organization in epigenetic memory.
R
emembering gene expression states— that is, which genes are “on” or “off”—is a remarkable capability of living cells. It is well established that this “epigenetic” memory can be stably encoded in the abundances of freely diffusing transcription factors (TFs) that regulate each other’s synthesis (1–3). But in eukaryotes, such as ourselves, in addition to TF-based memory, there is evidence that memory can be held locally to the genes, in the chromatin (4–7). It has been suggested that a seat of this chromatin-based epigenetic memory could be the chemical modifications (marks) of the DNA-bound histones, which vary across the genome in patterns correlated with gene expression. However, chromatin and its marks are subject to large disruptions through the cell cycle, and it is not clear what is required to make stable memories out of mark patterns. In this study, we identified three qualitative elements that together are sufficient for stable epigenetic memory. Our minimal theoretical model incorporating these elements unites a battery of classic observations ascribed to epigenetic memory of heterochromatin, makes predictions that emerging experimental techniques can test, and suggests a functional role for a hallmark of nuclear organization: its three-dimensional (3D) compartmentalization. Heterochromatin—the transcriptionally silent, denser nuclear compartment—is rich in particular histone marks, especially the lysine trimethylations H3K9me3 and H3K27me3. These marks are made by so-called reader-writer enzymes (8, 9), which can bind marked histones allosterically, stimulating their marking activity on neighboring histones, effectively 1
Department of Physics, Massachusetts Institute of Technology, Cambridge, MA, USA. 2Department of Mechanical and Aeronautical Engineering, UCLA, Los Angeles, CA, USA. *Corresponding author. Email: [email protected] Present address: Department of Chemistry, Princeton University, Princeton, NJ, USA.
Owen et al., Science 382, eadg3053 (2023)
spreading marks between neighbors. Marked histones can be retained locally when the replication fork passes (10, 11), but they are (by necessity) diluted in the process by newly synthesized, unmarked histones. The combination of these two features is highly suggestive of a stable memory system, in which local mark spreading accurately restores mark patterns after their partial erasure at replication. However, simple mathematical models (12, 13) of this mechanism reveal a basic instability: If mark spreading is strong enough to restore a partially erased pattern, marks also spread ectopically to the rest of the chromosome. Recent experiments suggest that reader-writer enzymes may be able to spread histone marks in three dimensions (9, 14, 15), that is, between histones that are nearby in space because of how chromatin is folded, not just in one dimension along the chromatin polymer. Because histone marks also contribute to the spatial compartmentalization of the genome, this raises the tantalizing possibility of a bidirectional coupling between the 3D folding of chromatin and the marks on the chromatin polymer (16–19). Could this help stabilize memory? Recent theoretical work (19–22) has explored some consequences of this putative coupling, but generally these studies have difficulty achieving a self-sustaining memory of mark patterns. An understanding of the qualitative conditions required for chromatinbased epigenetic memory is yet to emerge. Model
In search of design principles for epigenetic memory, we developed and studied a simple biophysical model in which memory is held autonomously in mark patterns. In many prior models, mark patterns are sustained by external reinforcement—for example, by nucleation sites or genomic bookmarks (12, 13, 23) that recruit modifying enzymes—or by a static 3D contact structure (24). But a pattern determined
17 November 2023
by external influences is not itself a seat of memory, and so we excluded such elements from our model. We modeled chromatin as a long polymer of 104 monomers that is confined within a sphere (Fig. 1A). This could represent, in a coarsegrained manner, all the chromatin in the nucleus, or just a chromosomal region of 2 Mb (104 nucleosomes). Monomers in the polymer can be in one of two states, A or B, with B monomers representing marked, heterochromatic regions and A monomers representing unmarked, euchromatic ones. To represent the “stickiness” of heterochromatin (25–28), B monomers experience a short-range attraction (fig. S14) to one another of magnitude a, which leads B monomers to spatially segregate from the A monomers, forming a denser compartment. To model the 3D spreading of marks (Fig. 1B), we supposed that A monomers turn into B monomers at a rate SnB , where nB is the number of neighboring B monomers within a 3D interaction radius, rc (1.5 times the diameter of a monomer), and S is the spreading rate. B monomers turn back into A monomers at a constant rate L , uniformly at all sites, representing in aggregate the loss of marked histones owing to the activity of demodifying enzymes (e.g., demethylases), histone exchange, and replicative dilution. Our core results proved insensitive to precisely how the loss of marks is modeled (fig. S2). To represent the cell cycle (Fig. 1C), we ran our model in two alternating phases. During “interphase,” we assumed that the chromatin is frozen in place while marks are spread and lost, reaching a steady state. By contrast, in “mitosis,” we assumed that marks remain unchanged while the chromatin polymer is compacted into a condensed state. Then, to establish a new interphase state, we allowed the polymer to expand subject to interactions between marked regions, naturally leading to compartmentalization (Materials and methods). We called each round of polymer dynamics followed by mark dynamics one “cell generation.” Our assumptions about the dynamics in each phase reflect experimental observations. In interphase, the gross 3D organization of chromatin is quite stable (29), whereas some marks can turn over completely on a timescale of minutes to hours (30, 31), a time over which chromatin loci may displace by just ~ 0.2 to 0.4 mm (32). By contrast, in mitosis, repressive marks appear to remain stable (33, 34) even while chromatin undergoes dramatic refolding. Several factors may account for this stability of marks, including inhibition of modifying enzymes by mitotic phosphorylation of the H3 tail (35, 36), decreased accessibility of mitotic chromatin, and the short duration of mitosis. Later in the study, we loosened the assumptions we made about the phases (figs. S3 and S4). 1 of 9
RES EARCH | R E S E A R C H A R T I C L E
A
C Establish marked regions
B Initial state A
in the densities grows (Fig. 2C, inset), further broadening this range. The analogy to epidemic spreading shows quantitatively how the density difference between the compartments underlies the sharp localization of marks to the dense compartment, providing robust recovery of the initial mark pattern within one cell generation. Memory is lost over multiple cell generations
B One cell generation
Mark dynamics on fixed polymer
Interphase
B
A
B
B Polymer dynamics (decompaction) while marks are fixed
B
Compact state of polymer
Mitosis
A
Fig. 1. Model of mark and chromatin dynamics. (A) Chromatin in the nucleus modeled as a spherically confined copolymer with monomers of two types, A (pale yellow) and B (blue), representing a varying pattern of histone marks. Monomers of type B, which represent regions bearing heterochromatic marks, self-attract. (B) Marks spread to 3D neighbors at rate S and are lost everywhere uniformly at rate L. (C) The overall dynamics of our model consist of alternating phases of polymer dynamics and mark dynamics, representing the cell cycle.
In our simulations, we set an initial pattern of A and B monomer identities before the first interphase and allowed it to evolve over one or many cell generations. If, at later times, the pattern resembles the initial pattern, and would do so for several possible initial patterns, then the system can be said to exhibit memory. Results Marks localize to dense regions providing stable memory for one cell generation
Over a single cell generation, we found that there is an extremely good memory of mark patterns. The steady state of the mark dynamics reached in the interphase closely resembles the initial mark pattern used to fold the polymer (Fig. 2, A and B). The steady-state pattern can recover after large perturbations, such as a complete randomization of the pattern (every monomer is randomly set to A or B) (Fig. 2A) or wholesale erasure of half of the pattern (Fig. 2B). The reason for the recovery of the pattern is that spreading marks tend to localize to dense regions; the remaining marks spread in 3D and restore the marks in the spatially dense compartment that was formed by the originally marked regions. It is as if the mark pattern has been “memorized” in the 3D configuration of the polymer. An analogy to epidemic spreading can help us understand the localization of marks to dense regions quantitatively. The mark dynamics of Owen et al., Science 382, eadg3053 (2023)
our model are identical to a susceptibleinfected-susceptible (SIS) epidemic model on a network (37). The monomers of our polymer are like individuals whose “social” contact network (Fig. 2D) is defined by the polymer configuration, and the marked state is like the infected state. The infection, like marks, spreads at rate S, and infected individuals recover (lose marks) at rate L. A key parameter for epidemic spreading dynamics is the average number of neighbors, d, of an individual (monomer). Roughly, there is an “epidemic threshold,” 1=d, such that if S=L is below 1=d, the infection will die out. Returning to our model with a dense and diffuse compartment, with average numbers of neighbors dþ and d , respectively, this suggests that when S=L lies in the range 1 S 1 ≤ ≤ dþ L d there should be sharp localization of marks to the dense compartment, with very few marks in the diffuse compartment (a more detailed discussion is provided in the supplementary text). Intuitively, this condition says that the system must be above the epidemic threshold in the dense region and below it in the diffuse region. Consistently, simulations (Fig. 2C) show localization of marks in an even broader range of S=L. As the strength of self-attraction, a, increases, the difference
17 November 2023
However, over multiple generations (Fig. 3), something very different happens. Sweeping through the parameter space of our model (Fig. 3B), what we found is an unstable, all-ornone behavior. When S=L is greater than a critical value, lc ðaÞ (which depends on a), an initially marked region grows uncontrollably until it covers the entire polymer. When S=L is less than lc ðaÞ, marks are instead lost globally. In both cases, memory of the initial state is lost within a few generations. When there is strong self-attraction and S=L is fine-tuned to very near lc ðaÞ, memory lasts longer, but even then, there is a clear tendency to uncontrolled spread or global loss of marks. The same basic instability is apparent in the closely related model of Sandholtz et al. (21), who found that fine-tuning of parameters was required to achieve just five generations of mark-pattern memory. Taken together, 3D spread of marks, even when coupled with 3D genome folding through the self-attraction of marks, is not enough to provide lasting epigenetic memory. Enzyme limitation stabilizes epigenetic memory
However, so far we have neglected a key biological fact, often omitted in biophysical models of mark dynamics. Marks do not spread by themselves; spreading requires the action of a reader-writer enzyme, which in the nucleus is likely to be limited relative to its histone substrates. Estimates of the abundances of the histone methyltransferases PRC2 and SETB1 (38–40), for example, suggest that they are hundreds to thousands of times less abundant than nucleosomes, which number in the tens of millions. To account for the limitation of the reader-writer enzymes, we introduced a Michaelis-Menten-type scheme (41, 42) in which A-B pairs that are within the interaction radius act as the substrate (Fig. 3F). Remarkably, we found that adding enzyme limitation to our model stabilizes the memory of the initial mark pattern (Fig. 3, D and E) for hundreds of cell generations and over a broad range of parameters. The effect of enzyme limitation is to replace the spreading rate, S, by an effective spreading rate, Seff , that depends on the number of A-B pairs, NAB (supplementary text) ( Seff ¼
S SET NAB
if NAB < ET if NAB ≥ ET 2 of 9
RES EARCH | R E S E A R C H A R T I C L E
A
C 8 6
Fixed polymer
One generation
10
4
Randomize marks
2 0
0.
0.6
1.2
1.8
2.4
B
Erase right half of pattern
Fixed polymer
One generation
D
polymer
Fig. 2. Spreading marks sharply localize to dense regions. (A) Mark dynamics, with S=L ¼ 0:5, over a single cell generation, on a fixed polymer folded according to an initial pattern, with a ¼ 2:4 kB T (where kB is the Boltzmann constant and T is the temperature). Time advances from top to bottom. (Left) Inset circles show snapshots of the polymer configuration (2D projection) over time. (B) As in (A), with a different initial pattern and perturbation. (C) (Top) When the polymer folds, marked regions tend to be denser (red) than unmarked ones (green) because of the self-attraction of marks [density is quantified by the number of
where ET is the total amount of enzyme. Intuitively, the enzyme sets a maximum global modification rate: the “Vmax” of the enzyme, which equals SET. In those cases in which previously, marks spread uncontrollably across the whole polymer, now the total number of marks is set by the balance ofVmax and L to be NB ¼ SET =L. This fixing of the number of marks is sufficient to yield a stable memory of the mark pattern—for example, of the position of a marked domain (Fig. 3D and fig. S1). Stability of the mark pattern is also seen when loss occurs purely by replicational dilution (modeled as random loss of half the marks) once every cell cycle period, Tdiv , instead of at a constant loss rate, L (fig. S2). The stable memory is seen across a broad range of parameters as long as self-attraction is strong enough (Fig. 3E), and it works without external reinforcement or fine-tuning, as required by other models.
monomer neighbors (color bar)]. The plot shows the average number of monomer neighbors in each compartment as a function of the strength of B-B self-attraction, a. (Bottom) In turn, when marks evolve according to their dynamics of spreading and loss, they tend to localize in dense regions for a range of S=L values. (D) An analogy to epidemic spreading, in which marked monomers are equivalent to infected people, predicts correctly that this localization will occur at least in the red interval, whose width is set by the number of neighbors in each region.
leading to nuclear compartmentalization and densification of marked regions; (ii) 3D spread of marks; and (iii) limitation of the readerwriter enzyme relative to its substrates. We propose that the presence of these elements together amounts to a basic design principle for epigenetic memory systems that exploit 3D genome structure for their function. Our results suggest that heterochromatin may be dense not necessarily to sterically exclude transcriptional machinery [heterochromatin is likely highly permeable to polymerase-size particles (43)] but rather as a way to maintain the memory of heterochromatin. A rich observable phenomenology follows directly from these elements, providing strong support for our model, as well as many predictions to be tested by emerging experimental modalities.
Design principles and model phenomenology
The number of marks scales linearly with the enzyme concentration
To summarize our findings so far, we have found a memory system that depends on three key ingredients, all characteristic of heterochromatin: (i) strong self-attraction of marked regions,
First and most basically, our model relates the abundance of a mark to the activity (S) and concentration (ET ) of a reader-writer enzyme that makes it; in particular, we found a broad
Owen et al., Science 382, eadg3053 (2023)
17 November 2023
contact network
regime in which the number of marks is linear in both of these quantities: NB ¼ SET =L. This prediction is at least consistent with the measured effects of EZH2 inhibition (44) and activating mutations (45, 46) on H3K27me3 levels (supplementary text), although a definitive test of linearity will require careful quantitation of both sides of the equation. Perhaps more unexpectedly, our model reveals that sometimes changing the concentration of an enzyme is different than uniformly changing its activity. This could shed light on mechanistic puzzles, such as the question of how the oncogenic mutant H3K27M histone reduces H3K27me3: Does it sequester limited PRC2 (effectively reducing ET ) (47), or does it persistently reduce its activity (e.g., S) after transient contact (38)? Our model predicts that varying ET should change the number of marks smoothly, whereas reducing S=L below a critical value can cause a sharp, global loss of marks (fig. S5). Stable domains remain only partly marked
Second, our model predicts that only about half of the monomers in marked regions are marked. As S=L or ET is varied, the stable mark domains 3 of 9
RES EARCH | R E S E A R C H A R T I C L E
A
B
10,000 monomers
t
t (generations)
0
100
200
C
Unlimited enzyme
10,000 monomers
D
E
t
t (generations)
0
100
200
F
Limited enzyme
Fig. 3. Limited enzyme stabilizes memory over multiple generations. (A) Time evolution of a mark pattern, with unlimited enzyme, over 200 generations (a ¼ 2:4 kB T, S=L = 0.42), starting from an initial pattern consisting of a single domain of 1000 marked monomers. Inset circles show snapshots in time of the polymer configuration. For this choice of parameter, marks spread everywhere and the polymer collapses. (B) Time evolution of the mark pattern, with unlimited enzyme,
(Fig. 3E) arising in the limited enzyme regime vary in length, but the fraction of monomers within the domain that are marked remains roughly constant, around 0.55 (fig. S6A). This “semimarking” phenomenon is consistent with Owen et al., Science 382, eadg3053 (2023)
as a function of a and S=L. There is no stable memory. (C) With unlimited enzyme, the global marking rate in the nucleus, VB , is proportional to the number NAB of A-B pairs. (D and E) Just as in (A) and (B), but with limited enzyme, ET ¼ 1000. Stable memory is achieved for hundreds of generations, over a broad range of parameters. (F) With limited enzyme, VB is proportional to NAB when it is small, but then saturates at the value Vmax ¼ SET , preventing uncontrolled spreading of marks.
several experimental results. Semimarking leads to a density difference of two- to threefold between compartments (fig. S6B), which is consistent with observed differences between heterochromatic and euchromatic regions in the
17 November 2023
nucleus (48). Semimarked domains also fold into irregular structures (fig. S6B)—as recently observed by super-resolution microscopy targeting Polycomb-repressed Hox genes (49)—instead of spheres, as would fully marked domains. 4 of 9
RES EARCH | R E S E A R C H A R T I C L E
Additionally, semimarking explains the counterintuitive findings of Alabert et al. (50) showing that certain histone marks require several cell generations to be fully established on new histones after replicational dilution and that old histones keep getting marked; our model naturally reproduces these observations (fig. S7). Mark redistribution and error correction
Third, our model predicts a coupling between distant genomic regions that is mediated by the titration of the limited enzyme. The plainest consequence of this is that if marks are lost somewhere, they tend to be gained elsewhere. In capturing this, our model agrees with the numerous observations of such titration effects in epigenetic systems (51–53). As one illustration of this, we show that our model (fig. S8) can emulate the findings of Kraft et al. (15), which demonstrated that genomic deletion of PRC2 nucleation sites can cause loss of H3K27me3 local to the deletion but gain of the mark elsewhere. We found that this “mark redistribution” has a natural directionality to it: Marks tend to flow from smaller domains to larger domains. A pattern consisting of multiple, noncontiguous mark domains can be remembered for hundreds of generations (Fig. 4A). However, we observed that over a longer timescale, the separate domains compete with one another for the limited enzyme (even when “infinitely” far apart) (fig. S9), inexorably leading to the formation of a single large domain. The spontaneous formation of a large marked domain by mark redistribution after many cell generations is reminiscent of the formation of senescence-associated heterochromatin foci (SAHF) in senescent cells, which is associated with loss of heterochromatin elsewhere (54). Present accounts of SAHF formation suggest an orchestrated process regulated by many specific effectors (55), but our findings highlight the possibility that similar behavior could be a primitive tendency of mark spreading coupled to 3D genome organization. Longer domains are more stable against mark redistribution in direct proportion to their length (fig. S9). This effect extends to clusters of domains; as the interdomain separation is decreased, they begin to act as a single larger domain, lasting longer in competition with a larger domain (Fig. 4B). These predictions could be tested by observing the fate of artificial ectopic mark domains of differing lengths (20, 56, 57) and clusters of small marked domains. Because tiny domains are lost quickly, mark redistribution can be viewed as a form of error correction. If “errors” appear in the form of a background rate at which monomers spontaneously switch from A to B, creating “domains” consisting of individual monomers, these errors are corrected immediately by redistribution of the marks to a larger domain (Fig. 4C and fig. Owen et al., Science 382, eadg3053 (2023)
S10). Resistance to this kind of error is important for any model of epigenetic memory that is based on spreading by reader-writer enzymes because these enzymes have (at some low, but nonzero rate) nonspecific writing activity, unstimulated by the reader domain (9). Conversely, this finding suggests that mechanisms other than ours must be at work in small, unusually stable mark domains, such as the three-nucleosome FLC nucleation region of Arabidopsis (58, 59). Our model is compatible with such mechanisms; introducing small, permanently marked regions to our model does not alter the basic story (fig. S11). Epigenetic heterogeneity can emerge stochastically and then remain stable
The final category of tests for our model stems from its ability to capture the emergence of epigenetic heterogeneity in a cell population. We considered the case in which marks are initially present in a small contiguous region and then S=L or ET is suddenly increased (Fig. 4, D to I). This could represent a developmental event, such as an increase in the duration of the cell cycle (which effectively decreases L), or the overexpression and activation of a readerwriter enzyme (an increase inET). Immediately, new marks emerge randomly along the polymer, but over a few cell generations they redistribute to form one or a few large domains, strongly biased to include the small initially marked domain (Fig. 4D and movie S1). At the level of a population average (Fig. 4E), the initial domain appears to simply expand linearly into a larger one. But in fact, there is large singlecell variation involving noncontiguous domains (Fig. 4F), a prediction that single-cell epigenomic techniques (60) could test. This behavior means that our model, without any modification or additional elements, can reproduce both classic and emerging aspects of the position-effect variegation (PEV). In PEV, translocations of the white gene of Drosophila to a genomic position near or within heterochromatin lead to stochastic but mitotically heritable silencing of the gene (51). This results in a variegating phenotype characterized by mottled red-white eyes, in which clonal patches bear the same coloration. Thus, the state of the locus is stochastic yet memorized over many cell divisions. To see PEV in our model, we created cell lineage trees by simply duplicating our simulation after every generation and then continuing the simulation of the copies independently. We then interrogated the marking status of a small regulatory region somewhere along the polymer to read out the silencing status over time in the lineage (Fig. 4, G to I). Solely by varying the position of this region, relative to the initially marked domain, our model reproduces drastically different observed phenotypes— including both the sectored (Fig. 4H) and “saltand-pepper” (Fig. 4I) modes of variegation
17 November 2023
(61, 62)—thus providing a mechanistic rationale for this classic phenomenon of stochastically established yet memorized epigenetic states. Discussion
We have shown how three ingredients— self-attraction of marked regions, 3D spread of marks, and limited enzyme—give rise to stable mark patterns that could serve as seats of epigenetic memory. Of these ingredients, we want to especially highlight the importance of the limitation of reader-writer enzymes— an element which is biologically very plausible, often neglected in models of mark dynamics, and which completely changes the system’s behavior. The mechanism that we identified accounts for the stable maintenance of mark domains after their establishment but does not address the question of why certain regions get marked in the first place. Mark patterns can certainly be strongly influenced by processes that we do not model here, such as nucleation regions to which modifying enzymes are recruited (14, 23) or actively transcribed regions that are impervious to repressive marks (63–66). Our memory mechanism is compatible with such exogenous influences. Simulations in which we imposed the condition that some regions are “pinned” to be permanently marked (fig. S11), or conversely are unmarkable (fig. S12), exhibit stable memory of mark domains away from pinned regions. It is also possible that some genomic elements are required to license certain regions for memory (67, 68). Such conditional nucleation sites (20) could be modeled as markable regions separated by large unmarkable ones. In fact, in our model, this kind of architecture may help memory by slowing mark redistribution (fig. S12B). Our model makes a number of assumptions and has several limitations, including its consideration of just a single epigenetic mark rather than competing or successive levels of modification, as well as the absence of attraction of heterochromatin to the nuclear lamina. We explored variants of our model that relax two of our central assumptions: the absence of mitotic mark dynamics and the absence of interphase chromatin dynamics. We found that spreading of marks during mitosis has only a small effect, most fundamentally because mitosis is short in duration relative to the length of the cell cycle (fig. S3 and supplementary text). This effect compounds (figure S3B) with any reduced activity of modifying enzymes on mitotic chromatin (e.g., due to reduced chemical or physical accessibility), further diminishing the role of possible mitotic spread of marks. Interphase dynamics can accelerate the loss of mark patterns (especially when loss is solely due to replicative dilution), but we found that this can be rescued by increasing the strength of self-attraction (fig. S4 and supplementary 5 of 9
RES EARCH | R E S E A R C H A R T I C L E
Single realization
A
Average over realizations
B
0
0
(generations)
25
50
decreasing inter-domain separation
500
C (generations)
0
1000
“Errors”:
A
B
50
100 Single realization (e.g. one cell lineage)
D
G
(generations)
0
100
200
E
Average over realizations
H
(generations)
0
100
200
F
Samples from distinct realizations
I
Fig. 4. Dynamics of complex patterns and position-effect variegation. (A) Time evolution of a pattern consisting of three equally sized mark domains (a ¼ 2:4 kB T, S=L ¼ 0:5, ET ¼ 3000). The pattern is stable for hundreds of generations, although marks eventually redistribute to form a single contiguous domain. In a population average (right), the three domains instead appear to merge. (B) Multiple small domains (a ¼ 2:4 kB T, S=L ¼ 0:5; population average shown) competing with a much bigger one (not pictured) survive redistribution longer the closer they are together. (C) Error correction: Tiny “domains” introduced by spontaneous marking at rate k are lost immediately (a ¼ 2:4 kB T, S=L ¼ 0:5, ET ¼ 1000, k=L ¼ 0:003). (D) Expansion of marks from a small domain leads to the random formation of new domains that can be remembered for hundreds of Owen et al., Science 382, eadg3053 (2023)
17 November 2023
generations (a ¼ 2:4 kB T, S=L ¼ 0:5, ET ¼ 3000). (E) The population average of (D) hides the large cell variation shown in (F). (G to I) Position-effect variegation: To visualize the consequences of this, we consider a gene regulatory region of five consecutive monomers (red) somewhere along the polymer and take the presence of a single B monomer in this region to silence the gene. Investigating silencing status (white indicates silenced; red indicates not silenced) in a lineage tree generated with our model, we find different phenotypes reminiscent of the classic position-effect variegation: (G) wild-type, (H) sectored variegation, and (I) “salt-and-pepper” variegation, depending on the position of the regulatory region. [The clip art of fly eyes is from the Database Center for Life Science (CC BY 4.0).] 6 of 9
RES EARCH | R E S E A R C H A R T I C L E
text). Tethering of heterochromatin to the lamina might also play this role, hinting at a mechanism that could link disruptions of the lamina to deterioration of epigenetic memory (69, 70). A natural question to ask about epigenetic memory systems is “How much information can be stored and for how long?” As a first step toward addressing this question for our system, we chose a scheme for recording and reading out “bits” in a mark pattern and then investigated how the probability of a bit error grows over successive cell generations in our model (fig. S13 and supplementary text). We uncovered a capacity-stability tradeoff: The more bits one seeks to encode (in our polymer of fixed length), the shorter the memory. For example, our system can reliably memorize at least 8 bits for 50 generations or at least 17 bits for 20 generations. These are only lower bounds on the capacity because we have not shown that our scheme for encoding bits is optimal. Errors arise through mark redistribution, and we only poorly understand what controls the redistribution timescale, aside from the expectation that it will increase with system size. Despite these caveats, our estimate gives us a sense of scale: A mechanism such as ours (using only 104 monomers) could provide stability to 250 (~28) alternative cellular states over 50 generations. Intuitively, the memory mechanism that we uncovered relies on the encoding of memory in different forms in different phases of the cell cycle. In interphase, memory is held in the 3D structure of the genome in the form of density differences because dynamic marks sharply localize to dense regions. During mitosis, when the 3D structure is being totally reorganized, memory is held in the 1D sequence of marks. The mark dynamics on a fixed polymer in our model clearly have some affinity to the protein-sequence design problem (71–73), for which the goal is to find an amino acid sequence that will fold into a target 3D structure. Classically, the design may be accomplished by choosing a sequence that minimizes the energy of the target configuration relative to all other configurations (74, 75). Analogously, our mark dynamics—although not directly minimizing the energy of a target structure—nevertheless perform a kind of sequence design, giving rise to a mark sequence that refolds into a similar polymer structure. The dynamics of our model could then be thought of as iterated rounds of design and refolding, with the goal of preserving the sequence—a problem that is different from sequence design and that to our knowledge has not been considered in the proteinfolding field. The encoding of a mark pattern through folding of the polymer, within one cell generation, could also be thought of as the “learning rule” of an associative memory in a Hopfield network (76). Learning by the Hebb rule in such networks strengthens connections between Owen et al., Science 382, eadg3053 (2023)
active neurons (77, 78); here, connections between marked regions are established by folding them together (“mark together, park together”). In this analogy, the mark dynamics are like the “update rule” that allows recovery of a stored memory. This lens is particularly relevant in view of the growing recognition that single cells (79–81) and simple chemical systems (82, 83) are capable of remarkably complex behaviors and memory. The possibility that epigenetic systems are capable not just of memory but also of more sophisticated information processing, such as associative learning, should be kept in mind; it may be the key to understanding them in their full complexity. Materials and methods
To simulate our model, we used polychrom (84), a lab-developed wrapper of OpenMM (85), for the polymer dynamics, and EoN (Epidemics on Networks) (86) for the mark dynamics. Our simulated polymer consists of 10,000 monomers connected by harmonic bonds with natural length l ¼ 1 . Monomers are of two types, A and B. Every pair of monomers additionally interacts according to an interparticle potential (supplementary text) that is repulsive for r < 1 and that for B-B pairs models shortranged attraction between r = 1 and r = 1.5 as a smoothed square well of depth a. Finally, our polymer is confined to a sphere with radius chosen so that the volume fraction occupied by monomers is 5%. This value may be compared with a rough estimate for the volume fraction of nucleosomes in the nucleus, for example, ~ð30 million 500 nm3 Þ=300 mm3 ¼ 0:05. In our simulations, the polymer was initialized with the “grow_cubic” function of polychrom (84), which generates a compact, unknotted random walk on cubic lattice. It was then relaxed from this state by using the variable timestep Langevin integrator of OpenMM (error tolerance = 0.0005). Note that our simulation was performed in the underdamped (ballistic) regime (frictionCoeff = 0.01 inverse OpenMM “picoseconds”) to enable use of the efficient variable timestep integrator. This did not change the equilibrium distribution of the monomer positions, from which we were interested in sampling. For all our simulations, we relaxed the polymer for 1000 OpenMM “picoseconds,” which is long enough for the polymer to expand to fill the confining sphere and to develop compartmentalization according to the monomer identities. As a control, we varied this relaxation time by a factor of 10 in either direction and observed no significant differences in the time evolution of mark patterns (fig. S15). For the mark dynamics, we generated a contact graph, G, from the relaxed polymer configuration, where the vertices of the graph are all the monomers and there is an edge be-
17 November 2023
tween two vertices if the distance between the associated monomers is less than the spreading radius, 1.5 (as a control, we show consequences of varying this value in fig. S16). We then simulated the dynamics of spreading and loss on this graph G with the fast_SIS (87) function of EoN, which performs an exact stochastic simulation of the Markovian SIS model on the graph G. Note that these mark dynamics are nonequilibrium (violating detailed balance). The marks relaxed, according to their dynamics, toward an extremely long-lived metastable state from which we sought to sample (the true steady state of these dynamics is always the absorbing state where there are no marks at all). The mark relaxation time that we used in all our simulations was 200 divided by the loss rate, L, which appears more than sufficient to reach the metastable state. We tested this by varying the relaxation time in both directions and saw no appreciable changes (fig. S15). In the case of limited enzyme, in which there is an effective spreading rate dependent on the number of marks (supplementary text), we updated the spreading rate according to the evolving number of marks 200 times for each cell generation.
REFERENCES AND NOTES
1. M. Ptashne, A Genetic Switch: Gene Control and Phage l (Cell Press and Blackwell, 1986). 2. T. S. Gardner, C. R. Cantor, J. J. Collins, Construction of a genetic toggle switch in Escherichia coli. Nature 403, 339–342 (2000). doi: 10.1038/35002131; pmid: 10659857 3. R. Zhu, J. M. Del Rio-Salgado, J. Garcia-Ojalvo, M. B. Elowitz, Synthetic multistability in mammalian cells. Science 375, eabg9765 (2022). doi: 10.1126/science.abg9765; pmid: 35050677 4. I. B. Dodd, M. A. Micheelsen, K. Sneppen, G. Thon, Theoretical analysis of epigenetic cell memory by nucleosome modification. Cell 129, 813–822 (2007). doi: 10.1016/ j.cell.2007.02.053; pmid: 17512413 5. S. Berry, M. Hartley, T. S. G. Olsson, C. Dean, M. Howard, Local chromatin environment of a Polycomb target gene instructs its own epigenetic inheritance. eLife 4, e07205 (2015). doi: 10.7554/eLife.07205; pmid: 25955967 6. I. B. Dodd, K. Sneppen, “Modeling Bistable Chromatin States” in Epigenetics and Systems Biology (Elsevier, 2017), pp. 145–168. 7. R. Bonasio, S. Tu, D. Reinberg, Molecular signals of epigenetic states. Science 330, 612–616 (2010). doi: 10.1126/ science.1191078; pmid: 21030644 8. R. Margueron et al., Role of the polycomb protein EED in the propagation of repressive histone marks. Nature 461, 762–767 (2009). doi: 10.1038/nature08398; pmid: 19767730 9. M. M. Müller, B. Fierz, L. Bittova, G. Liszczak, T. W. Muir, A two-state activation mechanism controls the histone methyltransferase Suv39h1. Nat. Chem. Biol. 12, 188–193 (2016). doi: 10.1038/nchembio.2008; pmid: 26807716 10. N. Reverón-Gómez et al., Accurate recycling of parental histones reproduces the histone modification landscape during DNA replication. Mol. Cell 72, 239–249.e5 (2018). doi: 10.1016/ j.molcel.2018.08.010; pmid: 30146316 11. T. M. Escobar et al., Active and repressed chromatin domains exhibit distinct nucleosome segregation during DNA replication. Cell 179, 953–963.e11 (2019). doi: 10.1016/ j.cell.2019.10.009; pmid: 31675501 12. N. A. Hathaway et al., Dynamics and memory of heterochromatin in living cells. Cell 149, 1447–1460 (2012). doi: 10.1016/j.cell.2012.03.052; pmid: 22704655 13. C. Hodges, G. R. Crabtree, Dynamics of inherently bounded histone modification domains. Proc. Natl. Acad. Sci. U.S.A. 109, 13296–13301 (2012). doi: 10.1073/pnas.1211172109; pmid: 22847427
7 of 9
RES EARCH | R E S E A R C H A R T I C L E
14. O. Oksuz et al., Capturing the onset of PRC2-mediated repressive domain formation. Mol. Cell 70, 1149–1162.e5 (2018). doi: 10.1016/j.molcel.2018.05.023; pmid: 29932905 15. K. Kraft et al., Polycomb-mediated genome architecture enables long-range spreading of H3K27 methylation. Proc. Natl. Acad. Sci. U.S.A. 119, e2201883119 (2022). doi: 10.1073/pnas.2201883119; pmid: 35617427 16. E. E. Dormidontova, A. Y. Grosberg, A. R. Khokhlov, Intramolecular phase separation of a copolymer chain with mobile primary structure. Macromol. Theory Simul. 1, 375–385 (1992). doi: 10.1002/mats.1992.040010603 17. D. Michieletto, E. Orlandini, D. Marenduzzo, Polymer model with epigenetic recoloring reveals a pathway for the de novo establishment and 3D organization of chromatin domains. Phys. Rev. X 6, 041047 (2016). doi: 10.1103/ PhysRevX.6.041047 18. R. Cortini et al., The physics of epigenetics. Rev. Mod. Phys. 88, 025002 (2016). doi: 10.1103/RevModPhys.88.025002 19. D. Jost, C. Vaillant, Epigenomics in 3D: Importance of longrange spreading and specific interactions in epigenomic maintenance. Nucleic Acids Res. 46, 2252–2264 (2018). doi: 10.1093/nar/gky009; pmid: 29365171 20. F. Erdel, E. C. Greene, Generalized nucleation and looping model for epigenetic memory of histone modifications. Proc. Natl. Acad. Sci. U.S.A. 113, E4180–E4189 (2016). doi: 10.1073/ pnas.1605862113; pmid: 27382173 21. S. H. Sandholtz, Q. MacPherson, A. J. Spakowitz, Physical modeling of the heritability and maintenance of epigenetic modifications. Proc. Natl. Acad. Sci. U.S.A. 117, 20423–20429 (2020). doi: 10.1073/pnas.1920499117; pmid: 32778583 22. M. Katava, G. Shi, D. Thirumalai, Chromatin dynamics controls epigenetic domain formation. Biophys. J. 121, 2895–2905 (2022). doi: 10.1016/j.bpj.2022.07.001; pmid: 35799447 23. D. Michieletto et al., Shaping epigenetic memory via genomic bookmarking. Nucleic Acids Res. 46, 83–93 (2018). doi: 10.1093/nar/gkx1200; pmid: 29190361 24. A. Z. Abdulla, C. Vaillant, D. Jost, Painters in chromatin: A unified quantitative framework to systematically characterize epigenome regulation and memory. Nucleic Acids Res. 50, 9083–9104 (2022). doi: 10.1093/nar/gkac702; pmid: 36018799 25. H. Bauer, Structure and arrangement of salivary gland chromosomes in Drosophila species. Proc. Natl. Acad. Sci. U.S.A. 22, 216–222 (1936). doi: 10.1073/pnas.22.4.216; pmid: 16577698 26. A. R. Strom et al., Phase separation drives heterochromatin domain formation. Nature 547, 241–245 (2017). doi: 10.1038/ nature22989; pmid: 28636597 27. A. G. Larson et al., Liquid droplet formation by HP1a suggests a role for phase separation in heterochromatin. Nature 547, 236–240 (2017). doi: 10.1038/nature22822; pmid: 28636604 28. M. Falk et al., Heterochromatin drives compartmentalization of inverted and conventional nuclei. Nature 570, 395–399 (2019). doi: 10.1038/s41586-019-1275-3; pmid: 31168090 29. H. Strickfaden, A. Zunhammer, S. van Koningsbruggen, D. Köhler, T. Cremer, 4D chromatin dynamics in cycling cells: Theodor Boveri’s hypotheses revisited. Nucleus 1, 284–297 (2010). doi: 10.4161/nucl.11969; pmid: 21327076 30. C. Kadoch et al., Dynamics of BAF-Polycomb complex opposition on heterochromatin in normal and oncogenic states. Nat. Genet. 49, 213–222 (2017). doi: 10.1038/ng.3734; pmid: 27941796 31. P. Dobrinić, A. T. Szczurek, R. J. Klose, PRC1 drives Polycombmediated gene repression by controlling transcription initiation and burst frequency. Nat. Struct. Mol. Biol. 28, 811–824 (2021). doi: 10.1038/s41594-021-00661-y; pmid: 34608337 32. M. Gabriele et al., Dynamics of CTCF- and cohesin-mediated chromatin looping revealed by live-cell imaging. Science 376, 496–501 (2022). doi: 10.1126/science.abn6583; pmid: 35420890 33. K. Ito, K. S. Zaret, Maintaining transcriptional specificity through mitosis. Annu. Rev. Genomics Hum. Genet. 23, 53–71 (2022). doi: 10.1146/annurev-genom-121321-094603; pmid: 35440147 34. Y. Liu et al., Widespread mitotic bookmarking by histone marks and transcription factors in pluripotent stem cells. Cell Rep. 19, 1283–1293 (2017). doi: 10.1016/ j.celrep.2017.04.067; pmid: 28514649 35. S. Rea et al., Regulation of chromatin structure by site-specific histone H3 methyltransferases. Nature 406, 593–599 (2000). doi: 10.1038/35020506; pmid: 10949293 36. P. N. I. Lau, P. Cheung, Histone code pathway involving H3 S28 phosphorylation and K27 acetylation activates transcription
Owen et al., Science 382, eadg3053 (2023)
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
and antagonizes polycomb silencing. Proc. Natl. Acad. Sci. U.S.A. 108, 2801–2806 (2011). doi: 10.1073/ pnas.1012798108; pmid: 21282660 R. Pastor-Satorras, C. Castellano, P. Van Mieghem, A. Vespignani, Epidemic processes in complex networks. Rev. Mod. Phys. 87, 925–979 (2015). doi: 10.1103/RevModPhys.87.925 J. M. Stafford et al., Multiple modes of PRC2 inhibition elicit global chromatin alterations in H3K27M pediatric glioma. Sci. Adv. 4, eaau5935 (2018). doi: 10.1126/sciadv.aau5935; pmid: 30402543 R. Leicher et al., Single-molecule and in silico dissection of the interaction between Polycomb repressive complex 2 and chromatin. Proc. Natl. Acad. Sci. U.S.A. 117, 30465–30475 (2020). doi: 10.1073/pnas.2003395117; pmid: 33208532 M. A. Gillespie et al., Absolute Quantification of Transcription Factors Reveals Principles of Gene Regulation in Erythropoiesis. Mol. Cell 78, 960–974.e11 (2020). doi: 10.1016/ j.molcel.2020.03.031; pmid: 32330456 L. Michaelis, M. L. Menten, K. A. Johnson, R. S. Goody, The original Michaelis constant: Translation of the 1913 MichaelisMenten paper. Biochemistry 50, 8264–8269 (2011). doi: 10.1021/bi201284u; pmid: 21888353 J. Gunawardena, Time-scale separation—Michaelis and Menten’s old idea, still bearing fruit. FEBS J. 281, 473–488 (2014). doi: 10.1111/febs.12532; pmid: 24103070 A. Bancaud et al., Molecular crowding affects diffusion and binding of nuclear proteins in heterochromatin and reveals the fractal organization of chromatin. EMBO J. 28, 3785–3798 (2009). doi: 10.1038/emboj.2009.340; pmid: 19927119 M. T. McCabe et al., EZH2 inhibition as a therapeutic strategy for lymphoma with EZH2-activating mutations. Nature 492, 108–112 (2012). doi: 10.1038/nature11606; pmid: 23051747 W. Béguelin et al., Mutant EZH2 induces a pre-malignant lymphoma niche by reprogramming the immune response. Cancer Cell 37, 655–673.e11 (2020). doi: 10.1016/ j.ccell.2020.04.004; pmid: 32396861 G. P. Souroullas et al., An oncogenic Ezh2 mutation induces tumors through global redistribution of histone 3 lysine 27 trimethylation. Nat. Med. 22, 632–640 (2016). doi: 10.1038/ nm.4092; pmid: 27135738 K. L. Diehl et al., PRC2 engages a bivalent H3K27M-H3K27me3 dinucleosome inhibitor. Proc. Natl. Acad. Sci. U.S.A. 116, 22152–22157 (2019). doi: 10.1073/pnas.1911775116; pmid: 31611394 H. D. Ou et al., ChromEMT: Visualizing 3D chromatin structure and compaction in interphase and mitotic cells. Science 357, eaag0025 (2017). doi: 10.1126/science.aag0025; pmid: 28751582 S. Murphy, A. N. Boettiger, Polycomb repression of Hox genes involves spatial feedback but not domain compaction or demixing. bioRxiv 2022.10.14.512199 [Preprint] (2022). https://doi.org/10.1101/2022.10.14.512199. C. Alabert et al., Two distinct modes for propagation of histone PTMs across the cell cycle. Genes Dev. 29, 585–590 (2015). doi: 10.1101/gad.256354.114; pmid: 25792596 S. C. R. Elgin, G. Reuter, Position-effect variegation, heterochromatin formation, and gene silencing in Drosophila. Cold Spring Harb. Perspect. Biol. 5, a017780 (2013). doi: 10.1101/cshperspect.a017780; pmid: 23906716 C. M. Weber et al., mSWI/SNF promotes Polycomb repression both directly and through genome-wide redistribution. Nat. Struct. Mol. Biol. 28, 501–511 (2021). doi: 10.1038/ s41594-021-00604-7; pmid: 34117481 E. Conway et al., BAP1 enhances Polycomb repression by counteracting widespread H2AK119ub1 deposition and chromatin condensation. Mol. Cell 81, 3526–3541.e8 (2021). doi: 10.1016/j.molcel.2021.06.020; pmid: 34186021 X. Zhang et al., The loss of heterochromatin is associated with multiscale three-dimensional genome reorganization and aberrant transcription during cellular senescence. Genome Res. 31, 1121–1135 (2021). doi: 10.1101/gr.275235.121; pmid: 34140314 I. Olan, M. Narita, Senescence: An Identity Crisis Originating from Deep Within the Nucleus. Annu. Rev. Cell Dev. Biol. 38, 219–239 (2022). doi: 10.1146/annurev-cellbio-120420-013537; pmid: 35804478 P. I. Thakore et al., Highly specific epigenome editing by CRISPR-Cas9 repressors for silencing of distal regulatory elements. Nat. Methods 12, 1143–1149 (2015). doi: 10.1038/ nmeth.3630; pmid: 26501517 J. Pulecio, N. Verma, E. Mejía-Ramírez, D. Huangfu, A. Raya, CRISPR/Cas9-based engineering of the epigenome. Cell Stem Cell 21, 431–447 (2017). doi: 10.1016/j.stem.2017.09.006; pmid: 28985525
17 November 2023
58. H. Yang et al., Distinct phases of Polycomb silencing to hold epigenetic memory of cold in Arabidopsis. Science 357, 1142–1145 (2017). doi: 10.1126/science.aan1121; pmid: 28818969 59. C. Lövkvist et al., Hybrid protein assembly-histone modification mechanism for PRC2-based epigenetic switching and memory. eLife 10, e66454 (2021). doi: 10.7554/eLife.66454; pmid: 34473050 60. S. Preissl, K. J. Gaulton, B. Ren, Characterizing cis-regulatory elements using single-cell epigenomics. Nat. Rev. Genet. 24, 21–43 (2022). doi: 10.1038/s41576-022-00509-1; pmid: 35840754 61. B. Y. Lu, C. P. Bishop, J. C. Eissenberg, Developmental timing and tissue specificity of heterochromatin-mediated silencing. EMBO J. 15, 1323–1332 (1996). doi: 10.1002/j.1460-2075.1996. tb00474.x; pmid: 8635465 62. F. Bughio, G. R. Huckell, K. A. Maggert, Monitoring of switches in heterochromatin-induced silencing shows incomplete establishment and developmental instabilities. Proc. Natl. Acad. Sci. U.S.A. 116, 20043–20053 (2019). doi: 10.1073/ pnas.1909724116; pmid: 31527269 63. F. W. Schmitges et al., Histone methylation by PRC2 is inhibited by active chromatin marks. Mol. Cell 42, 330–341 (2011). doi: 10.1016/j.molcel.2011.03.025; pmid: 21549310 64. S. Henikoff, A. Shilatifard, Histone modification: Cause or cog? Trends Genet. 27, 389–396 (2011). doi: 10.1016/ j.tig.2011.06.006; pmid: 21764166 65. S. Berry, C. Dean, M. Howard, Slow chromatin dynamics allow polycomb target genes to filter fluctuations in transcription factor activity. Cell Syst. 4, 445–457.e8 (2017). doi: 10.1016/ j.cels.2017.02.013; pmid: 28342717 66. C. Lövkvist, M. Howard, Using computational modelling to reveal mechanisms of epigenetic Polycomb control. Biochem. Soc. Trans. 49, 71–77 (2021). doi: 10.1042/BST20190955; pmid: 33616630 67. R. T. Coleman, G. Struhl, Causal role for inheritance of H3K27me3 in maintaining the OFF state of a Drosophila HOX gene. Science 356, eaai8236 (2017). doi: 10.1126/science. aai8236; pmid: 28302795 68. T. A. Shafiq, D. Moazed, Three rules for epigenetic inheritance of human Polycomb silencing. bioRxiv2023.02.27.530239 [Preprint] (2023). .doi: 10.1101/2023.02.27.530239 69. D. K. Shumaker et al., Mutant nuclear lamin A leads to progressive alterations of epigenetic control in premature aging. Proc. Natl. Acad. Sci. U.S.A. 103, 8703–8708 (2006). doi: 10.1073/pnas.0602569103; pmid: 16738054 70. A. Karoutas, A. Akhtar, Functional mechanisms and abnormalities of the nuclear lamina. Nat. Cell Biol. 23, 116–126 (2021). doi: 10.1038/s41556-020-00630-5; pmid: 33558730 71. V. S. Pande, A. Y. Grosberg, T. Tanaka, Heteropolymer freezing and design: Towards physical models of protein folding. Rev. Mod. Phys. 72, 259–314 (2000). doi: 10.1103/ RevModPhys.72.259 72. A. Y. Grosberg, A. R. Khokhlov, Giant Molecules: Here, There, and Everywhere (World Scientific Publishing, ed. 2, 2010). 73. T. M. Fink, R. C. Ball, How many conformations can a protein remember? Phys. Rev. Lett. 87, 198103 (2001). doi: 10.1103/ PhysRevLett.87.198103; pmid: 11690459 74. E. I. Shakhnovich, A. M. Gutin, Engineering of stable and fastfolding sequences of model proteins. Proc. Natl. Acad. Sci. U.S.A. 90, 7195–7199 (1993). doi: 10.1073/pnas.90.15.7195; pmid: 8346235 75. V. S. Pande, A. Y. Grosberg, T. Tanaka, Thermodynamic procedure to synthesize heteropolymers that can renature to recognize a given target molecule. Proc. Natl. Acad. Sci. U.S.A. 91, 12976–12979 (1994). doi: 10.1073/pnas.91.26.12976; pmid: 7809158 76. J. J. Hopfield, Neural networks and physical systems with emergent collective computational abilities. Proc. Natl. Acad. Sci. U.S.A. 79, 2554–2558 (1982). doi: 10.1073/ pnas.79.8.2554; pmid: 6953413 77. D. O. Hebb, The Organization of Behavior: A Neuropsychological Theory (Wiley, 1949). 78. S. Löwel, W. Singer, Selection of intrinsic horizontal connections in the visual cortex by correlated neuronal activity. Science 255, 209–212 (1992). doi: 10.1126/science.1372754; pmid: 1372754 79. J. P. Dexter, S. Prabakaran, J. Gunawardena, A complex hierarchy of avoidance behaviors in a single-cell eukaryote. Curr. Biol. 29, 4323–4329.e2 (2019). doi: 10.1016/ j.cub.2019.10.059; pmid: 31813604 80. S. J. Gershman, P. E. M. Balbi, C. R. Gallistel, J. Gunawardena, Reconsidering the evidence for learning in single cells. eLife 10, e61907 (2021). doi: 10.7554/eLife.61907; pmid: 33395388
8 of 9
RES EARCH | R E S E A R C H A R T I C L E
81. M. Kramar, K. Alim, Encoding memory in tube diameter hierarchy of living flow network. Proc. Natl. Acad. Sci. U.S.A. 118, e2007815118 (2021). doi: 10.1073/pnas.2007815118; pmid: 33619174 82. A. Murugan, Z. Zeravcic, M. P. Brenner, S. Leibler, Multifarious assembly mixtures: Systems allowing retrieval of diverse stored structures. Proc. Natl. Acad. Sci. U.S.A. 112, 54–59 (2015). doi: 10.1073/pnas.1413941112; pmid: 25535383 83. W. Zhong, D. J. Schwab, A. Murugan, Associative pattern recognition through macro-molecular self-assembly. J. Stat. Phys. 167, 806–826 (2017). doi: 10.1007/s10955-017-1774-2 84. M. Imakaev, A. Goloborodko, H. B. Brandão, mirnylab/ polychrom: v0.1.0, Zenodo (2019); https://doi.org/10.5281/ ZENODO.3579473. 85. P. Eastman et al., OpenMM 7: Rapid development of high performance algorithms for molecular dynamics. PLOS Comput. Biol. 13, e1005659 (2017). doi: 10.1371/journal. pcbi.1005659; pmid: 28746339 86. J. C. Miller, T. Ting, EoN (Epidemics on Networks): A fast, flexible Python package for simulation, analytic approximation,
Owen et al., Science 382, eadg3053 (2023)
and analysis of epidemics on networks. J. Open Source Softw. 4, 1731 (2019). doi: 10.21105/joss.01731 87. I. Z. Kiss, J. C. Miller, P. L. Simon, Mathematics of Epidemics on Networks: From Exact to Approximate Models (Springer, 2017). 88. J. Owen, jaowen/3d-epigenetic-memory: First release, Version v1, Zenodo (2023); https://dx.doi.org/10.5281/ ZENODO.8322781. ACKN OWLED GMEN TS
The authors are very grateful to M. Kardar for valuable scientific discussions. Funding: This research was funded by the National Human Genome Research Institute, NIH grant 3UM1HG011536 (L.M.); the National Institute of General Medical Sciences, NIH grant GM114190 (L.M.); and NSF award 2044895 (L.M.). Author contributions: Conceptualization: J.A.O., D.O., and L.M. Methodology: J.A.O., D.O., and L.M. Investigation: J.A.O., D.O., and L.M. Funding acquisition: L.M. Project administration: L.M. Supervision: L.M. Writing – original draft: J.A.O., D.O., and L.M. Writing – review and editing: J.A.O., D.O., and L.M. Competing
17 November 2023
interests: The authors declare that they have no competing interests. Data and materials availability: To simulate our model, we used polychrom (84), a lab-developed wrapper of OpenMM (85) for the polymer dynamics, and EoN (Epidemics on Networks) (86) for the mark dynamics. Code is available at https://github.com/ jaowen/3d-epigenetic-memory/ and deposited at Zenodo (88). License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/about/science-licenses-journalarticle-reuse SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adg3053 Supplementary Text Figs. S1 to S16 References (89–94) Movie S1 Submitted 14 December 2022; accepted 28 September 2023 10.1126/science.adg3053
9 of 9
Your Legacy to Science A N E STAT E G I F T TO T H E A M E R I CA N A S S O C I AT I O N FO R T H E A DVA N C E M E N T O F S C I E N C E
Since 1848, our founding year, the American Association for the Advancement of Science (AAAS) has been deeply committed to advancing science, engineering and innovation around the world
“As a teacher and instructor, I bear responsibility for the younger generations. If you have extra resources, concentrate them on organizations, like AAAS, that are doing work for all.” —Prof. Elisabeth Ervin-Blankenheim, 1848 Society member
for the benefit of all people. By making AAAS a beneficiary of your will, trust, retirement plan or life insurance policy, you become a member of our 1848 Society, joining Thomas Edison, Alexander Graham Bell and the many
If you intend to include AAAS in your estate plans, provide this information to your lawyer or financial adviser: Legal Name: American Association for the Advancement of Science
distinguished individuals whose vision led to the creation of AAAS
Federal Tax ID Number: 53-0196568
and our world-renowned journal, Science, so many years ago.
Address: 1200 New York Avenue, NW, Washington, DC 20005
Unlike many of its peers, Science is not for-profit. Your estate gift
If you would like more information on making an estate gift to
would provide long-term financial stability and durable annual income that will support operations and competitive innovation for years to come. This support is vital.
AAAS, cut out and return the form below or send an email to [email protected]. Additional details are also available online at www.aaas.org/1848Society.
cut here
Yes, I would like more information about joining the AAAS 1848 Society. PLEASE CONTACT ME AT: Name: Address: City: Email:
State:
Zip code:
Country:
Phone:
RETURN THIS FORM TO: AAAS Office of Philanthropy and Strategic Partnerships • 1200 New York Avenue, NW • Washington, DC 20005 USA
RES EARCH
RESEARCH ARTICLE SUMMARY
◥
NEUROSCIENCE
A tool kit of highly selective and sensitive genetically encoded neuropeptide sensors Huan Wang†, Tongrui Qian†, Yulin Zhao, Yizhou Zhuo, Chunling Wu, Takuya Osakada, Peng Chen, Zijun Chen, Huixia Ren, Yuqi Yan, Lan Geng, Shengwei Fu, Long Mei, Guochuan Li, Ling Wu, Yiwen Jiang, Weiran Qian, Li Zhang, Wanling Peng, Min Xu, Ji Hu, Man Jiang, Liangyi Chen, Chao Tang, Yingjie Zhu, Dayu Lin, Jiang-Ning Zhou, Yulong Li*
INTRODUCTION: Neuropeptides are small chains
RATIONALE: Recent endeavors from our group
of amino acids that play vital roles in the endocrine and nervous systems, regulating diverse functions such as metabolism, pain perception, sleep and circadian rhythm, mood, and learning. Malfunctions in neuropeptide signaling have been implicated in many diseases, including insomnia, diabetes, and depression. Monitoring neuropeptides with high spatiotemporal resolution in vivo could provide insights into their functions in physiological conditions and disease pathophysiology and advance new drug development. Recent progress led to the development of fluorescent sensors for selected neuropeptides, including orexin and oxytocin, which helped reveal their in vivo release patterns and dynamics. However, given the large number of neuropeptides, the scalable development of neuropeptide sensors is appealing yet challenging.
and others in combining G protein–coupled receptors (GPCRs) with circularly permutated green fluorescent protein (cpGFP) have led to the development of several GPCR activation– based (GRAB) sensors that can detect neuromodulators with high spatiotemporal resolution. As most neuropeptide receptors belong to the GPCR superfamily, which endogenously has high affinity and specificity for their ligands, one could, in principle, develop genetically encoded fluorescence sensors for each neuropeptide. However, given the diversity of neuropeptides and their cognate GPCRs, developing and optimizing a GRAB sensor with each GPCR de novo could be labor-intensive. Despite variations in their sequences and structures, neuropeptide GPCRs do share a common structural change upon ligand-induced activation. Therefore, a streamlined approach by transplanting the entire cpGFP-containing
Report module from optimized GRAB sensors SST CCK NTS OX NPY CRF
cpEGFP
Cultured neurons
Linker and cpEGFP optimization
Cultured pancreas Pancreas
In vivo
Tail lift
LiCl injection
development of neuropeptide sensors, we have developed an ICL3 grafting method that efficiently generates genetically encoded fluorescent sensors. Leveraging this method, we developed a panel of sensors to detect the real-time dynamics of six commonly studied neuropeptides. We demonstrated that our GRAB SST and CRF sensors can be used to monitor neuropeptide dynamics in vitro, ex vivo, and in vivo with good sensitivity, selectivity, and spatiotemporal resolution. This flexible engineering strategy and toolkit of optimized peptide sensors pave the way for studying the release, regulation, and functions of diverse neuropeptides under both physiological and pathophysiological states.
▪
Acute brain slices Tail shock
Forced running
Pancreas islet
Development and applications of fluorescent sensors for detecting neuropeptides. OX, orexin; SP, substance P; GHRL, ghrelin; UCN, urocortin; PTH, parathyroid hormoneÐrelated peptide. [Created with BioRender.com] 786
RESULTS: Using this grafting strategy, we developed a series of highly selective and sensitive genetically encoded neuropeptide sensors for detecting somatostatin (SST), corticotropinreleasing factor (CRF), cholecystokinin (CCK), neuropeptide Y (NPY), neurotensin (NTS), and vasoactive intestinal peptide (VIP). These sensors are engineered by replacing the ICL3 in each new peptide GPCR with the entire cpGFPcontaining ICL3 from previously optimized norepinephrine (NE) sensor GRABNE. This array of peptide sensors enables the detection of specific neuropeptides at nanomolar concentrations with minimal disruptions to neuronal activity, transcriptional profiling, and animal behaviors. We demonstrated the utilities of SST and CRF sensors in detail. Specifically, we used the SST1.0 sensor to detect activity-dependent SST release in cultured rat cortical neurons and mice pancreatic islets and to reveal the SST dynamics during conditioned learning in mice. Moreover, the CRF1.0 sensor reliably reported the electrical stimulation evoked release of CRF in acute brain slices from mice, monitored in vivo changes in CRF levels in the hypothalamus, and visualized spatially resolved cortical CRF dynamics in response to stress-inducing stimuli in mice. CONCLUSION: To permit the fast and scalable
Peptide sensors
SP GHRL UCN VIP PTH
intracellular loop 3 (ICL3) from existing GRAB sensors to new neuropeptide GRAB sensors could give the new sensors the ability to change fluorescence upon ligand binding without modifying and optimizing each sensor individually. This strategy could greatly accelerate the development of a wide variety of GRAB sensors tailored to detect the real-time dynamics of diverse endogenous neuropeptides with minimal optimization needed.
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
The list of author affiliations is available in the full article online. *Corresponding author. Email: [email protected] †These authors contributed equally to this work. Cite this article as H. Wang et al., Science 382, eabq8173 (2023). DOI: 10.1126/science.abq8173
READ THE FULL ARTICLE AT https://doi.org/10.1126/science.abq8173 science.org SCIENCE
RES EARCH
RESEARCH ARTICLE
◥
NEUROSCIENCE
A tool kit of highly selective and sensitive genetically encoded neuropeptide sensors Huan Wang1,2†, Tongrui Qian1,2†, Yulin Zhao1,2, Yizhou Zhuo1,2, Chunling Wu1,2, Takuya Osakada3, Peng Chen4,5, Zijun Chen6, Huixia Ren7, Yuqi Yan1,2,7, Lan Geng1,2, Shengwei Fu1,2,7, Long Mei3, Guochuan Li1,2, Ling Wu1,2, Yiwen Jiang3, Weiran Qian8, Li Zhang9, Wanling Peng10, Min Xu10, Ji Hu11, Man Jiang9, Liangyi Chen7, Chao Tang7, Yingjie Zhu6, Dayu Lin3, Jiang-Ning Zhou4,5, Yulong Li1,2,7,12* Neuropeptides are key signaling molecules in the endocrine and nervous systems that regulate many critical physiological processes. Understanding the functions of neuropeptides in vivo requires the ability to monitor their dynamics with high specificity, sensitivity, and spatiotemporal resolution. However, this has been hindered by the lack of direct, sensitive, and noninvasive tools. We developed a series of GRAB (G protein–coupled receptor activation‒based) sensors for detecting somatostatin (SST), corticotropin-releasing factor (CRF), cholecystokinin (CCK), neuropeptide Y (NPY), neurotensin (NTS), and vasoactive intestinal peptide (VIP). These fluorescent sensors, which enable detection of specific neuropeptide binding at nanomolar concentrations, establish a robust tool kit for studying the release, function, and regulation of neuropeptides under both physiological and pathophysiological conditions.
N
europeptides were first identified nearly seven decades ago as hormone regulators in the endocrine system and have since been recognized as highly effective signaling molecules in both central and peripheral tissues (1–4). In the brain, neuropeptides regulate many types of physiological functions, such as digestion, metabolism, sleep and circadian rhythm, reproduction, and higher cognitive processes (5–8). Thus, neuropeptide signaling—which is mediated primarily by G protein–coupled receptors (GPCRs)—provides a key site for drug targeting for a wide range of diseases and conditions such as insomnia, pain, obesity, and diabetes (9–11). 1
State Key Laboratory of Membrane Biology, Peking University School of Life Sciences, Beijing 100871, China. 2IDG/McGovern Institute for Brain Research, Peking University, Beijing 100871, China. 3Department of Psychiatry and Department of Neuroscience and Physiology, New York University Langone Medical Center, New York, NY 10016, USA. 4Institute of Brain Science, The First Affiliated Hospital of Anhui Medical University, Hefei 230022, China. 5Chinese Academy of Sciences Key Laboratory of Brain Function and Diseases, Division of Life Sciences and Medicine, University of Science and Technology of China, Hefei 230026, China. 6Shenzhen Key Laboratory of Drug Addiction, Shenzhen Neher Neural Plasticity Laboratory, Brain Cognition and Brain Disease Institute, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, China. 7 Peking-Tsinghua Center for Life Sciences, Academy for Advanced Interdisciplinary Studies, Peking University, Beijing 100871, China. 8Institute of Molecular Medicine, Peking University, Beijing 100871, China. 9Department of Physiology, School of Basic Medicine and Tongji Medical College, Huazhong University of Science and Technology, Wuhan 430030, China. 10Chinese Academy of Sciences Center for Excellence in Brain Science and Intelligence Technology, Chinese Academy of Sciences, Shanghai 200031, China. 11 School of Life Science and Technology, ShanghaiTech University, Shanghai 201210, China. 12National Biomedical Imaging Center, Peking University, Beijing 100871, China. *Corresponding author. Email: [email protected] These authors contributed equally to this work.
Wang et al., Science 382, eabq8173 (2023)
The ability to measure the spatial and temporal dynamics of neuropeptides in vivo is essential for understanding their functions and the mechanisms that regulate these key signaling molecules. However, current methods for detecting peptides in the brain either lack the necessary spatiotemporal resolution or are not suitable for in vivo application. Thus, the precise spatiotemporal dynamics and release patterns of endogenous peptides remain poorly understood. Genetically encoded fluorescent indicators have proven suitable for measuring the dynamics of signaling molecules with high spatiotemporal resolution in vivo. For example, bacterial periplasmic binding protein (PBP) based sensors have been developed to detect neurotransmitters such as glutamate, acetylcholine, and serotonin (12–14). However, corresponding PBPs for peptides and proteins are unlikely to exist. Generating peptide-sensing PBPs with high affinity and selectivity will therefore require considerable bioengineering and screening. Notably, most neuropeptide receptors are GPCRs, and peptide or protein GPCR ligands make up 70% of all nonolfactory GPCR ligands in the human body (Fig. 1A) (15, 16). Peptide GPCRs can provide a valuable opportunity for generating genetically encoded sensors with high sensitivity and selectivity. Previously, our group and others developed and characterized several GPCR activation–based (GRAB) intensiometric biosensors, using GPCRs as the ligandsensing unit and circularly permutated green fluorescent protein (cpGFP) as the reporter module, for detecting small-molecule transmitters (17–21) and some peptides, including oxytocin and orexin (22–24). The strategy for developing these GRAB sensors includes screen-
17 November 2023
ing for the optimal cpGFP placement site within the receptor’s third intracellular loop 3 (ICL3); however, given the large number of peptide and protein GPCRs (with 131 expressed in humans) and the high variability of ICL3 among GPCRs (ranging from 2 to 211 amino acids), developing and optimizing a GRAB sensor for each GPCR would be highly labor-intensive (25, 26). Despite this structural variation in the ICL3, however, peptide GPCRs undergo a common structural change upon activation, with an outward movement of transmembrane 6 (TM6) observed in both class A and class B1 peptide GPCRs (Fig. 1B) (27–29). Thus, peptide GPCRs generated using the entire cpGFPcontaining ICL3 in previously optimized GRAB sensors may retain the ability to couple the activation-induced conformational change with an increase in fluorescence, thereby accelerating the development of a wide variety of GRAB peptide sensors. In this study, we used this strategy to develop a series of GRAB sensors for detecting neuropeptides with nanomolar affinity. These sensors are able to reveal peptide dynamics with singlecell spatial resolution and subsecond temporal resolution. Developing a generalized method for engineering fluorescent sensors to detect neuropeptides
Neuropeptides and peptide receptors were widely expressed in the brain (30–32), and we chose these highly expressed peptide GPCRs as scaffolds for sensor development (Fig. 1C). To develop a scalable method for generating a series of genetically encoded peptide sensors, we replaced the ICL3 domains in various peptide GPCRs with the ICL3 in several existing sensors, including GRABNE1m, GRABDA2m, GRAB5-HT1.0, GRABACh3.0, and dLight1.3b (19–21, 33, 34). These sensor-derived ICL3s vary in length with respect to the number of amino acids that flank the cpGFP module (table S1); thus, GRAB peptide sensors were generated by replacing the ICL3 in the GPCR with the linker sequences and cpGFP derived from the inner membrane regions of TM5 and TM6, located at sites around 5.70 and 6.28, respectively (Fig. 1C). Each newly generated candidate peptide sensor was then expressed in human embryonic kidney (HEK) 293T (HEK293T) cells together with a plasma membrane–targeted mCherry (as a marker of surface expression) (fig. S1A). Each candidate’s performance was measured with respect to trafficking to the plasma membrane and the change in the sensor’s fluorescence in response to the appropriate ligand (Fig. 1D and fig. S1, B and C). Candidates with a trafficking index of >80% (measured as the Pearson correlation coefficient between the expression of a candidate and mCherry) and a fluorescence increase of >30% upon ligand application were considered as responsive peptide sensors. 1 of 16
RES EARCH | R E S E A R C H A R T I C L E
A Ca2+
~78
Ca2+ Monoamines Amino acids Nucleotides Lipids Peptides …
LDCVs
SVs
Glu, GABA Ionotropic
D 68 AAs
GRAB-NE cpEGFP
NPY
ICL3 of GPCR-based sensors
GRAB-DA 14
32
12 AAs
TM5
TM6
Inactive Active
2
cpEGFP
Peptide GRAB sensors TM5 5.68 5.70 5.72 5.74
PTH
SP GHRL CRF UCN VIP
13/16 81%
3
ΔF/F0
OX
Class B1 CRF1R
TM6
TM5
SST CCK CRF NPY NTS VIP …
Peptide GPCRs SST CCK NTS
Class A SSTR2
~187
GPCR
C
B
Peptide/ protein ligands
Nonpeptidergic ligands
GRAB-ACh
10/15 67%
cpEGFP
4
5
GRAB-5-HT
2/14 14%
cpEGFP
TM6 6.26 6.28 6.30 6.32
6
5
2/14 14%
dLight cpEGFP
5
4/14 28%
2 1
threshold
TM5
cpEGFP
TM6
3. cpEGFP optimization TM5
cpEGFP
TM6
17 November 2023
1500
ΔF/F0
1000
1000
2000
NTS1.0 500 200
100
1000
1500
NPY1.0
400
VIP1.0
200
Number of variants
300
SSTR5 NPY1R GHSR AVPR2 NTSR1 HCRTR2 OPRM1 GRPR TACR1 TRHR VIPR1 VIPR2 CRF1R PTH1R
Class B1
15
1500
CCK1.0
CRF1.0
Fig. 1. A general method for engineering fluorescent indicators for neuropeptides. (A) (Left) Illustration of peptide-containing large dense-core vesicles (LDCVs), neurotransmitter-containing synaptic vesicles (SVs), and their receptors in a synapse. (Right) Proportion and number of peptide or protein GPCR ligands and nonpeptidergic GPCRs in humans, with corresponding examples. (B) Superposition of active (blue) and inactive (gray) structures of class A SSTR2 [Protein Data Bank (PDB) IDs 7XMR and 7XN9] and class B1 CRF1R (PDB IDs 4K5Y and 6PB0). The dashed arrows indicate the movement of the sixth transmembrane domain (TM6). (C) Schematic diagram depicting the ICL3 transplantation strategy for developing GRAB sensors. (D) Fluorescence responses (DF/F0) of peptide GPCR chimeras with ICL3 transplanted from the indicated sensors. The amino acid (AA) numbers flanking cpGFP are labeled. The number and percentage of GPCRs with a maximum Wang et al., Science 382, eabq8173 (2023)
SSTR5 NPY1R GHSR AVPR2 NTSR1 HCRTR2 OPRM1 GRPR TACR1 TRHR VIPR1 VIPR2 CRF1R PTH1R
SSTR5 NPY1R GHSR AVPR2 NTSR1 HCRTR2 OPRM1 GRPR TACR1 TRHR VIPR1 VIPR2 CRF1R PTH1R
500
1000
Class A
10 5 0 1
0.5 0
GRAB
EGFP -CAAX
2. Linker optimization
500
G
SST CCK NTS OX NPY SP GHRL CRF UCN VIP PTH
TM6
SST1.0
Norm. Brightness
cpEGFP
0 10 5 0 0 20 10 0 0 4 2 0 0 3 1.5 0 0 4 2 0 0
ΔF/F0 ΔF/F0 ΔF/F0
TM5
4 2 0
ΔF/F0
1. ICL3 L3 replacing sites
F
ΔF/F0
E GRAB sensors optimization
ΔF/F0
SSTR5 NPY1R GHSR AVPR2 NTSR1 CCKBR HCRTR2 OPRM1 GRPR TACR1 TRHR VIPR1 VIPR2 CRF1R PTH1R
SSTR5 NPY1R NPY5R GHSR AVPR2 NTSR1 CCKBR HCRTR2 OPRM1 GRPR TACR1 TRHR VIPR1 VIPR2 CRF1R PTH1R
0
response exceeding 0.3 (dashed horizontal line) are shown, and these sensors are shaded in yellow. Each data point represents the average of between 100 and 300 cells measured in one well. (E) Schematic diagram showing the steps for GRAB sensor optimization, including the ICL3 replacement site, linker optimization, and cpEGFP optimization. Black triangles indicate the optimization sites. (F) Optimization of the SST, CCK, CRF, NTS, NPY, and VIP sensors. In each plot, the black dot indicates the initial version after ICL3 transplantation. After optimization, candidates with the highest DF/F0 were selected as the first-generation (1.0) sensors (blue dots). (G) Summary of the peak fluorescence response (top) and maximum brightness (bottom) of the indicated peptide sensors developed by transplanting ICL3 into the indicated class A and class B1 GPCRs (n = 4 wells containing 100 to 300 cells per well). 2 of 16
RES EARCH | R E S E A R C H A R T I C L E
We found that peptide sensors containing the ICL3s derived from GRABNE1m and GRABDA2m—both of which contain relatively long linkers (table S1)—had a higher trafficking index and a larger response than did sensors with ICL3s derived from GRAB5-HT1.0, GRABACh3.0, or dLight1.3b (Fig. 1D). For further development, we chose the more generally applicable ICL3 in GRABNE1m and optimized the peptide sensor prototypes in four steps, including modifying the ICL3 replacing sites, modifying the linker sequences, and modifying critical residues both in the cpGFP module and in the GPCR (fig. S2). The critical residues were chosen from the cpGFP alignment of different cpGFP-based sensors (20) and from the potential interface with cpGFP in the GPCRs (fig. S2). The first three steps are depicted schematically in Fig. 1E, and the complete optimization processes are shown for six peptide GRAB sensors in Fig. 1F, in which the optimal version of the CRF sensor yielded a >10-fold increase in fluorescence upon CRF binding compared with the original candidate (Fig. 1, E and F, and fig. S2). Using the same strategy for both class A and class B1 peptide GPCRs—including the receptors SSTR5 (somatostatin receptor type 5), CCKBR (gastrin/ cholecystokinin type B receptor), NTSR1 (neurotensin receptor type 1), HCRTR2/OX2R [hypocretin (orexin) receptor 2], NPY1R (neuropeptide Y receptor type 1), TACR1 (tachykinin receptor 1), GHS-R (growth hormone secretagogue receptor type 1), CRF1R (corticotropin releasing factor type 1 receptor), CRF2R, VIPR2 (vasoactive intestinal polypeptide receptor 2), and PTH1R (parathyroid hormone 1 receptor)—we then developed and optimized a series of GRAB peptide sensors for detecting somatostatin (SST), cholecystokinin (CCK), neurotensin (NTS), orexin/hypocretin (OX), neuropeptide Y (NPY), substance P (SP), ghrelin (GHRL), corticotropinreleasing factor (CRF), urocortin (UCN), vasoactive intestinal peptide (VIP), and parathyroid hormone–related peptide (PTH) (Fig. 1G). Characterization of GRAB peptide sensors in cultured cells
Next, we characterized the properties of the SST1.0, CCK1.0, CRF1.0, NPY1.0, NTS1.0, and VIP1.0 sensors (table S2); the full amino acid sequences of these sensors are shown in fig. S3. When expressed in HEK293T cells, all six GRAB sensors localized primarily to the plasma membrane and produced a robust change in fluorescence (ranging from a 2.5- to 12-fold increase in fluorescence) in response to their respective ligand (table S2), and each response was blocked by the corresponding GPCR antagonist (fig. S4, A and B). These sensors also retained the ligand selectivity of their respective GPCR scaffolds and had high sensitivity, with apparent half-maximum effective concentrations (EC50) of approximately 10 to 100 nM (fig. S4C and table S2). For example, the CRF1.0 Wang et al., Science 382, eabq8173 (2023)
sensor was based on CRF1R, which has a higher affinity for CRF than for UCN (35). As expected, the CRF1.0 sensor’s EC50 for CRF was 33 nM, compared with 69 nM for urocortin 1 (UCN1), whereas the peptides UCN2 and UCN3 had no effect on the CRF1.0 sensor (fig. S4C2). We then tested the ligand specificity of CRF1.0 and SST1.0 sensors and found that none of the sensors responded to glutamate (Glu), g-aminobutyric acid (GABA), dopamine (DA), or any other neuropeptides tested (Fig. 2E). We also measured the single-photon spectra of these six peptide sensors and found a common excitation peak at ~500 nm and a common emission peak at ~520 nm, with an isosbestic point of the excitation wavelength at ~420 nm (fig. S4E and table S2). The two-photon excitation cross section of the SST and CRF sensors showed excitation peaks at 920 to 930 nm in the presence of the respective ligands (fig. S4F). The kinetics of the peptide sensors’ responses were also measured by locally applying the corresponding peptide ligands and antagonists and then recording the change in fluorescence using line-scan confocal microscopy. The resulting time constants of the rise in the signal (ton) ranged from approximately 0.3 to 0.9 s, and the time constants of the signal decay (toff) ranged from approximately 3 to 12 s (fig. S4D and table S2). Next, we measured the properties of our GRAB peptide sensors expressed in cultured rat cortical neurons. All six neuropeptide sensors, including the SST1.0 and CRF1.0 sensors, localized to the neuronal membrane both at the cell body and in extended ramified neurites and also responded robustly to ligand application (Fig. 2, A and B, and fig. S5). Moreover, when expressed in cultured neurons, the peptide sensors’ responses and apparent EC50 values were measured, and the responses were again blocked by the respective GPCR antagonists (Fig. 2, B and C, and table S3). Finally, for most of the peptide sensors, the ligandinduced change in fluorescence remained stable for up to 120 min in neurons exposed to a saturated ligand concentration (Fig. 2D), indicating minimal internalization of the peptide sensors. We then tested whether our GRAB peptide sensors couple to downstream signaling pathways by measuring G protein–mediated signaling and b-arrestin recruitment. Although wild-type peptide receptors activated both signaling pathways, their corresponding GRAB sensors elicited significantly reduced or virtually no downstream signaling (Fig. 2, F and G). Coexpression of peptide sensors showed no detectable alteration to the affinity and efficacy of corresponding wild-type receptors in response to their ligands (fig. S6, A and B), and no significant differences were observed in neuronal Ca2+ response (fig. S6, C to G). Additionally, RNA sequencing (RNA-seq) analysis shows that GRAB peptide sensors did not alter the
17 November 2023
cellular transcriptome in either cultured neurons or the mouse cortex (Fig. 2H and fig. S7). Furthermore, expressing GRAB peptide sensors exhibits no detectable change in the expression level and localization of native GPCRs (figs. S8 and S9), highlighting that overexpression of peptide sensors does not disrupt endogenous signaling. Our SST, CCK, CRF, NPY, NTS, and VIP sensors were all highly sensitive, specific, and produced a robust real-time increase in fluorescence in response to their corresponding ligands, without activating downstream signaling pathways. We chose the SST and CRF sensors for further study. The SST1.0 sensor can be used to detect the release of endogenous SST in cortical neurons
Neuropeptides are widely used as markers to categorize various types of neurons, with SSTexpressing neurons representing subsets of interneurons in the cerebral cortex (31, 36). Although used as a marker for neuronal subpopulations, whether SST is actually released from cortex neurons—and the spatiotemporal pattern of its potential release—has not been well investigated. Applying trains of electrical field stimuli to cultured mouse hippocampal neurons can induce the fusion of peptidecontaining dense-core vesicles (37, 38). To detect SST release from these neurons, we expressed the SST1.0 sensor in cultured primary rat cortical neurons. Applying increasing numbers of pulse trains elicited increasingly strong responses then reached a plateau (Fig. 3, A to C, and table S4). Application of 75 mM K+ to depolarize the neurons also induced a robust increase in SST1.0 fluorescence that was blocked by the SST receptor antagonist BIM 23056; moreover, no increase in response was measured in neurons expressing the membranetargeted EGFP-CAAX (Fig. 3, A to C, and fig. S10A). The signal was reversible, and the rise and decay half-times of the SST1.0 signal induced by stimulation and K+ application are summarized in fig. S10B. Furthermore, the SST1.0 response was directly correlated with the corresponding increase in cytosolic Ca2+ levels measured using the fluorescent Ca2+ indicator Calbryte-590 (fig. S10C). The SST1.0 sensor can be used to detect glucose-stimulated SST release in isolated pancreatic islets
SST plays an essential role in feeding and energy expenditure by affecting central and peripheral tissues (39). In pancreatic islets, the release of SST from delta (d) cells is critical for regulating the activity of glucagon-releasing a cells and insulin-releasing b cells (40, 41). However, the spatiotemporal pattern of SST release in individual islets has not been investigated. To measure SST release in islets, we expressed SST1.0 under the control of a cytomegalovirus 3 of 16
RES EARCH | R E S E A R C H A R T I C L E
sa l 0 60 12 0
Ba
Luminescence (AU)
CRF1.0
Log10(FPKM+1)
Log10(FPKM+1)
SST1.0
cAMP level (AU)
Ba
Luminescence (AU)
***
Luminescence (AU)
***
17 November 2023
n.s.
D to G). We then examined whether SST1.0 can detect the release of endogenous SST in islets in response to high glucose stimulation (42, 43). Application of 20 mM glucose caused a progressive increase in SST1.0 fluorescence (Fig. 3, D and E, and fig. S10, H to J). Moreover, the increase in SST1.0 fluorescence had a dis-
***
n.s.
n.s.
Wang et al., Science 382, eabq8173 (2023)
(66/3)
Norm. ΔF/F0
Norm. ΔF/F0
sa l 0 60 12 0
Norm. ΔF/F0
Norm. ΔF/F0
***
***
**
***
(CMV) promoter in mouse pancreatic islets cultures using adenovirus infection. Application of the peptide SST-14—but not CCK— caused a robust increase in SST1.0 fluorescence, and this response was blocked by the SST receptor antagonist BIM 23056 but not by the CCK receptor antagonist YM 022 (fig. S10,
(66/3)
ΔF/F0
(72/4)
(115/4)
ΔF/F0
Fig. 2. Characterization of CRF1.0 sensor on culture neurons A1 A2 SST1.0 sensor on culture neurons SST1.0 and CRF1.0 senExpression Response Soma Expression Response Soma sors in vitro. (A) Primary cultured rat cortical neurons expressing SST1.0 (A1) or CRF1.0 (A2), showing sensor expression (left), pseudocolor responses (middle), and cell membrane localization (right). ΔF/F0 ΔF/F0 Scale bars, 100 mm 0 20 0 8 (left) and 20 mm (right). B1 E B2 *** Norm. ΔF/F0 Norm. ΔF/F0 1 μM SST-14 *** 0.3 μM CRF 30 (B) Example fluorescence 9 0 0.5 1 1 0.5 0 traces (left) and summary 6 Saline data (right) of neurons 5 15 1 SST-14 ΔF/F0 ΔF/F0 3 Ctrl CRF expressing SST1.0 (B1) or Ctrl 50 s 50 s UCN2 CRF1.0 (B2); where indicated, 0 0 UCN3 Ctrl BIM Ctrl Anta peptides and antagonists Glu BIM 23056 (10 μM) Antalarmin (10 μM) GABA SST1.0 CRF1.0 were applied (n = 66 to DA 115 ROIs from three or four C1 D1 NE C2 D2 n.s. n.s. 5-HT coverslips). (C) Normalized 1 1 1 HA 1 CRF dose-response curves of ATP SST-14 EC50 ADP neurons expressing SST1.0 0.5 0.5 18 nM 0.5 EC50 0.5 Ado (C1) or CRF1.0 (C2) in 130 nM CCK-8s 0 NPY response to the indicated 0 0 0 NTS ligands; n = 3 cultures VIP -10 -8 -6 -10 -8 -6 each with 20 to 40 ROIs. PACAP [CRF] (LogM) Time (min) [SST-14] (LogM) Time (min) ACTH (D) Summary of the fluoβ-MSH rescence change measured F2 G-protein F1 G-protein G1 β-arrestin γ-MSH G2 β-arrestin GHRH 6 in neurons expressing 6 3 OT 2 CRFR1 SST1.0 (D1) or CRF1.0 (D2) CRFR1 AVP 4 SSTR5 SSTR5 Vasotocin 2 in response to a 2-hour Isotocin 3 1 continuous application of 2 CRF1.0 Glucagon Ctrl Ctrl 1 Ctrl GLP-1 1 mM SST-14 or 300 nM CRF, CRF1.0 Ctrl SST1.0 SST1.0 0 DAMGO 0 0 respectively; n = 4 cultures 0 DynA -16 -13 -10 -8 -6 -4 -10 -8 -6 -10 -8 -6 each with 20 to 40 ROIs. Enk [CRF] (LogM) [SST-14] (LogM) [SST-14] (LogM) [CRF] (LogM) Goserelin (E) Summary of normalized Ghrelin Pearson’s Pearson’s DF/F0 in HEK293T cells Galanin H1 Transcriptome H2 Transcriptome OX-A 4 r = 0.995 4 r = 0.989 expressing SST1.0 (left) or AAV-hSynAAV-hSynSP CRF1.0/ SST1.0/ CRF1.0 (right) in response to NKB 2 2 EGFP-CAAX EGFP-CAAX BB the indicated compounds TPTD applied: SST-14, CRF, UCN2, 0 2-week expression 0 2-week expression Taltirelin 0 2 4 0 2 4 UCN3, CCK-8s, NPY, NTS, Cortex tissue Cortex tissue Log10(FPKM+1) Log10(FPKM+1) CRF1.0 EGFPSST1.0 EGFPVIP, pituitary adenylate RNA-seq CAAX RNA-seq CAAX EGFP-CAAX EGFP-CAAX cyclase–activating polypeptide (PACAP), adrenocorticotropic hormone (ACTH), b-melanocyte-stimulating hormone (b-MSH), assay (F1), a cyclic adenosine monophosphate (cAMP) reporter (F2), and the g-MSH, growth hormone–releasing hormone (GHRH), oxytocin (OT), vasopressin Tango assay (G1 and G2) in cells expressing either the wild-type peptide (AVP), vosotocin, isotocin, glucagon, glucagon-like peptide (GLP-1), DAMGO, receptor (red), sensor (green), or no receptor (Ctrl; gray) in the presence of the dynorphin A (DynA), enkephalin (Enk), goserelin, ghrelin, galanin, orexin A indicated concentrations of the ligand; n = 3 wells each. AU, arbitrary units. (OX-A), SP, neurokinin B (NKB), and bombesin (BB) were applied at 1 mM, while (H) RNA-seq analysis shows that GRAB peptide sensors did not alter the Glu, GABA, DA, norepinephrine (NE), serotonin (5-HT), histamine (HA), ATP, cellular transcriptome. Comparisons of transcriptomes between cortex tissue ADP, adenosine (Ado), teriparatide (TPTD), and taltirelin were applied at 10 mM expressing SST1.0 or EGFP-CAAX (H1) and between CRF1.0 or EGFP-CAAX (n = 4 wells containing 100 to 300 cells per well). (F and G) G protein and (H2). Pearson’s correlation coefficient analysis was used to evaluate the b-arrestin coupling were measured using the split-luciferase complementation differential RNA expression.
tinct spatial pattern within the islet, with regions that could be classified as either nonburst or burst regions (Fig. 3E and movie S1). Analyzing these regions separately revealed that burst regions exhibited a phasic SST1.0 response in the presence of 20 mM glucose, with a higher burst rate and larger peak responses 4 of 16
RES EARCH | R E S E A R C H A R T I C L E
SST1.0
Cortical neurons
3 X Train
10 X Train
75 mM KCl
ROI1 Boutons
ROI1 AAV-hSyn-SST1.0
ROI2
∆F/F0 0.6 0
Elec. stim. or 75 mM KCl
ROI2 Soma
Train stim. @ 50 Hz 1X 2X
…
∆F/F0 1
0.5 s
0
F
3 min
11
23
16
30
Glucose
E
SST1.0
ROI1
ROI1
Non-burst Region
ROI2 ROI2
ROI3
ROI3
I
15
Time
0s
2 min
Burst Region
3
5
40
10
0
-7.5 0
50 100 150
Time (s)
0
Fig. 3. Imaging SST release in cultured neurons and pancreatic islets. (A) (Left) Schematic diagram depicting the experimental strategy. (Middle) Fluorescence image of rat cortical neurons expressing SST1.0. (Right) Pseudocolor images of zoom-in boutons or soma areas bounded by yellow rectangles in the middle image; where indicated, train electrical stimuli (50 pulses delivered at 50 Hz, 0.5-s intertrain interval) or 75 mM K+ were applied. The white arrow indicates soma regions. Scale bars, 100 mm (middle) and 50 mm (right). (B) Example traces of the change in SST1.0 (green) and Calbryte-590 (red) in response to electric stimuli; SST1.0 (with or without antagonist BIM) and EGFP-CAAX fluorescence in response to 75 mM K+; yellow shading indicates the 75 mM KCl perfusion time. (C) Summary of the peak change in fluorescence measured in neurons expressing SST1.0 or EGFP-CAAX in response to burst stimuli or 75 mM K+. (D) (Left) Schematic diagram depicting the experimental strategy in which pancreatic islets were isolated, infected with adenoviruses Wang et al., Science 382, eabq8173 (2023)
17 November 2023
0
2
∆F/F 0
***
***
8 4
7.5
-15
4
G
0
J
0.36 µm2/s
D (µm2/s)
Distance (µm)
H
1 ∆F/F 0
8
(32/3)
Pancreatic islet ADV-SST1.0
Train stim. (n X) 75 mM KCl
20 mM Glucose
3 mM Glucose Time
0 1 3 5 10 16 Ctrl BIM GFP
EGFP-CAAX
Pancreas
n.s.
(30/3)
D
200 s
0.2
**
(55/3)
0.1 / 0.4 ∆F/F 0
20 s
**
(55/3)
SST1.0 Cal-590
16 X
n.s. 0.4
Frequency (burst /10 min)
10 X
5X
C
Amplitude (Peak ∆F/F 0)
1X
75 mM KCl SST1.0 0.1 ∆F/F 0 200 s BIM
Peak ∆F/F 0
B
**
1 0.1
(12/3)
A
0.01
expressing SST1.0, and treated with high (20 mM) glucose. (Right) Example fluorescence images of an SST1.0-expressing pancreatic islet before and after application of 20 mM glucose. Scale bar, 50 mm. (E) SST1.0 fluorescence was measured at the indicated ROIs in the same pancreatic islet shown in (D). On the basis of the response patterns (right panel), ROI1 and ROI2 are classified as nonburst regions (blue), whereas ROI3 is classified as a burst region (red). Scale bar, 50 mm. (F and G) Summary of the burst frequency (F) and peak response (G) measured for nonburst and burst regions; n = 30 to 55 ROIs from three islets. (H) Representative spatial-temporal profile of the SST1.0 fluorescence response measured during a single burst. (I) Example time-lapse pseudocolor images of SST1.0 fluorescence measured in the burst region. The white arrow indicates the location from which the signal originates. Scale bar, 10 mm. (J) Summary of the effective diffusion coefficient (D); note that the y axis is a log scale (n = 12 burst events from three islets). 5 of 16
RES EARCH | R E S E A R C H A R T I C L E
compared with nonburst regions (Fig. 3, F and G). During a single burst event, the SST1.0 signal first increased at a focal hotspot and then propagated over time to neighboring cells (Fig. 3, H and I, and fig. S10K). The response measured near the initial hotspot was more rapid and robust than the responses measured farther away from the hotspot (fig. S10L). Moreover, at the 10 s time point, this propagation of the SST1.0 signal had an average half-width of ~6 mm (fig. S10, M to O), and this half-width increased over time, with an average diffusion coefficient of ∼0.4 mm2/s (Fig. 3J). Detection of endogenous SST release in a Pavlovian conditioning process
SST is widely distributed throughout the central nervous system (44). Previous studies indicate a correlation between endogenous SST levels and learning and memory performance. Mice with SST deficiencies exhibit impaired associative learning (45–47). The basolateral amygdala (BLA) is known as the hub for converging inputs of different modalities during associative learning, with BLA SST interneurons involved in these learning processes (48–50). However, the SST dynamics during the learning process remains unknown. We recorded the SST1.0 signal in mouse BLA during olfactory Pavlovian learning (Fig. 4, A and B). In the early training phase, reward but not punishment unconditional stimulus (US) evoked a transient increase in SST1.0 signals (Fig. 4C). After conditioning by pairing the conditional stimulus (CS; odor) with the US (reward), a fluorescence response developed to the rewardpredictive CS, lasting until reward delivery and then decreasing after the US. In contrast, no observable SST1.0 response was found during “nothing” or “punishment” trials (Fig. 4, C to E). To evaluate whether expression of SST1.0 would affect normal animal behaviors, we recorded body weight and food and water consumption and performed open field test and elevated plus maze test. We found no significant behavioral difference between mice expressing enhanced yellow fluorescent protein (EYFP; control) or SST1.0 at BLA (fig. S14, A to D). Characterization of the CRF1.0 sensor expressed in acute brain slices
CRF is an anxiogenic neuropeptide, and CRF neurons in the central amygdala (CeA) play an important role in several conditions related to fear, anxiety, and alcohol addiction (51–54). To test whether the CRF1.0 sensor can be used to measure the release of endogenous CRF in the CeA, we expressed the CRF1.0 sensor in the CeA and then recorded the response in acute brain slices using two-photon fluorescence microscopy (Fig. 5A). Electric stimuli delivered at 20 Hz induced a robust increase in CRF1.0 Wang et al., Science 382, eabq8173 (2023)
fluorescence, with larger responses induced by increased numbers of pulses, and this response was significantly blocked by treating the slices with the CRF receptor antagonist AHCRF (a-helical CRF) (Fig. 5, B and C); in contrast, no response was measured in slices expressing EGFP-CAAX (Fig. 5C). Additionally, slices expressing the CRF-insensitive mutant (CRFmut), which harbors a point mutation at the ligand-binding pocket (28) (see fig. S13, A to D, for expression in HEK293T cells), showed no observable response compared with CRF1.0 (fig. S11, A to D), and CRISPR-mediated knockout of the Crh gene at CeA lead to significantly reduced CRF1.0 response to electrical stimulation (fig. S11, E to H). To examine whether expression of GRAB peptide sensors would alter the physiological properties of neurons, we compared the calcium signals and the GIRK (G protein–coupled inwardly rectifying potassium) channel currents between sensor or control fluorescent protein–expressing neurons in acute slices (fig. S12). There was no significant difference in the electric stimuli– or high potassium–induced calcium signals (fig. S12, A to G) nor significant alterations to the currentvoltage curves of GABABR agonist baclofen– induced GIRK currents (fig. S12, H to L). The rise and decay half-times increased with increasing pulse numbers, with on and off t50 values of approximately 0.6 to 1.8 s and 3.5 to 6.4 s, respectively (Fig. 5D). Finally, the CRF1.0 signal propagated during electrical stimulation (Fig. 5, E to H) with an average diffusion coefficient of 3.5 × 103 mm2/s (Fig. 5, I and J). The CRF1.0 sensor can be used to measure CRF release in vivo
CRF neurons in the paraventricular nucleus of the hypothalamus (PVN) play an essential role in regulating the stress response via the endocrine axis (8). In addition, these neurons also respond rapidly to both aversive and appetitive stimuli (55–57). To investigate the specificity of our CRF1.0 sensor in vivo, we expressed CRF1.0 or a CRFmut in the mouse PVN. We recorded the signal using fiber photometry while infusing CRF and/or AHCRF through an intracerebroventricular cannula (Fig. 6A). CRF1.0 fluorescence increased in a dose-dependent manner after CRF infusion (Fig. 6B), and the increase was blocked by coadministration of AHCRF (Fig. 6D); in contrast, CRFmut expressed in the PVN showed virtually no response to CRF, even at the highest concentration (Fig. 6C). Next, we measured the dynamics of CRF release in the PVN during stressful experiences in mice expressing CRF1.0 (Fig. 6, E and F). Suspending the mouse by the tail for 30 s induced a robust time-locked increase in CRF1.0 fluorescence, whereas mice expressing CRFmut or EGFP-CAAX in the PVN showed no visible response (Fig. 6, G1 to J1, and fig. S13, E to H).
17 November 2023
Similarly, an intraperitoneal injection of lithium chloride (LiCl), an abdominal malaiseinducing stimulus, but not saline, elicited a long-lasting increase in CRF1.0 fluorescence, whereas no response was measured in mice expressing CRFmut or EGFP-CAAX (Fig. 6, G2 to J2, and fig. S13, E to H; the rise halftimes are shown in fig. S13I). We also observed no significant alteration to animal behaviors, including metabolism, performance in open field test, immobility time in tail suspension and forced swimming tests, and sucrose preference (fig. S14, E to I), indicating that expression of CRF1.0 had no negative effects on animal behaviors. CRF is expressed abundantly in neocortical interneurons, and CRF receptors are present in pyramidal cells (58, 59). In the frontal cortex, CRF mediates stress-induced executive dysfunction (60, 61). We therefore investigated the role of CRF in the mouse cortex during various behavioral paradigms. We injected virus expressing CRF1.0 into the motor cortex and prefrontal cortex (PFC) and then performed two-photon imaging of CRF1.0-expressing layer 2/3 neurons in head-fixed mice (Fig. 7A). We observed a transient reversible increase in CRF1.0 fluorescence in both the motor cortex and PFC in response to tail shocks; in contrast, no response was detected in mice expressing CRFmut or EGFP-CAAX (Fig. 7, B1 to D1 and G, and fig. S15, A to E; the kinetics and time constants are shown in Fig. 7, E and F). Finally, head-fixed mice were forced to run on a treadmill. In response to this stressful stimulus, CRF1.0 fluorescence was monitored using two-photon microscopy (Fig. 7A2). At the onset of forced running, CRF1.0 fluorescence first increased, then reached a plateau within ∼5 s, and finally returned to baseline after the treadmill stopped; in contrast, no response was measured in mice expressing CRFmut or EGFPCAAX (Fig. 7, B2 to D2 and G, and fig. S15, A to E; the kinetics and time constants are shown in Fig. 7, E and F). Discussion
We developed and characterized a series of highly selective and sensitive genetically encoded neuropeptide sensors. Moreover, as proofof-principle, we demonstrated that our SST and CRF sensors can be used to monitor their corresponding peptides in vitro, ex vivo, and in vivo. We used our SST sensor to monitor activity-dependent SST release in cultured cortical neurons as well as pancreatic islets. SST1.0 sensor also revealed the SST dynamic changes in the process of conditioned learning. In acute brain slices, our CRF sensors reliably reported the electrical stimulation evoked release of CRF in the central amygdala. Moreover, the CRF sensor was successfully used to measure in vivo changes in CRF levels in response to stress-inducing stimuli. 6 of 16
RES EARCH | R E S E A R C H A R T I C L E
Trials Time (s) 0
1
2
Nothing
A
Reward
B
Punishment
C
D
4
Trials ΔF/F0 (%)
1
1
1
0
0
0
2
2
2
1
1
1
0
0
0
4 % -1
2
4
6
8
Time (s)
0 0
2
4
6
8
10
0
2
2
4
4
6
8
10
6
Time (s)
8
10
0
2
8
10
Early-training (E)
n.s. 2
Well-learned (W)
2 n.s.
n.s.
6
CS Response
US Response
1
4
Time (s)
0
1
* n.s.
n.s.
0 E W
E W
E W
E W
E W
E W
Time (s)
Time (s)
Fig. 4. Detection of endogenous SST release in a Pavlovian conditioning process. (A) Schematic for fiber photometry recording of SST1.0-expressing neurons from the BLA of a mouse. (B) Schematic for olfactory Pavlovian conditioning tasks. (C) Exemplar time-aligned lick trials, pseudocolor images, and averaged traces from a mouse in early-training and well-learned sessions. (D) Averaged traces of SST1.0 signals in early-training and well-learned sessions (n = 5 mice). (E) Group analysis of the peak DF/F0 of SST1.0 signals to US
The ICL3 with relatively long linkers derived from GRABNE1m, GRABDA2m, empirically showed higher membrane trafficking index and fluorescence response, which may accommodate the folding of TM5, TM6, and cpGFP. Further detailed structural study could help to understand its mechanism. Using our peptide sensors, we observed electrically evoked CRF and CCK (fig. S16) release in acute brain slices and measured their average apparent diffusion coefficients during signal propagation. This signal spread may derive both from the increase of peptide release and from the diffusion of released peptides. Our calculated diffusion coefficients are relatively higher than those Wang et al., Science 382, eabq8173 (2023)
0
E
Nothing Reward Punish.
0.5
0
10
US
1
0.5
US
Trials
2
CS
1
CS
2
Well-learned (W)
US
Punishment
US
2
0
Early-training (E) CS
ΔF/F0 (%)
US
Reward CS
4 % -1
Peak ΔF/F0 (%)
Odor CS
Well-learned (W)
BLA
US
CS
Early-training (E)
AAV-hSyn-SST1.0
B
Nothing
C
Fiber photometry
ΔF/F0 (%)
A
and CS in different sessions. Each trace represents data from one animal (n = 5 mice); StudentÕs t test, n.s., not significant; *P < 0.05. (US responses between early-training and well-learned: P = 0.6336 in nothing trial, P = 0.0184 in reward trial, and P = 0.8859 in punishment trial; CS responses between early-training and well-learned: P = 0.6517 in nothing trial, P = 0.0900 in reward trial, and P = 0.3499 in punishment trial.) Values with error bars indicate mean ± SEM.
of glutamate in the synaptic cleft (~330 mm2/s) (62), dopamine in the rat brain (~68 mm2/s) (63), and GFP-tagged tissue plasminogen activator (~0.02 mm2/s) (64) estimated by other methods. Further studies could apply optogenetic and chemogenic tools to drive the release from peptidergic neurons. The development and optimization of red sensors for neuropeptides will facilitate research of this field, which could be applied in multiplexed dual-color recording with green sensors in the future. By combining these GRAB peptide sensors with neurotransmitter sensors, it may be possible to monitor the real-time release of both neuropeptides and neurotransmitters, providing new insights into
17 November 2023
the mechanisms and functions of neuropeptide corelease. In addition to its use in vitro in cultured neurons, we also measured the endogenous SST release in isolated pancreatic islets, consisting of cell types that secrete glucagon and insulin to maintain blood glucose levels (40). The finest temporal resolution of pulsatile SST release measured in previous studies was on the order of 30 s (42, 43). Using our SST sensor, we measured changes in SST levels in response to high glucose at the single-cell level with high temporal resolution on the order of seconds. SST released from d cells functions as a paracrine regulator to integrate signals from 7 of 16
RES EARCH | R E S E A R C H A R T I C L E
A AAV-hSyn-
B
CRF1.0/ EGFP-CAAX
Brain slices CRF1.0
Stim. @ 20 Hz
Baseline
CeA
1p
5p
20 p
100 p
100 p
AHCRF
ROI diameter 100µm
∆F/F0 50%
Electrode
Pulses
CeA
Stim. @ 20 Hz, 1 ms/ pulse
5s
10
ACSF AHCRF (100 p)
0
EGFP-CAAX (100 p) CRF1.0 100
Stim. @ 20 Hz, 100 p ∆F/F 0 10%
Off t50 ~9.1 s On t50 5s
~2.3 s
9
Stim. @ 20 Hz
6 3 0
on off on off on off on off
Pulses 10
Pulse number
20
50
100
Stim. 100 pulses, 5 s 0s
0.25
0.5
0.75
1
2
3
5
6
10
20 ∆F/F0 40%
0 µm
0
10
∆F/F0 20%
-50
0
10
20
Time (s)
30
0
0 -10
0
10
20
Time (s)
30
Fig. 5. Detection of endogenous CRF release in acute brain slices using CRF1.0. (A) Schematic illustration depicting the experimental design in which CRF1.0 or EGFP-CAAX was expressed virally in the CeA; after 3 weeks, acute slices were prepared. (B) Two-photon fluorescence images of acute slices, showing expression of CRF1.0 (left). Example pseudocolor images of acute slices expressing CRF1.0 at baseline and in response to 1, 5, 20, and 100 electric stimuli [i.e., pulses (P)] delivered at 20 Hz, and the response to 100 pulses measured in the presence of 100 nM AHCRF. The dashed white circles indicate the ROI used to calculate the response, and the approximate position of the stimulating electrode is indicated. Scale bars, 100 mm. (C) Representative traces (left) and summary (right) of the change in CRF1.0 fluorescence in response to electric stimuli delivered at 20 Hz in ACSF and 100 pulses delivered in the presence of AHCRF; also shown is the response measured in slices expressing EGFP-CAAX
ghrelin, dopamine, acetylcholine, and leptin (65). Moreover, pancreatic islets receive regulatory input that affects Ca2+ fluctuations in a and b cells. These fluctuations are subsequently translated into the appropriate release of glucagon and insulin (66). Thus, our SST sensor and other hormone and/or transmitter sensors, such as ghrelin, UCN3, DA, and adenosine triphosphate (ATP) sensors, can be combined with Ca2+ indicators to study pancreatic islets Wang et al., Science 382, eabq8173 (2023)
I
20
1s 2s 5s 10 s
10
0
0
50
100
Distance (μm)
4 0
10
0 1
2
3 4
Time (s)
5
3.5
5
0 CRF1.0
(n = 3 to 6 slices from one to three mice). (D) Fitted curves (top) and summary (bottom) of on and off t50 of the change in CRF1.0 fluorescence; n = 2 to 6 slices. (E) Example time-lapse pseudocolor images of CRF1.0 expressed in the CeA; during the first 5 s, 100 pulses were delivered at 20 Hz. Scale bar, 100 mm. (F to H) Spatial-temporal profile (F), temporal dynamics (G), and spatial dynamics (H) of the fluorescence change shown in (E). The profile in (F) shows the average response of three trials conducted in one slice. The traces in (G) and (H) correspond to the indicated distances and times, respectively, and the data in (H) were fitted with a Gaussian function. (I) Square of the full width at half maximum (FWHM2) plotted against time on the basis of the data shown in (H); the diffusion coefficient (D) was measured as the slope of a line fitted to the data. (J) Summary of the diffusion coefficient (D) measured CRF in the CeA; n = 6 slices from three mice.
in healthy conditions and in diabetic animal models. Finally, our in vivo experiments show that these sensors can be used to directly monitor neuropeptide release within specific brain regions during behaviors, supporting their utility in freely moving animals. Although the peptideexpressing cortical neurons are well established (67), it remains unclear whether these peptides are released in a behaviorally relevant manner.
17 November 2023
8
J
Diffusion Coefficient D = 1.6 x 103 µm2/s
(x 103 µm2/s)
10 µm 20 µm 50 µm 100 µm
Diffusion Coefficient
0
H
Stim. @ 20 Hz
∆F/F 0 (%)
50
-100 -10
20
ΔF/F0 (%)
G
100
CRF1.0 Distance (µm)
F
FWHM2 (x103 µm2)
-100
(6/4)
E
(AHCRF) (EGFP-CAAX)
∆F/F 0 5%
100 20 5 1
D
*** ***
20
Peak ∆F/F 0 (%)
2PM
Electrode
Stim. @ 20 Hz
t50 (s)
C
0
1 2 5 10 20 50 100 100 100
3 weeks later
In addition to the axonal release, neuropeptides can also be released from large dense-core vesicles in the somatodendritic compartment, likely contributing to volume transmission and exerting their function through paracrine modulation (37, 68, 69). The CRF1.0 sensor exhibits similar signals when expressed in different cell types in brain slices (fig. S17). Different cell types may have different regulations of neuropeptide release. These peptide sensors may be helpful 8 of 16
RES EARCH | R E S E A R C H A R T I C L E
Optic fiber
AAV-hSynCRF1.0/ CRFmut
PVN
C
CRF1.0 2 1.5
0.5
** **
0
1000
2000
1
0
G1 Tail Lift
Time (s)
H1
AAV-hSyn-CRF1.0/ CRFmut
ΔF/F0 ΔF/ 5%
470-nm LED
0.5
I1
0.03
CRF1.0 CRFmut
0.02
n.s.
0 0
G2
H2
i.p. injection
Fiber Track
I2
i.p. injection ΔF/F0 5%
PVN
0.25
10 min
0.1
0.5
0.05
LiCl
Saline
Fig. 6. Using fiber photometry to measure endogenous CRF release in vivo. (A) (Left) Schematic diagrams depicting the strategy for virus injection, fiber and cannula implantation, and measurement of CFR1.0 or CRFmut in the PVN. (Right) image showing the expression of CRF1.0 (green) in the PVN and the approximate location of the optic fiber above the PVN; the nuclei were counterstained with 4′,6-diamidino-2-phenylindole (DAPI) (blue). Scale bar, 200 mm. (B to D) Traces (left panels) and summary of the response (right panels) measured in mice expressing CRF1.0 [(B) and (D)] or CRFmut (C); the indicated concentrations of CRF and a-helical CRF 9-41 (AHCRF) were infused through
to shed light on the understanding of the peptide release mechanism in the future. Although sensor fluorescence does not directly represent endogenous receptor activation, when and where these neuropeptides are released can nonetheless be examined using these sensors, thus helping elucidate their regulatory role on neural circuits. Our peptide GRAB sensors were designed to take advantage of the native peptide receptors, inheriting their high selectivity and sensitivity.
-0.1
0.1
*
0.05
LiCl
0
Saline 0
10
0
-0.05 20
Time from injection (min)
Saline LiCl
the cannula. (E) Schematic diagram depicting the strategy for virus injection and fiber photometry recording. (F) Image showing the expression of CRF1.0 (green) and the approximate location of the imaging fiber; the nuclei were counterstained with DAPI (blue). Scale bars, 300 mm (left) and 40 mm (right). (G to J) Illustration (G), representative traces (H), average traces per stimulusresponse (I), and summary data (J) of the change in CRF1.0 and CRFmut fluorescence measured before and during a 30-s tail lift (G1 to J1) and before and after an intraperitoneal injection of LiCl or saline (G2 to J2); n = 3 to 6 animals.
For example, CRF1.0 and SST1.0 sensors showed selectivity profiles similar to those of their native receptors, CRF1R and SSTR5 (Fig. 2E and fig. S4C). We further validated the selectivity of CRF1.0 by CRISPR-mediated knockout of the corresponding Crh gene in mice (fig. S11, E to H), and the provisional selectivity of SST1.0 could be validated by SST deletion experiment. Peptides bind to receptors with high affinity and potency, with the median inhibition constant (Ki) and EC50 at nanomolar range, targeting class
17 November 2023
J2
i.p. injection
-0.05
LiCl/ Saline
Wang et al., Science 382, eabq8173 (2023)
0.04
Time from lift (min)
ΔF/F0
CRF1.0 DAPI
*
0.08 0.06
0
F
J1
Tail Lift
0.09
PVN
1
0 CRF 0 1.6 1.6 0 200 400 600 AHCRF 0.25 0.25 0 Time (s) (mg/mL)
0
4.8
*
0.5
n.s. 0
*
1.5
0 0.25 1.6 0.25 1.6 0
0.06
1 min
2
CRF AHCRF
(mg/mL)
CRF (mg/mL)
CRF1.0 CRFmut
Tail Lift
Photoreceiver
1
1
0 0 200 400 600
CRF (mg/mL)
1.5
0.5
0
0 0.05 0.5 1.6 4.8
2 1.5
0.5
Time (s)
E
4.8 0
1
0.5 0
CRF(mg/mL)
ΔF/F0
1
**
1.5
CRF1.0
ΔF/F0
4.8 1.6 0.5 0.05 0
D
CRFmut
ΔF/F0
CRF(mg/mL)
ΔF/F0
ΔF/F0
1.5
** **** **
CRF1.0 PVN P PV VN DAPI
C57BL/6 male
ΔF/F0
PVN
ΔF/F0
Optic fiber
B
Fiber track
Cannula
ΔF/F0
Photoreceiver 390 Hz 535 nm 470 nm LED Dichroic
ΔF/F0
A
A GPCRs (10, 70). The difference in sensor EC50 values for HEK cells and neurons may be due to the differences in lipid and cholesterol compositions of cell membranes. Although the sensorsÕ affinity remains lower than that of native receptors at present (table S3), it is sufficient to detect endogenous neuropeptide changes ex vivo and in vivo (Figs. 4 to 7). Our series dilution experiment further suggests that the virus titer is not a major factor in determining the sensor fluorescence change upon 9 of 16
RES EARCH | R E S E A R C H A R T I C L E
A1
B1
Expression
Before
Tail shock
C1
After
Motor Cortex
AAV-hSyn-CRF1.0 /CRFmut
3
CRF1.0
PFC Motor Cortex
2
3 4
2
PFC
1
1
4
2 3
3
CRFmut
∆F/F0
120 s Tail shock: 0.7 mA, 3 s
A2
B2
Motor Cortex
1
0.5
2
4
3
Before
C2 2
CRF1.0
4
3 4
PFC 4
1
2
2
1
3
0.05 ∆F/F0
0.5
PFC 2 4
1
Running start
Running stop
0.05 ∆F/F0 10 s
Off t50 ~7 s
Fig. 7. Spatially resolved measurements of CRF in vivo using two-photon imaging. (A to D) Schematic diagrams (A), representative expression and pseudocolor response images (B), representative traces measured at the indicated ROIs (C), and average traces per stimulus (D) measured in the head-fixed mice expressing CRF1.0 or CRFmut (B1 to D1, top and bottom panels, respectively). The mice were subjected to tail shock (A1 to D1) and forced running on a treadmill (A2 to D2),
neuropeptide binding given sufficient incubation time, and the titer is within a reasonable range (3 × 1012 to 1013 genome copies per milliliter in our case) (fig. S13, J to M). Improving neuropeptide sensors to achieve picomolar sensitivity in the future will be the key to cap-
0.05 ∆F/F0 30 s
F
4 -0.1
Tail shock
0.5
G
Forced running
10 5 0
0.3
Tail shock
CRF1.0 0.2 *** CRFmut 0.1
***
n.s.
Forced running **
CRF1.0 EGFP-CAAX ***
n.s. n.s.
n.s.
0 onoff on off on off on off Motor PFC Motor PFC
Stim. - + - + - + - + Motor PFC
-+ -+
-+ -+
Motor PFC
and two-photon imaging was performed in the motor cortex and PFC. Scale bars, 100 mm. (E and F) Representative traces (E) and summary (F) of the rise and decay t50 values of the CRF1.0 signal in response to tail shock and forced running; n = 10 to 12 trials from three mice. (G) Summary of the peak fluorescence response measured in the motor cortex and PFC in mice expressing CRF1.0, CRFmut, or EGFPCAAX in response to tail shock and forced running; n = 3 to 7 mice each.
ture neuropeptide release under diverse natural conditions. The “on” kinetics of GRAB peptide sensors are 300 to 400 ms and “off ” kinetics are 3 to 12 s, which is in a similar range to that of naïve peptide receptors. There is relatively
17 November 2023
0.1 ∆F/F0
3
t50 (s)
On t50 ~2.4 s
10 s
Wang et al., Science 382, eabq8173 (2023)
30 s
2
3
∆F/F0
Off t50 ~7 s
4
1
15
On t50 ~0.4 s
-0.1
∆F/F 0
EGFP-CAAX
∆F/F0
Tail shock
Averaged response Forced running
D2
Forced running
1
3
3
E
30 s
0.5
After
Forced running
2
2-P imaging
69 s 0s Speed: 15 cm/s
0.1 ∆F/F0
4 -0.1
1
Forced running
0.1 ∆F/F0
1
2
3
Motor Cortex
AAV-hSyn-CRF1.0 /EGFP-CAAX
Motor Cortex
Expression
4 -0.1
30 s
∆F/F0
PFC
Averaged response
2
4
1
2-P imaging
0s
D1
Tail shock
ROI 1
limited information for the kinetics of native peptide GPCRs (71, 72). The ton and toff time constants of the CRF receptor are 0.1 s (at 1 mM concentration) and 142 min, respectively (73). A time constant of ~1 s for the parathyroid hormone receptor was measured 10 of 16
RES EARCH | R E S E A R C H A R T I C L E
by intramolecular fluorescence resonance energy transfer in vivo (74). The ton and toff time constants of gonadotropin-releasing hormone receptor are 1 to 20 s (at 1 mM concentration) and 2 to 6 min, respectively (75). Given the lack of direct kinetics data of native peptide GPCRs, we also compared the reported equilibrium dissociation constant (Kd) of native peptide receptors with that of sensors. The Kd values of high- and low-affinity CRF binding sites are approximately 0.2 and 20 nM, respectively (73). The EC50 of CRF1.0 sensor is 18.6 ± 1.6 nM, indicating that there is not much difference in sensors’ EC50 with native receptors’ Kd values. The kinetics of GRAB peptide sensor’s signal reflect the GPCR structure change upon ligand binding, which is at the first level of signal transduction. It would be even slower if the measurement was at secondary effector levels that require activation of cellular signaling cascades. Neuropeptides bind to endogenous GPCRs and transduce signals. If peptide sensors recapitulate the signal transduction or compete the ligand binding with native receptors, the endogenous signal fidelity and normal animal behavior could be affected when using peptide sensors. Most of our peptide sensors show minimal downstream coupling; for example, the SST1.0 sensors exhibit virtually no coupling (Fig. 2, F and G), suggesting that the expression of these peptide sensors will not affect the normal functions of cells. However, the CRF1.0 sensor still shows significant coupling, albeit with orders of magnitude lower affinity and 60% reduced efficacy (Fig. 2, F and G). The structures of peptide GPCRs bound to G proteins and b-arrestin have been solved, and the interaction sites have been identified (28, 76); altering these sites in the CRF1.0 sensor will allow future modifications to further reduce downstream coupling. At cellular signaling levels, expressing peptide sensors showed no obvious alterations to the cellular transcriptome (Fig. 2H and fig. S7), neural calcium activity in culture neurons or acute slices (figs. S6 and S11). At behavior outcome levels, we found no significant changes between mice expressing peptide sensors or control fluorescence proteins (fig. S14). In summary, this series of newly generated GRAB peptide sensors can be used both in vitro and in vivo to monitor the rate and range of peptide release with a high spatiotemporal resolution. These tools have the potential to advance our understanding of the roles of neuropeptides in health and disease.
transfecting cells with pCS7-PiggyBAC (S103P, S509G) (77) together with vectors containing a 5′ PiggyBac inverted terminal repeat sequence (ITR), CAG promoter, the GRAB peptide sensor coding region, internal ribosomal entry site (IRES) sequence, a puromycin-encoding gene, and a 3′ PiggyBac ITR; 24 hours after transfection, the cells were selected by culturing in 1 mg/ml puromycin. The HTLA cell line for the Tango assay was a gift from B. L. Roth (78). All cell lines were cultured in Dulbecco’s modified Eagle’s medium (Biological Industries, 06-1055-57-1ACS) supplemented with 10% (v/v) fetal bovine serum (FBS; CellMax, SA301.02) and 1% (v/v) penicillinstreptomycin (Gibco, 15140122) at 37°C in humidified air containing 5% CO2.
Materials and methods Cell lines
Molecular cloning was conducted using the Gibson assembly method. Primers for Gibson assembly were synthesized by Tsingke Biotechnology Co., Ltd., with 30–base pair overlap. The coding sequences for the GPCRs were polymerase chain reaction (PCR)–amplified from the
HEK293T cells (CRL-3216, ATCC) were used to generate cell lines stably expressing the CRF1.0, SST1.0, CCK1.0, NPY1.0, NTS1.0, and VIP1.0 sensors. These stable cell lines were generated by Wang et al., Science 382, eabq8173 (2023)
Cultured rat primary cortical neurons
Rat cortical neurons were obtained from postnatal day 0 (P0) Sprague–Dawley rat pups of both sexes (Beijing Vital River Laboratory Animal Technology Co., Ltd.). In brief, the brain was removed, and the cortex was dissected, dissociated in 0.25% trypsin-EDTA (Gibco, 25200056), and plated on glass coverslips precoated with poly-D-lysine hydrobromide (Sigma, P7280). The neurons were cultured in Neurobasal medium (Gibco, 21103049) supplemented with 2% B-27 (Gibco, A3582801), 1% GlutaMAX (Gibco, 35050061), and 1% penicillin-streptomycin (Gibco, 15140122) at 37°C in humidified air containing 5% CO2. Mice
C57BL/6N mice of both sexes (6 to 8 weeks of age and 10 to 12 weeks of age) were obtained from Beijing Vital River Laboratory Animal Technology Co., Ltd., and group-housed (up to five mice per cage) under a 12-hours/12-hours light/dark cycle with the ambient temperature maintained at 25°C. CaMKIIa-Cre mice (JAX Strain 005359), Vglut2-Cre mice (JAX Strain 028863), and Gad2-Cre mice (JAX Strain 019022) were obtained from Jackson Laboratory. All surgical and experimental protocols were approved by the Animal Care and Use Committee at Peking University, the University of Science and Technology of China, New York University, the Institute of Neuroscience, and the Chinese Academy of Sciences and were performed in accordance with the standards established by the Association for the Assessment and Accreditation of Laboratory Animal Care. Detailed information about the sex and littermate status of the mice used is given below, in the section pertaining to each experiment. Molecular biology
17 November 2023
corresponding full-length human GPCR cDNAs (hORFeome database 8.1) using GoldenStar T6 DNA Polymerase (Tsingke, TSE102). The ICL3 from the GRAB-NE (19), GRAB-DA (33), GRABACh (20), GRAB-5-HT (34), and dLight (21) sensors were PCR-amplified from the corresponding sensors. Chimeric GPCRs and GRAB sensors were cloned into the modified pDisplay vector (Invitrogen) with an upstream immunoglobulin K-chain leader sequence and followed by an IRES sequence and mCherry-CAAX. Sanger sequencing was performed to verify the sequence of all clones. GPCR/Sensor-SmBit was constructed from b2AR-SmBit, and LgBit-mGs/ mGsi/mGsq was a gift from N. A. Lambert (79). The GRAB peptide sensors were cloned into the pAAV vector under the control of the human Synapsin promoter and used for AAV packing. Transfection of cell lines and virus infection of primary cultures
HEK293T cells and HTLA cells at 50 to 60% confluency were transfected with a mixture of polyethylenimine (PEI) and plasmid DNA at a 3:1 (w/w) ratio; after 6 to 8 hours, the transfection reagent was replaced with standard culture medium, and the cells were cultured for an additional 24 to 36 hours for expression of the transfected plasmids. AAV9 viruses expressing the indicated GRAB peptide sensors were packaged at WZ Biosciences and BrainVTA (Wuhan) Co., Ltd. Each virus [at a titer of 3 × 1013 to 5 × 1013 viral genomes per milliliter (vg/ml) was added to cultured rat cortical neurons at DIV5-7, and the neurons were imaged 7 to 10 days later. Fluorescence imaging of cultured cells and primary neurons
HEK293T cells and primary neurons were imaged using a Ti-E A1 inverted confocal microscope (Nikon) and an Opera Phenix HighContent Screening System (PerkinElmer). The confocal microscope was equipped with a 10×/0.45 numerical aperture (NA) objective, a 20×/0.75 NA objective, and a 40×/1.35 NA oil-immersion objective. A 488-nm laser and 525/50-nm emission filter were used to image green fluorescence, and a 561-nm laser and 595/50-nm emission filter were used to image red fluorescence. Cells were cultured on glass coverslips in 24-well plates and imaged in a custom-made chamber. The Opera Phenix system was equipped with 20×/1.0 NA and 40×/ 1.15 NA water-immersion objectives. A 488-nm laser and 525/50-nm emission filter were used to image green fluorescence, and a 561-nm laser and 600/30-nm emission filter were used to image red fluorescence. Cells were cultured and imaged in CellCarrier Ultra 96-well plates (PerkinElmer). The cells were imaged in Tyrode’s solutions containing (in millimolar concentrations): 150 NaCl, 4 KCl, 2 MgCl2, 2 CaCl2, 10 HEPES, 11 of 16
RES EARCH | R E S E A R C H A R T I C L E
and 10 glucose (pH adjusted to between 7.35 and 7.45 with NaOH). Where indicated, the following compounds were applied to the cells in Tyrode’s solution by bath application or a custommade perfusion system: SST-28 (Anaspec), SST-14 (Anaspec), CCK-8s (Abcam), CCK-4 (Abcam), CRF (Anaspec), UCNI (MedChemExpress), UCNII (MedChemExpress), UCNIII (Abcam), NTS (Anaspec), NPY (Abcam), VIP (Anaspec), PACAP(1-38) (MedChemExpress), PACAP(1-27) (MedChemExpress), Orexin-B (GL Biochem), Substance P (Tocris), Ghrelin (Tocris), teriparatide [human parathyroid hormone-(1-34)] (MedChemExpress), Glu (Sigma-Aldrich), GABA (Tocris), DA (Sigma-Aldrich), NE (Tocris), 5-HT (Tocris), HA (Tocris), ATP (Tocris), ADP (MedChemExpress), Ado (Tocris), ACTH (MedChemExpress), b-MSH (MedChemExpress), g-MSH (MedChemExpress), GHRH (Anaspec), OT (Anaspec), AVP (Tocris), vasotocin (MedChemExpress), isotocin (KS-V peptide), glucagon (GLPBIO), GLP-1 (MedChemExpress), DAMGO (Tocris), dynorphin A (Tocris), [Leu5]-enkephalin (MedChemExpress), goserelin (MedChemExpress), galanin (Tocris), orexin-A (Tocris), NKB (Tocris), bombesin (MedChemExpress), taltirelin (MedChemExpress), BIM23056 (Abcam), YM 022 (Tocris), NBI 27914 (Santa Cruz), antalarmin (Cayman), a-helical CRF (Tocris), SR142948 (Tocris), BIBO 3304 (Tocris), and PACAP(6-38) (Tocris). For high K+ stimulation, Tyrode’s solution contained 79 mM NaCl and 75 mM KCl. For screening candidates using SSTR5, NPY1, NPY5, GHS-R, AVPR2, NTSR1, CCKBR, HCRTR2 (OX2), OPRM1, GRPR, TACR1 (NK1), TRHR, VIPR1, VIPR2, CRF1R, or PTH as scaffolds in Fig. 1, the following compounds were applied respectively (in micromolar concentrations): 1 SST-28, 1 NPY, 1 NPY, 1 ghrelin, 5 desmopressin (Tocris), 1 NTS, 1 CCK-8s, 1 orexin-B, 1 DAMGO (Tocris), 10 bombesin (Tocris), 10 substance P, 20 taltirelin (Tocris), 1 VIP, 1 VIP, 1 CRF, and 1 teriparatide. Neuron cultures were incubated in culture medium with 5 mM Calbryte 590 AM (ATT Bioquest) at 37°C for 30 min before calcium imaging. Field electric stimuli were delivered by GRASS S88 stimulator. The pulse duration is 1 ms, 50 pulses were delivered at 50 Hz for 1 s, which are considered as 1× burst stimulation. The time interval between each burst is 0.5 s. Spectra measurements
The linear optical properties of the GRAB peptide sensors expressed in HEK293T cells were measured using a Safire 2 plate reader (Tecan). Cells were harvested and transferred to blackwall 384-well plates containing either saline alone or saline containing the corresponding peptides. Emission spectra were measured using an excitation wavelength of 455 nm with a bandwidth of 20 nm, and emissions were collected using an emission wavelength step size of 5 nm. Excitation spectra were measured using excitation light ranging from 300 to 520 nm with a Wang et al., Science 382, eabq8173 (2023)
wavelength step size of 5 nm, and emission light was collected at 560 nm with a bandwidth of 20 nm. The two-photon fluorescence spectra of the GRAB peptide sensors expressed in HEK293T cells were measured at 10-nm increments from 700 to 1050 nm using a Bruker Ultima Investigator two-photon microscope equipped with Spectra-Physics Insight X3. Cells were measured in Tyrode’s solutions or Tyrode’s solutions containing the corresponding peptides. The two-photon laser power at various wavelengths was calibrated, and the fluorescence measured in untransfected cells was subtracted as background. Tango assay
HTLA cells were cultured and transfected in six-well plates and placed in 96-well plates (white with a clear flat bottom), and solutions containing various concentrations of peptides were applied; 12 hours after induction, the medium was discarded, and 40 ml of Bright-Glo Luciferase Assay Reagent (Promega) diluted 20-fold in phosphate-buffered saline (PBS) was added to each well at room temperature. After a 10-min reaction in the dark, luminescence was measured using a Victor X5 multi-label plate reader (PerkinElmer). Mini G protein luciferase complementation assay
HEK293T cells were cultured and transfected in six-well plates and grown to between 80 and 90% confluency. The cells were then dissociated using a cell scraper, resuspended in PBS, and placed in 96-well plates (white with a clear flat bottom) containing Nano-Glo Luciferase Assay Reagent (Promega) diluted 1000-fold in PBS at room temperature. Solutions containing various concentrations of peptides were added to the wells. After a 10-min reaction in the dark, luminescence was measured using a Victor X5 multi-label plate reader (PerkinElmer). Pancreatic islet isolation and imaging of SST1.0 sensor
Male C57BL/6N mice (10 weeks of age) were obtained from Beijing Vital River Laboratory Animal Technology Co., Ltd. The mice were sacrificed by cervical dislocation, and primary pancreatic islets were isolated using collagenase P digestion and purified by hand-picking under a dissecting microscope. After isolation, the islets were cultured overnight in RPMI1640 medium containing 10% FBS (10099141C, Gibco), 8 mM D-glucose, 100 units/ml penicillin, and 100 mg/ml streptomycin for overnight culture at 37°C in a 5% CO2 humidified air atmosphere. Adenovirus (ADV) expressing the SST1.0 sensor (pAdeno-MCMV-SST1.0) was prepared by OBiO Technology (Shanghai) Corp., Ltd. The islets were infected with pAdeno-MCMV-SST1.0 by 1 hour exposure in 200 ml culture medium
17 November 2023
(∼4 × 106 plaque-forming units (PFU) per islet), followed by addition of regular medium and further culture for 16 to 20 hours before use. All fluorescence images were acquired using Dragonfly 200 series (Andor) with a Zyla4.2 sCMOS camera (Andor) and the Fusion software. All channels were collected with a 40×/ 0.85 NA Microscope Objective (Warranty Leica HCX PL APO). The SST concentrations in culture medium were measured according to the protocol of the Somatostatin (Human, Rat, Mouse, Porcine)–EIA Kit (Phoenix Pharmaceuticals, EK-060-03). Briefly, batches of five islets were incubated in Krebs-Ringer buffer containing 3 or 20 mmol/liter glucose for 0.5 hours. At the end of incubation, the media were removed for enzyme-linked immunosorbent assay. Fluorescence imaging of peptide sensors in acute brain slices
Male C57BL/6N mice (6 to 8 weeks of age) were anesthetized with an intraperitoneal injection of tribromoethanol (Avertin; 500 mg/kg body weight), and the AAV9-hSyn-CRF1.0, AAV9hSyn-CCK1.0, AAV9-hSyn-EGFP-CAAX, or AAV9CMV-saCAS9-sgRNA virus (300 nl, 3 × 1013 to 5 × 1013 vg/ml, WZ Biosciences), AAV9-hSynCRFmut, hsyn-NES-jRGECO1a (300 nl, 3 × 1012 vg/ml, BrainVTA) virus was injected into the left CeA (AP: −1.2 mm relative to Bregma; ML: −2.5 mm relative to Bregma; DV: −4.4 mm from the dura) or the left CA1 (AP: −2.0 mm relative to Bregma; ML: −1.5 mm relative to Bregma; DV: −1.5 mm from the dura) at a rate of 30 nl/min. After 3 weeks, to allow for virus expression, the mice were anesthetized with Avertin and perfused with ice-cold oxygenated slicing buffer containing (in millimolar concentrations): 110 choline-Cl, 2.5 KCl, 7 MgCl2, 1 NaH2PO4, 0.5 CaCl2, 25 NaHCO3, and 25 glucose (pH 7.4). The brains were dissected, and 300-mm-thick coronal slices were cut in icecold oxygenated slicing buffer using a VT1200 vibratome (Leica). The slices were transferred and allowed to recover for at least 40 min at 34°C in oxygenated artificial cerebrospinal fluid (ACSF) containing (in millimolar concentrations): 125 NaCl, 2.5 KCl, 1.3 MgCl2, 1 NaH2PO4, 2 CaCl2, 25 NaHCO3, and 25 glucose (pH 7.4). The brain slices were then transferred to a custom-made perfusion chamber and imaged using an FV1000MPE twophoton microscope (Olympus) or Bruker two-photon microscope. CRF1.0, CCK1.0, and EGFP-CAAX were excited using a 920-nm two-photon laser, and dual-color imaging used a 950-nm two-photon laser for excitation, and electrode tips were placed near the CeA or CA1 region expressing CRF1.0, CCK1.0, or EGFPCAAX. Electrical stimuli were applied using an S88 stimulator (Grass Instruments), with a stimulation voltage of 5 to 8 V and pulse duration of 1 ms. For CRF1.0 imaging experiments in acute brain slices, the electrode was placed on 12 of 16
RES EARCH | R E S E A R C H A R T I C L E
the surface of the brain slice, and the averaged depth of imaging regions was 52 ± 5 mm below the surface, which is calculated by aligning imaging regions to z-axis scanning images. The gRNA sequences for Crh knockout or control scramble gRNA are as follows: sa1: CCTCAGCCGGTTCTGATCCGC; sa2: GAAGAATACTTCCTCCGCCTG; sa3: GAGCCGGCCGAACGCGGCGCC; sa4: CCCAACTCCACGCCCCTCACC; sa5: CGCGACCCGCGGTGAGGGGCG; and saCtrl: GTGTAGTTCGACCATTCGTG. Two-photon in vivo imaging in mice
Female C57BL/6N mice (6 to 8 weeks of age) were anesthetized with Avertin, and AAV9-hSynCRF1.0, AAV9-hSyn-CRFmut, or AAV9-hSynEGFP-CAAX (200 nl, full titer, WZ Biosciences) was injected into the motor cortex (AP: 1.0 mm relative to Bregma; ML: 1.5 mm relative to Bregma; DV: −0.5 mm from the dura) and PFC (AP: 2.8 mm relative to Bregma; ML: 0.5 mm; DV: −0.5 mm from the dura). A high-speed drill was then used to open a 4 mm by 4 mm square in the skull. After virus injection, craniotomies were installed with a glass coverslip affixed to the skull surface. A stainless-steel head holder was attached to the animal’s skull using dental cement to help restrain the animal’s head and reduce motion-induced artifacts during imaging. The imaging experiments were performed ∼3 weeks after surgery. An awake mouse with head mounts was habituated for 10 min in the treadmill-adapted imaging apparatus to minimize the stress associated with head restraint and imaging. The motor cortex or PFC was imaged 100 to 200 mm below the pial surface to measure sensor fluorescence. A Bruker Ultima Investigator two-photon microscope equipped with Spectra-Physics Insight X3 was used for in vivo imaging. A 920-nm laser was used for excitation, and a 490- to 560-nm filter was used to measure green fluorescence. All experiments were performed using a 16×/0.8 NA objective immersed in saline, and images were acquired at a frame rate of 1.5 Hz. For the forced running model, the running speed was set at ∼15 cm/s; for the tail shock model, a 0.7-mA shock was delivered for a duration of 3 s. After imaging, any motion-related artifacts were corrected using the Non-Rigid Motion Correction (NoRMscorre) algorithm. The fluorescence time course was measured using ImageJ software by averaging all pixels within the regions of interest (ROIs). DF/F0 was calculated using the following equation: DF/F0 = [(F − F0)/F0], in which F0 is the baseline fluorescence signal averaged over a 10-s period before the onset of the forced running or tail shock. Fiber photometry recording of CRF1.0 with in vivo drug application
Male C57BL/6N mice bred at the NYULMC animal facility (10 to 12 weeks of age) were Wang et al., Science 382, eabq8173 (2023)
anesthetized with isoflurane and placed in a stereotaxic frame. AAV expressing hSyn-CRF1.0 or hSyn-CRFmut (Vigene Biosciences) was injected (160 nl per animal) into the PVN (AP: −0.75 mm relative to Bregma; ML: +0.22 mm relative to Bregma; DV: −4.7 mm from the dura). An optical fiber (400-mm diameter) was implanted 150 mm above the virus injection site (either at the time of virus injection or 2 weeks later). At the same time that the optical fiber was implanted, a bilateral cannula (Plastics One) for drug infusion was also implanted in the dorsal third ventricle or the left lateral ventricle. At least 4 weeks after virus injection, fiber photometry recording was performed in the PVN. Before fiber photometry recording, a ferrule sleeve (ADAL1-5, Thorlabs) was used to connect a matching optic fiber to the implanted fiber, and recordings were performed on the head-fixed wheel. For recording, a 390-Hz sinusoidal 488-nm blue LED light (35 mW; M470F1; Thorlabs) driven by a LEDD1B driver (Thorlabs) was bandpass-filtered (passing band: 472 ± 15 nm, Semrock, FF02-472/30-25) and delivered to the brain to excite CRF1.0 or CRFmut. The emission light passed through the same optic fiber, through a bandpass filter (passing band: 534 ± 25 nm, Semrock, FF01-535/50), and into a Femtowatt Silicon Photoreceiver, which recorded the CRF1.0 or CRFmut emission using an RZ5 real-time processor (TuckerDavis Technologies). The 390-Hz signals from the photoreceiver were extracted in real-time using a custom-written program (Tucker-Davis Technologies) and used to determine the intensity of the CRF1.0 or CRFmut fluorescence signal. For generating dose-response curves, CRF (C3042, Sigma or AS-24254, Eurogentec) or AHCRF 9-41 (1184, Tocris) was infused into one of the ventricles through the implanted cannula using a syringe (65457-02, Hamilton). For the data shown in Fig. 6, B and C, 250 nl of CRF diluted to indicated concentrations (4.8, 1.6, 0.5, or 0.05 mg/ml) or 250 nl saline was infused; for the data shown in Fig. 6D, 100 nl of 1.6 mg/ml CRF and/or 300 nl of 0.25 mg/ml AHCRF 9-41 was infused. CRF was diluted in distilled water, and AHCRF 9-41 was diluted in distilled water containing 0.1 M NH4OH. For the data shown in Fig. 6, B and D, Friedman’s test was performed, followed by Horm correction. For the data shown in Fig. 6C, the two-sided paired Wilcoxon signed rank test was performed. The peak values obtained after applying 1.6 mg/ml CRF in Fig. 6, B and D, were the average of all trials from each animal. Fiber photometry recording of CRF1.0 during behavioral testing
Male C57BL/6N mice (10 to 12 weeks of age, from River Vital Laboratory) were anesthetized with an intraperitoneal injection of sodium pentobarbital (80 mg/kg body weight) and
17 November 2023
AAV9-hSyn-CRF1.0 or AAV9-hSyn-EGFP-CAAX (300 nl, 3 × 1013 to 5 × 1013 vg/ml, WZ Biosciences) was injected into the PVN (AP: −0.80 mm relative to Bregma; ML: −0.25 mm relative to Bregma; DV: −4.60 mm from the dura) at a rate of 40 nl/min. The optic fiber (200 mm inner core diameter, 0.37 fiber numerical aperture, Thinkerbiotech) was implanted 0.20 mm above the injection site and sealed with dental cement. After 4 to 5 weeks (to allow the mice to recover and to allow for virus expression), a Multi-Channel Fiber Photometry Device (Inper, OPT-FPS-410/470/561) was used for recording. Signals were acquired at a frame rate of 50 Hz, with an exposure time of 9 ms, with gain 0, using 470-nm light at 30 to 40% power. For the tail lift experiments, the mouse was suspended by the tail 50 cm above the floor for 30 s per trial. Three 30-s tail lift trials were performed at an interval of ~220 s; the signal recorded 150 s before the first lift was used as the baseline, and the average of the three responses recorded during the 30-s lifts was used as the lift signal. For LiCl or saline injection, the signal recorded 500 s before injection was recorded as the baseline. The mice were then briefly anesthetized with isoflurane and given an intraperitoneal injection of saline (0.1 ml/10 g body weight) or LiCl (125 mg/kg body weight) dissolved in saline. The signals were recorded for 2400 s after intraperitoneal injection, and the average response measured during the first 1500 s was used as the LiCl or saline signal. Fiber photometry recording of SST1.0 during olfactory Pavlovian learning
Male adult C57BL/6J mice (8 to 13 weeks of age, from River Vital Laboratory) were anesthetized under ketamine and xylazine (100 and 10 mg/kg intraperitoneally, respectively) and AAV9-hSynSST1.0 (300 nl, 3 × 1013 to 5 × 1013 vg/ml, WZ Biosciences) was injected into the BLA (AP: −1.5 mm relative to Bregma; ML: −3.25 mm relative to Bregma; DV: −4.6 mm from the dura) at a rate of 100 nl/min. The optic fiber (200 mm, 0.39 NA, Thorlabs) was implanted 0.10 mm above the injection site and sealed with dental cement. Mice were allowed at least 3 weeks to recover and to express the virus before behavioral training. Signals were recorded using FiberOptoMeter (FOM-02M, C-Light, SooChow, China), using a beam from a 470-nm LED reflected with a dichroic mirror, focused with a lens coupled to a photomultiplier tube. The LED power at the tip of the patch cord was between 25 and 30 mW. Transcriptome-wide RNA-seq analysis
Cultured rat cortical neurons at DIV 5-7 were transfected with AAV virus (at a titer of 3 × 1013 to5 × 1013 vg/ml) expressing CRF1.0, SST1.0, or EGFP-CAAX, and the neurons were lysed by Trizol after 7-day expression for RNA extraction. 13 of 16
RES EARCH | R E S E A R C H A R T I C L E
Male C57BL/6N mice (6 to 8 weeks of age) were anesthetized with an intraperitoneal injection of tribromoethanol (Avertin; 500 mg/kg body weight), and the AAV9-hSyn-CRF1.0, AAV9hSyn-SST1.0, AAV9-hSyn-mApple-CAAX, or AAV9-hSyn-EGFP-CAAX virus (300 nl, 3 × 1013 to 5 × 1013 vg/ml, WZ Biosciences) was bilaterally injected into the motor cortex (AP: 1.0 mm relative to Bregma; ML: ±1.5 mm relative to Bregma; DV: −0.5 mm from the dura). After 2-week expression, motor cortex regions were dissected and frozen by liquid nitrogen for RNA extraction. The mRNA library constructing and sequencing were conducted by AZENTA Life Sciences, performed in Illumina HiSeq/ Illumina Novaseq/ MGI2000 instrument. Sequencing qualities were filtered by Cutadapt (V1.9.1, phred cutoff: 20, error rate: 0.1, adapter overlap: 1bp, min. length: 75, proportion of N: 0.1). Data were aligned to reference genome by the software Hisat2 (v2.0.1) and gene expression levels were estimated from the pair-end clean data by the software HTSeq (v0.6.1). The FPKM (fragments per kilobase of transcript per million mapped reads) heatmaps of selected CREB responsive genes were plotted by R (v 4.2.2). Western blot
Primary antibodies to SSTR2 (Abcam, ab134152, 1:2000), primary antibodies to CB1R (Abcam, ab23703, 1:500) GFP antibody (Abcam, ab6556, 1:1000), b-actin antibody (CWBIO, CW0096M, 1:2000), Goat anti Rabbit IgG-HRP conjugated (CWBIO, CW0103S, 1:3000), and Goat anti Mouse IgG-HRP conjugated (CWBIO, CW0102S, 1:3000) were used in this study for Western blots. Standard Western blot protocols were applied. Male C57BL/6N mice (10 to 12 weeks of age, from River Vital Laboratory) expressing SST1.0 or CRF1.0 sensors and EGFP-CAAX for 2 weeks were used. In brief, mice motor cortex were lysed, protein concentration was quantified by bicinchoninic acid assay, and an equal amount of each lysate was loaded for SDS–polyacrylamide gel electrophoresis. Then, sample proteins were transferred onto hybridization nitrocellulose filter (Millipore, HATF00010) and immunoblotted with primary antibodies, followed by secondary antibody incubation and exposure using cECL Western Blot Kit (CWBIO, CW0048M) by ChemiDoc XRS System (BIO-RAD). Immunohistochemistry
Mice were anaesthetized (using Avertin) and intracardially perfused with saline followed by 4% paraformaldehyde (PFA) in 0.1 M PBS buffer, and brains were dissected and fixed at 4 °C overnight by 4% PFA in 0.1 M PBS buffer. Brains were sectioned at 40 mm thickness using a VT1200 vibratome (Leica). Sections were placed in blocking solution containing 5% normal goat serum (NGS) and 0.1% Triton X-100 and 2 mM MgCl2 in 1× PBS for 1 hour, then incubated with primary antibodies to CRF (Peninsula LaboWang et al., Science 382, eabq8173 (2023)
ratories International, T-4037, 1:1000), primary antibodies to SSTR2 (Abcam, ab134152, 1:2000), and diluted AGT solution (0.5% NGS, 0.1% Triton, and 2 mM MgCl2 in 1× PBS) overnight at 4 °C. Sections were rinsed three times in AGT and incubated for 1 hour at room temperature with secondary antibodies Alexa-555conjugated goat-anti-rabbit IgG (H+L) (AAT Bioquest, 16690, 1:500) and Nissl Stained using NeuroTrace 640/660 (Invitrogen, N21483, 1:300). Sections were rinsed three times in AGT and mounted on slides using DAPI Fluoromount-G (SouthernBiotech, 0100-20) mounting medium. Sections were imaged on Olympus VS120 slide scanner. Slice electrophysiology
The procedures for preparing acute brain slices were similar to the parts of fluorescence imaging of peptide sensors in acute brain slices. Briefly, male C57BL/6N mice (6 to 8 weeks of age) were anesthetized with Avertin, and the AAV9hSyn-SST1.0, AAV9-hSyn-CRF1.0, or AAV9-hSynEYFP virus (300 nl, 3 × 1013 to 5 × 1013 vg/ml, WZ Biosciences) was injected into the mPFC (AP: 2.0 mm relative to Bregma; ML: 0.5 mm; DV: −2.0 mm from the dura) at a rate of 30 nl/min. After 3 weeks of virus expression, coronal medial PFC slices were prepared in a solution containing (in millimolar concentrations): 228 sucrose, 26 NaHCO3, 11 glucose, 2.5 KCI, 1 NaH2PO4, 7 MgSO4, and 0.5 CaCI2 and recovered at 35°C in ACSF containing (in millimolar concentrations): 119 NaCI, 26 NaHCO3, 11 glucose, 2.5 KCI, 1 NaH2PO4, 1.3 MgSO4, and 2.5 CaCI2. After 1 hour incubation, the slices were transferred to a recording chamber bathed with oxygenated ACSF at 35.5°C. Layer 5 pyramidal neurons were visualized with an upright infrared differential interference contrast (IR-DIC) microscope (BX51WI; Olympus). Whole-cell recordings were performed with a MultiClamp 700B amplifier and Digidata 1550B4 (Molecular Devices, USA). The resistance of patch pipette was 4 to 6 MW. Signals were filtered at 10 kHz and then sampled at 20 kHz using Clampex v10.4 (Molecular Devices). Layer 5 pyramidal neurons expressing SST1.0, CRF1.0, or EYFP in mPFC were recorded using an internal solution containing (in millimolar concentrations): 140 K-gluconate, 10 HEPES, 0.25 EGTA, 2 MgATP, 0.3 Na3GTP, 7 phosphocreatine (pH 7.25 to 7.3; osmolarity 294 to 298). To isolate voltage-dependent potassium currents, 0.5 mM TTX and 300 mM Cd2+ were included in the bath. We initially held the recorded neuron at −60 mV and applied a series of 200-ms test potentials (−140 to 0 mV) to activate voltagedependent potassium currents. The current recorded under pretreatment conditions was defined as the constitutive potassium current. Perfusion of 20 mM baclofen was used to activate the GIRK channel currents. The baclofen-activated currents (IBac) were defined by subtracting the
17 November 2023
constitutive current from those recorded during baclofen perfusion. Liquid junction potential (~16 mV) was not corrected for the membrane potential values in the text and figures. Behavioral assays
Male C57BL/6J mice [8 weeks of age, from SPF (Beijing) Biotechnology Co., LTD.] were anesthetized, and AAV9-hSyn-CRF1.0 or AAV9-hSynEGFP-CAAX was injected into the PVN, similar to the procedure for the PVN fiber photometry recording experiment. The AAV9-hSyn-SST1.0 or AAV9-hSyn-EYFP was injected into the BLA, similar to the procedure for the BLA fiber photometry recording experiment. The same cohort of mice was subjected to various behavioral tests at least 26 days after surgery in the following order: open-field test, elevated plus maze test, tail suspension test, forced swimming test, and sucrose preference test. Mice were handled for 3 days for 3 min each day before experiments. During behavioral procedures, all of the mice were singly housed. The behavior assays were performed as previously described (60, 80). Olfactory Pavlovian learning
The olfactory Pavlovian conditioning assays were performed using protocols described in previous research (81). In brief, mice were trained to associate odors with reinforcing outcomes in head-fixed configuration using a custom-made apparatus. The odors used in this study were ethyl acetate, 2-pentanone, and (R)-(+)-limonene, which have neutral value to mice. Mice were water-deprived and habituated to the head-fixed recording configuration. In the reward trials, water was delivered. In the neutral trials, nothing happened after cue delivery. In the punishment trials, an air puff was delivered to the eye of mice. Mice were trained for 4 days, and each day’s session contained 180 trials. Metabolism
Four or five mice were co-housed, and body weights were measured every 4 days after the virus injection. Twenty-nine days after virus injection, mice were singly housed, and water and food consumption was recorded from day 29 12:00 to day 30 11:00. Open-field test
The open-field test was performed in a nontransparent square box (50 cm by 50 cm by 40 cm), with smooth interior walls. The center area of the open field was defined as a 25 cm by 25 cm zone centered in the arena. At the start of the test, mice were placed in one of four corners of the arena and were allowed to freely explore the environment for 5 min. Locomotion traces were recorded by video camera for 10 min for each mouse. Time and entries in the center area and total distance were analyzed by EthoVision XT 8.5 software. 14 of 16
RES EARCH | R E S E A R C H A R T I C L E
Elevated plus maze test
The elevated plus maze has two opposite open arms without walls (30 cm by 6 cm), two opposite closed arms (30 cm by 6 cm by 15 cm) and a central platform (5 cm by 5 cm). The maze was elevated 80 cm above the floor. At the start of the session, animals were first placed in the center zone facing one of the open arms. Mice were allowed to explore the maze for 6 min. Locomotion traces were recorded by video camera. The time and entries were quantified and analyzed. Data were analyzed using EthoVision XT 5.1 software. Tail suspension test
The mouse was suspended by the tail 50 cm above the floor, ensuring that the mouse could not make any other contact or climb during the assay. The immobility behavior in a 6-min session was recorded, and immobility time for 2 to 6 min was analyzed. Forced swimming test
The mouse was placed in a transparent cylinder (25 cm high, 25 cm in diameter) filled with water to a depth of 18 cm and maintained at between 24° and 25°C for each 6-min session. After each session, the mouse was dried with a towel and returned to its home cage. The total immobility time for 2 to 6 min was recorded and analyzed. Mice were considered immobile when they did not make any struggling movements. Sucrose preference test
The mice were singly housed with two bottles of 1% sucrose water and adapted for 1 day before the test. The mice were singly housed with one bottle of 1% sucrose water and one bottle of water. The consumption from day 35 20:00 to day 36 12:00 was recorded, and the sucrose preference [sucrose water consumption/(sucrose water consumption + water consumption)] was analyzed. Quantification and statistical analysis
Summary data with error bars are presented as the mean ± SEM. Except where indicated otherwise, groups were compared using StudentÕs t test or a one-way analysis of variance with post hoc test, and differences were considered significant at P ≤ 0.05. Where applicable, *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, and n.s. means not significant. RE FE RENCES AND N OT ES
1. D. De Wied, E. R. De Kloet, Pro-opiomelanocortin (POMC) as homeostatic control system. Ann. N. Y. Acad. Sci. 512, 328–337 (1987). doi: 10.1111/j.1749-6632.1987.tb24971.x; pmid: 2831780 2. J. Spiess, J. Rivier, C. Rivier, W. Vale, Primary structure of corticotropin-releasing factor from ovine hypothalamus. Proc. Natl. Acad. Sci. U.S.A. 78, 6517–6521 (1981). doi: 10.1073/pnas.78.10.6517; pmid: 6273874 3. A. V. Schally, A. Arimura, A. J. Kastin, Hypothalamic regulatory hormones. Science 179, 341–350 (1973). doi: 10.1126/ science.179.4071.341; pmid: 4345570
Wang et al., Science 382, eabq8173 (2023)
4. V. Du Vigneaud, Hormones of the posterior pituitary gland: Oxytocin and vasopressin. Harvey Lect. 50, 1–26 (1954-1955). pmid: 13306033 5. A. L. Alhadeff et al., A neural circuit for the suppression of pain by a competing need state. Cell 173, 140–152.e15 (2018). doi: 10.1016/j.cell.2018.02.057; pmid: 29570993 6. L. de Lecea et al., The hypocretins: Hypothalamus-specific peptides with neuroexcitatory activity. Proc. Natl. Acad. Sci. U.S.A. 95, 322–327 (1998). doi: 10.1073/pnas.95.1.322; pmid: 9419374 7. T. Sakurai et al., Orexins and orexin receptors: A family of hypothalamic neuropeptides and G protein-coupled receptors that regulate feeding behavior. Cell 92, 573–585 (1998). doi: 10.1016/S0092-8674(00)80949-6; pmid: 9491897 8. W. Vale, J. Spiess, C. Rivier, J. Rivier, Characterization of a 41-residue ovine hypothalamic peptide that stimulates secretion of corticotropin and b-endorphin. Science 213, 1394–1397 (1981). doi: 10.1126/science.6267699; pmid: 6267699 9. A. S. Hauser, M. M. Attwood, M. Rask-Andersen, H. B. Schiöth, D. E. Gloriam, Trends in GPCR drug discovery: New agents, targets and indications. Nat. Rev. Drug Discov. 16, 829–842 (2017). doi: 10.1038/nrd.2017.178; pmid: 29075003 10. A. P. Davenport, C. C. G. Scully, C. de Graaf, A. J. H. Brown, J. J. Maguire, Advances in therapeutic peptides targeting G protein-coupled receptors. Nat. Rev. Drug Discov. 19, 389–413 (2020). doi: 10.1038/s41573-020-0062-z; pmid: 32494050 11. T. Hökfelt, T. Bartfai, F. Bloom, Neuropeptides: Opportunities for drug discovery. Lancet Neurol. 2, 463–472 (2003). doi: 10.1016/S1474-4422(03)00482-4; pmid: 12878434 12. J. S. Marvin et al., An optimized fluorescent probe for visualizing glutamate neurotransmission. Nat. Methods 10, 162–170 (2013). doi: 10.1038/nmeth.2333; pmid: 23314171 13. E. K. Unger et al., Directed evolution of a selective and sensitive serotonin sensor via machine learning. Cell 183, 1986–2002.e26 (2020). doi: 10.1016/j.cell.2020.11.040; pmid: 33333022 14. P. M. Borden et al., A fast genetically encoded fluorescent sensor for faithful in vivo acetylcholine detection in mice, fish, worms and flies. bioRxiv 2020.02.07.939504 [Preprint] (2020); https://doi.org/10.1101/2020.02.07.939504. 15. S. R. Foster et al., Discovery of human signaling systems: pairing peptides to G protein-coupled receptors. Cell 179, 895–908.e21 (2019). doi: 10.1016/j.cell.2019.10.010; pmid: 31675498 16. V. Isberg et al., GPCRdb: An information system for G proteincoupled receptors. Nucleic Acids Res. 44, D356–D364 (2016). doi: 10.1093/nar/gkv1178; pmid: 26582914 17. M. Jing et al., A genetically encoded fluorescent acetylcholine indicator for in vitro and in vivo studies. Nat. Biotechnol. 36, 726–737 (2018). doi: 10.1038/nbt.4184; pmid: 29985477 18. F. Sun et al., A genetically encoded fluorescent sensor enables rapid and specific detection of dopamine in flies, fish, and mice. Cell 174, 481–496.e19 (2018). doi: 10.1016/ j.cell.2018.06.042; pmid: 30007419 19. J. Feng et al., A genetically encoded fluorescent sensor for rapid and specific in vivo detection of norepinephrine. Neuron 102, 745–761.e8 (2019). doi: 10.1016/j.neuron.2019.02.037; pmid: 30922875 20. M. Jing et al., An optimized acetylcholine sensor for monitoring in vivo cholinergic activity. Nat. Methods 17, 1139–1146 (2020). doi: 10.1038/s41592-020-0953-2; pmid: 32989318 21. T. Patriarchi et al., Ultrafast neuronal imaging of dopamine dynamics with designed genetically encoded sensors. Science 360, eaat4422 (2018). doi: 10.1126/science.aat4422; pmid: 29853555 22. T. Qian et al., A genetically encoded sensor measures temporal oxytocin release from different neuronal compartments. Nat. Biotechnol. 41, 944–957 (2023). doi: 10.1038/ s41587-022-01561-2; pmid: 36593404 23. D. Ino, Y. Tanaka, H. Hibino, M. Nishiyama, A fluorescent sensor for real-time measurement of extracellular oxytocin dynamics in the brain. Nat. Methods 19, 1286–1294 (2022). doi: 10.1038/s41592-022-01597-x; pmid: 36138174 24. L. Duffet et al., A genetically encoded sensor for in vivo imaging of orexin neuropeptides. Nat. Methods 19, 231–241 (2022). doi: 10.1038/s41592-021-01390-2; pmid: 35145320 25. J. M. Otaki, S. Firestein, Length analyses of mammalian G-protein-coupled receptors. J. Theor. Biol. 211, 77–100 (2001). doi: 10.1006/jtbi.2001.2272; pmid: 11419953 26. H. Unal, S. S. Karnik, Domain coupling in GPCRs: The engine for induced conformational changes. Trends Pharmacol. Sci. 33, 79–88 (2012). doi: 10.1016/j.tips.2011.09.007; pmid: 22037017
17 November 2023
27. K. Hollenstein et al., Structure of class B GPCR corticotropinreleasing factor receptor 1. Nature 499, 438–443 (2013). doi: 10.1038/nature12357; pmid: 23863939 28. S. Ma et al., Molecular basis for hormone recognition and activation of corticotropin-releasing factor receptors. Mol. Cell 77, 669–680. e4 (2020). doi: 10.1016/j.molcel.2020.01.013; pmid: 32004470 29. W. Zhao et al., Structural insights into ligand recognition and selectivity of somatostatin receptors. Cell Res. 32, 761–772 (2022). doi: 10.1038/s41422-022-00679-x; pmid: 35739238 30. A. Paul et al., Transcriptional architecture of synaptic communication delineates GABAergic neuron identity. Cell 171, 522–539.e20 (2017). doi: 10.1016/j.cell.2017.08.032; pmid: 28942923 31. S. J. Smith et al., Single-cell transcriptomic evidence for dense intracortical neuropeptide networks. eLife 8, e47889 (2019). doi: 10.7554/eLife.47889; pmid: 31710287 32. W. Zhong et al., The neuropeptide landscape of human prefrontal cortex. Proc. Natl. Acad. Sci. U.S.A. 119, e2123146119 (2022). doi: 10.1073/pnas.2123146119; pmid: 35947618 33. F. Sun et al., Next-generation GRAB sensors for monitoring dopaminergic activity in vivo. Nat. Methods 17, 1156–1166 (2020). doi: 10.1038/s41592-020-00981-9; pmid: 33087905 34. J. Wan et al., A genetically encoded sensor for measuring serotonin dynamics. Nat. Neurosci. 24, 746–752 (2021). doi: 10.1038/s41593-021-00823-7; pmid: 33821000 35. F. M. Dautzenberg et al., Cell-type specific calcium signaling by corticotropin-releasing factor type 1 (CRF1) and 2a (CRF2(a)) receptors: phospholipase C-mediated responses in human embryonic kidney 293 but not SK-N-MC neuroblastoma cells. Biochem. Pharmacol. 68, 1833–1844 (2004). doi: 10.1016/ j.bcp.2004.07.013; pmid: 15450949 36. B. Wamsley, G. Fishell, Genetic and activity-dependent mechanisms underlying interneuron diversity. Nat. Rev. Neurosci. 18, 299–309 (2017). doi: 10.1038/nrn.2017.30; pmid: 28381833 37. C. M. Persoon et al., Pool size estimations for dense-core vesicles in mammalian CNS neurons. EMBO J. 37, e99672 (2018). doi: 10.15252/embj.201899672; pmid: 30185408 38. S. Arora et al., SNAP-25 gene family members differentially support secretory vesicle fusion. J. Cell Sci. 130, 1877–1889 (2017). doi: 10.1242/jcs.201889; pmid: 28404788 39. U. Kumar, S. Singh, Role of somatostatin in the regulation of central and peripheral factors of satiety and obesity. Int. J. Mol. Sci. 21, 2568 (2020). doi: 10.3390/ijms21072568; pmid: 32272767 40. A. C. Hauge-Evans et al., Somatostatin secreted by islet d-cells fulfills multiple roles as a paracrine regulator of islet function. Diabetes 58, 403–411 (2009). doi: 10.2337/ db08-0792; pmid: 18984743 41. P. Rorsman, M. O. Huising, The somatostatin-secreting pancreatic d-cell in health and disease. Nat. Rev. Endocrinol. 14, 404–414 (2018). doi: 10.1038/s41574-018-0020-6; pmid: 29773871 42. A. Salehi, S. S. Qader, E. Grapengiesser, B. Hellman, Pulses of somatostatin release are slightly delayed compared with insulin and antisynchronous to glucagon. Regul. Pept. 144, 43–49 (2007). doi: 10.1016/j.regpep.2007.06.003; pmid: 17628719 43. B. Hellman, A. Salehi, E. Grapengiesser, E. Gylfe, Isolated mouse islets respond to glucose with an initial peak of glucagon release followed by pulses of insulin and somatostatin in antisynchrony with glucagon. Biochem. Biophys. Res. Commun. 417, 1219–1223 (2012). doi: 10.1016/j.bbrc.2011.12.113; pmid: 22227186 44. T. Reisine, G. I. Bell, Molecular biology of somatostatin receptors. Endocr. Rev. 16, 427–442 (1995). pmid: 8521788 45. O. Nakagawasai et al., Immunohistochemical fluorescence intensity reduction of brain somatostatin in the impairment of learning and memory-related behaviour induced by olfactory bulbectomy. Behav. Brain Res. 142, 63–67 (2003). doi: 10.1016/S0166-4328(02)00383-2; pmid: 12798266 46. P. Davies, R. Katzman, R. D. Terry, Reduced somatostatin-like immunoreactivity in cerebral cortex from cases of Alzheimer disease and Alzheimer senile dementia. Nature 288, 279–280 (1980). doi: 10.1038/288279a0; pmid: 6107862 47. C. Kluge, C. Stoppel, C. Szinyei, O. Stork, H. C. Pape, Role of the somatostatin system in contextual fear memory and hippocampal synaptic plasticity. Learn. Mem. 15, 252–260 (2008). doi: 10.1101/lm.793008; pmid: 18391186 48. J. M. Stujenske et al., Prelimbic cortex drives discrimination of non-aversion via amygdala somatostatin interneurons. Neuron 110, 2258–2267.e11 (2022). doi: 10.1016/ j.neuron.2022.03.020; pmid: 35397211 49. J. Gründemann, A. Lüthi, Ensemble coding in amygdala circuits for associative learning. Curr. Opin. Neurobiol. 35, 200–206 (2015). doi: 10.1016/j.conb.2015.10.005; pmid: 26531780 50. P. H. Janak, K. M. Tye, From circuits to behaviour in the amygdala. Nature 517, 284–292 (2015). doi: 10.1038/ nature14188; pmid: 25592533
15 of 16
RES EARCH | R E S E A R C H A R T I C L E
51. Y. S. Jo, V. M. K. Namboodiri, G. D. Stuber, L. S. Zweifel, Persistent activation of central amygdala CRF neurons helps drive the immediate fear extinction deficit. Nat. Commun. 11, 422 (2020). doi: 10.1038/s41467-020-14393-y; pmid: 31969571 52. C. A. Sanford et al., A central amygdala CRF circuit facilitates learning about weak threats. Neuron 93, 164–178 (2017). doi: 10.1016/j.neuron.2016.11.034; pmid: 28017470 53. M. B. Pomrenze et al., A corticotropin releasing factor network in the extended amygdala for anxiety. J. Neurosci. 39, 1030–1043 (2019). doi: 10.1523/JNEUROSCI.2143-18.2018; pmid: 30530860 54. G. de Guglielmo et al., Inactivation of a CRF-dependent amygdalofugal pathway reverses addiction-like behaviors in alcohol-dependent rats. Nat. Commun. 10, 1238 (2019). doi: 10.1038/s41467-019-09183-0; pmid: 30886240 55. J. Kim et al., Rapid, biphasic CRF neuronal responses encode positive and negative valence. Nat. Neurosci. 22, 576–585 (2019). doi: 10.1038/s41593-019-0342-2; pmid: 30833699 56. Y. Yuan et al., Reward inhibits paraventricular CRH neurons to relieve stress. Curr. Biol. 29, 1243–1251.e4 (2019). doi: 10.1016/j.cub.2019.02.048; pmid: 30853436 57. N. Daviu et al., Paraventricular nucleus CRH neurons encode stress controllability and regulate defensive behavior selection. Nat. Neurosci. 23, 398–410 (2020). doi: 10.1038/ s41593-020-0591-0; pmid: 32066984 58. T. Gallopin, H. Geoffroy, J. Rossier, B. Lambolez, Cortical sources of CRF, NKB, and CCK and their effects on pyramidal cells in the neocortex. Cereb. Cortex 16, 1440–1452 (2006). doi: 10.1093/cercor/bhj081; pmid: 16339088 59. J. M. Deussing, A. Chen, The corticotropin-releasing factor family: physiology of the stress response. Physiol. Rev. 98, 2225–2286 (2018). doi: 10.1152/physrev.00042.2017; pmid: 30109816 60. P. Chen et al., Prefrontal cortex corticotropin-releasing factor neurons control behavioral style selection under challenging situations. Neuron 106, 301–315.e7 (2020). doi: 10.1016/ j.neuron.2020.01.033; pmid: 32101698 61. A. Uribe-Mariño et al., Prefrontal cortex corticotropin-releasing factor receptor 1 conveys acute stress-induced executive dysfunction. Biol. Psychiatry 80, 743–753 (2016). doi: 10.1016/ j.biopsych.2016.03.2106; pmid: 27318500 62. T. A. Nielsen, D. A. DiGregorio, R. A. Silver, Modulation of glutamate mobility reveals the mechanism underlying slowrising AMPAR EPSCs and the diffusion coefficient in the synaptic cleft. Neuron 42, 757–771 (2004). doi: 10.1016/ j.neuron.2004.04.003; pmid: 15182716 63. M. E. Rice, G. A. Gerhardt, P. M. Hierl, G. Nagy, R. N. Adams, Diffusion coefficients of neurotransmitters and their metabolites in brain extracellular fluid space. Neuroscience 15, 891–902 (1985). doi: 10.1016/0306-4522(85)90087-9; pmid: 2866468 64. A. N. Weiss, M. A. Bittner, R. W. Holz, D. Axelrod, Protein mobility within secretory granules. Biophys. J. 107, 16–25 (2014). doi: 10.1016/j.bpj.2014.04.063; pmid: 24988337 65. M. R. DiGruccio et al., Comprehensive alpha, beta and delta cell transcriptomes reveal that ghrelin selectively activates delta cells and promotes somatostatin release from pancreatic islets. Mol. Metab. 5, 449–458 (2016). doi: 10.1016/ j.molmet.2016.04.007; pmid: 27408771 66. M. O. Huising, T. van der Meulen, J. L. Huang, M. S. Pourhosseinzadeh, G. M. Noguchi, The difference d-cells make in glucose control. Physiology 33, 403–411 (2018). doi: 10.1152/physiol.00029.2018; pmid: 30303773
Wang et al., Science 382, eabq8173 (2023)
67. S. J. Smith, M. Hawrylycz, J. Rossier, U. Sümbül, New light on cortical neuropeptides and synaptic network plasticity. Curr. Opin. Neurobiol. 63, 176–188 (2020). doi: 10.1016/j.conb.2020.04.002; pmid: 32679509 68. M. Ludwig, G. Leng, Dendritic peptide release and peptidedependent behaviours. Nat. Rev. Neurosci. 7, 126–136 (2006). doi: 10.1038/nrn1845; pmid: 16429122 69. A. N. van den Pol, Neuropeptide transmission in brain circuits. Neuron 76, 98–115 (2012). doi: 10.1016/j.neuron.2012.09.014; pmid: 23040809 70. J. P. Overington, B. Al-Lazikani, A. L. Hopkins, How many drug targets are there? Nat. Rev. Drug Discov. 5, 993–996 (2006). doi: 10.1038/nrd2199; pmid: 17139284 71. I. Nederpelt, J. Bunnik, A. P. IJzerman, L. H. Heitman, Kinetic profile of neuropeptide–receptor interactions. Trends Neurosci. 39, 830–839 (2016). doi: 10.1016/j.tins.2016.09.008; pmid: 27793433 72. A. Kaiser, I. Coin, Capturing peptide-GPCR interactions and their dynamics. Molecules 25, 4724 (2020). doi: 10.3390/ molecules25204724; pmid: 33076289 73. E. B. De Souza, Corticotropin-releasing factor receptors in the rat central nervous system: Characterization and regional distribution. J. Neurosci. 7, 88–100 (1987). doi: 10.1523/ JNEUROSCI.07-01-00088.1987; pmid: 3027279 74. M. Castro, V. O. Nikolaev, D. Palm, M. J. Lohse, J. P. Vilardaga, Turn-on switch in parathyroid hormone receptor by a two-step parathyroid hormone binding mechanism. Proc. Natl. Acad. Sci. U.S.A. 102, 16084–16089 (2005). doi: 10.1073/ pnas.0503942102; pmid: 16236727 75. I. Nederpelt et al., Characterization of 12 GnRH peptide agonists – a kinetic perspective. Br. J. Pharmacol. 173, 128–141 (2016). doi: 10.1111/bph.13342; pmid: 26398856 76. W. Yin et al., A complex structure of arrestin-2 bound to a G protein-coupled receptor. Cell Res. 29, 971–983 (2019). doi: 10.1038/s41422-019-0256-2; pmid: 31776446 77. K. Yusa, L. Zhou, M. A. Li, A. Bradley, N. L. Craig, A hyperactive piggyBac transposase for mammalian applications. Proc. Natl. Acad. Sci. U.S.A. 108, 1531–1536 (2011). doi: 10.1073/ pnas.1008322108; pmid: 21205896 78. W. K. Kroeze et al., PRESTO-Tango as an open-source resource for interrogation of the druggable human GPCRome. Nat. Struct. Mol. Biol. 22, 362–369 (2015). doi: 10.1038/ nsmb.3014; pmid: 25895059 79. Q. Wan et al., Mini G protein probes for active G protein– coupled receptors (GPCRs) in live cells. J. Biol. Chem. 293, 7466–7473 (2018). doi: 10.1074/jbc.RA118.001975; pmid: 29523687 80. X. Y. Qin et al., PSD-93 up-regulates the synaptic activity of corticotropin-releasing hormone neurons in the paraventricular nucleus in depression. Acta Neuropathol. 142, 1045–1064 (2021). doi: 10.1007/s00401-021-02371-7; pmid: 34536123 81. Y. Zhu et al., Dynamic salience processing in paraventricular thalamus gates associative learning. Science 362, 423–429 (2018). doi: 10.1126/science.aat0481; pmid: 30361366 ACKN OWLED GMEN TS
We thank Y. Rao for sharing the two-photon microscope. We thank X. Lei at PKU-CLS, the National Center for Protein Sciences at Peking University, and State Key Laboratory of Membrane
17 November 2023
Biology at Tsinghua University for providing support for the Opera Phenix high-content screening system. Cartoons in figures were created with BioRender.com. We thank members of the Li lab for helpful suggestions and comments on the manuscript. Funding: This research was supported by grants from the National Natural Science Foundation of China (31925017), the National Key R&D Program of China (2019YFE011781), the Beijing Municipal Science and Technology Commission (Z220009), and the NIH BRAIN Initiative (1U01NS120824) to Y.L. Support was also provided by the Feng Foundation of Biomedical Research, the Clement and Xinxin Foundation, the New Cornerstone Science Foundation through the New Cornerstone Investigator Program and the XPLORER PRIZE (to Y.L.) and by grants from the Peking-Tsinghua Center for Life Sciences and the State Key Laboratory of Membrane Biology at Peking University School of Life Sciences to Y.L., the NIH BRAIN Initiative (1U01NS113358) to Y.L. and D.L., grants from the National Natural Science Foundation of China (82171492) to Y.Z., the Osamu Hayaishi Memorial Scholarship to T.O., and the Levy Leon Postdoctoral Fellowship to M.L. Author contributions: Y.L. designed and supervised the project. H.W. performed the experiments related to developing, optimizing, and characterizing neuropeptide sensors in cultured cells, with contributions from T.Q., Y.Y., S.F., L.G., G.L. and L.W. Y.Zhao and T.Q. performed the two-photon imaging of sensors in acute brain slices. C.W. and Y.Zhuo performed the in vivo two-photon imaging of mice cortex. H.R. and W.Q. performed the experiments related to pancreatic islets under the supervision of L.C. and C.T. T.O., L.M., and Y.J. performed in vivo intracerebroventricular infusion and fiber photometry recording experiments under the supervision of D.L. P.C. performed in vivo fiber photometry recording experiments and some behavior experiments under the supervision of J.-N.Z. Z.C. and Y.Zhu performed in vivo fiber photometry recording experiments and some behavior experiments. L.Z. and M. J. performed the slice electrophysiology experiments. All authors contributed to the interpretation and analysis of the data. H.W. and Y.L. wrote the manuscript, with input from all coauthors. Competing interests: All authors declare no competing interests. Data and materials availability: DNA plasmids of GRAB peptide sensors used in this study have been deposited to Addgene (Addgene ID 208654-208685). All other data needed to evaluate the conclusions in the paper are available in the main text or the supplementary materials. License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/ about/science-licenses-journal-article-reuse SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.abq8173 Figs. S1 to S17 Tables S1 to S4 References (82–97) MDAR Reproducibility Checklist Movie S1 Data S1 and S2 Submitted 3 May 2022; resubmitted 6 July 2023 Accepted 2 October 2023 10.1126/science.abq8173
16 of 16
RESEAR CH
RESEARCH ARTICLES
◥
ATMOSPHERIC AEROSOLS
Particle-phase accretion forms dimer esters in pinene secondary organic aerosol Christopher M. Kenseth1* , Nicholas J. Hafeman1, Samir P. Rezgui1, Jing Chen2, Yuanlong Huang3, Nathan F. Dalleska3, Henrik G. Kjaergaard2, Brian M. Stoltz1, John H. Seinfeld1,4, Paul O. Wennberg3,4 Secondary organic aerosol (SOA) is ubiquitous in the atmosphere and plays a pivotal role in climate, air quality, and health. The production of low-volatility dimeric compounds through accretion reactions is a key aspect of SOA formation. However, despite extensive study, the structures and thus the formation mechanisms of dimers in SOA remain largely uncharacterized. In this work, we elucidate the structures of several major dimer esters in SOA from ozonolysis of a-pinene and b-pinene—substantial global SOA sources—through independent synthesis of authentic standards. We show that these dimer esters are formed in the particle phase and propose a mechanism of nucleophilic addition of alcohols to a cyclic acylperoxyhemiacetal. This chemistry likely represents a general pathway to dimeric compounds in ambient SOA.
S
econdary organic aerosol (SOA) contributes substantially (15 to 80% by mass) to the global burden of atmospheric fine particulate matter (PM2.5) (1), which exerts large but uncertain effects on climate (2) as well as adverse impacts on air quality and human health (3, 4). The oxidation of monoterpenes (C10H16), emitted in appreciable quantities from forested regions (~150 Tg year−1) (5), represents a dominant source of SOA (6–8). For more than two decades, high–molecular weight dimeric compounds, notably those proposed to
-Pinene
Abundance
oxidation -Pinene
emission
contain ester linkages, have been identified using advanced mass spectrometric techniques as important components of both laboratoryderived and ambient monoterpene SOA and have been implicated as key players in particle formation and growth, volatility, viscosity, and cloud condensation nuclei (CCN) activity (9–39). Particle-phase reactions of closed-shell monomers (e.g., esterification and peroxyhemiacetal or diacyl peroxide decomposition) and gas-phase reactions involving early-stage oxidation products and/or reactive intermediates [e.g., stabilized LC Monomers
1 Division of Chemistry and Chemical Engineering, California Institute of Technology, Pasadena, CA 91125, USA. 2 Department of Chemistry, University of Copenhagen, DK-2100 Copenhagen, Denmark. 3Division of Geological and Planetary Sciences, California Institute of Technology, Pasadena, CA 91125, USA. 4Division of Engineering and Applied Science, California Institute of Technology, Pasadena, CA 91125, USA.
*Corresponding author. Email: [email protected] Present address: Department of Atmospheric Sciences, University of Washington, Seattle, WA 98195, USA.
Dimers
Proposed Structure O HO
SOA
OH
O O
O
OH
Time
O
MS/MS Intensity
Intensity
MS
Proposed Formation Mechanism O R1
m/z Pinene SOA Formation
Criegee intermediates (SCIs), carboxylic acids, and organic peroxy radicals (RO2)] have been advanced as possible dimer ester formation pathways (fig. S1). Owing to a lack of authentic standards (40), however, the structures of the dimer esters are not known but only inferred from accurate mass and fragmentation data. As a result, mechanistic understanding of dimer ester formation, in particular the relevance of gas- versus particle-phase chemistry, remains unconstrained (Fig. 1). In this work, informed by detailed structural analyses, we synthesize the first authentic standards of several major dimer esters identified in SOA from ozonolysis of a-pinene and b-pinene, which together account for >50% of total monoterpene emissions (5). On the basis of targeted experiments in the Caltech dual 24-m3 Teflon environmental chambers (CTEC) using ultraperformance liquid chromatography coupled to negative electrospray ionization quadrupole time-of-flight mass spectrometry [UPLC/(−)ESIQ-TOF-MS] for analysis of SOA molecular composition (41), we demonstrate that these
R2 OH
O
OH H2O
R1
O
R2
m/z
Analysis via Liquid Chromatography (LC)/Mass Spectrometry (MS)
CURRENT APPROACH Ambiguous structural assignment
O H18O
OH O
Informs targeted SOA experiments
O
OH
OH
O
Guided by LC/MS and MS/MS analyses
O
Synthesis of Authentic Standards THIS RESEARCH Unambiguous structural assignment Enables elucidation of formation mechanisms
Fig. 1. Characterization of dimeric compounds in pinene SOA. Currently proposed structures, and by extension formation mechanisms, of dimeric compounds identified in pinene SOA using advanced mass spectrometric techniques are inferred from accurate mass and fragmentation data and are SCIENCE science.org
thus uncertain. Independent synthesis of authentic standards, guided by mass spectrometric analysis, affords unambiguous structural assignment and enables elucidation of formation mechanisms by informing targeted SOA experiments. 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
787
RES EARCH | R E S E A R C H A R T I C L E S
dimer esters are formed through particle-phase chemistry and propose a unifying mechanism that accounts for the observed regioselectivity, dynamics, and environmental dependencies (e.g., oxidant type and RO2 fate) of ester formation. Identification of the chemistry underlying dimer ester production provides a missing link that ties the atmospheric degradation of monoterpenes to the formation of low-volatility accretion products capable of driving aerosol formation and growth.
A
synthesized bPdiol (fig. S3) did not, however, result in detectable production of dimer ester I, which suggests, in line with past studies (27, 30, 33), that dimer formation does not occur through conventional esterification (i.e., carboxylic acid + alcohol) in either the gas or particle phase. To constrain the potential involvement of bPdiol in forming dimer ester I, b-pinene ozonolysis experiments were carried out in the absence of cyclohexane (CHX) as a scavenger for OH—which is formed as a by-product of
Dimer esters in pinene SOA
Guided by our previous work on dimers formed from synergistic O3 + OH oxidation (33), the structure of one of the major dimers identified in SOA from b-pinene ozonolysis (Fig. 2A, dimer ester I) was proposed to consist of an ester of cis-pinic acid, the most abundant carboxylic acid measured in pinene SOA (34) and a commonly reported dimer subunit (15, 17, 27, 30, 33), and b-pinanediol (bPdiol) (fig. S2). CTEC experiments featuring synthesized cis-pinic acid and
160
120 100 80
O
357
P + O3 P + O3 + CHX P + O3 + CHX + Pdiol
140
(C17H26O8)
HO
185
O
OH +
O
O
(C9H14O4)
299
OH
HO
OH
HO
O
OH
O
OH
O
OH
O
O
(C15H24O6)
O
O
OH
HO
+
O
60
HO
4
OH
337 (C19H30O5)
40 20 0
B
160 (C9H14O4)
P + O3 P + O3 + CHX P + O3 + CHX + Pdiol P + O3 + CHX + Pdiol
OH
O
HO O
120
Percent
O
O
185
140
O
O
100
OH
HO
357
80
O
(C17H26O8)
O
OH
HO
OH OH
+
O
60
199
367
(C10H16O4)
40
(C19H28O7)
337
20
(C19H30O5)
0
C
140 185
P + O3 + CHX P + O3 + CHX + OH-Pinonic
(C9H14O4)
120
357 (C17H26O8)
100
O
80
O OH
HO O
199
60
OH
+ HO O
(C10H16O4)
40 367 (C19H28O7)
20 0 3.50
4.00
4.50
5.00
5.50
6.00
6.50
7.00
7.50
Time (min) Fig. 2. Formation of dimer esters in pinene SOA. (A to C) Base peak ion (BPI) chromatograms of SOA formed from ozonolysis of b-pinene (bP) [(A) and (C)] and a-pinene (aP) (B) after ~4 hours of reaction in the CTEC. Experiments were conducted in the absence of CHX as an OH scavenger, in the presence of CHX, and in the presence of both CHX and aPdiol, bPdiol, or OH-pinonic acid. Numbers correspond to nominal mass/charge ratio (m/z) values of [MÐH]− ions, and molecular formulas are given in parentheses. Chromatograms are normalized to the total organic carbon (TOC) content of the corresponding SOA filter samples, 788
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
reported as averages of duplicate samples collected in parallel for each experiment, and scaled such that the largest peak in the control experiments (gray shading) is 100%. cis-Pinic acid (C9H14O4) and OH-pinonic acid (C10H16O4) were identified through comparison with authentic standards. Structures in dashed boxes denote those proposed in past studies (12, 15). Structures in shaded boxes are of monomeric subunits identified for dimer esters I to IV. For clarity, all structures are drawn as the (+) enantiomers, despite experiments being carried out with (+)-aP, (−)-bP, (+)-aPdiol, (−)-bPdiol, and (−)-OH-pinonic acid. science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
produced from reaction of bPdiol and a derivative of cis-pinic acid. Based on the formation of dimer ester I in b-pinene photooxidation experiments performed in the presence of commercial cis-pinic acid, we had previously suggested that dimer ester I is produced through reaction of an OH-derived product, tentatively identified as either bPdiol or a derivative, and cis-pinic acid (33). However,
ozonolysis—in the presence of CHX, and in the presence of both CHX and bPdiol (Fig. 2A). Because dimer ester I is one of several dimers shown to form from accretion of O3- and OHderived products and/or intermediates (33), its production was inhibited by CHX. Unlike the other synergistic O3 + OH dimers, however, dimer ester I was observed to form on addition of bPdiol, which demonstrates that it is
O
R
O
OH
O
O
HO
O
80
167 (C9H11O3)
60 40
100
(C9H13O4)
141 (C8H13O2)
Percent
A
Primary Ester
185
100
R
O
Secondary Ester
80
OH
60
OH
40
337
20
123
(C19H29O5)
0
(C8H11O)
7.20 7.30 7.40
20
Time (min)
0 100
150
200
100
250
(C9H13O4)
80 (C8H13O2) 60
350
80 60
O
40
HO
OH
4
20
40 20
300
100
185
141
Percent
B
0
167
131
(C9H11O3)
(C6H11O3)
5.85 5.95 6.05
299
Time (min)
(C15H23O6)
0 100
C
150
100
200
169
60 40
167 (C9H11O3)
123
300
350
100
(C10H17O2)
80
250
337
80
Percent
Percent
Fig. 3. Determination of dimer ester structures. (A to D) Extracted ion chromatograms (EICs) and MS/MS spectra of synthesized secondary (red) and primary (blue) dimer esters of (+)-cis-pinic acid and (+)-bPdiol (A), 6hydroxyhexanoic acid (B), (+)-aPdiol (C), and (+)-OHpinonic acid (D) as well as of dimer esters identified in SOA formed from ozonolysis of a-pinene and/or b-pinene (black diamonds and dashed lines): dimer ester I (C19H30O5) (A), dimer ester II (C15H24O6) (B), dimer ester III (C19H30O5) (C), and dimer ester IV (C19H28O7) (D). Numbers in MS/MS spectra correspond to nominal m/z values of [MÐH]− ions, and ionic formulas [CxHyOz]− are given in parentheses.
(C8H11O) 151 (C10H15O)
(C19H29O5)
60
OH OH
40 20 0 7.40 7.50 7.60
20
Time (min)
0 100
185
100 80
141 (C8H13O2)
167 (C9H11O3)
250
300
OH
HO
80
O
60 40
367 (C19H27O7)
20
123 (C8H11O)
349
0 6.00 6.10 6.20
20
350
O
100
(C9H13O4)
60 40
200
Percent
D
150
(C19H25O6)
Time (min)
0 100
150
200
250
m/z SCIENCE science.org
300
350
the above findings, together with unsuccessful attempts to replicate the b-pinene photooxidation experiments using synthesized cis-pinic acid, exclude cis-pinic acid as the dimer ester source. We hypothesize that dimer ester I was formed in the earlier photooxidation experiments as a result of an impurity present in the commercial cis-pinic acid, likely the cis-pinic acid derivative or an oxidizable precursor. These insights were made possible only through synthesis of high-purity cis-pinic acid and bPdiol standards. b-Pinene ozonolysis experiments conducted in the presence of CHX and alcohols of varying structure and volatility (fig. S4) clarify that the cis-pinic acid derivative forms dimer esters only with alcohols of sufficiently low volatility to undergo gas-particle partitioning (42) and, therefore, that the accretion reaction occurs in the particle phase. Motivated by tandem mass spectrometry (MS/MS) analysis suggesting that the major dimer only present in SOA from b-pinene ozonolysis with CHX (Fig. 2A, dimer ester II) is an ester of cis-pinic acid and 6hydroxyhexanoic acid (an OH oxidation product of CHX), 6-hydroxyhexanoic acid was added to a b-pinene ozonolysis experiment performed without CHX (fig. S5), which led to appreciable formation of dimer ester II. Together with the production of dimer ester I from bPdiol, this result illustrates the generality of the particlephase reaction between the cis-pinic acid derivative and semivolatile or low-volatility alcohols. The currently accepted mechanism for the production of cis-pinic acid from ozonolysis of both a-pinene and b-pinene proceeds through a common acyl peroxy radical (fig. S6). This commonality suggests that dimer esters proposed to contain cis-pinic acid subunits in each SOA system may be formed from the same cispinic acid derivative. a-Pinene ozonolysis experiments without CHX, with CHX, and with both CHX and either bPdiol or a-pinanediol (aPdiol) (Fig. 2B) confirm that the cis-pinic acid derivative is produced from both a-pinene and b-pinene ozonolysis, as evidenced by the formation of dimer ester I, the analogous ester of cis-pinic acid and aPdiol (Fig. 2B, dimer ester III), and dimer ester II. These experiments also underscore the role of alcohol volatility in dimer ester production given that the less volatile bPdiol yielded a higher abundance of dimer ester I than aPdiol yielded dimer ester III, although the impact of structural differences between the diols cannot be discounted. The two most abundant dimers identified in SOA from a-pinene ozonolysis (Fig. 2, dashed boxes) (34) have been the subject of extensive study (9, 12, 13, 15–18, 21, 25–30, 37, 38, 43, 44) and are both proposed to consist of esters with cis-pinic acid subunits. Together with dimer ester II, their abundances were affected by the formation of dimer esters I and III in a-pinene and b-pinene ozonolysis experiments with CHX and either bPdiol or aPdiol (Fig. 2, A and B). 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
789
RES EARCH | R E S E A R C H A R T I C L E S
A
100
185
P + O3 + CHX P + O3 + MeOH
(C9H14O4)
367
80
(C19H28O7)
357
(C17H26O8)
Percent
60
O
O
O
O
40
OH
O
HO
199
(C10H16O4)
*
20
*
*
0 3.50
B
4.00
4.50
O
O
5.00
OH
O
R
OH
O
O
C1
O R1
OOH
6.50
7.00
7.50
H2O
+
O
O
O
+
6.00
O R OH
O
HOO
5.50
Time (min)
R2
OH
R1
H
O O
H
R3
O
OH
R2
R1
O
+
R3
O HO
+
R2
H 2O
C2 O R1
OH O O
H
R3
R2
OH OH
R1 R3
O
O O
OH H
R2
O R1
O
R3
+ HO
Fig. 4. Formation mechanism of dimer esters. (A) BPI chromatograms of SOA formed from ozonolysis of aP after ~4 hours of reaction in the CTEC in the presence of either CHX or MeOH as an OH scavenger. Numbers correspond to nominal m/z values of [MÐH]− ions, and molecular formulas are given in parentheses. Chromatograms are normalized to the TOC content of the corresponding SOA filter samples, reported as averages of duplicate samples collected in parallel for each experiment, and scaled such that the largest peak in the control experiments (gray shading) is 100%. cis-Pinic acid (C9H14O4), OH-pinonic acid (C10H16O4), dimer
This behavior implies that the two dimers are also formed from the cis-pinic acid derivative, the particle-phase abundance of which limits ester production. Consistent with expectations, b-pinene ozonolysis with CHX carried out in the presence of synthesized cis-10-hydroxypinonic acid (OH-pinonic) (Fig. 2C), an abundant constituent of SOA from a-pinene ozonolysis (34) not produced by ozonolysis of b-pinene, was observed to yield the corresponding major dimer (Fig. 2C, dimer ester IV). Conversely, detected amounts of dimer ester IV were negligible in CTEC experiments featuring cispinic acid and OH-pinonic acid (fig. S7). These results establish that the well-characterized dimer ester IV is not formed through conventional esterification but through particlephase accretion of OH-pinonic acid and the cis-pinic acid derivative. Structures of dimer esters
The CTEC experiments conclusively demonstrate that dimer esters I to IV are formed through particle-phase reaction of a cis-pinic 790
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
O
OH R2 O
H
R1
O
R3
+
O HO
R2
+
H2O
ester II, and dimer ester IV were identified through comparison with authentic standards. Asterisks indicate dimers (C19H28O9 and C18H26O6) proposed to consist of esters with cis-pinic acid subunits based on MS/MS analysis. (B) Proposed formation mechanism of dimer esters in pinene SOA through particle-phase nucleophilic addition of a semivolatile or low-volatility alcohol to the cyclic acylperoxyhemiacetal derived from particle-phase tautomerization of cis-3peroxypinalic acid. (C) General mechanism of dimer ester formation through condensed-phase reaction of an alcohol with an acylperoxyhemiacetal.
acid derivative and the corresponding semivolatile alcohol. Because cis-pinic acid is an asymmetric dicarboxylic acid, however, structures containing either a primary or secondary ester are possible for each dimer ester (Fig. 3) and cannot be definitively resolved by MS/MS analysis. To address this ambiguity, primary and secondary esters of cis-pinic acid and the identified alcohol subunits of dimer esters I to IV were prepared using modular synthetic strategies in 10 to 13% yield [8 steps, longest linear sequence (LLS)] and 5 to 17% yield (6 to 8 steps, LLS), respectively (fig. S8). Comparison of the LC retention times and MS/MS fragmentation patterns of dimer esters I to IV with those of the synthesized primary and secondary esters (Fig. 3) reveals that dimer ester production from the cis-pinic acid derivative is regioselective and forms the more sterically hindered secondary ester in each case. To our knowledge, this represents the first synthesis of authentic standards of dimer esters identified in SOA from ozonolysis of a-pinene and b-pinene, including the extensively studied dimer ester IV.
Formation mechanism of dimer esters
To assess the reactivity of the cis-pinic acid derivative, a-pinene ozonolysis experiments with CHX were conducted in which bPdiol was added either before or 10 hours after the onset of ozonolysis (fig. S9). As before (Fig. 2B), considerable formation of dimer ester I was observed in the former experiment, whereas only trace amounts of dimer ester I were detected in the latter. The small but nonnegligible yield of dimer ester I with delayed bPdiol addition indicates that the cis-pinic acid derivative is a short-lived and most likely closed-shell species, which had largely reacted away over the 10-hour interval but was still present in trace quantities to react with bPdiol. As a means of elucidating a central feature of the dimerization reaction—namely which monomeric species contributes the O atom in the ester linkage—b-pinene ozonolysis with CHX was carried out in the presence of synthesized 18OH-pinonic acid, labeled at the hydroxy group (fig. S10). Formation of 18O-labeled dimer ester IV at the same retention time as its 16O science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
isotopologue establishes that the ester O atom in dimer ester IV, and by extension dimer esters I to III, originates from the alcohol and suggests that the particle-phase accretion reaction proceeds through nucleophilic addition of the semivolatile or low-volatility alcohol to the reactive cis-pinic acid derivative. Additional mechanistic evidence is provided by ozonolysis experiments with CHX and bPdiol featuring one of two synthesized unsaturated carbonyl isomers (C11H18O), an enone or enal, that respectively produce only one of the two Criegee intermediates (CIs) on ozonolysis that form concurrently from ozonolysis of a-pinene (fig. S11). As anticipated, dimer esters I, II, and IV, as well as the major C17H26O8 dimer, were observed to form only from ozonolysis of the enal, given that only the CI arising from the enal is understood to yield the acyl peroxy radical common to a-pinene and b-pinene ozonolysis (fig. S6) from which the cis-pinic acid derivative is hypothesized to stem. In light of the inferred short lifetime of the cis-pinic acid derivative, of the same order as those measured for organic peroxides in SOA from a-pinene ozonolysis (45), cis-peroxypinic acid was investigated as a likely candidate. CTEC experiments with synthesized cis-peroxypinic acid and either bPdiol or OH-pinonic acid proved inconclusive owing to the instability of the peracid standard. However, detection of cis-peroxypinic acid in SOA from ozonolysis of the enone but not the enal, confirmed through comparison with the authentic standard (fig. S11), implies that cis-peroxypinic acid is not involved in dimer ester formation. The key insight into the chemistry underlying dimer ester production comes from a-pinene ozonolysis experiments in which the ratio of RO2 to HO2 concentrations ([RO2]:[HO2]) was modulated through the use of either CHX or methanol (MeOH) as an OH scavenger (Fig. 4A) (46–48). In the presence of MeOH, under reduced [RO2]:[HO2] relative to CHX, higher abundances of dimer ester IV, together with three additional dimers also proposed to consist of esters with cis-pinic acid subunits based on MS/MS analysis, were observed. Because elevated HO2 concentrations shift the fate of the common acyl peroxy radical and increase the fraction that forms cis-3-peroxypinalic acid (fig. S6), the corresponding enhancement in dimer ester abundance indicates that cis-3peroxypinalic acid is either the cis-pinic acid derivative or a direct precursor. We propose that dimer esters I to IV and, by extension, dimer esters in a-pinene and b-pinene SOA that contain cis-pinic acid subunits are formed through particle-phase nucleophilic addition of the requisite semivolatile or lowvolatility alcohol to the cyclic acylperoxyhemiacetal derived from particle-phase tautomerization of cis-3-peroxypinalic acid, with subsequent decomposition of the addition product yielding the SCIENCE science.org
primary carboxylic acid and water (Fig. 4B). This mechanism accounts for the observed regioselectivity, dynamics, isotopic labeling, alcohol volatility, and [RO2]:[HO2] dependence of ester production. Acylperoxyhemiacetal formation is known to occur readily in solution (49), and it has recently been shown that cis-3peroxypinalic acid is a major gas-phase product of a-pinene ozonolysis (50). Moreover, quantum chemical calculations evaluating the reactivity of a series of carboxylic acid derivatives [CH3C(=O)–X] toward esterification with MeOH (fig. S12) reveal that the acylperoxyhemiacetal reaction is the most energetically favorable (i.e., has the lowest reaction barrier). Given the ubiquity of peracid, aldehyde, and alcohol functionalities in SOA from both biogenic and anthropogenic sources, it is expected that ester formation through condensed-phase reaction of alcohols with acylperoxyhemiacetals followed by Baeyer-Villiger decomposition of the resulting peroxyhemiacetals (Fig. 4C) represents a general route to the production of low-volatility dimeric compounds in ambient SOA. Atmospheric implications
The formation of dimer esters I to IV—shown through synthesis of authentic standards to be secondary esters of cis-pinic acid—through particle-phase accretion of alcohols with an acylperoxyhemiacetal resolves a long-standing puzzle in atmospheric aerosol chemistry and rationalizes a number of experimental and ambient observations from past studies (supplementary materials, section S2). This chemistry incorporates otherwise semivolatile oxidation products into the particle phase as low-volatility dimeric compounds, providing an important source of irreversibly condensed mass for SOA growth. Because alcohols, aldehydes, and, in turn, peracids are formed from the oxidation of almost all biogenic and anthropogenic SOA precursors (49), this chemistry is likely a general feature of SOA formation. Notably, the production of dimer esters of cis-pinic acid and meso-erythritol in a-pinene and b-pinene ozonolysis experiments with CHX featuring meso-erythritol as a surrogate for semivolatile isoprene tetrols (fig. S13) establishes this chemistry as a synergistic pathway between major oxidation products of pinene and isoprene, the most abundant nonmethane hydrocarbons emitted to the atmosphere (5). Longer aerosol residence times in the real atmosphere as compared with CTEC experiments are expected to increase the importance of this particle-phase reactivity in ambient SOA. Further investigation is needed to determine the dependence of this chemistry on aerosol physicochemical properties (e.g., pH, chemical composition, and phase state) and environmental conditions (e.g., temperature and relative humidity). Quantitative understanding of these complex interac-
tions is essential to assessing the effects of dimer ester formation on the abundance, composition, properties, and associated impacts of SOA. REFERENCES AND NOTES
1. J. L. Jimenez et al., Science 326, 1525–1529 (2009). 2. Intergovernmental Panel on Climate Change, Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, T. F. Stocker et al., Eds. (Cambridge Univ. Press, 2013). 3. A. J. Cohen et al., Lancet 389, 1907–1918 (2017). 4. R. Burnett et al., Proc. Natl. Acad. Sci. U.S.A. 115, 9592–9597 (2018). 5. A. B. Guenther et al., Geosci. Model Dev. 5, 1471–1492 (2012). 6. M. Hallquist et al., Atmos. Chem. Phys. 9, 5155–5236 (2009). 7. C. L. Heald et al., J. Geophys. Res. 113, D05211 (2008). 8. H. Zhang et al., Proc. Natl. Acad. Sci. U.S.A. 115, 2038–2043 (2018). 9. T. Hoffmann, R. Bandur, U. Marggraf, M. Linscheid, J. Geophys. Res. 103, 25569–25578 (1998). 10. M. P. Tolocka et al., Environ. Sci. Technol. 38, 1428–1434 (2004). 11. A. Reinhardt et al., Anal. Chem. 79, 4074–4082 (2007). 12. L. Müller, M.-C. Reinnig, J. Warnke, T. Hoffmann, Atmos. Chem. Phys. 8, 1423–1433 (2008). 13. L. Müller, M.-C. Reinnig, H. Hayen, T. Hoffmann, Rapid Commun. Mass Spectrom. 23, 971–979 (2009). 14. M. Camredon et al., Atmos. Chem. Phys. 10, 2893–2917 (2010). 15. F. Yasmeen et al., Atmos. Chem. Phys. 10, 9383–9392 (2010). 16. F. Yasmeen et al., J. Mass Spectrom. 46, 425–442 (2011). 17. F. Yasmeen et al., Environ. Chem. 9, 236–246 (2012). 18. Y. Gao, W. A. Hall IV, M. V. Johnston, Environ. Sci. Technol. 44, 7897–7902 (2010). 19. W. A. Hall IV, M. V. Johnston, Aerosol Sci. Technol. 45, 37–45 (2011). 20. A. L. Putman et al., Atmos. Environ. 46, 164–172 (2012). 21. B. Witkowski, T. Gierczak, Atmos. Environ. 95, 59–70 (2014). 22. I. Kourtchev et al., Atmos. Chem. Phys. 14, 2155–2167 (2014). 23. I. Kourtchev et al., Atmos. Chem. Phys. 15, 5683–5695 (2015). 24. I. Kourtchev et al., Sci. Rep. 6, 35038 (2016). 25. K. Kristensen et al., Atmos. Chem. Phys. 13, 3763–3776 (2013). 26. K. Kristensen et al., Atmos. Chem. Phys. 14, 4201–4218 (2014). 27. K. Kristensen et al., Environ. Sci. Technol. Lett. 3, 280–285 (2016). 28. K. Kristensen, L. N. Jensen, M. Glasius, M. Bilde, Environ. Sci. Process. Impacts 19, 1220–1234 (2017). 29. K. Kristensen et al., Atmos. Chem. Phys. 20, 12549–12567 (2020). 30. X. Zhang et al., Proc. Natl. Acad. Sci. U.S.A. 112, 14168–14173 (2015). 31. K. Sato et al., Atmos. Environ. 130, 127–135 (2016). 32. A. Mutzel, M. Rodigast, Y. Iinuma, O. Böge, H. Herrmann, Atmos. Environ. 130, 136–144 (2016). 33. C. M. Kenseth et al., Proc. Natl. Acad. Sci. U.S.A. 115, 8301–8306 (2018). 34. C. M. Kenseth et al., Environ. Sci. Technol. 54, 12829–12839 (2020). 35. R. Zhao, C. M. Kenseth, Y. Huang, N. F. Dalleska, J. H. Seinfeld, Environ. Sci. Technol. 52, 2108–2117 (2018). 36. Y. Huang, C. M. Kenseth, N. F. Dalleska, J. H. Seinfeld, Environ. Sci. Technol. 54, 13238–13248 (2020). 37. A. Kahnt et al., Atmos. Chem. Phys. 18, 8453–8467 (2018). 38. Y. Zhao et al., Environ. Sci. Technol. 56, 14249–14261 (2022). 39. O. Peräkylä et al., J. Am. Chem. Soc. 145, 7780–7790 (2023). 40. M. A. Upshur et al., Nat. Prod. Rep. 40, 890–921 (2023). 41. Materials and methods are available as supplementary materials online. 42. J. F. Pankow, Atmos. Environ. 28, 185–188 (1994). 43. Y. Iinuma, S. Ramasamy, K. Sato, A. Kołodziejczyk, R. Szmigielski, Atmosphere 12, 17 (2021). 44. M. Beck, T. Hoffmann, Atmos. Environ. 130, 120–126 (2016). 45. M. Krapf et al., Chem 1, 603–616 (2016). 46. K. S. Docherty, P. J. Ziemann, Aerosol Sci. Technol. 37, 877–891 (2003). 47. M. D. Keywood et al., Environ. Sci. Technol. 38, 3343–3350 (2004). 48. D. M. Bell et al., Environ. Sci. Atmos. 3, 115–123 (2023). 49. P. J. Ziemann, R. Atkinson, Chem. Soc. Rev. 41, 6582–6605 (2012). 50. W. Zhang, H. Zhang, Anal. Chem. 93, 8595–8602 (2021).
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
791
RES EARCH | R E S E A R C H A R T I C L E S
ACKN OW LEDG MEN TS We thank J. D. Crounse and J. A. Thornton for useful discussions and the Thomson group at Northwestern University for providing one of the dimer ester standards. UPLC/(−)ESI-Q-TOF-MS was performed in the Resnick Sustainability Institute Water and Environment Lab at the California Institute of Technology. Funding: This work was supported by the National Science Foundation (AGS-1523500, CHE-1800511, and CHE-1905340), the Independent Research Fund Denmark (9040-00142B), the Villum Fonden (VIL50443), and the Alfred P. Sloan Foundation (G-2019-12281). C.M.K. acknowledges support from a National Science Foundation Atmospheric and Geospace Sciences Postdoctoral Research Fellowship (AGS-2132296).
Author contributions: C.M.K. designed research; C.M.K. and Y.H. performed research; C.M.K., N.J.H., S.P.R., and B.M.S. contributed new reagents; J.C. and H.G.K. performed theoretical calculations; C.M.K., J.C., Y.H., N.F.D., and P.O.W. analyzed data; and C.M.K., J.H.S., and P.O.W. wrote the paper. Competing interests: The authors declare no competing interests. Data and materials availability: All data are available in the main text or supplementary materials. License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/about/sciencelicenses-journal-article-reuse
SOLID-STATE PHYSICS
Emergent symmetry in a low-dimensional superconductor on the edge of Mottness P. Chudzinski1,2†*, M. Berben3,4†, Xiaofeng Xu5, N. Wakeham6, B. Bernáth3,4, C. Duffy3,4, R. D. H. Hinlopen7, Yu-Te Hsu3,4, S. Wiedmann3,4, P. Tinnemans4, Rongying Jin8, M. Greenblatt9, N. E. Hussey3,4,7* Upon cooling, condensed-matter systems typically transition into states of lower symmetry. The converse—i.e., the emergence of higher symmetry at lower temperatures—is extremely rare. In this work, we show how an unusually isotropic magnetoresistance in the highly anisotropic, one-dimensional conductor Li0.9Mo6O17 and its temperature dependence can be interpreted as a renormalization group (RG) flow toward a so-called separatrix. This approach is equivalent to an emergent symmetry in the system. The existence of two distinct ground states, Mott insulator and superconductor, can then be traced back to two opposing RG trajectories. By establishing a direct link between quantum field theory and an experimentally measurable quantity, we uncover a path through which emergent symmetry might be identified in other candidate materials.
S
ymmetry is one of the most inspiring concepts in mathematics and physics, with an impact that spans multiple disciplines from art to applied physics and chemistry. In the context of quantum mechanics, symmetry enables us to identify conservation laws (through Noether’s theorem) and to group states according to their symmetry classes. Frequently, this is the only strictly exact information that we have about a given many-body system. Among the plethora of symmetry-related phenomena, the notion of symmetry breaking (i.e., the lowering of symmetry) plays a particularly prominent role. It provides us with a framework to understand phase transitions in condensed-matter systems (1) and, on a more fundamental level, helps explain mass generation through the Higgs
1
School of Mathematics and Physics, Queen’s University Belfast, Belfast, UK. 2Institute of Fundamental Technological Research, Polish Academy of Sciences, Warsaw, Poland. 3High Field Magnet Laboratory (HFML-EMFL), Radboud University, Nijmegen, Netherlands. 4Institute for Molecules and Materials, Radboud University, Nijmegen, Netherlands. 5Key Laboratory of Quantum Precision Measurement of Zhejiang Province, Department of Applied Physics, Zhejiang University of Technology, Hangzhou, China. 6Center for Space Sciences and Technology, University of Maryland Baltimore, Baltimore, MD, USA. 7H. H. Wills Physics Laboratory, University of Bristol, Bristol, UK. 8Center for Experimental Nanoscale Physics, Department of Physics and Astronomy, University of South Carolina, Columbia, SC, USA. 9Department of Chemistry and Chemical Biology, Rutgers University, Piscataway, NJ, USA. *Corresponding author. Email: [email protected] (P.C.); [email protected] (N.E.H.) †These authors contributed equally to this work.
792
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
mechanism (2). Recently, a counter idea has appeared (3): Can the symmetry of the system become higher upon decreasing the temperature? The name emergent symmetry reflects the fact that a more symmetric state emerges at low energies from an initial high-energy state that does not explicitly reveal such symmetry. The idea that mutual cooperation of various strong correlations can lead to a lowenergy state of higher symmetry has become one of the cornerstones for understanding the exotic ordering in general and indicates that multiple order parameters may be united (4) under the umbrella of a single parent state (5). Despite sustained interest, it has proved very difficult to obtain such a state experimentally, with the notable exception of interacting spins on an insulating one-dimensional (1D) chain tuned by a magnetic field (6, 7). In condensed matter, there is always the problem of additional symmetry-breaking terms becoming relevant—e.g., upon Mott gap or pseudogap opening—and obscuring the detectability of any emergent symmetry. As a result, the main experimental focus has shifted to coldatom systems that offer full control over the parameters of the model (8), thereby providing hope of engineering any quantum simulator, albeit with the added difficulty of reaching sufficiently low temperatures. In any many-body system with a continuous spectrum, the notion of symmetry can be rather subtle. Let us imagine a system of spins
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adi0857 Materials and Methods Supplementary Text Figs. S1 to S17 Table S1 Synthetic Procedures NMR and IR Spectra References (51–104) Submitted 6 April 2023; accepted 11 October 2023 10.1126/science.adi0857
freely fluctuating in plane. In the charge sector, this corresponds to freely flowing charges described by a free-particle theory. Adding an Ising-type perturbation g can drive the system toward localization, but just before it enters the gapped state, there is a special point in parameter space where the symmetry is enlarged— i.e., where the spins can rotate equally well in and out of plane. Likewise, the charge can become equally localized and itinerant, spanning the entire manifold of available states. Checking whether the symmetry is global (i.e., holding at all length and timescales) is an immense task. If the corresponding quantum field theory (QFT) is renormalizable, however, then we can apply renormalization group (RG) methods and inspect the corresponding RG flow (Fig. 1A)—that is, whether the perturbation g increases or decreases as we integrate out the highest energy degrees of freedom. If g is unaffected, then the system must be at the special point (the so-called separatrix in RG parlance), and thus we have proof of such hidden or emergent symmetry. Physical realization of emergent symmetry
The challenge, however, is to access this flow experimentally. Our proposed solution is a model in which g depends on an external (e.g., magnetic, electric, or strain) field X, whereas the hydrodynamic, free theory—with its velocities vi, compressibilities Ki, etc.—stays unaffected. Because the system is a perfect conductor in a hydrodynamic description, its resistivity r can be expressed as g ðX Þn Mðvi ; Ki Þ, where n is a finite number and M is a correlation function for the free theory.Mðvi ; Ki Þ is usually a complicated functional, a result of years of study aimed at solving a given QFT, but it has one important feature: It does not depend on g and so is independent of X. In this circumstance, the relative change in the resistivity in response to the external field becomes dependent solely on g and the dependence of g on any other variable parameter, such as temperature. In this work, we consider a correlated metal in which all of these theoretical conditions are seemingly met, with magnetic field H playing the role of X. Li0.9Mo6O17 (LMO) is a lowdimensional system that both lies close to the Mott transition (9, 10) and can host superconductivity (11, 12). The fact that two distinct science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
Fig. 1. Emergent symmetry in a 1D metal. (A) Parametric plot of tentative RG flow of the system n o on the g3 ½l; Kr Kr ½l plane, where Kr Kr is the distance from the critical point for the TLL parameter Kr. With lowering energy, the standard (BKT) trajectory (plus incommensurability) drives the system away from the separatrix (green dashed line) and toward either the gapless, metallic state or the gapped (Mott) insulating state (black dotted lines). (B) The emergence of the highly relevant, random component of g3 steers the system back toward to the separatrix. (C) Scattering of fermions in the charge channel (irrespective of spin orientation): In real space, an electron (blue) or hole (red) propagating in the lattice (purple circles) encounters a slower obstacle particle (green) and may either undergo forward scattering (f-outgoing, dashed arrows) or umklapp scattering (u-outgoing solid arrows). In real space, the arrows correspond to currents of probability. For the charge-SU(2) symmetric case, the amplitudes of both processes (f/u) are equal for both electrons and holes. (Inset) Reciprocal space image of the same processes. Summing over all f-events gives a term proportional to a gradient of the charge density bosonic field (times Kr), whereas summing over all u-events gives a cosine of this field (times g3).
ground states appear so close energetically raises the prospect of bringing the two states to a degenerate point (the separatrix in Fig. 1A) and thereby realizing emergent symmetry. Being electronically 1D (12), LMO is also a viable candidate for the realization of the 1D QFT solution known as the TomonagaLuttinger liquid (TLL) state. This is promising for several reasons. First, the exact solution for the TLL is well established, and thus its manifestation can be robustly verified in LMO if and only if g[X] behaves the same for resistivities along different geometrical directions. Second, LMO is nearly ¼ filled, and its parametrization implies that the relevant TLL (compressibility) parameter Kr must be very close to the critical value Kr∗ ¼ 1 =4 , thus placing LMO close to the RG flow separatrix (9). Third, as described below, it has already been shown (13, 14) that the remaining incommensurability in LMO (from the noninteger Li) can be compensated by an auxiliary potential whose amplitude depends strongly on the magnitude of H. This indicates not only that the relevant perturbation (labeled hereafter g3 to reflect the fact that it is umklapp-like) has a strong H dependence, but it also provides a SCIENCE science.org
mechanism to steer the system toward the separatrix (Fig. 1B). At this special point, the charge has equal probability of being localized (back-scattered) or itinerant (forward-scattered), as shown schematically in Fig. 1C. Below, we provide evidence for this approach to the separatrix in LMO. Structurally, LMO comprises stacks of conducting chains (Fig. 2A). A pair of 1D dxy bands cross the Fermi level close to commensurate ¼ filling. Only the slight incommensurability (from the noninteger Li) prevents the Mott gap from forming, and at high T, LMO exhibits metallic (T-linear) resistivity. Although proximity to commensurability makes the status quo fragile, in the canonical Berezinskii-KosterlitzThouless (BKT) picture of phase transitions in a TLL with a cosine perturbation (the sineGordon model), the expectation is that g3 will become increasingly irrelevant with lowering T and that the metallic state will be preserved, as indicated by the black dotted line in Fig. 1B. One of the distinctive elements of the LMO band structure is the propensity of the other 4d bands to form (dark) excitons (Fig. 2C) (13). These excitons introduce new scattering centers below a temperature scale TH ≈ 100 to
150 K and, through these, an auxiliary potential (Fig. 2D, red crosses) that divides the chains into segments of randomized lengths, thereby bringing the mobile carriers closer to or further away from commensurability. Indirect evidence for these excitons and their influence on the mobile carriers was reported in a recent angledependent magneto-resistance study (14). As a result of this interaction, the amplitude of g3 also becomes randomized. We argue that the emergence of this highly relevant, random component of g3 is the key perturbation that brings the system back toward the separatrix. In those chain segments closest to the separatrix, the amplitudes of backward and forward scattering of fermions (whether they be electrons or holes) become equal (Fig. 1C) as will the likelihood of any instability associated with these scattering amplitudes. Whereas backward scattering favors instability and a gapped phase with ℤ2 symmetry between occupied and unoccupied sites, forward scattering favors a metal with O(2) symmetry with a free choice of gauge field. Together at the separatrix, they form a state with broader SU(2) symmetry (15). In this way, an approach to the separatrix can be interpreted as a manifestation of emergent symmetry—the symmetry in question being the equal probability to form a Mott-localized phase or a metallic and ultimately superconducting (SC) phase. Obtaining experimental evidence for such cooperative “steering” rests on finding a quantity that will directly track the RG flow of g3 itself. In the present case, the relevant quantity is the magnetoresistance (MR). The scattering centers (excitons) randomly modify the amplitude of g3, and the randomness of this effect increases with H, resulting in a positive normal state (i.e., non-SC state) MR. This fieldinduced effect enables us to detect when the RG flow of the system becomes steered toward the separatrix (Fig. 1B). As we will show below, for the specific circumstances found in LMO, a pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi linear T dependence of rð0Þ=DrðH Þ —the inverse square root MR—is expected with an offset that is proportional to its initial distance from the separatrix. Experimental evidence for emergent symmetry
Having sketched out the hypothesis, we now turn to the experimental study. Figure 3, A, C, and E, shows, respectively, the in-chain resistivity rb(T) for two SC single crystals and one non-SC crystal of LMO. In all three samples, rb(T) is T-linear down to T = TH (marked by vertical arrows in Fig. 3, A, C, and E), below which rb(T) develops upward curvature. At a lower temperature scale, Tmin ≈ 25 to 30 K, rb(T) passes through a minimum [the origin of which has remained a mystery for several decades (16–19)] before reaching a maximum at the onset of superconductivity or diverging (in the case of the insulating sample). The lack of 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
793
RES EARCH | R E S E A R C H A R T I C L E S
any evidence for lattice dynamics associated with a charge density wave below Tmin (18), as well as the divergent Lorenz ratio (20) (see below), allows us to exclude electron-phonon scattering as the origin of the T-linear resistivity in LMO. According to TLL theory, T-linear r(T) implies that the TLL parameter Kr = ¼. The upturn at Tmin then identifies the energy scale at which the system evolves from a regime dominated by scattering (on excitonic fluctuations) to one dominated by tunneling (through excitonic roadblocks) (21). Whereas the correlation function M and its T dependence change at this point, crucially, the functional dependence of the resistivity on g3 stays the same. As a result, the relation between the MR and g3 is unaffected across Tmin. At low fields, the transverse MR (H//c) of all three samples is positive and varies as H2 over the entire temperature range studied (22). Taking the initial slopes A of the H2 MR at each T (normalized to 1 T), obtain theffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi we pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi resultant plots of rð0Þ=A ¼ rð0Þ=DrðH Þ for the three samples shown, respectively, in Fig. 3, B, D, and F. Despite the complex and pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi varied form of rb(T), rð0Þ=DrðH Þ displays a simple T-linear dependence from 300 K down to 2 K with an absolute magnitude that is comparable in all three samples. Notably, the magnitude of DrðH Þ=rð0Þ is orders of magnitude larger than what one would expect from Boltzmann transport theory (22). Moreover, as shown in Fig. ffi3H, the form and magnitude of p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rð0Þ=DrðH Þ are found to be markedly insensitive to the relative orientation of the current and/or the applied field [differing by only a factor of 4, whereas r(0) along the different crystallographic axes differs by 3 orders of magnitude], which is in marked contrast with expectations for and observations (23) in a quasi-1D Fermi-liquid. Such anomalies compel us to seek an explanation for this MR behavior that lies beyond standard Boltzmann theory. In a TLL, the resistivity can be expressed as an amplitude for back-scattering multiplied by a correlation function of the bosonic fields rðT ; H Þ ¼ g32 ðT ; H ÞMij ðT Þ
ð1Þ
The correlation function Mij contains information about multiple parameters associated with the TLL, including the velocity of the charge mode vr+ and the corresponding TLL parameter Kr+. Any field dependence in Kr+ would lead to a nonmonotonous and rapidly changing Mij and, in turn, a very complicated MR signal. This is not what we see experimentally. Thus, we can assume that Kr+ and vr+ are independent of H and that the MR depends only on g3 dr=dH ¼ 2g3 ðdg3 =dH ÞMij ðT Þ
ð2Þ
and there is one dominant mechanism scattering the 1D carriers out of their chiral trajectory. To obtain the low-field MR, we calculate 794
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
Fig. 2. Crystallographic and electronic structure of LMO. (A) 3D crystal structure showing isolated, conducting, and zigzag chains of MoO6 octahedra along the b axis in dark purple and nonconducting octahedra and MoO4 tetrahedra in light orange. Li ions are shown as green spheres. (B) Simplified Fermi surface of LMO showing the weakly dispersive, quasi-1D bands along the a axis caused by the weak interchain hopping energy. Note that the actual Fermi surface is close to being ½-filled owing to the zigzag nature of the chains. (C) This small interchain hopping nevertheless causes an energy gap around the Fermi surface (schematic orange curves). The small energy gap easily allows excitation of electron-hole pairsÑi.e., excitons. (D) Schematic structure of LMO including the 1D chains (boxes) and lattice periodicity (circles). Below TH ≈ 150 K, the system couples to an auxiliary potential (crosses) associated with these excitons, which tends to fix its small incommensurability and hence reactivate the umklapp g3 terms. Owing to the random distribution of scattering centers, different regions fall closer to or further away from incommensurability, as indicated by the intensity of the yellow background. As a result, the amplitude of g3 also becomes randomized.
DrðH Þ ¼ ðdr=dHÞ DH and then divide by the zero-field resistivity r(0) to obtain DrðH Þ dg3 ¼ 2@H DH rð0Þ g3 ¼ 2@H fln½g3 ðH ÞgDH
ð3Þ
Hence, if only g3 depends on H, Mij —the most complicated term in r(T,H)—drops out of the expression for DrðH Þ=rð0Þ. The Mij functions are different for rxx, ryy, and rzz, but the
g32 prefactor remains the same. From this, one deduces that in the TLL framework, DrðH Þ= rð0Þ exhibits similar behavior for all orientations of current and field with only the prefactor depending on H. Thus, the MR becomes effectively the same for all orientations, as observed (Fig. 3H), which confirms the validity of the TLL description. Because the running variable l of the RG can be linked with temperature ½l∼ lnðT =LÞ (where L is the high-energy cutoff in the science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
system), the T dependence of the MR also tracks the RG flow of g3[l] and thus allows direct access to the RG trajectory itself (Fig. 1B). More precisely, the inverse square ratio of pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi the MR rð0; T Þ=DrðH; T Þ ∼ @H ðlnfg3 ½H; l ¼ lnðT =LÞgÞ 1=2 . This statement is actually quite strong: No matter what the prefactor’s dependence on g3 (it might even change as the system goes from one conductivity regime to another), the MR will always depend on this logarithmic derivative. Therefore, if there exists any characteristic energy scale l0 below which the system opens a gap (or alternatively, g3 disappears exponentially), then this must be reflected in the T dependence of@l fln½g3 ðl Þg 1=2. The only way this quantity can stay linear is on approach to the separatrix. Within this picture, based on a modification of the known BKT flow (Fig. 3G,p inset), the offset in the MR [the interffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi cept in rð0; T Þ=DrðH; T Þ at zero temperature] can be related to an initial distance from the separatrix before g3 randomness emerges, as illustrated in Fig. 3G, where we observe that the system tends to move closer to the separatrix with decreasing temperature—an opposite trend to that usually found in a doped 1D Mott system (Fig. 1A) (24). This explains one further notable detail of the experimental data. The 0 K intercept in the more metallic (i.e., SC) samples is finite (Fig. 3, B, D, and H, insets), whereas in the non-SC sample with a divergent resistivity, it is effectively zero (Fig. 3F, inset), in agreement with the expectation that the latter lies closer to the separatrix (i.e., closer to the Mott state) already at elevated T. This in turn suggests that SU(2) symmetry in the charge sector does not appear accidentally (e.g., owing to a coincidence of parameters in LMO) but rather emerges during the RG flow itself. The validity of this interpretation of the MR rests on two conjectures: (i) that the only term responsible for finite resistance is g3 and (ii) that in the expression for r(T,H), only the amplitude of g3 is H dependent, whereas the TLL parameters Ki are not (i is the TLL mode index, e.g., r+). Two further transport properties determined for SC LMO—the Hall coefficient RH(T) (25, 26) and the Lorenz ratio L = k/sT of the thermal k to the electrical s conductivity (20, 27)—appear to confirm these conjectures. First, RH(T) exhibits a marked T dependence (Fig. 4A), whose dominant contribution is a single power law that is the same for all fields. In TLL theory, the value of the power-law exponent depends on Ki, and the prefactor depends on g3. The fact that RH(T) fits at different field strengths contain the same exponent allows us to infer that the only field dependence is indeed in g3. Because k contains all back-scattering terms present in the system, the notable coincidence of L(T) and RH (T), shown in Fig. 4B, informs us that neither lattice nor neutral bosonic modes contribute to the resistivity—only g3 contributes (22). SCIENCE science.org
Fig. 3. MR as a probe of emergent symmetry in LMO. (A and C) In-chain resistivity rb(T) for two SC samples with Tc = 2.15 K (A) and 2.2 K (C). Dashed lines are linear fits to the high-T data, and colored arrows indicate roughly where rb(T) deviates from linearity. The inset in (C) shows a zoomed-in view near Tc of rc(T) taken on a piece of the same crystal (12). (E) rb(T) for a non-SC sample. The deviation from the pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi high-T T-linear behavior (dashed line) sets in at TH ≈ 150 K. (B, D, and F) rð0Þ=DrðHÞ extracted from the inverse square root of the coefficient A of the low-field H2 b axis MR [and divided by the zero-field resistivity r(0)] for the samples shown in (A), (C), and (E), respectively. Dashed lines highlight the T-linearity pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi of rð0Þ=DrðHÞ. Vertical arrows again indicate where rb(T) deviates from linearity [as shown in (A)]. (Insets) Zoomed-in views of the low-T region for each sample. Note that for both SC samples [(B) and (D)], the intercept is finite, whereas for the non-SC sample (F), it is negligible, as expected were it to locate closer to the separatrix (see text). (G) Connection between the MR, as expressed through the derivative @H fln½g3 ðHÞg, and RG flow. (Inset) Parametric plot of g3[l] flow as it gradually approaches the separatrixÑthe lower the line at Kr Kr e 0:1, the further away the system is initially from the separatrix [green dashed pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi line in (B)]. (Main panel) Resultant T dependence of @H fln½g3 ðHÞg 1=2 e rð0Þ=DrðHÞ (for the yellow and blue trajectories) to be compared with experiment. The closer that the system is initially to the separatrix, the lower the intercept at zero temperature. [Corresponding curves for the green and red trajectories in the pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi inset (not shown in this figure) simply have a larger offset.] Arb. Units, arbitrary units. (H) rð0Þ=DrðHÞ plot for various SC samples (Tc = 2.0 ± 0.2 K) with different orientations of current and field, normalized to their (extrapolated) absolute value at 300 K. The normalization factors are (in parentheses): I//a; H//c (64); I//c; H//a (119); I//c; H//b (150); I//c; and H//c (160). (Inset) Zoomed-in view of the low-T region showing, again, the finite intercept for SC samples. 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
795
RES EARCH | R E S E A R C H A R T I C L E S
The experimental result for RH(T) not only confirms that the charge TLL parameters are H independent, but it also identifies the mechanism steering the system closer toward the separatrix. Specifically, we find that the powerlaw exponent in RH(T) below 150 K closely matches the characteristic value for random umklapp processes. This drives the emergent symmetry in LMO; being more relevant than standard umklapp processes, these random processes push the system toward the Mott phase (Fig. 1B). At the same time, however, they can only be defined within the metallic phase, when the electrons are able to propagate freely and explore various regions with differing strengths of umklapp events. As a result, the scattering becomes a self-limiting process, whose asymptote is located right on the separatrix of the flow. Discussion and outlook
We make the following minimal statements on the basis of our experimental findings. The monotonous, single–power law T dependence of the MR indicates that there is only one mechanism of scattering (otherwise it would have to result from some serendipitous compensation of various terms), and its overall isotropy confirms that our choice of 1D QFT is the correct description (coherent motion is only along one direction, and we measure this scattering vertex independently of the sample orientation). This and supporting evidence from the Hall effect and the Lorenz ratio (that Kr does not depend on H) imply that LMO is a good realization of the QFT proposed above, and the conjecture that the MR reveals the RG flow of g3[l] seems valid. Because the T dependence of the (inverse square root) MR is purely linear, the RG flow must stay on the separatrix down to the SC Tc , the latter providing sufficient proof of the emergent symmetry. What is notable is the fact that LMO not only hosts these two ground states (Mott insulating and SC) but that the RG trajectories extracted from a MR study are capable of distinguishing between them. The relation between the emergence of superconductivity and symmetry is now more transparent. Mottness is a tendency of charges to localize—a tendency that suppresses any SC instability by diminishing the spectral weight available for condensation. Emergent symmetry, by creating a larger manifold of degenerate states, provides a linear combination of states capable of evading localization and thus enhances the spectral weight for superconductivity. The experimental realization of emergent symmetry in SC LMO may also have implications for our understanding of other unconventional superconductors proximate to a Mott state, such as the high-Tc cuprates, the quantum spin liquids, and the two-leg ladders. Superconductivity at the edge of Mottness is 796
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
Fig. 4. Hall effect and Lorenz ratio in LMO. (A) RH(T) of SC LMO (Tc = 2.15 K) at two different field strengths: 1.5 T (blue circles) and 8.0 T (red circles) plotted on a log-log scale. Note the strengthening field dependence of RH below 30 K. The dashed lines are fits to the expression for a TLL (22): RH ðT Þ ¼ R0H ð1 þ aT m Þ, where R0H ¼ 1:5mm3 =C represents the band value of the Hall coefficient, m = −1.5, and a = 3600 and 2000 for m0H = 1.5 and 8.0 T, respectively. (B) Comparison of the low-field RH(T) (above Tmin ≈ 25 K) (circles) and the zero-field Lorenz ratio L/L0 (squares) measured on different SC LMO crystals taken from the same growth batch (where the Lorenz number L0 = 2.44 × 10−8 V2 K−2). The Lorenz data are taken from Wakeham et al. (20).
an unsolved problem, largely because the simplest model to treat strong correlations, the Hubbard model, is intractable in dimensions greater than one. In LMO, however, TLL theory provides a robust theoretical footing from which to explore the origins of pair condensation on the border of localization. Although the degeneracy or near-degeneracy of multiple ground states provides a natural setting for emergent symmetry to occur (5, 28), the role played by fluctuations between such states in promoting pairing is an interesting avenue for future research. Finally, the role of disorder in the vicinity of the transition that we have been able to uncover is a very general aspect that connects to other areas, including Griffiths phases (29) in systems with generic slow dynamics (30). In this regard, the notion that disorder or randomness on the interaction level can drive a system toward the separatrix (between metallicity and Mott localization) sheds light on this complex transition. Notably, the notion itself is not specific to 1D systems. RE FERENCES AND NOTES
1. P. Coleman, Introduction to Many-Body Physics (Cambridge Univ. Press, 2015). 2. P. W. Anderson, Phys. Rev. 130, 439–442 (1963).
3. D. González-Cuadra, A. Bermudez, P. R. Grzybowski, M. Lewenstein, A. Dauphin, Nat. Commun. 10, 2694 (2019). 4. A. J. A. James, R. M. Konik, P. Lecheminant, N. J. Robinson, A. M. Tsvelik, Rep. Prog. Phys. 81, 046002 (2018). 5. S. C. Zhang, Science 275, 1089–1096 (1997). 6. R. Coldea et al., Science 327, 177–180 (2010). 7. H. Zou et al., Phys. Rev. Lett. 127, 077201 (2021). 8. M. A. Cazalilla, A. M. Rey, Rep. Prog. Phys. 77, 124401 (2014). 9. P. Chudzinski, T. Jarlborg, T. Giamarchi, Phys. Rev. B 86, 075147 (2012). 10. M. Nuss, M. Aichhorn, Phys. Rev. B 89, 045125 (2014). 11. M. Greenblatt, W. H. McCarroll, R. Neifeld, M. Croft, J. V. Waszczak, Solid State Commun. 51, 671–674 (1984). 12. J.-F. Mercure et al., Phys. Rev. Lett. 108, 187003 (2012). 13. P. Chudziński, Eur. Phys. J. B 90, 148 (2017). 14. J. Lu et al., Sci. Adv. 5, eaar8027 (2019). 15. M. Nakamura, Phys. Rev. B 61, 16377–16392 (2000). 16. G. Gweon et al., Phys. Rev. Lett. 85, 3985 (2000). 17. J. Choi et al., Phys. Rev. B 69, 085120 (2004). 18. C. A. M. dos Santos, B. D. White, Y.-K. Yu, J. J. Neumeier, J. A. Souza, Phys. Rev. Lett. 98, 266405 (2007). 19. X. Xu et al., Phys. Rev. Lett. 102, 206602 (2009). 20. N. Wakeham et al., Nat. Commun. 2, 396–399 (2011). 21. A. Furusaki, N. Nagaosa, Phys. Rev. B 47, 3827–3831 (1993). 22. See the supplementary materials. 23. N. E. Hussey et al., Phys. Rev. Lett. 89, 086601 (2002). 24. T. Giamarchi, Phys. Rev. B 44, 2905–2913 (1991). 25. H. Chen et al., Europhys. Lett. 89, 67010 (2010). 26. J. L. Cohn, B. D. White, C. A. M. dos Santos, J. J. Neumeier, Phys. Rev. Lett. 108, 056604 (2012). 27. C. L. Kane, M. P. A. Fisher, Phys. Rev. Lett. 76, 3192–3195 (1996). 28. X. Montiel, T. Kloss, C. Pépin, Sci. Rep. 7, 3477 (2017). 29. K. Agarwal, S. Gopalakrishnan, M. Knap, M. Müller, E. Demler, Phys. Rev. Lett. 114, 160401 (2015). 30. M. Schiulaz, M. Müller, AIP Conf. Proc. 1610, 11–23 (2014). 31. N. Hussey, Emergent symmetry in lithium molybdate (LMO), dataset, Dryad (2023); https://doi.org/10.5061/dryad. qfttdz0pj. 32. ElasticScattering, SmallAngleScatteringLMO, SmallAngleScatteringLMO/SmallAngleScatteringLMO: Small angle Boltzmann transport scattering in Li0.9Mo6O17, version v1.0, Zenodo (2023); https://doi.org/10.5281/zenodo.8252644. AC KNOWLED GME NTS
We acknowledge stimulating discussions with A. Ghosh, M. Katsnelson, M. Grüning, and M. Rösner and experimental assistance from A. F. Bangura, J.-F. Mercure, and A. Narduzzo. We also acknowledge C. Xu for assistance with preparing Fig. 2. This work was partially carried out at HFML-RU/NWO, a member of the European Magnetic Field Laboratory (EMFL). Funding: This study was supported by Netherlands Organisation for Scientific Research grant 16METL01 (N.E.H. and M.B.), the European Research Council under the European Union’s Horizon 2020 research and innovation program grant 835279-Catch-22 (N.E.H., B.B., C.D., R.D.H.H., and Y.-T.H.), European Union’s Horizon 2020 research and innovation program grant no. 847639 (P.C.), Engineering and Physical Sciences Research Council (UK) grant EP/V02986X/1 (P.C. and N.E.H.), and National Science Foundation of China grants 12274369 and 11974061 (X.X.). Author contributions: Conceptualization: N.E.H., P.C., and X.X. Methodology: P.C., M.B., X.X., and N.W. Sample synthesis and characterization: R.J., M.G., and P.T. Investigation: X.X., N.W., M.B., B.B., C.D., Y.-T.H., and S.W. Analysis: M.B., P.C., and R.D.H.H. Funding acquisition: N.E.H. and P.C. Supervision: N.E.H. and P.C. Writing – original draft: P.C., M.B., and N.E.H. with input from all coauthors. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data used to generate the figures are available at Dryad (31), and the code used to generate fig. S9 is available at Zenodo (32). License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www. science.org/about/science-licenses-journal-article-reuse SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.abp8948 Materials and Methods Supplementary Text Figs. S1 to S11 Table S1 References (33–62) Submitted 7 March 2022; accepted 29 September 2023 10.1126/science.abp8948
science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
GLOBAL WARMING
Curbing global solid waste emissions toward net-zero warming futures Zheng Xuan Hoy1, Kok Sin Woon1*, Wen Cheong Chin2, Yee Van Fan3, Seung Jick Yoo4 No global analysis has considered the warming that could be averted through improved solid waste management and how much that could contribute to meeting the Paris Agreement’s 1.5° and 2°C pathway goals or the terms of the Global Methane Pledge. With our estimated global solid waste generation of 2.56 to 3.33 billion tonnes by 2050, implementing abrupt technical and behavioral changes could result in a net-zero warming solid waste system relative to 2020, leading to 11 to 27 billion tonnes of carbon dioxide warming–equivalent emissions under the temperature limits. These changes, however, require accelerated adoption within 9 to 17 years (by 2033 to 2041) to align with the Global Methane Pledge. Rapidly reducing methane, carbon dioxide, and nitrous oxide emissions is necessary to maximize the short-term climate benefits and stop the ongoing temperature rise.
A
nthropogenic activities have caused a global surface temperature increase of 1.1°C relative to preindustrial levels as of the end of 2020 (1). Amid the continued emissions of greenhouse gases (GHGs), the world will not be on track to meet the Paris Agreement’s 1.5° and 2°C pathway goals. The global warming observed to date is mostly caused by the emissions of carbon dioxide (CO2), methane (CH4), and nitrous oxide (N2O) (1). Steep reductions in these GHGs are essential to limiting climate change. CH4 was determined to be of particular importance at the 26th United Nations Climate Change Conference of the Parties (COP26) (2) because it has a short atmospheric lifetime of approximately a decade, so cutting its emissions can rapidly decelerate near-term global warming (3). More than 100 countries have signed the Global Methane Pledge to curb 30% of global CH4 emissions by 2030, using a 2020 baseline (4). In 2022, atmospheric CH4 levels reached 1911.8 parts per billion (ppb), more than double preindustrial levels, and the recorded 14.1 ppb increase was the third fastest since the early 1980s (5). Approximately 90% of CH4 emissions from the solid waste industry, a major source of atmospheric CH4, could be eradicated by 2050 using readily available technologies (6). Behavioral changes and decision making regarding consumption patterns, dietary choices, and solid waste handling practices from separation to collection and treatment have direct implications for the successful implementation of these technologies (7). Unless immediate action is taken, municipal solid 1
New Energy Science and Engineering Department, School of Energy and Chemical Engineering, Xiamen University Malaysia, Bandar Sunsuria 43900, Malaysia. 2Department of Mathematics, Xiamen University Malaysia, Bandar Sunsuria 43900, Malaysia. 3Sustainable Process Integration Laboratory (SPIL), NETME Centre, Faculty of Mechanical Engineering, Brno University of Technology, Brno 61669, Czech Republic. 4Department of Climate and Environmental Studies, Sookmyung WomenÕs University, Seoul 04310, Korea. *Corresponding author. Email: [email protected]
SCIENCE science.org
waste (MSW)–related emissions are anticipated to nearly double by 2050 compared with 2016 (8). The rapid decarbonization of the global MSW system may provide short-term relief for negotiations to cut emissions from hard-to-abate sectors (9), including heavy-duty transport (e.g., aviation, shipping, and trucking) and heavy industries (e.g., chemical, cement, and steel manufacturing) (10). Previous assessments have explored the potential of the global MSW system to mitigate emissions (11–13); however, there has been no global analysis considering the global warming that could be averted to put us on track with achieving the
Paris Agreement goals and Global Methane Pledge (see the supplementary materials, table S1). In confronting this unprecedented climate crisis, defining future emission pathways for the global MSW system to track its progress toward long-term climate goals can assist policymakers in designing CH4 mitigation strategies that would complement CO2 and N2O abatement strategies (14). In this study, we forecasted the disaggregated GHG emissions of the global MSW system to assess its potential for alleviating global warming and meeting climate goals. We developed a historical data inventory based on a bottom-up approach for the 43 highest MSWgenerating countries, representing ~86% of global MSW generation in 2016 (8). The selected countries comprised four income groups: 14 high-income countries (HICs), 13 uppermiddle-income countries (UMICs), 13 lowermiddle-income countries (LMICs), and 3 low-income countries (LICs) (see the supplementary materials, table S2). The data inventory covered MSW generation, composition, and treatment facility allocation from 1990 to 2020. We filled the data gaps for MSW generation using a panel data regression model with gross domestic product (GDP) per capita because MSW generation primarily grows with GDP and population growth (8). We calculated GHG emissions from MSW disposal and treatment facilities during 1990 to 2020 based on the Intergovernmental Panel on
Fig. 1. Cumulative GHG s s s s emissions reduction potential of the mitigation strategies from 2020 to 2050 compared with BAU expressed in billion tonnes (Gt) CO2-we. Adoption of the mitigation strategies begins in 2023 and is to be completely adopted by 2050. The reduction potentials are colored by mitigation strategy and categorized based on four income groups: HICs, UMICs, LMICs, and LICs. The net cumulative GHG emissions, indicated by the solid black circle, is the sum of the cumulative GHG emissions under BAU and the reduction in GHG emissions under each income group. Negative cumulative CH4 emissions mean that the amount of CH4 removed from the atmosphere exceeds the quantity being emitted over 2020 to 2050. For readability, values in the figure are rounded up to the nearest whole number (with the exception of reduction in LICs); refer to data S1 for exact values. 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
797
RES EARCH | R E S E A R C H A R T I C L E S
Climate Change (IPCC)’s “2006 IPCC Guidelines for National Greenhouse Gas Inventories” (15) and considering the “2019 Refinement to the 2006 IPCC Guidelines for National Greenhouse Gas Inventories” (16). We predicted MSW generation by 2050 using the panel data regression model with projected population and GDP per capita based on Shared Socioeconomic Pathways (SSPs) (17, 18). We estimated that global MSW generation will reach 2.56 to 3.33 Gt by 2050 (see the supplementary materials, fig. S2). This growth is equivalent to a 1.4 to 2.8% average annual growth from 2020 to 2050, consistent with the World Bank’s projected global average annual growth of 2.4% (19). Considering the uncertainty in modeling GHG emissions, we applied an ensemble forecasting technique using an artificial neural network coupled with Bayesian hyperparameter optimization (20). The model used country-specific population, GDP per capita, and MSW generation to forecast the CO2, CH4, and N2O emissions from the MSW disposal and treatment for 2021 to 2050. We forecasted using a 10-member ensemble (20) and analyzed it based on the ensemble median while referring to the 5th and 95th percentiles (21). We then determined the maximum allowable cumulative GHG emissions from 2020 for the global MSW system in meeting the 1.5° and 2°C targets. We followed the 50 and 67% likelihood thresholds set in the IPCC 6th Assessment Report to err on the side of caution without attempting an unrealistic near certainty (21). The conventional GWP100 metric may misrepresent the warming potential of short-lived GHGs such as CH4 when assessing the cumulative emissions budget because its warming potential peaks in the first 20 years after its release and it has an average lifetime of approximately a decade, compared with centuries for CO2 (22). To allow both short-lived and long-lived GHG emissions to be commensurable in a common cumulative framework regardless of their warming impacts, we expressed the GHG emissions as CO2 warming-equivalents (CO2-we) using the GWP* metric (23). We determined that the global MSW system has a cumulative emissions budget of 12 and 27 Gt CO2-we to meet the 1.5° and 2°C targets, respectively, with a 50% likelihood and a cumulative emissions budget of 11 and 23 Gt CO2-we to meet the respective targets with a 67% likelihood (see the supplementary materials, table S9). Emissions pathways of the global MSW system
Our forecasts show that the current global MSW system will exceed the 1.5°C emissions budget between 2027 and 2028 and the 2°C emissions budget between 2037 and 2044 (see the supplementary materials, fig. S4). The cumulative GHG emissions from 2020 to 2050 are expected to reach 32 to 35 Gt CO2-we. 798
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
Fig. 2. Cumulative GHG emissions pathway of the mitigation strategies from 2020 to 2050 with respect to the IPCC cumulative emissions budget to achieve the 1.5° and 2°C targets expressed in Gt CO2-we. The solid line shows the ensemble median, and the shaded region shows the 5th and 95th percentiles of the 10-member ensemble. The horizontal dashed lines indicate the cumulative emissions budget to meet the 1.5° and 2°C temperature limits. The 50 and 67% chances are represented by the 50th and 67th percentiles of the transient climate response to cumulative carbon emissionsÕ normally distributed range, respectively. All strategies represent the combination of the three mitigation strategies: retrofitting landfills, composting organics, and halving waste.
Given that the global MSW system would exceed the emissions budget under business-as-usual (BAU) conditions, we explored four mitigation strategies for the abatement of GHG emissions among the four income groups: (i) retrofitting landfills with biogas capture, (ii) diverting organic waste for composting, (iii) diverting organic waste for anaerobic digestion, and (iv) halving MSW generation. These mitigation strategies were selected based on their representation of diverse situations regarding required cost, technology readiness, skilled worker requirements, and standing in the waste management hierarchy (see the supplementary materials, table S10). We emphasized our scenario analysis on the SSP2 (Middle of the Road) pathway because of its higher likelihood of occurring given the current energy policy and investment choices mapped out by the International Energy Agency (24). We assumed a decarbonizing future in which the adoption of the mitigation strategy would begin in 2023 with a linearly increasing annual emissions reduction potential rate until complete adoption in 2050. We found that individual mitigation strategies would be able to reduce the cumulative GHG emissions from 2020 to 2050 by 27 to 70% relative to BAU. Their effectiveness in descending order is as follows: digesting organ-
ics, 70%; halving waste, 63%; composting organics, 57%; and retrofitting landfills, 27% (Fig. 1). Halving waste generation and diverting organic waste to be treated biologically will result in negative cumulative CH4 emissions by 2050 (–5 to –1 Gt CO2-we) relative to 2020. Achieving these strategies will require behavioral changes at the consumer level to consciously practice waste minimization and organic waste segregation at the source. The effectiveness of the mitigation strategies varies by income group. Halving waste generation is the most effective among HICs, whereas diverting organic waste for anaerobic digestion is the most effective among the remaining income groups. The GHG emissions reduction potentials are intricately linked to income groups because of their differences in biodegradable and combustible MSW compositions (25). Driven by disparate consumption patterns and economic structures (8), HICs exhibit greater consumption of packaged goods and disposables, resulting in higher production of nonbiodegradable MSW. By contrast, UMICs, LMICs, and LICs predominantly generate biodegradable MSW because of the low availability of processed and packaged products. HICs also have better access to advanced technologies and expertise for efficient waste management practices (8), so the effectiveness of reducing science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
Fig. 3. Reduction in CH4 emissions of the mitigation strategies with respect to the Global Methane Pledge expressed in Mt CO2-eq. The CH4 emissions are evaluated in CO2-eq to estimate the relative difference in emitted amounts without considering its warming impact. The horizontal dashed line shows the threshold based on the Global Methane Pledge to reduce 30% of CH4 emissions by 2030 relative to 2020 (422 Mt CO2-eq). (A) CH4 emissions pathway of each mitigation strategy with a linear increase in emissions reduction rate starting in 2023 until complete adoption in 2050. The pathway of composting organics overlaps with that of halving waste due to a nearly similar reduction in CH4 emissions; refer to data S1 for exact values. (B) Comparison of the resulting CH4 emissions in 2030 among scenarios of BAU, completely adopting the mitigation strategies by 2050, and accelerating the year of complete adoption to be on track with the Global Methane Pledge. Retrofitting landfills is excluded in (B) because it cannot achieve the Global Methane Pledge even at complete adoption (i.e., it can only reduce 18% of CH4 emissions relative to 2020).
waste is more apparent than diverting organic waste for biological treatment or retrofitting landfills. UMICs, LMICs, and LICs generally prioritize low-cost landfills, so the effectiveness of diverting organic waste for biological treatment is more apparent. Although treating organic waste by anaerobic digestion nets lower cumulative GHG emissions (10 Gt CO2-we) than halving waste generation (12 Gt CO2-we) by the end of 2050, halving waste generation at its source has to be prioritized considering that (i) it may be financially and technologically infeasible to build mass anaerobic digestion plants in LMICs and LICs to treat their organic waste (26), (ii) it is the most effective mitigation strategy for HICs, (iii) source reduction has the highest priority in the waste management hierarchy that considers the perspective of life cycle thinking (27, 28), (iv) additional emissions associated with handling and transporting the MSW may incur before treating it in anaerobic digestion plants (29), and (v) it aligns with the SDG 12, target 12.3 under the United Nations Sustainable Development Goal to halve per capita global food waste at the retail and consumer levels by 2030. To illustrate how GHG emissions from the global MSW system could be maintained below the emissions limit, we considered the adoption of one or more mitigation strategies in the linearly decarbonizing future (Fig. 2). Considering the financial and technological feasibility of implementing the mitigation strategies (30), we focused on diverting the organic SCIENCE science.org
waste to composting instead of anaerobic digestion and present a comparison between them in the supplementary materials (fig. S5). We found that focusing solely on retrofitting landfills may jeopardize the achievement of the Paris Agreement goals because it only marginally stays within the 2°C limit until 2050. The remaining mitigation pathways show that meeting the 2°C target is plausible because of the immediate climate benefit induced by CH4 reduction, which is equivalent to atmospheric CO2 removal (31). Staying within the 1.5°C limit requires adopting more than one strategy because individual strategies alone are insufficient. Aiming for an absolute minimal MSW generation may also be challenging to achieve because it might require drastic changes in consumer behavior and lifestyle (13). Integrated MSW frameworks, which encompass multiple strategies to tackle the heterogeneity of MSW composition, engage a wider range of stakeholders, resulting in a more adaptable approach to reducing GHG emissions (32). Our findings show that adopting all three strategies, retrofitting landfills, composting organics, and halving waste, could prevent overshooting the temperature limits while resulting in a global MSW system with net-zero warming relative to 2020. The pathway permits one-off CO2 emissions of 12 ± 5 Gt (50% likelihood) and 11 ± 5 Gt (67% likelihood) under the 1.5°C target, and 27 ± 5 Gt (50% likelihood) and 23 ± 5 Gt (67% likelihood) under the 2°C target. This will contribute to a near-term cool-
ing effect relative to 2020 by 2050, with continuous efforts to reduce GHG emissions. Our scenario analyses exemplify the promising potential of the global MSW system in cutting its GHG emissions within the IPCC cumulative emissions budget using a reporting framework to properly account for CH4’s impact on global temperatures. The urgency of adopting these mitigation strategies can be seen by illustrating their reduction in CH4 emissions with respect to the Global Methane Pledge (Fig. 3A). Upon complete adoption of mitigation strategies by 2050, CH4 emissions can be reduced by up to 80% compared with the BAU. However, the designed mitigation strategies cannot reduce 30% of the CH4 emissions in a timely manner from the global MSW system by 2030 relative to 2020 [422 million tonnes (Mt) Mt CO2-eq] because the mitigation strategies are only to be fully adopted by 2050, which would only be able to achieve the reduction target by then (except for retrofitting landfills). We found that it would be necessary to accelerate the complete adoption of mitigation strategies by at least 9 to 17 years (by 2033 to 2041) to be on track with the progress to achieve the Global Methane Pledge under a linearly decarbonizing future (Fig. 3B). Implications for controlling global warming
Because we are on course for ~2.6°C of warming by 2100 based on the current policy implementation (33), fulfilling the Global Methane Pledge is essential. Previous analysis showed that curbing 57% of global CH4 emissions could result in decreases of up to 0.25 °C by 2050 and 0.5°C by 2100 (9), slowing the amplifying climate feedback such as the ice-albedo feedback and the permafrost thaw, which would otherwise further reduce the remaining emissions budget by 8 to 25% (34). Our results show that rapid and far-reaching changes in the global MSW system within the next two decades are necessary to achieve the Paris Agreement goals and the Global Methane Pledge. These changes require policies synchronizing with the GHG emissions mitigation strategies expected within the framework of the Paris Agreement. Therefore, waste stakeholders must coordinate with governments and educational institutions to articulate a vision for revamping the precarious state of MSW disposal. An efficient waste management system necessitates the ongoing utilization of three types of policy tools in a cohesive blend: (i) direct regulation encompassing laws enforced rigorously, (ii) economic instruments offering incentives and disincentives for particular MSW management practices, and (iii) social tools grounded in communication and engagement with stakeholders (35). Although mitigation approaches vary by country, the key levers identified for minimizing 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
799
RES EARCH | R E S E A R C H A R T I C L E S
GHG emissions are reducing the MSW volume and managing organic waste sustainably. These levers would be plausible by encouraging home composting and starting educational and awareness campaigns to inculcate environmental values in communities to generate less waste at the source (36). Overcoming behavioral barriers in altering the status quo in consumption and disposal patterns requires leveraging the power of both rational motives (e.g., monetary incentives) and nonrational motives (e.g., moral obligations such as ethical consumerism and environmental stewardship) (7). Because disposal sites are still the backbone of MSW management in LMICs and LICs, disposal of MSW in open dumps has to transition into managed landfills (37). Isolation of organic waste in LMICs and LICs can be done through decentralized composting and small-scale biodigesters since they have the potential to achieve environmental benefits similar to those of centralized biological treatment plants at a lower cost (38). It is imperative to use facilitating tools to empower social control of public policies with a focus on strategic directives, organizational structures, legal considerations, and funding mechanisms (32). Implementing extended producer responsibility policies holds manufacturers responsible for the life cycle of their products, encouraging them to design products with minimal waste and better end-of-life management (8). Landfill operators can retrofit their facilities for biogas energy recovery but they first require government incentives (e.g., the renewable fuel standard and the low carbon fuel standard in the United States) to increase the price of biogas (6). Policy considerations regarding pricing emissions and allowing emissions credits for reducing landfill gas emissions are also important for the economic viability of waste-to-energy facilities (39). The specific design and implementation of incentives should be tailored to the socioeconomic context of each country to ensure equitable access to MSW-related mitigation practices. Charging systems prevailing in HICs (e.g., flat rates and the “pay-as-you-throw” system) gather funds to finance advanced treatment services and offer an incentive for MSW reduction (40). The World Bank’s development policy loan effectively reforms the MSW management system of a LMIC (i.e., Morocco), increasing the collection of MSW for disposal in managed landfills from 10% (2008) to 32% (2011) while closing or rehabilitating 21 open dump sites (41). The resulting benefits of reducing GHG emissions must not be undermined by socioeconomic disparities between and within regions. Countries leading in MSW management sharing their expertise with countries with less expertise will substantially affect global emissions. We have demonstrated opportunities to maintain the global MSW system within the Paris 800
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
Agreement goals and achieve net-zero warming by 2050 relative to 2020. The short-term cooling of CH4 is not a low-hanging fruit for combating climate change because it does not exactly “buy time” for delayed CO2 mitigation. Concentrating CH4 has a huge initial effect (i.e., a considerable reversal of warming); however, the longer we fail to decarbonize, the higher the temperature we are locked into. The climatically optimal strategy is to capitalize on the rapid cooling from ceasing CH4 emissions and stop the ongoing temperature increases that would occur from sustained CO2 and N2O emissions (31). RE FERENCES AND NOTES
1. Intergovernmental Panel on Climate Change, “Summary for policymakers,” in Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, V. Masson-Delmotte et al., Eds. (Cambridge Univ. Press, 2021). 2. E. Masood, J. Tollefson, Nature (2021). 3. W. Cornwall, “‘It is the methane moment.’ How a once ignored greenhouse gas moved to center stage” (ScienceInsider, 2021). 4. International Energy Agency, “Global methane tracker 2022,” (IEA, 2022); https://www.iea.org/reports/global-methanetracker-2022. 5. X. Lan, K. W. Thoning, E. J. Dlugokencky, “Trends in globallyaveraged CH4, N2O, and SF6 determined from NOAA Global Monitoring Laboratory measurements, version 2023-10” (Global Monitoring Laboratory, 2022); https://doi.org/10.15138/ P8XG-AA10. 6. S. DeFabrizio et al., “Curbing methane emissions: How five industries can counter a major climate threat” (McKinsey, 2021); https://www.mckinsey.com/capabilities/sustainability/ our-insights/curbing-methane-emissions-how-five-industriescan-counter-a-major-climate-threat. 7. K. Parajuly, C. Fitzpatrick, O. Muldoon, R. Kuehr, Resour. Conserv. Recycl.: X 6, 100035 (2020). 8. S. Kaza, L. Yao, P. Bhada-Tata, F. Van Woerden, “What a waste 2.0: A global snapshot of solid waste management to 2050” (World Bank, 2018); http://hdl.handle.net/ 10986/30317. 9. Control methane to slow global warming - fast. Nature 596, 461 (2021). 10. Energy Transitions Commission, “Mission Possible: Reaching netzero carbon emissions from harder-to-abate sectors” (ETC, 2018); https://www.energy-transitions.org/publications/missionpossible/. 11. J. T. Powell, M. R. Chertow, D. C. Esty, Waste Manag. 80, 137–143 (2018). 12. S. B. Borrelle et al., Science 369, 1515–1518 (2020). 13. A. Gómez-Sanabria, G. Kiesewetter, Z. Klimont, W. Schoepp, H. Haberl, Nat. Commun. 13, 106 (2022). 14. D. Shindell et al., Science 356, 493–494 (2017). 15. R. Pipatti, S. M. M. Vieira, “2006 IPCC guidelines for national greenhouse gas inventories” (IPCC, 2006); https://www.ipcc.ch/ report/2006-ipcc-guidelines-for-national-greenhouse-gasinventories/. 16. E. Calvo Buendia et al., “2019 refinement to the 2006 IPCC guidelines for national greenhouse gas inventories” (IPCC, 2019); https://www.ipcc-nggip.iges.or.jp/public/2019rf/ index.html. 17. R. Dellink, J. Chateau, E. Lanzi, B. Magné, Glob. Environ. Change 42, 200–214 (2017). 18. S. Kc, W. Lutz, Glob. Environ. Change 42, 181–192 (2017). 19. S. Kaza, S. Shrikanth, S. Chaudhary, “More growth, less garbage” (World Bank, 2021); https://openknowledge. worldbank.org/entities/publication/ba7feea4-0abe-59fbbc60-ce6b60eb1ceb. 20. Z. X. Hoy, K. S. Woon, W. C. Chin, H. Hashim, Y. V. Fan, Comput. Chem. Eng. 166, 107946 (2022). 21. J. G. Canadell et al., “Global carbon and other biogeochemical cycles and feedbacks,” in Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate
22. 23. 24. 25. 26. 27.
28. 29. 30. 31. 32. 33. 34. 35.
36. 37. 38. 39. 40. 41.
42.
Change, V. Masson-Delmotte et al., Eds. (Cambridge Univ. Press, 2021). M. Cain et al., NPJ Clim. Atmos. Sci. 2, 29 (2019). M. R. Allen et al., NPJ Clim. Atmos. Sci. 1, 16 (2018). Z. Hausfather, G. P. Peters, Nature 577, 618–620 (2020). R. Prateep Na Talang, S. Sirivithayapakorn, J. Clean. Prod. 312, 127761 (2021). L. Breitenmoser et al., J. Environ. Manage. 236, 396–412 (2019). European Parliament, “Directive 2008/98/EC of the European Parliament and of the council of 19 November 2008 on waste and repealing certain directives” (European Parliament, 2018); http://data.europa.eu/eli/dir/2008/98/ 2018-07-05. D. Lazarevic, N. Buclet, N. Brandt, J. Clean. Prod. 29-30, 199–207 (2012). S. Chen et al., Sci. Total Environ. 717, 137193 (2020). S. Sondh, D. S. Upadhyay, S. Patel, R. N. Patel, J. Clean. Prod. 356, 131908 (2022). J. Lynch, M. Cain, R. Pierrehumbert, M. Allen, Environ. Res. Lett. 15, 044023 (2020). M. Batista et al., J. Clean. Prod. 312, 127516 (2021). Nat. Geosci. 14, 875–875 (2021). T. Gasser et al., Nat. Geosci. 11, 830–835 (2018). D. C. Wilson et al., “Global waste management outlook” (UN Environment Programme, 2015) https://www.unep.org/ resources/report/global-waste-management-outlook. G. B. Zamri et al., J. Clean. Prod. 246, 118969 (2020). Z. Lin, J. K. Ooi, K. S. Woon, Sci. Total Environ. 816, 151541 (2021). L. Nanlin, L. Fan, Z. Hua, S. Liming, H. Pinjing, Sci. Total Environ. 884, 163724 (2023). A. K. Townsend, M. E. Webber, Waste Manag. 32, 1366–1377 (2012). B. R. Alzamora, R. T. V. Barros, Waste Manag. 115, 47–55 (2020). World Bank, “Morocco: Improving municipal solid waste management through development policy operations” (World Bank, 2013); https://www.worldbank.org/en/results/2013/ 05/22/morocco-improving-municipal-solid-wastemanagement-through-development-policy-operations. Z. X. Hoy, K. S. Woon, W. C. Chin, Y. V. Fan, S. J. Yoo, Data for: Curbing global solid waste emissions toward net-zero warming futures, version 1, Dryad (2023); https://doi.org/10. 5061/dryad.bvq83bkfv.
AC KNOWLED GME NTS
We thank Z. X. Phuang, M. Y. Chin, J. K. Ooi, and W. L. Ng for providing comments on improving the manuscript and Editage (www. editage.com) for English language editing. Funding: This work was supported by the Ministry of Higher Education Malaysia (Fundamental Research Grant Scheme FRGS/1/2020/TK0/XMU/02/2 to K.S.W.); Xiamen University Malaysia (Xiamen University Malaysia Research Fund XMUMRF/2019-C4/IENG/0022 to K.S.W.); and the Korea Environment Industry & Technology Institute (KEITI) of the Korea Ministry of Environment (MOE) (grant 2022003560007 to S.J.Y.). Author contributions: Conceptualization: Z.X.H., K.S.W.; Formal analysis: Z.X.H., K.S.W.; Funding acquisition: K.S.W.; Investigation: Z.X.H., K.S.W.; Methodology: Z.X.H., K.S.W., W.C.C., Y.V.F., S.J.Y.; Project administration: K.S.W.; Supervision: K.S.W., W.C.C.; Validation: Z.X.H., K.S.W., W.C.C., Y.V.F., S.J.Y.; Visualization: Z.X.H.; Writing – original draft: Z.X.H.; Writing – review and editing: K.S.W., W.C.C., Y.V.F., S.J.Y. Competing interests: The authors declare no competing interests. Data and materials availability: All data are available in the main text or the supplementary materials and have been archived in Dryad (42). License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/about/science-licensesjournal-article-reuse SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adg3177 Materials and Methods Supplementary Text Figs. S1 to S7 Tables S1 to S10 References (43Ð79) Data S1 to S3 Submitted 2 January 2023; accepted 22 September 2023 10.1126/science.adg3177
science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
ELECTROCALORICS
High cooling performance in a double-loop electrocaloric heat pump Junning Li1, Alvar Torelló1 , Veronika Kovacova1, Uros Prah1, Ashwath Aravindhan1,2, Torsten Granzow1, Tomoyasu Usui3, Sakyo Hirose3, Emmanuel Defay1* Cooling through solid-state electrocaloric materials is an attractive replacement for vapor compression. Despite recent efforts, devices that are potentially commercially competitive have not been developed. We present an electrocaloric cooler with a maximum temperature span of 20.9 kelvin and a maximum cooling power of 4.2 watts under the moderate applied electric field of 10 volts per micrometer without any observed breakdown. Moreover, the maximum coefficient of performance, even taking into account energy expended on fluid pumping, reaches 64% of CarnotÕs efficiency as long as energy is properly recovered. We believe that this demonstration shows electrocaloric cooling to be a very promising alternative to vapor compression cooling.
F
ive billion refrigeration, air-conditioning, and heat pump systems are in operation, which are estimated to consume ~20% of global electricity (1). Most of these systems work through vapor-compression technology with hydrofluorocarbons inside and are responsible for producing up to 10% of the overall greenhouse gas emissions in the world (2). The latter includes indirect carbon dioxide emissions coming from the electricity production, as well as the direct leakage of fluorinated refrigerants. To solve this issue, solid-state caloric materials have been pinpointed as one promising alternative because they can be highly efficient and have no direct global warming potential (3–9). Caloric materials undergo reversible adiabatic temperature (or isothermal entropy) changes upon modifications in external applied fields, which can be magnetic (magnetocalorics), hydrostatic pressure (barocalorics), uniaxial stress (elastocalorics), and electric [electrocalorics (EC)] (10). Another advantage of these materials is that most of the work invested into driving the changes is stored and can be recovered. Thus, the coefficient of performance (COP) of the resulting devices can be notably improved. In the case of EC devices, this is particularly easy and low cost through the manipulation of electrical charges and has already been experimentally proven by using appropriate electrical circuitry (11). In addition, the direct use of electricity makes EC coolers compact in volume and suitable for miniaturization. This process is more challenging for magnetocaloric devices, which require more space to accommodate the voluminous permanent magnets. In 1
Materials Research and Technology Department, Luxembourg Institute of Science and Technology, Belvaux L-4422, Luxembourg. 2University of Luxembourg, Esch-sur-Alzette L-4365, Luxembourg. 3Murata Manufacturing Co., Nagaokakyo, Kyoto 617–8555, Japan. *Corresponding author. Email: [email protected] †Present address: Group of Characterization of Materials, Department of Physics, Universitat Politècnica de Catalunya, Barcelona 08019, Spain.
SCIENCE science.org
the case of elastocaloric devices, one option is large hydraulic compressors, although smaller electric motors that provide mechanical loading can also be used. Recently, adiabatic EC temperature changes of more than 5 K have been reported in poly(vinylidene fluoride)–based polymers (12, 13) and in PbSc0.5Ta0.5O3 (PST) multilayer capacitors (MLCs) (14). In 2020, an electrocaloric cooler based on the latter crossed for the first time the 10 K barrier, producing a temperature span of 13 K (15). And the largest measured cooling power was 0.26 W, which was obtained in another device (15). Despite these notable advances, performances of EC devices still do not meet the demands of current cooling applications, and their industrial development is on hold. These performances also lag behind most of their magnetocaloric and elastocaloric counterparts, for which temperature spans of more than 20 K and cooling powers of more than 50 W have been repeatedly reported (16–18), even up to 2000 W in a specific magnetocaloric regenerator with 1.52 kg of active material (19). We aimed to develop an EC-based device with exceptional cooling performance (20, 21). By devising a different configuration of the PSTMLC parallel plate matrix used by Torello et al. and meticulously optimizing its design (15),
the prototype developed in this work succeeded in producing consistent temperature spans of more than 20 K. Additionally, we implemented a double-loop–based cooling power measurement, reporting up to 4.2 W. We also measured the energy needed to pump the fluid inside the regenerator. Last, we computed the COP and second-law efficiency from direct measurements of all different power contributions, including pumping. Under the assumption that all the recoverable energy is actually recovered (11, 22, 23), we found a maximum exergy efficiency (COP/ COPCarnot) of 64%. Temperature span and cooling power are respectively 50% and 15 times larger than those of the previous best electrocaloric device (15). These results position this device among the best caloric coolers, with the added value of being very compact overall compared with magneto- and elastocaloric devices, if one includes power supply devices. Having now an electrocaloric device exhibiting a cooling power in the watt range with an active regenerator of only 10 cm3 made of scalable cooling modules— MLCs—helps point the way toward a potential alternative for the cooling and air conditioning industry. Optimized regeneration
Our fluid-based EC regenerator was designed on the basis of magnetocaloric regenerative cycles, which were first introduced by Brown in 1976 (24). The regenerator structure consists of 0.5-mm-thick PST-MLCs stacked on top of each other, with spacers positioned on both sides and electrically connected in parallel. A heat transfer fluid can then flow through and displace heat accordingly (Fig. 1A). The fluid movement is synchronized with the electrocaloric heating and cooling steps (electric field on and off). By continuously repeating these steps (Fig. 1A), the temperature difference between the hot and cold sides of the regenerator (DTspan) can become larger than the adiabatic EC temperature change of the material (DTEC), with the corresponding ratio, DTspan/DTEC, defined as the regeneration factor. Further details on heat regeneration can be consulted elsewhere (25).
Table 1. Summary of the specific device cooling power density of caloric devices reported in the literature. MC, magnetocaloric; eC, elastocaloric; and EC, electrocaloric.
Caloric prototype
Cooling ChamberÕs power (W) volume (liters)
Cooling power Specific cooling density power Reference (W liter−1) (W kg−1)
2502 3.3 747 1646 (19) MC/ La(Fe,Si)13H ..................................................................................................................................................................................................................... eC/NiTi 60 0.2 300 4400 (18) ..................................................................................................................................................................................................................... EC/P(VDF-TrFE-CFE) 0.3 0.013 23 2800 (34) ..................................................................................................................................................................................................................... EC/PMN-PT 0.23 0.003 78 16 (38) ..................................................................................................................................................................................................................... EC/PST 0.26 0.004 66 22 (15) ..................................................................................................................................................................................................................... EC/PST 4.2 0.010 420 161 This work .....................................................................................................................................................................................................................
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
801
RES EARCH | R E S E A R C H A R T I C L E S
Our regenerator design relies on the wellestablished lead scandium tantalate MLCs (14). These structures possess metal terminals, which are good thermal conductors placed on opposite sides for electrical connections (26). This structure can be advantageous to transfer heat to the fluid (27). In our device, terminals can be either placed outside (Fig. 1B, REG-TO) or inside (Fig. 1B, REG-TI) the fluid channels. We prepared one regenerator of each type with 32
erators were tested under the same electric field of 7.5 V mm−1 (300 V). The operating frequency was 0.33 Hz. The displaced fluid volume ratios (v*) in REG-TO and REG-TI were 0.9 and 1, respectively. Those experiments clearly show the benefit of using the REG-TI configuration, which stems from reducing the thermal conductivity produced by the alignment of the terminals in the REG-TO configuration. We therefore focused on the REG-TI geometry.
0.5-mm-thick PST-MLCs, distributed in matrices of four columns by eight rows (28). The slit thickness between each MLC in a column was 0.25 mm in both cases. We ran REG-TO and REG-TI under various parameters (tables S1 to S4). After setting the starting temperature at room temperature and after 180 s of operation, we measured a maximum DTspan of 3.3 and 5.1 K in REG-TO and REG-TI, respectively (Fig. 1C). The regen-
B
A
C
REG-TO
Regenerator (REG)
PST MLC
Active area
26
5 mm
1. Field ON
@ 300 V
27
REG-TO Hot end Cold end
2. Field ON with fluid pumping
Copper wires
Void Terminals of channels PST MLCs
Silver epoxy
T (°C)
25
Spacer
24
REG-TI Hot end Cold end
23
3. Field OFF
22
21
REG-TI
18 16 14 12 10 8 6 4
Number of Rows
14 13 12 11
B
@ 400 V
40
30
10 5 0
20
200
BTO-based PVDF-based BST@PVDF-based PMN-PT PST
This work
15
15
0
2000
4000
10
5
Cold end temperature 0
8 4
6
8
10
12
14
16
0
1000
Number of Columns Fig. 2. Optimization and characterization of the EC prototype. (A) Temperature differences between hot and cold ends (DTspan) in REG-TI as a function of the number of columns and rows. The electric field is 7.5 V mm−1 (300 V). The cycling period for each REG-TI was optimized to maximize DTspan. (B) Temperature time evolution of hot and cold sides for a REG-TI with 10 columns and 14 rows at the electric field of 10 V mm−1 (400 V) and cycling period of 4 s. 802
C
20
35
Time (s) 9
150
20
25
10
100
Time (s)
Hot end temperature
Tspan (K)
Tspan (K)
@ 300 V
15
T (°C)
16
50
time, the syringe pump pushes the fluid to the other side of the regenerator. (iv) The regenerator absorbs the heat from the fluid moving to the cold end at zero field. A regenerative cycle can be achieved by repeating the above steps. (B) Two EC regenerators (four columns by eight rows) with different geometries in terms of the position of MLCs’ terminals. REG-TO refers to the regenerator with terminals outside. REG-TI refers to the regenerator with terminals inside. (C) Temperature changes as a function of time for hot and
Fig. ALERT 1. Design STREAM of theCUT EC regenerator. !!! (A) Schematics of an electrocaloric cycling by using an EC regenerator with a stack of PST-MLCs. (i) An electric field is applied rapidly, resulting in a temperature increase of the EC regenerator. Simultaneously, a syringe pump pushes the dielectric fluid through the regenerator. (ii) Heat is transferred from the hot regenerator to the fluid moving to the hot end while the electric field remains constant. (iii) The electric field is removed, leading to a temperature drop of the EC regenerator. At the same
A
0
Tspan (K)
4. Field OFF with fluid pumping
Kapton film
17 NOVEMBER 2023 ¥ VOL 382 ISSUE 6672
2000
Time (s)
3000
4000
0
1
2
3
4
5
TEC (K)
(Inset) The corresponding DTspan. (C) DTspan as a function of EC temperature change (DTEC) of different EC-based prototypes. BTO, BaTiO3 (29–32); PVDF, poly(vinylidene fluoride-trifluoroethylene) [P(VDF-TrFE)] (33) and poly(vinylidene fluoride-trifluoroethylene-chlorofluoroethylene) [P(VDF-TrFE-CFE)] (12, 34, 35); BST@PVDF, BaxSr1–xTiO3 nanoparticles-filled P(VDF-TrFE-CFE) (36, 37); PMN-PT, 0.9Pb(Mg1/3Nb2/3)O3–0.1PbTiO3 (38); PST, PbSc0.5Ta0.5O3 (15, 39, 40). science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
To find out the optimal regenerator size, we built several versions with different numbers of columns and rows, and we measured their temperature span at the moderate field of 7.5 V mm−1 (300 V) (Fig. 2A) [(28), section 3]. We found the optimum to be a regenerator of 10 columns by 14 rows, producing a maximum temperature span of 17.4 K (Fig. 2A). The presence of such an optimum is likely due to thermal losses that increase with the length
of regenerators, as we observed with finite element modeling [(28), section 4]. Then, we increased the electric field for this configuration to 10 V mm−1 (400 V), and we measured a temperature span of 20.0 K after 4000 s of operation (Fig. 2B). To achieve such a large temperature span at zero load, the operating frequency was 0.25 Hz, and the displaced fluid volume ratio was 0.4. The starting temperature of this experiment was set
A
to 29°C. Considering that we measured a DTEC of 1.7 K at 400 V in PST-MLCs, the resulting regeneration factor was 11.8. We also assessed the reproducibility of this experiment by dismantling and building again the regenerator four times from scratch [(28), section 5]. The maximum temperature span varied between 18.3 and 20.9 K. We compiled the temperature span of all electrocaloric devices reported (Fig. 2C) (29–40). The temperature span reached by our design is more than twice as large as that of any other electrocaloric regenerator, except one that reached 13 K by using similar PST-MLCs (15). However, this DTspan has been generated from a lower intrinsic DTEC (105). We performed ex situ accelerated lifetime tests on individual MLCs [(28), section 8], which indicate stable electrocaloric behavior up to at least 4 million voltage switching cycles. This results in an estimated lifetime of 31.7 years under dc field application. 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
803
RES EARCH | R E S E A R C H A R T I C L E S
804
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
A 20
Maximum temperature span (K)
To have a fair assessment of the COP, we measured the energy needed to pump the fluid inside the regenerator by developing a specific setup that enabled the observation of the fluid level while pumping in both directions [(28), section 9]. The pressure reached 9.9 mbar. The corresponding flow rate was 80 ml min−1 (1.33 × 10−6 m3 s−1), and the period of the whole cycle was 4 s. Therefore, the power due to the pressure drop for each cycle was 1.3 mW, which is extremely low compared with the generated cooling power. Last, we calculated the COP and the COP/ COPCarnot of our device (i) with no charge recovery and (ii) by considering that all the electric energy left in the MLCs when we discharge them can be recovered (28). To be more specific, we measured how much electric energy is required to charge a 10-by-14 regenerator (1.64 J) and how much energy can be recovered (1.39 J) [(28), section 10]. We also systematically included the pumping power in all COP values. We found a linear increase of the COP with cooling power, reaching a maximum of 10 (66 when recovering the energy) (Fig. 3D). At 2.6 W cooling power and 4.7 K temperature span, COP reached 6 (40 with energy recovery). This represents an exergy efficiency (COP/COPCarnot) of 10 and 64%, respectively, without and with energy recovery. To be more realistic, we worked out numbers of efficiency by considering two representative experimental energy recovery studies in the context of electrocaloric coolers from the literature, namely 75 and 99% of recovery efficiency reported in (11) and (22), respectively. With a recovery rate of 75%, the maximum exergy efficiency was 26% with a cooling power of 2.6 W, and 12% at 4.2 W. With a recovery rate of 99%, the maximum exergy efficiency was 60% with a cooling power of 2.6 W, and 28% at 4.2 W. An energy recovery rate of 75% does not enable increasing exergy efficiency beyond 25%. Therefore, these results show that energy recovery is key to improve the efficiency of electrocaloric systems. Very recently, Mönch et al. showed that an energy recovery as high as 99.7% is possible, which is very promising (41). Because such a configuration can also work as a heat pump, we calculated heating power and efficiency when the device runs in heating mode [(28), section 7]. The maximum heating power was 3.1 W, with a temperature span of 4 K; COP reached 8 (50 with energy recovery). At 2.2 W heating power and 5 K temperature span, COP reached 5 (35 with energy recovery). It represents an exergy efficiency (COP/COPCarnot) of 9 and 58%, respectively, without and with energy recovery. To test reproducibility, we ran complementary experiments for cooling and heating power [(28), section 7], which showed similar results and trend.
15
PbSc0.5Ta0.5O3 ceramics [Refs. 39, 42, 43] BaTiO3-based MLCs [Refs. 11, 29-32, 44, 45] P(VDF-TrFE)-based films [Refs, 12, 33-35, 46] Pb(Mg1/3Nb2/3)O3-PbTiO3 ceramics/MLCs [Refs. 38, 47-49]
This work
PbSc0.5Ta0.5O3 MLCs [Refs. 15, 40] Ba0.6Sr0.4TiO3/BN@P(VDF-TrFE-CFE) composite [Refs. 36, 37, 50, 51]
10
5
0 1985
B 4
Cooling power (W)
High efficiency
3
1990
1995
2010
2015
2020
PbSc0.5Ta0.5O3 ceramics [Ref. 42] BaTiO3-based MLCs [Refs. 11, 30, 32, ] P(VDF-TrFE-CFE)-based films [Refs. 12, 34, 35] Pb(Mg1/3Nb2/3)O3-PbTiO3 ceramics/MLCs [Ref. 38]
2025
This work
PbSc0.5Ta0.5O3 MLCs [Refs. 15, 40] Ba0.6Sr0.4TiO3@P(VDF-TrFE-CFE) composite [Refs. 36, 37]
2
1
0 1985
1990
1995
2010 Year
2015
2020
2025
Fig. 4. Summary of the performance of the electrocaloric cooling devices reported in the literature. (A) Maximum temperature span without thermal load reported over the years. (B) Measured maximum cooling power in each year. Data were taken from (11, 12, 15, 29–40, 42–51).
Additionally, we estimated the capability of this regenerator to extract heat from the EC modules. The maximum heat that can be generated in a cycle is essentially the mass of the active EC material times its specific heat times the EC adiabatic temperature change (mCpDT). We can compare this maximum heat with the experimental cooling power times the cycling period. We called this parameter regenerator efficiency hreg. (28). Hence, hreg is 33 and 25% for cooling and heating, respectively. This means that although the results are extremely encouraging, there is still room for improvement regarding the extraction of heat from the electrocaloric modules. Comparing electrocaloric prototypes
We carried out a thorough characterization of the two most representative metrics in our prototype—namely, the maximum temperature span without thermal loads and the measured maximum cooling power. This allowed us to compare the performance of our prototype with previous electrocaloric prototypes from the literature represented over the years (Fig. 4). Hence, the temperature span measured (20.9 K)
is the highest among all electrocaloric systems, the second best being at 13 K (Fig. 4A) (15). The measured cooling power is more than one order of magnitude larger than any of ceramic and MLC-based systems and five times larger than P(VDF-TrFE-CFE)–based coolers (Fig. 4B). In Table 1, we summarize the cooling performances of representative magnetocaloric, elastocaloric, and electrocaloric devices in terms of specific cooling power in W kg−1 and of specific device cooling power density in W liter−1. The volume used in this figure of merit refers to the volume of the chamber that contains the caloric material, which is in practice what matters when it comes to device final dimensions [(28), section 11]. Magnetocaloric and elastocaloric coolers exhibit absolute cooling powers up to nearly two orders of magnitude larger than our device because the amount of active material is much larger [for example, 227 cm3 of La(Fe,Si)13H (19) compared with 5.2 cm3 of PST]. However, the cooling power density of our electrocaloric regenerator reaches 420 W liter−1, which is comparable with that of its magnetocaloric (747 W liter−1) and elastocaloric (300 W liter−1) counterparts. science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
Our regenerator needs dielectric fluid rather than water to prevent short circuits. It exhibits poor thermal properties compared with those of water, which is used in magnetoand elastocaloric regenerators. The direct consequence is that heat exchange takes time. Hence, the heat flow frequency is 0.25 Hz, whereas it reaches 1.7 Hz in magnetocalorics (17) and 2 Hz in elastocalorics (18). This is why the specific cooling power of our electrocaloric device is one order of magnitude lower than that of magnetocalorics and elastocalorics (Table 1). Consequently, we can anticipate that heat exchange could be one order of magnitude larger if water were used. Therefore, further improvement is possible by developing waterproof MLCs and by scaling up their number. As pointed out previously (15), additional modifications allow for improvement. These include the use of water and thinner and flatter electrocaloric modules, and if safe, the application of higher electric fields is possible. These obvious improvements would immediately boost performance and are all reasonably attainable goals. RE FE RENCES AND N OT ES
1. J. L. Dupont, P. Domanski, P. Lebrun, F. Ziegler, “The role of refrigeration in the global economy-38. Informatory note on refrigeration technologies,” INIS-FR–20-0278 (International Institute of Refrigeration, 2019). 2. Y. Dong, M. Coleman, S. A. Miller, Annu. Rev. Environ. Resour. 46, 59–83 (2021). 3. A. Kitanovski, Adv. Energy Mater. 10, 1903741 (2020). 4. L. Mañosa, A. Planes, Appl. Phys. Lett. 116, 050501 (2020). 5. P. Lloveras, J. L. Tamarit, MRS Energy Sustain 8, 3–15 (2021). 6. A. S. Mischenko, Q. Zhang, J. F. Scott, R. W. Whatmore, N. D. Mathur, Science 311, 1270–1271 (2006). 7. B. Neese et al., Science 321, 821–823 (2008). 8. J. Shi et al., Joule 3, 1200–1225 (2019). 9. H. Hou, S. Qian, I. Takeuchi, Nat. Rev. Mater. 7, 633–652 (2022). 10. X. Moya, N. D. Mathur, Science 370, 797–803 (2020). 11. E. Defay et al., Nat. Commun. 9, 1827 (2018). 12. Y. Meng et al., Nat. Energy 5, 996–1002 (2020). 13. X. Qian et al., Nature 600, 664–669 (2021). 14. B. Nair et al., Nature 575, 468–472 (2019). 15. A. Torelló et al., Science 370, 125–129 (2020). 16. A. Greco, C. Aprea, A. Maiorino, C. Masselli, Int. J. Refrig. 106, 66–88 (2019). 17. M. Masche, J. Liang, K. Engelbrecht, C. R. H. Bahl, Appl. Therm. Eng. 204, 117947 (2022). 18. Ž. Ahčin et al., Joule 6, 2338–2357 (2022). 19. S. Jacobs et al., Int. J. Refrig. 37, 84–91 (2014). 20. Y. Meng, J. Pu, Q. Pei, Joule 5, 780–793 (2021). 21. A. Torelló, E. Defay, Adv. Electron. Mater. 8, 2101031 (2022). 22. S. Moench et al., IEEE Access 10, 46571–46588 (2022). 23. D. E. Schwartz, Int. J. Refrig. 131, 970–979 (2021). 24. G. V. Brown, J. Appl. Phys. 47, 3673–3680 (1976). 25. A. Torelló, E. Defay, “Basics of design and modeling of regenerative electrocaloric coolers,” in The Electrocaloric Effect (Woodhead, 2023), pp. 333–357. 26. S. Kar-Narayan, N. D. Mathur, J. Phys. D Appl. Phys. 43, 032002 (2010). 27. Y. Liu, B. Dkhil, E. Defay, ACS Energy Lett. 1, 521–528 (2016). 28. Materials and methods are available as supplementary materials. 29. T. Zhang, X. S. Qian, H. Gu, Y. Hou, Q. M. Zhang, Appl. Phys. Lett. 110, 243503 (2017). 30. Y. Jia, Y. Sungtaek Ju, Appl. Phys. Lett. 100, 242901 (2012). 31. S. Bellafkih, A. Hadj Sahraoui, P. Kulinski, P. Dumoulin, S. Longuemart, J. Therm. Sci. Eng. Appl. 14, 061011 (2022).
SCIENCE science.org
32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42. 43. 44.
45. 46. 47. 48. 49. 50. 51.
Y. D. Wang et al., Appl. Phys. Lett. 107, 134103 (2015). H. Gu et al., Appl. Phys. Lett. 102, 122904 (2013). R. Ma et al., Science 357, 1130–1134 (2017). Y. Bo et al., Adv. Energy Mater. 11, 2003771 (2021). H. Cui et al., Joule 6, 258–268 (2022). P. Bai et al., Adv. Mater. 35, e2209181 (2023). U. Plaznik et al., Int. J. Refrig. 98, 139–149 (2019). Y. V. Sinyavsky, V. M. Brodyansky, Ferroelectrics 131, 321–325 (1992). Y. Wang et al., Science 370, 129–133 (2020). S. Mönch et al., IEEE J. Emerg. Sel. Top. Power Electron. 11, 4491–4507 (2023). Y. V. Sinyavsky, N. D. Pashkov, Y. M. Gorovoy, G. E. Lugansky, L. Shebanov, Ferroelectrics 90, 213–217 (1989). Y. V. Sinyavskii, Chem. Petrol. Eng. 31, 295–306 (1995). W. Sato, A study of a new cooling device based on the electrocaloric effect. Proceedings of ASME 2013 International Mechanical Engineering Congress, San Diego, 15–21 November 2013, IMECE2013-62258. D. Sette et al., APL Mater. 4, 091101 (2016). M.-D. Li et al., Nat. Commun. 13, 5849 (2022). U. Plaznik et al., Appl. Phys. Lett. 106, 043903 (2015). P. Blumenthal, C. Molin, S. Gebhardt, A. Raatz, Ferroelectrics 497, 1–8 (2016). P. Blumenthal, A. Raatz, Energy Technol. 6, 1560–1566 (2018). X. Kang, S. Jia, J. Peng, H. Yu, X. Zhou, Compos., Part B Eng. 227, 109391 (2021). P. Bai et al., Next Mat. 1, 100001 (2023).
ACKN OWLED GMEN TS
We thank H. Kuramoto and K. Sasaki for their assistance in fabricating the MLCs. Funding: J.L., A.T., T.G., U.P., V.K., and E.D. acknowledge the Fonds National de la Recherche (FNR) of Luxembourg for supporting this work through the projects BRIDGES2021/MS/16282302/CEC0HA/Defay, THERMODIMAT C20/MS/14718071/Defay, and BRIDGES2020/MS/15410586/
CALPOL/Defay. Author contributions: E.D. suggested the experimental study with A.T. and J.L.; T.U. and S.H. prepared the PST-MLC samples. J.L. prepared the experimental setup and ran the experiments that led to Figs. 1 and 2, with A.T. and V.K.; A.T. designed the double-loop setup that led to Fig. 3A, with V.K.; A.T. and J.L. ran the experiments that lead to Fig. 3B. J.L. ran the experiments that led to figs. S2 to S8, with A.T.; J.L. measured the pressure change that led to fig. S9, with U.P.; A.T. collected the data in fig. S10. T.G. and A.A. ran the experiments related to fatigue. J.L. gathered all the data, with A.T.; E.D., J.L., and A.T. elaborated the equations in the supplementary materials, materials and methods. J.L. built tables and prepared the figures. S.H. and T.G. made valuable contributions to the productive discussions and provided insightful comments. E.D. wrote the manuscript, with A.T. and J.L. All authors contributed to the final version of the manuscript. E.D. obtained the funding and supervised the project. Competing interests: A.T. and E.D. are inventors on the patent WO2021123460A1. The other authors declare no competing interests. Data and materials availability: All data are available in the manuscript or in the supplementary materials. Correspondence and requests for materials should be addressed to E.D. License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www. science.org/about/science-licenses-journal-article-reuse SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adi5477 Materials and Methods Supplementary Text Figs. S1 to S11 Tables S1 to S9 References (52–54) Submitted 8 May 2023; accepted 22 September 2023 10.1126/science.adi5477
ANTHROPOLOGY
Cooperation across social borders in bonobos Liran Samuni1,2,3* and Martin Surbeck2,4* Cooperation beyond familial and group boundaries is core to the functioning of human societies, yet its evolution remains unclear. To address this, we examined grooming, coalition, and food-sharing patterns in bonobos (Pan paniscus), one of our closest living relatives whose rare out-group tolerance facilitates interaction opportunities between groups. We show that, as in humans, positive assortment supports bonobo cooperation across borders. Bonobo cooperative attitudes toward ingroup members informed their cooperative relationships with out-groups, in particular, forming connections with out-group individuals who also exhibited high cooperation tendencies. Our findings show that cooperation between unrelated individuals across groups without immediate payoff is not exclusive to humans and suggest that such cooperation can emerge in the absence of social norms or strong cultural dispositions.
M
ore than any other species, humans cooperate across vast contexts, numbers, and social scales. We live in complex multilevel societies that promote the formation of strong cooperative relationships not only with kin, allies, and friends but also with distant acquaintances and even strangers. The extent of human nonkin coop1
Cooperative Evolution Lab, German Primate Center, Göttingen, Germany. 2Department of Human Evolutionary Biology, Harvard University, Cambridge, MA, USA. 3School of Psychology and Neuroscience, University of St Andrews, St Andrews, UK. 4Department of Human Behavior, Ecology and Culture, Max Planck Institute for Evolutionary Anthropology, Leipzig, Germany.
*Corresponding author. Email: [email protected] (L.S.); [email protected] (M.S.)
eration is unmatched, with trade and sharing of commodities, knowledge, and skills (1–3) taking place not only within human residential units (hereafter “groups”) but also across these units. Kin relations and repeated interaction opportunities are important foundations of withingroup cooperation across taxa (4, 5). However, these appear insufficient in explaining human large-scale cooperation (6). In humans, theoretical and empirical models consistently identify population structures, individual attributes, and cultural processes as critical components that promote cooperation (2, 7–13). These models predict that cooperation, such as the sharing of food resources, can emerge if the population 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
805
RES EARCH | R E S E A R C H A R T I C L E S
structure permits clustering of similar individuals, for example, when cooperators co-reside or interact preferentially with one another (10, 11, 13). Tolerance and cooperation across groups are apparent in various animal taxa, including insects, birds, and mammals (14–17). However, the ability to cooperate with unrelated outgroup partners with no immediate return is considered to be an exclusively human feature (18, 19). Human cooperation across residential groups facilitates resource and information transfer and the accumulation of knowledge, both of which support our long life spans and prolonged development and allow our species to thrive across the globe (2, 20, 21). But how distinctive is this human capacity? And what are the evolutionary foundations of human broad cooperation? Observations of the between-group relations of bonobos (Pan paniscus), our closest living relatives together with chimpanzees, challenge the notion that cooperation with no immediate return between distantly related individuals across groups is exclusive to humans. Within certain bonobo populations, individuals from distinct groups engage in a diverse range of interactions that span from aggression to cooperation, including grooming, forming alliances, and sharing high-value food resources
Grooming
High within-group cooperator Medium within-group cooperator Low within-group cooperator
(22–31). Despite prolonged and tolerant interactions between different groups, bonobos still maintain a clear in-group–out-group distinction (32). However, our understanding of bonobo cooperation across groups largely relies on descriptive information, which lacks empirical examination of the frequency, social structures, and underlying mechanisms that are involved. In this study, we characterized bonobo cooperation across groups by observing within- and between-group interactions of 31 adults living in two social groups (Ekalakala: three adult males and eight adult females; and Kokoalongo: seven adult males and 13 adult females) in the Kokolopori Bonobo Reserve, Democratic Republic of the Congo (DRC). During a 2-year observation period, we documented 95 encounters between the two groups, which lasted for around 20% of the total observation time. The durations of bonobo between-group encounters varied substantially, from less than 1 hour to 14 consecutive days [12.5 ± 17.2 hours (mean ± SD)], which emphasizes that bonobo groups can associate in a nontransient manner, thereby facilitating opportunities for consistent cross-group exchange. Results
Bonobo within- and between-group interactions consisted of a variety of cooperative acts,
Coalition
Within-group Between-group
Food sharing
Female Male
Fig. 1. Cooperation assortment in bonobo social networks. Social networks indicate that high within-group cooperators tend to be located more centrally and form the same types of connections across groups. Within-group cooperative tendencies were operationalized for each cooperative interaction: (i) grooming network, by dividing the amount of time a bonobo groomed an in-group member (average individual grooming time: Ekalakala, 2268 ± 881 min; Kokoalongo, 704 ± 439 min) by the average grooming duration within their group; (ii) coalition network, by dividing the number of times a bonobo formed a coalition with an in-group member (average individual coalition: Ekalakala, 36 ± 27; Kokoalongo, 63 ± 53) by the average coalition times within their group; and (iii) food-sharing network, by dividing the number of times a bonobo donated food to an in-group member (average individual sharing: Ekalakala, 27 ± 18; Kokoalongo, 16 ± 15) by the average food-sharing times within their group. Node colors indicate high (66th 806
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
including 3744 grooming interactions, 592 coalitions, 2920 cases of noncoalitionary aggressions, and 650 cases of food transfers (see supplementary materials for definitions). Betweengroup interactions represented 10% (N = 383, between 115 dyads) of all cases of grooming, 15% (N = 87, between 43 dyads) of all coalitions, 14% (N = 402, between 126 dyads) of all noncoalitionary aggressions, and 6% (N = 41, between 28 dyads, including 16 donors and 15 recipients) of all food transfers observed. Although kin selection is a powerful driver of cooperation and between-group connections, interactions in this bonobo population are unlikely to be driven solely by genetic relatedness. Bonobos are a male philopatric species, and the only mother-offspring pairs in the two groups are four mothers and their sons who reside in the same group. Whereas female migration can produce close familial connections across groups, permanent female immigration between the two groups has not been observed since the establishment of the research site in 2016, despite numerous (N = 22) immigrant females arriving to or leaving from the two groups. Finally, analysis of 15-loci autosomal microsatellite genotypes revealed that only 6% of both within- and between-group dyads are second-degree relatives or higher (see supplementary materials).
Unrelated Kin
percentile; red), medium (between the 33rd and 66th percentile; blue), and low (below the 33rd percentile; white) within-group cooperators. Node shape indicates males (square) or females (circles), and node size indicates the number of both incoming and outgoing connections possessed by an individual, with larger nodes signifying a greater number of connections. Edges represent connections with values equal to or greater than the population mean, with their colors indicating whether partners reside in the same (yellow) or different (gray) groups or whether partners are related (dashed line) or not (solid line). Only individuals that appeared in the data during both observation years (N = 27 out of the 31 individuals in the data) are depicted in the social network illustrations. The individuals that are not connected to the main network are three males from the Kokoalongo group (food-sharing network) and one male from Kokoalongo and three nulliparous females from Ekalakala (coalition network). science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
The bonobo cooperative acts vary in their susceptibility to cheating and delay of potential payoff. Grooming in bonobos is a frequent behavior that necessitates little initial investment by actors and allows immediate return; it therefore involves opportunities to both test partners’ willingness to cooperate and to receive immediate reward. Bonobo coalition formation requires joint action against a common opponent, which may provide benefits to all partners. Finally, food transfer (hereafter “sharing”) is an act that can incur an initial cost to actors (reduced energetic or nutrient intake) and offers little guarantee of a future return, which results in an uncertain payoff to actors. Increased defection opportunities in between-group food sharing relative to grooming are visible in our dataset. Across the 2 years of observation, food
sharing was reciprocated only among 4 of the 28 between-group dyads that shared food. In comparison, grooming investment was highly reciprocated within and between groups (fig. S1), and >70% of between-group grooming interactions involved immediate return (within the same interaction). The variability of the risk that benefits will be reciprocated among these cooperative interactions provides a platform through which to explore the underlying structures of cooperation across borders in a nonhuman species. In human networks, population structures that allow for the positive assortment of cooperators support the proliferation of cooperation by pooling the benefits among those who cooperate (9–11, 13). Cooperation assortment refers to the proclivity of individuals to selec-
A
B
Grooming
Food sharing
Cooperation score Food sharing Coalition Female-Male Male-Male
Cooperation score Grooming Coalition Female-Male Male-Male
-4 -3 -2 -1 0 1
D
Within-groups 1.00 0.75 0.50 0.25 0.00 0.25
0.50
0.75
1.00
Sharing in-degree
Grooming in-degree
C
-4 -3 -2 -1 0 1 Within-groups 1.00 0.75 0.50 0.25 0.00 0.00
Grooming out-degree
F
Between-groups 1.00 0.75 0.50 0.25 0.00 0.00
0.25
0.50
0.75
Grooming out-degree
0.50
0.75
1.00
Sharing out-degree
1.00
Sharing in-degree
Grooming in-degree
E
0.25
Between-groups 1.00 0.75 0.50 0.25 0.00 0.0
0.1
0.2
0.3
Sharing out-degree
Fig. 2. Positive assortment underlies bonobo cooperation across borders. (A and B) Estimates based on models that test whether the joint within-group cooperative tendencies of individuals (i.e., cooperation score) or interactions in other currencies explain their likelihood to groom (A) or share food (B) across groups. The cooperation score is calculated as the sum of the cooperative tendencies of a dyad based on either within-group grooming (A) or food donations (B). Shown are the model estimates (black diamonds) with 50% (yellow rectangles) and 95% (black lines) credible intervals derived from a Bayesian Poisson regression. (C to F) Generous bonobos are more attractive as interaction partners. The relationships between grooming [(C) and (E)] and food-sharing [(D) and (F)] in- and out-degree within [(C) and (D)] and between [(E) and (F)] groups are shown. The in-degree represents the number of partners from whom an individual received grooming or food, whereas the out-degree represents the number of partners that an individual groomed or donated food to. The in- and out-degrees are proportional to the potential number of partners that one can interact with within and between groups. Shown are the data points (N = 31 individuals; females are represented by circles and males by squares), model estimates (yellow, within-group; blue, betweengroup), and the 95% credible intervals (gray) derived from a Bayesian regression. SCIENCE science.org
tively interact with others who have similar cooperative tendencies or traits and requires interaction strategies to be nonrandom. To investigate whether nonrandom assortment underlies bonobo cooperation across social borders, we first established that bonobos interacted nonrandomly and exhibited a preference for specific partners both within and between groups. Specifically, using data permutations (see supplementary materials), we calculated the expected variance of bonobo within- and between-group grooming, coalition, and food-sharing interactions if interactions were randomly distributed among available partners and compared this variance against the “true” interaction variance (observed variance). In accordance with assortativity predictions, the observed variance of all interaction types significantly exceeded what is expected by chance, both within and between groups (all P < 0.001; fig. S2). Partner selectivity in our bonobo population provides a basis upon which cooperation assortment can emerge. If bonobos selectively interact with partners who are more likely to cooperate, they can increase their chances for a positive net gain. To examine cooperation assortment, we operationalized bonobo cooperative tendencies using interindividual variation in grooming, coalition formation, and food-sharing acts expressed within groups (fig. S3). We constructed a social network for each of the bonobo interactions and examined whether individuals categorized as high cooperators within their groups are more likely to connect groups in the different networks (Fig. 1). The social network analyses illustrate that bonobos who groom, form coalitions, or donate more food to in-group members are more likely to form the same kinds of connections with out-group members (Fig. 1 and fig. S4). By contrast, within-group cooperative tendencies within one form of cooperation do not appear to predict connections between groups in another form of cooperation (figs. S5 to S7). These patterns may emerge if bonobos have consistent cooperative tendencies within, rather than across, the different cooperation forms—whether interacting with in-group or out-group. It is unlikely that kin relations explain the observed cooperative patterns between groups because none of the between-group dyads were parentoffspring and only 5, 12, and 11% of the dyads that groomed, formed a coalition, and shared food between the groups, respectively, were identified as at least second-degree relatives (see supplementary materials). Overall, the social networks suggest that bonobos who more frequently cooperate within their groups are (i) more likely to engage in the same behavior with out-group individuals relative to less-frequent within-group cooperators and (ii) engage with out-group members 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
807
RES EARCH | R E S E A R C H A R T I C L E S
who are also frequent within-group cooperators (assortment of cooperators). To test these two characteristics of the bonobo networks, we used Bayesian Poisson regression models. First, we tested how individual within-group cooperative tendencies affected cooperative relationships with out-group members. Accounting for interaction opportunities, we found that bonobos who showed higher within-group food-sharing or coalitionary tendencies were also more likely to form the same kind of connections with out-group individuals [food sharing: estimate = 0.74, 95% credible interval (CI95%) = 0.38 to 1.14, odds ratio 2.1 per 1–standard deviation (SD) increase, table S1; coalition: estimate = 1.11, CI95% = 0.78 to 1.47, odds ratio 3 per 1-SD increase, table S2]. We did not find the same pattern when using grooming as the cooperative behavior (table S3). Identical analyses conducted to examine the impact of the within-group cooperative tendencies on between-group interactions across behaviors supported the idea of consistent cooperative tendencies within, but not across, currencies (tables S1 to S3). Finally, we also found that bonobos who had higher sharing tendencies, but not higher grooming or coalition tendencies, within groups were less likely to engage in aggression with out-group members (estimate = −0.36, CI95% = −0.53 to −0.20; fig. S5 and table S4). Competition and aggression may jeopardize cooperative relationships, particularly with a delay between action and reward, as with food sharing. As such, the reduced likelihood of between-group aggressions in those with high food-sharing tendencies may offer a pathway to foster the maintenance of food-sharing relationships across groups. Cooperation assortment can emerge if individuals interact with those who are more able and willing to confer benefits (33). Accordingly, we would expect individuals who are higher cooperators within their groups to be more likely to cooperate with out-group members who are themselves higher cooperators. Therefore, in a second step, we tested whether the combined within-group cooperation tendencies of dyads (examining each cooperation form separately) predicted interactions between groups. Consistent with this prediction, cooperation patterns between groups were best explained by the sum of the within-group cooperative tendencies of partners (i.e., “cooperation score”). A 1-SD increase in the food sharing, coalition, or grooming cooperation scores increased betweengroup food sharing, coalition, or grooming odds by factors of 1.77, 2.63, and 1.35, respectively (Fig. 2, fig. S8, and table S5). Further, partners who groomed more frequently also had a higher relative sharing probability (estimate = 0.33, CI95% = 0.05 to 0.61, odds ratio 1.39 per 1-SD increase), and vice versa (estimate = 0.35, CI95% = 0.25 to 0.46, odds ratio 1.42 per 1-SD increase). The sex combination 808
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
of the partners had no obvious consistent impact on their cooperation patterns (Fig. 2, fig. S8, and table S5). The between-group connections of cooperators in the different networks confirm the presence of cooperation assortment in bonobos and that reciprocity might be at play in supporting bonobo cooperative acts. Nonetheless, strong connections of high cooperators between groups may also arise as a result of random processes. For example, in a population with random partner selection and varying interindividual sharing tendencies, food sharing is expected to accumulate between those individuals who share more. The same is true for the other cooperative interactions. Such a process can generate stronger cross-group connections between high within-group cooperators by chance, without the need for preferential interactions. However, given that partner selection in Kokolopori bonobos is not random (see permutation procedure), the observed assortment of cooperators in the different networks is unlikely to be a mere artifact of random processes. Further, in the context of food sharing, that only four (14%) between-group dyads reciprocated food sharing further strengthens the idea that cooperation assortment is not a product of chance. Finally, cooperation can evolve when moregenerous individuals are also more likely to obtain benefits, whether through reciprocity or because of their attractiveness as interaction partners. Subsequently, it is predicted that those who are more able or willing to benefit others will be more readily chosen by others as interaction partners. We therefore examined whether bonobos who groomed or donated food to more partners (“high outdegree”) also received grooming or food from more partners (“high in-degree”) by dividing the number of partners each individual groomed or donated food to or received food from by the total number of potential partners with whom one could have interacted. Following this procedure, we generated values between 0 and 1, with 1 indicating that an individual interacted with all potential partners. We could not evaluate the same question for coalition formation because there is no clear actor or receiver in this type of interaction. We found a strong relationship (Fig. 2) between the grooming and food sharing in- and out-degrees of bonobos both within groups [grooming: regression coefficient (R) = 0.99; food sharing: R = 0.81] and between groups (grooming: R = 0.94; food sharing: R = 0.41). Overall, individuals that were more generous, by grooming more partners or donating food to more individuals, were also more likely to receive benefits from more individuals. The in- and out-degree grooming patterns were similarly strong within and between groups, likely because of high grooming return in bonobos (fig. S1). In comparison, the relation-
ship between food-sharing in- and out-degree was especially evident within groups, where repeated interaction opportunities, and hence reciprocity or knowledge about partners, are more certain (53% of within-group dyads shared food reciprocally versus 14% of between-group dyads). Discussion
Bonobos, our closest living relative together with chimpanzees, maintain stable but variable cooperative relationships that transcend group boundaries. The bonobo population structure permits repeated interactions and partner selectivity, which are fundamental components for the emergence of cooperative relationships. Bonobo nonkin cooperation across groups included food sharing, a prosocial act with uncertain returns and a high defection probability that is considered to be a key aspect of the human collaborative foraging niche. Food sharing in humans promotes cooperative relationships that support our expensive life histories by offsetting the risk of food shortfalls (34). Although the relevance of nonkin food sharing to the survival and reproduction of bonobos is unknown, we find that, like humans, bonobo sharing networks rely on the positive assortment of cooperators. Nonetheless, the processes that support cooperation assortment in humans and bonobos likely differ. Our results suggest that various mechanisms, including reciprocal altruism (i.e., cooperative investments based on past return) and partner choice, may explain cooperation assortment between bonobo groups. Whereas reciprocal altruism requires repeated interactions between the same partners and some form of bookkeeping, partner choice assumes stable interindividual differences in cooperative characteristics and that individuals have knowledge of the characteristics of potential partners to make effective choices (35). Both repeated interactions and knowledge about partners are nontrivial when interactions occur between members of different social groups. As such, the high group-membership fluidity, transient interactions, and large social network in human societies at present may limit the efficiency and explanatory power of these mechanisms. Instead, cultural processes and norm psychology are suggested to stabilize assortment and largescale cooperation in humans (2, 13, 19, 20). In bonobos, the small social network relative to that of humans and the frequent associations between individuals from different groups (32) facilitate interactions and the accumulation of social knowledge beyond one’s own group (e.g., which out-group individuals are more able and willing to confer benefits than others). Such a social structure provides a basis upon which partner familiarity can enforce reciprocity beyond group boundaries and cooperation between groups can be sustained. science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
Although social norms and culture are considered the main cooperation mechanisms that sustain human large-scale cooperation, it is reasonable to assume that reciprocity may have played a more crucial role in our evolutionary past when living in smaller societies. Bonobo society offers a rare opportunity to study a social system in which individuals from different groups engage in resource and commodity exchange. Our research builds upon previous experiments in captivity that show that bonobo food sharing with unrelated and unfamiliar individuals is not solely motivated by selfish interests or immediate rewards (30, 31). The convergence of evidence from both wild and captive studies suggests that the xenophilic tendencies of bonobos may be intrinsic to the species as a whole. The tolerant and cooperative between-group relations of bonobos (22–26, 28, 32) stand in contrast to the ubiquitously hostile betweengroup relations and the strong in-group favoritism observed in their sister species, chimpanzees [Pan troglodytes (36–38)]. A leading hypothesis in Pan speciation posits that bonobos have experienced a selection against aggression [the self-domestication hypothesis (39, 40)], which has led to a reduction of in-group favoritism and an overall increase in prosocial tendencies toward others, whether in-group or out-group (41). Therefore, it is often assumed that bonobos are inclined to interact prosocially with everyone, both strangers and familiar individuals, and that it is their nondiscriminatory prosocial tendencies that permit out-group relationships and cooperation. Although the marked prosocial tendencies that the Kokolopori bonobos exhibit toward out-groups support the selfdomestication hypothesis, our findings additionally suggest that bonobo prosocial tendencies are discriminatory. Instead of generally high prosocial tendencies, we propose that it is the strategic social ties that bonobos form with those who are more able and likely to confer benefits to others that stabilizes bonobo cooperation across borders. Bonobos are not the only nonhuman species that exhibit cooperative relationships between nonkin across groups (14–17). For example, in bottlenose dolphins, males form “third-order alliances” with unrelated out-group males that allow them to successfully compete over females (16). Third-order social alliances in dolphins are akin to the bonobo coalitions that are formed between males and females of different groups. In both species, these interactions confer improved access to contested resources to all allies and/or increased social status, which is therefore better categorized as mutualism. However, bonobo cooperation across borders also includes a behavior that cannot be explained by mutualism, thereby incorporating cooperation aspects that are considered exclusive to humans, such as the SCIENCE science.org
ability to act prosocially toward unrelated out-group members with no guarantee of a return. Owing to habituation status and data collection constraints, our study only included two out of at least four bonobo groups that are observed to regularly interact within the Kokolopori population (32), resulting in an investigation of a relatively small social network (31 individuals). Although these bonobos maintain a wider social network than that represented here, the overall size of this bonobo social network is limited and residential mobility is considerably reduced compared with humans (32), making it challenging to examine cooperation mechanisms of bonobos and humans on an equal footing. Further, unlike bonobos, cooperation across groups in humans extends beyond pair-wise interactions to also include large-scale cooperation (1–3). Nevertheless, our bonobo results suggest that the higher-order social connections and cooperative pair-wise relationships that humans form across groups may have a different evolutionary history than is often assumed. Theories in human evolution posit that pair bonding, exogamy, and the ability to recognize maternal and paternal kin and affines (i.e., relatives by marriage) are necessary components of between-group bonds and cooperation (19, 42). Here, bonobos offer an alternative scenario, in which cooperative ties between groups are formed in the absence of exogamy or strong bilateral kin recognition. It is therefore plausible that an ancestral state of human between-group, pair-wise cooperation is that of a bonobo-like social system, in which tolerance toward out-groups facilitates the emergence of cooperation in the absence of high degrees of genetic relatedness. Consequently, bonobos offer a key comparative model to human social systems and a rare opportunity to reconstruct the ancestral conditions of human large-scale cooperation. RE FERENCES AND NOTES
1. R. Boyd, P. J. Richerson, Evol. Anthropol. 31, 175–198 (2022). 2. J. Henrich, M. Muthukrishna, Annu. Rev. Psychol. 72, 207–240 (2021). 3. M. B. Brewer, Am. Psychol. 62, 726–738 (2007). 4. R. Axelrod, W. D. Hamilton, Science 211, 1390–1396 (1981). 5. T. Clutton-Brock, Nature 462, 51–57 (2009). 6. R. Boyd, P. J. Richerson, J. Theor. Biol. 132, 337–356 (1988). 7. I. Eshel, L. L. Cavalli-Sforza, Proc. Natl. Acad. Sci. U.S.A. 79, 1331–1335 (1982). 8. S. Bowles, Science 314, 1569–1572 (2006). 9. M. Dyble et al., Curr. Biol. 26, 2017–2021 (2016). 10. C. L. Apicella, F. W. Marlowe, J. H. Fowler, N. A. Christakis, Nature 481, 497–501 (2012). 11. A. Romano, D. Balliet, T. Yamagishi, J. H. Liu, Proc. Natl. Acad. Sci. U.S.A. 114, 12702–12707 (2017). 12. H. Ohtsuki, C. Hauert, E. Lieberman, M. A. Nowak, Nature 441, 502–505 (2006). 13. C. Handley, S. Mathew, Nat. Commun. 11, 702 (2020). 14. E. J. H. Robinson, J. L. Barker, Biol. Lett. 13, 20160793 (2017). 15. C. C. Grueter et al., Trends Ecol. Evol. 35, 834–847 (2020).
16. R. C. Connor, M. Krützen, S. J. Allen, W. B. Sherwin, S. L. King, Proc. Natl. Acad. Sci. U.S.A. 119, e2121723119 (2022). 17. I. Krams, T. Krama, K. Igaune, R. Mänd, Behav. Ecol. Sociobiol. 62, 599–605 (2008). 18. E. Fehr, U. Fischbacher, Nature 425, 785–791 (2003). 19. C. L. Apicella, J. B. Silk, Curr. Biol. 29, R447–R450 (2019). 20. R. Boyd, P. J. Richerson, Philos. Trans. R. Soc. Lond. Ser. B 364, 3281–3288 (2009). 21. A. B. Migliano et al., Sci. Adv. 6, eaax5913 (2020). 22. L. Cheng et al., Horm. Behav. 128, 104914 (2021). 23. T. Furuichi, Int. J. Primatol. 41, 203–223 (2020). 24. G. Itani, Afr. Study Monogr. 11, 153–186 (1990). 25. T. Sakamaki, H. Ryu, K. Toda, N. Tokuyama, T. Furuichi, Int. J. Primatol. 39, 685–704 (2018). 26. N. Tokuyama, T. Sakamaki, T. Furuichi, Am. J. Phys. Anthropol. 170, 535–550 (2019). 27. L. Cheng, L. Samuni, S. Lucchesi, T. Deschner, M. Surbeck, Anim. Behav. 187, 319–330 (2022). 28. B. Fruth, G. Hohmann, Hum. Nat. 29, 91–103 (2018). 29. L. Samuni, F. Wegdell, M. Surbeck, eLife 9, e59191 (2020). 30. J. Tan, B. Hare, PLOS ONE 8, e51922 (2013). 31. J. Tan, D. Ariely, B. Hare, Sci. Rep. 7, 14733 (2017). 32. L. Samuni, K. E. Langergraber, M. H. Surbeck, Proc. Natl. Acad. Sci. U.S.A. 119, e2201122119 (2022). 33. K. M. Smith, C. L. Apicella, Evol. Hum. Behav. 41, 354–366 (2020). 34. H. S. Kaplan, P. L. Hooper, M. Gurven, Philos. Trans. R. Soc. Lond. Ser. B 364, 3289–3299 (2009). 35. R. McElreath, R. Boyd, Mathematical Models of Social Evolution: A Guide for the Perplexed (Univ. Chicago Press, 2007). 36. R. W. Wrangham, Am. J. Phys. Anthropol. 110, 1–30 (1999). 37. M. L. Wilson et al., Nature 513, 414–417 (2014). 38. L. Samuni, C. Crockford, R. M. Wittig, Nat. Commun. 12, 539 (2021). 39. B. Hare, V. Wobber, R. Wrangham, Anim. Behav. 83, 573–585 (2012). 40. R. W. Wrangham, Proc. Natl. Acad. Sci. U.S.A. 115, 245–253 (2018). 41. J. Tan, B. Hare, in Bonobos, B. Hare, S. Yamamoto, Eds. (Oxford Univ. Press, 2017), pp. 140–154. 42. B. Chapais, Evol. Anthropol. 22, 52–65 (2013). 43. L. Samuni, LiranSamuni/Bonobo-cooperation-across-borders: Cooperation across borders in bonobos. Zenodo (2023); https://doi.org/10.5281/zenodo.8156761. AC KNOWLED GME NTS
We thank the Institut Congolais pour la Conservations de la Nature (ICCN) and the Ministry of Scientific Research and Technology in the DRC for their support and permission to work in the Kokolopori Bonobo Reserve, DRC. We also thank the Bonobo Conservation Initiative and Vie Sauvage, especially S. Coxe, A. Lotana Lokasola, A. Menante, and staff members of the Kokolopori Bonobo Research Project for supporting our work. We thank L. Vigilant and V. Städele for conducting the genetic analysis and J. Henrich, C. Curtin, and E. Wessling for helpful discussions and comments. Funding: This work was funded by Harvard University (M.S.), the Max Planck Society (M.S., L.S.), the British Academy (L.S.), and Deutsche Forschungsgemeinschaft (L.S.). Author contributions: Conceptualization: L.S., M.S.; Methodology: L.S., M.S.; Investigation: L.S.; Funding acquisition: L.S., M.S.; Project administration: M.S.; Writing – original draft: L.S.; Writing – review and editing: L.S., M.S. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All processed data and code used in the analyses are available on Zenodo (43). License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/ about/science-licenses-journal-article-reuse SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adg0844 Materials and Methods Figs. S1 to S9 Tables S1 to S5 References (44–55) MDAR Reproducibility Checklist Submitted 2 December 2022; accepted 29 September 2023 10.1126/science.adg0844
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
809
RES EARCH | R E S E A R C H A R T I C L E S
SOLAR CELLS
Bimolecularly passivated interface enables efficient and stable inverted perovskite solar cells Cheng Liu1†, Yi Yang1†, Hao Chen1,2†, Jian Xu2†, Ao Liu1†, Abdulaziz S. R. Bati1, Huihui Zhu1, Luke Grater2, Shreyash Sudhakar Hadke3, Chuying Huang1, Vinod K. Sangwan3, Tong Cai1,4, Donghoon Shin3,4, Lin X. Chen1, Mark C. Hersam1,3,5, Chad A. Mirkin1,3,4, Bin Chen1*, Mercouri G. Kanatzidis1*, Edward H. Sargent1,2,5* Compared with the n-i-p structure, inverted (p-i-n) perovskite solar cells (PSCs) promise increased operating stability, but these photovoltaic cells often exhibit lower power conversion efficiencies (PCEs) because of nonradiative recombination losses, particularly at the perovskite/C60 interface. We passivated surface defects and enabled reflection of minority carriers from the interface into the bulk using two types of functional molecules. We used sulfur-modified methylthio molecules to passivate surface defects and suppress recombination through strong coordination and hydrogen bonding, along with diammonium molecules to repel minority carriers and reduce contact-induced interface recombination achieved through field-effect passivation. This approach led to a fivefold longer carrier lifetime and one-third the photoluminescence quantum yield loss and enabled a certified quasi-steadystate PCE of 25.1% for inverted PSCs with stable operation at 65°C for >2000 hours in ambient air. We also fabricated monolithic all-perovskite tandem solar cells with 28.1% PCE.
C
ertified power conversion efficiencies (PCEs) > 25% have been widely reported for perovskite solar cells (PSCs) in the regular (n-i-p) structure (1–3). Although inverted (p-i-n) PSCs have potential advantages because of their stability, lowtemperature processing, and compatibility with integration into tandem solar cells (4–8), their reported PCEs rarely surpass 24% under the stringent quasi-steady-state (QSS) protocol (9–11). This efficiency gap is primarily attributed to higher recombination rates at the interface between the perovskite and the charge transport materials (12). The detrimental impact of buried perovskite/hole transport layer interface losses has been addressed through the development of self-assembled monolayers (13–15). However, the top interface between the perovskite and the electron transport layer (ETL), typically made from C60 and its derivatives, suffers from interfacial recombination that results from minority carriers in the vicinity of the interface as well as the effect of incompletely passivated trap states (16). Surface passivation can suppress interface charge recombination and has been accomplished with organohalides (4, 17–19), Lewis bases (20, 21), and dipolar compounds (22, 23). 1
Department of Chemistry, Northwestern University, Evanston, IL 60208, USA. 2Department of Electrical and Computer Engineering, University of Toronto, Toronto, ON M5S 1A4, Canada. 3Department of Materials Science and Engineering, Northwestern University, Evanston, IL 60208, USA. 4 International Institute for Nanotechnology, Northwestern University, Evanston, IL 60208, USA. 5Department of Electrical and Computer Engineering, Northwestern University, Evanston, IL 60208, USA. *Corresponding author. Email: [email protected] (B.C.); [email protected] (M.G.K.); ted.sargent@ northwestern.edu (E.H.S.) These authors contributed equally to this work.
810
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
We noted that reliance on a single species of molecule may fail to address simultaneously both surface and interface recombination processes (Fig. 1A) (24, 25). Specifically, the existence of near-interface minority carriers (holes in the perovskite layer) leads to direct interface recombination with majority carriers (electrons in the ETL), a process that can occur even at nondefect sites (26). In addition, defects at the perovskite surface induce surface recombination through trapping of carriers. The most common defect, the halide vacancy, has a low formation energy (27, 28). Exploring class 1 and 2 molecule combinations
We sought to address complex interface carrier recombination issues using a combination of different molecules, each with distinct functionalities. The first class of molecule we incorporated repelled hole carriers to reduce interface recombination through field-effect passivation (Fig. 1B). The second class of molecule interacted with defect sites to form chemical bonds to reduce surface recombination through chemical passivation. Diammonium ligands, in which one -NH3+ group anchors to the perovskite surface and the other extends away from it, can induce a surface dipole and n-type doping (29, 30) and provide effective field-effect passivation for both narrow bandgap (NBG) (~1.2 eV) and wide bandgap (WBG) (~1.8 eV) PSCs (26). We explored the passivation effect of different diammonium ligands on normal bandgap (~1.5 eV) PSC devices. The device architecture consisted of fluorine-doped tin oxide (FTO)/ NiOx/[4-(3,6-dimethyl-9H-carbazol-9-yl)butyl] phosphonic acid (Me-4PACz)/perovskite/ passivation layer/C60/bathocuproine (BCP)/ Ag (fig. S1). Optical constants of materials
are shown in fig. S2. The current-voltage characteristics showed that ethane-1,2-diammonium (EDAI2) and propane-1,3-diammonium iodide (PDAI2), with high binding energy with the perovskite surface (fig. S3), enabled a device performance improvement compared with the control devices (without passivation), from PCEs of ~22.8% to ~23.9% with active areas of 0.05 cm2 (Fig. 1, C and D). Thus, diammonium ligands work well in normal bandgap PSCs, and the PCE improvement could be explained by field-effect passivation that repels minority carriers (26). We then sought a second molecule to add a chemical passivation function. We first examined n-butylammonium iodide (BAI), which is widely used as a chemical passivating agent (31, 32), in combination with PDAI2. Its addition increased PCE to ~24.3% compared with PDAI2 alone. Extending the chain length to amylamine hydroiodide (AAI) further improved the average efficiency to ~24.5%, thus providing a baseline roughly at parity with efficient previously reported inverted PSCs (12, 17). We then tuned the electrical dipole moment by incorporating sulfur as a donor atom in the alkyl chain by synthesizing methylthio-based ammonium ligands, namely 2-(methylthio) ethylamine hydroiodide (2MTEAI) and 3(methylthio)propylamine hydroiodide (3MTPAI). The use of both diammonium and methylthio molecules led to improved PCE across several combinations, namely EDAI2/2MTEAI, PDAI2/ 2MTEAI, EDAI2/3MTPAI, and PDAI2/3MTPAI, in comparison to both the control device and single-molecule passivated devices. The highest average PCE (>25.5%) was achieved with PDAI2/3MTPAI (Fig. 1D). We thus focused on the PDAI2/3MTPAI combination for further investigation. PDAI2/3MTPAI characteristics
We used density function theory (DFT) to compare 3MTPA+ versus AA+ by modeling ligand orientations of 3MTPA and AA on the perovskite surface (fig. S4). The binding energy difference (DEclean) between the parallel (Eclean-parallel) and vertical configurations (Eclean-vertical) was used as a measure of ligand orientation. A larger DEclean value of −0.22 eV for 3MTPA indicated a stronger preference for the parallel orientation compared with −0.13 eV for AA (Fig. 2A). This difference corresponded to greater occupation on the vacancy defect position. Electrostatic potentials in the ligands (Fig. 2B) showed that 3MTPA, when compared with AA, had a lower minimum electrostatic potential (ϕmin) because of its electron-rich center surrounding the S atom that facilitated binding with the positively charged iodide vacancy. The higher maximum electrostatic potential (ϕmax) at the -NH3+ side also added increased binding strength between the ligand and the surface cation vacancy site of the perovskite. science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
Fig. 1. Passivation at the perovskite/ETL interface. (A and B) Schematic of the perovskite surface without passivation (A) and with both chemical and field-effect passivation (B). (C) Chemical structures of the diammonium and ammonium ligands investigated in this study. (D) PCEs of control versus passivated PSCs using different passivation ligands.
The passivation effect of the ligands was further evaluated by considering the presence of iodide vacancies, which are the predominant defects on the perovskite surface (fig. S5) (33). We assessed the binding energy difference (DErelative) between that of the defective surface (EVI-parallel) and the clean surface (Eclean-parallel). DErelative of AA with the perovskite remains nearly unchanged, regardless of the presence or absence of iodide vacancy. In contrast, DErelative = −0.38 eV was obtained for 3MTPA, indicating a favorable interaction with the defective sites. 3MTPA induced a notable charge redistribution that accumulated charges at iodide vacancy, assigned to S-Pb coordination bonding (Fig. 2C). Charge transfer between 3MTPA and formamidinium (FA) was also observed (Fig. 2D), accompanied by a shorter distance of 2.72 Å between the sulfur atom in 3MTPA and the hydrogen atom in FA that indicated the formation of a hydrogen bond. In contrast, the distance between the carbon atom at the corresponding site in AA and the hydrogen atom in FA was 3.33 Å. Thus, the formation energy of the FA vacancy increased from −0.79 eV to −0.71 eV (fig. S6). Proton nuclear magnetic resonance (1H NMR) spectra showed that the amino proton peak of FAI at d = 8.82 parts per SCIENCE science.org
million (ppm) exhibited increased broadening and shifted to a lower field after mixing with 3MTPAI compared with AAI (Fig. 2E). These changes again indicated stronger hydrogen bonding interactions between 3MTPA and FA than between AA and FA (34). Computation work suggested that the methylthio group provided stronger binding—viewed in some studies as a proxy for stronger passivation— compared with ligands that relied on ammonium functional groups alone. We assessed diammonium-methylthio dual passivation (DMDP) using x-ray photoelectron spectroscopy (XPS). The Pb 4f peaks of the passivated perovskite film shifted toward a lower binding energy of 0.23 eV compared with the control film (Fig. 2F), which we attributed to an increased electron density at Pb2+ (35). Time-of-flight secondary ion mass spectrometry (ToF-SIMS) was used to analyze the surface ligand distribution. Comparing signal ratios of PDA:3MTPA and PDA:AA under identical conditions on perovskite films with PDAI2/3MTPAI and PDAI2/AAI bimolecular passivation, we found that the signal ratio of 1:2.7 for PDA: 3MTPA was lower than the ratio of 1:1.1 for PDA:AA (Fig. 2G). This suggests that 3MTPA has a stronger binding affinity to the perovskite surface and a better passivation effect
on defects, consistent with its higher binding energy (fig. S7). Figure 2H illustrates the centimeter-scale photoluminescence (PL) intensity distribution of a perovskite film with Gaussian-distributed passivators on the surface (fig. S8) (36). The region surrounding the PDAI2/3MTPAI center exhibited higher PL emission than the corresponding region for PDAI2/AAI; and the contour region with the lowest PL intensity was skewed toward the PDAI2/AAI center. Scanning electron microscopy (SEM) images revealed dense polygonal grains with sizes of ~500 nm for the control perovskite film; the morphologies were unchanged after DMDP passivation (fig. S9). Grazing-incidence wideangle x-ray scattering (GIWAXS) did not reveal any peaks at low scattering vectors q in the DMDP-based film, which indicated that no lowdimensional perovskite formed (Fig. 3A). We ascribed the peak at ~0.84 Å−1 in the control sample to the presence of d-FAPbI3 formed in the ambient humid air during the measurement. The suppression of d-FAPbI3 in the DMDP-based film indicates improved ambient stability. The unchanged surface dimensionality was further corroborated by transient absorption (TA) spectra where the passivated film displayed a single bleach spectral 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
811
RES EARCH | R E S E A R C H A R T I C L E S
Fig. 2. The process of passivation through the methylthio group on perovskite surfaces. (A) Binding energies of ammonium ligands with a clean or defective perovskite surface with the typical iodide vacancy. (B) Electrostatic potential (ϕ) of 3MTPA and AA ligands (ϕmax, blue color; ϕmin, red color). (C) Calculated charge density difference (blue, depletion; yellow, accumulation) of anchoring ammonium ligands onto the perovskite surface with I vacancies. The red open circles indicate the positions of I vacancies. The atoms in the structures are differentiated by different colors: S is represented by grass green, C by brown, Pb by gray,
feature from the three-dimensional perovskite (fig. S10). To examine the chemical passivation effect of the DMDP strategy on film optoelectronic properties, we conducted time-resolved photoluminescence (TRPL) measurements (Fig. 3B and table S1). Control perovskite films showed a sharp decrease in emission characteristic of high levels of nonradiative carrier recombination on the bare perovskite surface. Treatment with PDAI2 showed little improvement in lifetime, reflecting its limited suppression of defect-induced surface recombination. In contrast, the perovskite film treated with 3MTPAI displayed a sustained plateau in the decay curve, reflecting increased carrier lifetime. This might be due to a combined effect of reduced nonradiative traps and enhanced photon recycling (37). The reemission of photons from the WBG subcell might serve to augment photon absorption of the adjacent NBG subcell when it is integrated in the tandem configuration. We used ultraviolet photoelectron spectroscopy (UPS) to characterize band edge energies 812
17 NOVEMBER 2023 ¥ VOL 382 ISSUE 6672
and I by purple. (D) Calculated charge density difference showing the hydrogen bond formation between 3MTPA and FA. (E) The proton NMR spectra of FAI, FAI with AAI, and FAI with 3MTPAI. (F) High-resolution Pb 4f XPS peaks of the perovskite films. (G) SIMS mapping of signal ratios of 3MTPA:PDA and AA:PDA for perovskite samples with PDAI2/3MTPAI (1:2 molar ratio) and PDAI2/AAI (1:2 molar ratio) bimolecular passivation. (H) PL intensity distribution of the 1 cm by 1 cm perovskite film postsynthetic treated by spray coating with ink 1 of PDAI2/3MTPAI and ink 2 of PDAI2/AAI solution centered around the diagonal corners.
(Fig. 3C and fig. S11). PDAI2 treatment reduced the energy level difference between the conduction band minimum (ECBM) and the Fermi level (EF) of the perovskite surface to 0.10 eV, compared with 0.20 and 0.17 eV for the control and 3MTPAI treatments, respectively. The stronger n-type doping effect of PDAI2 was attributed to the additional -NH3+ group extending away from the perovskite matrix that induced a surface dipole that repelled the minority carrier at the interface (26, 38). This treatment enabled field-effect passivation and reduced interface recombination (fig. S12). We expect this passivation effect to be retained upon incorporation of 3MTPAI because n-type doping was also observed. Photovoltaic performance
PDAI2 did not enhance the PL quantum yield (PLQY) (Fig. 3D) of the perovskite film before C60 deposition, and there was little PLQY loss after coating with C60, consistent with its fieldeffect passivation role. Increased PDAI2 concentration led to a decrease in both PLQY and PCE,
which we attributed to increased surface recombination. For 3MTPAI, in the absence of C60, PLQY increased as 3MTPAI processing solution concentration increased from 3 mM to 15 mM. On contact with C60, noticeable PLQY and PCE losses were seen, and these losses became more pronounced at higher concentrations, indicating increased interface recombination. The DMDP strategy improved PLQY of the perovskite/C60 samples and increased the PCE to >26% even at 12 mM concentration of 3MTPAI. In sum, 3MTPAI and PDAI2 could increase passivation and decrease carrier recombination without interfering with one another (Fig. 3E). The photovoltaic parameters of devices with different treatments at the optimized concentration are summarized in fig. S13. The DMDP-based devices showed an improved PCE from 22.8 ± 0.4 to 25.5 ± 0.3% compared with the control device, accompanied by enhancements in open-circuit voltage (VOC) from 1.12 ± 0.01 to 1.16 ± 0.01 V and fill factor (FF) from 78.5 ± 1.3 to 83.8 ± 1.3%. The science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
Fig. 3. DMDP strategy working principle. (A) GIWAXS patterns of the control and DMDP-based perovskite films. (B) TRPL of the perovskite films treated with different ligands. (C) The energy level difference between the conduction band maximum and Fermi level for the perovskite films treated with different ligands. (D) Ligand concentrationÐdependent PLQY of the perovskite films, PLQY loss of
diode characteristics of the devices in the absence of light showed that the devices in which we used the DMDP strategy presented an average dark saturation current (J0) reduction by two orders of magnitude compared with the control devices, demonstrating effective inhibition of carrier recombination (fig. S14) (39). Figure 4A shows the current density–voltage (J-V) curves for the champion DMDP device, which exhibited a PCE of 26.4%, with a short-circuit current (JSC) of 26.2 mA cm−2, VOC of 1.17 V, and FF of 85.8%. We focused on QSS measurement in certification. Here, the highest performance based on maximum power point tracking (MPPT) was a PCE of 25.5% for 100 s (fig. S15). A National Renewable Energy Laboratory (NREL) certification that used the asymptotic maximum power scan protocol (Fig. 4B and fig. S16) reported a QSS PCE of 25.1% for an illuminated area of SCIENCE science.org
the perovskite films after C60 deposition, and PCEs of devices. For DMDP, the concentration of 3MTPAI was varied, while the PDAI2 concentration was optimized and maintained at 6 mM. (E) Schematic diagram showing the inhibition of interface recombination by PDAI2 and the suppression of defectinduced recombination by 3MTPAI.
0.05 cm2 along with a fast-scan PCE of 25.9%, compared with other reported certified QSS PCEs that did not exceed 25% (Fig. 4C and table S2) (9, 10, 40, 41). We also fabricated 1.5 cm2 devices using the DMDP treatment that delivered a PCE of 24.0% (fig. S17), consistent with increased film homogeneity and reduced localized nonradiative recombination (fig. S18). Longevity studies
In our studies of the thermal stability of encapsulated devices, we found that after 1600 hours of thermal aging at 85°C in nitrogen (ISOS-D-2 protocol, where ISOS is the International Summit on Organic PV Stability), the DMDP-based devices retained 95% of initial PCE, surpassing the retention of 84% for the control device (Fig. 4D and fig. S19). We investigated the operating stability under MPPT under 1 sun of an encapsulated device operating in ambient air. After
2000 hours of continuous operation under 1 sun illumination at 65°C (ISOS-L-3 protocol), the DMDP-based device maintained 96% of original PCE, whereas the control device was reduced to 70% of initial PCE (Fig. 4E and fig. S20). A comparison with other PSCs tested using the ISOS-L-3 protocol is provided in table S3. Discussion
To investigate the applicability of the DMDP strategy on other perovskite compositions, we fabricated PSCs with both WBG and NBG perovskite materials. Notably, the average PCE was improved by 14 and 13% when the DMDP strategy was used for NBG and WBG PSCs, respectively (Fig. 4F). We applied the DMDP strategy to monolithic all-perovskite tandem solar cells with the structure of FTO/NiOx/ Me-4PACz/WBG perovskite/C60/SnOx/Au/ poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS)/NBG perovskite/C60/ 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
813
RES EARCH | R E S E A R C H A R T I C L E S
Fig. 4. Device performance and stability. (A) J-V characteristics of the bestperforming DMDP-based device. (B) QSS J-V curve of one representative DMDP-based device certified at NREL. (C) Certified QSS and fast-scan PCE statistics of inverted PSCs. (D) Thermal stability of encapsulated PSCs annealed at 85°C in nitrogen. (E) Operational stability of encapsulated PSCs
SnOx/Ag (Fig. 4G). The J-V characteristics of the champion tandem device (Fig. 4H) with an illuminated area of 0.05 cm2 exhibited a PCE of 28.1% with a VOC of 2.14 V, JSC of 15.6 mA cm−2, and FF of 84.0%, and a stabilized PCE of 27.1% under MPPT. A well-matched current response is seen in the external quantum efficiency (EQE) spectra (fig. S21). Realizing both chemical and field-effect passivation by the combined use of methylthio and diammonium molecules has mitigated complex carrier recombination issues at the perovskite/ ETL interface. We consider the multimolecule passivation approach, along with diverse functionalities, as a promising direction for exploring next-generation passivation strategies to achieve improved performance and stability in perovskite optoelectronics. 814
17 NOVEMBER 2023 ¥ VOL 382 ISSUE 6672
under constant 1-sun illumination at 65°C in air with 50% relative humidity. (F) PCEs of the WBG and NBG PSCs showing the universality of the DMDP strategy. (G) Architecture of the tandem device. (H) J-V characteristics of the best-performing tandem device based on the DMDP strategy. The inset shows the stabilized PCE under MPPT.
RE FERENCES AND NOTES
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20.
M. Kim et al., Science 375, 302–306 (2022). Y. Zhao et al., Science 377, 531–534 (2022). Y. Ding et al., Nat. Nanotechnol. 17, 598–605 (2022). T. Li et al., Nat. Energy 8, 610–620 (2023). X. Zheng et al., Nat. Energy 8, 462–472 (2023). R. Lin et al., Nature 620, 994–1000 (2023). J. Tong et al., Nat. Energy 7, 642–651 (2022). J. Xu et al., Science 367, 1097–1104 (2020). W. Peng et al., Science 379, 683–690 (2023). Q. Jiang et al., Nature 611, 278–283 (2022). Q. Cao et al., Sci. Adv. 7, eabg0633 (2021). S. Liu, V. P. Biju, Y. Qi, W. Chen, Z. Liu, NPG Asia Mater. 15, 27 (2023). F. Ali, C. Roldán-Carmona, M. Sohail, M. K. Nazeeruddin, Adv. Energy Mater. 10, 2002989 (2020). K. Almasabi et al., ACS Energy Lett. 8, 950–956 (2023). Q. Tan et al., Nature 620, 545–551 (2023). F. Ye et al., Nat. Commun. 13, 7454 (2022). H. Chen et al., Nat. Photonics 16, 352–358 (2022). R. Lin et al., Nature 603, 73–78 (2022). D. H. Kim et al., Joule 3, 1734–1745 (2019). Z. Li et al., Science 376, 416–420 (2022).
21. X. Li et al., Science 375, 434–437 (2022). 22. P. Caprioglio et al., Energy Environ. Sci. 14, 4508–4522 (2021). 23. Z. Zhu et al., Joule 6, 2849–2868 (2022). 24. W. Yang et al., Research Square [Preprint] (2022); https://doi. org/10.21203/rs.3.rs-2147188/v1. 25. Z. Zhang et al., Chem. Soc. Rev. 52, 163–195 (2023). 26. H. Chen et al., Nature 613, 676–681 (2023). 27. D. Meggiolaro, F. De Angelis, ACS Energy Lett. 3, 2206–2222 (2018). 28. A. Walsh, S. D. Stranks, ACS Energy Lett. 3, 1983–1990 (2018). 29. S. Hu et al., Energy Environ. Sci. 15, 2096–2107 (2022). 30. C. Quarti, F. De Angelis, D. Beljonne, Chem. Mater. 29, 958–968 (2017). 31. S. Sidhik et al., Science 377, 1425–1430 (2022). 32. A. R. Mohd Yusoff et al., Energy Environ. Sci. 14, 2906–2953 (2021). 33. J. Jeong et al., Nature 592, 381–385 (2021). 34. L. Zhu et al., Nat. Commun. 12, 5081 (2021). 35. Z. Wang et al., Nano Energy 59, 258–267 (2019). 36. E. J. Kluender et al., Proc. Natl. Acad. Sci. U.S.A. 116, 40–45 (2019).
science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
37. D. W. deQuilettes et al., Chem. Rev. 119, 11007–11019 (2019). 38. A. Liu et al., InfoMat 5, e12386 (2023). 39. A. Cuevas, Energy Procedia 55, 53–62 (2014). 40. X. Wu et al., Adv. Mater. 35, e2208431 (2023). 41. F. Li et al., Nat. Photonics 17, 478–484 (2023). ACKN OW LEDG MEN TS
Part of the research described in this paper was performed at the Canadian Light Source, a national research facility of the University of Saskatchewan, which is supported by the Canada Foundation for Innovation (CFI), the Natural Sciences and Engineering Research Council (NSERC), the National Research Council (NRC), the Canadian Institutes of Health Research (CIHR), the Government of Saskatchewan, and the University of Saskatchewan. A.S.R.B. acknowledges support from King Abdullah University of Science and Technology (KAUST) through the Ibn Rushd Postdoctoral Fellowship Award. Funding: This work was supported under award number OSR-CRG2020-4350.2. E.H.S. acknowledges support from
the Office of Naval Research (ONR) grant N00014-20-1-2572). M.G.K. was supported by ONR grant N00014-20-1-2725. C.A.M. was supported by the Army Research Office under grants W911NF23-1-0141 and W911NF-23-1-0285 and by the Sherman Fairchild Foundation, Inc. This work made use of the SPID, EPIC, and Keck-II facilities of Northwestern University’s NUANCE Center, which has received support from the SHyNE Resource (NSF ECCS2025633); the International Institute of Nanotechnology, Northwestern University; and Northwestern’s MRSEC program (NSF DMR-1720139). Charge transport characterization was supported by the National Science Foundation (NSF) Materials Research Science and Engineering Center (MRSEC) at Northwestern University under award number DMR-1720319. This work was partially supported by award 70NANB19H005 from the US Department of Commerce, National Institute of Standards and Technology, as part of the Center for Hierarchical Materials Design (CHiMaD). Author contributions: Conceptualization: C.L. and Y.Y. DFT calculation: J.X. Device fabrication: C.L., Y.Y., and H.C. Writing – original draft: C.L. and Y.Y. Writing – review & editing: A.L. and H.Z. XPS and UPS characterization: A.S.R.B. GIWAXS characterization: L.G. Electrical
CATALYSIS
Nickel-catalyzed ester carbonylation promoted by imidazole-derived carbenes and salts Changho Yoo1 , Shrabanti Bhattacharya1, Xin Yi See2, Drew W. Cunningham1, Sebastian Acosta-Calle1, Steven T. Perri2, Nathan M. West2, Dawn C. Mason2, Chris D. Meade2, Christopher W. Osborne2, Phillip W. Turner2, Randall W. Kilgore2, Jeff King2, Jeffrey H. Cowden2, Javier M. Grajeda2*, Alexander J. M. Miller1* Millions of tons of acetyl derivatives such as acetic acid and acetic anhydride are produced each year. These building blocks of chemical industry are elaborated into esters, amides, and eventually polymer materials, pharmaceuticals, and other consumer products. Most acetyls are produced industrially using homogeneous precious metal catalysts, principally rhodium and iridium complexes. We report here that abundant nickel can be paired with imidazole-derived carbenes or the corresponding salts to catalyze methyl ester carbonylation with turnover frequency (TOF) exceeding 150 hour–1 and turnover number (TON) exceeding 1600, benchmarks that invite comparisons to state-of-the-art rhodium-based systems and considerably surpass known triphenylphosphine-based nickel catalysts, which operate with TOF ~7 hour–1 and TON ~100 under the same conditions.
P
rocesses using carbon monoxide for the production of organic chemicals containing the acetyl (–C(O)CH3) group are foundational to the chemical industry (1–6). Methanol carbonylation reactors produce over 13 million tons of acetic acid per year (1, 2). Analogous carbonylation of esters generates carboxylic acid anhydrides that are widely applied in industrial acetylation and dehydration reactions (3–7). Organic esters of cellulose, for example, are produced from acetylation of cellulose with anhydrides (8–10). These cellulose esters have wide-ranging applications including in coatings, optical films, and membrane structures (11). Commercial alcohol and ester carbonylations are amongst the highest-volume applications 1
Department of Chemistry, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA. 2Eastman Chemical Company, Kingsport, TN, USA.
*Corresponding author. Email: [email protected] (J.M.G.); [email protected] (A.J.M.M.) †Present address: Department of Chemistry, Ulsan National Institute of Science and Technology (UNIST), Ulsan 44919, Republic of Korea.
SCIENCE science.org
of homogeneous catalysis in current practice, yet these catalysts rely on the rarest transition metal elements on the planet: rhodium and iridium. The price of rhodium in particular increased from ~$700 per troy ounce in 2016 to ~$29,000 per troy ounce in 2021 (~$25 per gram to $1000 per gram) and has remained volatile since then (12). Beyond the obvious financial benefits of moving away from scarce resources and their attendant economic price volatility, alternative Earth-abundant metal catalysts would help address ethical concerns associated with mining precious metals (13–15). On a per-gram basis, rhodium mining is estimated to be the largest contributor to global warming of all metals; it is also amongst the most ecologically and toxicologically damaging (14, 16). Nickel complexes were once considered promising homogeneous carbonylation catalysts, as reflected in patents (17–22) and the peer-reviewed literature (23–33). There are inherent economic advantages of an abundant and inexpensive nickel catalyst: a metric ton
characterization: S.S.H., V.K.S., and M.C.H. TA measurement: C.H. and L.X.C. Ink spray coating: T.C. and D.S. Supervision on ink spray coating: C.A.M. Supervision: B.C., M.G.K., and E.H.S. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data are available in the main text or the supplementary materials. License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www. science.org/about/science-licenses-journal-article-reuse SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adk1633 Materials and Methods Figs. S1 to S21 Tables S1 to S3 References (42–50) Submitted 7 August 2023; accepted 6 October 2023 10.1126/science.adk1633
of nickel can cost the same as a troy ounce of rhodium. However, a long list of chemical disadvantages is apparent in the earlier carbonylation studies. Most of these revolve around the supporting ligand, usually a tertiary phosphine, which can dissociate under the high CO pressures, generating highly toxic and catalytically inactive Ni(CO)4 (29, 30). To mitigate this, high catalyst loadings were employed (often 0.1 M nickel concentration, >1 mol% loading) and a large excess of tertiary phosphine (>4 equivalents, or 0.4 M) was added. Even then, the promoter MeI, which facilitates formation of a nickel methyl intermediate, reacts with the free phosphine ligand to form the phosphonium salt [MePR3][I], reducing effective phosphine concentration (29, 30). The unwanted phosphine methylation also complicates ligand tuning: more electron-donating ligands would accelerate MeI oxidative addition but any advantage is lost because these ligands are also more prone to formation of the phosphonium salt (29). Figure 1A summarizes the prior state of the art, which requires high loadings of nickel and methyl iodide, and additional salt promoters (e.g., LiI, LiOAc) and reductants [e.g., H2, Mo(CO)6]. As rhodium and iridium carbonylation catalysts advanced to commercial viability, research in nickel-catalyzed carbonylation was largely abandoned (2). Reflection on the historical period in which nickel-catalyzed carbonylation reactions were first explored revealed an opportunity. During the main burst of research activity into nickel carbonylation catalysis, phosphines were unrivaled as supporting ligands in organometallic catalysis. Not long after interest in industrially relevant nickel carbonylation catalysis had faded, however, came the advent of N-heterocyclic carbenes (NHCs) as ligands par excellence for catalysis. Although NHC-based catalysts have now replaced many traditional phosphine-based catalysts (34–38), we are not aware of any studies of NHC-supported nickel catalysts for acetyl synthesis (39). As a result of 17 NOVEMBER 2023 ¥ VOL 382 ISSUE 6672
815
RES EARCH | R E S E A R C H A R T I C L E S
D
Fig. 1. Nickel-catalyzed methanol and methyl ester carbonylation. (A) Summary of prior state-of-the-art nickel carbonylation catalysis (X is commonly iodide or acetate). (B) Nickel-based carbonylation catalysts reported here. (C) Summary of methanol carbonylation with (IPr)Ni(CO)3. (D) Summary of methyl ester carbonylation with nickel-based catalysts. (E) Initial optimization of methyl propionate carbonylation comparing the turnover number (TON) of acetyl anhydrides, methyl acetate, and acetic acid with various nickel catalysts. Error bar is the standard deviation of total acetyl TON from five independent experiments (table S4). Conditions: 200°C,
their strong electron-donating nature (40, 41), NHC ligands have higher binding affinity to nickel (42) and better promote CO dissociation relative to phosphines (42, 43). Using NHC-supported nickel catalysts (Fig. 1B) as a starting point has led us to conditions for the carbonylation of alkyl esters with high activity at low catalyst loading. Up to 60% conversion of the neat ester to acetyls is observed, corresponding to a turnover number (TON) exceeding 1600 and a turnover frequency (TOF) 816
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
50 bar CO, 15 hours (except for entry 5, 45 hours), 20 mL methyl propionate, 0.075 mmol catalyst, and 7.5 mmol MeI. (F) Comparison of various nickel systems and a typical rhodium precursor under the conditions of (E) (except for entry 5, 37 mmol MeI with one repressurization of CO). (G) Gas uptake traces during catalysis with Ni(OAc)2•4H2O/2IPr•HCl with a single pressurization (blue), and with Ni(OAc)2•4H2O/2IPr with two repressurizations (black) under the conditions of (E).The initial turnover frequency (TOFinit) was determined from a linear fit to the pressure change (red dashed lines).
over 150 hour–1, with good selectivity for anhydride while maintaining low loadings of nickel and methyl iodide. Preliminary mechanistic studies are consistent with ligated nickel active catalysts accessible from a surprisingly wide range of diarylimidazole derivatives. Reaction optimization
Initial studies focused on methanol (MeOH) carbonylation in methyl propionate (EtCO2Me) solvent with the known NHC complex (IPr)Ni(CO)3
(42) (IPr is 1,3-bis(2,6-diisopropylphenyl)imidazol2-ylidene, Fig. 1B). Methyl propionate was strategically chosen over methyl acetate (MeOAc) to ensure accurate product analysis. Hydrolysis of MeOAc would produce an acetyl (acetic acid) without any carbonylation taking place, but with EtCO2Me, acetyls can only be produced by carbonylation (fig. S8). A solution of (IPr)Ni(CO)3 and MeI in 1:1 MeOH:EtCO2Me was pressurized to 50 bar CO and heated to 200°C for 15 hours, producing MeOAc and HOAc with a science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
Fig. 2. Mechanistic studies. (A) General reaction pathway mediated by a ligated nickel complex. (B) General reaction pathway mediated by simple nickel salt ion pair and LiI. (C) Plausible interactions of imidazole derivatives with nickel. (D) Species detected by IR during catalysis and by mass spectrometry after catalysis (R = Et), and IR spectrum obtained during catalysis (R = iPr).
total TON of 63 (Fig. 1C and fig. S9). However, substantial solvent loss was observed during the reaction, attributed to ~60% conversion of MeOH to the unwanted byproduct dimethyl ether (major product by 1H NMR spectroscopy). Hypothesizing that MeOH activation was obscuring the true catalytic activity, we removed MeOH from the reaction mixtures and turned our attention to direct methyl ester carbonylation. Excellent carbonylation activity was observed when methyl propionate alone was used as the neat substrate (Fig. 1D). Heating 20 mL (208 mmol) EtCO2Me at 200°C for 15 hours in the presence of 0.075 mmol (IPr)Ni(CO)3 (0.036 mol% catalyst loading), 7.5 mmol MeI, and 50 bar CO generated ~13 mmol acetic anhydrides [some scrambling of anhydrides occurs (44–46), figs. S11 and S12], along with ~2 mmol acetic acid and ~0.5 mmol methyl acetate (Fig. 1E). The turnover number corresponding to the total of all acetyls, TONtot, was 218 ± 19 (average of five independent experiments; based on this and other replicate data, we estimate a 7% uncertainty in all TONtot valSCIENCE science.org
ues, table S4). The reaction parameters of temperature, CO pressure, MeI concentration, and IPr loading were systematically varied, revealing strong temperature and MeI concentration dependence and nonlinear CO and IPr dependences (figs. S14 to S19). An increase in activity by a factor of three was achieved by using (IPr) Ni(CO)3 and one equivalent of free IPr, TONtot = 603 (22% acetyl yield). The selectivity for anhydrides was over 80% in these reactions. Commercial reagents can be employed equally well for this reaction, providing a convenient alternative to synthesizing and isolating an organometallic complex ahead of time (Fig. 1E). Using nickel(II) acetate hydrate (Ni(OAc)2•4H2O) and two equivalents of the free IPr gave TONtot = 685 (25% yield). Lowering the catalyst loading to 0.025 mmol (only 0.012 mol%, 120 ppm) Ni(OAc)2•4H2O while maintaining the amount of IPr enhanced the productivity, increasing TONtot to 1509 (~18% yield). Furthermore, the simple imidazolium salt IPr•HCl proved to be an equally viable precursor (TONtot = 755, 27% yield). Imidazolium salts, the synthetic pre-
cursors to NHCs, are attractive as they require one less synthetic step and are air-stable. Durable performance was apparent in preliminary studies. Online pressure monitoring shows relatively linear pressure drops until below ~30 bar, at which point reactions slowed. The average TOF (TOFavg) of Ni(OAc)2•4H2O/2IPr• HCl over 15 hours was ~50 hour–1, based on 1H NMR analysis (table S13). The initial TOF was 59 hour–1, calibrated to gas uptake kinetics (which provides initial TOF values for any reaction, even those that have nonlinear gas uptake; see SM for details). In a 45-hour experiment in which the reactor was repressurized at 15 and 30 hours, TOFinit decreased from 32 to 22 hour–1 (Fig. 1G). The decrease could be due to changes in concentration of methyl propionate and methyl iodide [MeI concentration drops by ~15% in a typical 15-hour reaction, as expected as ester is consumed (3, 4, 7)], and/or some ligand or catalyst decomposition. Even after 45 hours, however, the catalyst was still active; after the longer reaction time, the acetyls yield approximately doubled, reaching 40% (TONtot = 1100, Fig. 1E). 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
817
RES EARCH | R E S E A R C H A R T I C L E S
Fig. 3. Understanding how imidazole structure influences carbonylation. (A) Under the standard conditions shown at the top, various imidazole derivatives were tested in methyl propionate carbonylation. Note that an isolated sample of (IAd)Ni(CO)2 (51) gave TONtot = 141. The area of the circles is proportional to TONtot; blue circles correspond to NHCs and orange circles correspond to imidazolium salts. Full details are given in table S11, and correlations with various ligand parameters (41, 52) are found in figs. S21 to S24. (B) Gas uptake traces during EtCO2Me carbonylation catalysis with 0.075 mmol Ni(OAc)2•4H2O and 0.15 mmol of IPr (black), IPr•HCl (blue), and IPr-CH3+ (orange) showing differences in activity over 12 hours at 200°C, 7.5 mmol MeI. The initial turnover frequency (TOFinit) was determined from a linear fit to the pressure change (red dashed lines).
Benchmarking comparisons to nickel phosphine catalysts
Comparisons to the Ni/PPh3 catalyst system were carried out next (Fig. 1F). Under the standard conditions employed above, but with the phosphine-based catalyst Ni(PPh3)2(CO)2, TONtot was 108 (3.9% yield) and TOFavg was 7 hour–1. A mixture of NiI2/LiI/2PPh3, chosen based on prior reports (28–30), gave similar performance to the isolated bis(phosphine) complex. In the case of imidazole derivatives with LiI, NiI2/ LiI/2IPr catalyzed carbonylation with TOFinit of 153 hour–1 (TONtot = 883, 32% yield; Fig. 1F), a more than 20-fold increase in activity relative to PPh3-based catalysis. LiI cannot replace MeI, however, as no acetyls were observed when including LiI but omitting MeI. The imidazolebased catalysts have TON, yield, and TOF values approximately one order of magnitude higher than phosphine-based catalysts in these experimental comparisons. In previous reports of Ni carbonylation of MeOAc, 30 to 50% acetyl yields were achieved at approximately 100-fold higher catalyst and ligand loadings, for example 1 mol% (0.05 M) Ni and 0.1 M PPh3, 55 bar CO, plus an additional 13 bar H2, a very high 1.3 M MeI content, and 0.4 M LiOAc (31). The high Ni loading limits the TON below 90; the TOF was ~50 hour–1. The present Ni/NHC catalyst system achieves higher yield and higher activity at much lower loading of nickel, ligand, and MeI, and without requiring H2 gas or other additives. The low nickel loading not only leads to much 818
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
higher TON values than prior reports but also minimizes risk associated with Ni(CO)4 formation. To address the possible formation of highly toxic Ni(CO)4 in this reactor configuration, carbonylation reactions were vented into a solution of iodine in acetone for quenching prior to work up and analyses (see SM for details). Inductively coupled plasma optical emission spectroscopy (ICP-OES) analysis of the quench solutions showed ~1 ppm Ni after quenching 7 reactors. The low ligand loading is also helpful given that imidazolium salts are more expensive than triphenyl phosphine. The low loading of toxic methyl iodide further improves the safety profile of the reaction. To more directly compare activity and yield, an experiment was conducted at the same MeI content as the prior literature (31). With 37 mmol MeI, with or without LiI, carbonylation proceeded with TOFinit 120 to 170 hour–1 (table S13); after 22 hours (with 2 repressurizations to replenish CO), the TON was ~1600 and the acetyls yield reached 60% (table S12). Scale-up under continuous CO delivery
The nickel-catalyzed carbonylation was then run at mole scale in larger reactors using industrially relevant esters. Methyl isobutyrate is a feedstock for carbonylation catalysis, with resulting anhydrides used for cellulose functionalization (8, 9, 47) and the generation of polyester monomer precursors (48). Heating 1.7 moles of methyl isobutyrate in the presence of 0.75 mmol (0.044 mol%) Ni(OAc)2•4H2O,
3.75 mmol of IPr ligand, and 75 mmol of MeI at 200°C under continuous 50 bar CO for 15 hours produced acetyls with TONtot up to 620 (fig. S37 and table S17). Similar results were obtained with IPr•HCl, even when the reactor was charged in air before pressurization. These larger-scale reactions were run with a continuous CO feed, such that a constant pressure was maintained. Under continuous pressures of 50 bar CO and 33 bar H2, the reaction afforded comparable yields of acetyls (table S17). Nickel/phosphine catalysts often require substantial partial pressures of H2 gas to achieve optimum performance (23, 25–28, 30–32). The imidazole-based catalyst system is unaffected by H2, simplifying the setup and confirming that the system tolerates CO with >3% H2, the typical quality requirement for applications in the chemical industry (49). The CO uptake was steady and aligned nicely with the yield of acetyls (fig. S38), confirming that the catalyst remained active during the experiment, just as in the smaller scale reactions. Operando IR spectroscopic monitoring showed that the anhydride C=O stretching frequency increased with the same kinetics as gas uptake (figs. S33 and S38). Mole-scale methyl acetate carbonylation worked similarly well, with TOFinit = 45 hour–1 (table S16). Acetic anhydride is the major product, formed with a turnover number of 545 (average of two runs), corresponding to ~17% yield. This is a lower limit of total acetyls yield because acetic acid was not included as some of it could have been produced from hydrolysis. The yield science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
of acetic anhydride will likely be limited in batch reactions because the overall reaction is isoergic at 200°C (see SM for details) (50). Efficient separation in a continuous process helps overcome the equilibrium limitation in industrial ester carbonylation (3, 7). Industrial carbonylation is currently dominated by precious metal rhodium and iridium catalysts, so it is useful to compare the performance of different metal complexes under the present conditions. Methyl propionate carbonylation under the standard conditions, but with the Ni/NHC catalyst system replaced by Rh(acetylacetonate)(CO)2, resulted in lower activity (TONtot of 509, Fig. 1F and table S10). Methyl isobutyrate carbonylation by RhCl3 in the presence of LiI and LiOAc promoters gave approximately four times faster CO uptake than the Ni/NHC system (fig. S39). Although comprehensive comparisons between Rh and Ni catalyst systems are challenging, particularly given that industrial applications might involve different optimized process conditions and recycling considerations, our studies show that under the present conditions the nickel system promoted by imidazole affords rates that are comparable to a rhodium catalyst. Mechanistic considerations
Next, preliminary mechanistic studies were carried out to provide initial guidance for future catalyst development. The similar activity of carbene and imidazolium derivatives led us to consider two general pathways in Fig. 2, A and B: (A) a nickel complex catalyst, ligated by an imidazole derivative and (B) the anionic nickel species [Ni(CO)3I]– ion pairing with imidazolium. In each case, a thermally promoted organocatalytic cycle regenerates methyl iodide, consistent with the sharp drop in activity at reduced temperatures (fig. S19) (3, 4, 7). Moser et al. ruled out an ion-pairing mechanism in the Ni/PPh3 ester carbonylation system based on in situ IR spectroscopy (29, 30). Schaub and coworkers proposed that both phosphine-ligated and ion-pairing mechanisms were operative in the carbonylation of higher alcohols with Ni/PR3 systems, with large amounts of LiI shifting the system toward the ion-pairing mechanism (32). Ligand identity has a substantial impact on performance only when a ligated pathway is at play. Several factors suggest that ligation is essential in the Ni/imidazole system. First, the identity of the ligand is critical (Fig. 3A). Structural variations in aryl-containing imidazole derivatives result in greater than twofold differences in TON (and TOF). Alkylcontaining ligands give much lower TON values. Such a major influence of structure variations on activity is highly unlikely in an ion pair. Prior studies have also shown that ligands with extremely bulky alkyl groups directly bound to the imidazole nitrogen atoms, such as 1,3-bis (1-adamantyl)imidazol-2-ylidene (IAd) bind only SCIENCE science.org
weakly to Ni, generating catalytically inactive Ni(CO)4 under CO (51). So IAd would seem likely to promote [Ni(CO)3I]– formation, yet the activity is lower than aryl-substituted imidazole derivatives (52). Second, added iodide salts are not required for efficient catalysis. Without ample free iodide to displace CO and form [Ni(CO)3I]–, a ligated pathway is more likely. Adding a full 100 equiv LiI only increased the TOFinit from ~57 hour–1 to 91 hour–1 (table S13), whereas substantial differences were observed as a function of iodide content in cases where the ion-pairing pathway is operating (32). Third, there is no evidence for formation of Ni(CO)4 (2044 cm–1) or [Ni(CO)3I]– (1955 cm–1) during operando IR spectroscopic monitoring of methyl isobutyrate carbonylation (Fig. 2D). Instead, signals at 2058 and 1978 cm–1 are observed for the C≡O stretches previously assigned to (IPr)Ni(CO)3 (42). This key observation establishes that the imidazolium salt IPr•HCl readily undergoes metalation to Ni under the reaction conditions. The two other signals are tentatively assigned as (IPr)2Ni(CO)2 based on comparisons to other bis(NHC) nickel carbonyl complexes (53). Although there is clear evidence for ligated nickel under the reaction conditions starting from Ni(OAc)2 and IPr•HCl, it is also possible that the active catalyst is not (IPr)Ni(CO)3. One possibility is that the imidazole unit may be methylated during catalysis, which could lead to a different mode of ligation. Free NHCs are rapidly methylated in the reaction medium, even at room temperature. Mass spectrometry analysis of post catalysis reaction mixtures shows that although the imidazole core remains intact, all nonligated IPr•HCl is converted to IPr-CH3+ and other products with additional methylation (Fig. 2D). These methylation reactions occur even without nickel, upon heating imidazolium precursors with MeI under 50 bar CO at 200°C. Less alkylation is observed when nickel is present (figs. S29 and S30), however, suggesting that some of the imidazole derivatives are bound to nickel during catalysis and thus protected from alkylation. At room temperature, no reaction is observed between (IPr)Ni(CO)3 and MeI, consistent with some degree of protection of the carbene. Catalytic reactions using isolated IPr-CH3+ exhibit good activity, albeit with TON and TOFinit values ~20 to 30% lower than reactions using IPr or IPr•HCl (Fig. 3B and table S14). The viability of 2-methyl-imidazolium salts in the catalytic system prompted us to consider mechanisms by which these salts could ligate nickel. Figure 2C shows three possibilities. First, IPrCH3+ species can be deprotonated (C–H activation) to form N-heterocyclic olefins (NHOs) (54, 55), which are known to coordinate various transition metal ions. Alternatively, IPrCH3+ could undergo C–C bond activation to
directly access the same (IPr)Ni intermediates expected from (IPr)Ni(CO)3 precatalysts. Finally, the dominance of aryl-substituents raises the possibility of arene p-coordination to nickel(0) or nickel(II). To distinguish between these possibilities, a labeling study employing IPr-13CH3+ was carried out. After catalysis, 7% natural abundance imidazolium derivatives were present by mass spectrometry (fig. S28), indicating that some C–C bond activation was occurring (a reaction that has been observed for other metals) (56, 57). Thus, at least some degree of C–C activation appears to be possible, although more work will be needed to further probe the structure of the active catalyst. The mechanistic studies provide initial insight into the viability of not only free carbenes but also protonated or alkylated imidazole groups to promote high activity carbonylation and point toward aryl substituents as key features in catalyst design. Given that metal-mediated interconversions between binding modes of imidazoles are known (58, 59), perhaps this versatility itself plays a key role in engendering excellent carbonylation performance relative to traditional phosphine ligands. Outlook
The nickel catalyst system can leverage a wide variety of imidazole derivatives to reach performance levels far beyond prior nickel catalysts with phosphine ligands, and even comparable to precious metal catalysts. This work opens the door for careful reconsideration of this abundant and inexpensive first-row transition metal for large-scale carbonylation reactions. Future efforts in this direction would likely require not only advances in chemistry, such as enhanced mechanistic understanding and development of a continuous process, but also iterative risk assessments, technoeconomic analyses, and life-cycle assessments. REFERENCES AND NOTES
1. C. Le Berre, P. Serp, P. Kalck, G. P. Torrence, in Ullmann’s Encyclopedia of Industrial Chemistry (Wiley, 2014), vol. 10, pp. 1–34. 2. P. Kalck, C. Le Berre, P. Serp, Coord. Chem. Rev. 402, 213078 (2020). 3. M. Beller et al., in Applied Homogeneous Catalysis with Organometallic Compounds (Wiley, 2017), pp. 91–190. 4. J. R. Zoeller, Catal. Today 140, 118–126 (2009). 5. G. Morris, in Mechanisms in Homogeneous Catalysis (Wiley, 2005), vol. 54, pp. 195–230. 6. H. Held, A. Rengstl, D. Mayer, in Ullmann’s Encyclopedia of Industrial Chemistry (Wiley, 2000), vol. 8, pp. 239–255. 7. J. R. Zoeller, Org. Process Res. Dev. 20, 1016–1025 (2016). 8. M. El-Sakhawy, S. Kamel, A. Salama, H.-A. Sarhan, J. Drug Deliv. 2014, 575969 (2014). 9. K. Balser et al., in Ullmann’s Encyclopedia of Industrial Chemistry (Wiley, 2004), vol. 7, pp. 333–380. 10. C. L. Elkins, R. Lin, M. J. Rodig, R. J. Sharpe, Regioselectively Substituted Cellulose Esters, US Patent 11,136,414 B2 (2021). 11. K. J. Edgar et al., Prog. Polym. Sci. 26, 1605–1688 (2001). 12. A. Cowley, “PGM market report” (Johnson Matthey, 2023); https://matthey.com/en-US/products-and-markets/ pgms-and-circularity/pgm-markets/pgm-market-reports. 13. J. R. Ludwig, C. S. Schindler, Chem 2, 313–316 (2017).
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
819
RES EARCH | R E S E A R C H A R T I C L E S
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. 44. 45. 46. 47. 48.
49. 50. 51. 52. 53. 54. 55. 56. 57. 58. 59.
R. M. Bullock et al., Science 369, eabc3183 (2020). B. J. Glaister, G. M. Mudd, Miner. Eng. 23, 438–450 (2010). P. Nuss, M. J. Eckelman, PLOS ONE 9, e101298 (2014). A. N. Naglieri, N. Rizkalla, Preparation of Carboxylic Acid Anhydrides, US Patent 4,002,678 (1977). T. Isshiki, Y. Kijima, Y. Miyauchi, Process for Preparing Carboxylic Anhydrides, US Patent 4,239,698 (1980). J. Gauthier-Lafaye, R. Perron, Process for producing acetic acid by carbonylation, European Patent 0035458 (1981). J. Gauthier-Lafaye, R. Perron, Preparation of Anhydrides by Carbonylation of Esters, US Patent 4,353,844 (1982). A. N. Naglieri, N. Rizkalla, Preparation of carboxylic acids, US Patent 4,356,320 (1982). J. S. Kanel, S. J. Okrasinski, Nickel-Catalyzed Carbonylation Process, US Patent 5,900,504 (1999). N. Rizkalla, in ACS Symposium Series (ACS, 1987), Vol. 328, pp. 61–76. H. Adkins, R. W. Rosenthal, J. Am. Chem. Soc. 72, 4550–4553 (1950). A. A. Kelkar, R. S. Ubale, R. V. Chaudhari, J. Catal. 136, 605–608 (1992). A. A. Kelkar, R. S. Ubale, R. V. Chaudhari, J. Mol. Catal. 80, 21–29 (1993). A. A. Kelkar, R. S. Ubale, R. M. Deshpande, R. V. Chaudhari, J. Catal. 156, 290–294 (1995). R. S. Ubale, A. A. Kelkar, R. V. Chaudhari, J. Mol. Catal. Chem. 118, 9–19 (1997). W. R. Moser, B. J. Marshik-Guerts, S. J. Okrasinski, J. Mol. Catal. Chem. 143, 57–69 (1999). W. R. Moser, B. J. Marshik-Guerts, S. J. Okrasinski, J. Mol. Catal. Chem. 143, 71–83 (1999). J. Gong, Q. Fan, D. Jiang, J. Mol. Catal. Chem. 147, 113–124 (1999). S. Sabater et al., Organometallics 39, 870–880 (2020). N. Lichtenberger et al., Organometallics 41, 1184–1196 (2022). S. Díez-González, N. Marion, S. P. Nolan, Chem. Rev. 109, 3612–3676 (2009). R. Corberán, E. Mas‐Marzá, E. Peris, Eur. J. Inorg. Chem. 2009, 1700–1716 (2009). M. C. Jahnke, F. E. Hahn, in Transition Metal Complexes of Neutral eta1-Carbon Ligands, R. Chauvin, Y. Canac, Eds. (Springer, 2010), vol. 30 of Topics in Organometallic Chemistry, pp. 95–129. M. N. Hopkinson, C. Richter, M. Schedler, F. Glorius, Nature 510, 485–496 (2014). V. Ritleng, M. Henrion, M. J. Chetcuti, ACS Catal. 6, 890–906 (2016). M. Jeletic, A. Veige, in N-Heterocyclic Carbenes in Transition Metal Catalysis and Organocatalysis, C. Cazin, Ed. (Springer, 2010), vol. 32 of Catalysis by Metal Complexes, pp. 217–235. D. G. Gusev, Organometallics 28, 763–770 (2009). D. G. Gusev, Organometallics 28, 6458–6461 (2009). R. Dorta et al., J. Am. Chem. Soc. 127, 2485–2495 (2005). M. T. Lee, C. H. Hu, Organometallics 23, 976–983 (2004). C. D. Hurd, M. F. Dull, J. Am. Chem. Soc. 54, 3427–3431 (1932). K. Kikukawa, K. Kono, K. Nagira, F. Wada, T. Matsuda, Tetrahedron Lett. 21, 2877–2878 (1980). K. Kikukawa, K. Kono, K. Nagira, F. Wada, T. Matsuda, J. Org. Chem. 46, 4413–4416 (1981). C. J. Malm, L. W. Blanchard, Mixed cellulose esters containing isobutyryl groups, US Patent 2,828,303 (1958). C. E. Sumner, B. L. Gustafson, J. R. Knight, Process for the manufacture of 2,2,4,4-tetramethylcyclobutanediol, US Patent 5,258,556 (1993). W. Boll et al., in UllmannÕs Encyclopedia of Industrial Chemistry (John Wiley & Sons, Ltd, 2011), vol. 16, pp. 531–532. W. M. Haynes, Ed., CRC Handbook of Chemistry and Physics (CRC Press, Boca Raton, ed. 95, 2014). R. Dorta, E. D. Stevens, C. D. Hoff, S. P. Nolan, J. Am. Chem. Soc. 125, 10490–10491 (2003). H. Clavier, S. P. Nolan, Chem. Commun. 46, 841–861 (2010). N. M. Scott, H. Clavier, P. Mahjoor, E. D. Stevens, S. P. Nolan, Organometallics 27, 3181–3186 (2008). S. Naumann, Chem. Commun. 55, 11658–11670 (2019). S. M. I. Al-Rafia et al., Chem. Commun. 47, 6987–6989 (2011). A. R. Chianese, B. M. Zeglis, R. H. Crabtree, Chem. Commun. 2176–2177 (2004). E. A. Denisova, D. B. Eremin, E. G. Gordeev, A. M. Tsedilin, V. P. Ananikov, Inorg. Chem. 58, 12218–12227 (2019). K. Cavell, Dalton Trans. 6676–6685 (2008). A. Schumann, C. Hering-Junghans, Eur. J. Inorg. Chem. 2018, 2584–2588 (2018).
820
17 NOVEMBER 2023 ¥ VOL 382 ISSUE 6672
ACKN OWLED GMEN TS
The authors acknowledge H. Dodge for assistance with chemical inventory, Y. Li and J. Halderman for assistance setting up the operando IR system, and B. Ehrmann for assistance with mass spectrometry. Funding: This work was supported by the Eastman Chemical Company. NMR spectroscopy carried out at the University of North Carolina at Chapel Hill NMR Core Laboratory was supported by the National Science Foundation under grant CHE-1828183. Mass spectrometry was carried out at the University of North Carolina’s Department of Chemistry Mass Spectrometry Core Laboratory, supported by the National Science Foundation under grant CHE-1726291. Author contributions: Conceptualization: A.J.M.M., J.M.G., and C.Y. Methodology: A.J.M.M., J.M.G., C.Y., and X.Y.S. Investigation: C.Y., S.B., X.Y.S., D.W.C., S.A.C., S.T.P., C.D.M., C.W.O., P.W.T., R.W.K., J.K., and J.H.C. Funding acquisition: A.J.M.M. and J.M.G. Project administration: A.J.M.M., J.M.G., N.M.W., and D.C.M. Supervision: A.J.M.M. and J.M.G. Writing – original draft: C.Y. Writing – review and editing: C.Y., A.J.M.M., and J.M.G. Competing interests: A PCT has been filed and has published (WO 2023/049476 A1). Inventors: A.J.M.M., J.M.G., C.Y., D.W.C., N.M.W., X.Y.S., S.T.P., D.C.M., C.D.M. Applicants: The University of North Carolina at Chapel Hill and
Eastman Chemical Company. A provisional application has been filed. Inventors: A.J.M.M., J.M.G., S.B., and S.A.C. Applicants: The University of North Carolina at Chapel Hill and Eastman Chemical Company. Data and materials availability: All data are available in the main text or the supplementary materials. License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.sciencemag.org/about/science-licensesjournal-article-reuse SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.ade3179 Materials and Methods Supplementary Text Figs. S1 to S39 Tables S1 to S18 Equations S1 to S2 Submitted 9 August 2022; resubmitted 9 August 2023 Accepted 4 October 2023 10.1126/science.ade3179
METABOLISM
Autoregulatory control of mitochondrial glutathione homeostasis Yuyang Liu1, Shanshan Liu1, Anju Tomar2,3, Frederick S. Yen1, Gokhan Unlu1, Nathalie Ropek4, Ross A. Weber1 , Ying Wang1, Artem Khan1, Mark Gad1,5, Junhui Peng6, Erdem Terzi7, Hanan Alwaseem8, Alexandra E. Pagano8, Søren Heissel8, Henrik Molina8, Benjamin Allwein5, Timothy C. Kenny1, Richard L. Possemato7, Li Zhao6, Richard K. Hite5, Ekaterina V. Vinogradova4, Sheref S. Mansy2, Kıvanç Birsoy1* Mitochondria must maintain adequate amounts of metabolites for protective and biosynthetic functions. However, how mitochondria sense the abundance of metabolites and regulate metabolic homeostasis is not well understood. In this work, we focused on glutathione (GSH), a critical redox metabolite in mitochondria, and identified a feedback mechanism that controls its abundance through the mitochondrial GSH transporter, SLC25A39. Under physiological conditions, SLC25A39 is rapidly degraded by mitochondrial protease AFG3L2. Depletion of GSH dissociates AFG3L2 from SLC25A39, causing a compensatory increase in mitochondrial GSH uptake. Genetic and proteomic analyses identified a putative iron-sulfur cluster in the matrix-facing loop of SLC25A39 as essential for this regulation, coupling mitochondrial iron homeostasis to GSH import. Altogether, our work revealed a paradigm for the autoregulatory control of metabolic homeostasis in organelles.
C
ells require the ability to sense changes in the abundance of nutrients to ensure their efficient use for survival and growth under environmental perturbations (1). Although several nutrient sensing mech-
1
Laboratory of Metabolic Regulation and Genetics, The Rockefeller University, New York, NY, USA. 2Department of Chemistry, University of Alberta, Edmonton, Alberta, Canada. 3 Department of Cellular, Computational and Integrative Biology, Università di Trento, Trento, TN, Italy. 4Laboratory of Chemical Immunology and Proteomics, The Rockefeller University, New York, NY, USA. 5Structural Biology Program, Memorial Sloan Kettering Cancer Center, New York, NY, USA. 6 Laboratory of Evolutionary Genetics and Genomics, The Rockefeller University, New York, NY, USA. 7Department of Pathology, New York University Grossman School of Medicine, New York, NY, USA. 8The Proteomics Resource Center, The Rockefeller University, New York, NY, USA. *Corresponding author. Email: [email protected] †Present address: Department of Radiation Oncology, Memorial Sloan Kettering Cancer Center, New York, NY, USA.
anisms have been described for cytosolic metabolites such as amino acids and cholesterol (2), whether organelles sense and regulate metabolite availability is not fully understood. Mitochondria are semiautonomous compartments with endosymbiotic origins. As a source of oxidative reactions and with an independent genetic system, mitochondria must maintain optimal concentrations of nucleotides, amino acids, and antioxidant molecules to perform critical protective and biosynthetic functions. Indeed, previous work provided evidence for such homeostatic mechanisms for a subset of redox-active molecules (3, 4). Among these, glutathione (GSH) is a small-molecule thiol that is highly abundant in mitochondria and is required for antioxidant defense and iron homeostasis (5). Mitochondrial GSH availability is largely controlled by its import through the science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
SLC25A39
P2A
HEK293T-3xFLAG-SLC25A39-P2A-RFP
XBP1 ATF4 SLC25A39 OAZ2 OAZ1
10 4 10 3 10 2
HIF1A DELE1
10 1
FBXL5 10 0 -2 -1 10 10 10 0 10 1 10 2 10 3 10 4 10 5 RNA-TPM OpenCell Database
-40
RFP
-25
kDa -40
FLAG
-
+
-40
-Tubulin
-50
FLAG
-40
-Tubulin
-50
40
20
0
H
GSH -50
FLAG
kDa
SLC25A39
5-oxo-L-proline mitoCHAC1 ( aa1-8) L-cysteinylglycine
-25
-70
SLC25A12 4 -Tubulin 2
-50
cDNA:
0
HEK293T 3xHA-mCherry-OMP25 B/E treated
Mito IP
Lysis
Isolated mitochondria 0’ 30’ 60’ 120’
Immunoblotting
GSH added
E
HEK293T-MitoCHAC1-HA
Untreated
CHX/min: 0 20 40 60 80 100 SLC25A39 protein levels, relative to initial
-40
CHAC1
6
B/E
B/E + GSHee
HEK293T
8
D
DMSO -Tubulin
60
cDNA:
-70
SLC25A12
HEK293T-3xFLAG-SLC25A39
kDa
FLAG
BSO:
C
RFP
80
Ve M c su itoC tor pp H l. AC w / G 1, SH
3xFLAG-SLC25A39
G
P=0.0130 GSH abundance (normalized to NAD+)
10
5
C trl G S G H SH Tr ee o Li lox p C rox ys st ti at to ne in-1 N cop AC h er M ol ito BH Q 4
Protein-Copy Number
10
Mitochondrial
ns Translation
10 7 6
Whole Cell
RFP
Ve M c su itoC tor pp H l. AC w / G 1, SH
10 8
F 3xFLAG
Ve M su itoC ctor pp H l. AC w / G 1, SH
B
SLC25 family members Known posttranscriptionally regulated proteins
GSH abundance (normalized to NAD+)
A
1.0
GSH added
SLC25A39
kDa -40 -35
SLC25A12
-70
t1/2 = 326.3 min DMSO B/E B/E + GSHee
0.5
-35
MitoCHAC1-HA
ATP5A1
SLC25A11 Time/min:
-25
0
30
60 120
0
30
60 120
t1/2 = 14.24 min t1/2 = 16.84 min
0.0 0
50 CHX/min
100
Merge
Fig. 1. SLC25A39 is a short–half life protein regulated by mitochondrial GSH availability. (A) Scatter plot showing the protein copy number versus mRNA abundance (TPM) for all genes in HEK293T cells detectable across the proteome. Original data were retrieved from the OpenCell database. Green dots denote SLC25 family proteins, and purple dots denote representative proteins known to be regulated post transcriptionally. (B) (Top) Schematic showing the construct for cotranslational expression of 3xFLAG-tagged SLC25A39 and RFP, separated by a self-cleaving P2A peptide. (Bottom) Immunoblots of the indicated proteins in HEK293T cells expressing the aforementioned construct. Cells were treated with BSO (1 mM) for 48 hours and were then treated with GSH (10 mM), GSH ethyl ester (GSHee, 10 mM), Trolox (50 mM), Liproxstatin-1 (1 mM), cystine (200 mM), N-acetylcysteine (NAC, 1 mM), a-tocopherol (5 mM), MitoQ (30 nM), or BH4 (4 mM) for 8 hours. RFP was used as an internal control for the translational levels of the construct, and SLC25A12 was used as a loading control. (C) (Top) Immunoblots of the indicated proteins in HEK293T cells expressing 3xFLAG-SLC25A39 cDNA treated with cycloheximide (CHX, 50 mg/ml) for the indicated times. Prior to CHX treatment, cells were treated with BSO (1 mM) and erastin (5 mM) for 24 hours and GSH ethyl ester (GSHee, 10 mM) for 8 hours.
SLC25A39 transporter on the mitochondrial inner membrane (6, 7). SLC25A39 protein accumulates upon GSH depletion, which strongly indicates a potential feedback mechanism to control the availability of mitochondrial GSH (6). In line with this observation, cells treated with buthionine sulfoximine (BSO; a GSH synSCIENCE science.org
DMSO was used as the control. b-tubulin was used as a loading control. (Bottom) Quantification of FLAG band signal intensity from the immunoblots above. Halflife (t1/2) was calculated by the nonlinear fitting of FLAG band signal intensity versus time to one phase decay exponential model. (D) Schematic showing the localization and the catalytic reaction of engineered MitoCHAC1 protein. (E) Immunofluorescence images of MitoCHAC1 (HA, green), ATP5A1 (red), and DAPI (blue) in HEK293T cells. (F) Whole-cell and mitochondrial abundance of GSH (normalized to NAD+ abundance) in HEK293T cells expressing empty vector or MitoCHAC1 (in the presence of exogenous GSH). Data are mean ± SD representing three biologically independent samples. P values were calculated from WelchÕs t test. (G) Immunoblots showing the amounts of the indicated proteins in HEK293T cells that express an empty vector or MitoCHAC1 (in the presence of exogenous GSH). SLC25A12 and b-tubulin were used as loading controls. (H) (Top) The schematic of the cell-free assay that uses immunopurified mitochondria (mito IP) from HEK293T cells to analyze SLC25A39 stability. (Bottom) Immunoblots of the indicated proteins from purified mitochondria after treating them with GSH (20 mM) for the indicated times. SLC25A12 and SLC25A11 were used as loading controls.
thesis inhibitor) alone or in combination with erastin (an antagonist of uptake of the GSH precursor cystine) increased the abundance of SLC25A39 commensurate with the intensity of GSH depletion (fig. S1, A and B). However, how GSH levels are sensed and maintained through SLC25A39 in mitochondria is unknown.
SLC25A39 is a short–half-life protein regulated by mitochondrial GSH availability
To understand how SLC25A39 is regulated in mitochondria, we compared global mRNA and protein abundance using the OpenCell database (8). This revealed a small subset of posttranscriptionally regulated proteins with low 17 NOVEMBER 2023 ¥ VOL 382 ISSUE 6672
821
RES EARCH | R E S E A R C H A R T I C L E S
A
IMS
SLC25A1 SLC25A3 SLC25A5 SLC25A6 SLC25A10 SLC25A11 SLC25A39
B
Matrix
C
Overlay AlphaFold2 Structural Predictions
SLC25A39
3xFLAGSLC25A39 ( 42-106)
3xFLAGSLC25A39
cDNA: 135°
aa(42-106)
-50
B/E:
-
+ HEK293T
IMS
-35
-Tubulin
-50
+ B/E:
-
+
+ HEK293T
-
+
E
1.0 0.5
G
CT sg
kDa -40
2 2_
sg
G AF
sg
AF
L G3
kDa -40
FLAG
-100
AFG3L2
-70
-70
-
+
I
HEK293T: AFG3L2(E408Q)-HA FLAG
kDa
-50
FLAG
-70
Ve ct or 25 AG A3 + 3xF -S 42- 9 SL L L 1 C AG C2 06 25 5 A3 SL A1 9( C2 1 42 5A -1 1 06 1 )
cDNA:
-50
FLAG
-40
HA
-70
LC
B/E: -
-S
AG
Ve c
FL
cDNA:
K
L
-
+
cDNA:
3
FLAG
A 25 LC -S 8S) G 8 A / FL 78 3x (C
S GLA xF
Input
-Tubulin
kDa
FLAG
-40
-Tubulin
-50
B/E:
-
+
-
HA
M
HEK293T SLC25A39_KO 3xHA-mCherry-OMP25 cells
-40
cDNA: Vector SLC25A39 SLC25A39 (C78/88S) -/+ B/E treatment
-70 -50
Quantify GSH uptake FLAG
-40
HA
-70
Isolated mitochondria
IP: FLAG
+
C88
C94
+
HEK293T: AFG3L2(E408Q)-HA
39
39
5A
2 LC
100 •
- - + 3xFLAGSLC25A39
kDa
HEK293T
100
90 •
80 •
C74 C78
GSHee: -
3x
FL
-Tubulin
IP: FLAG
HA
80
SLC25A39, human SLC25A39(Ancestral), primates SLC25A39(Ancestral), placental SLC25A39(Ancestral), amniotes SLC25A39(Ancestral)
-70
-40
IP: FLAG
3x
Amino Acid 70 •
HA
-70
-Tubulin
40 60 CHX/min
-40
FLAG Input
HA
Input
20
J
HEK293T: AFG3L2(E408Q)-HA
kDa -40
0
0 30 60 90 0 30 60 90 0 30 60 90
CHX/min:
B/E: - + - + - + cDNA: Vector 3xFLAG- 3xFLAG-
GSH-(glycine-13C2, 15N)
P=0.019
3
2 P=0.064 P=0.169 1
0 B/E: cDNA:
+
-
SL C
SLC25A39 SLC25A39 (C78/88S)
Mitochondrial GSH uptake 4
+
-
+
SL (C C2 78 5A /8 39 8S )
+
t1/2 = 23.03 min 0.0
A3 9
-
-50
r
+
r
-
to
B/E:
H
-Tubulin
-50
t1/2 = 169.7 min 0.5
25
-Tubulin
sgCTRL sgAFG3L2_1 sgAFG3L2_2 t1/2 = 575.2 min
1.0
to
-100
AFG3L2
-1.5 -1.0 -0.5 0.0 0.5 Guide score (SLC25A39-lo fraction)
1 2_ 3L
RL
G AF
FLAG
-0.5
Red: AFG3L2 guides
HEK293T-3xFLAG-SLC25A39
2 2_ 3L
sg
Compare sgRNAs with NGS
Ve c
G AF
Fixation and FLAG staining
GSH(m+3) abundance normalized by NAD+
sg
0.0
FLAG-SLC25A39 signal
Transduced cells
SLC25A39 protein levels, relative to initial
RL
CT
sg
1.5
FACS sorting
HEK293T-3xFLAG-SLC25A39 1 2_ 3L
2.0
sgRNA pool against Mitochondrial Proteases
HEK293T cells 3xFLAG-SLC25A39
F
Guide scroe (SLC25A39-hi fraction)
D AFG3L2 IMMP1L MPRP1 PYCARD AGTPBP1 IMMP2L NLN RHBDD1 ATG4D INPP5E PARK7 SPG7 CASP4 KLK6 PITRM1 UQCRC2 LACTB CLPP PMPCA USP15 METAP1D PMPCB USP30 CLPX MIPEP CPS1 PRSS15 XPNPEP3 HTRA2 MMP2 PSARL YME1L1 Matrix Inner Membrane IMS Others
kDa -40
FLAG
-Tubulin
Matrix
cDNA:
kDa -40
FLAG
3xFLAGSLC25A11 +(aa42-106)
3xFLAGSLC25A11
3xFLAGSLC25A39
HEK293T SLC25A39_KO
822
17 NOVEMBER 2023 ¥ VOL 382 ISSUE 6672
science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
Fig. 2. AFG3L2 binds and degrades SLC25A39 through a matrix loop domain in a GSH-dependent manner. (A) Schematic showing the alignment of AlphaFold2-predicted structural models of the indicated proteins. SLC25A39 is highlighted in pink, with aa42–106 highlighted in green. (B) Immunoblots of the indicated proteins in HEK293T cells expressing 3xFLAG-SLC25A39 or 3xFLAGSLC25A39 without the matrix-facing loop (Daa42–106) after 24-hour treatment with BSO (1 mM) and erastin (5 mM) or DMSO as the control. (C) Immunoblots of the indicated proteins in HEK293T cells expressing 3xFLAG-SLC25A39, 3xFLAGSLC25A11, or a chimeric protein in which aa42–106 of SLC25A39 is spliced into SLC25A11 after 24-hour treatment with BSO (1 mM) and erastin (5 mM) or DMSO as the control. (D) Schematic showing the library design of the mitochondrial peptidase sgRNA library and the workflow of the FACS-based CRISPR screen for 3xFLAG-SLC25A39 stability. (E) Scatter plot showing the enrichment of sgRNAs targeting mitochondrial proteases in the SLC25A39-lo cell fraction (x axis) and SLC25A39-hi cell fraction (y axis). Red dots represent sgRNAs targeting AFG3L2. (F) Immunoblots of the indicated proteins in HEK293T cells expressing 3xFLAG-SLC25A39 and sgRNAs targeting control or AFG3L2 after 24-hour treatment with BSO (1 mM) and erastin (5 mM) or DMSO as the control. (G) (Left) Immunoblots of the indicated proteins in HEK293T cells expressing 3xFLAG-SLC25A39 and sgRNAs targeting AFG3L2 or control upon treatment with cycloheximide (CHX, 50 mg/ml) for the indicated times. b-tubulin was used as a loading control. (Right) Quantification of FLAG bands signal intensity from the immunoblots. Half-life (t1/2) was calculated by the nonlinear fitting of FLAG band signal intensity versus time to one phase decay exponential model. (H) Immunoblots of the indicated proteins from whole-cell lysates or
protein-to-mRNA ratios, such as Hypoxia Inducible Factor 1 Subunit Alpha (HIF1A), DAP3 binding cell death enhancer 1 (DELE1), Activating Transcription Factor 4 (ATF4), ornithine decarboxylase antizyme (OAZ1), and F-box and leucine-rich repeat protein 5 (FBXL5). These proteins are rapidly degraded or translationally repressed under basal conditions but accumulate upon diverse environmental stimuli, which enables their dynamic response to cellular stress. Among these proteins, SLC25A39 was the only mitochondrial transporter, indicating that a posttranscriptional mechanism may underlie its regulation (Fig. 1A). To determine the mode of SLC25A39 regulation in response to GSH depletion, we generated a reporter construct in which a 3xFLAG-tagged SLC25A39 cDNA was cotranslated with an internal control (RFP), separated by a 2A selfcleaving peptide from porcine teschovirus-1 (P2A). GSH depletion in human embryonic kidney HEK293T cells expressing this reporter strongly induced the accumulation of the 3xFLAGSLC25A39 protein relative to the abundance of RFP (Fig. 1B). This effect was independent of oxidative stress, as only supplementation of GSH, but not other antioxidants, suppressed SLC25A39 accumulation (Fig. 1B and fig. S1C). Because these observations point to a proteinlevel control, we measured the half-life of SLC25A39 in cycloheximide chase assays. SLC25A39 protein was unstable, with an estimated half-life of 15 min (Fig. 1C). GSH depletion substantially extended the half-life of SLC25A39 (>300 min), and supplementing cells with GSH restored this rapid degradation (Fig. 1C). These data indiSCIENCE science.org
FLAG-immunoprecipitation from HEK293T cells stably expressing cDNAs for vector, 3xFLAG-SLC25A39, 3xFLAG-SLC25A39 lacking matrix-facing loop (Daa42–106), 3xFLAG-SLC25A11, or a chimeric protein in which aa42–106 of SLC25A39 is spliced into SLC25A11 and transiently transfected with AFG3L2 (E408Q)-HA cDNA. (I) Immunoblots of the indicated proteins from whole-cell lysates or FLAG immunoprecipitation from HEK293T cells stably expressing cDNA for empty vector or 3xFLAG-SLC25A39 and transiently transfected with AFG3L2(E408Q)-HA cDNA. Cells were treated for 24 hours with BSO (1 mM) and erastin (5 mM) or DMSO as the control. Indicated cells were then treated for 8 hours with GSHee (10 mM). (J) Multiple sequence alignment between SLC25A39 and the inferred ancestral sequence of SLC25A39 reconstructed from amino acid sequences of SLC25A39 homologs in the indicated taxa. Four conserved cysteines in the matrixfacing loop are highlighted. (K) Immunoblots for the indicated proteins in HEK293T cells that express 3xFLAG-SLC25A39 or 3xFLAG-SLC25A39(C78/88S). Cells were treated for 24 hours with BSO (1 mM) and erastin (5 mM) or DMSO as the control. (L) Immunoblot of the indicated proteins from whole-cell lysates or FLAG immunoprecipitation from HEK293T cells stably expressing cDNAs for vector, 3xFLAG-SLC25A39, or 3xFLAG-SLC25A39(C78/88S) and transiently transfected with AFG3L2(E408Q)-HA cDNA. Cells were treated for 24 hours with BSO (1 mM) and erastin (5 mM) or DMSO as the control. (M) (Left) Schematic showing the GSH uptake assay that uses immunopurified mitochondria from HEK293T-SLC25A39_KO cells expressing cDNAs for empty vector, SLC25A39, or SLC25A39(C78/88S). (Right) Abundance of GSH-(glycine-13C2,15N) taken up by isolated mitochondria. Data are mean ± SD representing three biologically independent samples. P values were calculated from WelchÕs multiple t test with the Holm-Šídák method.
cate that SLC25A39 has a short half-life and becomes stabilized when GSH is depleted. Given that the mitochondrial matrix harbors a distinct GSH pool (9), we sought to determine whether SLC25A39 stability is regulated specifically by changes in the availability of local mitochondrial GSH, thus forming a feedback loop in which the substrate controls its own uptake. To test this possibility, we engineered the human GSH-specific gamma-glutamylcyclotransferase (CHAC1), a cytosolic enzyme that catalyzes the conversion of GSH into 5oxoproline and L-cysteinylglycine (10, 11), to be expressed in mitochondria (hereby referred to as MitoCHAC1) (Fig. 1, D and E, and fig. S1D). This engineered enzyme allowed us to specifically deplete mitochondrial GSH without significantly altering whole-cell GSH levels (Fig. 1F and fig. S1, E to G). Confirming the robust enzymatic activity of MitoCHAC1, we observed 5-oxoproline accumulation in the mitochondria (fig. S1H). Depletion of mitochondrial GSH with MitoCHAC1 expression was sufficient to induce SLC25A39 accumulation (Fig. 1G and fig. S1I). Expression of MitoCHAC1 did not affect the abundance of SLC25A39 mRNA and only slightly decreased cell proliferation (fig. S1, J and K). Moreover, expression of MitoCHAC1 did not impact the abundance of the master regulator of the antioxidant response, nuclear factor erythroid 2–related factor 2 (NRF2), or that of its downstream transcriptional targets (12), indicating that NRF2 activity is not controlled by mitochondrial GSH availability (fig. S1L). To formally test whether SLC25A39 stability can be
regulated independently of any signaling input from the cytosol, we developed a cell-free system with immunopurified mitochondria (Fig. 1H). Although isolated mitochondria from GSH-depleted cells displayed highly stable SLC25A39, direct supplementation with exogenous glutathione led to rapid degradation of SLC25A39, indicating that signaling input from cytosol or nucleus is dispensable for such regulation. Thus, a compartmentalized feedback mechanism appears to enable mitochondria to rapidly respond to the changes in the concentration of GSH in the matrix by tuning SLC25A39 stability. AFG3L2 degrades SLC25A39 in a GSH-dependent manner
SLC25A39, unlike other SLC25 family members, has an exceptionally low protein copy number to mRNA-TPM ratio and is distinctly sensitive to GSH availability. We therefore considered that a particular structural feature on SLC25A39 might allow this regulation. We aligned AlphaFold (13)–predicted structures of SLC25A39 with those of highly expressed SLC25 family members as well as that of SLC25A40, a paralog of SLC25A39 that is not sensitive to GSH availability (Fig. 2A and fig. S2, A and B). This analysis revealed a protruding loop on the matrix side of the SLC25A39 protein that corresponded to amino acids 42 to 106 (aa42–106) (Fig. 2A and fig. S2A). Although this domain is not critical for GSH transport activity (fig. S2C), its deletion completely abrogated GSH sensitivity and extended protein half-life (>250 min), thereby uncoupling the GSH-mediated regulation 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
823
RES EARCH | R E S E A R C H A R T I C L E S
of SLC25A39 from its transport function (Fig. 2B and fig. S2D, lanes 1 to 4). Splicing this fragment into a distant SLC25 family member (SLC25A11) that does not respond to perturbations in GSH levels rendered it GSH responsive (Fig. 2C). The differential responses of SLC25A39 and its close paralog, SLC25A40, to GSH availability raised the possibility that the matrix loop may be the critical attribute for the diversification of these two genes. Indeed, a significant portion of the SLC25A39 loop do-
E
C
Mitochondrial sgRNA library
Iron
B/E
SLC25A28 SLC25A37
FLAG-SLC25A39 signal
1.5
[2Fe-2S] proteins ABCB7
Compare sgRNAs with NGS
cytosolic Fe-S proteins ABCB7
0 FDX1L HSCB SLC25A28 NFS1 GLRX5 ISCU
HEK293T-3xFLAG-SLC25A39 _1 _2 CB CB RL CT HS HS g g g s s s FLAG
-1
-2 -1 1 0 -3 -2 3 2 log2 FC: shNFS1 vs shGFP (w. B/E) kDa -40
HSCB
-25
-Tubulin
-50
[2Fe-2S] cluster -containing proteins
-Log10 P value
B/E: [2Fe−2S] cluster assembly Protein insertion into mitochondrial inner membrane
No. of Genes 3 6 9 12 15
Iron−sulfur cluster assembly Metallo−sulfur cluster assembly Inner mitochondrial membrane organization Protein targeting to mitochondrion Establishment of protein localization to organelle
-
+
-
+
-
F
+
FLAG
kDa -40
Input
HA
-70
HSCB
-25
-Tubulin
-50
FLAG
sg C AB TR C L sg B7 AB _1 C B7 _2
IP: FLAG
-40
HA
-70
sgCTRL: sgHSCB: B/E: cDNA:
•
0
10 20 30 Fold Enrichment
Fig. 3. CRISPR screen identifies [2Fe-2S] cluster assembly as essential for the regulation of SLC25A39 stability. (A) Schematic of the CRISPR screen workflow with mitochondrial sgRNA library for SLC25A39 stability under GSH depletion. (B) (Top) Dot plot showing the distribution of differential CRISPR gene score calculated as (median guide enrichment in the SLC25A39-hi fraction) – (median guide enrichment in the SLC25A39-lo fraction). Pink dots indicate genes in the [2Fe-2S] cluster assembly pathway. The green dot indicates the putative mitochondrial [2Fe-2S] cluster exporter ABCB7. (Bottom) Gene Ontology enrichment analysis of genes with a differential CRISPR gene score lower than −1. (C) Schematic of the mitochondrial Fe-S cluster assembly pathways that highlights the genes that scored (pink and green) in the CRISPR screen. (D) Immunoblots of the indicated proteins in HEK293T cells that express 3xFLAGSLC25A39. (Top) Cells were transduced with lentivirus expressing Cas9 and control sgRNA or sgRNAs that target iron-sulfur cluster assembly factor HSCB. Cells were treated for 24 hours with BSO (1 mM) and erastin (5 mM) or DMSO as the control. 17 NOVEMBER 2023 ¥ VOL 382 ISSUE 6672
• • •
Ve c
Intracellular transport
-50
-Tubulin
sg
Protein localization to organelle
kDa -40
-100
−log10(FDR)
Protein targeting
HEK293T: AFG3L2(E408Q)-HA FLAG
HEK293T-3xFLAG-SLC25A39
IRP2 2.2 2.3 2.4 2.5 2.6 2.7
0 7 shNFS1 vs shGFP
• •
SL 3xF C LA 25 G A3 9
1500
to r
500 1000 Gene rank
sh G AB FP C sh B7 AB _1 C B7 _2
0
824
0 -0.5
0
-4
1
0.5
7
D
-2
SLC25A39
-1.5
sh
Differential gene score (SLC25A39-hi - SLC25A39-lo)
2
mitochondrial proteomics
2
NFS1 NFU1 ISCU ISCA1 Fe-S Assembly HSCB GLRX5
Sort SLC25A39-hi/ SLC25A39-lo cells
-/+ B/E treatment
shNFS1
FDX1L
FACS sorting
B
shGFP
[4Fe-4S] proteins
B/E vs DMSO
HEK293T cells 3xFLAG-SLC25A39
porters (fig. S2H). Thus, the conserved matrixfacing loop domain is necessary and sufficient for the GSH-mediated regulation of SLC25A39. We next sought to determine the proteolytic machinery involved in SLC25A39 degradation. Inhibition of proteasomal or lysosomal function did not impact SLC25A39 stability (fig. S3A). Given these results and the mitochondrialocalized regulation of SLC25A39 (Fig. 1H), we focused on proteolysis systems in the mitochondria. We designed a fluorescence-activated
log2 FC: B/E vs DMSO
A
main responsible for GSH sensitivity is evolutionarily new, with no homologs outside vertebrates (data S1). Using a maximum likelihood method, we reconstructed the ancestral sequence of SLC25A39 (fig. S2, E to G). When spliced into SLC25A11, the loop domain in the ancestral SLC25A39 sequence conferred GSH responsiveness, indicating that this feature arose early in the evolution of SLC25A39 and might be the key driving force behind the diversification of mitochondrial GSH trans-
(Bottom) Cells were transduced with lentivirus expressing control sgRNA or sgRNAs targeting ABCB7 (left), or shRNAs targeting GFP or ABCB7 (right). (E) Scatter plot showing log2 fold change and −log10(P value) of proteomics analysis from immunopurified mitochondria of the indicated samples. The x axis represents log2 protein fold change (FC) in isolated mitochondria from HEK293T cells that express shRNA targeting GFP or cysteine desulfurase NFS1 after 24-hour treatment with BSO (1 mM) and erastin (5 mM). The y axis represents log2 fold change (FC) in mitochondrial protein abundance after treating cells with BSO (1 mM) and Erastin (5 mM) versus DMSO as the control. The color grid indicates −log10(P values) and green circles represent [2Fe-2S] cluster–containing proteins. (F) Immunoblots of the indicated proteins from the whole-cell lysates or FLAG-immunoprecipitation from HEK293T cells stably expressing cDNAs for vector or 3xFLAG-SLC25A39, are infected with lentivirus-expressing control sgRNA or sgRNAs for HSCB, and transiently transfected with AFG3L2(E408Q)-HA cDNA. Cells were treated for 24 hours with BSO (1 mM) and erastin (5 mM) or DMSO as the control. science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
D
55
Fe
HEK293T cells 3xFLAG-SLC25A39
Chemical reconstitution of Fe-S cluster
Fe-S cluster Liquid scintillation assay Fe
-70
-Tubulin
-50
FLAG
-40
Fe CPM, FLAG-IP
-40
AFG3L2
IP: FLAG
400 500 600 700 Wavelength(nm)
0.0 -0.5 -1.0
800
-4
-2
0 2 Velocity (mm/s)
4
Reactive cysteine labeling with IA-DTB
P < 0.0001
• • kDa
FLAG Input
0.00% 99.46% 0.00% 0.23% 0.32%
0.0 300
O
I
DTB linker NH
1000
DMSO
P = 0.0006
B/E
500
55
•
[1Fe-0S]2+ [2Fe-2S]2+ [4Fe-4S]2+ Iron Sulfide Peptide
1500
• •
Contribution
E
to r 3 SL xFL C AG 25 A3 9 •
0.5
Mössbauer spectrum
Species
KKCLLYCNGVLEPLYLCPNGARCAKK
Ve c
sgCTRL: sgAFG3L2: B/E:
1.0
aa73-95 55
HEK293T cDNA:
Absorbance
Immunoprecipitation
B
UV-Vis spectrum
1.5
Absorption (%)
A
0 • sgCTRL: sgAFG3L2: B/E: cDNA:Vector
•
HEK293T AFG3L2-KO cells 3xFLAG-SLC25A39
Immunoprecipitation
B/E + DFO
Quantify cysteine reactivity via TMT-proteomics
• • • • 3xFLAGSLC25A39
P = 0.9923 P = 0.4804
0.5
B/ E D FO
0.0
+
B/ E
+
D
E FO
0.0
1.0
B/ E
0.5
1.5
D M SO
1.0
Relative levels of cysteine labeling
P = 0.0647
2.0
B/
FL
0.0
1.5
P = 0.2810
Cys 94
Cys 88
Cys 202
Within GSH-responsive loop
Outside
3x
3x
FL
AG -S L
C
AG -S L
25
3x
FL Ve A3 AG 9( -SL cto C 74 C2 r 3x /78 5A3 FL /8 9 AG 8/9 -S 4S L2 ) 5A 11
cDNA:
0.5
2.0
D M SO
C 3x 25 FL A3 A 9( G-S V C e 3x 74/ LC2 cto FL 78 5A r AG /88 39 -S /94 L2 S 5A ) 11
0
P = 0.5148 Relative levels of cysteine labeling
-35
E FO
FLAG cDNA:
1.0
D
IP: FLAG
1000
+
-40
2000
1.5 P = 0.0002
E
-50
P < 0.0001
SO
-Tubulin
3000
B/
-35
P < 0.0001
P = 0.4224
B/
-40
55
FLAG Input
Fe CPM, FLAG-IP
HEK293FS kDa
P = 0.0019 2.0
M
P = 0.7720
F
D
P = 0.7964
Relative levels of cysteine labeling
C
Fig. 4. A GSH-sensitive iron-sulfur cluster associates with SLC25A39 and mediates its regulation. (A) Schematic showing the method of tracking iron-sulfur cluster bound to SLC25A39 by means of 55Fe tracing. (B) (Left) Immunoblots of the indicated proteins from whole-cell lysates or FLAG immunoprecipitation from HEK293T cells stably expressing cDNAs for empty vector or 3xFLAG-SLC25A39 and infected with lentivirus-expressing control sgRNA or sgRNA targeting AFG3L2. Cells were labeled with 55FeCl3 in the culture media and treated for 24 hours with BSO (1 mM) and erastin (5 mM) or DMSO as the control. (Right) The amount of 55Fe bound to FLAG immunoprecipitant, from the identical cells as those in the immunoblot, quantified by liquid scintillation assay. (C) (Left) Immunoblots of the indicated proteins from whole-cell lysates or FLAG immunoprecipitation from HEK293FS cells stably expressing cDNAs for empty vector, 3xFLAG-SLC25A39, 3xFLAG-SLC25A39(C74/78/88/94S), or 3xFLAG-SLC25A11. Cells were labeled with 55FeCl3 in the culture media and treated for 24 hours with BSO (1 mM) and erastin (5 mM). (Right) The amount of 55 Fe bound to FLAG immunoprecipitant, from the identical cells as those used for the immunoblot, quantified by liquid scintillation assay. (D) (Left) Schematic SCIENCE science.org
showing the location and peptide sequence of SLC25A39(aa73–95) used to reconstitute peptide-[2Fe-2S] cluster complex in vitro. (Right) UV-visible spectrum (with the calculated contribution of the indicated species) and Mössbauer spectrum of the reconstituted peptide-[2Fe-2S] complex. The Mössbauer spectrum was least-squares fit to extract hyperfine parameters: isomer shift (d), quadrupole splitting (D), full width at half-maximum (FWHM), and intensities (I). The red and green curves of the Mössbauer spectrum plot represent two Fe3+ doublets with slightly different shifts and quadrupole splitting fitted to the spectral data. Red: d = 0.23 mm/s, D = 0.51 mm/s, and I = 70%. Green: d = 0.30 mm/s, D = 0.90 mm/s, and I = 30%. (E) Schematic for cysteine reactivity profiling for SLC25A39 with an iodoacetamide-desthiobiotin (IA-DTB) probe. (F) Reactivity of SLC25A39-Cys94, Cys88, and Cys202 after a 24-hour treatment with indicated reagents, quantified by mass spectrometry as the intensity ratio between IA-DTB–labeled versus unlabeled peptide containing SLC25A39-Cys94 and normalized to DMSO-treated samples. [(B), (C), and (F)] Data are mean ± SD and represent three biologically independent samples. P values were calculated from one-way analysis of variance. 17 NOVEMBER 2023 ¥ VOL 382 ISSUE 6672
825
RES EARCH | R E S E A R C H A R T I C L E S
cell sorting (FACS)–based CRISPR screen for 3xFLAG-SLC25A39 stability with a single guide RNA (sgRNA) library targeting all annotated mitochondrial proteins with peptidase activity (34 genes; 7 sgRNAs per gene) (Fig. 2D and fig. S3, B and C). Transduced cells were stained with a FLAG antibody, and sgRNA sequences from cells with the highest (SLC25A39-hi) and the lowest (SLC25A39-lo) FLAG signal were quantified by next-generation sequencing. Among all candidates, AFG3L2, a mitochondrially localized protease, was the only target whose loss significantly altered the abundance of SLC25A39 (q value = 0.009) (Fig. 2E, fig. S3D, and data S2). Consistent with the screen results, the loss of AFG3L2 completely abolished its GSH-mediated regulation (Fig. 2F and fig. S3E) and stabilized SLC25A39 even at normal GSH concentrations (Fig. 2G). SPG7, the paralog of AFG3L2, was dispensable for this process (fig. S3, F to H). Furthermore, immunoprecipitation experiments with an ATP triphosphatase– mutant AFG3L2(E408Q), which allows the detection of transient protease-target interactions (14), revealed that SLC25A39 associated with AFG3L2 through its loop domain (aa42–106) (Fig. 2H). When this loop was removed from SLC25A39, the loss of AFG3L2 did not further stabilize the protein (fig. S2D). This interaction was highly sensitive to changes in GSH availability and was abrogated by GSH depletion (Fig. 2I). Unbiased proteomic profiling further confirmed that SLC25A39, but not other mitochondrial membrane proteins, was specifically targeted for degradation by AFG3L2 in a GSHdependent manner (fig. S4, A to F, and data S3). GSH availability therefore determines the turnover of SLC25A39 protein by enabling the recruitment of AFG3L2 through the matrix-facing loop domain of SLC25A39. CRISPR screen identifies [2Fe-2S] cluster assembly as essential for SLC25A39 stability
To further dissect the function of the loop domain in the GSH-mediated regulation of SLC25A39, we used two orthogonal approaches. We used a reporter assay for mitochondrial protein stability (fig. S5A) to identify a short fragment on the GSH-responsive loop of SLC25A39 (aa72–86) necessary for its recruitment to AFG3L2 and proteolysis (fig. S5, B to E). Additionally, a detailed conservation analysis revealed four highly conserved cysteines—Cys74, Cys78, Cys88, and Cys94—within the vicinity of this fragment (Fig. 2J and fig. S5F). Introducing individual cysteine-to-serine mutations, particularly on Cys88 and Cys94, partially abrogated SLC25A39 stabilization upon GSH depletion (Fig. 2J and fig. S6A). More notably, mutating two cysteines in the loop domain was sufficient to completely abolish the regulation (Fig. 2K and fig. S6, B and C). The absence of these conserved cysteines led to the constitutive association of SLC25A39 and AFG3L2, independent of GSH 826
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
abundance (Fig. 2L). To investigate the functional role of these cysteines in maintaining GSH homeostasis in the mitochondria, we performed uptake assays with isotope-labeled GSH [GSH-(glycine-13C2,15N)] in mitochondria isolated from control cells or cells in which GSH was depleted. Consistent with the strong feedback mechanism, in the presence of wildtype SLC25A39, GSH depletion led to a 3.5-fold compensatory increase in mitochondrial GSH uptake (Fig. 2M). By contrast, mutating cysteine residues (C78/ 88S) completely abolished this homeostatic mechanism (Fig. 2M and fig. S6D), confirming the essential role of conserved cysteines in the feedback regulation of SLC25A39-mediated GSH uptake. Cysteine residues on proteins have critical roles in redox sensing, posttranslational modifications, and cofactor binding (15). However, our initial interrogation showed that disulfide bond formation and relay or hypoxic response did not impact SLC25A39 regulation (fig. S7, A to E). To identify the precise mechanism by which cysteines in the loop domain enable GSH-mediated regulation of SLC25A39 stability, we performed a FACS-based CRISPR screen for SLC25A39 stability under GSH depletion in HEK293T cells expressing a 3xFLAG-tagged SLC25A39 cDNA (Fig. 3A). Given that regulation of SLC25A39 can occur independently of any cytosolic machinery (Fig. 1H), for these screens, we generated an sgRNA library that contained all annotated mitochondrial proteins (MITO-sgRNA) (16, 17). After immunostaining with a FLAG antibody, transduced cells with the highest (SLC25A39-hi) and lowest (SLC25A39-lo) FLAG signal intensity were isolated and their sgRNA abundances were quantified. Among the genes whose disruption led to a lower amount of SLC25A39 were many assembly factors for [2Fe-2S] clusters such as NFS1, ISCU, HSCB, and GLRX5 (Fig. 3, B and C, fig. S8A, and data S2). Additionally, one of the genes whose loss most strongly enhanced SLC25A39 stability was ABCB7, a putative transporter involved in the export of iron-sulfur clusters from mitochondria (18–20). A gene ontology analysis further confirmed [2Fe-2S] cluster assembly as the most enriched pathway for the scoring genes in the SLC25A39-lo fraction (FDR = 1.8 × 10−5) (Fig. 3B and fig. S8A). Consistent with the screen results, acute loss of Fe-S cluster assembly factors (ISCU, GLRX5, or HSCB) abolished SLC25A39 stabilization under GSH depletion (Fig. 3D and fig. S8, B to F), whereas ABCB7 loss constitutively stabilized SLC25A39, likely owing to the mitochondrial accumulation of [2Fe-2S] clusters (Fig. 3D and fig. S8G) (21). Unbiased proteomics experiments under GSH-depleted conditions also revealed SLC25A39 as the only mitochondrial transporter whose abundance decreased upon blocking synthesis of [2Fe-2S] clusters (Fig. 3E, fig. S8H, and data S4). The loss of [2Fe-2S] as-
sembly factor HSCB (22) led to the constitutive association of SLC25A39 with AFG3L2 (Fig. 3F). Altogether, these results suggest that [2Fe-2S] cluster synthesis is necessary for stabilizing SLC25A39 when cells are depleted of GSH. Iron-sulfur clusters are typically coordinated by cysteines and are essential for the stability of the holoprotein. Because iron-sulfur cluster assembly and conserved cysteines are indispensable for SLC25A39 stability, we considered the possibility that SLC25A39 might be associated with a [2Fe-2S] prosthetic group. To test the presence of potential Fe-containing cofactors associated with SLC25A39, we labeled HEK293T cells expressing 3xFLAG-tagged SLC25A39 cDNA with radioactive 55FeCl3 (Fig. 4A). Although we did not observe evidence of increased iron-sulfur cluster synthesis or accumulated mitochondrial iron (fig. S9, A to C), the amount of 55Fe immunoprecipitated with SLC25A39 was increased upon GSH depletion (Fig. 4B). This increase of bound iron appeared to precede the stabilization of SLC25A39 because similar results were observed in AFG3L2-knockout cells with a comparable amount of immunoprecipitated SLC25A39 protein (Fig. 4B). This factor is unlikely to exist as heme iron or [4Fe-4S] clusters as depleting cells of ferrochelatase (FECH), a critical enzyme of heme synthesis or [4Fe-4S] cluster-assembly factor NFU1 had no effect on SLC25A39 regulation (fig. S9, D and E). By contrast, the loss of [2Fe-2S] assembly factor HSCB almost completely blocked iron binding to SLC25A39 (fig. S9F). The iron-binding activity of SLC25A39 lies within its glutathione-sensing matrix loop. Deleting this loop abolished the iron-binding of SLC25A39, and splicing the loop domain of SLC25A39 to SLC25A11, which normally does not bind iron, enabled it to associate with iron (fig. S9G). Mutating the critical cysteines in the loop domain (Cys74/78/88/94) prevented the association of 55Fe to SLC25A39 upon glutathione depletion (Fig. 4C). Furthermore, with a synthetic peptide encompassing the GSHresponsive cysteines of SLC25A39 (aa73–95), we were able to reconstitute the peptide-[2Fe-2S] cluster complex in vitro that displayed characteristic spectral features of a [2Fe-2S] cluster (Fig. 4D) in which all four cysteines could engage in metal coordination (fig. S9H). We conclude that SLC25A39 associates with a GSH-sensitive ironsulfur cluster through its cysteine residues. As a small-molecule thiol, GSH could exchange with the cysteine ligands of a [2Fe-2S] cluster, resulting in partial or complete dissociation of the cluster from the holoprotein (21, 23, 24) (fig. S9I). In line with these previous observations, GSH supplementation induced a notable change in the spectral feature of the SLC25A39 (aa73–95)–[2Fe-2S] cluster complex (fig. S9, J and K), indicating the displacement of one or more cysteine residues in the loop domain. Additionally, supplementation of cell-permeable GSH to GSH-depleted cells science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
B
kDa
E
HEK293T: 3xFLAG-SLC25A39
-35
SLC25A28
-35
FLAG
-40
IRP2
-100
Vector
[Fe/GSH]:
• •
-50
-
B/E: DFO: FAC:
Ve ct or 25 SL A2 8 C 25 A3 7
cDNA:
+ -
+ + -
+ + +
C
-40 SLC25A28 -35 n.s -25
SLC25A28
NDUFS1
MRM1
-35
Mito. translation
-50 -40 SLC25A28 -35 n.s
SLC25A28 FLAG (MitoGSHf)
G
-70
-35 -25
MRPS23 MRPL11
-15
-100
•
-1.5 cDNA:
-50
•
Vector
SLC25A28 SLC25A28 SLC25A39
Vector
SLC25A28 SLC25A39
SLC25A28
SLC25A39
Off
O
Proteolysis
Fe2+
Matrix
ff
Proteolysis
[2Fe-2S]
[2Fe-2S] GSH import
Fe
GSH
GSH
GSH
Fig. 5. SLC25A39-mediated GSH import maintains iron/GSH balance in mitochondria. (A) Immunoblots of the indicated proteins from HEK293T cells overexpressing cDNAs of Mitoferrin 1 (SLC25A37), Mitoferrin 2 (SLC25A28), or an empty vector. (B) Immunoblot of the indicated proteins in HEK293T cells expressing 3xFLAG-SLC25A39 cDNA after 24-hours treatment with BSO (1 mM) and Erastin (5 mM); BSO, Erastin, and deferoxamine (50 mM); or BSO, Erastin, deferoxamine (50 mM), and Ferric Ammonium Citrate (FAC, 10 mg/ml). (C) Immunoblots of the indicated proteins from HEK293T cells overexpressing cDNAs of MitoCHAC1 or empty vector after 4 hours of treatment with 50-mM iron chelator deferoxamine (DFO) or control. (D) Immunoblots of the indicated proteins from HEK293T cells overexpressing cDNAs of SLC25A28, MitoCHAC1, or FLAG-tagged MitoGSHf, an engineered bacterial GSH synthase localized to the mitochondria. (E) (Top) Schematic of the experiment setup of TMT proteomics
significantly reduced the amount of iron bound to SLC25A39 (fig. S9L). To further support these findings, we used chemical proteomics to assess the reactivity of cysteines with an iodoacetamidedesthiobiotin probe, given that coordination with
SLC25A39
On
Cys Cys CysCys
GSH AFG3L2
GSH
SLC25A39
Matrix
SCIENCE science.org
-1.5 cDNA:
IMS
GSH AFG3L2
Fe2+ Fe
0
Upon depletion of Mitochondrial GSH
IMS GSH
O n
Iron-sulfur cluster-containing
•
Physiological levels of Mitochondrial GSH
SLC25A39
1.5
0
3xFLAG-SLC25A39 cDNA: + + + + SLC25A28 cDNA: HEK293T SLC25A39_KO cells
-25
• •
Mitochondrial Translation
1.5
-40
kDa
Protein Z-score
SLC25A39
−2 −1 0 1 2 z−score of differentially expressed proteins
-15
XPD
0 6 -Log10 p-Value
Fe-S proteins
• • • • • •
•
1.5 2.0 2.5 3.0 Enrichment Ratio
-70
• •
NDUFS7
•
Amide biosynthetic proc.
-25
SDHB
HEK293T
CHAC1 SLC25A28: cDNA: MitoGSHf: MitoCHAC1:
Mitochondrial ATP synthesis coupled electron transport
-15
NDUFB8
-50
hi:
Mitochondrial gene expression
COX II ETC components
-25
CHAC1
D
Mitochondrial translation
-40
FLAG
-100
IRP2
•
Vector SLC25A28 SLC25A39 + SLC25A28
Electron transport chain
-35
•
• •
Gene Count −logFDR
SL
SLC25A39
MitoCHAC1 cDNA: DFO:
•
•
TMT proteomics
F
• • • • kDa
hi:
SLC25A39+SLC25A28
Top enriched GO Terms in differentially expressed proteins 40 60 80 0 1 2 3
HEK293T [Fe/GSH], lo
SLC25A28
SLC25A28 n.s
-55
[Fe/GSH], lo
cDNA:
-35
SLC25A37
C
HEK293T SLC25A39_KO cells
kDa
3.5 0 -Log10 p-Value
HEK293T SLC25A39
Protein Z-score
A
Maintenance of mitochondrial Fe/GSH balance
for cells with different iron/GSH ratios. Three biologically independent samples per condition from HEK293T-SLC25A39_KO cells expressing the indicated cDNAs were used. (Middle) Gene ontology enrichment analysis of the most differentially expressed proteins across the three conditions that shows the top five most significantly enriched biological processes. The violin plots indicate the relative protein abundance (z-scores) of the differentially expressed proteins in the indicated biological processes. (Bottom) Dot plot representing protein levels (z-scores) of the mitochondrial translation machinery and iron-sulfur clusterÐ containing proteins. The darkness of the lines represents the statistical significance of the changes in protein abundance. (F) Immunoblots of the indicated proteins in HEK293T-SLC25A39_KO cells overexpressing empty vector, SLC25A39 cDNA, Mitoferrin 2 (SLC25A28) cDNA, or both. (G) Schematic for the model describing the autoregulatory control of mitochondrial iron/GSH balance by SLC25A39.
iron-sulfur clusters could limit cysteine reactivity (Fig. 4E) (25). Cys94, a conserved cysteine in the SLC25A39 matrix loop that was most reliably detected by mass spectrometry, displayed a strong decrease in reactivity upon GSH deple-
tion, which was restored by iron chelation (Fig. 4F and data S5). We observed a similar trend for Cys88; by contrast, Cys202, a cysteine residue that lies outside of the matrix-facing loop, showed no significant change in reactivity. These 17 NOVEMBER 2023 ¥ VOL 382 ISSUE 6672
827
RES EARCH | R E S E A R C H A R T I C L E S
observations support a model in which GSH depletion leads to the tight association of a [2Fe2S] cluster to SLC25A39 through cysteines (C74/ 78/88/94) in the loop domain, which in turn prevents the recruitment of protease AFG3L2, thus stabilizing SLC25A39.
function and impaired viability under GSH limitation (fig. S12, A to D). Thus, a feedback mechanism for SLC25A39 stability enables cells to maintain optimal mitochondrial functions by restoring iron/GSH balance.
SLC25A39-mediated GSH import maintains mitochondrial Fe-GSH balance
Metabolic homeostasis is maintained by negative feedback regulation at both the cellular and organismal levels. We provide evidence that organelles use similar principles to control their internal metabolite pools. In mitochondria, autoregulatory control of GSH availability occurs by coupling GSH sensing in the matrix to the degradation of its transporter, SLC25A39. This regulation pattern mirrors the homeostatic mechanisms for redox potential (28) or pH (29) in other organelles. Conceivably, other cellular compartments, the internal chemical environments of which differ from that of cytosol, may harbor similar mechanisms for sensing and regulating metabolites abundance (30). Concentrations of metabolites are often maintained within strict limits. Deviations from the optimal range, either through deficiency or excess, can damage cellular functions. We propose that feedback regulation of mitochondrial GSH may primarily serve to maintain an appropriate amount of GSH to accompany free iron. This is consistent with the chemical properties of GSH, which is predicted to be the major iron ligand in cells (26). Labile iron in mitochondria is essential for enzymatic activities and macromolecular complex assembly in many pathways. By contrast, excess iron can cause oxidative damage to proteins, DNA, and membrane lipids (31). This dual effect of free iron implies that its concentration in the mitochondria must be properly buffered. Feedback regulation of the availability of mitochondrial GSH is absent in lower organisms and appears to arise more recently in vertebrates. Given the essential role of SLC25A39 in erythropoiesis (6, 32), this feature may have evolved to cope with more complex iron metabolism in vertebrates. Understanding how such closed-loop regulation functions in other physiological or pathological processes may provide important insights into the systemic role of metabolic compartmentalization.
We considered why such a feedback mechanism might have evolved and how it might contribute to mitochondrial function. In addition to its antioxidant role, GSH is a major endogenous iron ligand in cells (26). In yeast, the absence or an excess of GSH can both lead to abnormal iron metabolism and the activation of cellular stress responses (27). Given the essential role of iron-sulfur clusters in SLC25A39 stability, this feedback mechanism might help maintain a proper balance between the availability of iron and GSH in mitochondria. Consistent with this idea, when we increased the mitochondrial iron/GSH ratio by overexpressing Mitoferrin 1 (SLC25A37) or Mitoferrin 2 (SLC25A28), the two major mitochondrial iron importers, SLC25A39 protein levels increased (Fig. 5A and fig. S10A). This regulation occurs through the GSH-responsive cysteines on SLC25A39 because mutating these cysteines in situ dampened the response of SLC25A39 to iron overload (fig. S10B). The increase in mitochondrial iron accompanied a compensatory increase in the abundance of GSH in the mitochondria (fig. S10C). By contrast, iron chelation, which decreases the mitochondrial iron/GSH ratio, abolished the stabilization of SLC25A39 upon GSH depletion (Fig. 5B and fig. S10D). We used multiple orthogonal approaches to perturb mitochondrial iron and GSH pools, and the abundance of SLC25A39 responded in accordance with the change in iron/GSH ratio, indicating that a feedback mechanism maintains iron/GSH balance (Fig. 5, C and D). We tested whether GSH limitation and iron overload imposed similar stresses on mitochondrial function by comparing mitochondrial proteomes from cells depleted of GSH depletion (6) or exposed to excess mitochondrial iron after the overexpression of Mitoferrin 2 (fig. S11A). These conditions led to the depletion of similar proteins, particularly those that function in the mitochondrial translation machinery, ETC components, and iron-sulfur proteins (fig. S11, B and C, and data S6). Decreases in the abundance of mitochondrial proteins induced by mitochondrial iron overload could be largely restored by boosting mitochondrial GSH uptake, supporting a critical buffering role of GSH in iron overload (Fig. 5, E and F; fig. S11, D and E; and data S7). SLC25A39 appeared to protect mitochondrial function at least partially through its adaptive response to iron/GSH balance because defects in the GSH-sensing function of SLC25A39 led to suboptimal mitochondrial 828
17 NOVEMBER 2023 ¥ VOL 382 ISSUE 6672
Discussion
RE FERENCES AND NOTES
1. A. Efeyan, W. C. Comb, D. M. Sabatini, Nature 517, 302–310 (2015). 2. M. S. Brown, J. L. Goldstein, Science 232, 34–47 (1986). 3. J. Mårtensson, J. C. Lai, A. Meister, Proc. Natl. Acad. Sci. U.S.A. 87, 7185–7189 (1990). 4. Z. Gu et al., Mol. Microbiol. 51, 149–158 (2004). 5. S. M. Beer et al., J. Biol. Chem. 279, 47939–47951 (2004). 6. Y. Wang et al., Nature 599, 136–140 (2021). 7. X. Shi et al., Nat. Commun. 13, 2483 (2022). 8. N. H. Cho et al., Science 375, eabi6983 (2022). 9. O. W. Griffith, A. Meister, Proc. Natl. Acad. Sci. U.S.A. 82, 4668–4672 (1985). 10. R. R. Crawford et al., J. Biol. Chem. 290, 15878–15891 (2015). 11. S. Tsunoda et al., eLife 3, e03421 (2014). 12. L. Torrente, G. M. DeNicola, Annu. Rev. Pharmacol. Toxicol. 62, 279–300 (2022). 13. J. Jumper et al., Nature 596, 583–589 (2021). 14. C. Puchades et al., Mol. Cell 75, 1073–1085.e6 (2019).
15. U. Topf et al., Nat. Commun. 9, 324 (2018). 16. S. E. Calvo, K. R. Clauser, V. K. Mootha, Nucleic Acids Res. 44 (D1), D1251–D1257 (2016). 17. A. C. Smith, A. J. Robinson, Nucleic Acids Res. 47 (D1), D1225–D1228 (2019). 18. V. Srinivasan, A. J. Pierik, R. Lill, Science 343, 1137–1140 (2014). 19. J. Y. Lee, J. G. Yang, D. Zhitnitsky, O. Lewinson, D. C. Rees, Science 343, 1133–1136 (2014). 20. P. Li et al., Nat. Commun. 13, 4339 (2022). 21. J. Li, J. A. Cowan, Chem. Commun. 51, 2253–2255 (2015). 22. N. Maio et al., Cell Metab. 19, 445–457 (2014). 23. L. Que Jr., M. A. Bobrik, J. A. Ibers, R. H. Holm, J. Am. Chem. Soc. 96, 4168–4178 (1974). 24. L. Que Jr., R. H. Holm, L. E. Mortenson, J. Am. Chem. Soc. 97, 463–464 (1975). 25. D. W. Bak, E. Weerapana, Nat. Chem. Biol. 356–366 (2023). 26. R. C. Hider, X. L. Kong, Biometals 24, 1179–1187 (2011). 27. C. Kumar et al., EMBO J. 30, 2044–2056 (2011). 28. C. S. Sevier et al., Cell 129, 333–344 (2007). 29. M. Hu et al., Cell 185, 2292–2308.e20 (2022). 30. L. Bar-Peled, N. Kory, Nat. Metab. 4, 1232–1244 (2022). 31. P. B. Walter et al., Proc. Natl. Acad. Sci. U.S.A. 99, 2264–2269 (2002). 32. R. Nilsson et al., Cell Metab. 10, 119–130 (2009). 33. E. F. Pettersen et al., Protein Sci. 30, 70–82 (2021). AC KNOWLED GME NTS
We thank all members of the Birsoy laboratory for helpful suggestions. Data were generated by the Proteomics Resource Center (RRID:SCR_017797), Flow Cytometry Resource Center, Drug Discovery Resource Center, and Genomics Resource Center at The Rockefeller University. We thank T. Carroll and all staff of the Bioinformatics Resource Center for their help in data analysis. We thank Y. Shen for the illustration of the summary figure 5G. Funding: National Cancer Institute F99CA284249 (Y.L.); Medical Scientist Training Program, National Institute of General Medical Sciences award T32GM007739 (F.S.Y. and R.A.W.); Damon Runyon Cancer Research Foundation DRG-2431-21 (G.U.); NIH/NIDDK F32 fellowship DK127836 (T.C.K.); Merck Postdoctoral Fellowship at The Rockefeller University (T.C.K.); The Shapiro-Silverberg Fund for the Advancement of Translational Research (T.C.K.); Pershing Square Sohn Foundation (R.K.H.); National Cancer Institute Cancer Center Support Grant P30-CA008748 (R.K.H.); Rockefeller University start-up funds (E.V.V. and N.R.); Robertson Foundation (E.V.V.); Simons Foundation 290358FY18 and 290358FY19 (S.S.M.); Natural Sciences and Engineering Research Council of Canada RGPIN-2020-04375 (S.S.M.); European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska–Curie grant agreement no. 813873 (S.S.M.); Mark Foundation Emerging Leader Award (K.B.); Searle Scholar (K.B.); Pew-Stewart Scholar (K.B.); National Cancer Institute, R01CA273233 (K.B.); Pershing Square Sohn Foundation (R.K.H.); NIH National Cancer Institute Cancer Center Support grant P30-CA008748 (R.K.H.); Rockefeller University start-up funds (E.V.V. and N.R.); Robertson Foundation (E.V.V.); Boehringer Ingelheim Fonds PhD fellowship (A.K.). Author contributions: Conceptualization: Y.L. and K.B.; Methodology: Y.L., S.L., A.T., F.S.Y., G.U., J.P., E.T., H.A., A.E.P., S.H., R.L.P., L.Z., E.V.V., S.S.M., and K.B.; Investigation: Y.L., S.L., A.T., F.S.Y., G.U., N.R., R.A.W., Y.W., J.P., E.T., H.A., A.E.P., S.H., T.C.K., and E.V.V.; Visualization: A.K., M.G., and B.A.; Funding acquisition: Y.L., F.S.Y., G.U., N.R., R.A.W., T.C.K., R.K.H., E.V.V., S.S.M., and K.B.; Supervision: K.B., S.S.M., E.V.V., R.K.H., L.Z., R.L.P., and H.M.; Writing – original draft: K.B. and Y.L.; Writing – review and editing: Y.L., F.S.Y., G.U., A.K., T.C.K., E.V.V., and K.B. Competing interests: K.B. is a scientific advisor to Nanocare Pharmaceuticals and Atavistik Bio. The other authors declare no competing interests. Data and materials availability: All data are available in the main text or the supplementary materials. License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/about/ science-licenses-journal-article-reuse SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adf4154 Materials and Methods Figs. S1 to S12 Data S1 to S7 References (34–49) MDAR Reproducibility Checklist Data S1 to S7 Submitted 30 October 2022; resubmitted 18 June 2023 Accepted 18 October 2023 10.1126/science.adf4154
science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
BIOMATERIALS
A strong quick-release biointerface in mussels mediated by serotonergic cilia-based adhesion Jenaes Sivasundarampillai1†, Lucia Youssef1†, Tobias Priemel1‡, Sydney Mikulin1, E. Deniz Eren1, Paul Zaslansky2, Franziska Jehle1,3, Matthew J. Harrington1* The mussel byssus stem provides a strong and compact mechanically mismatched biointerface between living tissue and a nonliving biopolymer. Yet, in a poorly understood process, mussels can simply jettison their entire byssus, rebuilding a new one in just hours. We characterized the structure and composition of the byssus biointerface using histology, confocal Raman mapping, phase contrastÐ enhanced microcomputed tomography, and advanced electron microscopy, revealing a sophisticated junction consisting of abiotic biopolymer sheets interdigitated between living extracellular matrix. The sheet surfaces are in intimate adhesive contact with billions of motile epithelial cilia that control biointerface strength and stem release through their collective movement, which is regulated neurochemically. We posit that this may involve a complex sensory pathway by which sessile mussels respond to environmental stresses to release and relocate.
P
roducing mechanically stable biointerfaces between living tissues and nonliving materials is a challenge with relevance for applications in tissue engineering (1, 2), wearable sensors (3), neural implants (4), and advanced cellular diagnostics (5). In addition to concerns about biocompatibility and functionality, mechanical factors are central in designing effective biointerfaces (4, 6). For example, modulus mismatch occurs when stiff-
ness and/or Poisson’s ratios of materials in contact are appreciably different, leading to localized stresses that can cause interfacial failure (7, 8). Designing effective interfaces for bionic implant materials is particularly tricky given the large disparity in mechanical properties between soft tissues [e.g., extracellular matrix (ECM), brain tissue] and typical implant materials (e.g., TiO2, semiconductors) (4). Additionally, implant exchange or device removal,
Fig. 1. Mussel byssus stem root microscale hierarchical structure. (A) Opened mussel with magnified area highlighting (B) the generator—the region of the foot that produces the stem and in which the stem root is embedded. (C) Image of cleanly released byssus highlighting the intact stem root—the region of the stem embedded in the generator. (D) Image of forcibly released byssus highlighting the damage to the stem root. White dashed lines in (C) and (D) indicate the approximate boundary between stem and stem root. (E) Section through a false-color 3D reconstruction of a PCE-mCT dataset acquired from SCIENCE science.org
when necessary, can also cause tissue damage (9). There is thus a need for biointerface designs that are mechanically stable, yet easily removed on demand. Nature provides valuable role models for bioinspired design of biointerfaces between mechanically mismatched tissues (e.g., squid beak, marine worm jaws). In these examples, mechanical mismatch is mitigated through compositional, structural, and mechanical gradients (7, 8). However, these solutions are not conducive to on-demand release of the interface. By contrast, we focus here on the biointerface between the mussel byssus stem root and the mussel foot (the byssus-producing organ), which provides a natural example of a strong, yet removable and rebuildable mechanical interface between a living tissue and a nonliving biopolymeric material (Fig. 1, A and B) (10–12). The byssus comprises an array of proteinaceous attachment threads used by mussels to 1 Department of Chemistry, McGill University, 801 Sherbrooke Street West, Montreal, Quebec H3A 0B8, Canada. 2Department for Operative, Preventive and Pediatric Dentistry, CharitéUniversitätsmedizin Berlin, Berlin 14197, Germany. 3Department of Biomaterials, Max Planck Institute of Colloids and Interfaces, Am Mühlenberg 1, 14476 Potsdam, Germany.
*Corresponding author. Email: [email protected] †These authors contributed equally to this work. ‡Present address: Department of Sustainable and Bio-inspired Materials, Max Planck Institute of Colloids and Interfaces, Am Mühlenberg 1, 14476 Potsdam, Germany.
the generator region, showing the wavy lamellar sheets of the stem root (movie S1). The space between lamellae is where the ECM is located. Inset gives approximate orientation of the stem root in the mCT sample. (F) Trichrome-stained histological section from the generator region showing interdigitation of the wavy lamellar sheets that comprise the stem root. Black dashed lines in (E) and (F) indicate the approximate boundary between stem and stem root. (G and H) Zoomed-in images from (F) showing structural details of (G) the lamellae-cilia interface and (H) the generator-muscle interface. 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
829
RES EARCH | R E S E A R C H A R T I C L E S
Fig. 2. Structural and compositional analysis of the stem generator biointerface. (A) Single image from FIB-SEM image stack, showing key features with false color. (B) Reconstructed features in 3D from a FIB-SEM image stack. (C) Zoomed-in and slightly rotated image of the leftmost lamella from (B), highlighting cilia interaction with lamella. (D) Overview STEM image from the generator tissue. (E to G) Higher-magnification STEM images from lamella, vesicles, and septa, respectively. (H) Trichrome-stained generator section showing region similar to that in the confocal Raman image in (I). (J) Averaged Raman spectra acquired from regions of matching color in (I).
anchor in seashore habitats against forces from crashing waves and predators (13). For Mytilus mussels, each byssal thread is glued at its distal end to a hard surface by means of an adhesive plaque and is attached at its proximal end to the stem, like branches on a tree trunk (Fig. 1C). At its base, the cylindrical external stem transitions into the more flattened and tapered internal stem root (14). The stem root anchors the entire byssus into the living tissue at the base of the foot in a region known as the generator, which is also the secretory tissue responsible for forming the stem (Fig. 1, A to C) (12). Given the forces exerted by crashing waves whose velocities can exceed 30 m/s (15), this necessitates a strong interface. However, Mytilus mussels are inexplicably able to jettison their entire byssus on demand and then fabricate a new one in hours (14, 16, 17). Although the cues that induce this behavior are still unknown (17), byssus release enables sessile mussels to regain mobility and crawl across substrates (14) and even to scale vertical surfaces (fig. S1). Such movements may allow mussels to relocate and escape unfavorable conditions (e.g., elevated temperature, predators, wave exposure, unsuitable substrates) (14, 18–21). Thus, although the interface between the stem biopolymer and the generator tissue must be very strong to 830
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
resist wave forces, its strength must also be dynamically tunable to enable quick release. Recent investigations of a distantly related mussel, Atrina Pinctada, revealed a chemical interaction at the stem root–tissue interface mediated by metal coordination and sugar binding (10, 11). However, little is understood about the byssus release mechanism of Mytilidae mussels and its relationship to stem root ultrastructure (12, 17). Dynamic stem interface mechanics
To investigate the mechanical integrity of the stem root biointerface, we forcibly removed the byssus using a tensile tester and recorded maximum force values of 6.4 ± 1.6 N (figs. S2 and S3). Consistent with previous reports (14), the tips of the majority of forcibly extracted stem roots appeared torn and frayed (Fig. 1D), indicating that part of the stem root broke off. However, mussels occasionally released the entire stem root, leaving a smooth and tapered appearance at the tips, similar to that of stem roots that are jettisoned by mussels spontaneously (Fig. 1C and fig. S1) (14). To investigate this clean-release phenomenon further, we manually applied a sustained subcritical force (i.e., below the breaking force) on the byssus, revealing that numerous mussels could be induced to release
the byssus cleanly (fig. S4 and movies S2 and S3), leaving intact stem roots. Although the conditions that favor byssus release in nature remain unknown (17), these observations indicate that the stem–tissue interface can sustain large forces before rupture; yet, under certain conditions the entire interface can be cleanly released under subcritical forces without obvious damage (14). Hierarchal structure and composition of the stemÐgenerator biointerface
To better understand the paradoxical high strength and quick release of the stem root biointerface, we analyzed the hierarchical structure of the generator region of the foot—the soft tissue in which the stem root is formed and anchored (Fig. 1, A and B). Phase contrast– enhanced microcomputed tomography (PCEmCT) images of the generator region were computationally reconstructed and segmented to generate a three-dimensional (3D) image of the stem–generator biointerface with micrometer resolution (Fig. 1E and movie S1). The reconstructed image reveals that within the generator region, the stem root comprises more than 40 individual sheets, previously named lamellae (12), each of which possesses a thickness of 2 to 3 mm and a characteristic wavy morphology (Fig. 1E). Although the spaces between the science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
Fig. 3. Histological and spectroscopic characterization of stem release. An opened mussel before (A) and after (B) removal of the byssus, highlighting the stem root. (C) Trichrome-stained section of the generator tissue with embedded stem root, highlighting complete lamella and surrounding cilia. (D) Trichromestained section of the generator tissue after induced stem root release, showing no signs of lamella between cilia. (E) Trichrome-stained section of the generator tissue after forced stem root release, showing remnants of lamella where the
lamellae observed in the PCE-mCT reconstruction are filled with the soft tissue of the generator, they become compressed together at the exit of the generator, forming the cylindrical external region of the stem (Fig. 1E). This highlights the continuity between the soft living generator tissue and stiff nonliving byssus, indicating that forces transmitted through the threads to the stem will converge and concentrate in the generator at the interface between the lamellae and the soft tissue. Histological cryosections cutting through the stem to the base of the generator reveal additional compositional and structural details (Fig. 1, F to H). Staining with Masson’s trichrome was used to localize collagenous (blue) and noncollagenous (red) protein components. The wavy morphology of the lamellar sheets interdigitated into the generator is evident in the histological sections, consistent with PCE-mCT and an earlier ultrastructural study (Fig. 1F) (12). At higher magnification (Fig. 1G), the lamellae possess a blue-staining core, suggesting a primarily collagenous nature of the byssus stem tissue. The area between two lamellae, named the septa, also stains positive (blue) for collagen with purple-staining nuclei peppered throughout, indicating its ECM-like nature. The lamellae and septa ECM are separated by a thin red-staining region with a feathery morphology (Fig. 1G), which is consistent with cilia, as previously SCIENCE science.org
stem root broke off. (F) Trichrome-stained section of generator tissue after clean stem root release, showing empty space between cilia where lamella was (indicated by black arrows). (G and H) Raman image and spectra, showing lack of lamella. (I) Sirius redÐstained section of cleanly removed stem lamellae, showing biphasic composition of collagenous and noncollagenous components. (J and K) Raman image and spectra, highlighting biphasic composition of lamella.
indicated by Tamarin (12). The collagenous septa ECM is connected directly to red-staining byssus retractor muscles at the base of the generator, which can apply a tensile force to the entire byssus (Fig. 1, F and H) (16), showing the continuity between the byssus and the musculature. To observe the interface between the generator septa and the stem root lamellae at higher resolution, we used focused ion beam–scanning electron microscopy (FIB-SEM), which enables 3D reconstruction of small tissue volumes with ~20-nm resolution. Figure 2A shows a single image from the FIB-SEM image stack, highlighting key features (i.e., septa, cilia, lamellae, and secretory vesicles) in false color, which were then reconstructed in 3D using the FIB-SEM image stack (Fig. 2, B and C; fig. S5; and movie S4). This highlights the direct and intimate interaction between the cilia and the lamellar material, with indentations apparent where the cilia contact the surface of the lamellae (fig. S5). Previously identified secretory vesicles containing protein precursors for building the stem (12) are seen lined up at the septa–lamellar interface (Fig. 2C and fig. S5), with the cilia spreading apart, likely to enable vesicle secretion. It has been proposed that with each new byssal thread produced, the stem extends further out from the foot, suggesting an extrusionlike mechanism by which the stem grows (12, 22).
The vesicles have a shape similar to that of the well-characterized secretory vesicles that contain liquid crystalline collagenous precursors that form the byssal threads (23). Scanning transmission electron microscopy (STEM) imaging of the generator tissue provides additional details of the internal structure of the lamellae and the cilia–lamella interface (Fig. 2, D to G). Although the lamellae appeared homogeneous with FIB-SEM, the higher resolution of STEM reveals a heterogeneous structure with a fibrous core and a more amorphous outer region that is in direct contact with cilia. The cilia can be assumed to be motile on the basis of the characteristic organization of microtubules into a so-called 9+2 axoneme (24), in which nine microtubule doublets surround two single microtubules in the center of the cilia. STEM images indicate that the cilia are tightly pressed into the outer layer of the lamellae, consistent with FIB-SEM imaging, suggesting that the lamella surface is relatively soft (Fig. 2E), whereas secretory vesicles can be seen pressed between cilia poised for secretion (Fig. 2F). The surrounding septa ECM exhibits randomly oriented thin fibrils with characteristic banding that we assume are collagenous on the basis of histological staining (Fig. 2G). Confocal Raman spectroscopic imaging within the generator reveals further compositional details of the biointerface between the septae and 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
831
RES EARCH | R E S E A R C H A R T I C L E S
Fig. 4. Role of cilia and neurotransmitters in stem release. (A) Maximum stem root pullout force measured with a tensile tester under various treatments (5-HT, serotonin; MET, metergoline; DA, dopamine). The asterisk (*) for the 0.2 mM 5-HT treatment indicates that in all cases, the byssus came out during sample loading on the tensile tester, indicating very low pullout forces. Data are mean ± SD with N ≥ 5 specimens for each treatment. Inset show example forceextension curves. (B) DIC light microscopy still images from cilia video, showing tracking of cilia beating in dissected sections of generator tissue. Movement
lamellae (Fig. 2, H to J). Raman spectra acquired from the stem and lamellae resemble spectra acquired from the proximal byssal thread (25), confirming a primarily collagenous composition, whereas spectra acquired from the septa ECM region are highly similar to spectra of type I collagen as seen by comparison with a mouse tail tendon control (fig. S6). The region between the septa ECM and lamellae, where the cilia are observed with electron microscopy, is dominated by spectra consistent with a-helical protein conformation based on the positions of the amide I and amide III bands, similar to previous Raman measurements of specialized cilia in the secretory ducts that form the byssus adhesive plaque (26). Stem quick release response
Given the intimate interaction between the stem root lamellae with the cilia of the septa ECM, we investigated the tissue-level effect of stem root release (Fig. 3). Byssus release was induced by 832
17 NOVEMBER 2023 ¥ VOL 382 ISSUE 6672
of four specific cilia is illustrated with colored circles. (C) Mean beating speed of cilia after consecutive treatments of the same region of tissue. Number (N) of cilia detected and tracked in each treatment is indicated on the plot. (D) Schematic showing how wave forces acting on the mussel are transmitted through the byssus into the stem and generator. (E) Schematic showing intimate interaction of lamellae with cilia on generator epithelial cells. (F and G) Schematic representations of the effects of (F) dopamine and (G) serotonin on biointerface strength mediated by cilia beating.
applying a sustained subcritical force on the byssus as already described, showing no signs of damage to the stem root (Fig. 3, A and B; fig. S2; and movie S2). Histological analysis of the generator tissue immediately after induced stem release shows empty spaces between the cilia (Fig. 3D) where the lamellar sheets were observed in the native samples (Figs. 1G and 3C). Images indicate no damage to the redstaining cilia and no trace of the blue-staining lamellar material, supporting a clean release. Raman imaging of the generator tissue after induced stem root release further supports a clean, quick release with no associated tissue damage (Fig. 3, F to H). Conversely, during forced removal where the stem root shows visible damage, we observe remnants of the broken lamellae in histological sections of the generator (Fig. 3E). Cleanly released stem root lamellae were further analyzed by staining with Sirius red (Fig. 3I), revealing intact undamaged sheets
with no remnants of cilia or cellular debris attached. Consistent with STEM imaging, released lamellae show a red-staining squiggly fiber embedded in a nonstaining material, confirming that lamellae are indeed biphasic with a fibrous collagen core component and a noncollagen component that interacts directly with the cilia. Raman spectroscopic imaging of the released lamellar material further supports a biphasic composition with the collagenous core spectra strongly resembling that of the proximal thread and the outer noncollagen component showing a more disordered secondary structure (Fig. 3, J and K) (25). Role of cilia motility and neurotransmitters in stem release
Based on the clean release of the stem lamellae, we formulate two hypotheses: (i) The interaction between the septa surface and lamellae is strong, yet reversible and noncovalent. (ii) The release of the stem appears to be under science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
biological control and likely involves the cilia, given their tight interaction with the lamellae. To test these hypotheses, we exploit the fact that beating of motile cilia in mussel gills and the digestive system has been previously shown to be up-regulated and down-regulated in vitro by addition of serotonin and dopamine, respectively (27–30). To test if cilia beating movement might influence stem release, we independently injected serotonin and dopamine in a dose-dependent manner into the open gap between stem and byssus without disturbing the tissue, and the maximum force to remove the stem was measured with mechanical testing (Fig. 4A and figs. S2 and S3). After injection of 200 mM serotonin, the byssus stem could be easily removed with such little force, that it was not possible to measure because the act of attaching the byssus to the tensile tester resulted in stem release (Fig. 4A and movie S5). Serotonin showed a dose-dependent effect with higher pullout force associated with lower serotonin concentration. By contrast, dopamine injection significantly increased the force to remove the stem to 20.1 ± 5.9 N at 70 mM dopamine (compared to 6.7 ± 1.7 N for controls in which only water was injected). Histological analysis of the generator at these high forces indicates remnants of ruptured lamellae in between septae (Fig. 3E). Dopamine also shows a clear dose dependence (Fig. 4A), but there was a large drop in the maximum force at 200 mM dopamine to 3.2 ± 0.3 N. Further inspection reveals that in these samples, failure occurs not because the stem is released, but rather because the entire generator ruptured and pulled away from the underlying musculature—although the stem–generator biointerface remains intact. This may suggest an important toughening role associated with partial cilia movement (supplementary text). These findings strongly implicate neurotransmitter-regulated cilia movement in controlling the mechanical interaction between the living and nonliving tissues, suggesting that the release process is serotonergic. To further explore this hypothesis, we dissected tissue from the generator and used light microscopy in differential interference contrast (DIC) mode to visualize and quantify the cilia beating velocity (Fig. 4B and movie S5) (31). Freshly dissected tissue exhibited cilia movement for up to 1 hour after dissection, which is consistent with previous reports that dissection can lead to the activation of the serotonin pathways in mussels (32). Recording cilia maximum velocity using image tracking software (33, 34) (Fig. 4, B and C, and fig. S7), we observed that addition of serotonin to the tissue in vitro resulted in a substantial increase in the number of beating cilia and an increased beating velocity, whereas addition of metergoline—a nonselective serotonin receptor antagonist (35)—resulted in a concomitant decrease both in the number of SCIENCE science.org
beating cilia detected and the beating velocity (Fig. 4C, fig. S8, and movie S6). A subsequent second addition of serotonin again increased both the number of cilia beating and beating velocity, strongly supporting the role of the serotonin pathway in the stem-release process. The generator is a strong and serotonergic quick-release biointerface
Mytilus mussels have evolved a distinctive solution for fabricating a strong, quick-release biointerface by interdigitating stiff, wavy biopolymer sheets with living tissue and carpeting the surfaces with soft motile cilia. Yet, it is highly unusual for cilia to be used for such a mechanically demanding task. ECM tissue is typically quite soft and viscoelastic (36), and the flexural stiffness of cilia is predicted to be extremely low (37). By contrast, given the similarity of lamella Raman spectra to byssal thread collagen proteins, we predict a high lamella stiffness of 50 to 500 MPa (13). How can we then reconcile the ability of this mechanically mismatched, submillimeter-scale biointerface to sustain repeated loadings of up to 20 N? Using extracted values from light microscopy images, we can roughly estimate a value of 4.1 cm2 for the interfacial area between the stem lamellae and the cilia on the basis of their tight interdigitation (supplementary text). Given this surface area, a force of 20 N translates to a stress of only ~50 kPa distributed over the interface. For comparison, this same force would translate to a stress of 1.1 GPa in a single byssal thread, which is about 50 times the proximal thread strength for M. edulis (38) [incidentally, there are normally at least 50 threads in a mature byssus, and failure most commonly occurs at the substrate (39)]. We estimate that there are ~5.7 billion cilia in contact with the lamellar surface (supplementary text). Therefore, if an applied force of 20 N is evenly distributed, each cilium would experience a force of ~3.5 nN. Similarly, reversible adhesion by gecko toe pads is dependent on billions of densely packed tiny hairs called spatulae (similar to cilia in terms of size, aspect ratio, and lateral packing density), each of which achieves adhesive forces of 2 to 16 nN primarily through van der Waals and other weak interactions (40–42). The collective action of many billions of such interactions on gecko toe pads enables large frictional forces when loaded in parallel (40). Similarly, the cilia-lamellar interface in the generator will be loaded parallel to the stem axis when a force acts on the byssus (Fig. 4, D and E), enhancing the number of interactions and thus, force capacity. Furthermore, the interdigitation of multiple lamellae and septae is superficially analogous to the interleaving of pages from two phone books. In this popularized feat, shockingly large forces are required to pull apart the two books owing to amplification of friction between the individual sheets
enhanced by their bending angle (43). Similarly, the bending and interleaving of many lamellar sheets in the spaces between septae may further increase the pullout forces that can be sustained before failure. Thus, the key question is not why the stem root biointerface is so strong, but rather how it can be released on demand without failure. To disengage from surfaces, geckos change the angle of applied force by rolling back their toe pads, which focuses stress on fewer individual bonds, precipitating quick release (40). Perhaps analogous to gecko toe rolling (40), we have observed that the motile cilia at the stem root biointerface can actively move relative to the lamellar surface under an appropriate neurochemical trigger (Fig. 4, F and G). This oscillating motion apparently disrupts the adhesive interaction between the cilia and lamellae surface, enabling the release of the entire stem root under subcritical forces (Fig. 4G). Presumably, the number of cilia moving and/or the velocity at which they are beating can modulate the overall force required to remove the byssus, enabling the observed serotonin and dopamine dose dependency of pullout force (Fig. 4, A, F, and G) and allowing mussels to jettison their byssus without damaging the tissue within (fig. S1) (30). Consistent with this hypothesis, it has been shown that cilia on the epithelial lining of Mytilus edulis foot and gill tissue are regulated by dopaminergic and serotonergic nerves, with serotonin having an excitatory effect and dopamine having an inhibitory effect on ciliary beating (28, 29). Thus, it is reasonable to posit that serotonergic and dopaminergic nerve fibers in the generator comprise a complex feedback system for regulating biointerface strength and release. As support, induced pullout experiments (fig. S4 and movies S2 and S3) may indicate the presence of a mechanosensory pathway, and it was previously shown that heat stress treatments increased serotonin levels in the M. edulis central nervous system (44) and that gill cilia beating rate increases with increasing temperature (45). This is in line with suggestions that mussels in hightemperature tide pools will release their byssus to relocate (18). We thus hypothesize that mussels may convert sensory information about their environment into a neurochemical signal that is translated into increased cilia beating that enables byssus release and increased mobility, which could enhance mussel survival under unfavorable conditions. Coupled to increased cilia beating, a force applied by the byssus retractor muscles may be sufficient for inducing spontaneous release (14, 16). Without a byssus attachment, mussels can freely pull themselves around by their foot to find a suitable location before putting down a new byssus (14). Indeed, increased mussel mobility has been reported in response to many different environmental factors, 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
833
RES EARCH | R E S E A R C H A R T I C L E S
including high temperature (18), chemical cues from predators (21), wave exposure (19), and even substrate chemistry (20). This implies a complex sensory system that can translate physical, chemical, and mechanical stimuli into specific neurochemical signals in the generator. Although the cellular mechanism by which this is mediated is still being elucidated, we demonstrate here that the Mytilid byssus stem root provides an example of an effective quick-release mechanism for interfacing living tissues with nonliving materials that is not found in human technology. RE FE RENCES AND N OT ES
1. Y. Fang et al., Chem. Rev. 122, 5233–5276 (2022). 2. L.-L. Li, G.-B. Qi, F. Yu, S.-J. Liu, H. Wang, Adv. Mater. 27, 3181–3188 (2015). 3. Y. Luo et al., ACS Nano 17, 5211–5295 (2023). 4. S. M. Wellman et al., Adv. Funct. Mater. 28, 1701269 (2018). 5. R. Elnathan et al., Nat. Rev. Mater. 7, 953–973 (2022). 6. M. W. Tibbitt, C. B. Rodell, J. A. Burdick, K. S. Anseth, Proc. Natl. Acad. Sci. U.S.A. 112, 14444–14451 (2015). 7. Z. Liu, M. A. Meyers, Z. Zhang, R. O. Ritchie, Prog. Mater. Sci. 88, 467–498 (2017). 8. J. H. Waite, H. C. Lichtenegger, G. D. Stucky, P. Hansma, Biochemistry 43, 7653–7662 (2004). 9. T. R. Dunson, R. N. Amatya, S. L. Krueger, Obstet. Gynecol. 85, 543–548 (1995). 10. H. Y. Yoo et al., Nat. Commun. 7, 11923 (2016). 11. J. Choi, E. Hennebert, P. Flammang, D. S. Hwang, Biomater. Sci. 8, 3751–3759 (2020). 12. A. Tamarin, J. Morphol. 145, 151–177 (1975). 13. J. H. Waite, M. J. Harrington, Can. J. Chem. 100, 197–211 (2022). 14. H. A. Price, J. Molluscan Stud. 49, 9–17 (1983). 15. M. W. Denny, B. Gaylord, Annu. Rev. Mar. Sci. 2, 89–114 (2010). 16. E. Carrington, J. H. Waite, G. Sarà, K. P. Sebens, Annu. Rev. Mar. Sci. 7, 443–469 (2015). 17. J. H. Waite, C. C. Broomell, J. Exp. Biol. 215, 873–883 (2012). 18. E. A. Roberts et al., Funct. Ecol. 35, 1222–1239 (2021). 19. H. L. Hunt, R. E. Scheibling, Veliger 45, 273–277 (2002). 20. J. Choi, S. Lee, K. Ohkawa, D. S. Hwang, ACS Nano 15, 18566–18579 (2021). 21. I. M. Côté, E. Jelnikar, J. Exp. Mar. Biol. Ecol. 235, 201–211 (1999). 22. J. H. Waite, “The formation of mussel byssus: Anatomy of a natural manufacturing process” in Structure, Cellular Synthesis and Assembly of Biopolymers, S. T. Case, Ed. (Springer, 1992), pp. 27–54. 23. F. Jehle et al., ACS Nano 15, 6829–6838 (2021). 24. T. Ishikawa, Cold Spring Harb. Perspect. Biol. 9, a028076 (2017). 25. L. Youssef, M. Renner-Rao, E. D. Eren, F. Jehle, M. J. Harrington, ACS Nano 17, 2294–2305 (2023). 26. T. Priemel et al., Science 374, 206–211 (2021). 27. R. Beiras, J. Widdows, Mar. Biol. 122, 597–603 (1995). 28. G. B. Stefano, E. Aiello, Biol. Bull. 148, 141–156 (1975). 29. L. Vitellaro-Zuccarello, S. De Biasi, A. Amadeo, Cell Tissue Res. 261, 467–476 (1990). 30. S. L. Schor, Science 148, 500–501 (1965). 31. R. Francis, C. Lo, J. Vis. Exp., 50343 (2013). 32. E. L. Aiello, Physiol. Zool. 33, 120–135 (1960). 33. D. Ershov et al., Nat. Methods 19, 829–832 (2022). 34. K. Jaqaman et al., Nat. Methods 5, 695–702 (2008). 35. M. Lima-Maximino et al., J. Neurochem. 153, 495–509 (2020). 36. J. R. Staunton, W. Y. So, C. D. Paul, K. Tanner, Proc. Natl. Acad. Sci. U.S.A. 116, 14448–14455 (2019). 37. D. A. Hoey, M. E. Downs, C. R. Jacobs, J. Biomech. 45, 17–26 (2012). 38. E. Bell, J. Gosline, J. Exp. Biol. 199, 1005–1017 (1996). 39. E. C. Bell, J. M. Gosline, Mar. Ecol. Prog. Ser. 159, 197–208 (1997). 40. Y. Tian et al., Proc. Natl. Acad. Sci. U.S.A. 103, 19320–19325 (2006). 41. S. Singla et al., Sci. Adv. 7, eabd9410 (2021). 42. G. Huber et al., Proc. Natl. Acad. Sci. U.S.A. 102, 16293–16296 (2005).
834
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
43. H. Alarcón et al., Phys. Rev. Lett. 116, 015502 (2016). 44. G. B. Stefano, L. Hiripi, E. J. Catapane, J. Therm. Biol. 3, 79–83 (1978). 45. G. B. Stefano, E. J. Catapane, J. M. Stefano, Biol. Bull. 153, 618–629 (1977). ACKN OWLED GMEN TS
We thank the staff at FEMR (McGill) for their support in imaging, the staff at Helmotz Zentrum Berlin including H. Markötter at the BAM beamline for support with PCE-mCT, and T. van de Ven for use of the tensile tester. We thank H. F. Hollinger and J. H. Waite for helpful discussion. Funding: Natural Sciences and Engineering Research Council of Canada, NSERC Discovery Grant RGPIN-201805243 (M.J.H.), Canada Research Chair award, CRC Tier 2 950231953 (M.J.H.), Max Planck Society (F.J.), FQRNT Quebec Merit Fellowship for Foreign Students (L.Y., T.P.). Author contributions: Conceptualization: J.S., L.Y., T.P., M.J.H. Methodology: J.S., L.Y., T.P., P.Z., F.J. Investigation: J.S., L.Y., T.P., S.M., D.E., P.Z., F.J. Visualization: J.S., L.Y., T.P., P.Z., F.J. Funding acquisition: L.Y., T.P., M.J.H. Project administration: M.J.H. Supervision: M.J.H. Writing –
original draft: J.S., L.Y., M.J.H. Writing – review and editing: All authors Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data are available in the main text or the supplementary materials. License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.sciencemag.org/about/science-licensesjournal-article-reuse SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adi7401 Materials and Methods Supplementary Text Figs. S1 to S8 Movies S1 to S6 MDAR Reproducibility Checklist Submitted 15 May 2023; accepted 29 September 2023 10.1126/science.adi7401
OCEAN HEAT
Surface climate signals transmitted rapidly to deep North Atlantic throughout last millennium Wanyi Lu1*, Delia W. Oppo1, Geoffrey Gebbie1, David J. R. Thornalley1,2 Instrumental observations of subsurface ocean warming imply that ocean heat uptake has slowed 20th-century surface warming. We present high-resolution records from subpolar North Atlantic sediments that are consistent with instrumental observations of surface and deep warming/freshening and in addition reconstruct the surface-deep relation of the last 1200 years. Sites from ~1300 meters and deeper suggest an ~0.5 degrees celsius cooling across the Medieval Climate Anomaly to Little Ice Age transition that began ~1350 ± 50 common era (CE), whereas surface records suggest asynchronous cooling onset spanning ~600 years. These data suggest that ocean circulation integrates surface variability that is transmitted rapidly to depth by the Atlantic Meridional Ocean Circulation, implying that the ocean moderated EarthÕs surface temperature throughout the last millennium as it does today.
E
arth’s surface has generally warmed over the last century (1), but the ocean has slowed this warming by taking up over 90% of the excess thermal energy since 1955 (2) and increasingly storing it in the deep oceans (3–5). Climate reconstructions provide a baseline for assessing the anomalous nature of 20th-century change and the role of the deep ocean in moderating surface climate on longer time scales. The common era (CE) (i.e., the last ~2000 years) contained substantial climate variability, including a cooling trend from the peak of the Medieval Climate Anomaly [(MCA) around 850 to 1250 CE] to the Little Ice Age [(LIA) around 1400 to 1850 CE], and rapid industrial warming since ~1850 (1, 6). The high-latitude North Atlantic Ocean is an important region where deep water forms and surface temperature anomalies are expected to be transported efficiently to depth through the Atlantic Meridional Overturning Circula1
Woods Hole Oceanographic Institution, Woods Hole, MA, USA. 2Department of Geography, University College London, London, UK. *Corresponding author. Email: [email protected]
tion (AMOC) (7). However, few high-resolution records from the deep North Atlantic span the MCA-LIA transition (8) and the hypothesis that an active AMOC has moderated surface climate on centennial time scales across this transition has not been evaluated with deep North Atlantic proxy records. Here we present data from well-dated sediment cores that form a depth transect spanning ~1000 to 2300 m and sample several important water masses in the subpolar North Atlantic. These data permit us to compare changes in the properties of the Nordic Overflows with those of waters formed south of the Nordic Seas and, with the insights of a model, place aspects of modern ocean warming in a longer-term context. We used 11 marine sediment cores from south of Iceland along the eastern flank of Reykjanes Ridge, collected in 2014 on the research vessel R/V Endeavor (cruise EN539) (table S1 and Fig. 1). Iceland Scotland Overflow (ISOW) entering the northern Iceland Basin is much denser than the ambient Atlantic intermediate waters near the sill depths (~500 to 800 m), resulting in vigorous mixing and entrainment, especially during its initial descent but also as it science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
A
B
Location map
Sea surface temperature
o
C
C
Sea surface salinity
psu
W DSO
SP
D
MC22A MC28A MC26A MC21A MC25A MC20A MC14A MC19A MC10A MC9A MC13A
LS W
NA
G
C
IS
O
W
EGC
Temperature section
o
C
E
Salinity section
F δ18Ocalcite section
psu
27.6 SPMW
SPMW
MC22A MC21A MC20A MC19A ISOW MC10A MC9A MC13A MC14A
Reykjanes Ridge
MC28A MC26A MC25A
LSW MC21A MC19A
ISOW MC10A
MC9A MC13A MC14A
Fig. 1. Locations of 11 EN539 multicores. Schematic illustrations of (A) ocean circulation, (B) sea surface temperature, (C) sea surface salinity, (D) temperature, (E) salinity, and (F) d18Ocalcite sections along the eastern flank of Reykjanes Ridge. The isopycnal layers are shown by dashed lines in (F) with the 28 kg/m3 isopycnal approximately defining ISOW near the ridge (fig. S1) (38). The ocean circulation paths were adapted from (39), with thin red arrows showing warm salty surface Atlantic waters, thin green arrows showing cold fresh surface water sourced from the Arctic, and thick dashed lines showing deep overflow pathways. Red shaded
flows along the flank of the Reykjanes Ridge (9). A water mass decomposition from an inversion of oceanographic data (figs. S1 and S2) (10) indicates that all our core sites contain a mixture of ISOW, colder, fresher Labrador Sea Water (LSW), and warmer, saltier Subpolar Mode Water (SPMW). SPMW is most prevalent at the shallowest site whereas the ISOW contribution increases with depth, reaching a maximum of nearly 60% at our deepest site. The LSW contribution is relatively constant with depth, with a maximum at about 1300 m. Higher seawater density north of the sills, in comparison to south of them, has largely driven the transport of overflow waters over the last century (11). We measured the oxygen and carbon isotope ratios, d18O and d13C (the 18O/16O and 13C/12C in each sample, relative to that of an international standard), in two species of planktic foraminifera [Globigerina bulloides, which calcifies in the upper ~50 m (12); Globorotalia inflata, which calcifies as deep as ~300 m (13)] and in one of two species of benthic foraminifera (Cibicidoides wuellerstorfi or Uvigerina peregrina), SCIENCE science.org
MC22A MC20A
[m]
MC28A MC26A MC25A
LSW
27.8 LSW 28.0
‰
SPMW
MC28A MC26A MC25A MC22A MC21A MC20A MC19A ISOW MC10A MC9A MC13A MC14A
area in (A) marks the location of the sections shown in (D) to (F). Temperature and salinity data are the average of six decadal climatologies from 1955 to 2017 from the World Ocean Atlas 2018 (40). The d18Ocalcite section was converted from an oceanographic data inversion (10) using empirical calibrations (14). The figures were generated using Ocean Data View software (41). SPG, Subpolar Gyre; NAC, North Atlantic Current; EGC, East Greenland Current; ISOW, Iceland-Scotland Overflow Water; DSOW, Denmark Strait Overflow Water; LSW, Labrador Sea Water; SPMW, Subpolar Mode Water.
depending on their availability. Variations in the d18O of foraminifera reflect variations in calcification temperature and the d18O of seawater, the latter in turn influenced by salinity (14); foraminiferal d18O increases with increasing seawater d18O and with decreasing temperature. In addition, C. wuellerstorfi calcifies in equilibrium with seawater, but d18O values in U. peregrina are ~0.47‰ higher (14). In the modern subpolar North Atlantic below 1000 m (15), variations in the d18O of calcite are largely due to temperature variability (R2 = 0.93) rather than salinity (R2 = 0.00) (fig. S3), and modern measured benthic d18O closely follows the predicted d18O of calcite (fig. S2). Chronologies and their uncertainties were constrained by radiocarbon data and determined using Bayesian methods (16) (see Methods). All cores have high sediment accumulation rates (~25 to 70 cm per 1000 years) and except for MC22A and MC13A, have modern core tops as indicated by radiocarbon [fraction modern (Fm) >1)] (table S1 and fig. S4). The records span the last ~500 to 2250 years (Fig. 2, data S1 to S3, and figs. S5 to S14).
Rapid transmission of common era surface climate trends to depth
All planktic d13C records but one from cores with Fm >1 show a sharp decrease in recently deposited sediments (figs. S9, S11, and S12), reflecting the oceanic uptake of isotopically light anthropogenic carbon released by fossil fuel burning since the early 19th century (17). The rapid decrease of planktic d13C near the top of MC13A suggests that its top is modern despite its relatively low Fm value. Out of the 10 benthic records from cores with modern tops, 8 have the lowest values near their tops. Of these, the d13C decrease is significant relative to the post-1850 period in six records (figs. S7 and S12). The amplitude of the d13C decrease in the benthic records ranges from about 0.1 to 0.4‰, consistent with modeled amplitude (18). The finding of low core-top d13C in the benthic records confirms an important role for AMOC in sequestering anthropogenic carbon into the deep ocean. Most planktic d18O records from cores with modern tops show a decrease since the early 20th-century, suggesting warming and/or freshening (Fig. 2, fig. S8, and fig. S10). At least half 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
835
Model OPT-0015 1.50
−1
C
G. inflata
1.75
1.25
MC26A (1190 m) 2.75
1.75
MC25A
2.00
1.50 1.75
MC22A MC21A
1.75
2.75
MC22A (1535 m)
2.50 1.75
δ18O (‰)
MC20A
MC20A
2.00
MC19A
MC19A
1.50
MC10A MC9A
1.75
1.75
δ18O (‰)
1.50
MC10A MC9A
1.50
Warm, fresh
1.75
MC13A MC14A MCA 0
400
800
LIA
δ18O (‰)
2.00
1.25
1200 1600 2000
17 NOVEMBER 2023 ¥ VOL 382 ISSUE 6672
3.00
MC20A (1711 m) MC19A (1810 m)
2.75
2.00
MC10A (1999 m) MC9A (2103 m)
3.00
2.50
1.75
Warm, fresh
MC13A MC14A
2.00
MCA 0
400
2.75 3.00
800
LIA
Cold, salty
1200 1600 2000
Year (CE)
Fig. 2. Planktic and benthic d18O records compared with global mean surface air temperature (GMST) history and temperature anomalies from the OPT-0015 inversion model (25). (A) Global mean surface air temperature anomalies with respect to 1961 to 1990 CE (1). (B) to (D) 11 foraminiferal d18O records arranged by increasing core water depth. Thick colored lines correspond to the median d18O ensemble member and colored shading shows the 90% highest-density probability ranges determined by R software package
of the benthic d18O records also show a decrease during this time period (Fig. 2 and fig. S6), although the signals are smaller than in the planktic records and a statistical test, discussed below, was used to establish their significance.
2.75
2.50
1.50
Cold, salty
Year (CE)
3.00
MC21A (1647 m)
δ18O (‰)
1.50
MC25A (1310 m)
2.50
2.00
1.25
δ18O (‰)
3.25
δ18O (‰)
1.25
1.75
836
3.00
1.50
δ18O (‰)
δ18O (‰)
MC22A MC21A
δ18O (‰)
1.25 1.50
2.75 3.00
δ18O (‰)
MC25A
2.50
δ18O (‰)
1.75
D
Benthic Model OPT-0015 MC28A (1031 m)
1.50 1.50
−1
2.00
MC28A MC26A δ18O (‰)
δ18O (‰)
1.75
MC28A MC26A
0
1.50
Model OPT-0015
LIA
δ18O (‰)
1.25
0
MCA
1
δ18O (‰)
B
G. bulloides
LIA
δ18O (‰)
−1
MCA
1
GMST anomaly (oC)
0
A
δ18O (‰)
LIA
Instrumental Reconstruction
GMST anomaly (oC)
MCA
1
δ18O (‰)
GMST anomaly (oC)
RES EARCH | R E S E A R C H A R T I C L E S
Warm, fresh
MC13A (2129 m) MC14A (2274 m) MCA 0
400
800
LIA
Cold, salty
1200 1600 2000
Year (CE)
“geoChronR” (16) (Methods). To compare the modeled temperature trends (black dashed lines, OPT-0015) (25) with the d18O changes, we scaled a 1°C decrease to correspond to a 0.22‰ d18O increase (14). Thick color-coded arrows indicate mean ensemble change point ages where they are significant and cyan-colored bars denote the ranges of these ages. Benthic records in MC28A, 26A, and 25A were generated on U. peregrina whereas the other benthic records were generated on C. wuellerstorfi.
Most benthic and planktic records extending into the MCA appear to show a d18O increase across the MCA-LIA transition, suggesting cooling and/or increasing salinity. However, at the shallowest site (MC28A, ~1000 m) the benthic
d18O decreases across the MCA-LIA transition and through most of the LIA. The amplitude of the MCA-LIA benthic d18O increase is also relatively small at the next deepest site (MC26A, ~1200 m) compared with the deeper sites. science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
A
Pre-1850 (LIA minus MCA)
Benthic
Δδ18OLIA − MCA (‰)
−0.1
G. bulloides
G. inflata MC25A
0.0
MC26A MC13A MC25A
0.1
OPT−0015
MC22A OPT−0015 MC22A
MC26A
MC13A
MC25A
Cold, salty LIA
MC22A
OPT−0015
MC26A
Warm, fresh LIA
MC13A
RAPiD-17-5P
MC9A
MC9A
0.2 MC9A
1000
1500
2000
1000
1500
2000
1000
1500
2000
Δδ18Opost-1950 − pre-1950 (‰)
B Industrial era (post-CP minus pre-CP) Benthic
−0.50
G. bulloides
Benthic G. bulloides G. inflata
MC19A MC13A
MC20A MC10A OPT−0015 MC13A
0.00
MC10A MC26A
MC21A MC25A
Data Model Species
G. inflata MC28A
−0.25
Type
MC20A MC10A MC26A MC28A Warm, fresh MC14A
MC25A MC20A MC14A
OPT−0015 MC13A
OPT−0015
MC25A
post-1950s
Cold, salty 0.25 1000
1500
2000
1000
1500
2000
1000
1500
2000
Change point estimate (CE) Fig. 3. Change point (CP) estimates versus d18O differences (after minus before the CP). This figure includes both d18O records and OPT-0015 inversion for the MCA-LIA transition (pre-1850 dataset) (A) and industrial era (post1850 dataset) (B). Note that the y axes are reversed. Points below the zero
To determine whether and when significant changes in the mean of d18O time series occurred, we computed the change points of the d18O records, adapting a method that accounts for age uncertainty and data variability and was previously used to detect AMOC strength change (19) (Methods). We divided the d18O records into post-1850 and pre-1850 datasets and conducted change point analyses on each time interval. For the post-1850 interval we excluded MC22A, which does not have a modern core top. For the pre-1850 dataset we only included the six cores with Bayesian ensemble median ages older than 1200 CE (table S1). For each core and foraminifera species we computed the average d18O difference before and after the change point if the significance test was passed (Fig. 3). From these change point analyses, our key findings are that (i) most planktic (15 out of 20) and half of the benthic (5 out of 10) records show statistically significant 20th-century d18O decreases; (ii) most planktic (10 out of 15) and benthic (5 out of 6) records show statistically significant d18O increases across the MCA-LIA transition; and SCIENCE science.org
dashed line in (A) imply colder saltier LIA conditions; points above the zero dashed line in (B) imply warmer, fresher post-1950s conditions. In the benthic records, open circles denote U. peregrina and closed circles denote C. wuellerstorfi. The G. inflata d18O record of nearby core RAPiD-17-5P is from (42).
(iii) of the records with a significant d18O increase across the MCA-LIA transition, the average ages of the benthic d18O change points are in a narrow range (~1346 ± 49 CE) whereas the planktic d18O change points occur across a ~600-year range, between 1100 and 1700 CE (average ~1357 ± 216 CE) (data S4). The large range of planktic change points may reflect earlier cooling at our northern sites in contrast to our southern sites (fig. S15). Bioturbation, coupled with higher abundances of the planktic foraminifera near the tops of the cores, may have resulted in a small (~2 to 3 cm) downcore shift of the recent planktic d18O decrease (figs. S16 and S17), implying that the post-1850 change point may have been more recent than implied by our analyses. Composites of raw d18O data from all cores on their median Bayesian ages confirm larger variability in the planktic d18O than benthic d18O records (fig. S14). We interpret the 20th-century d18O decrease recorded in the planktic records as rapid warming and freshening of subpolar surface and nearsurface North Atlantic waters, consistent with planktic faunal changes in the same cores (20)
(fig. S18) and instrumental evidence indicating surface or near-surface warming and freshening trends since the 1950s both basin-wide (21, 22) and locally (23) (fig. S19). The smaller magnitude of the benthic (0.04 to 0.22‰) than planktic d18O decreases (0.06 to 0.43‰) in the late 20th-century (Fig. 3B) is consistent with a recent, rapid, high-amplitude surface signal that was diluted by mixing with older waters in transit to the deep core sites. Thus, the corresponding trends of decreasing d18O and direct observations of recent warming and freshening imply that we can use the d18O signals in these cores to infer past changes related to seawater density. With the exception of the MC25A G. inflata record, average planktic d18O was ~0.05 to 0.25‰ higher after the MCA-LIA change point than before it (Fig. 3A). At sites deeper than 1300 m the mean benthic d18O is 0.05 to 0.14‰ higher after the MCA-LIA change point than before it. If these changes were driven by temperature then the benthic d18O increases would correspond to 0.2 to 0.6°C average cooling (14). The smoothed benthic composite 17 NOVEMBER 2023 • VOL 382 ISSUE 6672
837
RES EARCH | R E S E A R C H A R T I C L E S
A
0.1
0
0
0.5 05
0 -0.1
500 0.3
-0.2
0
0.4
-0 0.4 -0.5 05
0.6
-0.1
0.6
0.4
0.1
0.
4
-0.1
0.3
0
-0.3
0.2
1
.2 0.2 0
0.6 7 0.7
0.3
0
0..1
0.5
1500
0.3 0
0.5
1000
-0. 0
0.1
Depth (m)
Arctic amplification of MCA-LIA cooling
θ [oC]: OPT-0015 0.4
0
1 0.1
-0 - .1
2000
0.8 0.6
2500
200
400
600
800
1000
1200
1400
1600
1800
2000
B
0.2 0 0.2 0.3 0.5
0.1
0.1 01
0.2
-0.5
-0.4
0.1
-0.6 -0.8 -1 01 0.1
0 -0.3
1500
0.2 .
-0.1
Depth (m)
0.6
-0.2
0
.5 -0
0 0.4
0.3
0.1 -0
.3
-0
-0.4
1000
0
-0.2
-0.1
0
500
0.4
-0.2
04 -0.4
-0.5
2000
2500 1850
1900
1950
2000
Year (CE) Fig. 4. Anomalous temperature evolution from south of Iceland (grid box of 61.5°N, 20.5°W). From the OPT-0015 inversion (25) during the last 2000 years (A) and from 1850 to 2015 CE (B).
record suggests that on average, the deep sites (>1300 m) increased by ~0.1‰ or cooled by ~0.5°C, similar to the change point results from individual records (Fig. 3A and fig. S14). Although temperature likely dominated the benthic d18O increases at these deep sites (>1300 m) we cannot rule out that the d18O of one or more water masses influencing these sites changed across this transition. Rapid transfer of surface signals to our deep core sites is consistent with young water mass ages at the sites (~35 to 65 years; fig. S2) implied by a global inversion of modern oceanographic data (10). We infer that the greater range in the timing of change points in the planktic records across the MCA-LIA transition (1076 to 1712 CE) as compared with the benthic records (1275 to 1395 CE) (Fig. 3A) is due to several factors, including larger sea838
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
sonal and depth-habitat variations of planktic foraminifera, the time-transgressive nature of surface change, larger temporal variability in the surface (for example the meandering of zonal fronts on seasonal-interannual-decadal timescales), and a potential contribution due to bioturbation coupled with planktic foraminifera abundance changes (see SM). Furthermore, the narrower range of benthic than planktic change points is the expected consequence of interior ocean mixing that damps surface variability. Globally averaged temperature anomalies, for example, exhibit their greatest interannual variability above 500 m depth, but the deeper ocean is more representative of the longer-term ocean heat gain (24). Thus, the benthic records integrate the surface variability and more reliably record the overall timing of the MCA-LIA cooling/salinification.
We compare our d18O records with results from an ocean model inversion (25) (referred to as OPT-0015 hereinafter), which fits an empirical ocean circulation model to modern-day tracer observations, historical deep sea temperatures in the 1870s (26), and global-mean Ocean2k sea surface temperatures (SST) reconstructed for the Common Era (6) (see SM). The OPT-0015 inversion solves for the threedimensional (3D) evolution of temperature throughout the Common Era under the assumption of a fixed, modern-day ocean circulation. In the inversion SST is allowed to vary regionally to fit the subsurface constraints. We extracted the OPT-0015 model temperature simulated at the model grid nearest our cores and converted the temperature into d18O changes using empirical calibrations (14). The model exhibits temporal trends consistent with those in the foraminiferal d18O records of the surface and the deeper sites (Figs. 2 and 4), showing both LIA cooling and 20th-century warming. To compare data and model change points, we first obtained model variability by computing 100-year moving averages in the surface mixed layer and at 2000 m depth, and then performed change point analyses on the moving averages (fig. S20). For both the surface and deep sites, the post-1850 change point in the model and data are within error, especially considering chronological uncertainty in our records (Fig. 3B). On the other hand, there is a large datamodel change point mismatch in the timing of significant LIA cooling/salinification, with the model change points occurring ~350 years before the average benthic and planktic change points (Fig. 3A). This occurs because the model was constrained with the Ocean2k SSTs (25), which contain cooling before and during the MCA, and which the model then faithfully reproduces. By contrast, other independent estimates of the timing of the LIA from regions proximal to our study area yield ages that are more consistent with the change point derived from our benthic d18O records (~1346 ± 49 CE), such as an Arctic temperature reconstruction (27) (~1258 ± 2 CE) and Greenland ice cap growth records (28) (~1353 ± 9 CE). The model inversion suggests that for multicentennial variability in the pre-1850s, including the MCA-LIA transition, temperature change in the deep sea (>1000 m) was greater than that of the upper ocean (1300 m are consistent with the model prediction of ~0.5°C LIA cooling in the deep sea (Fig. 3A). The reason that the deep sea cools more than the upper ocean during the simulated LIA is related to Arctic amplification of the LIA cooling, science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
which is present in the SST inversion (25) and was also simulated in the Community Earth System Model-Last Millennium Ensemble (29). Thus, cooling in the Nordic Seas where the overflows originated was greater than in waters formed south of the sills (25, 29). With an active AMOC, larger cold anomalies from the Nordic Seas are transmitted to depth by the overflows. In contrast to the MCA-LIA cooling, modern warming is (thus far) concentrated in the upper ocean, both in observations and in OPT0015 (Figs. 3B and 4B). During the MCA–LIA transition, the deep ocean had sufficient time to record the cooling which lasted ~600 years whereas the time interval of 20th-century warming was shorter. Thus the core sites, having average water mass ages of 35 to 65 years (modern, fig. S2), have not yet had sufficient time to fully record the warming at depth and anomalies at depth will always lag those at the surface unless warming ceases. MCAÐLIA SPMW freshening/warming
The model inversion suggests a large cooling across the MCA-LIA transition that is not evident in the benthic d18O data of the two shallowest sites (MC28A and 26A) (Fig. 2). This benthic data-model mismatch likely reflects MCA-LIA oceanographic changes (i.e., changes in circulation, seawater d18O, salinity), that were not considered in the inversion. The shallowest site, MC28A, which shows a trend of decreasing d18O from the MCA through the end of the LIA, is currently within the high salinity zone of SPMW (Fig. 1), which has been diluted by mixing with LSW and overflows (figs. S1 and S2). It is possible that a cooling trend at site MC28A was compensated by freshening that is not considered in the model and that the temperature-related d18O increase at MC26A was also dampened by freshening. Fresh, lowd18O polar waters may have been incorporated into SPMW, which most affects the two shallowest sites. This hypothesis is consistent with evidence of increased sea ice export from the Arctic that began at ~1300 CE and continued through the LIA (8, 30, 31). SPMW freshening during the LIA was also inferred downstream of the eastern subtropical gyre (32). Alternatively, or in addition, a greater contribution of a fresher, lower-d18O upper LSW relative to SPMW to our shallow sites could have resulted in the observed d18O decrease. Given that overflows entrain less dense waters during their descent along the ridge (9), it is possible that all our core sites were fresher during the LIA than MCA. If the LIA cooling from the OPT0015 inversion (0.4 to 0.6°C) is assumed accurate, cooling at all but one site would have been partially compensated by a decrease in d18OSW (freshening) (fig. S21). Such a freshening at depth may also explain why the benthic d18O data do not record the higher amplitude of the MCA-LIA deep cooling compared with SCIENCE science.org
the surface suggested by OPT-0015 (Figs. 3 and 4 and fig. S20). However, we cannot rule out the possibility that the MCA-LIA trend of decreasing benthic d18O at our shallow site reflects warming of the SPMW. Possible warming mechanisms include greater transport of warm subtropical waters to the SPMW formation regions (33, 34) or a weakening AMOC, which results in upper ocean subsurface warming due to reduced convection and exchange with the overlying cold atmosphere (35). Conclusions
Our data provide strong support for a persistent role of the AMOC in transferring anomalous upper ocean heat and freshwater to depth during the last ~1200 years. The records indicate a deep ocean that cooled and lost heat during the LIA, implying that the heat was transferred to the upper ocean and atmosphere (25). Thus, the ocean acted to dampen MCA-LIA surface change much like it has dampened surface warming during the industrial era. Whereas modern warming is surface intensified—typically in the upper ~700 m (2)— our model simulation suggests that on longer time scales temperature change in the deep subpolar North Atlantic exceeds that in the upper ocean, consistent with polar amplification of temperature change and an active overflow. The model simulation we used assumes that the intensity of the AMOC was unchanged from the modern era. If the AMOC has declined during the 20th-century as several studies suggest (36, 37), and AMOC during most of the Common Era was stronger, then the pre–20thcentury AMOC may have played a larger role in transferring surface climate signals to depth than in the modern. RE FERENCES AND NOTES
1. R. Neukom et al., Nat. Geosci. 12, 643–649 (2019). 2. S. Levitus et al., Geophys. Res. Lett. 39, L10603 (2012). 3. G. A. Meehl, J. M. Arblaster, J. T. Fasullo, A. Hu, K. E. Trenberth, Nat. Clim. Chang. 1, 360–364 (2011). 4. X. Chen, K. K. Tung, Science 345, 897–903 (2014). 5. M.-J. Messias, H. Mercier, Commun. Earth Environ. 3, 118 (2022). 6. H. McGregor et al., Nat. Geosci. 8, 671–677 (2015). 7. B. Dickson et al., Nature 416, 832–837 (2002). 8. P. Moffa-Sánchez et al., Paleoceanogr. Paleoclimatol. 34, 1399–1436 (2019). 9. H. R. Langehaug et al., Paleoceanography 31, 399–415 (2016). 10. G. Gebbie, P. Huybers, J. Phys. Oceanogr. 42, 291–305 (2012). 11. T. Rossby, L. Chafik, L. Houpert, Geophys. Res. Lett. 47, e2020GL087456 (2020). 12. L. Jonkers, S. van Heuven, R. Zahn, F. J. C. Peeters, Paleoceanography 28, 164–174 (2013). 13. A. Rebotim et al., J. Micropalaeontol. 38, 113–131 (2019). 14. T. M. Marchitto et al., Geochim. Cosmochim. Acta 130, 1–11 (2014). 15. G. A. Schmidt, G. R. Bigg, E. J. Rohling, Global Seawater Oxygen-18 Database, v1.22 (NASA, 1999); https://data.giss. nasa.gov/o18data/. 16. N. P. McKay, J. Emile-Geay, D. Khider, Geochronology 3, 149–169 (2021). 17. C. D. Keeling, Environ. Int. 2, 229–300 (1979). 18. M. Eide, A. Olsen, U. S. Ninnemann, T. Eldevik, Global Biogeochem. Cycles 31, 492–514 (2017). 19. L. Caesar, G. D. McCarthy, D. J. R. Thornalley, N. Cahill, S. Rahmstorf, Nat. Geosci. 14, 118–120 (2021).
20. P. T. Spooner et al., Geophys. Res. Lett. 47, e2020GL087577 (2020). 21. R. Curry, C. Mauritzen, Science 308, 1772–1774 (2005). 22. A. R. Friedman, G. Reverdin, M. Khodri, G. Gastineau, Geophys. Res. Lett. 44, 1866–1876 (2017). 23. Hafrannsóknastofnun (Marine and Freshwater Research Institute), Sjórannsóknir – Oceanography. (EMODnet, 2022); https://sjora.hafro.is/. 24. D. Roemmich et al., Nat. Clim. Chang. 5, 240–245 (2015). 25. G. Gebbie, P. Huybers, Science 363, 70–74 (2019). 26. D. Roemmich, W. J. Gould, J. Gilson, Nat. Clim. Chang. 2, 425–428 (2012). 27. N. P. McKay, D. S. Kaufman, Sci. Data 1, 140026 (2014). 28. M. B. Osman et al., Nat. Geosci. 14, 756–761 (2021). 29. B. L. Otto-Bliesner et al., Bull. Am. Meteorol. Soc. 97, 735–754 (2016). 30. M. W. Miles, C. S. Andresen, C. V. Dylmer, Sci. Adv. 6, eaba4320 (2020). 31. M. Alonso-Garcia et al., Clim. Past 13, 317–331 (2017). 32. A. Morley et al., Earth Planet. Sci. Lett. 308, 161–171 (2011). 33. D. Desbruyères, L. Chafik, G. Maze, Commun. Earth Environ. 2, 48 (2021). 34. D. J. R. Thornalley, H. Elderfield, I. N. McCave, Nature 457, 711–714 (2009). 35. Z. Liu et al., Science 325, 310–314 (2009). 36. D. J. R. Thornalley et al., Nature 556, 227–230 (2018). 37. L. Caesar, S. Rahmstorf, A. Robinson, G. Feulner, V. Saba, Nature 556, 191–196 (2018). 38. A. Bower, H. Furey, J. Geophys. Res. Oceans 122, 6989–7012 (2017). 39. M. I. García-Ibáñez et al., Prog. Oceanogr. 135, 18–36 (2015). 40. T. P. Boyer et al., World Ocean Atlas 2018 (Temperature and salinity) (NOAA National Centers for Environmental Information, 2018); https://www.ncei.noaa.gov/products/ world-ocean-atla. 41. R. Schlitzer, Ocean Data View (2021); https://odv.awi.de/. 42. P. Moffa-Sánchez, A. Born, I. R. Hall, D. J. R. Thornalley, S. Barker, Nat. Geosci. 7, 275–278 (2014). 43. W. Lu, D. W. Oppo, G. Gebbie, D. J. R. Thornalley, Planktic and benthic foraminiferal isotope and abundance records from the subpolar North Atlantic during the last 1,200 years, Zenodo (2023); https://doi.org/10.5281/zenodo.8429225. AC KNOWLED GME NTS
We thank the WHOI Seafloor Samples Repository for curating the samples, and WHOI NOSAMS for radiocarbon analyses. We thank K. Pietro and S. Wang for technical assistance. Funding: This work was funded by NSF grants OCE-2031929 (to G.G. and D.W.O.) and OCE-1304291 (to D.W.O. and D.J.R.T.) and by WHOI’s Edna McConnell Clark Foundation Fund (to D.W.O.). W.L. was supported by the WHOI Postdoctoral Scholar Program, with funding provided by the Weston Howland Jr. Postdoctoral Scholarship, and by NSF OCE-2114579 (to D.W.O.) Author contributions: Conceptualization: D.W.O. and G.G. Methodology: W.L., D.W.O., G.G. Investigation: W.L., D.W.O., G.G., D.J.R.T. Visualization: W.L. and D.W.O. Funding acquisition: D.W.O., G.G., D.J.R.T., W.L. Project administration: D.W.O. Supervision: D.W.O. and G.G. Writing – original draft: W.L. Writing – review and editing: W.L. and D.W.O., G.G., and D.J.R.T. Competing interests: The authors declare no competing interests. Data and materials availability: Source data for Figs. 1 to 3 and figs. S6 to S15 are provided in the supplementary materials. The raw radiocarbon, isotope data and R-scripts to reproduce the results are publicly available at the National Center for Environmental Data (https://www. ncei.noaa.gov/access/paleo-search/study/38185) and Zenodo repository (https://zenodo.org/record/8429225) (43). License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www. sciencemag.org/about/science-licenses-journal-article-reuse SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adf1646 Materials and Methods Supplementary Text Figs. S1 to S21 Table S1 Data S1 to S4 References (44–49) Submitted 3 October 2022; accepted 11 October 2023 10.1126/science.adf1646
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
839
RES EARCH | R E S E A R C H A R T I C L E S
INORGANIC CHEMISTRY
An all-metal fullerene: [K@Au12Sb20]5− Yu-He Xu1, Wen-Juan Tian2, Alvaro Muñoz-Castro3, Gernot Frenking4,5, Zhong-Ming Sun1* The C60 fullerene molecule has attracted tremendous interest for its distinctive nearly spherical structure. By contrast, all-metal counterparts have been elusive: Fullerene-like clusters composed of noncarbon elements typically suffer from instability, resulting in more compact geometries that require multiple embedded atoms or external ligands for stabilization. In this work, we present the synthesis of an all-metal fullerene cluster, [K@Au12Sb20]5−, using a wet-chemistry method. The cluster's structure was determined by single crystal x-ray diffraction, which revealed a fullerene framework consisting of 20 antimony atoms. Theoretical calculations further indicate that this distinct cluster exhibits aromatic behavior.
T
he study of all-metal clusters reveals the delicate balance between electronic shells and structural geometry: The quantum confinement of electrons (1, 2) gives rise to a rich diversity of atomic arrangements with intriguing bonding characteristics (3–5). The discovery of buckminsterfullerene (C60), which marked a major milestone in the exploration and application of stable threedimensional cages, has sprouted new research disciplines in chemical, physical, and material science (6, 7). The distinct near-spherical structure of fullerenes along with the surface of delocalized p electrons produces many notable properties and enables a wide range of applications in biology, medicine, electronics, and photovoltaics (8, 9). What’s more, the internal cavity of fullerenes provides a space for hosting a variety of atoms and molecules, giving rise to a class of endohedral clusters termed endofullerenes (10–13). The rapid progress in fullerene-related clusters and extensive applications of fullerene-based materials have
prompted the exploration of analogous hollow sphere molecules composed of other maingroup or transition metal elements known as inorganic fullerenes (14). In74 with D3h symmetry and In48Na12 with D3d symmetry are fullerene-like constructs found in the solidstate Zintl phase Na96In97Z2 (Z = Ni, Pd, Pt); both constitute the outermost shell of a fourlayer onion-like structure rather than existing as hollow cages (15). Additionally, theoretical calculations have predicted the stability of an all-gold fullerene Au32, structurally very similar to C60 (16). However, experiments produced an Au328+ cluster featuring a compact configuration of Au12@Au20, which was different from the previously anticipated fullerene, and the cationic cluster was protected by organic ligand (17, 18). Another ligand-protected dodecahedral silafullerane was also reported, which encapsulated one chloride ion (19). Obtaining ligand-free C60 analogs with heavier atoms may be constrained by their susceptibility to rearrangement into alternative, more stable
structures, as evidenced by previous theoretical studies. In this work, we report the isolation and characterization of an all-metal endohedral fullerene, [K@Au12Sb20]5− through a solution-based method in which only one K+ ion resides in a bare dodecahedral cage comprising 12 Au and 20 Sb atoms with distinct structural features. Each Au atom sits in the center of an Sb pentagonal plane without breaking the structure of Sb20 cage, but rather stretching the cage size. The [K@Au12Sb20]5− cluster is held together exclusively by Au–Sb bonds exploiting the icosahedron-dodecahedron duality, thereby retaining an icosahedral, near-spherical geometry with similar size to C60, but composed of 32 atoms. Synthesis and characterization
Compound [K(2,2,2-crypt)]5[K@Au12Sb20] was synthesized by reacting the Zintl phase K8SnSb4 with precursor Au(PPh3)Me in an ethylenediamine solution, which was facilitated by the presence of [2.2.2]crypt (see the supplementary materials). After stirring at room temperature for 7 hours, the color of the reaction
1 State Key Laboratory of Elemento-Organic Chemistry, Tianjin Key Lab of Rare Earth Materials and Applications, School of Materials Science and Engineering, Nankai University, Tianjin 300350, China. 2Institute of Molecular Science, Shanxi University, Taiyuan 030006, China. 3 Facultad de Ingeniería, Arquitectura y Diseño, Universidad San Sebastián, Bellavista 7, Santiago 8420524, Chile. 4 Institute of Advanced Synthesis, School of Chemistry and Molecular Engineering, State Key Laboratory of MaterialsOriented Chemical Engineering, Nanjing Tech University, Nanjing 211816, China. 5Fachbereich Chemie, PhilippsUniversität Marburg, Hans-Meerwein-Strasse 4, 35043 Marburg, Germany.
*Corresponding author. Email: [email protected]
Fig. 1. Molecular structure of the [K@Au12Sb20]5Ð cluster. (A) Thermal ellipsoid plot (50% probability) of the cluster. (B) Front side view of (A). (C) Space-filling representation of the crystal structure. (D) Top view of (A). (E) A typical Sb5 pentagon face centered by a gold atom in the cluster (with SbÐSb bond lengths marked). K is represented by cyan, Au by gold, and Sb by blue. 840
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
mixture was observed to change from brownred to brown-gold, and the stirring was stopped. The reaction vial was stored in a refrigerator at 10°C for about 5 days, and black block-shaped crystals were isolated from the bottom of the vial in a yield of 25% based on Au(PPh3)Me. The resulting complex was characterized by singlecrystal x-ray diffraction, which revealed its crystallization in the triclinic space group P-1. The asymmetric unit contained five [K(2,2,2-crypt)]+ charge-balancing cations. Additionally, energydispersive x-ray spectroscopy (EDS) was used to determine the elemental composition of the compound. The obtained atom values for Au and Sb were in good agreement with the theoretical values calculated for Au12Sb20. Many attempts were made to obtain mass spectral information, but to no avail. Because of the high negative charge of the cluster, it is extremely unstable in the air, and a distinct gray precipitate was observed in the crystal solution once exposed to air, which generated great difficulties in mass spectral detection. As shown in Fig. 1, the anion [K@Au12Sb20]5− exhibits the overall structure of a slightly flattened dodecahedron with each Sb5 pentagonal face centered by one Au atom. The average
short axis of the cage measures 7.30 Å (the distance between two opposing Au@Sb5 pentagonal faces) (Fig. 1, A to C), slightly exceeding the diameter of C60 (7.1 Å) (20), whereas the longest axis measures 9.03 Å (the distance between the two farthest Sb atoms on the cage) (Fig. 1D), which is comparable to the calculated diameter of the theoretically predicted Au32 (9.0 Å) (16). In all twelve planes, the sum of the angles between the central Au atom and the five vertices of the Sb5 ring is close to 360°, indicating that the gold atom lies on the surface of cyclo-Sb5 (in the Sb5 plane in Fig. 1E, for example, the sum of the angles is 359.9°). Such a planar pentacoordinate motif is unusual. Theoretical studies predicted a planar hexacoordinate carbon atom in the anion CB62−, which has not yet been synthesized (21). Similarly, an iron-centered planar cation [FeSb5]+ was predicted, in which the Sb5 ring is aromatic with equal-length Sb–Sb bonds of 2.973 Å (22). However, the Sb–Sb bond distances in [K@Au12Sb20]5− span a wide range from 3.114 to 3.436 Å (average of 3.227 Å), which are significantly longer than that of [FeSb5]+ as well as typical Sb–Sb single bonds (2.81 to 2.98 Å) (23–25), denoting distinct structural features.
Additionally, the Sb20 dodecahedral shell is expanded compared with that of the reported [Sb@Ni12@Sb20]− compounds (where the average Sb–Sb distance is 3.11 Å), potentially owing to the influence of the Au atom in the plane enlarging the Sb5 pentagon (26). All Au–Sb bonds are located on the faces of the dodecahedron, with a relatively narrow range of 2.698 to 2.798 Å, which is considerably longer than the bonds observed in [Au2Sb16]4− (2.67 to 2.71 Å) and [Sb3Au3Sb3]3− (2.59 to 2.61 Å) (27, 28). The central K+ ion is coordinated by twelve Au atoms, supporting their positioning on the Sb5 faces. The K–Au contacts range between 3.493 and 3.756 Å, indicating predominantly electrostatic interactions. Nevertheless, the presence of K+ as template cations remains crucial for the overall cluster stability. Theoretical analysis
To gain insights into the chemical bonding in the [K@Au12Sb20]5− cluster, theoretical analysis was conducted. The optimized structure for [K@Au12Sb20]5– revealed Au–Au distances of 4.002 Å, mediated by Au–Sb bonds of 2.773 Å, which compared well to the x-ray structure (Au–Au, 3.914 Ǻ; Au–Sb, 2.747 Ǻ). Comparison
Fig. 2. AdNDP bonding patterns and canonical molecular orbitals for [K@Au12Sb20]5−. (A) Sb 5s, Au 5d lone pairs, and the four-center two-electron (4cÐ2e) s bonds over each Sb2Au2 quadrilateral. ON, occupation number. (B) Selected canonical molecular orbitals with their superatomic features (S, P, and D) indicated. SCIENCE science.org
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
841
RES EARCH | R E S E A R C H A R T I C L E S
Fig. 3. Magnetic behavior for [K@Au12Sb20]5Ð. (A) Three-dimensional and (B) contour-plot representation of NICS and the induced magnetic field under certain orientations of the external field. (C) Streamline representation from GIMIC calculations of the current density over spheres with radii of 3.0, 4.5, and
between the calculated energies of a D5d-symmetry and Ih-symmetry structure indicated that the latter is favored by 5.4 kcal mol−1, suggesting that the experimentally characterized D5dstructure may be influenced by counterions and crystal packing effects. The resulting Au12 icosahedron enclosed an inner spherical cavity of diameter 7.491 Ǻ, which is significantly larger than the Au12 cage found in ligandprotected gold clusters of about 5.4 Å (29, 30). This suggests that the cage structure is supported by Au–Sb bonds, providing a larger interior volume, which thus presents a promising strategy for designing larger hollow clusters. The endohedral K+ atom was stabilized by a calculated encapsulation energy of –375.8 kcal mol−1, which was primarily driven by electrostatic interactions that accounted for 90% of the stabilizing forces (table S6). The calculated highest occupied molecular orbital–lowest occupied molecular orbital (HOMO-LUMO) gap amounted to 2.57 eV at the hybrid PBE0 level. Vibrational analysis denoted a bouncing motion for the endohedral K+ atom between 70 and 30 cm−1. A theoretical comparison between [K@Au12Sb20]5– and its hypothetical compact counterpart with Au–Au distances of 3.045 Å (fig. S7b) reveals an energetic preference for the characterized structure of 55.9 kcal mol−1. To effectively allocate the 238 valence electrons, we used the adaptive natural partitioning (AdNDP) analysis with the AdNDP 2.0 code (31). The advantage of using the AdNDP method is its capacity to elucidate the chem842
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
6.0 Å, with the 6.0-Å sphere given in side and top views, and a cut plane at the center of the spherical cluster that denotes the magnitude of current density vector field in nA/T. Isosurface values set at ±3 ppm. Blue represents shielding and red represents deshielding.
ical bonding arrangement, encompassing both Lewis bonding constituents [including lone pairs (1c2e) and two-center two-electron (2c2e) bonds] and delocalized bonding constituents. In addition to the 20 Sb 5s lone pairs and 60 Au 5d lone pairs, there are 30 pairs of 4c–2e s bonds distributed evenly over each butterflyshaped Au2Sb2 quadrilateral, covering the surface of the [K@Au12Sb20]5– cluster (Fig. 2A). The occupation numbers of these bonds range from 1.91 to 1.94 |e|. The remaining 18 electrons are allocated to nine orbitals with superatomic features (S, P, and D), satisfying the 3D aromatic requirement of 2(n + 1)2 (n = 2) (Fig. 2B). These electron distributions contribute to the overall stability and distinct properties of the cluster. The natural atomic orbitals analysis provided valuable insights into the contribution of each atom to the orbitals (32). Table S4 presents the total contribution of each atom to these orbitals, revealing that nearly all atoms make substantial contributions. The natural population analysis conducted on the optimized structure of [K@Au12Sb20]5– revealed that the central K atom carries a charge of +0.85 |e|, indicating the presence of electrostatic interactions between the inner K atom and the outer Au12Sb20 shell. Moreover, the detailed energy decomposition analysis results at the PBE0/STO-TZ2P-ZORA level, as shown in table S6, further support the ionic nature of the system. The analysis revealed that electrostatic interactions dominate the K+-cage bonding, contributing more significantly (DEelstat,
89.6%) compared with orbital interactions (DEorb, 7.4%) in attracting local charges and stabilizing the system. To evaluate the aromatic properties of [K@Au12Sb20]5–, the overall magnetic behavior was given by the three-dimensional representation of nucleus-independent chemical shift (NICS) which accounted for the orientationally averaged behavior resulting from the experimental molecular tumbling in solution (Fig. 3). The NICS isosurface exhibits a shielding contour at the spherical cage, which suggests a spherical aromatic behavior (33, 34). To overcome the NICS exaltation near to heavy nuclei, we focused our analysis on the long-range characteristics of the induced magnetic field at the low-electron density limit. (35) Moreover, the representation of the magnetic response under specific orientation of the external field (Bindi; i = x, y, z) provides a picture of the shielding and deshielding regions that account for the possible global aromatic characteristics in [K@Au12Sb20]5–. As a result, from Bindi, an enhanced long-range shielding region aligned to the applied field for different orientations was obtained, complemented with a deshielding region in a perpendicular plane, which accounts for the shielding cone property inherent to aromatic species (29, 36). The long-range shielding region exhibited calculated values of –20.0 ppm at 7.5 Å from the center of the structure, and of –2.6 ppm at 15.0 Å (Fig. 3b), thus supporting the spherical aromatic behavior of [K@Au12Sb20]5– and leading to an enhanced science.org SCIENCE
RESE ARCH | R E S E A R C H A R T I C L E S
shielding region. In addition, the current density upon a z-aligned external field was given from gauge-including magnetically induced currents (GIMIC) calculations, denoting a collective of parallel currents around the cluster that were observed at inner regions (3.0 Å of radius), at the structure contour (4.5 Å), and outside of the spherical shell. This analysis supports the formation of a long-range shielding region owing to the presence of aromatic currents upon an external field. Integration of the induced current strength denoted values of 9.8 nanoamperes per tesla (nA/T), contributed by +158.4 nA/T from diatropic and –148.6 nA/T from paratropic currents, which is sizable in comparison to the prototypical planar aromatic species given by benzene, with a value of 12.1 nA/T at the PBE0/def2-tvpz level. At the PBE0/LanL2DZ level, a value of 21.2 nA/T was obtained, denoting dependence of the level of theory. The Au–Sb heterobonds play a crucial role in maintaining the structural integrity of the cage, whereas the endohedral cation acts as a template for structure formation. Future investigations will focus on exploring alternative synthetic strategies that leverage the interplay between cage composition and endohedral templates, thereby enabling the rational and controlled synthesis of larger all-metal fullerenes. These superatoms hold great potential for the design and fabrication of precisely engineered nanostructures owing to their atomically precise near-spherical structures. RE FE RENCES AND N OT ES
1. A. W. Castleman Jr., S. N. Khanna, J. Phys. Chem. C 113, 2664–2675 (2009). 2. H. Häkkinen, Chem. Soc. Rev. 37, 1847–1859 (2008).
SCIENCE science.org
3. F. Lips, R. Clérac, S. Dehnen, Angew. Chem. Int. Ed. 50, 960–964 (2011). 4. J.-Q. Wang, S. Stegmaier, T. F. Fässler, Angew. Chem. Int. Ed. 48, 1998–2002 (2009). 5. M. Ruck, V. Dubenskyy, T. Söhnel, Angew. Chem. Int. Ed. 42, 2978–2982 (2003). 6. H. W. Kroto, J. R. Heath, S. C. Obrien, R. F. Curl, R. E. Smalley, Nature 318, 162–163 (1985). 7. W. Kratschmer, L. D. Lamb, K. Fostiropoulos, D. R. Huffman, Nature 347, 354–358 (1990). 8. K. Tanigaki et al., Nature 352, 222–223 (1991). 9. X. Lu, L. Feng, T. Akasaka, S. Nagase, Chem. Soc. Rev. 41, 7723–7760 (2012). 10. J. Zhao, Q. Du, S. Zhou, V. Kumar, Chem. Rev. 120, 9021–9163 (2020). 11. S. Aoyagi et al., Nat. Chem. 2, 678–683 (2010). 12. C. M. Beavers et al., J. Am. Chem. Soc. 133, 15338–15341 (2011). 13. S. Stevenson et al., Nature 401, 55–57 (1999). 14. M. J. Moses, J. C. Fettinger, B. W. Eichhorn, Science 300, 778–780 (2003). 15. S. C. Sevov, J. D. Corbett, Science 262, 880–883 (1993). 16. M. P. Johansson, D. Sundholm, J. Vaara, Angew. Chem. Int. Ed. 43, 2678–2681 (2004). 17. S. Kenzler et al., Angew. Chem. Int. Ed. 58, 5902–5905 (2019). 18. S.-F. Yuan, C.-Q. Xu, J. Li, Q.-M. Wang, Angew. Chem. Int. Ed. 58, 5906–5909 (2019). 19. J. Tillmann et al., Angew. Chem. Int. Ed. 54, 5429–5433 (2015). 20. W. I. F. David et al., Nature 353, 147–149 (1991). 21. K. Exner, P. von Ragué Schleyer, Science 290, 1937–1940 (2000). 22. M. Lein, J. Frunzke, G. Frenking, Angew. Chem. Int. Ed. 42, 1303–1306 (2003). 23. M. Westerhausen, S. Weinrich, P. Mayer, Z. Anorg. Allg. Chem. 629, 1153–1156 (2003). 24. H. Breunig, R. Rösler, Coord. Chem. Rev. 163, 33–53 (1997). 25. S. C. Critchlow, J. D. Corbett, Inorg. Chem. 23, 770–774 (1984). 26. Y. Wang et al., J. Am. Chem. Soc. 139, 619–622 (2017). 27. I. A. Popov et al., Angew. Chem. Int. Ed. 55, 15344–15346 (2016). 28. F.-X. Pan et al., J. Am. Chem. Soc. 137, 10954–10957 (2015). 29. X. Kang, H. Chong, M. Zhu, Nanoscale 10, 10758–10834 (2018).
30. H. Yang et al., Nat. Commun. 4, 2422–2429 (2013). 31. D. Y. Zubarev, A. I. Boldyrev, Phys. Chem. Chem. Phys. 10, 5207–5217 (2008). 32. T. Lu, F. Chen, J. Comput. Chem. 33, 580–592 (2012). 33. N. V. Tkachenko, I. A. Popov, A. Muñoz-Castro, Eur. J. Inorg. Chem. 41, 4239–4250 (2021). 34. N. Fedik, A. I. Boldyrev, A. Muñoz-Castro, Phys. Chem. Chem. Phys. 21, 25215–25219 (2019). 35. C. Foroutan-Nejad, Theor. Chem. Acc. 134, 8 (2015). 36. R. Islas, T. Heine, G. Merino, Acc. Chem. Res. 45, 215–228 (2012). AC KNOWLED GME NTS
We are grateful for the valuable discussions with X.-B. Wang (PNNL), J. Xu (NKU), and N. Li (Rigaku Beijing Co., Ltd.). Funding: This work was supported by the National Natural Science Foundation of China (nos. 92161102 and 22371140), the Natural Science Foundation of Tianjin City (no. 21JCZXJC00140), and 111 project (B18030) from China (MOE). A.M.-C. acknowledges financial support from ANID FONDECYT Regular 1221676. Author contributions: Conceptualization: Z.-M.S.; Methodology: A.M.-C. and Z.-M.S.; Investigation: Y.-H.X.; Visualization: Y.-H.X., W.-J.T., A.M.-C., and Z.-M.S.; Funding acquisition: A.M.-C. and Z.-M.S.; Project administration: Z.-M.S.; Supervision: Z.-M.S.; Writing – original draft: Y.-H.X, W.-J.T., A.M.-C., and Z.-M.S.; Writing – review and editing: Y.-H.X, W.-J.T., A.M.-C., G.F., and Z.-M.S. Competing interests: The authors declare no competing interests. Data and materials availability: X-ray data are available free of charge from the Cambridge Crystallographic Data Centre under reference numbers CCDC 2269174 (method 1) and 2291088 (method 2). All other experimental, spectroscopic, crystallographic, and computational data are included in the supplementary materials. License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/about/science-licenses-journalarticle-reuse SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.adj6491 Materials and Methods Figs. S1 to S9 Tables S1 to S6 References (37–53) Submitted 8 July 2023; resubmitted 3 September 2023 Accepted 4 October 2023 10.1126/science.adj6491
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
843
ScienceCareers.org
Science Careers helps you advance your career. Learn how !
Register for a free online account on ScienceCareers.org.
Search hundreds of job postings and find your perfect job.
Download our career booklets, including Career Basics, Careers Beyond the Bench, and Developing Your Skills.
Sign up to receive e-mail alerts about job postings that match your criteria.
Complete an interactive, personalized career plan at “my IDP.”
Upload your resume into our database and connect with employers.
Visit our Employer Profiles to learn more about prospective employers.
Read relevant career advice articles from our library of thousands.
Watch one of our many webinars on different career topics such as job searching, networking, and more.
Visit ScienceCareers.org today — all resources are free
SCI E N CECA RE E RS.O RG
1117Recruitment_RC.indd 844
11/14/23 9:09 AM
The Department of Biomedical Data Science (BMDS) and the Center for Precision Health & Artificial Intelligence (CPHAI) at the Geisel School of Medicine at Dartmouth invite applications for a tenure-track appointment at the rank of Assistant Professor. BMDS offers a dynamic and interactive environment with a commitment to research excellence. BMDS faculty members benefit from highly collaborative and collegial interactions across Dartmouth College and Dartmouth Health. CPHAI is a new initiative that will foster novel, interdisciplinary AI and machine learning research, advancing public health and healthcare delivery. As an Ivy League research university and the birthplace of AI, Dartmouth is a leading teaching and research institution in the United States, dedicated to finding solutions to the world’s most challenging problems and preparing students for leadership roles.
ScienceCareers.org
TENURE-TRACK FACULTY POSITION IN BIOMEDICAL DATA SCIENCE, PRECISION HEALTH & ARTIFICIAL INTELLIGENCE
Successful applicants will develop vigorous, innovative research programs focused on biomedical informatics, machine learning in healthcare, personalized medicine, learning healthcare systems, medical image analysis, clinical natural language processing, or multimodal healthcare data analytics, with the goal of having a practical impact on healthcare delivery and patient outcomes. Expertise in interpretable machine learning, ethical and equitable use of AI, and algorithmic bias mitigation is strongly encouraged. A generous start-up package, as well as access to state-of-the-art research facilities, will be provided. Individuals will be given opportunities for mentorship and teaching in relevant PhD and MS graduate programs (e.g., Program in Quantitative Biomedical Sciences). Opportunities for innovative research methods development and application are facilitated through Dartmouth’s high-performance computing resources (e.g., Discovery), access to Dartmouth’s Shared Resources, and interactions with existing NIH Centers for Biomedical Research Excellence (COBREs) at Dartmouth. Successful individuals will be provided a tenure-track faculty appointment in the Department of Biomedical Data Science at the Geisel School of Medicine at Dartmouth College. Secondary appointments in other departments at Dartmouth College and Dartmouth-Health may be available depending on expertise and experience. Candidates are expected to have a strong record of scholarship, to develop an independent research program, and to participate in graduate-level teaching. Evidence of an ability to secure extramural funding is desirable. Furthermore, candidates should have a PhD and/or MD (or equivalent) in a related field, and a current faculty appointment at the level of Assistant Professor will be preferred, but Associate Professors may also be considered. Dartmouth College is an equal opportunity employer with a strong commitment to diversity and inclusion. We prohibit discrimination on the basis of race, color, religion, sex, age, national origin, sexual orientation, gender identity or expression, disability, veteran status, marital status, or any other legally protected status. Applications by members of all underrepresented groups are encouraged. BMDS and CPHAI at Geisel, and Dartmouth are committed to fostering a diverse, equitable, and inclusive population of students, faculty, and staff. Dartmouth recently launched a new initiative, Toward Equity, that embraces shared definitions of diversity, equity, inclusion, and belonging as a foundation for our success in institutional transformation. We are especially interested in applicants who can work effectively with students, faculty, and staff from all backgrounds and with different identities and attributes. Applicants should upload a cover letter addressed to the Search Committee Chair, Professor Saeed Hassanpour, along with a CV, a 3-page research statement, a 1-page statement of teaching philosophy, a 1-page statement of contributions to diversity, and three letters of recommendation to apply.interfolio.com/134886. Screening of applications will begin December 1, 2023 and continue until the position is filled.
POSTDOCTORAL RESEARCH ASSOCIATES NRC Research Associateship Programs
POSTDOCTORAL RESEARCH FELLOWS Air Force Science & Technology Fellowship Program
The National Academies of Sciences, Engineering, and Medicine administers postdoctoral and senior research awards at participating federal laboratories and affiliated institutions at locations throughout the U.S and abroad.
The National Academies of Sciences, Engineering, and Medicine administers postdoctoral and senior fellowship awards at the U.S. Air Force Research Laboratory (AFRL), the U.S. Air Force Institute of Technology (AFIT), and the U.S. Air Force Academy (USAFA) under the Air Force Science & Technology Fellowship Program (AF STFP).
We are seeking highly qualified candidates who hold, or anticipate earning, a doctorate in a variety of fields of science or engineering. Degrees from foreign universities should be equivalent in training and research experience to a doctoral degree from a U.S. institution. Citizenship eligibility varies among the sponsoring laboratories. Application deadline dates (four annual review cycles): • February 1 • May 1 • August 1 • November 1 Awardees have the opportunity to: • Conduct independent research in an area compatible with the interests of the sponsoring laboratory • Devote full-time effort to research and publication • Access the excellent and often unique facilities of the federal research enterprise • Collaborate with leading scientists and engineers at the sponsoring laboratories Awardee benefits include: • Stipends ranging from $45,000 to $97,490; may be higher based on experience • Health insurance (including dental and vision), relocation benefits, and a professional travel allowance For detailed program information, to search Research Opportunities, and to contact prospective Research Adviser(s) visit www.nas.edu/rap.
1117Recruitment_RC.indd 845
We are seeking highly qualified candidates who are U.S. citizens and hold, or anticipate earning, a doctorate in a variety of fields of science or engineering. Application deadline dates (four annual review cycles): • February 1 • May 1 • August 1 • November 1 Awardees have the opportunity to: • Conduct independent research in an area compatible with the interests of the Air Force laboratories • Devote full-time effort to research and publication • Access the excellent and often unique Air Force research facilities • Collaborate with leading scientists and engineers Awardee benefits: • Base stipend starting at $76,542; may be higher based on experience • Health insurance (including dental and vision), relocation benefits, and professional travel allowance For detailed program information, to search for AFRL, AFIT, and USAFA Research Opportunities, and to contact prospective Research Adviser(s), visit www.nas.edu/afstfp.
11/14/23 9:09 AM
WORKING LIFE By Katie Suleta
Credentials aren’t everything
“Y
ou’re so smart! Why don’t you have a doctorate?” The familiar question was offered as a compliment, but it always felt like a slap in the face. I’d been working at the organization for 2 weeks, and everyone seemed to like the work I was doing, the perspective I brought, and the direction I envisioned for my team. But there was just one little problem: I hadn’t completed my doctorate. All the highly educated people I worked with seemed to know it—and they weren’t going to let me forget. I knew not having a doctorate didn’t make me any less capable. It didn’t mean I wasn’t smart enough to be in the room. In fact, I was proud that earlier in my career I had left an ill-fitting Ph.D. program I felt pushed into. But that missing doctorate still haunted me. I first thought about going to graduate school when I was an undergraduate and developed a taste for research by working in a psychology lab. I was hesitant about a Ph.D.: I didn’t want to commit to one esoteric area or become a professor, and I was concerned a Ph.D. would pigeonhole me and limit my career options. So, I opted for “just” a master’s. But once I graduated, I felt the weight of my mentors’ expectations. “What school do you want to go to? We can make it happen!” “You’re going to be a wonderful colleague one day.” I felt lucky to have such a supportive network, and I didn’t want to let down those who had invested so much time and energy into me and my development. So, I told myself I would give a Ph.D. a try. From the start, it wasn’t a good fit. I wasn’t passionate about what I was researching. I wasn’t excited by the prospect of a career in the field. I felt I had stumbled into a world where I didn’t belong. I shouldn’t have been surprised when I didn’t pass my preliminary exam. After 3 years, I left the program and set out to explore what I could do with a strong background in infectious disease epidemiology but no doctorate. I found a plethora of opportunities: conducting HIV/ AIDS research in the United States and Kenya, digging into the nuances of medical claims and Medicaid payments at a nonprofit health plan, teaching medical residents about research. A sense that I had unfinished business sometimes nagged at me, as the unmet expectations of overachievers can. But for the most part, I was at peace with my choice. Then the pandemic happened. I quickly encountered more job opportunities than I ever imagined—amazing jobs that I wanted, in health care and big data, seats at the table where big decisions are made. But there it was
again, that one problem: no doctorate. I interviewed with multiple people and organizations and they always wanted to know, “Why not?” Although I didn’t regret my decision all those years ago to leave my Ph.D., I realized that the kinds of jobs I wanted and was otherwise qualified for required the degree. At this point, it was 10 years since my initial foray into the land of doctoral studies. I knew myself better, I had specific goals, and I had a better idea of what I would want out of a doctorate. So, I bit the bullet and decided to pursue a doctorate once more—on my terms this time. I am now in my final year, working toward a doctorate of health sciences while maintaining my day job. It’s a professional research degree that will qualify me for the types of leadership positions I ultimately want, and offers more versatility and flexibility than a Ph.D. I don’t know where I’ll end up— but that’s kind of the point. I’ve always wanted to explore a range of professional opportunities and experiences, and I think this doctorate will open doors rather than pigeonhole me. I’m proud of the work I’ve done for my doctorate, and that I did it when and how it was right for me. And I’m excited about my future, whatever it may hold. But even though the doctorate qualifies me for some opportunities that were previously out of reach, I know that doesn’t mean I wasn’t capable or worthy before I had it. Credentials aren’t everything, and everyone deserves respect— regardless of their degree, or lack thereof. j
846
17 NOVEMBER 2023 • VOL 382 ISSUE 6672
1117WorkingLife_18241975.indd 846
ILLUSTRATION: ROBERT NEUBECKER
“Not having a doctorate didn’t make me any less capable.”
Katie Suleta is a DHSc candidate at George Washington University and regional director of research in graduate medical education for HCA Healthcare. Send your career story to [email protected].
science.org SCIENCE
11/10/23 4:46 PM
The 2024 AAAS Annual Meeting will take place in person in Denver at the Colorado Convention Center, February 15-17. See meeting program and register today!
aaas.org/meetings
1117Product.indd 847
|
#AAASmtg
11/9/23 8:21 AM
science.org/journal/stm
PUT HUMAN HEALTH AT THE HEART OF YOUR RESEARCH Submit your research: cts.ScienceMag.org
Twitter: @ScienceTM Facebook: @ScienceTranslationalMedicine
1117Product.indd 848
11/9/23 8:21 AM