How the threat of Brexit has NextGen Voices: Ask Approaching autism from

changed researchers’ lives p. 24 a peer mentor p. 28 the outside-in p. 45

$15
4 OCTOBER 2019
SPECIAL ISSUE
sciencemag.org

LANGUAGE
AND THE BRAIN
 2019-2020
SCIENTIFIC
CONFERENCES
Presenting the most significant research on cancer etiology, prevention, diagnosis, and treatment

Tumor Immunology and Immunotherapy The Evolving Landscape of Cancer Modeling

Conference Cochairs: Timothy A. Chan, Conference Cochairs: Cory Abate-Shen, Andrea
Charles G. Drake, Marcela V. Maus, Califano, Jos Jonkers, and Calvin J. Kuo
and Arlene H. Sharpe March 2-5, 2020 | San Diego, CA
November 17-20, 2019 | Boston, MA
EACR-AACR Basic and Translational Research
San Antonio Breast Cancer Symposium Conference: Tumor Microenvironment
Codirectors: Carlos L. Arteaga, In partnership with ASPIC
Virginia G. Kaklamani, and C. Kent Osborne (Portuguese Association for Cancer Research)
December 10-14, 2019 | San Antonio, TX Scientific Committee Cochairs: Carlos M. Caldas,
Luís Costa, and Lisa M. Coussens
Advancing Precision Medicine
March 2-4, 2020 | Lisbon, Portugal
Drug Development: Incorporation of
Real World Data and Other Novel Strategies
Evolutionary Dynamics in Carcinogenesis
Conference Cochairs: David M. Hyman,
and Response to Therapy
Elaine R. Mardis, Lillian L. Siu,
Conference Cochairs: James DeGregori,
and Eliezer M. Van Allen
Marco Gerlinger, Robert Gillies, and Andriy Marusyk
January 9-12, 2020 | San Diego, CA
March 12-15, 2020 | Denver, CO
Sixth AACR-IASLC International Joint
Conference: Lung Cancer Translational Advances in Prostate Cancer Research
Science from the Bench to the Clinic Conference Cochairs: Felix Y. Feng,
Conference Cochairs: Trever G. Bivona, Karen E. Knudsen, and Scott A. Tomlins
John V. Heymach, Christine M. Lovly, March 12-15, 2020 | Denver, CO
and Katerina A. Politi
January 11-14, 2020 | San Diego, CA AACR Annual Meeting 2020
Program Committee Chair: Antoni Ribas
Advances in Liquid Biopsies April 24-29, 2020 | San Diego, CA
Conference Cochairs: Luis A. Diaz Jr.,
Maximilian Diehn, Irene M. Ghobrial, Second AACR International Meeting:
and Nicholas C. Turner Advances in Malignant Lymphoma:
January 13-16, 2020 | Miami, FL Maximizing the Basic-Translational
Interface for Clinical Application
The Microbiome and Viruses In cooperation with the International Conference
Conference Cochairs: Cynthia L. Sears, on Malignant Lymphoma (ICML)
Giorgio Trinchieri, Jennifer A. Wargo, Scientific Committee Chair: Ari M. Melnick
and Laurence Zitvogel June 25-28, 2020 | Boston, MA
February 21-24, 2020 | Orlando, FL

Learn more and register at

AACR.org/Calendar

1004Product.indd 2 9/25/19 1:11 PM

 R I S I N G S T A R AWA R D
I N B R A I N S C I E N C E T E C H N O L O GY

I N T E R N AT I O N A L C A L L F O R S U B M I S S I O N S
The Mahoney Institute for Neurosciences (MINS) at the University of Pennsylvania is proud to announce
its annual call for submissions for the Rising Star Award in neuroscience research.

The Rising Star Award celebrates the convergence on Penn’s campus of multiple disciplines – from advanced imaging
and computation to bioengineering – in support of the “Year of Brain Science Technology,” and it honors a young
researcher for outstanding contributions to Brain Science Technology with a USD 10,000 honorarium at the
36th Annual MINS Retreat and Symposium on April 28th, 2020.

“Technological advances often precipitate paradigm shifts in knowledge. Novel methods for measuring the living brain with ever
higher spatial and temporal resolution are providing new insights into brain organization and function. The 2020 MINS Rising
Star Award will honor a researcher whose focus on the latest acquisition and analysis methods is expanding the frontiers of
basic and clinical neuroscience research.”
– John A. Detre, M.D., Professor and Vice Chair for Research,
Department of Neurology, and Director of the new Brain Science Center at Penn

Eligibility • Researchers who received their first advanced degree, such as the Ph.D., in 2006 or more recently are invited to submit an entry.
• Winner must be able to travel to the University of Pennsylvania to present a research seminar.
Award • In addition to the USD 10,000 personal honorarium, the award recipient will present a research seminar at the symposium.
• Also at the symposium, Polina Anikeeva, Ph.D., of MIT will give the Sprague Lecture, and Anthony Zador, M.D., Ph.D.,
of the Cold Spring Harbor Laboratory, will give the Adler Lecture.
Submission • Applicants should submit a single PDF file with a one-page description of their contribution to Brain Science Research,
full curricula vitae, and the names of three references.
• The deadline for submissions is December 10, 2019 to MINSRisingStarAward@lists.upenn.edu.
More information can be found at www.med.upenn.edu/ins.

© 2019 University of Pennsylvania - MINS

1004Product.indd 3 9/25/19 1:11 PM

 Advertorial

Nankai’s new campus

Nankai’s president reflects on the future

Just a year and a half after taking up his position as president of Nankai been published 12,564 times and cited 210,847 times in the world’s most
University, former president of the Chinese Academy of Medical Sciences consistently high-impact journals.
Cao Xuetao has no doubt he made the right decision. “This makes Nankai the most highly cited university in China in
In January 2019, President Xi Jinping visited the university’s campus cumulative terms, as we average 16 citations per paper,” says Cao.
in Tianjin, northern China. He toured Nankai’s 100-year anniversary As a comprehensive research university, Nankai’s 26 colleges cover a
exhibition, which showcased its historic achievements as a center of balanced representation of humanities and science disciplines, including
learning, and delivered a speech emphasizing the importance of both basic the natural sciences, law, medicine, and art. Among these subjects,
and cutting-edge research. It seems that Cao’s ambition to transform Nankai seven disciplines stand out as top performers in their fields: chemistry,
into a state-of-the art research hub that attracts global talent is in line with mathematics, history, economics, statistics, materials science, and business
the President’s vision. administration.
Deeply aware of the university’s distinctive place in China’s educational In September 2017, Nankai was chosen to become part of a nationwide
system, Cao plans to build on Nankai’s reputation for excellence with higher-education initiative called Double First Class—a program to
a series of reforms, including the establishment of 10 interdisciplinary accelerate the rate at which China’s top universities develop their core areas
research centers and 10 international research centers that partner with of expertise and improve the quality of their research, in order to become
overseas institutes. world-class educational and research hubs.
“President Xi praised one of our founders, Zhang Boling, who was Nankai has “12 research areas ranked in the top 1% of their fields
influential in developing higher education in China and provided the basis worldwide according to Essential Science Indicators,” says Cao, including
for the development of China’s new university system. I want to continue chemistry, mathematics, materials science, physics, engineering,
that visionary tradition,” says Cao. environmental science and ecology, biology and biochemistry, molecular
biology and genetics, clinical and medicine, pharmacology and toxicology,
A proud history agricultural sciences, plant and animal science.
Nankai University was founded in 1919 as one of China’s first private “During President Xi’s visit, he praised the university as an institution
universities by education pioneers Yan Xiu and Zhang Boling, and produced that cultivates leading academic talent, and offers a delightful academic
the republic’s first premier, Zhou Enlai. Famous alumni have also included environment and an impressive, patriotic education. He also encouraged
the winner of the Wolf Prize in Mathematics Shiing-Shen Chern, playwright professors and students to work for a greater cause and to step up their
Cao Yu, and aerodynamics scientist Guo Yonghuai. endeavor to help Nankai become a world-class university. This is precisely
In 1937, the campus was bombed by the Japanese army, and the faculty the vision I hope to support,” Cao says.
was forced to relocate to Kunming in southern China, where it merged with Nankai has always been known for its dedication to serving the wider
Peking University and Tsinghua University to form the National Southwest needs of China and its people, adds Cao. In the 1950s, university president
Associated University until 1946, when it returned to Tianjin. Yang Shixian changed his research focus from pharmaceutical chemistry
“The eight years in Kunming was a proud period in Nankai’s history: Not to pesticides to help ensure the country’s food security during a period of
only was education preserved in wartime, but the university also nurtured economic uncertainty.
a large number of important and influential leaders and scholars who
PHOTO: JIANG BINGJUN
contributed significantly to their fields and to society,” says Cao. Looking to a bright future
After the establishment of the People’s Republic of China in 1949, Cao has ambitious plans for the university that start with launching
Nankai reopened as a comprehensive university. the 4211 New Nankai Excellence Initiative—a program to bolster the
development of four research areas: social sciences, natural sciences,
Growing influence engineering, and biomedical sciences.
Over the past decade, the university’s reputation has continued to In terms of engineering, during the past two years, Nankai has launched
grow. According to the Science Citation Index, Nankai’s researchers have four new engineering schools in the areas of artificial intelligence,

1004Product.indd 4 10/1/19 11:55 AM

 Advertorial

A global network
A cornerstone of Cao’s planned reforms is to build Nankai into a global
academic leader, with the launch of his “Global Nankai” initiative. To ensure
the university’s visibility on the world stage, research will focus on issues of
global importance; the university will establish a global network of research
partnerships; and staff and students will be encouraged to become more
globally minded.
An essential element of Cao’s plan is to encourage more prominent
scholars, academic leaders, and talented scientists from abroad to work and
study at Nankai.
A new program called “Studying in Nankai” aims to increase the
number of international students. There are also plans to enhance
the attractiveness of campus culture by improving salary levels for
faculty, enhancing the quality of research facilities, and improving the
management of everyday services.
“We will make this a world-class university, combining Nankai’s
character and Chinese attributes,” he says.
Cao went through some basics of China’s higher education system with
us. In recent decades, the system has expanded quickly, with dramatic
results. The government is investing 4% of GDP per year in education. The
number of students in higher education is now 37.8 million—one-fifth of
the total number of students in the world. Between 2012 and 2018, the
number of Chinese universities ranked in the top 500 by Times Higher
Education increased from 10 to 72, while the number of universities grew
from 91 to 246.
“Now China’s higher education is developing from a model of scale
expansion to a model of quality improvement and structure optimization,”
Cao Xuetao, President says Cao.
He points to a recent collaboration with the University of Glasgow as an
example of how to create a global academic atmosphere that produces first-
rate students who are equipped to contribute to society at large. The newly
established two-year postgraduate dual Master’s qualification provides
students with degrees from both universities.
cyberspace security, statistics and data science, and materials science. In
May 2019, the faculty set up the CAMLiNK station—a collaboration between Anniversary celebrations
Nankai University and the University of Cambridge that aims to attract In October 2019, Nankai University will celebrate its 100-year
talent from the United Kingdom to work more closely with their Chinese anniversary. Alumni from all over the world will return to Tianjin to
counterparts. This new partnership marks the deepening of research celebrate the university’s achievements with a series of lectures, workshops,
relations between the two famous universities. A three-day international and social events, during what Cao believes will be a remarkable occasion.
seminar launched the research platform and included talks on subjects Since joining the university, Cao says he’s learned a great deal from his
such as solar energy and lithium battery materials, photoelectric materials, talented colleagues.
and the application of artificial intelligence in materials science. “I’ve fallen in love with Nankai,” he says. “I’ve learned so much, not only
In the area of biomedical research, the university has forged an alliance about its history but also about the research being done here. Our faculty
with 12 hospitals in Tianjin City. It is also in the process of setting up members are the most important asset of this university, and I want to
international joint research centers in the fields of medicine and biological create an academic atmosphere that is conducive to top-quality research
sciences with the University of Oxford and the University of Birmingham, and worthy of their talents.”
respectively.
Cao, who is also an immunologist, says he wants to create a state-of-
the-art pharmaceutical R&D system at Nankai: “I’m interested in fostering
a culture of interdisciplinary health research and integrating teaching,
research, and medical practice.” Sponsored by Produced by the Science/AAAS
Custom Publishing Office
PHOTO: WU JUNHUI

Cao also feels that the university could benefit from the cross-
fertilization of ideas between the liberal arts and natural sciences. Moving
forward, the university will focus on “macro-research topics,” such as
ecological civilization, big data, and media convergence, he says.
“These specialized topics will make the most of Nankai’s status as a
comprehensive university,” he observes.

1004Product.indd 5 10/1/19 11:55 AM

 Gain unique insights
for your scientific
discoveries.

ZEISS Microscopes for Life Sciences

Choose the ideal imaging technologies for your scientific question. The unique
ZEISS portfolio of light, electron, ion and X-ray microscopes lets you combine
complementary views to gain new insights into your sampleís biology.

zeiss.com/life-sciences

1004Product.indd 6 9/25/19 1:11 PM

 CONTENTS
4 O C T O B E R 2 01 9 • VO LU M E 3 6 6 • I S S U E 6 4 6 1
FEATURES
24 Split decisions
How Brexit has already taken a toll on five
researchers By E. Stokstad

INSIGHTS
LETTERS
28 NextGen advises “Trying to Manage”

PERSPECTIVES
31 Observing the cosmic web
The faint signature of gas filaments in the
intergalactic medium is finally detected
By E. Hamden
REPORT p. 97

32 RNA nucleosides built in

one prebiotic pot
Origins-of-life research models integrate
the synthesis of RNA building blocks
By N. V. Hud and D. M. Fialho
RESEARCH ARTICLE p. 76

33 Learning birdsong by imitation

Transforming sensory information into vocal
imitation allows young finches to sing
By D. F. Clayton
LANGUAGE AND THE BRAIN SECTION p. 48;
RESEARCH ARTICLE p. 83

35 Remodeling the genome with

24 DNA twists
Intricate protein machines repackage
DNA to turn genes on and off
By G. D. Bowman and S. Deindl

NEWS 22 Europe seeks broader support

and impact for science projects 37 Donald A. B. Lindberg
CREDITS: (ILLUSTRATION) DAVIDE BONAZZI/SALZMANART; (PHOTO) STEVE GSCHMEISSNER/SCIENCE SOURCE

The European Commission is asking the (1933–2019)

public to help design its new €94 billion Visionary medical informaticist who
IN BRIEF research program By T. Rabesandratana laid the groundwork for PubMed
By I. Kohane and J. M. Berg
14 News at a glance 23 DNA pushes back the
microbiome frontier POLICY FORUM
IN DEPTH New ways to culture microbes and decipher
18 Parched peatlands fuel their chemistries make invisible world 38 Collaborations and capacities
Indonesia’s blazes tangible By J. Cohen and E. Pennisi to transform fire management
RESEARCH ARTICLE BY Y. SUGIMOTO ET AL. Progress requires attention to governance
Policies to preserve and restore
10.1126/science.aax9176 at multiple levels
waterlogged forest help mitigate fires,
By C. A. Schultz and C. Moseley
scientists say By D. Normile
BOOKS ET AL.
19 NIH reveals its formula for
tracking foreign influences
Official highlights China’s talents program
23 42 Justice and genes
A new book offers an introduction to the
as agency watchdog calls for better vetting ethical dimensions of germline gene editing
of peer reviewers By J. Mervis By A. Hayden

20 Duplicated images point to fraud 43 The physicist and the dawn

in fish study, critics say of the double helix
A collage raises doubts about a 2014 behavior Erwin Schrödinger’s prescient musings
paper whose first author fabricated data in an on molecular biology turn 75
earlier study By M. Enserink By K. Sigmund

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 7

1004TOC.indd 7 10/1/19 5:04 PM

 (Actual size, 100 μl)

One and done

ALZET® Osmotic Pumps are small infusion pumps inserted at the
beginning of your study and then, that’s it. There’s no dosing
schedule. No repeated injections. Just a continuous, predictable
release of your test agent for the entire study. Days. Nights. Even
weekends. Anywhere from one day to six weeks. With no external
connections to monitor and less need for handling, it’s less stress
for you and your lab animals. Learn more about ALZET Osmotic
Pumps and just how needless injections can be at alzet.com.

VISIT US AT
EXHIBIT
#904
Find out more at ©2019 DURECT Corporation
alzet.com All Rights Reserved

1004Product.indd 8 9/26/19 11:22 AM

 CONTE NTS

PRIZE ESSAY
45 Outside-in SPECIAL SECTION
Rethinking the etiology of autism spectrum
disorders By L. Orefice
Language and the brain
RESEARCH INTRODUCTION
48 More than a tool for communication
ON THE COVER
Human beings are
social animals, and
REVIEWS language is central
IN BRIEF to our existence.
50 Evolution of vocal learning and
68 From Science and other journals spoken language E. D. Jarvis Both the large
human brain and the
55 The neurobiology of language language it enables
RESEARCH ARTICLES
beyond single-word processing have influenced
71 Wildlife trade P. Hagoort our lives, but which
Global wildlife trade across the tree has more effectively shaped the world
of life B. R. Scheffers et al. 58 From speech and talkers to the around us? This special issue highlights
social world: The neural processing some recent insights into the evolution
76 Prebiotic chemistry of human spoken language of language, the brain structures involved
Unified prebiotically plausible synthesis S. K. Scott in the production and interpretation of
language, and how this flexible function
of pyrimidine and purine
62 The neural basis of combinatory creates meaning. See page 48.
RNA ribonucleotides S. Becker et al.
syntax and semantics L. Pylkkänen Illustration: Eiko Ojala
PERSPECTIVE p. 32

83 Song learning EDITORIAL p. 13 PERSPECTIVE p. 33 RESEARCH ARTICLE p. 83

Inception of memories that
guide vocal learning in the songbird
W. Zhao et al.
PERSPECTIVE p. 33; LANGUAGE
AND THE BRAIN SECTION p. 48; PODCAST
93 Atomic physics 120 Biodiversity loss
Seconds-scale coherence on an Decline of the North American avifauna
optical clock transition in a tweezer array K. V. Rosenberg et al.
REPORTS
M. A. Norcia et al.
90 Star formation 124 Forest ecology
Gas flow and accretion via spiral 97 Intergalactic medium Differential soil fungus accumulation and
streamers and circumstellar disks in Gas filaments of the cosmic web located density dependence of trees in a subtropical
a young binary protostar around active galaxies in a protocluster forest L. Chen et al.
F. O. Alves et al. H. Umehata et al.
PERSPECTIVE p. 31 128 Lipid metabolism
A bacterial light response reveals an
orphan desaturase for human plasmalogen
31 & 97 100 Respiratory enzymes
Active site rearrangement and structural
divergence in prokaryotic respiratory
synthesis A. Gallego-García et al.

oxidases S. Safarian et al. 132 Quantum physics

Satellite testing of a gravitationally induced
105 3D printing quantum decoherence model P. Xu et al.
Scalable submicrometer additive
manufacturing S. K. Saha et al.
DEPARTMENTS
109 Structural biology 13 Editorial
Cryo-EM structure of a dimeric B-Raf:14-3-3 Language and the brain By Lera Boroditsky
complex reveals asymmetry in the active LANGUAGE AND THE BRAIN SECTION p. 48
sites of B-Raf kinases
Y. Kondo et al. 146 Working life
Review with care By Adriana L. Romero-Olivares
116 Systems biology
Stochastic antagonism between two Science Staff ................................................ 10
Galaxies (colors) and hard-to-detect gas filaments proteins governs a bacterial cell fate switch New Products ............................................. 136
(blue) observed with telescopes N. D. Lord et al. Science Careers ......................................... 137
IMAGE: HIDEKI UMEHATA

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 © 2019 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): $1971; Foreign postage extra: Mexico, Caribbean (surface mail) $55; other countries (air
assist delivery): $98. First class, airmail, student, and emeritus rates on request. Canadian rates with GST available upon request, GST #125488122. Publications Mail Agreement Number 1069624. Printed in the U.S.A.
Change of address: Allow 4 weeks, giving old and new addresses and 8-digit account number. Postmaster: Send change of address to AAAS, P.O. Box 96178, Washington, DC 20090–6178. Single-copy sales: $15 each plus shipping and
handling; bulk rate on request. Authorization to reproduce material for internal or personal use under circumstances not falling within the fair use provisions of the Copyright Act can be obtained through the Copyright Clearance Center
(CCC), www.copyright.com. The identification code for Science is 0036-8075. Science is indexed in the Reader’s Guide to Periodical Literature and in several specialized indexes.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 9

1004TOC.indd 9 10/1/19 5:04 PM

 BOARD OF REVIEWING EDITORS (Statistics board members indicated with S)
Adriano Aguzzi, U. Hospital Zürich Thomas Langer, U. of Cologne
Takuzo Aida, U. of Tokyo Mitchell A. Lazar, U. of Penn.
Leslie Aiello, Wenner-Gren Foundation Ottoline Leyser, U. of Cambridge
Editor-in-Chief Jeremy Berg, jberg@aaas.org Judith Allen, U. of Manchester Wendell Lim, U. of California, San Francisco
Sebastian Amigorena, Institut Curie Marcia C. Linn, U. of California, Berkeley
Executive Editor Monica M. Bradford James Analytis, U. of California, Berkeley Jianguo Liu, Michigan State U.
Editors, Research Valda Vinson, Jake S. Yeston Editor, Insights Lisa D. Chong Paola Arlotta, Harvard U.
Luis Liz-Marzán, CIC biomaGUNE
DEPUTY EDITORS Julia Fahrenkamp-Uppenbrink (UK), Stella M. Hurtley (UK), Phillip D. Szuromi, Sacha Vignieri SR. EDITORIAL FELLOW Johan Auwerx, EPFL
Jonathan Losos, Washington U. in St. Louis
David Awschalom, U. of Chicago
Andrew M. Sugden (UK) SR. EDITORS Gemma Alderton (UK), Caroline Ash (UK), Brent Grocholski, Pamela J. Hines, Paula A. Ke Lu, Chinese Acad. of Sciences
Clare Baker, U. of Cambridge
Kiberstis, Marc S. Lavine (Canada), Steve Mao, Ian S. Osborne (UK), Beverly A. Purnell, L. Bryan Ray, H. Jesse Smith, Jelena Stajic, Nenad Ban, ETH ZÜrich Christian Lüscher, U. of Geneva
Peter Stern (UK), Valerie B. Thompson, Brad Wible, Laura M. Zahn ASSOCIATE EDITORS Michael A. Funk, Priscilla N. Kelly, Tage S. Rai, Franz Bauer, Pontificia Universidad Católica de Chile Fabienne Mackay, U. of Melbourne
Seth Thomas Scanlon (UK), Keith T. Smith (UK), Yury V. Suleymanov LETTERS EDITOR Jennifer Sills LEAD CONTENT PRODUCTION Ray H. Baughman, U. of Texas at Dallas Anne Magurran, U. of St. Andrews
EDITORS Harry Jach, Lauren Kmec CONTENT PRODUCTION EDITORS Amelia Beyna, Jeffrey E. Cook, Chris Filiatreau, Nida Masiulis, Peter Bearman, Columbia U. Oscar Marín, King’s College London
Suzanne M. White SR. EDITORIAL COORDINATORS Carolyn Kyle, Beverly Shields EDITORIAL COORDINATORS Aneera Dobbins, Joi S. Carlo Beenakker, Leiden U. Charles Marshall, U. of California, Berkeley
Granger, Jeffrey Hearn, Lisa Johnson, Maryrose Madrid, Ope Martins, Shannon McMahon, Jerry Richardson, Alana Warnke, Alice Yasmine Belkaid, NIAID, NIH Christopher Marx, U. of Idaho
Whaley (UK), Anita Wynn PUBLICATIONS ASSISTANTS Jeremy Dow, Alexander Kief, Ronmel Navas, Hilary Stewart (UK), Brian White Philip Benfey, Duke U. Geraldine Masson, CNRS
EXECUTIVE ASSISTANT Jessica Slater ASI DIRECTOR, OPERATIONS Janet Clements (UK) ASI SR. OFFICE ADMINISTRATOR Jessica Waldock (UK) Gabriele Bergers, VIB C. Robertson McClung, Dartmouth College
Bradley Bernstein, Mass. General Hospital Rodrigo Medellín, U. of Mexico
News Editor Tim Appenzeller Alessandra Biffi, Harvard Med. School Graham Medley, London School of Hygiene &
Peer Bork, EMBL
NEWS MANAGING EDITOR John Travis INTERNATIONAL EDITOR Martin Enserink DEPUTY NEWS EDITORS Elizabeth Culotta, Lila Guterman, Tropical Med.
Chris Bowler, École Normale Supérieure
David Grimm, Eric Hand (Europe), David Malakoff SR. CORRESPONDENTS Daniel Clery (UK), Jon Cohen, Jeffrey Mervis, Elizabeth Ian Boyd, U. of St. Andrews
Jane Memmott, U. of Bristol
Pennisi ASSOCIATE EDITORS Jeffrey Brainard, Catherine Matacic NEWS REPORTERS Adrian Cho, Jennifer Couzin-Frankel, Jocelyn Edward Miguel, U. of California, Berkeley
Emily Brodsky, U. of California, Santa Cruz
Kaiser, Kelly Servick, Robert F. Service, Erik Stokstad (Cambridge, UK), Paul Voosen, Meredith Wadman INTERN Alex Fox Ron Brookmeyer, U. of California, Los Angeles (S) Tom Misteli, NCI, NIH
CONTRIBUTING CORRESPONDENTS Warren Cornwall, Ann Gibbons, Mara Hvistendahl, Sam Kean, Eli Kintisch, Kai Kupferschmidt Christian Büchel, UKE Hamburg Yasushi Miyashita, U. of Tokyo
(Berlin), Andrew Lawler, Mitch Leslie, Eliot Marshall, Virginia Morell, Dennis Normile (Shanghai), Elisabeth Pain (Careers), Dennis Burton, Scripps Research Alison Motsinger-Reif, NC State U. (S)
Charles Piller, Michael Price, Tania Rabesandratana (Barcelona), Emily Underwood, Gretchen Vogel (Berlin), Lizzie Wade Carter Tribley Butts, U. of California, Irvine Daniel Nettle, Newcastle U.
(Mexico City) CAREERS Donisha Adams, Rachel Bernstein (Editor), Katie Langin COPY EDITORS Julia Cole (Senior Copy Editor), György Buzsáki, New York U. School of Med. Daniel Neumark, U. of California, Berkeley
Cyra Master (Copy Chief) ADMINISTRATIVE SUPPORT Meagan Weiland Blanche Capel, Duke U. Beatriz Noheda, U. of Groningen
Annmarie Carlton, U. of California, Irvine Helga Nowotny, Austrian Council
Creative Director Beth Rakouskas Lars-Erik Cederman, ETH Zürich Rachel O’Reilly, U. of Warwick
Nick Chater, U. of Warwick Harry Orr, U. of Minnesota
DESIGN MANAGING EDITOR Marcy Atarod GRAPHICS MANAGING EDITOR Alberto Cuadra PHOTOGRAPHY MANAGING EDITOR William Douthitt
Zhijian Chen, UT Southwestern Med. Ctr. Pilar Ossorio, U. of Wisconsin
WEB CONTENT STRATEGY MANAGER Kara Estelle-Powers SENIOR DESIGNER Chrystal Smith DESIGNER Christina Aycock
Ib Chorkendorff, Denmark TU Andrew Oswald, U. of Warwick
GRAPHICS EDITOR Nirja Desai INTERACTIVE GRAPHICS EDITOR Xing Liu SENIOR SCIENTIFIC ILLUSTRATORS Valerie Altounian, Chris Bickel James J. Collins, MIT
SCIENTIFIC ILLUSTRATOR Alice Kitterman SENIOR GRAPHICS SPECIALISTS Holly Bishop, Nathalie Cary SENIOR PHOTO EDITOR Emily Petersen
Isabella Pagano, Istituto Nazionale di Astrofisica
Robert Cook-Deegan, Arizona State U.
Margaret Palmer, U. of Maryland
Alan Cowman, Walter & Eliza Hall Inst.
Elizabeth Levy Paluck, Princeton U.
Carolyn Coyne, U. of Pittsburgh
Interim Chief Executive Officer and Executive Publisher Alan Leshner Roberta Croce, VU Amsterdam Jane Parker, Max Planck Inst. Cologne
Jeff L. Dangl, U. of North Carolina Giovanni Parmigiani, Dana-Farber Cancer Inst. (S)
Publisher, Science Family of Journals Bill Moran Tom Daniel, U. of Washington Samuel Pfaff, Salk Inst. for Biological Studies
DIRECTOR, BUSINESS SYSTEMS AND FINANCIAL ANALYSIS Randy Yi DIRECTOR, BUSINESS OPERATIONS & ANALYSIS Eric Knott DIRECTOR OF Chiara Daraio, Caltech Julie Pfeiffer, UT Southwestern Med. Ctr.
ANALYTICS Enrique Gonzales MANAGER, BUSINESS OPERATIONS Jessica Tierney SENIOR BUSINESS ANALYST Cory Lipman, Meron Kebede Nicolas Dauphas, U. of Chicago Matthieu Piel, Institut Curie
FINANCIAL ANALYST Alexander Lee ADVERTISING SYSTEM ADMINISTRATOR Tina Burks SENIOR SALES COORDINATOR Shirley Young Frans de Waal, Emory U. Kathrin Plath, U. of California, Los Angeles
DIGITAL/PRINT STRATEGY MANAGER Jason Hillman QUALITY TECHNICAL MANAGER Marcus Spiegler ASSISTANT MANAGER DIGITAL/PRINT Claude Desplan, New York U. Martin Plenio, Ulm U.
Rebecca Doshi SENIOR CONTENT SPECIALISTS Steve Forrester, Jacob Hedrick, Antoinette Hodal, Lori Murphy DIGITAL PRODUCTION Sandra DÍaz, Universidad Nacional de CÓrdoba Elvira Poloczanska, Alfred-Wegener-Inst.
MANAGER Lisa Stanford CONTENT SPECIALIST Kimberley Oster DIGITAL AD-OPS SPECIALIST Patrick Gerrits ADVERTISING PRODUCTION Hong Ding, Inst. of Physics, CAS Julia Pongratz, Ludwig Maximilians U.
Jennifer Dionne, Stanford U. Philippe Poulin, CNRS
OPERATIONS MANAGER Deborah Tompkins DESIGNER, CUSTOM PUBLISHING Jeremy Huntsinger SR. TRAFFIC ASSOCIATE Christine Hall
Dennis Discher, U. of Penn. Jonathan Pritchard, Stanford U.
SPECIAL PROJECTS ASSOCIATE Sarah Dhere
Jennifer A. Doudna, U. of California, Berkeley
David Randall, Colorado State U.
ASSOCIATE DIRECTOR, BUSINESS DEVELOPMENT Justin Sawyers GLOBAL MARKETING MANAGER Allison Pritchard DIGITAL MARKETING Bruce Dunn, U. of California, Los Angeles
Félix A. Rey, Institut Pasteur
MANAGER Aimee Aponte MARKETING MANAGER Shawana Arnold MARKETING ASSOCIATES Tori Velasquez, Mike Romano, Ashley William Dunphy, Caltech
Trevor Robbins, U. of Cambridge
Hylton SENIOR DESIGNER Kim Huynh TRADE SHOW AND MEETINGS ASSOCIATE Andrew Clamp Christopher Dye, U. of Oxford
Todd Ehlers, U. of TÜbingen Amy Rosenzweig, Northwestern U.
DIRECTOR AND SENIOR EDITOR, CUSTOM PUBLISHING Sean Sanders ASSISTANT EDITOR, CUSTOM PUBLISHING Jackie Oberst Jennifer Elisseeff, Johns Hopkins U. Mike Ryan, U. of Texas at Austin
Tim Elston, U. of North Carolina Mitinori Saitou, Kyoto U.
DIRECTOR, BUSINESS STRATEGY AND PORTFOLIO MANAGEMENT Sarah Whalen ASSOCIATE DIRECTOR, PRODUCT MANAGMENT Kris Bishop
Andrea Encalada, U. San Francisco de Quito Shimon Sakaguchi, Osaka U.
ASSOCIATE DIRECTOR, PRODUCT DEVELOPMENT AND SPJ Hannah Heckner SR. PRODUCT ASSOCIATE Robert Koepke DIGITAL PRODUCT Miquel Salmeron, Lawrence Berkeley Nat. Lab
Nader Engheta, U. of Penn.
STRATEGIST Michael Hardesty SPJ ASSOCIATE Samantha Bruno Fuller Karen Ersche, U. of Cambridge Nitin Samarth, Penn. State U.
DIRECTOR, INSTITUTIONAL LICENSING Iquo Edim ASSOCIATE DIRECTOR, RESEARCH & DEVELOPMENT Elisabeth Leonard SENIOR INSTITUTIONAL Barry Everitt, U. of Cambridge Jürgen Sandkühler, Med. U. of Vienna
LICENSING MANAGER Ryan Rexroth INSTITUTIONAL LICENSING MANAGERS Marco Castellan, Christos Skoutas MANAGER, SYSTEMS AND Vanessa Ezenwa, U. of Georgia Alexander Schier, Harvard U.
Michael Feuer, The George Washington U. Wolfram Schlenker, Columbia U.
OPERATIONS Brian Holiahn MANAGER, AGENT RELATIONS & CUSTOMER SUCCESS Judy Lillibridge SENIOR OPERATIONS ANALYST Lana Guz
Toren Finkel, U. of Pittsburgh Med. Ctr. Susannah Scott, U. of California, Santa Barbara
FULFILLMENT COORDINATOR Melody Stringer
Gwenn Flowers, Simon Fraser U. Rebecca Sear, London School of Hygiene &
DIRECTOR, GLOBAL SALES Tracy Holmes ASSOCIATE DIRECTOR, ADVERTISING SALES, US Laurie Faraday US EAST COAST AND MID WEST SALES Peter Fratzl, Max Planck Inst. Potsdam
Tropical Med.
Glen Cox US WEST COAST SALES Lynne Stickrod US SALES MANAGER, SCIENCE CAREERS Claudia Paulsen-Young US SALES REP, SCIENCE Elaine Fuchs, Rockefeller U.
Vladimir Shalaev, Purdue U.
CAREERS Tracy Anderson ASSOCIATE DIRECTOR, ROW Roger Goncalves SALES REP, ROW Sarah Lelarge SALES ADMIN ASSISTANT, ROW Eileen Furlong, EMBL
Jay Gallagher, U. of Wisconsin Jie Shan, Cornell U.
Bryony Cousins DIRECTOR OF GLOBAL COLLABORATION AND ACADEMIC PUBLISHING RELATIONS, ASIA Xiaoying Chu ASSOCIATE DIRECTOR, Beth Shapiro, U. of California, Santa Cruz
Susan Gelman, U. of Michigan
INTERNATIONAL COLLABORATION Grace Yao SALES MANAGER Danny Zhao PROJECT MANAGER Kilo Lan ASCA CORPORATION, JAPAN Kaoru Jay Shendure, U. of Washington
Daniel Geschwind, U. of California, Los Angeles
Sasaki (Tokyo), Miyuki Tani (Osaka) COLLABORATION/CUSTOM PUBLICATIONS/JAPAN Adarsh Sandhu Karl-Heinz Glassmeier, TU Braunschweig Brian Shoichet, U. of California, San Francisco
DIRECTOR, COPYRIGHT, LICENSING AND SPECIAL PROJECTS Emilie David RIGHTS AND LICENSING COORDINATOR Jessica Adams RIGHTS AND Ramon Gonzalez, U. of South Florida Robert Siliciano, Johns Hopkins U. School of Med.
PERMISSIONS ASSOCIATE Elizabeth Sandler CONTRACTS AND LICENSING ASSOCIATE Lili Catlett Elizabeth Grove, U. of Chicago Lucia Sivilotti, U. College London
Nicolas Gruber, ETH ZÜrich Alison Smith, John Innes Centre
Kip Guy, U. of Kentucky College of Pharmacy Richard Smith, U. of North Carolina (S)
Taekjip Ha, Johns Hopkins U. Mark Smyth, QIMR Berghofer
MAIN HEADQUARTERS EDITORIAL PRODUCT ADVERTISING AAAS BOARD OF DIRECTORS Christian Haass, Ludwig Maximilians U. Pam Soltis, U. of Florida
Science/AAAS science_editors@aaas.org & CUSTOM PUBLISHING CHAIR Margaret A. Hamburg Sharon Hammes-Schiffer, Yale U. John Speakman, U. of Aberdeen
advertising.sciencemag.org/ Wolf-Dietrich Hardt, ETH ZÜrich
1200 New York Ave. NW NEWS PRESIDENT Steven Chu
Louise Harra, U. College London
Tara Spires-Jones, U. of Edinburgh
Washington, DC 20005 science_news@aaas.org products-services Allan C. Spradling, Carnegie Institution for Science
PRESIDENT-ELECT Claire M. Fraser Jian He, Clemson U.
science_advertising@aaas.org V. S. Subrahmanian, U. of Maryland
SCIENCE INTERNATIONAL INFORMATION FOR AUTHORS TREASURER Carolyn N. Ainslie Carl-Philipp Heisenberg, IST Austria
Ykä Helariutta, U. of Cambridge Ira Tabas, Columbia U.
Clarendon House sciencemag.org/authors/ CLASSIFIED ADVERTISING INTERIM CHIEF EXECUTIVE OFFICER
Sarah Teichmann, U. of Cambridge
science-information-authors Alan Leshner Janet G. Hering, Eawag
Clarendon Road advertising.sciencemag.org/ Rocio Titiunik, Princeton U.
Hans Hilgenkamp, U. of Twente
Cambridge, CB2 8FH, UK REPRINTS AND PERMISSIONS science-careers BOARD Cynthia M. Beall Shubha Tole, Tata Inst. of Fundamental Research
Kai-Uwe Hinrichs, U. of Bremen
sciencemag.org/help/ advertise@sciencecareers.org May R. Berenbaum David Hodell, U. of Cambridge Wim van der Putten, Netherlands Inst. of Ecology
SCIENCE CHINA
reprints-and-permissions Rosina M. Bierbaum Lora Hooper, UT Southwestern Med. Ctr. Reinhilde Veugelers, KU Leuven
Room 1004, Culture Square JOB POSTING CUSTOMER SERVICE
MEDIA CONTACTS Ann Bostrom Fred Hughson, Princeton U. Bert Vogelstein, Johns Hopkins U.
No. 59 Zhongguancun St. employers.sciencecareers.org Randall Hulet, Rice U. Kathleen Vohs, U. of Minnesota
scipak@aaas.org Stephen P. A. Fodor
Haidian District, Beijing, 100872 support@sciencecareers.org S. James Gates, Jr.
Auke Ijspeert, EPFL David Wallach, Weizmann Inst. of Science
MULTIMEDIA CONTACTS Akiko Iwasaki, Yale U. Jane-Ling Wang, U. of California, Davis (S)
SCIENCE JAPAN SciencePodcast@aaas.org MEMBERSHIP AND INDIVIDUAL Laura H. Greene Stephen Jackson, USGS and U. of Arizona
SUBSCRIPTIONS David Waxman, Fudan U.
ASCA Corporation ScienceVideo@aaas.org Kaye Husbands Fealing Kai Johnsson, EPFL Jonathan Weissman, U. of California, San Francisco
Sibaura TY Bldg. 4F, 1-14-5 sciencemag.org/subscriptions Maria Klawe Peter Jonas, IST Austria
INSTITUTIONAL SALES Chris Wikle, U. of Missouri (S)
Shibaura Minato-ku Robert B. Millard Matt Kaeberlein, U. of Washington
AND SITE LICENSES MEMBER BENEFITS Terrie Williams, U. of California, Santa Cruz
William Kaelin Jr., Dana-Farber Cancer Inst.
Tokyo, 108-0073 Japan sciencemag.org/librarian aaas.org/membercentral William D. Provine Daniel Kammen, U. of California, Berkeley Ian A. Wilson, Scripps Research (S)
V. Narry Kim, Seoul Nat. U. Yu Xie, Princeton U.
Science serves as a forum for discussion of important issues related to the advancement of science by publishing material on Robert Kingston, Harvard Med. School Jan Zaanen, Leiden U.
which a consensus has been reached as well as including the presentation of minority or conflicting points of view. Accordingly, Nancy Knowlton, Smithsonian Institution Kenneth Zaret, U. of Penn. School of Med.
all articles published in Science—including editorials, news and comment, and book reviews—are signed and reflect the individual Etienne Koechlin, École Normale Supérieure Jonathan Zehr, U. of California, Santa Cruz
views of the authors and not official points of view adopted by AAAS or the institutions with which the authors are affiliated. Alexander Kolodkin, Johns Hopkins U. Maria Zuber, MIT

10 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004Product.indd 11 9/25/19 2:15 PM
1004Product.indd 12 9/25/19 1:11 PM
 EDITORIAL

Language and the brain

L
anguages—exquisitely structured, complex, and the brain. People who were deprived of access to lan-
diverse—are a distinctively human gift, at the guage as children (e.g., deaf individuals without access
very heart of what it means to be human. As such, to speakers of sign languages) show patterns of neu-
language makes for both a particularly important ral connectivity that are radically different from those
and difficult topic in neuroscience. A dominant with early language exposure and are cognitively dif-
early approach to the study of language was to ferent from peers who had early language access. The
treat it as a separate module or organ within later in life that first exposure to language occurs, the
the brain. However, much modern empirical work has more pronounced and cemented the consequences. Lera Boroditsky
demonstrated that language is integrated with, and in Further, speakers of different languages develop dif-
is an associate
constant interplay with, an incredibly broad range of ferent cognitive skills and predispositions, as shaped
professor in the
neural processes. by the structures and patterns of their languages. Ex-
Department of
Unlike other areas of neuroscience investigation perience with languages in different modalities (e.g.,
(e.g., vision, motor action) that have relied heavily on spoken versus signed) also develops predictable dif- Cognitive Science
invasive techniques with animal models, the study of ferences in cognitive abilities outside the boundar- at the University
language lacks any such model. ies of language. For example, of California,
Furthermore, in language, the speakers of sign languages de- San Diego, CA, USA.
relationship between the form velop different visuospatial at- lera@ucsd.edu
of a signal and its meaning is
largely arbitrary. For example,
“…one cannot tention skills than those who
only use spoken language. Ex-
the sound of “blue” will likely
have no relationship to the prop-
understand the posure to written language also
restructures the brain, even
erties of light we experience as
blue nor to the visual written human brain without when acquired late in life. Even
seemingly surface properties,
form “blue,” will sound different
across languages, and have no understanding such as writing direction (left-
to-right or right-to-left), have
sound at all in signed languages.
No equivalent of “blue” will the contributions profound consequences for how
people attend to, imagine, and
even exist in many languages organize information.
that might make fewer or more of language…” The normal human brain that
or different color distinctions. is the subject of study in neuro-
With respect to language, the science is a “languaged” brain.
meaning of a signal cannot be It has come to be the way it is
predicted from the physical properties of the signal through a personal history of language use within an in-
available to the senses. Rather, the relationship is set dividual’s lifetime. It also actively and dynamically uses
by convention. linguistic resources (the categories, constructions, and
At the same time, language is a powerful engine of distinctions available in language) as it processes incom-
human intellect and creativity, allowing for endless re- ing information from across the senses.
combination of words to generate an infinite number Put simply, one cannot understand the human brain
of new structures and ideas out of “old” elements. Lan- without understanding the contributions of language,
guage plays a central role in the human brain, from how both in the moment of thinking and as a formative force
we process color to how we make moral judgments. It during earlier learning and experience. When we study
directs how we allocate visual attention, construe and language, we are getting a peek at the very essence of
remember events, categorize objects, encode smells and human nature. Languages—these deeply structured cul-
musical tones, stay oriented, reason about time, perform tural objects that we inherit from prior generations—
mental mathematics, make financial decisions, experi- work alongside our biological inheritance to make
ence and express emotions, and on and on. human brains what they are.
Indeed, a growing body of research is documenting
how experience with language radically restructures –Lera Boroditsky
CREDITS: (PHOTO) ASA MATHAT

10.1126/science.aaz6490

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 13

1004Editorial.indd 13 10/1/19 6:03 PM

 NEWS
There are those who seem to expect people
“ to work as if they didn’t have children.
”
Engineer Danielle George, in Times Higher Education, after colleagues faulted her for seeking
to present remotely at a conference, instead of in person, so she could care for her 4-year-old.

IN BRIEF Scientist jailed for pollution study

Edited by Jeffrey Brainard
HUMAN RIGHTS | A Turkish food scientist
was sentenced last week to 15 months in
PUBLIC HEALTH jail after disclosing findings from a study
he co-authored that linked toxic pollu-
Eastern equine encephalitis cases surge tion to a high incidence of cancer in
western Turkey. Bülent S˛ık, formerly of
Akdeniz University in Antalya, Turkey,

C
ases of brain infection in the United States caused by mosquito-
was convicted of releasing classified
borne eastern equine encephalitis (EEE) virus have jumped information in the study, commissioned
to a level not seen in decades, for reasons unknown. As of by the Ministry of Health, which has not
1 October, the Centers for Disease Control and Prevention disputed the accuracy of his findings. S˛ık
(CDC) has received reports of 31 confirmed and probable cases, and colleagues had discovered danger-
ous levels of pesticides, heavy metals,
including nine deaths. On average, seven EEE brain infections and polycyclic aromatic hydrocarbons in
occur annually in the United States, according to CDC. This year, multiple food and water samples from
most cases have occurred in Massachusetts (12) and Michigan (nine); several provinces in western Turkey. After
other affected states include Connecticut, North Carolina, and Rhode completing the study in 2015, S˛ık urged
government officials to take action. After
Island. People younger than 15 and older than 50 are at the highest 3 years of inaction, he published his
risk from EEE virus infection, which kills one-third of those who de- findings in an Istanbul newspaper. S˛ık,
velop the brain disease. Transmission, which continues until the first who lost his university position in 2016
hard frost, occurs most commonly in and around freshwater swamps after signing a petition calling for peace
between Turkish forces and Kurdish mili-
in Atlantic, Gulf Coast, and Great Lakes states. There is no treatment
tants in southeastern Turkey, remains free
and no commercially available vaccine. pending an appeal.

Satellite to watch Earth’s heat

First forensic genealogy rules Metaresearch consortium debuts | The European Space
C L I M AT E S C I E N C E
CRIMINAL JUSTICE | The U.S. Department SCIENCE OF SCIENCE | The Wellcome Agency (ESA) last week chose a new
of Justice (DOJ) last week released Trust, a major nonprofit biomedical satellite mission that will collect key
rules about when police can use genetic funder in London, is teaming up with missing data needed to predict more
ancestry databases to track down crime several partners to support metaresearch accurately the pace of climate change.
suspects. Law enforcement has used through a new consortium called the The Far-infrared Outgoing Radiation
these sites to help crack at least 60 cold Research on Research Institute. The Understanding and Monitoring (FORUM)
cases—and arrest the suspected Golden partnership includes Leiden University mission, to be launched in 2026, will
State Killer—by uploading a suspect’s DNA in the Netherlands, the University of take the first comprehensive measure-
profile and identifying distant relatives Sheffield in the United Kingdom, and ments from orbit at wavelengths up to
to develop leads. But the searches raise Digital Science, a research analytics 100 microns. Earth radiates most of its
concerns about the privacy of individu- company in London. Researchers at the energy into space in that part of the spec-
als in the database and their relatives. partner organizations hope to create trum, and global temperatures are rising
Starting on 1 November, DOJ says police new analytic methods; they also plan to as greenhouse gases trap the outgoing
can generally use forensic genetic geneal- study diversity and reproducibility in radiation. Humans’ growing emission of
ogy only to investigate homicides and sex research, how funders and institutions greenhouse gases has trapped heat and
crimes, and only after exhausting tradi- set research priorities, and how incen- decreased the outflow, causing tempera-
tional crime solving methods. Police must tive systems and career development tures to rise. Scientists plan to use the new
identify themselves to the ancestry site, can be improved. Funding agencies from data to improve climate model predic-
and the site itself must have informed its eight countries and several other large tions about the energy balance at the top
users that law enforcement agencies may foundations are planning to join the of the atmosphere, and how changes in
search their data. Although the interim consortium and share their data and water vapor, ice crystals, and clouds could
policy applies only to federally funded decision processes, the organizers say, affect surface temperatures. FORUM is the
searches, experts expect it to become a enabling metaresearchers to do analyses ninth mission selected under ESA’s Earth
national model. on a larger scale than ever before. Explorers program.

14 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004NewsInBrief.indd 14 10/1/19 6:28 PM

 ARCHAEOLOGY

Single strain of plague fingered in Europe’s Black Death

A
single, highly virulent strain of Yersinia pestis bacterium, Human History in Jena, Germany, identified the single strain
probably introduced from western Asia or eastern Europe, by analyzing 34 ancient genomes of Y. pestis from the teeth
was responsible for the Black Death pandemic that wiped of people buried at 10 sites in Russia and Europe from the
out as much as 60% of Europe’s population in the 14th 14th to 17th centuries. (A 1349 painting [above] by Gilles Le
century, researchers report this week. Until now, scien- Muisit shows the burial of Black Death victims in Belgium.)
tists didn’t know whether the deadly pathogen that causes In Nature Communications, the researchers describe a strain
plague came from a single area or was introduced by multiple from Russia’s Volga region ancestral to all other, later strains in
travelers carrying diverse strains from around the ancient world. Europe they studied, differing by only one mutation. But, they
Researchers at the Max Planck Institute for the Science of note, the pathogen could have reached Russia from elsewhere.

publications, including the Chinese 700 projects. PCORI-funded findings are

China plans top journals Medical Journal, Chinese Physics B, and expected to help physicians determine
PUBLISHING | China’s total output of sci- Earth and Planetary Physics. But papers what therapies to give patients and
entific papers surpassed that of the United in these journals rarely garner inter- make health care more efficient. For
PHOTO: PHOTO12/UNIVERSAL IMAGES GROUP VIA GETTY IMAGES

States in 2016, but some Chinese academ- national attention. example, one study showed that taking
ics have noted that many of the country’s a blood thinner after a stroke helped
top papers continue to appear in journals patients live at home and avoid further
published in other countries. Last week, End nears for health results group hospital stays. PCORI has $50 million
the China Association for Science and BIOMEDICINE | This week could mark left in its coffers to pay existing funding
Technology, a nongovernmental organiza- the beginning of the end for the Patient- commitments as it winds down over
tion representing 210 national scientific Centered Outcomes Research Institute 5 years. Bills are pending in Congress
societies, unveiled a 5-year Periodical (PCORI), a nonprofit organization in that would renew PCORI’s charter.
Excellence Action Plan to develop “world- Washington, D.C., that compares how well
class” science and technology journals treatments work; its charter expired on
managed by Chinese organizations. The 30 September. Created by the 2010 Grant rejectees later do well
number of journals and funding for the Affordable Care Act and funded partly by FUNDING | If you’ve had a grant
plan were not revealed. China already has a tax on health care insurers, PCORI has rejected, take heart: Early-career
more than 180 English-language scientific allotted more than $2.5 billion to some scientists whose grant applications fell

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 15

1004NewsInBrief.indd 15 10/1/19 6:28 PM

 NEWS | I N B R I E F

just short of the funding threshold for says it will sell a daily dose of insulin articles for publication. Each racked up
the U.S. National Institutes of Health’s made with recombinant technology for hundreds of citations from the articles
(NIH’s) main R01 research grant 10 cents, the lowest price offered by any they oversaw, more than 10 times the
went on to publish more highly cited manufacturer. According to the company, average for other PLOS ONE editors,
papers in the 10 years that followed the median price for a daily dose of insu- Petersen showed in his analysis of PLOS
than colleagues who narrowly won lin in these countries is 36 cents. Biocon ONE data from 2006 to 2015, published
funding. A study this week in Nature has sold 2 billion doses of human insulin in the November issue of the Journal
Communications analyzed information in the past 15 years. Some 80% of people of Infometrics. Petersen told Science
about 1184 scientists who submitted with diabetes who depend on insulin live he couldn’t prove a quid pro quo. But
grant applications between 1990 and in low- and middle-income countries, and he said the association deserved more
2005. The findings indicate that early many struggle to pay for it. research—both at PLOS ONE, which
career setbacks can serve as a motivating identifies articles’ handling editors and
force, driving future success in publish- is the only journal he studied, and at
ing impactful research—at least among Editors secured citation bump other publications that do not. Petersen
rejected applicants who tried again to PUBLISHING | Some of the more than 7000 also suggested that PLOS ONE limit the
get NIH grants. editors of the PLOS ONE megajournal number of articles that each subject
have accrued unusually large numbers of matter editor handles. In a statement to
citations, which a study suggests may be Science, PLOS ONE said that after 2015,
Maker deeply discounts insulin evidence that authors cited the editors it strengthened oversight of its editorial
| A drug manufacturer
P U B L I C H E A LT H as an incentive for favorable treatment. board and removed editors “that did not
in India announced last week that it has Alexander Petersen of the University of meet our high standards of manuscript
slashed the price of its human insulin California, Merced, found that three of handling, in some cases due to concerns
for poor and middle-income countries. the 10 most active PLOS ONE editors around the editorial practices that the
Biocon Biologics, based in Bengaluru, were also among the fastest at accepting study looked into.”

BY THE NUMBERS

BIOMEDICINE

Few NIH grantees have financial conflicts

805
F
inancial conflicts of interest that could bias researchers funded by the U.S. National
Institutes of Health (NIH) are rare, according to a report released last week. Only U.S. reported cases of lung injury,
about 3% of 55,600 grants awarded in 2018 involved a researcher with a “significant” including 12 deaths, associated
financial interest. Grantees’ institutions did not report monetary values for about half with vaping as of last week, a
of the financial interests, often equities in privately held companies, says the report near-quadrupling of a 27 August tally.
by the Department of Health and Human Services’s Office of Inspector General (OIG). (Centers for Disease
When institutions did disclose dollar values of financial interests, 85% were less than Control and Prevention)
$100,000. NIH tightened reporting rules and oversight after a 2008 OIG analysis found the
agency was not collecting adequate data on financial conflicts. The OIG report “doesn’t
tell us the whole story,” says conflict of interest expert Sheldon Krimsky of Tufts University
$750

CREDITS: (GRAPHIC) A. CUADRA/SCIENCE; (DATA) U.S. DEPARTMENT OF HEALTH AND HUMAN SERVICES
in Medford, Massachusetts, noting that in surveys, a much larger fraction of biomedical
researchers has reported industry relationships.
million
Gift to the California Institute of
Mostly modest conflicts Technology in Pasadena for research
In 2018, grantees’ institutions described 3978 “significant” financial interests but reported values for only
on environmental sustainability—the
55% of them. (Some grantees had more than one conflict per grant.)
largest of its kind—from billionaires
Values of fnancial conficts Stewart and Lynda Resnick, whose
business interests include agriculture
71% 14% 10% 2 4 and bottled water.

52
<$40,000 $40,000– $100,000– $300,000– ≥$600,000
99,999 299,999 599,999

Types of fnancial conficts Hours spent annually by researchers

to format scientific manuscripts to
49% 24% 14% 8%
meet journal-specific rules; many call
it a waste of time. (PLOS ONE)

Equity interests Payment Other Intellectual Salary Investments Travel

services property rights 2% 1% 1% SCIENCEMAG.ORG/NEWS
Percentages are rounded. Read more news from Science online.

16 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004NewsInBrief.indd 16 10/1/19 6:28 PM

 AAAS® and Science® are registered trademarks of the American Society for the Advancement of Science, USA. Massachusetts General Hospital® is a registered trademark of THE GENERAL HOSPITAL CORPORATION, USA.
Harvard Medical School® is a registered trademark of President and Fellows of Harvard College Corporation, USA. Eppendorf® and the Eppendorf logo are registered trademarks of Eppendorf AG, Germany.
Congratulations
to Lauren Orefice, Ph.D.
Assistant Professor
Massachusetts General Hospital
& Harvard Medical School

Meet the Winner

All rights reserved, including graphics and images. Copyright © 2019 by Eppendorf AG.
Eppendorf & Science Prize for Neurobiology
Congratulations to Lauren Orefice on winning the 2019 The annual US$25,000 Eppendorf & Science Prize for
Eppendorf & Science Prize for her work on the causes Neurobiology honors scientists, like Dr. Orefice, for their
and potential therapies for autism spectrum disorders. outstanding contributions to neurobiological research.
Dr. Orefice found that peripheral sensory neurons – Lauren Orefice is the 18th recipient of this international
neurons outside the brain – are key areas where autism- award. She will be presented with the Prize at a ceremony
associated gene mutations have a critical impact. She held during the week of the 2019 Annual Meeting of the
showed how abnormal function of peripheral sensory Society for Neuroscience in Chicago.
neurons causes touch over-reactivity. She demonstrated
how this over-reactivity during development contributes You could be next to win this prize.
to altered brain function and some autism-related If you are 35 years of age or younger and currently per-
behaviors in mice. Dr. Orefice’s work changes how we forming neurobiological research, you could be next to
think about the causes of autism spectrum disorders, win the 2020 Prize. Deadline for entries is June 15, 2020.
and highlights peripheral sensory neurons as a possible
novel therapeutic target.
Learn more at: www.eppendorf.com/prize

1004Product.indd 17 10/1/19 12:04 PM

 Indonesian firefighters try
to douse a peatland fire
IN DEP TH in Riau province on northern
Sumatra on 31 July.

ENVIRONMENT

Parched peatlands fuel Indonesia’s blazes

Policies to preserve and restore waterlogged forest help mitigate fires, scientists say

By Dennis Normile Small farmers in Indonesia have long The conflagrations are paced by the
practiced slash-and-burn agriculture, and weather. In 2015, an El Niño in the west-

O
nce again haze is suffocating Indone- in recent decades large corporations have ern Pacific Ocean combined with another
sia, but some scientists say it could industrialized the practice. They own long- irregularly recurring climatic phenomenon
have been worse. Acrid smoke from term concessions to develop plantations called the Indian Ocean dipole to make In-
fires set to clear land for agriculture within publicly owned forests, many in donesia’s typically dry summer even drier,
has sent scores to hospitals with swampy landscapes rich in organic mate- Field says. Fires raged from late June un-
respiratory problems and closed rial. Concentrated in the coastal plains of til the end of October, burning 2.6 million
thousands of schools in Indonesia and Sumatra, Borneo, and Papua, these peat for- hectares, an area half the size of Costa Rica.
neighboring Malaysia. At its thickest, in ests provide a habitat for rare species such The haze affected countries as far away as
mid-September, more than 100 flights had as orangutans, leopards, Sumatran tigers, Thailand and the Philippines.
to be canceled because of poor visibility. tapirs, white-winged ducks, and freshwater Stung by criticism, Indonesian President
Although the government has tried to seed turtles. But in the 1980s, concession hold- Joko Widodo promised action. In January
clouds for rain and dump water from the ers started to dig drainage canals through 2016, he established the Peatland Resto-
air, only the monsoon rains due later this peatlands to float out logs and dry out the ration Agency (BRG) in Jakarta, which is
month are likely to quench the fires. peat to plant dryland crops, especially oil trying to restore more than 2.6 million
Yet countermeasures Indonesia has taken palm and acacia trees for pulp and paper. hectares of degraded peatlands—two-
since the last major haze event, in 2015, Fires they set to clear the land can burn out thirds within corporate concession areas,
have helped limit this year’s disaster. A new of control. the rest in government hands—by 2020.
agency is restoring degraded peatlands, Peatlands play an outsize role in the hazes The agency blocks drainage canals, often
where agribusiness has drained and dried because dry underground peat deposits pro- with simple earthen or wooden dams de-
out meters-thick layers of waterlogged veg- vide “an inexhaustible supply of fuel,” says signed in some cases to allow small boats
etation, making it vulnerable to ground fires Robert Field, an atmospheric scientist at to pass, says BRG head Nazir Foead. It
that are almost impossible to stop. The gov- Columbia University who studies Southeast also replants degraded areas with native

PHOTO: WAHYUDI/AFP/GETTY IMAGES

ernment has also beefed up a moratorium on Asia’s haze. And because the very ground is vegetation and encourages local communi-
the conversion of prime forest land under- burning, “the fires can’t be contained until ties to use the lands in a sustainable way
lain by peat. The efforts “are providing some the monsoon.” They release not just smoke, for fishing and planting crops adapted to
positive results,” says Arief Wijaya of the but vast quantities of greenhouse gases. wetlands, such as sago palm. By the end of
Indonesian branch of the World Resources Indonesian tropical peatlands, 36% of the 2018, the agency had initiated restoration
Institute in Jakarta. But virtually all experts world’s total, hold an estimated 28.1 gigatons projects in 366 villages in seven provinces,
agree that more is needed, including stricter of carbon, according to a 2017 study—more Foead says.
enforcement of a ban on setting fires. than all the country’s upland forests. To remove the incentive to set fires—as

18 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004NewsInDepth.indd 18 10/1/19 5:34 PM

 NE WS

well as preserve Indonesia’s remaining SCIENCE & SECURITY

rainforests—the government in 2018 de-
cided not to grant new licenses for oil palm
plantations, instead focusing on boost-
ing yields from existing sites and lessen-
NIH reveals its formula for
ing their environmental impact (Science,
12 July, p. 112). And in August, Indonesia
made permanent a temporary moratorium
tracking foreign influences
on converting primary forests and peat- Official highlights China’s talents program as agency
lands to agricultural use. The government
has also promised stricter enforcement of
watchdog calls for better vetting of peer reviewers
laws that make concession holders respon-
sible for fires in their holdings, regardless By Jeffrey Mervis participating in the program have been en-
of whether they are deliberately set. Penal- couraged to pass along NIH grant proposals

T
ties can include criminal prosecution. o fight what it calls China’s theft of they were reviewing to Chinese colleagues,
Because the years since 2015 have been U.S.-funded intellectual property, the a violation of NIH rules on peer review.
relatively wet, the measures had never re- U.S. National Institutes of Health The program also has a provision for NIH-
ally been tested—until this year. The Indian (NIH) is focusing on China’s foreign funded scientists to set up parallel labs in
Ocean dipole again gave Indonesia an ex- talent recruitment programs. China, which Lauer says is aimed at allowing
tremely dry summer, Field says. When that NIH has concluded that the Chi- Chinese colleagues to sidestep U.S. techno-
happens, “You can’t expect no fires and nese government uses them to obtain con- logy export regulations if the research is
haze; the way to look at it is to see if there is fidential NIH grant applications and to eventually commercialized.
improvement,” he explains. establish so-called shadow labs in China, “It’s a very patient approach,” Lauer says
BRG claims there is, and Wijaya agrees. where NIH-funded research can be repli- about the so-called shadow labs. “You can
Field inspections found no indications of cated. And it is paying more attention not build a tremendous amount of knowledge
fires at 65% of the village restoration sites. just to grantees, but also to peer reviewers: about basic, preclinical science by setting
Between January and 15 September, fires the 27,000 researchers who vet- up shadow labs in China that
consumed 330,000 hectares, according to ted the 85,000 proposals NIH are mirrors of U.S. labs. And
Indonesia’s Ministry of Environment and received last year. “You can build then, once the research gets to
Forestry—less than 13% the toll of 2015.
But some restored peatlands are still
Many Chinese American sci-
entists fear the increased scru-
a tremendous a point of being translational,
it’s already in China. You don’t
burning, says Bambang Saharjo, a fire fo- tiny will result in unfair eth- amount of have to worry about tech trans-
rensics expert at IPB University in Bogor, nic profiling. “Akin to ‘driving fer—you’ve already transferred
Indonesia, pointing to shortcomings in the while black,’ there is growing knowledge … the technology to yourself.”
restoration program. For one thing, BRG
directly manages projects only on non-
fear of ‘researching while Chi-
nese,’” said David Ho, an AIDS
by setting Lauer declined to quantify
how often NIH believes these
concession and village lands; it provides
technical advice to oil palm concession
researcher at Rockefeller Uni-
versity in New York City, at a
up shadow labs tactics have been used. And
he said NIH is not looking
holders but can’t guarantee compliance, conference last week in Palo in China …” solely at scientists with links to
Saharjo says. He says the agency should Alto, California, on how tensions Michael Lauer, China. But some members of
monitor groundwater levels in peatlands between the United States and National Institutes Congress and President Don-
to check the progress of restoration. Mean- China are affecting scientists. of Health ald Trump’s administration
while, a different agency oversees restora- In August 2018, NIH began to have said China is by far their
tion efforts in logging concessions, further ask U.S. research institutions to investigate biggest concern, and Lauer repeatedly
confusing restoration efforts. whether some of their faculty members had cited the Thousand Talents Program when
Moreover, loopholes limit the effectiveness violated the confidentiality of peer review describing how NIH is dealing with threats
of both BRG’s efforts and the ban on forest or rules requiring disclosure of foreign af- to research integrity.
conversion, says Yuyun Harmono, climate filiations. The outcomes of most of those Knowing how the program operates has
justice campaign manager for the Indonesian investigations, affecting more than 60 in- helped NIH figure out a way to spot NIH-
Forum for Environment in Jakarta, an affili- stitutions, have not been made public. But funded scientists whose participation puts
ate of Friends of the Earth. For example, the this spring, MD Anderson Cancer Center in them at higher risk, Lauer says. One telltale
maps defining where the permanent mora- Houston, Texas, and Emory University in sign is when U.S.-based scientists list a Chi-
torium applies are incomplete and often Atlanta ousted several faculty members of nese institution as their primary affiliation
revised. And enforcement is lax, Harmono Asian descent after receiving such letters. on papers. “We have seen the contracts that
and others say. Indonesian courts found NIH has been tight-lipped about how it these scientists have signed with Chinese
a number of concession holders liable identified some 250 scientists who came institutions,” Lauer says, “in which they are
for damages from the 2015 fires, but the under suspicion. (One hundred and eighty explicitly told to make sure their Chinese af-
government has still not moved to collect are still being investigated.) But last week, filiation is listed first.”
payment, he says. “We need to ensure that Michael Lauer, director of NIH’s extramural NIH is also ramping up efforts to protect
companies [setting fires] fear law enforce- research program in Bethesda, Maryland, its peer-review system, urged on by a 27 Sep-
ment,” Harmono says. j suggested many were linked to China’s tember report from the Office of Inspector
Thousand Talents Program, which seeks to General (OIG) in Washington, D.C., within
With reporting by Dyna Rochmyaningsih in Deli build ties with researchers outside of China. the Department of Health and Human Ser-
Serdang, Indonesia. Lauer alleges that U.S.-based researchers vices, NIH’s parent body. One approach, it

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 19

1004NewsInDepth.indd 19 10/1/19 5:34 PM

 NEWS | I N D E P T H

says, is for NIH to pay “extra attention [to SCIENTIFIC INTEGRITY

scientists] who would be reviewing grant
applications with particularly sensitive sub-
ject matter or that have lucrative commer-
cial applications.”
Duplicated images point to
Lauer agrees that NIH needs to up its
game. “The leakage of information through
peer review is of great concern to us.” He
fraud in fish study, critics say
estimates that bad behavior by peer review- A collage raises doubts about a 2014 behavior paper whose
ers constitutes about 10% of the cases under
active investigation. But most reviewers are
first author fabricated data in an earlier study
also grantees, he notes, meaning any secu-
rity efforts aimed at grantees should also By Martin Enserink the number. (Chivers was also an external co-
address the conduct of reviewers. supervisor of Lönnstedt’s Ph.D. research.)

H
Erin Bliss, OIG’s assistant inspector ow many fish really appear in the The paper described a series of lab exper-
general for evaluation and inspections, photo collage on p. 21? The answer iments showing that zebra lionfish (Dendro-
says, “We have no basis to say” that being bears on whether a study about chirus zebra) flare their pectoral fins in a
a foreign national or working at a foreign lionfish social behavior, published striking display to recruit other fish for a
institution poses an inherently greater se- in Biology Letters in 2014, was collective hunt. The authors speculated that
curity risk. “And NIH has talked to us about fabricated—and whether Oona fin movements could even convey informa-
how important collaborations with foreign Lönnstedt, a marine biologist formerly at tion about the prey or its location, similar
researchers are, in terms of advancing sci- Uppsala University (UU) in Sweden who to the honey bee’s “waggle dance.” “These
ence. But we do recommend that NIH work made up data in a 2016 Science paper, com- are highly complex animals with advanced
with national security and intelligence ex- mitted an earlier fraud. The case also raises social behaviors, and they are ridiculously
perts to figure out what are the appropriate fresh questions about whether senior scien- good at catching prey,” Lönnstedt told Na-
risk factors that they should be considering tists working with Lönnstedt, who was then tional Geographic at the time.
in targeting their oversight.” a Ph.D. student, properly oversaw and took A whistleblower who asked not to be
NIH did exactly that in creating a pool responsibility for her work. identified because he’s worried it might
of grantees that warranted extra scrutiny, Last year, Lönnstedt and her co-authors hurt his career tells Science he first sent
Lauer says. By combining advice from se- posted the collage on the Biology Letters web- questions about the data to Biology Letters
curity experts with a review of the affilia- site in what appeared to be an attempt to end in May 2016, before the Science paper had
tions listed on grantee publications, NIH questions about whether the scientists really come out. Once the Science paper was re-
produced “a list of scientists who fit our caught enough fish to carry out their behav- tracted, he voiced concerns that the Biology
phenotype of who we might want to look ioral experiments. But critics say the colorful Letters paper, too, might be made up.
at in more detail,” he says. Expanding such ensemble appears to include many photos of The experiments described by Lönnstedt
screening to all peer-reviewer candidates the same fish, and in some cases doctored would have required catching 86 zebra lion-
would take “at least 6 months to 1 year,” duplicates of the same photo—which would fish and 16 spotfin lionfish (Pterois anten-
NIH told OIG. undermine the authors’ defense. nata). At the Lizard Island Research Station
At the Palo Alto conference, scientists The lionfish study was done in 2012, when on Australia’s Great Barrier Reef, where all
worried that NIH’s crackdown has already Lönnstedt was a student at James Cook three co-authors worked in 2012, research-
gone too far. The known cases “are largely University (JCU) in Townsville, Australia. ers are required to record their catch.
due to sloppiness and a degree of greed” But the suspicions about it resemble those When the whistleblower checked an online
by a few scientists, Ho argued at the event, that discredited the 2016 Science study of record of fish collections, he found that
put on by the Committee of 100, a group the effects of microplastics on fish larvae. Lönnstedt caught only 12 zebra lionfish and
of prominent Chinese Americans that pro- There, too, researchers questioned whether three spotfin lionfish during her 2012 trip.
motes U.S.-Chinese ties. “A small number Lönnstedt had collected the claimed number Zebra lionfish and spotfin lionfish aren’t
of bad apples does not connote a systemic of fish and wondered how she could have common around Lizard Island, says Anne
problem that requires federal intervention recorded reams of behavioral data without Hoggett, one of the station’s two directors,
when it could be addressed at the institu- videotaping the experiments. In 2017, both who has worked there for decades and says
tional level with policies already in place.” UU and a national Swedish ethics panel con- she’s “very familiar” with the marine fauna.
Brian Sun, a lawyer with Jones Day in firmed the doubts about the Science paper, “If I was advising someone proposing a proj-
Los Angeles, California, warned the audi- co-authored with UU biologist Peter Eklöv. ect that required such numbers, I would sug-
ence that participating in the Thousand Tal- It was retracted and Lönnstedt, who main- gest that it would be extremely difficult to
ents Program “puts a target on your back. tained her innocence, lost her job. obtain them in a reasonable period of time.”
So don’t be stupid.” NIH’s intense focus on Lönnstedt, who has left research, did not In February 2018, Biology Letters pub-
the program could even lead to changes in respond to requests for comment sent by lished an expression of concern about the
Thousand Talents, says Shan Ping Yu, an email and to her residence. Her co-authors on paper and said it was carrying out an investi-
anesthesiologist at Emory who follows the the Biology Letters paper, Maud Ferrari and gation. In November 2018, the three authors
issue. “What NIH is doing now,” Yu says, Douglas Chivers, both of the University of published a correction explaining that, in
“will push these Thousand Talent scholars Saskatchewan in Saskatoon, Canada, say they contrast to their earlier statement, they had
to make a final decision to stay in the U.S. don’t know whether Lönnstedt used as many reused some fish in the study, lowering the
without connections with China, or move lionfish as the paper claims. But they dispute numbers required to 40 zebra lionfish and
back to China to be a real, full-time em- that the collage, which they say Lönnstedt nine spotfin lionfish. To provide “evidence of
ployee there.” j produced, was meant to dispel doubts about the number of lionfish,” the correction said,

20 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004NewsInDepth.indd 20 10/1/19 5:34 PM

 of the Journal of Cell Biology who now runs
Image Data Integrity, a San Francisco–based
company, concurs that three pairs of images
in the collage “do indeed appear to be dupli-
cates derived from the same source images.”
The whistleblower says he brought these
problems up with Biology Letters in January.
A spokesperson for the Royal Society in Lon-
don, the journal’s publisher, says the matter
is still under investigation. “A summary of the
findings has been passed to the authors for
response and we cannot comment further at
this stage,” he wrote in an email to Science.
Chivers and Ferrari don’t dispute that
the same fish appear multiple times in the
collage. But in an email to Science, they say
Lönnstedt had produced the collage years
earlier for a presentation, and it wasn’t
meant to show the number of fish used in
the study. The published correction note mis-
takenly presents the collage as evidence, they
say, because of “an unfortunate error” during
the editing process at Biology Letters, which
they say the journal has acknowledged. They
want the correction corrected.
In their email, the two scientists say they
were at Lizard Island at the same time in
2012 as Lönnstedt and helped her design
the study and the fish tanks she used, but
did not see her actually conduct the ex-
periments. “We had no reason to think any-
thing was suspicious,” they write. Because
lionfish are nocturnal, they believed the
experiments took place “in the middle of
the night,” when they were asleep. “We have
no firsthand knowledge of the difficulty of
catching fish” at Lizard Island, Chivers and
Ferrari add. “We snorkel, but do not dive.”
A collage on the Biology Letters website shows 50 lionfish, but three pairs of photos (highlighted) are duplicates, Josefin Sundin, a former colleague of
image manipulation experts say. Other photos in the collage, though not identical, appear to show the same fish. Lönnstedt’s at UU who blew the whistle on
the Science study, says those answers re-
the authors published the collage of 50 fish (JCU dismissed Ridd in 2018 for violating mind her of Eklöv, who said he had almost
images in the supplementary material. its code of conduct after he repeatedly criti- no involvement in the actual work for the
But now the supposedly exonerating data cized research quality control at the Austra- Science paper or knowledge about the way
have come under suspicion as well. In a lian Research Council Centre of Excellence it was carried out. “Chivers was Lönnstedt’s
report he sent to Chivers earlier this year for Coral Reef Studies, which is based at JCU. supervisor,” Sundin says. “He should have
and that Science has seen, former JCU ma- In April, a judge ruled that his termination known what she was doing.” But Chivers
rine scientist Peter Ridd wrote that in two was unlawful, and earlier this month JCU says there is “nothing unusual about an ad-
instances, the same photo had been used was ordered to pay Ridd AU$1.2 million in vanced Ph.D. student, postdoctoral fellow,
IMAGES: O. LÖNNSTEDT ET AL., BIOLOGY LETTERS 10, 10.1098 (CC-BY 4.0)

twice in the collage. In each case, one of the damages. JCU has appealed.) or collaborator working on their own with-
photos was mirrored or its contrast or col- Science asked Elisabeth Bik, an indepen- out me or another senior researcher watch-
ors appeared to be altered. dent image expert based in San Francisco, ing over their shoulder.”
On top of that, Ridd wrote, many photos California, to study the collage. She, too, After the Science scandal, JCU promised
that weren’t identical appeared to be of the found two pairs of duplicate images. The to do an investigation of the research Lönnst-
same fish. From metadata in the PDF file fact that one-half of each pair was cropped edt published while at the university—a total
of the collage, he reconstructed a partial or mirrored “might suggest that these are not of 20 papers, 10 of them co-authored with
chronology of the original images. In sev- honest error, but a potential intention to mis- Chivers, Ferrari, or both. But that investiga-
eral instances, photos taken consecutively— lead,” she writes. (Later Bik confirmed a third tion has yet to start, a JCU spokesperson says.
but placed far apart in the collage—showed duplication.) Bik also found a set of 12 zebra The whistleblower says the Biology Let-
very similar-looking fish in almost the same lionfish photos that, although not identical, ters paper should have been retracted by
position. “It would be remarkable if one “look so similar that one could suspect they now. He says the failure of journals and
could get a second [different] fish into this were taken from the same animal,” and also universities to take misconduct allegations
precise position and take a photo in the very three very similar looking spotfin lionfish. seriously is “unethical.” For whistleblowers,
next shot from the camera,” Ridd wrote. Mike Rossner, a former managing editor he says, “It’s exhausting. And depressing.” j

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 21

1004NewsInDepth.indd 21 10/1/19 5:34 PM

 At the European Research and Innovation Days,
participants sat down to “codesign” Horizon Europe.

although he adds, “We will never infringe

on the autonomy of scientists and won’t
prescribe their day to day.”
The “cocreation” effort is focused largely
on Horizon Europe’s second and largest
“pillar,” called Global Challenges and Eu-
ropean Industrial Competitiveness and
worth €52.7 billion. It includes six thematic
“clusters” to fund the bulk of collaborative
research projects and five so-called missions,
a pet project of outgoing research commis-
sioner Carlos Moedas. Inspired by the Apollo
program to put a human on the moon, the
missions are headline targets, still to be de-
fined, in five broad areas: cancer, climate,
cities, soil, and oceans. “I don’t want pub-
lic servants to say what the missions are,”
Moedas said at the event. “I want to co-
RESEARCH FUNDING create these missions with the people …
and transform them into something very

Europe seeks broader support easy to understand.”

At a handful of “codesign” sessions at the
conference, participants sat in small circles

and impact for science projects with a member of the mission board to
share ideas in response to questions such as:
“Why are these challenges [for our oceans,
The European Commission is asking the public to help coasts, and inland waters] so removed from
design its new €94 billion research program public perception and how could a mission
best engage European citizens in tackling
these?” But many of the codesign sessions
By Tania Rabesandratana, in Brussels discuss research priorities “as if they were were standard panel discussions. And at-
hosting you at home,” in the words of Jean- tendees largely came from the policy circles

“W
e need to learn the art of listen- Éric Paquet, the commission’s director- and lobby groups that have helped shape
ing.” “There’s no taboo, no bad general for research and innovation. EU science policy in the past; laypeople
ideas.” “I can assure you your con- The shift comes at a time when grow- were virtually absent.
tribution will be considered.” ing populist movements are criticizing “My take is that the commission was
These sound like the messages the European Union as undemocratic and genuinely trying out a new format. ... There
you hear at a small-team brain- out of touch with the concerns of its citi- was a real willingness to listen,” says Jan
storming session—not a 3-day science pol- zens. Vandenberghe says the commission Palmowski, secretary-general of the Guild
icy congress with 5000 attendees. But the aims to reach beyond the usual “stakeholder of European Research-Intensive Universi-
encouraging words were spoken here last groups”—universities, researchers, busi- ties here. But it remains to be seen what the
week at the European Research and Inno- nesses—to include the interests and priori- commission will do with the input, he says.
vation Days, where policy bigwigs working ties of cities, pension funds, health insurers, The outcome of the codesign effort, along
with the European Commission invited the nongovernmental organizations, and science with an online survey that closes on 4 Octo-
audience to help shape Europe’s next 7-year museums, for example. Some attendees were ber, will feed into a “strategic plan” the com-
research funding program, Horizon Europe, skeptical. “It’s a search for legitimacy,” says mission is developing for the first 4 years of
set to begin in 2021. Their message: Any- one academic and government expert from Horizon Europe, whose proposed €94 bil-
one can help “cocreate” the program’s out- France. “Is it truly participatory? Not yet.” lion budget is still to be agreed on by the
lines. “Our instruction to staff was: Don’t The commission aims to fix problems in European Parliament and member states.
promote, listen,” says Kurt Vandenberghe, Horizon 2020, the current program, which Beyond helping define the clusters and
director for policy development and coor- a 2017 midterm review said is yielding missions, the exercise may help remedy
dination at the commission’s Directorate- good, incremental results but generating another problem identified by the Hori-
General for Research and Innovation here. few big breakthroughs and showing little zon 2020 review: Ordinary citizens hardly
It is a change of tone for the commission, tangible, large-scale effects on society. For know EU research programs. “It’s a prob-
PHOTO: EUROPEAN COMMISSION
whose research programs are often seen Horizon Europe, the commission aims to lem for the legitimacy of [research] bud-
as bureaucratic machines. The conference measure the effect of the whole program gets,” Vandenberghe says. That’s why the
took place in a drafty former car factory, or of a portfolio of projects, rather than for commission needs to promote its research
rather than in an anonymous commission single ones. It also wants the public to tell work more aggressively, Moedas empha-
venue, and it had the informal air of a tech it which research targets it values the most. sized last week. “Even if we do a bit of
startup convention. Civil servants in booths “We need more relevance and accountabil- oversell, don’t worry,” he said. “We are still
were instructed to welcome attendees to ity of science to society,” Vandenberghe says, underselling ourselves.” j

22 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004NewsInDepth.indd 22 10/1/19 5:34 PM

 NE WS | I N D E P T H

MICROBIOLOGY

DNA pushes back the microbiome frontier

New ways to culture microbes and decipher their chemistries make invisible world tangible

By Jon Cohen and Elizabeth Pennisi searchers a molecular tool for plucking out from the whole panoply of microbes in,
Saccharibacteria from the mix of microbes say, the mouth, gut, or an ecosystem—for

O
ver the past 15 years, researchers in human mouth fluids. To grow the few clues to the molecules produced by the
have come to appreciate how pro- cells they snagged, the researchers tried a microbes. Then, they made the mystery
foundly the diverse zoo of microbes wide variety of culture broths—variously molecules in well-studied microbes and
in the human gut, skin, and mouth made from body organs, sugars, soy, vita- tested their function. The work, reported
affects our health. But their identi- mins, and gastric juices—until they found a online in Science this week, “will greatly
ties and exactly how they exert their congenial growth medium, they report this expand our understanding of the chemi-
effects have remained mysterious. Now, two week in Nature Biotechnology. cal capabilities of microbiomes,” says
research groups have made this microbial “This is a wonderful study,” says Norman Lora Hooper, an immunologist at the
dark matter more visible, by starting with Pace, a microbiologist at the University of University of Texas Southwestern Medi-
its best-known aspect: its DNA. Colorado in Boulder who did groundbreaking cal Center in Dallas, who was not part of
One team harnessed DNA sequence in- work to discover new microbes in hot springs the project.
formation to isolate specific microbes they and other extreme environments. The new Donia’s team started by picking an en-
wanted to grow in lab dishes. The other work, he says, “uses thoroughly modern tech- zyme critical to making a particular type
used it to discover chemicals of molecule, then searching the
that microbes make to com- metagenome for the gene encod-
municate with each other and ing that enzyme. In the past, find-
influence their human host. By ing such genes in metagenomic
opening the way to a more de- data was daunting because those
tailed understanding of the tril- databases contain only short,
lions of microbes we contain, fragmentary DNA sequences.
the techniques could ultimately But the team developed what
lead to new treatments, says Eric Schmidt calls “an efficient ap-
Schmidt, a chemist at the Uni- proach to comb the microbiome
versity of Utah in Salt Lake City. for new chemicals.” Its computer
“Microbes make many of our algorithm divides a molecule-
best drugs, such as life-saving producing gene into small seg-
antibiotics,” he notes. More such ments. Then it tests those seg-
molecules could be waiting in ments and selects the best for
our own microbiomes, he says. searching for the gene sequences.
“It is amazing to think that the The sequences pulled out by
human microbiome has genes to the probe tend to come with se-
make countless complex chemi- quences of other enzymes needed
cals, but we don’t know what Microbes live in complex jumbles, making it difficult to isolate and grow to make the molecule.
those chemicals are.” individual species or determine what biomolecules they make. The researchers tested the
Mircea Podar, who studies evo- program by looking for aromatic
lutionary microbial genomics at Oak Ridge nology” to obtain uncultured species. “I find polyketides, a class of molecules that in-
National Laboratory in Tennessee, and his the study inspiring, something we couldn’t cludes the antibiotic tetracycline and the
co-workers used DNA to isolate and then even imagine in the early 1980s.” anticancer drug doxorubicin. Although soil
culture bacteria that, until now, have never Culturing previously obscure microbes microbes produce these compounds, no one
been grown in the lab. The team targeted should allow researchers to determine how had found them among the human micro-
Saccharibacteria, inhabitants of the human they function. It could also explain why biome’s products. The effort yielded 13 clus-
mouth. The dozen species living there make many strains won’t grow in mixed cultures. ters of polyketide-producing genes—all new
up less than 1% of the mouth’s microbiome “Once you separate your target organisms, to science—in the mouth, skin, and gut, the
PHOTO: STEVE GSCHMEISSNER/SCIENCE SOURCE

and are difficult to isolate and grow. then you can explore conditions under team reports. “These are not rare, yet no
Podar’s team first searched for genes which your microbe can flourish,” Podar one had seen these before,” Donia says.
in previously sequenced Saccharibacteria says. You might find, he says, that an ap- His group then put these sets of genes
DNA that likely code for proteins capable of parently “uncultivable” organism is actually into lab bacteria and tested the products.
jutting through the surfaces of cells, where inhibited by another species. None inhibited cancer cell growth, but
antibodies can “see” them. The research- Mohamed Donia didn’t want to wait two proved to be potent antibiotics. In the
ers then identified specific regions of those for a microbe to grow to find out how it human body, they may destroy microbial
surface proteins that would likely trigger worked. So the pharmacist at Princeton competitors or help protect the host from
strong antibody responses and injected University and his colleagues came up certain pathogens, Donia says—a possibility
these protein fragments into rabbits. Anti- with a way to comb through metagenomic that suggests bright prospects in medicine
bodies made by the animals gave the re- data—thousands of short DNA sequences for the microbial dark matter. j

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 23

1004NewsInDepth.indd 23 10/1/19 5:34 PM

24
FEATURES

ILLUSTRATION: DAVIDE BONAZZI/SALZMANART

 NE WS

SPLIT DECISIONS
How Brexit has already taken a toll on five researchers
By Erik Stokstad

23 June

O
n 23 June 2016, 34 million U.K. 2016 would immediately stop flowing. (How-
citizens took to the polls to vote United Kingdom votes in referendum to ever, Johnson has said he would fast-track
leave European Union.
on a simple, fateful question: visas for scientists, and the U.K. govern-
Should the United Kingdom re- 2017 29 March ment has long vowed to take over paying
United Kingdom informs European
main a part of the European Union of intent to leave, starting process. for existing EU grants.) “A no-deal Brexit
Union, or should it leave? Nearly would spell disaster for the country,” says
2018 25 November
52%—a majority of 1.3 million— European Union approves a deal with David Lusseau, a French-born behavioral
wanted out, and conservative United Kingdom for an orderly Brexit. biologist at the University of Aberdeen.
politicians vowed to carry out the 29 March
For some researchers, Brexit is not just a
2019
people’s wishes. Parliament rejects deal for third time. looming threat: It has already taken a toll
Sandra Arndt, a German biogeochem- on their lives and their science. Statistics
23 July
ist working at the University of Bristol, 2020 Boris Johnson becomes prime give the outlines. Last year, U.K. research-
couldn’t cast a ballot. So, she voted with minister, pledging to carry out Brexit. ers were involved in considerably fewer EU
her feet. In 2017, she got a job at the Free research projects than in the year before
University of Brussels and moved her fam- Troubling omens the Brexit vote (see middle graphic). Pres-
ily to Belgium. “Never in my wildest dreams In the years since the Brexit vote (top), U.K. tigious EU fellowships that give early-career
would I have imagined that one of the old- researchers have participated in a declining share of researchers 2 years of funding to work in
est democracies of the world—one of the grants from the main EU funding program (middle). another country offer another ominous in-
most tolerant, outward-looking countries, The country has also drawn a smaller fraction of dicator. From 2015 to 2019, the proportion of
one of the most pragmatic and successful postdocs from a top EU fellowship program (bottom). fellows choosing U.K. universities fell from
nations—would descend into such a dan- 33% to 22% (see bottom graphic). “None of
United Kingdom Germany France
gerous political mess,” Arndt says. Spain Italy this is good,” says James Wilsdon, a science
That mess is far from resolved, even policy expert at the University of Sheffield.
though the latest deadline for Brexit is Horizon 2020 participation But not all the news is bad. Sarah Main,
31 October. Last month, Parliament passed 14% director of the Campaign for Science and En-
a law effectively requiring Prime Minister 12 gineering, a London-based advocacy group,
Boris Johnson to seek an extension until says that after many years of flat or declin-
31 January 2020; EU leaders may grant the 10 ing budgets, U.K. politicians are signifi-
delay to give the United Kingdom time to 8 cantly boosting science spending in a bid to
approve a deal that would smooth the di- strengthen the country’s industrial prowess
vorce. But EU patience is wearing thin as 6 after Brexit. “We have broader and stronger
opponents in Parliament have repeatedly 4 political support for science than at any time
blocked that deal. And Johnson, who en- I can remember,” she says. In 2017, the gov-
2
tered office in July on a pledge to execute ernment set a 10-year goal of increasing R&D
Brexit with or without a deal, has said he’d 0 from 1.7% to 2.4% of gross domestic product,
“rather be dead in a ditch” than seek an- 2015 2016 2017 2018 2019* the average for wealthy nations.
other delay. A general election may offer Marie Skłodowska-Curie individual fellowships No one knows how it will all play out—or
the only path forward. 40% whether the initial bruises to U.K. science
The default trajectory is to crash out will turn into a deeper wound. “Like every-
of the European Union without any deal. one else, I’m holding my breath,” says Eske
30
Economists fear a no-deal Brexit could Willerslev, a geneticist at the University of
cause a nasty inflationary recession and, in Copenhagen who holds a half-time appoint-
the short term, food and medicine short- 20 ment at the University of Cambridge. “These
GRAPHICS: N. DESAI/SCIENCE

ages as trucks are delayed for days at the are the best research and intellectual envi-
borders. The government is even con- 10 ronments I have ever encountered in my ca-
cerned about civil unrest. For scientists, reer, and I pray it will stay that way.”
immigration rules imposed on Europeans 0
Here, Science presents stories of five re-
would make recruiting researchers harder 2015 2016 2017 2018 2019* searchers whose lives have already been
at all levels, and most EU research funding *Up to 20 September altered by Brexit.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 25

NEWS | F E AT U R E S

Preparing a university for Brexit

Uta Staiger, director of the European Institute at University College nently in the country after Brexit. The government had created an
London (UCL), didn’t exactly write the book on Brexit. But she edited app for that process, but it only worked on Android phones, so UCL
one. Last year, Staiger—an expert in the history of European inte- lent those to staff who needed them. Staiger herself, a German na-
gration and the role of emotions in politics—co-edited a 293-page tive, paid £1350 and swore allegiance to the queen to become a U.K.
scholarly book titled Brexit and Beyond: Rethinking the Futures of citizen, “just to be on the safe side.”
Europe. She calls Brexit “the most pressing issue in the country’s She and other administrators also worked to safeguard research.
postwar history.” To ensure EU grants would be underwritten by the U.K. government
That issue has already reshaped her job. The day after the refer- after a no-deal Brexit, details of every active research grant were
endum vote, Staiger joined UCL’s Brexit mitigation group—about entered into a web portal set up by UK Research and Innovation, the
15 top administrators and managers who funding agency that would take over the grants. The group contacted
meet monthly. Its first priority was to re- all 80 UCL research labs to find out what supplies might be threat-
assure students and staff in town hall meet- ened by no-deal disruptions—mouse bedding, for instance—and to
ings that the university would support its help arrange for stockpiles. UCL set up a legal presence in the EU 27,
international community. “We get quite required for the university’s 20 pan-EU clinical trials to continue after
emotional messages from staff and students,” a no-deal Brexit. The Brexit team also cataloged more than 4000
says Staiger, who speaks five European agreements that would be affected by data-transfer regulations, such
languages. Nationals from the other 27 EU as the exchange of research data with personal information.
countries besides the United Kingdom (EU So far, UCL has experienced minimal effects from Brexit, Staiger
27) make up 20% of academic staff, a typical says. Success rates for EU grants have stayed the same, as have
figure for U.K. research universities. numbers of staff and undergraduates from Europe. Recruitment of
“[It’s] the most By mid-2018, the risk of a no-deal Brexit postgraduates has also held firm, although a fellowship in biology
pressing issue in was rising, and Staiger was spending several and medicine attracted no applications from the EU 27 in the first
the country’s days a week dealing with the prospect. This year after the referendum. The task for universities, Staiger says, is
year, the Brexit team ran workshops to help to stay connected with European partners. “It will be crucial to main-
postwar history.” European nationals apply for settled status, tain the closest possible collaborations,” she says. “A no-deal Brexit,
Uta Staiger which confers the right to remain perma- in particular, would be a huge setback.”

An uneasy departure Van der Giezen recalls feeling stung when

former Prime Minister Theresa May said, “If
pay to keep successful EU grants going if
they were cut off in a no-deal Brexit. But like
you believe you are a citizen of the world, many others, van der Giezen was skeptical
When he checks the latest news about Brexit, you are a citizen of nowhere.” The process that the government could afford to keep
Mark van der Giezen’s blood pressure still of getting citizenship for his children, who that promise. “I would not like to be in the
shoots up—but at least he is reading from had been born in the country, also shook his position of opening my email on a Monday
a distance. In August, after 22 years in the faith in the U.K. government. For one child, morning and getting a message from the EU
United Kingdom, the Dutch microbiologist doing so involved £1250 in application fees and it says your funding has been cut off. And
moved his family, including two teenag- and a 5-centimeter stack of paperwork, then you have to look your postdoc in their
ers, to Norway and transferred his lab from including more than 20 years of financial eye and say, ‘Maybe you don’t have a job.’”
the University of Exeter to the University of records dredged from a bank archive. Van der Giezen made a list of countries
Stavanger. “The whole Brexit thing has been In 2018, because of Brexit, the CEFAS he’d rather live in—those with functioning
an emotional roller coaster,” he says, and the facility lost its status as the EU Refer- parliamentary democracies, such as Ger-
ride had become too wrenching. ence Laboratory for Fish and Crustacean many and Switzerland, he says—applied for
Van der Giezen studies how microbes Diseases, and the European Union trans- university jobs, and accepted the position in
affect human and animal health, particu- ferred the role to a lab in Denmark. Six other Norway. The move was trying. Because Nor-
larly livestock and fisheries. In 2017, he had reference labs, as well as the European way is not in the European Union, customs
become head of a collaborative effort on Medicines Agency, left the country. It felt like rules required that every piece of imported
aquatic diseases with the U.K. govern- a sign of turmoil to come. lab equipment be listed with date of pur-
ment’s Centre for Environment, Fisheries At Exeter, van der Giezen and colleagues chase, purchase price, and current value. He
& Aquaculture Science (CEFAS). Recently, were told to keep applying for EU funding has yet to sell his house in Exeter because
he helped improve diagnostics for the most because the U.K. government said it would the initial buyer backed out.
problematic parasite of crayfish in Europe. Norway has positives: It is a wealthy
He liked his job and life in Exeter. country with beautiful scenery, nearby
Then came the 2016 referendum cam- “The whole research opportunities in aquaculture, and
paign, and “cracks started appearing in an international community drawn by the ILLUSTRATIONS: N. DESAI/SCIENCE
Brexit thing has
British society,” van der Giezen says. Some country’s oil industry. To offset the higher
foreign lab members were told on the street been an emotional cost of living, Stavanger offered a 60% pay
to speak English or go back to their country. roller coaster.” raise. Van der Giezen says he is optimistic
“People told me, ‘Don’t worry, you will be Mark van about his prospects for research, but also
fine.’ But that made me feel uneasy. Is it der Giezen clear-eyed about what he has left behind.
because I’m white?” “The U.K. is a scientific hotbed,” he says.

26 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 Staying away Betting on the United Kingdom
Brexit convinced James Wasmuth not to go Isabel Rosa, a Portuguese landscape ecologist, weighed the prospect of her first permanent
home. He grew up in the United Kingdom, job against concerns about Brexit—and took the plunge. In spring 2018, while finishing a Marie
near Bristol, and he and his wife earned Skłodowska-Curie postdoctoral fellowship in Leipzig, Germany, she had applied for five posi-
Ph.D.s at the University of Edinburgh, where tions and got two offers. She took the one in the United Kingdom, at Bangor University. “The
he designed software to interpret genomic position here in Wales was exactly what I would like to be doing in academia.”
data from parasitic worms. They moved Rosa builds computer models of changes in land use, an interest sparked by Portugal’s fre-
to Canada in 2006 as postdocs and won quent wildfires. For her Ph.D., at Imperial College London, she had developed a model to predict
faculty positions at the University of Calgary. deforestation in the Amazon on the basis of historical patterns of land use and socioeconomic
With a postdoc and students, Wasmuth factors such as investment in protected areas, roads, or agriculture. She has since worked to
studies worm evolution. The lab also looks generalize the model, one of only a few to predict both where and how fast land use will change.
for drugs to control the worms, a major The job in Bangor matched her interests and offered 3 years of
problem for the cattle industry. Last year, startup funding. Recently, she sent a proposal to the Natural Environ-
he applied for a job in the Cambridge, U.K., ment Research Council to study reforestation in Wales, a way to slow
area; his wife considered looking for a bio- climate change. She has also submitted proposals to Horizon 2020,
tech job there. The work was appealing, and the European Union’s main funding program.
the move would bring the family closer to Rosa says she has never felt unwelcome. “I love living in the U.K. and
his parents and in-laws, who live in France. my job here, and I’d be happy to stay forever,” she says. But when she
He received an offer, but after a trip to the visits family in Portugal, she makes a point of returning with olive oil.
United Kingdom, he turned it down. The job Her only complaint about the United Kingdom is the food.
would have given him the chance to work That and Brexit. “It’s hard to be optimistic, especially under a no-deal
with leaders in genomics and bioinformat- Brexit scenario,” she says. In addition to professional anxieties—access
ics, but he and his wife had doubts about “There are these to EU funding and collaborations—Rosa wonders whether a no-deal
moving. “We kept thinking, ‘What’s the dark clouds always Brexit will make it hard to travel to Europe with her dog. Sometimes,
future? What will it be like in 10 years?’” Al- Rosa says, she feels she has traded the uncertain future of a postdoc
though the Cambridge region voted strongly over our heads.” for the uneasiness of being a scientist in Brexit Britain. “There are these
to remain in the European Union, Wasmuth Isabel Rosa dark clouds always over our heads.”
and his wife felt uneasy when visiting his
parents in a small town. “The undertones
were unfriendly. You’d hear people being a
bit spiteful,” he recalls. He and his wife won-
dered whether they wanted to raise their
young daughters in the United Kingdom.
Mourning lost ties
“This whole divisiveness isn’t healthy.” Ornella Corazza, an Italian scientist who studies addiction at the
The possibility of a post-Brexit recession University of Hertfordshire, had spent more than 10 years in the United
—and doubts that the country would remain Kingdom before the Brexit vote. “After the referendum, I felt very de-
flush enough to fully fund science—also moralized,” she says. “I didn’t feel valued in this country.” Her university
factored into Wasmuth’s decision. So did helped her get U.K. citizenship, but her scientific connections to Europe
worries that it would be hard to lure grad began to fray. For the first time, her grant proposals to the European
students and postdocs to a U.K. lab. Union were rejected. She never knew whether that was because of
Overall, U.K. universities are still doing Brexit. This summer, however, brought a clear signal.
well at recruiting faculty, according to Corazza takes part in a project funded by European Cooperation in
government figures. But the number of new Science and Technology (COST). The €520,000 grant supports dozens
Ph.D. students arriving at major research “[The United of partners studying problematic uses of the internet, such as exces-
universities from the EU 27 dropped by 9% Kingdom] is sive video gaming. On 2 April, COST told all U.K. coordinators to transfer
from 2016 to 2017, and lab leaders at even leadership of the grants to a partner in another nation, citing legal
the most prestigious institutions describe
already losing its constraints that would prevent the European Union from sending money
unprecedented difficulty in attracting Eu- leadership …” to the United Kingdom after a no-deal Brexit.
ropean postdocs. Wasmuth says graduate Ornella Corazza The overall project leader, psychiatrist Naomi Fineberg, also at the
school friends of his who are now U.K. aca- University of Hertfordshire, handed the reins to colleagues in Spain. That
demics live with extra stress: “Their anxiety change rankled Corazza, who points out that leaders of large research projects tend to get more
levels are 10-fold higher than mine.” grant money as well as a chance to network widely and build leadership skills. “It’s your idea. It’s
your vision,” Corazza says. “It’s extremely frustrating when it’s taken away.”
U.K. researchers have long done well in securing such positions. But the fraction of Horizon
“We kept
ILLUSTRATIONS: N. DESAI/SCIENCE

2020 collaborations headed by U.K. scientists fell from 12% in 2015 to 6% this year, by far the
thinking: largest drop among the six countries that coordinate the most projects.
Several U.K. scientists say invitations to join EU consortia have dried up. Some researchers,
‘What’s the like Corazza, no longer bother applying for EU grants, because they don’t know whether the
future?’” United Kingdom will be eligible. Instead, Corazza is expanding her collaborations into the United
James States and elsewhere. But that doesn’t diminish her sense of loss from weakening ties to Europe.
Wasmuth The United Kingdom, Corazza says, “is already losing its leadership, and it will be just one of
many countries that does excellent research.” j

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 27

 INSIGHTS
LET TERS
Dear NextGen Voices,
One of my lab responsibilities as a graduate student is to train
another graduate student to do certain parts of the team’s
research. The graduate student often neglects his duties,
which means I have to do the work myself. Despite meeting
with him to demonstrate exactly how to do the tasks, offer-
ing to answer any questions he has, and reminding him that
many of these tasks are time sensitive, I have not been able
to persuade him to complete his assignments. I always cover
for him, both because failure to do these tasks puts my own
research at risk, and because my principal investigator (PI)
has trusted me to train this student, so his failures reflect
poorly on me. How can I more effectively train this student and
also protect my research and my reputation as a manager?

Sincerely,
“Trying to Manage”

NEXTGEN VOICES: ASK A PEER MENTOR

NextGen advises “Trying to Manage”

Peer mentors can provide crucial support to young scientists by asking questions to facilitate further
reflection, assuring mentees that they are not alone, and suggesting potential next steps. We asked our readers
to act as peer mentors to “Trying to Manage,” the anonymous student who posed the question above. Follow
NextGen Voices on Twitter with hashtag #NextGenSci. Read previous NextGen Voices survey results at
https://science.sciencemag.org/collection/nextgen-voices. —Jennifer Sills

Share the big picture Is the student you are responsible for
facing any specific problem in the lab that
How can you make this student feel
like part of a team? When I supervised
Does your mentee understand the big
is hindering his tasks, or is this an issue interns as a young scientist, it made a
picture? As a fourth-year graduate student,
of motivation? A few years ago, as a lab huge difference when they felt their job
I have mentored eight students. Many
technician, I needed to delegate duties was important, not just a chore. Students
challenges I have seen—including procrasti-
nation, experimental failure, and neglect of that seemed repetitive and boring to lab are used to working independently; it
interns. However, when I explained that is important to show them the benefits

ILLUSTRATION: DAVIDE BONAZZI/SALZMAN ART

responsibilities—stem from a lack of under-
standing, a failure to see the big picture that the tasks were essential for disease diag- of assuming responsibilities as part of a
results in mentees feeling inadequately pre- nosis, they began to perform the duties team. During a lab meeting, you should
pared, undervalued, or even used. Mentees’ with greater enthusiasm. You can help announce that your mentee is fully
self-esteem and confidence thrive when they motivate your student by helping him to trained in a given task and will be fully
understand how their piece contributes to a identify the importance and meaning of responsible for it going forward.
larger puzzle. his work. Monika Lewinska
Douglas Steinhauff Joel Henrique Ellwanger Biotech Research and Innovation Centre,
Department of Biomedical Engineering, Department of Genetics, Universidade Federal do Faculty of Medicine, University of Copenhagen,
University of Utah, Salt Lake City, UT 84112, USA. Rio Grande do Sul, Porto Alegre, Rio Grande do Sul, Copenhagen, 2200, Denmark.
Email: dbsteinh@gmail.com 91501-970, Brazil. Email: joel.ellwanger@gmail.com Email: monika.lewinska@bric.ku.dk

28 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

How does your trainee hope to benefit persuade him to critically analyze his mentors and their willingness to show vul-
from the training you are providing? results before approaching your adviser. nerability have helped me understand that
During my first year in graduate school, Moreover, great mentor that you are, you making a mistake in an experiment is not
I assisted in a senior student’s research are training him to successfully present the end of the world. If you act as a supervi-
as part of my training. Because we never experimental findings. sor, the student may think that asking you
discussed authorship, I was not listed Buddini Iroshika Karawdeniya for help will diminish his worth in your eyes.
as a coauthor on the resulting publica- Department of Mechanical Engineering, Instead, invite him for coffee, talk about
Southern Methodist University, Dallas, TX 75206, the project (especially the exciting parts),
tion. Later, my application for a postdoc
USA. Email: bkarawdeniya@smu.edu
was rejected because I did not have any and ask for his input. Make him part of the
publications to demonstrate my exper- process. Once he is invested in the outcome,
Have you set clear expectations for your he will be more likely to participate.
tise in the techniques we used. As long
mentee and explained why it is impor-
as your trainee is unsure whether his Aminata Coulibaly
tant that he meet these expectations? It
contribution to your research work will Departments of Neurology and Neuroscience,
can be difficult as a junior researcher to University of Virginia, Charlottesville, VA 22908,
be adequately recognized, he will not
be in charge of others, particularly if the USA. Twitter: @AmiCoulibaly18
be motivated to complete any task you
mentees’ working styles differ from your
assign him. You should clearly explain
own, which has always been the case for Have you asked the student if he is okay?
that his level of contribution will deter-
me when I have mentored students. The I remember a mentee who had difficulty
mine whether he is listed as an author.
first steps in mentoring and supervising finishing his work. When his mentor
Edmond Sanganyado are to get a baseline assessment of your learned that he had an ill family mem-
Marine Biology Institute, Shantou University,
Shantou, Guangdong, 515063, China. mentees and to set clear expectations. ber, he allowed the mentee flexible work
Twitter: @ESanganyado Once the expectations are established, hours. The mentee came to the lab when-
continue to provide feedback to let the ever suited him best and soon caught up.
mentees know whether they are meeting Try to meet more often with your mentee.
Clarify expectations expectations or what else they need to do The main purpose of a mentoring rela-
Does your lab have clear standard operat- to succeed. tionship is mutual learning and growth,
ing procedures (SOPs)? When I started Naike Bochatay and the more you invest, the more you
my own lab, I found that even skilled Department of Pediatrics, Division of Pediatric will gain from the experience.
Critical Care, University of California, San
students sometimes did not train others Francisco, San Francisco, CA 94143, USA.
Shaima Ibrahim
Department of Chemistry, The American
effectively because they were too busy Twitter: @nbochatay
University in Cairo, Cairo, 11321, Egypt.
with their own experiments to provide Email: Rxshaima@hotmail.com
all the necessary details. I established Are you truly helping the graduate student
Lab SOPs explaining all experimental become a better professional by con- Have you thought about other reasons the
protocols, and at the beginning of each tinuously covering for his failures? I have student may be neglecting his responsibili-
new term, lab members (including new experienced similar situations in which ties? When working with people whom I
students) gather to update and improve those in my charge were unwilling to fully oversee, I often find it useful to get to know
the protocols together. Ask your PI if you accept and execute their responsibilities. them as individuals. This helps me to tailor
can take the lead in creating SOPs for In each of these situations, I knew that the sort of advice and instruction I give
your research. This will show initiative the student was fully capable; however, them so that we can all meet the specified
and minimize the time required to train they were not giving their full attention deadline. Perhaps your student feels over-
new lab members, for you and for those and focus to the tasks at hand. Acting in whelmed by the workload or never learned
in your position in the future. the best professional interests of those in proper time management as an under-
Yongsheng Ji our charge sometimes requires us to put graduate. Perhaps your method of offering
Department of Human Parasitology, Anhui aside feelings of social awkwardness. Talk help has to change to accommodate this
Medical University, Hefei, Anhui, 230032, China. directly to this student: Set clear expecta- individual’s background.
Email: jiyongshengkey@hotmail.com
tions, enforce rules, and, if possible, impose Déna Skye Jansen
consequences on his errant behavior. Department of Environmental & Geographical
Have you asked the student to demon- Science, University of Cape Town, Cape
Eric Britt Moore Town, Western Cape, 7700, South Africa.
strate how the work is done or to make a Department of Agronomy, Iowa State Email: djvirgo17@gmail.com
list of things he plans to do each day? As University, Ames, IA 50011, USA.
a graduate student, I mentored graduate, Email: ebm256@iastate.edu
undergraduate, and high-school students, What is your student’s current stake in
and I faced similar situations in which the research? Does he understand how
the mentee did not realize the time Build a relationship the lab work contributes to his personal
urgency of the work. Email your adviser Instead of showing the student how to do research goals? Often, new members come
(and copy your mentee) to arrange regu- the experiment, have you tried doing the into a lab with their own research inter-
lar project update meetings, and ask your experiment together? Because English is ests, which may not align precisely with
mentee to present the findings of his part not my first language, I used to be so afraid tasks that need to be done in the lab. Talk
himself. Emphasize that you want the to the student about his own research
of being seen as dull-witted that I would
interest, and help him find a way to relate
adviser to recognize his work. This will not ask questions. I would pretend to
his interest to the lab work.
remind him of his responsibility to com- understand what was asked of me and then
plete the work and the potential praise spend hours online trying to figure out Joshua I. James
Legal Informatics and Forensic Science Institute,
he could earn when successful. From that what to do. It made learning a laborious Hallym University, Chuncheon, Gangwon, 24252,
point on, do not cover for him. This will and frustrating process. The patience of my South Korea. Email: joshua.i.james@hallym.ac.kr

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 29

INSIGHTS | L E T T E R S

Motivation may be the problem, but supposed to be learning from your adviser nobody any good. I encourage you to
what are the alternative hypotheses? is how to mentor others. Try discussing marshal your empathy one more time
Before my doctoral training, I directed this situation with your adviser from this and talk to your labmate about why
a clinical center. The company’s CEO perspective. Your thesis committee may he’s floundering—what looks like lack
would have been rightfully worried if I also be able to step in and help you protect of interest may be something else. After
had reported smooth sailing week in and your research time. that, you’ve done what you can as a boss;
week out. Managing people is messy. I Beth McKinnon Adamowicz it’s time to ask your PI for help with this
instilled confidence by reporting sub- Department of Genetics, Cell Biology, delegated responsibility.
stantive problems as they arose, sharing and Development, University of Minnesota, Shervin Fatehi
Minneapolis, MN 55455, USA. Department of Chemistry, The University of
my progress with addressing ongoing Twitter: @BethA_z Texas, Rio Grande Valley, Edinburg, TX 78539–
problems, and acting confidently enough 2909, USA. Email: shervin.fatehi@utrgv.edu
to ask for support. In your laboratory, Is there something about your PI that
you systematically identify and modify makes you hesitant to discuss this situ- Would it be helpful to discuss your situa-
functional variables within your control. ation with him or her? Your dilemma tion with a senior colleague or with your
A scientific approach is no less useful appears to be that you were asked to PI? I have experienced similar situations,
when managing people. A fact-finding train this person (which you seem to have both as a grad student and as a postdoc.
mission might consist of seeking to done), but you now are trying to manage An important part of being a successful
understand the student’s situation over him, which you were never taught how supervisor is recognizing when you need
coffee, explaining yours, and persuad- to do. Most scientists (possibly including help. Document your experiences with
ing him that you are collaborators with your PI) have never gotten any manage- your coworker, keeping things as factual
a common objective. Communicate these ment training and are at a loss when as possible: dates, experiments, results
efforts with your adviser. Impressive faced with a team member who isn’t (or lack thereof ), and your actions.
managers do not hide problems; they show doing their job. Is there something going Then request a meeting with your PI
that they have a plan for managing them. on in this person’s personal life that’s and explain the situation. Be clear that
David M. Cole you are looking for a way forward with
Interdisciplinary Program in Neuroscience, Utah
preventing him from focusing? Has he
State University, Logan, UT 84322–7000, USA. lost interest in this project or the lab? It your student, keeping in mind that he
Email: dcole@observechange.org is the PI’s job to ask these questions, not may have a different view of events.
yours. But your PI can’t do that unless Supervising other students is a good step
you let him or her know what’s going on. in growing up as a researcher, but it pre-
Ask for help By keeping silent about the situation and sumes strong and constructive support
What can you learn about mentoring doing the work yourself, you are facilitat- from supervisors.
from this challenge? I experienced a sim- ing the student’s behavior. There is no Maria Helena Holmstrom
shame in not knowing how to do some- Department of Neuroscience, Gothenburg
ilar situation with a student who didn’t University, KlinKem, Sahlgrenska Akademin, GU,
respond to feedback. The best advice thing (managing others) that you were Gothenburg, 40530, Sweden.
I received was to use this situation as never taught. Email: maria.holmstrom@gu.se
an opportunity to learn how to men- Carl M. Cohen
tor rather than an interpersonal drama Science Management Associates, Newton, MA Why do you think this graduate student
02458, USA. Twitter: @carlmcohen
to solve, and to ask my mentor how he is not completing his duties? If there
would handle the situation. Framing the aren’t personal issues hampering his
situation this way made it less an issue What’s your first priority right now— work, then you should provide clear, doc-
of complaining about a fellow student becoming a better manager or becoming a umented instructions moving forward.
and more of a skill-building discussion better scientist? I’ve experienced a similar By email (so you have a paper trail), give
about learning how to manage others. situation, but I was the unproductive him a clearly bulleted list of assignments
labmate. Although I sincerely appreci- with generous deadlines. Note that you
Anna Lipkin
Department of Neuroscience, University
ate the work my peers put in to try to are happy to meet with him to provide
of California, San Francisco, San Francisco, help me succeed when I was struggling, a technique refresher or discuss time
CA 94158, USA. Twitter: @anna_lipkin what I really needed was some tough love management, provided he gets in touch
from my adviser and some encourage- by a specified date; otherwise, you will
Can you ask your PI for advice on how ment to get counseling to help overcome assume that he will be able to complete
to mentor this student more effectively? emotional obstacles to my success. Let’s his duties. If he still shirks his work, docu-
It seems like you’re in a tough spot: emphasize the positives: Your PI trusts ment the shortcomings and go to your PI.
You’re this graduate student’s supervi- in your lab skills, your ability to explain If you document the issue and what you’ve
sor and therefore accountable for his techniques thoroughly and safely, and already tried to fix it, you will come across
performance, but you’re simultaneously your overall reliability—that’s huge! as a capable manager stuck with a prob-
his peer and therefore have no power You’ve been working hard to repay that lem trainee. Continuing to cover for this
to make him do any work. This is an trust by making your mentoring rela- student is unsustainable, and it should not
unfair situation that you’ve been put in, tionship with your labmate a success, be your burden to bear alone!
particularly given that his failures pres- and you’ve been patient and empathetic, Rosa Li
ent risks to your own research. It’s worth which is wonderful. But look: You’ll Duke Institute for Brain Sciences,
remembering that even though you’re have the rest of your career to hone your Duke University, Durham, NC 27710, USA.
Twitter: @rosasbrain
the senior person in this relationship, skills as a manager. Right now, you’re
you too are still a trainee. As a gradu- sacrificing research progress to hide your
ate student, one of the things you’re mentee’s failure to thrive, and that does 10.1126/science.aaz5828

30 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 A simulation shows gas filaments (colored pink and
PERSPECTIVES blue) around a massive galaxy cluster, that absorb
and emit light. The brightest parts of the filaments
are detected with ground-based telescopes.

(60%) of the Universe’s content of normal

matter (baryons). Spectroscopic observations
of distant bright QSOs revealed a “forest” of
Ly-a absorption lines, indicating neutral hy-
drogen at many distances along the line of
sight. This observation is a signature of the
cosmic web. These initial indications were
augmented by more targeted absorption line
studies looking for evidence of filaments and
clouds in the halos and nearby environments
of galaxies. These studies, greatly increased in
sensitivity with the installation of the Cosmic
Origins Spectrograph on the Hubble Space
Telescope, led to detections of hydrogen and
other elements in the immediate circumga-
lactic medium of nearby galaxies. Absorption
line studies have their limits: Typically, only
a single sight line is available through any
one galaxy, which limits our ability to make
inferences about the environment of any in-
dividual galaxy.
The next step is direct detection of emis-
sion from the cosmic web. The Multi Unit
Spectroscopic Explorer (MUSE) on the Very
INTERGALACTIC MEDIUM Large Telescope and the Keck Cosmic Web
Imager on the Keck II telescope are both inte-

Observing the cosmic web gral field units (IFUs) designed to do imaging
spectroscopy across wide fields. They both
were designed to allow for the observation
The faint signature of gas filaments in the intergalactic of the faintest structures known, including
cosmic web emissions. This requires exqui-
medium is finally detected site sky subtraction (frequently the emission
detected is only a few percent of the level of
By Erika Hamden mic web, surrounding and illuminated by a the sky background), high throughput, and
cluster of forming galaxies. a well-understood point spread function to

A
strophysics is an exercise in expand- The galaxies themselves act as cosmic subtract the bright central galaxy or galax-
ing the known horizon, pushing the flashlights as the ultraviolet (UV) photons ies. The wavelength information generated
limits to observe fainter objects, more from recent star formation and quasi-stellar across the field is also integral to identifying
distant galaxies, and stranger things. objects (QSOs, ultraluminous active galactic these very faint structures.
This quest recently led to the design nuclei) within the galaxies provide the illu- Since coming online, these instruments
of clever instruments to make obser- minating source for intercluster medium hy- have opened a new discovery space in ob-
vations not just of galaxies but of the faint, drogen fluorescence via Lyman-alpha (Ly-a) servations of the circumgalactic medium,
barely detectable gas around galaxies. This emissions. Early observations of huge clouds the brightest parts of the cosmic web near-
circumgalactic and intergalactic gas is a of hydrogen around single or paired QSOs in- est to galaxies. The very wide field of view
prediction of the lambda cold dark matter clude UM287 (aka the Slug Nebula) (2, 3) and of MUSE, when combined with careful sky
(LCDM) model. In this model, primordial HS1549+19 (4). This hydrogen is potentially subtraction, has extended these observations
PHOTO: ILLUSTRIS COLLABORATION/ILLUSTRIS SIMULATION

hydrogen created in the Big Bang collapses flowing into the QSO host galaxy and provid- into intergalactic space (see the image). The
into sheets, which in turn collapse into fila- ing fresh fuel for new stars. Since these first observations from Umehata et al. are just
ments. Galaxies form where filaments either detections, observations of Ly-a–emitting the tip of the iceberg. Other similarly bright
cross or are overdense. The gas filaments feed structures have been made around galaxies at structures will eventually be identified and
galaxies as they grow. However, the evidence a range of distances, of QSO subtypes (radio- observed and, with longer observation times,
for this web of gas has remained circumstan- loud, radio-quiet, type II), and of overdensi- will reveal even fainter parts of the web.
tial. On page 97 of this issue, Umehata et al. ties. These observations paint a picture of Confirming the existence of these gas
(1) present evidence for a direct detection of the ubiquitous presence of hydrogen around clouds leaves open questions about their
light from the very brightest part of this cos- QSOs and galaxies in the early Universe. exact nature. The velocity of parts of these
These observations have been a long time observed clouds remains uncertain, given
in coming. Since LCDM was first formalized the resonant nature of the Ly-a line, because
Department of Astronomy, Steward Observatory,
University of Arizona, Tucson, AZ 85721, USA. in the late 1990s, a key prediction was a cos- disentangling the relative effect of resonance
Email: hamden@email.arizona.edu mic web that contains a substantial fraction from the true velocity signature is difficult.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 31

1004Perspectives.indd 31 9/30/19 12:56 PM

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

Eventual observation of other emission lines, CHEMISTRY

especially ones that trace a different phase of
gas, should validate the velocity structures
inferred by the Ly-a emission and create a
multidimensional picture of the galaxy sur-
RNA nucleosides built in
roundings. With this eventual multidimen-
sional view, we will trace gas inflows and
outflows, identify hot and cool gases, and
one prebiotic pot
map diffuse and dense structures. Origins-of-life research models integrate the synthesis
The determination of what is part of the
cloud and what isn’t is still uncharted ter-
of RNA building blocks
ritory: Each group of researchers makes a
determination of the cloud edge for their By Nicholas V. Hud and David M. Fialho “Prebiotic” processes are those that could
own observations. Developing a constant, conceivably have contributed to the origin

F
systematic way to identify these structures or decades, the RNA world has been of life on the primitive Earth. A fundamen-
and map their extents is important for con- one of the most influential hypotheses tal difficulty with the prebiotic formation of
sistency, the ability to perform comparative regarding the origins of life on Earth. RNA is that the RNA nucleobases (the pu-
studies, and understanding of the frequency In this hypothetical era before the rines, adenine and guanine; and the pyrimi-
of these structures throughout cosmic time. emergence of DNA and proteins, “life” dines, cytosine and uracil) are reluctant to
Future extremely large telescopes will image on primordial Earth consisted of RNA spontaneously form a bond with the sugar
Ly-a clouds to even lower surface brightness molecules that both store genetic informa- ribose; this glycosidic bond holds the nu-
limits and densities. To truly understand the tion and catalyze self-replication reactions cleobases to the phosphate-ribose backbone
environment around a galaxy, we will eventu- (1). Despite its popularity with researchers, of RNA. Within the past decade, a focus
ally need the community to make a clear defi- the RNA world hypothesis has suffered from on the pyrimidine nucleosides yielded two
nition of the cloud edges, potentially linking its own origins conundrum: A mechanism for completely different, experimentally sup-
that definition to simulations. the simultaneous prebiotic synthesis of RNA ported approaches for prebiotic nucleoside
Direct imaging of the cosmic web is limited nucleosides from both the purine and pyrimi- synthesis. In one, the pyrimidine base is
because the telescopes we use are ground- dine families has long eluded scientists (2). built along with ribose in a concerted fash-
based and can only observe these structures Although disparate prebiotic syntheses have ion (3). In the other, pyrimidines slightly
around very distant galaxies in the early Uni- been demonstrated for the two classes of different from those found in modern RNA
verse. One exception is an experimental bal- RNA nucleosides (3, 4), no single geochemi- formed alternative nucleosides that might
loon-borne far-UV spectrograph, FIREBall-2 cal scenario has generated both. Now, on have served as subunits of an ancestral
(5). The FIREBall mission can only observe in page 76 of this issue, Becker et al. (5) report form of RNA (6). Unfortunately, neither ap-
a very narrow range of light but can observe on chemistry that accomplishes this long- proach suggested a corresponding prebiotic
gas clouds from nearer galaxies, closer to the awaited goal. synthetic mechanism for purine nucleosides
present day. Probing the full history of the
Universe by observing gas emissions requires
a probe-class UV satellite with an IFU that is A possible solution to a very old problem
able to make similar observations across the Nucleosides, one of the building blocks of RNA, do not form spontaneously in a plausibly prebiotic manner
UV. Observing these gas structures in the UV from their chemical substructures, nucleobases and ribose. Becker et al. show that when non-natural
is easier in some ways because the sky back- nucleobases are mixed with ribose and subjected to environmental wet-dry cycles, intermediate molecular
ground is orders of magnitude lower (6, 7) species are formed that can easily be converted to the natural nucleosides in model prebiotic reactions.
and the emission signature is much brighter.
However, it requires getting a large UV tele-
Purine precursor Purine nucleoside Purine nucleoside
scope into space, which is challenging for lo- precursor
gistical and financial reasons.
These observations of the faintest, largest
structures in the Universe are a key to under-
standing how our Universe evolved through
time, how galaxies grow and mature, and
Dry
how the changing environments around gal-
axies created what we see around us. These
observations are of fundamental importance Unscrambling
to understanding the nature and composi-
tion of the Universe. j
REFERENCES AND NOTES Ribose
GRAPHIC: A. KITTERMAN/SCIENCE
1. H. Umehata et al., Science 366, 97 (2019).
2. S. Cantalupo et al., Nature 506, 63 (2014).
3. D. C. Martin et al., Nature 524, 192 (2015). Wet
4. D. C. Martin et al., Astrophys. J. 786, 106 (2014).
5. E. T. Hamden, Proc. SPIE 10982, 1098220 (2019).
6. Ch. Leinert et al., Astron. Astrophys. Suppl. Ser. 127, 1
(1998).
7. M. Giavalisco et al., STScI Instrument Science Report
WFC3-ISR 2002 – 02 10.
10.1126/science.aaz1318 Pyrimidine precursor Pyrimidine nucleoside precursor Pyrimidine nucleoside

32 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004Perspectives.indd 32 9/30/19 12:56 PM

 that is likely to succeed to a similar degree. been regular events on all exposed land of NEUROSCIENCE
Stepping up to this challenge, Becker et al. the prebiotic Earth, as they are driven by re-
(4) demonstrated, in 2016, a plausible pre-
biotic synthesis of purine nucleosides from
formamidopyrimidines (FaPys), a sort of “un-
liable fluctuations in temperature (e.g., day-
night or seasonal cycles) and water activity
(e.g., rainfall, tides, or geothermal activity in
Learning
raveled” form of a purine that readily reacts
with ribose in water to form nucleosides (4).
Once the FaPy nucleoside is formed, raising
hydrothermal fields). With the simultaneous
formation of purine and pyrimidine nucleo-
sides, Becker et al. provide further evidence
birdsong by
the solution’s pH induces a ring-closure reac-
tion that produces the intact purine nucleo-
side. Although this approach was considered
for the relevance of prebiotic wet-dry cycles
(see the figure).
In 2004, Orgel, a pioneer in the field of pre-
imitation
an important advance in the field of prebi- biotic chemistry, reviewed the previous four Transforming sensory
otic chemistry, there was no obvious way to decades of efforts to uncover plausible pre- information into vocal
reconcile this synthesis with the previously biotic reactions that would support the exis-
demonstrated disparate prebiotic syntheses tence of an RNA world (2). At that time, Orgel imitation allows young
of pyrimidine nucleosides by a dissimilar argued that the synthesis of nucleosides was finches to sing
chemical pathway (3). How could two totally “the weakest link” in the proposed chain of
different synthesis routes to the purine and prebiotic reactions that lead to RNA prod-
pyrimidine nucleosides occur in unison to ucts. As mentioned above, in the ensuing By David F. Clayton
produce the necessary subunits of RNA, pre- years, model prebiotic reactions were devel-

I
sumably in the same prebiotic milieu? oped in which extant pyrimidine nucleosides mitating an expert is seemingly a
In the new study, Becker et al. present an are prepared by building the nucleobases straightforward type of learning, one at
alternative prebiotic synthesis of pyrimidine along with the sugar in a concerted reaction the root of language, music, and other
nucleosides under geochemical conditions (rather than joining preformed bases and a culturally transmitted behaviors. But
designed to be compatible with their previ- sugar) (3). Furthermore, organic chemists learning to copy an action requires the
ously published synthesis of the purine nu- discovered heterocycles that react with ribose nervous system to perform some rather
cleosides (5). Using a similar strategy for the in wet-dry cycles to provide possible ances- unusual transformations. The expert’s ac-
pyrimidine nucleosides, the authors formed a tral RNA nucleosides (6). Now, Becker et al. tion must first be perceived and repre-
heterocyclic precursor (which can be viewed present what is arguably the most direct pre- sented as a sensory pattern. This sensory
as a “scrambled” pyrimidine nucleobase) in a biotic route to the RNA nucleosides. pattern must then be transformed into a set
model prebiotic reaction mixture, which then Given this progress, do origins-of-life re- of motor patterns sufficient to reproduce
reacted with ribose to form a non-natural searchers consider the problem of prebiotic the desired goal. Birdsong provides a model
ribonucleoside not found in modern RNA. nucleoside formation to be solved by a variety for this sort of “inverse” learning: Over
The subsequent addition of specific prebiotic of approaches? Alas, like many seminal sci- a span of ~2 months, a young zebra finch
reagents—an iron catalyst and hydrogen sul- entific problems, crucial advances often yield listens to the song of an adult tutor and
fide—induced a rearrangement in the hetero- more questions than answers. It is now fair gradually learns to reproduce it. On page
cyclic moiety to yield a cytosine nucleoside. to ask which experimentally demonstrated 83 of this issue, Zhao et al. (1) dissect the
This product can then be converted, under route reflects the true historical origin of neural circuitry involved in the sensorimo-
prebiotic reaction conditions, to a uracil nu- RNA in the primordial soup. Also, did the tor transformation underlying this process
cleoside (7). various proposed synthesis methods contrib- of imitation.
The non-natural nucleoside precursors ute to the emergence of life simultaneously A mature birdsong is a precisely struc-
used by Becker et al. for both the RNA purine or at different times during the hypothetical tured performance, consisting of a few short
and pyrimidine nucleosides are formed by RNA world era? Grappling with such issues sound elements (“syllables”) repeated in an
wet-dry cycling—a method used for inducing is far more gratifying than wallowing in the exact order, each bout lasting 1 to 2 s (2).
chemical reactions by evaporating an aque- despair that overshadowed the field two de- The syllables can have different complex
ous solution of the reactants at increased cades ago, when many prebiotic chemists be- acoustic structures, with stacks of harmonic
temperatures. The dried reactants coalesce lieved that a plausible, prebiotic solution to frequencies and varied amplitude enve-
into a highly concentrated, low–water activ- the synthesis of nucleosides was impossible. lopes. Although song is learned by copying
ity state, where their condensation (that is, Not anymore. j a tutor, each bird learns a slightly different
the joining through covalent bond forma- variation and the song is then used as a per-
REF ERENCES AND NOTES
tion with the release of water molecules) sonal calling card in courtship and other
1. G. F. Joyce, L. E. Orgel, in The RNA World, R. F. Gesteland, J.
becomes thermodynamically favorable. This F. Atkins, Eds. (Cold Spring Harbor Laboratory Press, Cold social interactions. For an accurate repro-
wet-dry method for driving model prebiotic Spring Harbor, NY, ed. 2, 1999), pp. 49–77. duction, a young zebra finch must learn
reactions spans the history of prebiotic chem- 2. L. E. Orgel, Crit. Rev. Biochem. Mol. Biol. 39, 99 (2004). how to shape the individual syllables, how
3. M. W. Powner, B. Gerland, J. D. Sutherland, Nature 459, 239
istry research and has proven its utility for (2009). to order them, and how they are spaced in
producing other chemical bonds necessary 4. S. Becker et al., Science 352, 833 (2016). time. The learning process begins with the
for formation of RNA (2, 8), lipid precur- 5. S. Becker et al., Science 366, 76 (2019). young bird listening to an adult tutor (often
6. M. C. Chen et al., J. Am. Chem. Soc. 136, 5640 (2014).
sors (9), and peptides (10, 11). Unlike volca- 7. R. Shapiro, Proc. Natl. Acad. Sci. U.S.A. 96, 4396 (1999). the father), laying down a sensory memory
nic eruptions and meteorite impacts—events 8. W. D. Fuller, R. A. Sanchez, L. E. Orgel, J. Mol. Evol. 1, 249 of the tutor’s performance. This normally
that often are proposed as drivers of prebi- (1972). occurs over several weeks as the young
9. C. L. Apel, D. W. Deamer, Orig. Life Evol. Biosph. 35, 323
otic reactions—wet-dry cycles would have (2005).
bird becomes independent from its parents,
10. J. G. Forsythe et al., Angew. Chem. Int. Ed. 54, 9871 (2015).
11. M. Rodriguez-Garcia et al., Nat. Commun. 6, 8385 (2015).
Department of Chemistry, Georgia Institute of Technology, Biological and Chemical Sciences, Queen Mary University of
Atlanta, GA 30332-0400, USA. Email: hud@chemistry.gatech.edu 10.1126/science.aaz1130 London, London E1 4NS, UK. Email: d.clayton@qmul.ac.uk

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 33

1004Perspectives.indd 33 9/30/19 12:56 PM

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

although under some experi- Sensorimotor transformation cess—at least with respect to
mental conditions a functional the duration of song elements.
sensory memory can be formed
in birdsong learning Zhao et al. observe that “the
A sensory memory of a tutor’s song is formed in a
with just a few minutes of expo- substrate for long-term storage
complex network of higher auditory brain circuits.
sure (3). This memory will then Later, the bird learns to reproduce this song of behavioral-goal memories is
serve as a reference “template” through a dedicated song motor control network. still unknown.” Their manipula-
(4) as the young tutee develops The nucleus interfacialis of the nidopallium (NIf) tions of the NIf-HVC pathway
its own song in the following projects onto HVC and is a key interface between had no measurable effect on
weeks (5), refining its vocal per- Zebra fnch these sensory and motor representations. other song features typically
formance through trial and er- imitated during song learning,
ror until it comes to match the such as pitch, mean frequency,
template closely. goodness of pitch, and entropy.
Sensory Motor
This two-phase process im- Where and how these features
plies the construction of at least Mesopallium HVC are represented in the network
two neural representations: remains unclear, but it seems
an initial sensory representa- entirely plausible that different

Neural connections
tion of the tutor’s performance, features of song are represented
and a motor representation Basal ganglia in different parts of the com-
Nidopallium NIf circuit for
for producing the young bird’s plex song control network (see
rehearsal
imitation (see the figure). The the figure). Indeed, their lesion
neural basis of the initial sen- results suggest that there may
sory template is still a subject not be a single lasting “engram”
of fascination and some mys- Robust nucleus of of the tutor song. Rather, tutor
Primary auditory the arcopallium
tery. Selective neurophysiologi- memories may be processed
cal responses to the sound of and transformed dynamically
particular songs emerge in across a shifting network to

Behavioral responses
Auditory input

forebrain sensory and associa- shape what ultimately matters

tive regions that receive input for the bird: to produce a stable,
Respiratory Syrinx
from primary auditory centers. Tutor song heard well-formed adult song.
By contrast, the neural basis of From an evolutionary perspec-
song-producing motor activities tive, birdsong learning ability
Song later produced
has been characterized with is a robust specialization shared
increasing precision, building in some form by ~3000 ex-
from the seminal finding that causing a le- and somehow transform them into motor tant songbird species (passerines). Human
sion to a forebrain nucleus now known for- patterns? If so, then selective alteration of speech development provides another ex-
mally as “HVC” disrupts song production the pattern of NIf-HVC firing only during ample of learning by imitation; might simi-
but not other behaviors (6). HVC sits high tutoring should show up later as predict- lar behavioral-goal memory circuits have
in the caudal telencephalon and projects able alteration of adult song. evolved in humans for speech learning? Re-
downward to the primary output nucleus of Zhao et al. tested this hypothesis using search has begun to reveal anatomical and
the song motor control system, the robust optogenetics (optical activation of specific molecular parallels between neural systems
nucleus of the arcopallium (RA), which in neurons). In juveniles prior to adult song controlling birdsong and human speech (11–
turn drives the muscles in the syrinx and exposure, they injected a virus expressing 13). Whether there is also a direct equivalent
respiratory systems that are responsible channelrhodopsin-2 into NIf (to confer in humans of the NIf-HVC pathway is an in-
for vocal sounds. Careful and sophisticated an electrophysiological response to light), teresting question for the future. j
analyses of electrophysiological firing in then implanted optical fibers to deliver
REF ERENCES AND NOTES
HVC suggest that it provides information light pulses onto NIf-HVC synapses for a
1. W. Zhao, F. Garcia-Oscos, D. Dinh, T. F. Roberts, Science
about the timing and sequencing of ele- few days, of durations somewhat shorter 366, 83 (2019).
ments in the learned song (7–9). or longer than typical song elements. As 2. R. A. Zann, The Zebra Finch: A Synthesis of Field and
How does the sensory template learned adults, the birds exposed to the short-du- Laboratory Studies (Oxford Univ. Press, 1996).
3. M. Deshpande, F. Pirlepesov, T. Lints, Proc. R. Soc. B 281,
from the tutor’s song get transformed into a ration pulses produced songs with signifi- 20132630 (2014).
set of muscular actions that reproduce that cantly shorter song elements than normal, 4. P. Marler, J. Comp. Physiol. Psychol. 71, 1 (1970).
sound? Zhao et al. focused on the nucleus whereas birds trained with long pulses had 5. K. Immelmann, in Bird Vocalizations, R. A. Hinde, Ed.
(Cambridge Univ. Press, 1969), pp. 61–74.
interfacialis of the nidopallium (NIf ), which songs with longer elements. These effects 6. F. Nottebohm, T. M. Stokes, C. M. Leonard, J. Comp.
appears to be, quite literally, an interface took time to emerge after the end of optoge- Neurol. 165, 457 (1976).
between sensory and motor processing sta- netic tutoring, observing the typical multi- 7. R. H. Hahnloser, A. A. Kozhevnikov, M. S. Fee, Nature 419,
65 (2002).
tions in the auditory-vocal control network. week time course of juvenile vocal practice 8. M. A. Long, M. S. Fee, Nature 456, 189 (2008).
Neurons in NIf show electrophysiological and refinement. Zhao et al. also found that 9. T. A. Nick, M. Konishi, J. Neurobiol. 62, 469 (2005).
responses to auditory signals (suggesting a lesions to the NIf-HVC pathway disrupted 10. A. L. Vyssotski, A. E. Stepien, G. B. Keller, R. H. R. GRAPHIC: C. BICKEL/SCIENCE
Hahnloser, PLOS Biol. 14, e2000317 (2016).
role in sensory coding), but also project to song development if induced before, but
11. M. S. Brainard, A. J. Doupe, Annu. Rev. Neurosci. 36, 489
HVC in the motor control pathway, where not after, tutor exposure. Hence, they con- (2013).
their synapses fire in temporal patterns that cluded that they had effectively imposed a 12. A. R. Pfenning et al., Science 346, 1256846 (2014).
immediately precede the production of spe- transient sensory memory (via the NIf-HVC 13. J. F. Prather, K. Okanoya, J. J. Bolhuis, Neurosci.
Biobehav. Rev. 81B, 225 (2017).
cific song syllables (10). Might NIf-HVC ac- synapse) that served as a behavioral goal
tivity initially encode aspects of tutor song, memory for the later motor rehearsal pro- 10.1126/science.aaz1552

34 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004Perspectives.indd 34 9/30/19 12:56 PM

 CHROMATIN BIOLOGY

Remodeling the genome with DNA twists

Intricate protein machines repackage DNA to turn genes on and off

By Gregory D. Bowman1 and Sebastian Deindl2 along DNA. Acting on either face of the nu- the binding site. When poised for hydrolysis,
cleosome disk, remodelers can move the his- the ATPase in its closed state eliminates this

I
n complex organisms such as humans, a tone core back and forth on DNA, changing bulge, corkscrewing DNA toward the nucleo-
single genetic blueprint can give rise to which parts of DNA are exposed and which some midpoint (dyad) on the other side. This
a multitude of different cell types, from are wrapped up in the nucleosome. Increased creation and elimination of a DNA bulge is
nerve to liver to muscle. Such cellular exposure of DNA occurs when remodelers equivalent to altering DNA twist.
diversity relies on restricting which por- shift adjacent nucleosomes into each other, Changes in DNA twist, known as twist
tions of genomic DNA are accessible and resulting in histone ejection (2). defects, have been proposed as a low-energy
therefore can be read by cellular machinery. The ability of remodelers to manipu- means of ratcheting DNA past the histone
Ultimately, access to DNA depends on place- late nucleosome structure stems from a core (6). Twist defects were observed in the
ment of a repetitive, spool-like structure highly conserved adenosine triphosphatase first nucleosome crystal structure, demon-
called the nucleosome, the basic packaging (ATPase) motor that belongs to a larger su- strating the inherent capability of nucleoso-
unit of chromosomes. The nucleosome oc- perfamily of helicase-like ATPases called mal DNA to sample different geometries (1).
cludes two tight loops of DNA and thus rep- superfamily 2 (SF2). On the basis of intense When DNA moves, a twist defect arises when
resents a fundamentally repressive element. biophysical and biochemical research on SF2 two adjacent segments of the same DNA du-
When and where nucleosomes are positioned and the related SF1 ATPases, understanding plex do not undergo a corkscrew shift in uni-
can affect complex transcriptional programs, how these enzymes move and interact with son. The twist defect can be thought of as the
and therefore disruptions in the factors re- DNA and RNA has revealed how remodeler junction between a stationary and a mobile
sponsible for nucleosome positioning often ATPases move and engage DNA (3). Both stretch of DNA: one side of the junction un-
result in cancers and multisystem develop- SF1 and SF2 ATPases consist of two distinct dergoes a corkscrew shift (mobile), while the
mental diseases. Although the mechanism domains that together form an ATP-binding other side remains stationary with respect
of shifting nucleosomes along DNA has pocket and a nucleic acid binding surface. to the histone core. Twist defects therefore
long proved elusive, a recent flurry of struc- Dictated by the occupancy of the nucleotide- reflect a gain or loss of a base pair resulting
tural, biophysical, and biochemical work binding pocket, the two domains open and from the corkscrew shift of DNA. Twist de-
has revealed a core mechanistic framework close like a clamshell, which in turn alters fects dissipate once one side of the junction
explaining how nucleosomes are actively re- interactions with the bound nucleic acid. For undergoes a compensatory corkscrew shift,
positioned throughout the genome. many SF1 and SF2 enzymes, alternating be- restoring DNA to its canonical conformation
Nucleosomes are the most ubiquitous pro- tween open and closed states ratchets the nu- on the nucleosome. If the corkscrew shift oc-
tein-DNA complexes in all eukaryotic cells. cleic acid past the ATPase in what is known curs on the previously stationary DNA seg-
The core of each nucleosome is a symmetric, as an inchworm-type mechanism (3). ment, another twist defect can be created
disk-like structure made of histone proteins Consistent with an inchworm-type trans- farther downstream, where the newly mobile
that provides a scaffold around which two location mechanism, nucleosomal DNA is DNA segment again runs up against a station-
loops of the DNA helix are snugly wrapped shifted by remodelers with an elementary ary segment. Through such discontinuous
(1). Histones are often modified through, for step of a single base pair (4). Given the spi- motions, propagation of twist defects all the
example, acetylation, methylation, and phos- ral structure of the duplex, nucleosomal way around the nucleosome repositions the
phorylation, which add an additional layer of DNA must shift around the histone core in entire length of nucleosomal DNA relative to
information on top of the genetic code. This a corkscrew fashion. Remodeler ATPases re- the histone core. Spontaneous nucleosome
epigenetic information demarcates function- main in a fixed location on the histone core sliding via twist defects has been visualized
ally distinct regions of the genome—for in- during DNA translocation, which means that through molecular simulations, suggesting
stance, whether a gene is active or designated DNA all around the nucleosome must shift that twist can be absorbed or buffered at mul-
to remain silent—for each cell type. in response to localized action at the ATPase tiple locations on the nucleosome (7).
Owing to their extensive protein-DNA binding site. According to the classic inter- Key stages in the nucleosome sliding cycle
interface, nucleosomes are relatively stable pretation of the inchworm model, DNA on have been captured by cryo–electron micros-
structures. Active placement and reorganiza- both sides of the ATPase should be shifted si- copy, adding essential mechanistic insight
tion of nucleosomes depend on chromatin re- multaneously with each ATP hydrolysis cycle, into how repositioning is achieved. In the
modelers. As the gatekeepers of nucleosome requiring that the motor be physically cou- initial stage, the open state of the ATPase
packaging, these enzymes participate both in pled to the histone core to generate force. In- bound to nucleosomal DNA creates a bulge
activating and repressing gene expression. stead, remodeler ATPases shift nucleosomal in only the tracking strand of the DNA duplex
Remodelers can assemble, disassemble, and DNA discontinuously, with DNA movement but not the complementary guide strand (8)
exchange histones within the nucleosome, as resulting from both open and closed states (half twist defect; step 1 in the figure). A con-
well as shift the position of the histone core of the ATPase that bypass the strict require- sequence of pulling only one of the two DNA
ment for a separate histone foothold to push strands is base tilting, which is necessary to
1
Thomas C. Jenkins Department of Biophysics, against (5). In its open (nucleotide-free and maintain base pairing between strands. Al-
Johns Hopkins University, Baltimore, MD, USA. adenosine diphosphate–bound) state, the though a shift of only one DNA strand was
2
Department of Cell and Molecular Biology, Science for Life
Laboratory, Uppsala University, Uppsala, Sweden. ATPase pulls entry-side DNA toward itself in unexpected, a similar tilting of base pairs was
Email: gdbowman@jhu.edu; sebastian.deindl@icm.uu.se a corkscrew fashion, creating a DNA bulge at observed in the DNA-RNA hybrid cradled

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 35

1004Perspectives.indd 35 9/30/19 12:56 PM

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

Entry Exit Shifting DNA around the nucleosome twist diffusion dependent on other charac-
A model of how twist defects, stimulated by the remodeler adenosine triphosphatase teristics of the nucleosome? The histone core
Dyad (ATPase), shift DNA around the nucleosome (shifted DNA is depicted in red). has dynamic properties (14, 15), and it will be
After a full cycle (steps 1 to 3), creation and passage of additional twist defects from interesting to see the degree to which plastic-
subsequent cycles push DNA out of the nucleosome exit side. ity of histone structure may affect formation
and propagation of twist defects within the
Remodeler nucleosome. Histone proteins come in a vari-
ATPase ety of flavors, and an important goal is identi-
ATP-bound fying whether distinct biophysical properties
1 A shift of only the tracking of histone variants and other epigenetic sig-
Nucleosome 0 strand of nucleosomal DNA
Half tilts base pairs, potentially natures can determine or bias the outcomes
twist creating strain in the DNA of remodeling reactions.
defect duplex on the entry side. The ability of remodeler-type ATPases to
create twist defects is used for more than
ATP just sliding nucleosomes. The SWR1 subclass
hydrolysis specializes in histone exchange, swapping
out canonical histone H2A-H2B dimers for
Closed state variant H2A.Z-H2B dimers, a universal epi-
ATP-bound genetic mark in gene promoters. A key issue
for this system will be uncovering the con-
1 2 nection between twist defects and histone
Open state ATP binding
dimer exchange.
Apo or adenosine and hydrolysis
diphosphate-bound cycle Full
Remodeler-type ATPases encompass en-
Tracking
strand twist zymes that specialize in non-nucleosomal
3 defect substrates. This class of enzymes includes
factors that are essential for DNA recombi-
nation and repair, as well as the recycling of
3 A corkscrew shift transcriptional machinery in both eukaryotes
of DNA over the dyad
and bacteria. Despite having distinct targets,
reverses the twist Full 2 A shift in the other strand
defect at the ATPase twist restores the canonical tilt of these remodeler-type ATPases likely catalyze
binding site while defect nucleosomal DNA, absorbing physical changes in their substrates by dis-
creating a new twist the additional base pair torting the DNA duplex, creating high-energy
at another site on the Closed state pulled onto the nucleosome intermediates analogous to nucleosomal
nucleosome. Hydrolysis-competent as a full twist defect.
twist defects. Just as insights for SF1 and SF2
ATPases have facilitated thinking about nu-
by RNA polymerase during transcription (9) DNA twist are complemented by the observa- cleosome sliding, the discovery that remodel-
and may create strain that allows the duplex tion of active movement of nucleosomal DNA ers create and eliminate twist defects on the
to translocate more easily. For remodelers, by remodelers during three-color single-mol- nucleosome will likely advance our under-
strain from base pair tilting would prime ecule FRET (Förster resonance energy trans- standing of these related yet functionally dis-
the duplex for the next stage, when a com- fer) experiments, where DNA movements tinct enzymes, which all wrestle the double
pensatory shift of the other strand creates were coupled to ATP hydrolysis (13). As dem- helix to transiently blaze a trail through the
a full twist defect. Such a remodeler-bound onstrated using FRET reporters on both sides energetic landscape of duplex DNA. j
state has recently been captured on a nu- of the nucleosome, remodelers pull DNA
REF ERENCES AND NOTES
cleosome bound to SWR1 (SWI/SNF-related) onto the entry side of the nucleosome before
1. K. Luger et al., Nature 389, 251 (1997).
(10), where both DNA strands bulge, accom- shifting DNA off the exit side. Observed as 2. M. L. Dechassa et al., Mol. Cell 38, P590 (2010).
modating an additional base pair and restor- a time delay, nucleosomes absorbed one or 3. M. R. Singleton et al., Annu. Rev. Biochem. 76, 23 (2007).
4. S. Deindl et al., Cell 152, 442 (2013).
ing the canonical base stacking outside the more base pairs of DNA pulled on by the re- 5. J. Winger et al., eLife 7, e34100 (2018).
remodeler binding site (step 2 in the figure). modeler, which would take the form of twist 6. K. E. van Holde, T. D. Yager, in Structure and Function
After remodelers have successfully shifted defects. Future studies should determine the of the Genetic Apparatus, C. Nicolini, P. O. P. Ts’o, Eds.
(Springer, 1985), pp. 35–53.
the twist defect toward the dyad, the DNA capacity and energetic cost of storing twist 7. G. B. Brandani et al., Nucleic Acids Res. 46, 2788 (2018).
at the remodeler binding site should return defects during remodeling and how buffer- 8. M. Li et al., Nature 567, 409 (2019).
to its canonical structure on the nucleosome ing twist defects may be influenced by DNA 9. A. L. Gnatt et al., Science 292, 1876 (2001).
10. O. Willhoft et al., Science 362, eaat7716 (2018).
(step 3 in the figure). Such a post-shifted state sequence. In addition, because a single nu- 11. L. Farnung et al., Nature 550, 539 (2017).
with the bound DNA in a canonical geometry cleosome can simultaneously bind a remod- 12. R. Sundaramoorthy et al., eLife 7, e35720 (2018).
has been visualized with remodeler ATPases eler on each side, it will also be of interest to 13. A. Sabantsev et al., Nat. Commun. 10, 1720 (2019).
14. K. K. Sinha et al., Science 355, eaaa3761 (2017).
trapped in hydrolysis-competent states (5, 8, determine whether such DNA buffering and 15. S. Bilokapic et al., Nat. Commun. 9, 1330 (2018).
11, 12). Because nucleosomal DNA with a ca- changes in twist enable communication be-
nonical twist aligns with the transition state tween pairs of remodelers vying to shift the ACKNOWL EDGMENTS GRAPHIC: V. ALTOUNIAN/SCIENCE

of the ATPase (5), elimination of the twist histone core in opposite directions (13). We thank I. Nodelman and A. Sabantsev for discussions
and critical comments. G.D.B. is supported by the National
defect may be sufficient to trigger ATP hy- With the core mechanism of nucleosome Institutes of Health (R01-GM084192). S.D. is supported by the
drolysis, which would initiate another round sliding finally coming into focus, this recently European Research Council, the Swedish Research Council,
of sliding and enforce directionality of twist developed framework provides a launching and the Knut and Alice Wallenberg Foundation. S.D. is a
European Molecular Biology Organization Young Investigator.
diffusion around the nucleosome. point for future research. In addition to DNA
These snapshots of remodelers altering geometry and energetics, to what extent is 10.1126/science.aay4317

36 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004Perspectives.indd 36 9/30/19 12:56 PM

 RETROSPECTIVE

Donald A. B. Lindberg (1933–2019)

Visionary medical informaticist who laid the groundwork for PubMed

By Isaac Kohane1 and Jeremy M. Berg2 was to be repeated throughout his career. At Richard Roberts and Joshua Lederberg, who
NLM, he used data to convince members of understood the importance of computing to

D
onald A. B. Lindberg, influential direc- the U.S. Congress and influential commercial biomedical investigation and helped him de-
tor of the United States National Li- stakeholders to better serve the public inter- velop long-term strategic plans for the NLM.
brary of Medicine (NLM) and unsung est by, for example, requiring prior registra- Don’s rare combination of strong scientific
hero of medical research and science, tion of expected outcomes for clinical trials. vision and leadership skills allowed him to
died on 17 August. He was 85. Thanks Don was a pioneer in medical computing. work well both on Capitol Hill and across U.S.
to his vision and leadership, more In the 1960s, he developed some of the earli- universities and academic health centers. Be-
than a million people a day can use PubMed est laboratory information systems, consist- cause he could bring parties with opposing
to look up biomedical information and nearly ing of a punch card–fed clinical reporting perspectives together to develop a shared
150,000 people a day can look up trials on program running on a Datatron-205 com- view of the public good in the applications
ClinicalTrials.gov. Researchers in artificial in- puter with 20 KB of memory. Twenty years of information technology, he was frequently
telligence, computational biology, and health later, he led the development of Artificial In- in demand for leadership challenges that ex-
care policy routinely use the mapping between telligence Rheumatology Consultant System tended beyond the remit of the NLM. From
hundreds of different biomedical vocabular- Ontology, an AI program that could diagnose 1992 to 1995, he served as founding director
ies and ontologies provided by the NLM’s complex cases of rheumatological disease. of the National Coordination Office for High
Unified Medical Language System. Many of Performance Computing and Communica-
the current leaders in biomedical computing tions in the Office of Science and Technology
have been trained by programs funded by the Policy, Executive Office of the President. In
NLM, which Don directed from 1984 to 2015, 1996, the Health and Human Services Secre-
starting well before big data, precision and tary appointed him to serve as the U.S. coor-
genomic medicine, or deep learning for medi- dinator for the G7 Global Health Applications
cine were even recognizable terms, let alone Project. He was also a founding member of
vibrant disciplines of their own. the American Medical Informatics Associa-
Don grew up in Brooklyn, New York. He tion and its first president.
became passionately interested in human Those who run libraries are frequently
biology at Amherst College in Massachu- underappreciated, as many take for granted
setts. After graduating in 1954 with a bach- the important and far-reaching services that
elor’s degree in biology, he considered going they provide to their communities. Even with
to graduate school. Persuaded that medical all of his accomplishments, this was true for
training would provide more opportunities, Don in many circles outside the informatics
he enrolled instead at Columbia University’s community. Many forget how different the
College of Physicians and Surgeons in New research world was before resources such as
York. It was a very good school, Don acknowl- PubMed were commonplace. These advances
edged, but “very far from the study of human One key to Don’s success was his ability have been driven by technological develop-
biology” and “a high-class trade school at to attract top talent to lead NLM initiatives ments, most notably the internet, but Don’s
best.” Although the medical school experience and implement his vision, including compu- vision predated them. Much of the difficult
was “thrilling,” the large amount of required tational biologist David Lipman, biomedical foundational work and a corps of well-trained
memorization was “almost antithetical to ex- informaticist Alexa McCray, medical librar- and engaged scientists and physicians were
perimental science.” After graduation in 1958 ian Betsy Humphreys, and health informati- already in place when the technological ca-
and a residency in pathology, Don served on cist Milton Corn. Don helped lead initiatives pabilities came to be. Don found the recent
the faculty of the University of Missouri for that spanned the mission of the National (re)discovery of the importance of data science
24 years. In 1984, he left to direct the NLM. Institutes of Health. He delegated many pro- in science and medicine somewhat amusing.
Early on, Don recognized that medicine gram responsibilities to his NLM colleagues Don’s wife of more than 60 years, Mary, be-
needed to be more data driven. During his and could seem detached. However, he was came well known to those of us who worked
residency training, his data-driven approach deeply knowledgeable about the programs with the NLM. Ever since their second date
to medicine landed him in hot water with the he oversaw and preferred to lead by helping in 1957, when they started planning their life
surgery department when he showed that to set the agenda and then empowering his together, Don and Mary embodied joint de-
silicon used for cardiac bypass was causing colleagues to work out the implementation cision-making. She survives him, along with
extensive embolization and death. Only upon details themselves, making himself avail- his brother, two sons, and two grandchildren.
PHOTO: JESSICA MARCOTTE

showing the surgeons the imaging data was able for guidance when needed. He had a dry Don’s work lives on with millions of patients,
he able to change their practice. This pattern sense of humor and used it to lighten what doctors, and scientists who depend on access
are often dry subjects, as when he named an to and investigations of the exponentially
1
Department of Biomedical Informatics, Harvard Medical early NLM resource Grateful Med. Don sur- growing corpus of biomedical knowledge. j
School, Boston, MA 02115, USA. 2Editor-in-Chief,
Science, Washington, DC 20005, USA. rounded himself with expert advisers, such
Email: isaac_kohane@harvard.edu; jberg@aaas.org as Nobel laureates in Physiology or Medicine 10.1126/science.aaz3644

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 37

1004Perspectives.indd 37 9/30/19 12:56 PM

 INSIGHTS

P OLICY FORUM
FOREST MANAGEMENT

Collaborations and capacities

to transform fire management
Progress requires attention to governance at multiple levels

By Courtney A. Schultz1 and forest restoration in fire-prone ecosystems

Cassandra Moseley2 typically involves mechanical fuel reduction
(thinning of trees and clearing of brush),

W
ildfires bring stark attention prescribed fire (see the photo), and/or the
to interactions among climate management of natural ignitions to reintro-
change, fire, forests, and liveli- duce fire. Although fires historically burned
hoods, prompting urgent calls for under a range of conditions, they have been
change from policy-makers and mostly suppressed when possible in the U.S.
the public. Management options West for the past century, excluding low- to
vary, but in many fire-adapted forests, the moderate-severity fire and perpetuating a
message from the scientific community is fire debt on the landscape. Understandably,
clear: Adapt to living with fire, reduce fu- there is social and political pressure to put
els and homes in the wildland-urban inter- fires out quickly to minimize risks to human
face (WUI), and strategically restore fire to health and economic well-being (such as for
ecosystems (1–4). Yet, changes to fire man- the tourism industry) and to protect homes ply having facilitative policies in place is
agement outcomes have been elusive. For and other infrastructure. Risks of allowing not enough. Progress requires coordination
example, across the primarily public forest- fire to burn are apparent in the short term, across sectors (such as air quality and land
lands of the U.S. West, prescribed fires (in- whereas benefits of supporting fire (and management), diverse actors, and multiple
tentionally lighted fires) constitute a small, downsides of exacerbating fire hazard by levels and jurisdictions.
inadequate fraction of forest treatments (5), suppressing fire or failing to conduct pre-
and fire managers rapidly contain over 95% scribed fire) often accrue after the tenure of PROMISING DEVELOPMENTS
of ignitions (2). Meanwhile, the WUI is the any given politician or land manager. Both the Collaborative Forest Landscape
fastest growing U.S. land-use type (6). Sub- This situation is exacerbated by con- Restoration Program (CFLRP; created by
stantial land-use changes that remove peo- flicting policy mandates within and across Congress in 2009 for work on National
ple and infrastructure from fire-prone areas governance levels and jurisdictions. For ex- Forest System lands) and the Joint Chiefs
are unlikely, making forest management a ample, the Oregon Department of Forestry Landscape Restoration Partnership (a simi-
critical piece of the puzzle. To inform the has a policy of putting out fires as quickly lar but smaller-scale, internal agency ini-
global challenge of living with fire, we dis- as possible to protect timber resources. tiative that began in 2014 to support work
cuss promising developments in U.S. federal This is at odds, partially, with federal pol- across both National Forests and private
fire management that rely on collaborative icy, which emphasizes both restoring fire forestlands) represent a new model of U.S.
governance, which is essential for grappling to ecosystems and suppressing fire when land management policy meant to facilitate
with complex environmental management necessary (7). The complexity of both our a more strategic approach to forest resto-
challenges to leverage diverse capacities, political system and the fire problem make ration. The programs are distinct in sup-
work across jurisdictions, and support col- it unlikely that policy conflicts will be re- porting a national process for prioritizing
lective action to plan for the long term in solved, although some policy reform has funding, which is important to focus invest-
the face of pressures to focus on short-term occurred. The National Cohesive Wildland ment of limited resources.
risks and objectives. Fire Management Strategy (mandated by Under these programs, the U.S. Forest
In the western United States and glob- Congress in 2009 and finalized in 2014) Service and Natural Resource Conversa-
ally, fire seasons have grown longer; fire and 2012 revisions of the National Forest tion Service allocate funding for mechanical
size, severity, and frequency have increased; Management Act regulations both empha- thinning, prescribed fire, and other restora-
and there has been an increase in costs and size the importance of restoring fire (7). In tion activities on the basis of science-based
losses of human life and infrastructure (1, 2016, Clean Air Act regulations were revised proposals that outline the ecological, social,
2). Scientists say that we must restore fire through collaboration among the U.S. En- and economic needs and opportunities in
to fire-adapted forest ecosystems in order vironmental Protection Agency and federal contiguous landscapes. Proposals are writ- PHOTO: LANCE CHEUNG/USDA

to reduce fire hazard, promote resilience, land management agencies to create more ten collaboratively by state and federal land
and support climate adaptation (2–4). U.S. opportunity for prescribed fire by increas- managers, scientists, community-based
ing flexibility around air-quality permitting, groups, and other nongovernmental orga-
1
Department of Forest and Rangeland Stewardship, Colorado something that scientists have suggested is nizations and are evaluated by a national
State University, Fort Collins, CO 80512, USA. 2Institute for a
Sustainable Environment, University of Oregon, Eugene, OR necessary to support more prescribed fire, committee. The programs offer multiyear
97403, USA. Email: courtney.schultz@colostate.edu although evidence is mixed (3, 8). But sim- funding commitments, which buffer against

38 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004PolicyForum.indd 38 9/27/19 5:54 PM

 The Geronimo Interagency Hotshot Crew from the
Bureau of Indian Affairs, San Carlos Agency, conducts a
prescribed burn near Galice, Oregon, in August 2013.

lutions (such as improved communication

and monitoring to find space to burn with-
out violating air-quality standards).
Similar venues exist or are emerging in
other states. Improved resource-sharing
tools and increased funding and human re-
source capacity, perhaps dedicated teams,
also are needed, along with consistent di-
rection, support, and incentives from Con-
gress and agency leadership to indicate that
prescribed fire is a priority, given that state
and federal policies focus on an array of
goals that may compete with increasing the
presence of fire on the landscape (8).
To build capacity and support collabo-
ration, states are increasing resources for
fuels and fire management. For example,
California is dedicating substantial fund-
ing, largely revenue from the state’s carbon
market, to fund teams to work across juris-
dictions to remove fuels mechanically and
with prescribed fire. New Mexico has appro-
priated funding to address fire hazard and
created a working group on prescribed fire.
fluctuations and uncertainties of annual ap- tem. Prescribed fire requires planning and Oregon has taken similar steps through its
propriations, support a coherent program of permitting, is logistically complicated to Federal Forest Restoration Program. Facing
work in a landscape, and draw in partners execute (requiring trained staff and equip- declining federal investments and the cross-
to leverage capacity. Both programs outline ment to be available during narrow burn jurisdictional nature of fire hazard, state
substantive goals (for example, reducing fire windows), and can be controversial. To and local governments need to act.
hazard and restoring fire-adapted ecosys- overcome these hurdles, federal and state The U.S. Forest Service is investing in col-
tems while engaging industry partners) but land managers and air-quality regulators laborative, preseason, and cross-boundary
allow flexibility to tailor projects to local ca- have emphasized that in addition to lever- planning for fire response. Using principles
pacities, socioeconomic conditions, and eco- aging local capacity, state-level interagency of risk management and new analytical
logical objectives (for example, focusing on a collaboration, because of the role of state tools that use machine learning to identify
municipal watershed versus restoring fire to regulatory and land management agen- potential fire control locations (10), manag-
an entire ecosystem). cies, is key for facilitating communication, ers of U.S. National Forests—engaging with
Although not every project has been suc- resource sharing, and problem solving (8). nongovernmental partners, tribes, state and
cessful, program-wide these approaches For example, the Montana-Idaho Airshed local agencies, and agency scientists—are
support larger-scale planning and imple- Group uses an online platform to track and determining fire management options in
mentation of mechanical thinning of prioritize burns, coordinates burners within advance of ignitions: priorities for suppres-
forests, innovations in monitoring and airsheds, identifies priorities for burning ac- sion or options to allow natural fires to burn
planning, leveraging of nonfederal capac- cording to the need and availability of burn when they may have benefits for valued re-
ity, and agreement-building in an arena windows, and uses a liaison who works on sources (11). These activities hold promise
that historically has been characterized by behalf of burners to communicate with state for getting more “good” fire on the ground
conflict over approaches to vegetation man- air-quality regulators. In California, the because changes in practice will be elusive
agement (9). At the same time, major chal- Fire MOU (Memorandum of Understand- until multiple actors build agreement about
lenges persist, including inadequate agency ing) Partnership brings together scientists; fire management approaches outside of the
capacity for planning and implementation, nonprofit, community-based, and tribal or- emergency management context and can
insufficient capacity in forestry-products in- ganizations; federal and state fire and land speak to collective goals during fire events.
dustry, and limited markets for wood prod- managers; and state air-quality regulators.
ucts (such as chips, pellets, firewood, and The Partnership and other statewide insti- OPPORTUNITIES FOR POLICY-MAKERS
lumber) that could offset high treatment tutions have helped land managers work To support collaboration and the use of
costs. Approaches that invest in places to address air regulators’ concerns with partner capacity, Congress should main-
where collaboration exists also may leave regard to managing pollutants to protect tain and fully fund programs such as the
behind communities without capacity, mak- human health, allowed regulators to bet- CFLRP, using it as a model for future pol-
ing it critical to address how to build capac- ter understand benefits of prescribed fire icy development (for example, a program
ity where it does not already exist. and commit to increasing permitted acres, geared toward cross-boundary prescribed
Despite these gains, it has been challeng- and identified barriers to burning (such as fire implementation). Congress and fed-
ing to implement prescribed fire without failure to use available burn days, often be- eral agencies should consider whether the
addressing barriers elsewhere in the sys- cause of lack of capacity) and possible so- Joint Chiefs Partnership should be estab-

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 39

1004PolicyForum.indd 39 9/27/19 5:54 PM

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

lished as a permanent program. Congress to increase the U.S. Forest Service’s appro- source sharing (8); and improve leadership
could improve and create grant opportuni- priations to fund restoration work. direction and performance measurement to
ties for state agencies, tribes, and commu- Congress should continue using its over- incentivize the application of fire, improve
nity-based groups, which would enhance sight role to understand barriers and op- accountability during fire response, focus
their ability to add capacity and advance portunities for change. Congress is now efforts on high-priority acres for restoration
solutions tailored to local conditions. For requiring greater information around cost treatments, and ensure that multiple-entry
example, organizations such as The Na- drivers and decision-making in wildland treatments (such as following mechanical
ture Conservancy and the Forest Stewards fire management and the effectiveness of thinning with prescribed fire) take place to
Guild have been supporting collaborations fuel treatments. Other challenges also need capitalize on prior investments (3, 7, 11).
and building capacity for forest restora- investigation; for example, fuel treatments Lessons from the U.S. West can be ex-
tion (for example, as part of efforts such as are not always placed strategically or main- tended to other contexts and inform the
the Western Klamath Restoration Partner- tained in a way that affects fire behavior and global challenge of living with fire. Multi-
ship, the Rio Grande Water Fund, and the management (13). Additional oversight ques- year funding commitments and laws and
Prescribed Fire Training Exchange, a na- tions could include, is the U.S. Forest Service policies that support collaboration within
tionwide program that builds capacity to using all scientific tools and policy options and across governance levels, facilitate ca-
conduct prescribed fire). Engagement with to improve strategic planning, access exist- pacity building and resource sharing, and
the forest products industry is important ing capacity, and track resource use? How is include objectives that can be adapted to
because difficult economics have slowed the agency capitalizing on opportunities to local conditions through participatory pro-
progress for restoration and kept treat- create and maintain desired conditions in cesses are all policy approaches that can
ment costs high (9). Policy-makers might locations that have burned or that have been promote collective action in a multilevel
request the Government Accountability Of- treated or where natural ignitions could be system. Collaborative governance is impor-
fice to investigate challenges faced by ex- managed to reintroduce fire? What more can tant at all system levels and for all aspects
isting and potential industry partners and be clarified about challenges and opportuni- of fire management, including building
to identify possible solutions. ties related to the pace of work? The Forest fire-adapted communities, given the impli-
States can build capacity to manage fire Service and Trump administration argue for cations of fire for health, safety, housing,
in partnership with federal agencies by in- reducing requirements around environmen- and growth and collaboration’s central role
creasing state funding for restoration work tal impact assessment to accelerate plan- in promoting effective community response
across jurisdictions, investing in air-quality ning, yet there is a substantial backlog of to disturbances and disasters (14, 15). Solu-
agency monitoring and permitting for fire planned-but-untreated acres. And, how does tions that embrace and navigate this com-
management, dedicating staff to participate the executive branch’s current direction to plexity have the potential to improve fire
in collaborations, and increasing funding focus on timber output square with the im- management by building the governance
and direction to state forestry agencies to portance of restoring forest resilience and processes and capacities necessary to trans-
create a larger prescribed fire and forest addressing fire hazard on federal lands? late policy goals into action. j
restoration workforce. State policy leaders Ongoing oversight and problem solving
REF ERENCES AND NOTES
can support creation of collaborative ven- relies on partner engagement and scientific
1. C. I. Millar, N. L. Stephenson, Science 349, 823 (2015).
ues such as the California Fire MOU Part- research. Problematically, the largest fed- 2. T. Schoennagel et al., Proc. Natl. Acad. Sci. U.S.A. 114,
nership or governors’ task forces to address eral funding source for applied fire research 4582 (2017).
place-specific challenges and identify solu- (including our work) has been cut. Fund- 3. M. P. North et al., Science 349, 1280 (2015).
4. M. D. Hurteau, M. P. North, G. W. Koch, B. A. Hungate,
tions. States may need to address whether ing for the Joint Fire Science Program, by Proc. Natl. Acad. Sci. U.S.A. 116, 10193 (2019).
the mandates of their forestry agencies or which interagency leadership sets priorities 5. C. A. Kolden, Fire 2, 30 (2019).
smoke management plans need updating in for much-needed ecological and social sci- 6. V. C. Radeloff et al., Proc. Natl. Acad. Sci. U.S.A. 115, 3314
(2018).
the face of a growing presence of fire. With ence research on fire management, should 7. C. A. Schultz, M. P. Thompson, S. M. McCaffrey, Fire Ecol.
the Forest Service’s current emphasis on be restored at fully authorized levels. 15, 13 (2019).
shared stewardship, there could be new op- Agencies have opportunities for internal 8. C. A. Schultz et al., Int. J. Wildland Fire (2019). 10.1071/
WF19040
portunities for state-level leadership. adjustments. The U.S. Forest Service is the 9. C. A. Schultz et al., Forests 9, 512 (2018).
To increase capacity, Congress and land- largest forest and fire management organi- 10. C. D. O’Connor et al., Int. J. Wildland Fire 26, 587 (2017).
management agencies must dedicate more zation in the country and faces steadily de- 11. M. P. Thompson, D. G. MacGregor, C. J. Dunn, D. E. Calkin,
J. Phipps, J. For. 116, 382 (2018).
funding to forest restoration and prescribed clining resources, increased fire costs, and
12. USDA Forest Service, “The rising cost of wildfire opera-
fire implementation. As a result of the ris- greater expectations for land management. tions: effects of the Forest Service’s non-fire work”
ing cost of fire, the majority of the U.S. For- These expectations are not commensurate (USDA Forest Service, 2015); www.fs.fed.us/sites/
est Service’s budget goes to fire response and with funding or staffing structures, necessi- default/files/2015-Rising-Cost-Wildfire-Operations.pdf
13. N. M. Vaillant, E. D. Reinhardt, J. For. 115, 300 (2017).
suppression. The consequence is that over tating a new model. The agency could con- 14. T. Steelman, B. Nowell, Int. J. Wildland Fire 28, 267
the past 20 years, funding for fuels reduction sider, for example, whether seasonal hiring (2019).
has not scaled with the scope of the prob- practices need to be adjusted to capitalize on 15. R. Djalante, C. Holley, F. Thomalla, Int. J. Disaster Risk Sc.
2, 1 (2011).
lem, and funding and personnel have de- burn windows throughout the year; address
clined substantially for everything else that whether there is need for a dedicated pre- ACKNOWL EDGMENTS
the agency does (12). Although private-sector scribed fire workforce; and limit leadership We thank our collaborators H. Huber-Stearns, S. McCaffrey,
contracting and partnerships with nongov- turnover and vacancies, which are problem- M. Thompson, and Z. Wurtzebach for their contributions to
our research and thinking. We also thank anonymous review-
ernmental organizations have increased, it atic for long-term collaboration. Scientists ers and A. Agrawal, P. Duffy, R. Gruby, L.Silva, and T. Steelman
has not made up for the loss of agency capac- have made other suggestions: Integrate stra- for helpful reviews of earlier drafts of this article. Funding:
ity. Addressing the agency’s ballooning fire tegic fire management planning more thor- This work was supported by Joint Fire Science Program
10 (16-1-02-8); C.A.S. also was supported by the National
suppression costs through the Consolidated oughly into land-management planning (3, Science Foundation under grant 1702676.
Appropriations Act of 2018 was part of the 7, 11); create national agreements between
solution. Now, Congress has the opportunity land-management agencies to streamline re- 10.1126/science.aay3727

40 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004PolicyForum.indd 40 9/27/19 5:54 PM

 A N N UA L M E E T I N G
Seattle, WA February 13–16, 2020

aaas.org/meetings

BE A PART OF THE CONVERSATION!

Climate Change Innovation Sustainability ArtiFcial Intelligence

Public Engagement Technology Diversity Psychology Global and Public Health

Chemistry Natural Hazards and Disasters Big Data

and more...

Advance registration now available online

Scientifc program released in October

1004Product.indd 41 9/25/19 1:11 PM

 Some skeptics of genetic cognitive enhancement
likely view conventional learning as more valuable.

demarcation between health-related edit-

ing and non–health-related editing is not as
clear as our intuitions may lead us to believe.
Cognitive enhancement through traditional
means, including reading and education,
may only be a difference of degree, not of
kind, from gene editing that selects for par-
ticular cognitive traits.
Baylis continues this important discussion
into chapter four, in which she argues that
all medical treatments are a form of enhance-
ment—improving or returning an aberrant
gene to its “normal” status. However, not all
enhancements, she argues, are treatments.
If one accepts this premise, the pertinent
ethical question is not whether treatments
should be allowed and nonmedical enhance-
ments prohibited but whether heritable gene
B O OKS et al . editing of any stripe may be structured to
promote equality, access, and fairness.
In chapter six, Baylis continues her dis-
GENE EDITING cussion of the possible harms and benefits
of germline editing through the lens of

Justice and genes potential moral wrongs. She worries, for

example, about the opportunity costs that
could occur when science and technology
A new book offers an introduction to the ethical funding is diverted toward gene editing re-
search, the benefits of which may then only
dimensions of germline gene editing be accessible to those with economic means.
Baylis encourages the public and scientists
By Adam Hayden the material accessible for scientifically cu- alike to ask first, how gene editing science
rious general readers, scientists interested will improve the human condition, and sec-

W
ith Altered Inheritance, bioethicist in the ethical dimensions of gene editing, ond, what kind of world they want to live in.
Françoise Baylis has authored a and bioethicists seeking an innovative Baylis introduces the idea of “impact eth-
vivid call to action that “aims to framework to direct theorizing and practice ics” in chapter nine. Following a heuristic
bridge the divides between theory, in the era of CRISPR. articulated in feminist ethics, in which one
science, politics, and practice” Baylis seamlessly threads definitional sets aside a traditional notion of individual
in response to increased public and conceptual content into the first three autonomy, impact ethics advances a rela-
awareness and scientific applica- chapters, which include critical tional understanding of autonomy, viewing
tions of CRISPR/Cas9 technol- information for readers who do people as interdependent. Here, Baylis also
ogy. She achieves her aim in this not have a strong background in resists the notion that experts have privi-
timely and important book. the terms and technology of gene leged access to the truth.
Baylis calls for broad societal editing. The first chapter analyzes “Decisions about the use of genetic tech-
consensus and shared responsibil- Huntington’s disease as a case nology are too important to be left to sci-
ity to guide heritable gene editing study for targeting a single gene. entists,” writes Baylis in chapter 10. This
toward the common good, which Chapter two distinguishes so- provocative claim is not intended as a nega-
she defines as that which is essen- matic from germline editing—the tive account of scientists’ involvement in
tial for survival and well-being, to Altered Inheritance former being an edit that affects the gene editing debate nor as a prohibition
Françoise Baylis
which the market, property, and Harvard University Press, an individual, the latter being one on their ongoing work. Rather, it is meant
liberty are subordinate. Her call 2019. 297 pp. that will be passed on to the in- as a call for bidirectional engagement be-
demands an urgent response, as dividual’s descendants. Heritable tween scientists and the public. Public
she notes that humanity sits precipitously on gene editing is the main focus of the book. empowerment in these integral policy dis-

PHOTO: KATE_SEPT2004/GETTY IMAGES

the verge of either a new beginning or the In chapter three, Baylis constructs a brief cussions is something “all of us” may and
beginning of the end. history of gene editing, using the case of should participate in.
A prologue and epilogue bookend the 10 “designer babies” as a central example. (In- Commitments to justice, responsibility,
chapters that comprise Altered Inheritance. cidentally, the term “designer baby” was first accountability, and consensus building are
The back matter includes an index to make coined to capture marketing-influenced, features of a socially just science and bio-
brand-heavy consumer behavior and only ethics. Toward this end, Altered Inheritance
later came to be applied to gene editing.) is a foundational tool in the path ahead. j
The reviewer is a philosopher of science and
freelance writer based in Indianapolis, IN, USA. She concludes this chapter with an expert
Email: adammarchayden@gmail.com philosophical discussion that shows that the 10.1126/science.aaz1167

42 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004Books.indd 42 9/30/19 12:46 PM

 INSI GHTS

CLASSICS REVISITED

The physicist and the dawn of the double helix

Erwin Schrödinger’s prescient musings on molecular biology turn 75

By Karl Sigmund quantum mechanics that Bohr,

Schrödinger, Jordan, and oth-

T
hree quarters of a century ers had developed in the 1920s.
ago, Nobel laureate Erwin Indeed, the copying mechanism
Schrödinger published What of genes is based on hydrogen
Is Life?, which described the bonds between complementary
forays of a “naïve physicist” nucleotides—territory ruled by
into biology and suggested the Schrödinger equation.
that hereditary properties are en- Six months after the discovery
coded in an “aperiodic crystal.” A of the double helix, and a mere
meme was born that changed the 10 years after the Dublin lec-
life sciences forever. tures, Francis Crick sent a letter to
A refugee from the Third Reich, Schrödinger, acknowledging the
Viennese-born Schrödinger had impact of What Is Life? on both
found shelter at Dublin’s Institute himself and Watson. Schrödinger
for Advanced Studies (1). Required did not reply. Even stranger, he
by statute to deliver public lectures never returned to biology.
every year, he chose in 1943 for Schrödinger lived until 1961,
his topic the physical aspects of when the search for the genetic
the living cell. He lectured on code was at its height, but seems
three consecutive Fridays, to a to have lost interest in “the most
packed house. The enthusiastic interesting science of our day.”
response spurred the publication Schrödinger’s book inspired early DNA research, but the physicist never He had assiduously studied ge-
of his notes in November 1944. returned to biology. netics for 20 years, but after the
The book did well. In the first publication of What Is Life? he
year, it sold 5000 copies and attracted 65 Gunter Stent) or physicians (such as Fran- turned to other things. In new editions of
reviews. The celebrated J. B. S. Haldane çois Jacob). Even a teenage bird watcher his seminal book—which never went out
mused that at a time when most physi- named James Watson became “polarized of print—Schrödinger did not address the
cists were engaged in war work and had a towards finding out the secret of the gene” breathtaking progress that it had caused,
hard time keeping up with their own sub- after reading Schrödinger’s booklet (4). nor Oswald Avery’s discovery that genes
ject, Schrödinger—in neutral Ireland—had Schrödinger’s prescient vision of genetic are made of DNA, nor John von Neumann’s
found leisure enough to turn to another coding is one message in What Is Life?. bold ideas on self-replicating automata.
branch of science: genetics (2). Schrödinger Another is that cells create order from With the advent of molecular biology,
had hailed the discipline as “easily disorder and transmit order from the centuries-old debate between “vital-
the most interesting of our days.” order. From this, Schrödinger con- ists” and “mechanists” turned decisively
Geneticist Haldane politely won- What Is Life? cluded that new physical laws were against the former. Schrödinger, however,
dered whether posterity would con- Erwin Schrödinger needed to explain life: “,,,present- punctured the latter, in a jaunty afterword
Cambridge
firm that judgment. It did, not least day physics and chemistry could tackling consciousness. Haldane quipped in
University Press,
thanks to Schrödinger’s book. 1944. 91 pp. not possibly account for what hap- his book review, “A mechanist must either
What Is Life? became one of the pens in space and time within a liv- give a mechanist account of life, or turn a
most important science books ever, not for ing organism.” He purported to prove this somersault. In his epilogue, Schrödinger
its content (whose limitations were soon by drawing on the results of mutagenesis does the latter with very great elegance.”
evident) but for its influence on a cohort experiments conducted by his erstwhile Schrödinger had preempted that objection
of brilliant scientists. Almost all of the pro- postdoc Max Delbrück, which offered by quoting the philosopher Unamuno: “If a
tagonists of the double helix saga acknowl- an estimate for gene size and implied, so man never contradicts himself, the reason
PHOTO: BETTMANN/CONTRIBUTOR/GETTY IMAGES

edged, at one time or another, the impact Schrödinger thought, that heredity relies must be that he virtually never says any-
of Schrödinger’s book (the one possible on a mechanism “that cannot be reduced thing at all.” j
exception being crystallographer Rosa- to the ordinary laws of physics.”
REF ERENCES AND NOTES
lind Franklin, who died before writing her Schrödinger was in good company: His
1. W. J. Moore, Schrödinger: Life and Thought (Cambridge
memoirs) (3). Many of them—including colleagues Niels Bohr and Pascual Jordan Univ. Press, 1989)
Francis Crick, Maurice Wilkins, and Sey- were also of the opinion that new laws of 2. J. B. S. Haldane, Nature 155, 375 (1945).
mour Benzer—were physicists, but some physics were needed to explain life. With 3. H. F. Judson, The Eighth Day of Creation (Cold Spring
Harbor, 1996).
were chemists (such as Erwin Chargaff and the discovery of the double helix, how- 4. J. Watson, The Double Helix: A Personal Account of the
ever, it turned out that there was no need Discovery of the Structure of DNA (Simon & Schuster,
The reviewer is at the Faculty for Mathematics,
for new laws. The chemical gadget could 1968).
University of Vienna, A-1090 Vienna, Austria.
Email: karl.sigmund@univie.ac.at be entirely explained, ironically, by the 10.1126/science.aaz4846

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 43

1004Books.indd 43 9/30/19 12:47 PM

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

Sign up at: webinar.sciencemag.org/stayinformed

1004Product.indd 44 9/25/19 1:11 PM

 INSIGHTS INSI GHTS

PRIZE ES SAY

NEUROBIOLOGY

Outside-in
Rethinking the etiology of
autism spectrum disorders

By Lauren Orefice commonly experience aberrant tactile sen- Transverse section of a mouse dorsal root
sitivity: a seemingly innocuous touch, such ganglion (DRG), showing neurons transduced with

A
utism spectrum disorders (ASDs) are as a gentle breeze or a hug, can be unpleas- a virus expressing the b3 subunit of the GABAA
thought to arise exclusively from ab- ant or even painful (1, 2). In fact, sensory receptor (green), NF200 expression in large-diameter
errant brain function. Our research overreactivity is so common that it is now a DRG neurons (yellow), and Hoechst (blue).
proposes a surprising revision of diagnostic factor for ASD (2).
this view. We have discovered that We therefore sought to determine whether drome model), Fmr1 (fragile X syndrome
peripheral sensory neurons—neu- somatosensory circuits were affected in ASD, model), and Cntnap2, as well as a maternal
rons outside the brain—are key sites at and whether altered tactile sensitivity might immune activation model (5–10).
which ASD-related gene mutations have a contribute to other ASD traits. Our goal was Taking advantage of the well-characterized
critical impact. Dysfunction of peripheral to focus on tractable symptoms—somatosen- somatosensory circuitry (11), we then sought
neurons in mouse models disrupts central sory abnormalities—as an entry into these to determine in which cell types ASD-related
nervous system development and causes complex, heterogeneous disorders. genes function for normal tactile reactivity. To
ASD-related phenotypes, including sen- do so, we used conditional mouse genetics to
sory overreactivity, social impairments, PERIPHERAL SENSORY NEURON selectively delete ASD-related genes in differ-
and anxiety-like behaviors. These unex- DYSFUNCTION UNDERLIES TACTILE ent cell types throughout the nervous system
pected findings have led us to propose and OVERREACTIVITY and assessed touch sensitivity. Surprisingly,
demonstrate that peripheral neurons are a We used genetic mouse models for ASD, loss of Mecp2 in excitatory neurons of the
tractable and effective therapeutic target to combined with behavioral testing, anatomy, forebrain produced no major ASD pheno-
improve some ASD-related behaviors. and electrophysiology, to define the etiology types. Instead, genetic mutations in periph-
of aberrant tactile sensitivity in ASD. Using eral somatosensory neurons (neurons that
TOO MUCH TOUCH: TACTILE assays to assess sensitivity to light touch, we receive touch signals at the skin) accounted
SENSITIVITY IN ASD found that monogenic and environ- for the touch overreactivity ob-
How did we come to this new perspective? mental mouse models of ASD dem- served in Mecp2 mutant mice (3, 4).
Our launching point was a curious, unex- onstrated tactile overreactivity (3, We found that the physiologi-
PHOTO: LAUREN OREFICE

plained clinical finding. People with ASD 4). The models used included mice cal deficits that cause tactile over-
with germline mutations in ASD- reactivity across different ASD
Department of Molecular Biology, Massachusetts related genes, including Mecp2 models are distinct. In our studies,
General Hospital, and Department of Genetics,
Harvard Medical School, Boston, MA 02114, USA. (Rett syndrome model), Gabrb3, peripheral sensory neurons lack-
Email: orefice@molbio.mgh.harvard.edu Shank3 (Phelan-McDermid syn- ing Mecp2 or Gabrb3 exhibited de-

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 45

1004Essay.indd 45 9/27/19 5:57 PM

 INSIGHTS | P R I Z E E S SAY

creased inhibitory signaling via the GABAA pairments in adult mice. But how? line mutations in Mecp2 and Shank3 showed
receptor (3), whereas those lacking Shank3 Similar to ASD, anxiety-like behaviors major improvements in body condition,
were hyperexcitable as a result of potassium and social impairments are traditionally at- brain development, anxiety-like behaviors,
channel loss (4). Peripheral neuron dysfunc- tributed to brain circuits. However, decades and some social impairments (4). Because
tion is thus a common feature of ASD mouse of research show that the brain does not de- the administered compound does not cross
models, although this dysfunction can arise velop in isolation. Rather, sensory inputs, in- the blood-brain barrier, it avoids the pro-
via multiple molecular mechanisms. cluding light, sound, touch, and many other found side effects associated with drugs that
environmental signals, guide brain develop- act on the central nervous system (15).
PERIPHERAL SENSORY NEURON ment (12, 13). Together with a growing body of other
DYSFUNCTION CONTRIBUTES TO SELECT We reasoned that if sensory perception is research (3, 4, 16–18), our work highlights
ASD-RELATED BEHAVIORS profoundly altered early in development, this that peripheral sensory neurons have a
The importance of peripheral sensory neu- may affect one’s experience of the world and major role in ASD and that selective treat-
rons for touch processing is interesting, but lead to large changes in behavior. Inspired by ment of these neurons has the potential to
the general behaviors of the conditional experiments conducted by Hubel and Wiesel improve some developmental and behav-
mutant mice we studied were even more re- (13), we found that ASD mutations in periph- ioral abnormalities associated with ASD.
markable. Mice with loss of Mecp2, Gabrb3, eral sensory neurons are sufficient to alter We are moving toward measuring touch
or Shank3 only in peripheral sensory neu- developmental and functional properties of overreactivity in humans with ASD and
rons demonstrated profound social impair- specific brain circuits (4). pursuing modulation of peripheral neuron
ments and anxiety-like behaviors reminiscent excitability as a potential clinical therapy.
of those observed in patients (3, 4). Further- A NOVEL THERAPEUTIC TARGET? Our work therefore proposes an exciting
more, selective restoration of gene function Rates of ASD diagnosis are increasing, with revision of ASD etiology and therapeutics,
only in peripheral sensory neurons, and not 1 in 59 people in the United States reported highlighting a genetic-environmental in-
in the brain, was sufficient to normalize tac- to be living with ASD. However, there are terplay at the center of ASD. j
tile behaviors, anxiety-like behaviors, and no FDA-approved treatments for core ASD
REF ERENCES AND NOTES
some social behaviors (3, 4). symptoms (14). Might it be possible to im-
1. C. J. Cascio, J. Neurodev. Disord. 2, 62 (2010).
Of course, peripheral sensory neurons are prove peripheral sensory neuron function 2. Diagnostic and Statistical Manual of Mental Disorders:
not the sole locus of dysfunction in ASD. and, by ameliorating touch overreactivity, DSM-5 (American Psychiatric Association, ed. 5, 2013).
Restoration of sensory neuron function did help relieve other related ASD symptoms? 3. L. L. Orefice et al., Cell 166, 299 (2016).
4. L. L. Orefice et al., Cell 178, 867 (2019).
not improve a number of ASD-related phe- Despite disparate pathophysiological 5. J. Guy, B. Hendrich, M. Holmes, J. E. Martin, A. Bird,
notypes, including motor dysfunction, early mechanisms, a common signature across Nat. Genet. 27, 322 (2001).
lethality, respiratory deficits, and memory ASD mouse models appears to be an in- 6. C. Ferguson et al., BMC Neurosci. 8, 85 (2007).
7. J. Peça et al., Nature 472, 437 (2011).
impairments in Mecp2 mutant mice, nor creased flow of information from pe- 8. C. M. Spencer, O. Alekseyenko, E. Serysheva, L. A. Yuva-
overgrooming or memory impairments in ripheral sensory neurons to the central Paylor, R. Paylor, Genes Brain Behav. 4, 420 (2005).
Shank3 mutants (3, 4). Nonetheless, our nervous system. We therefore reasoned 9. O. Peñagarikano et al., Cell 147, 235 (2011).
findings reveal a key locus of dysfunction that augmenting inhibitory signals via 10. G. B. Choi et al., Science 351, 933 (2016).
11. V. E. Abraira, D. D. Ginty, Neuron 79, 618 (2013).
underlying tactile overreactivity in distinct GABAA receptors at peripheral sensory 12. D. J. Simons, P. W. Land, Nature 326, 694 (1987).
ASD models, as well as a role for this tactile neurons might reduce tactile overreactiv- 13. T. N. Wiesel, D. H. Hubel, J. Neurophysiol. 26, 1003
overreactivity in the genesis of some aberrant ity. Acute treatment of adult mice with a (1963).
cognitive/social behaviors. GABAA receptor agonist that does not cross 14. J. Baio et al., MMWR Surveill. Summ. 67, 1 (2018).
15. P. R. Tata, J. Rollings, M. Collins, A. Pickering, R. R.
the blood-brain barrier directly reduced
Jacobson, Psychol. Med. 24, 203 (1994).
PERIPHERAL SENSORY NEURON somatosensory neuron excitability and di- 16. A. Bhattacherjee et al., Proc. Natl. Acad. Sci. U.S.A.
DYSFUNCTION ALTERS BRAIN minished tactile overreactivity in six differ- 114, E6952 (2017).
DEVELOPMENT AND FUNCTION ent ASD models (4). 17. J. M. Dawes et al., Neuron 97, 806 (2018).
The loss of ASD-related genes in peripheral We also saw substantial effects with 18. O. Perche et al., Front. Cell. Neurosci. 12, 96 (2018).

PHOTOS: (TLEFT TO RIGHT) COURTESY OF LAUREN OREFICE; ZSOFIA BOCZAN; HAROLD SHAPIRO
sensory neurons is therefore sufficient to chronic administration of this agonist start-
cause anxiety-like behaviors and social im- ing early in postnatal life. Mice with germ- 10.1126/science.aaz3880

GRAND PRIZE WINNER FINALIST FINALIST

Lauren Orefce András Szőnyi Zvonimir Vrselja
Lauren Orefice received her András Szőnyi received Zvonimir Vrselja received
B.S. in biology from Boston undergraduate degrees in his M.D. and Ph.D. from J. J.
College and her Ph.D. in medicine and a Ph.D. in neu- Strossmayer University in
neuroscience from George- rosciences from the Semmel- Croatia. He completed his
town University. After her weis University in Budapest, postdoctoral training in the
postdoctoral work at Harvard Medical School, Hungary. He performed research in the Insti- laboratory of Nenad Sestan at Yale School
Orefice started as an assistant professor in the tute of Experimental Medicine of the Hungar- of Medicine, where he continues to work as
Department of Molecular Biology at Massa- ian Academy of Sciences. Currently, Szőnyi is associate research scientist. His research
chusetts General Hospital and the Department a postdoctoral fellow in the Friedrich Miescher focuses on understanding how brain cells
of Genetics at Harvard Medical School in 2019. Institute for Biomedical Research in Basel, react to anoxic injury following circulatory
Her lab studies the development and function Switzerland. He studies the cellular mecha- arrest, and how such cells can be structur-
of somatosensory circuits and the ways in nisms of learning and memory formation in ally and functionally recovered by developing
which somatosensation is altered in develop- mice using in vivo imaging and optogenetics. a perfusion technology.
mental disorders. www.sciencemag.org/content/336/6461/46.1 www.sciencemag.org/content/336/6461/46.2

46 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004Essay.indd 46 9/27/19 5:57 PM

 INSIGHTS

NEUROBIOLOGY
PRIZE ES SAY

FINALIST
Conducting memory formation
András Szőnyi The nucleus incertus in the brainstem orchestrates the
András Szőnyi
received under- formation of contextual memories
graduate degrees
in medicine and a By András Szőnyi of Tamás Freund, I hypothesized that the
Ph.D. in neurosci-
little-known nucleus incertus (NI) would be
ences from the Semmelweis Uni-

L
istening to the harmonic performance ideal to fulfill this role, as it strongly proj-
versity in Budapest, Hungary. He
of a symphonic orchestra is a peculiar ects to the septo-hippocampal system (11).
performed research in the Institute
of Experimental Medicine of the
experience indeed. The music changes
Hungarian Academy of Sciences. dynamically: Sometimes the strenu- HIPPOCAMPAL DENDRITE–TARGETING
Currently, Szőnyi is a postdoctoral ous voice of the brasses dominates, INTERNEURONS ARE INHIBITED BY THE
fellow in the Friedrich Miescher In- sometimes only a faltering violin can NUCLEUS INCERTUS
stitute for Biomedical Research in be heard. The musicians, masters of their Using cell-type–specific anatomical tracing,
Basel, Switzerland. He studies the instruments, play together, but they form a electron microscopic analysis, and in vitro
cellular mechanisms of learning and perfect harmony only by following the in- optogenetics, we discovered that the NI se-
memory formation in mice using in structions of the conductor to play loudly or lectively targets SOM-positive dendrite-tar-
vivo imaging and optogenetics. softly, or to wait in silence. geting interneurons in the CA1 and that it
www.sciencemag.org/content/ The cofiring of subpopulations of pyra- establishes GABAergic inhibitory synapses
366/6461/46.1 midal cells in the CA1 region of the hip- on these interneurons (12). The NI also in-
pocampus encodes the episodic hibits glutamatergic and choliner-
memories of our life much like gic neurons in the MS that send
the voices of the instruments subcortical excitatory inputs onto
form a symphony. To fulfill this SOM-positive hippocampal inter-
role, CA1 pyramidal cells receive neurons. NI neurons frequently
multisensory information from project into the hippocampus and
the CA3 region and direct sen- MS simultaneously, suggesting
sory-related information from that they can inhibit hippocam-
the entorhinal cortex (1, 2). The pal SOM-positive interneurons
information carried by these inputs is as- through both a direct and an indirect path-
sociated by CA1 pyramidal cells to encode way at the same time.
episodic memories (3, 4).
The number of CA1 pyramidal cells that THE NUCLEUS INCERTUS IS RAPIDLY
encode a memory trace must be tightly regu- ACTIVATED BY SALIENT SENSORY STIMULI
lated. The participation of too many may On the basis of our research, NI neurons
engender memory interference, whereas the should “know” if something needs to be
activation of too few pyramidal cells will lead memorized. We therefore tested whether
to unstable memories (5, 6). γ-Aminobutyric the NI is activated by salient sensory inputs,
acid (GABA)–mediated inhibitory interneu- which would allow them to regulate SOM-
rons control how many CA1 neurons are re- positive hippocampal interneurons at the
cruited during this process (7). right time.
Somatostatin (SOM)–positive dendrite- We labeled NI cells with a fluorescent cal-
targeting interneurons play an essential cium indicator and performed two-photon
role in the recruitment of memory-encod- imaging of the NI fibers in the CA1 region of
ing CA1 pyramidal cells. SOM-positive in- head-fixed, awake mice. The mice ran on a
terneurons restrict direct sensory-related treadmill while they received different sen-
information from reaching most of the py- sory stimuli.
ramidal cells during memory formation and We found that NI fibers were activated
are activated by the excitatory glutamater- by different sensory stimuli at a different
gic and cholinergic cells of the medial sep- rate. Aversive air-puffs and water rewards
tum (MS) (8–10). evoked a larger response in NI fibers than
But who is the conductor of this sym- did light flashes or auditory tones, suggest-
phony? Which inputs help fine-tune the ing that the NI is fine-tuned by different
number of CA1 pyramidal cells that en- sensory inputs on the basis of their rel-
code a given memory trace? Together with evance and/or modality.
PHOTO: ZSOFIA BOCZAN

my mentor, Gábor Nyiri, in the laboratory Using rabies virus tracing, we also ex-
amined which brain areas modulate NI
GABAergic neurons directly. We found that
Laboratory of Cerebral Cortex Research, Institute of
Experimental Medicine, Szigony Str. 43, 1083 Budapest, areas that process relevant environmental
Hungary. Email: szonyi@koki.hu inputs, including the lateral habenula or

46-A 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004eEssay_Szonyi.indd 46-B 9/27/19 5:43 PM

 prefrontal cortex, innervated NI GABAergic trol mice, as indicated by higher levels of an aversive stimulus is presented prevents
neurons monosynaptically for potentially fear on the next day, in the environment contextual memory formation. This sug-
rapid activation (12). where they received foot-shocks. By con- gests that the NI may play a role in filtering
trast, there was no difference between the nonrelevant everyday experiences, the un-
THE NUCLEUS INCERTUS BIDIRECTION- non–hippocampus-dependent cued fear necessary encoding of which could lead to
ALLY CONTROLS CONTEXTUAL MEMORY levels of the two mouse groups when we cognitive symptoms (13).
FORMATION presented them with the auditory cue in a By contrast, because optogenetic inhibi-
To better understand how NI GABAergic different environment. These experiments tion of the NI causes pathologically strong
cells orchestrate behavior, we placed virus- confirmed that NI GABAergic cells can fear memory formation, dysfunction of NI
injected mice into an unfamiliar environ- regulate hippocampus-dependent episodic GABAergic neurons may contribute to gen-
ment, where they received mild, aversive memory formation bidirectionally (12). eralized anxiety-like syndromes or to post-
foot-shocks. NI GABAergic cells, or their fi- traumatic stress disorder. j
bers in the CA1 region, were activated opto- THE ROLE OF THE NUCLEUS INCERTUS IN
REF ERENCES AND NOTES
genetically precisely during the foot-shocks. NORMAL AND PATHOLOGICAL MEMORY
1. T. Kitamura et al., Science 343, 896 (2014).
On the next day, mice had to face the same FORMATION 2. T. Kitamura et al., Neuron 87, 1317 (2015).
environment. Whereas control mice showed We have shown that the NI rapidly pro- 3. S. Maren, M. S. Fanselow, Neurobiol. Learn. Mem. 67, 142
appropriate fear behavior, optogenetically cesses salient environmental stimuli. Our (1997).
4. S. Maren, K. L. Phan, I. Liberzon, Nat. Rev. Neurosci. 14,
stimulated mice showed no fear in the same data suggest that the NI tightly controls 417 (2013).
environment. This effect was absent when contextual memory formation by the direct 5. M. Lovett-Barron et al., Science 343, 857 (2014).
optogenetic stimulation was not precisely and indirect inhibition of SOM-positive 6. S. Siwani et al., Neuron 99, 404 (2018).
7. T. F. Freund, G. Buzsáki, Hippocampus 6, 347 (1996).
aligned with foot-shock presentation. dendrite-targeting interneurons in the CA1. 8. F. Fuhrmann et al., Neuron 86, 1253 (2015).
In a separate experiment, mice re- This action fine-tunes the recruitment of 9. R. N. Leão et al., Nat. Neurosci. 15, 1524 (2012).
ceived foot-shocks paired with auditory memory-encoding pyramidal cells on the 10. V. T. Takács et al., Nat. Commun. 9, 2848 (2018).
11. S. Ma et al., Neuroscience 144, 165 (2007).
cues while we optogenetically inhibited basis of the relevance and/or modality of 12. A. Szőnyi et al., Science 364, eaaw0445 (2019).
the NI. Optogenetically inhibited mice the different sensory stimuli. 13. S. Kamiya, Memory 22, 839 (2014).
formed enhanced hippocampus-dependent We have also shown that optogenetic
contextual memories compared to con- stimulation of the NI at the moment when 10.1126/science.aaz3883

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 46-A

1004eEssay_Szonyi.indd 46-C 9/27/19 5:43 PM

 INSIGHTS

NEUROBIOLOGY
PRIZE ES SAY

FINALIST
Destined for destruction?
Zvonimir Vrselja Brain circulation and cell function can be restored after
Zvonimir Vrselja re-
ceived his M.D. prolonged global anoxia
and Ph.D. from
J. J. Strossmayer By Zvonimir Vrselja production process, we possess a model for
University in Croa-
helping us understand how to recover cellu-
tia. He completed his postdoctoral

O
ne need not look beyond the stroke lar functions after prolonged anoxic injury.
training in the laboratory of Nenad
ward to appreciate the brain’s partic-
Sestan at Yale School of Medicine,
where he continues to work as associ-
ular vulnerability to interruptions in WHY TACKLE ONE PROBLEM, WHEN THERE
ate research scientist. His research blood flow. Cut off from circulation, ARE MANY?
focuses on understanding how brain brain function falls precipitously, Devoid of blood flow, brain cells activate
cells react to anoxic injury following leaving patients in critical, and some- multiple deleterious mechanisms that lead
circulatory arrest, and how such cells times, permanently disabled, states. Indeed, to cellular death, if left unchecked (9–11).
can be structurally and functionally this is not surprising, if we consider the Many “plumbing” issues arise when rein-
recovered by developing a perfusion monumental metabolic needs for the fatty stating blood flow, including thrombotic,
technology. organ to accomplish its world-altering tasks. vascular, and perivascular mechanisms
www.sciencemag.org/content/ (12–14). As such, anoxic or ischemic injury
366/6461/46.2 BRAIN CELLS REMAIN VIABLE AFTER is a “dirty” state.
PROLONGED GLOBAL ANOXIA With this in mind, we incorporated cy-
Much evidence has brought into question toprotective agents within the perfusate to
the inevitability of cell death after prolonged promote cellular recovery and to quench
cessation of cerebral blood flow. One obser- excitotoxicity, free radicals, oxidative stress,
vation is particularly important: Human and other processes invoked during cell
brain samples can be harvested death. To overcome the afore-
several hours postmortem and pro- mentioned difficulties that might
cessed for cell and tissue cultures accompany the use of blood, we
with sufficient viability (1, 2). In- specifically designed the perfusate
deed, cell and slice cultures have to be entirely synthetic and acel-
been shown to be experimentally lular. We also incorporated mic-
useful for studying not only indi- roparticles to allow us to monitor
vidual brain cells (3), but also brain perfusion dynamics by ultrasonog-
disorders, brain damage, and neu- raphy (5).
roprotection (4). How do we reconcile these Because the brain cannot survive without
observations with what is known about the support, we also developed a platform that
susceptibility of the brain to anoxia? mimics lungs, kidneys, and a heart. The de-
Because tissue culture studies show that toxification circuit, complete with a special
cells remain viable even hours postmortem, exchange solution, emulates renal function
we hypothesized that the same might be true and ensures proper maintenance of electro-
of the fully intact brain. My research there- lyte, metabolite, and acid-base homeostasis;
fore centered on restoring cellular functions the artificial lungs control dissolved gases;
in the fully intact porcine (pig) brain after and the simulated heart delivers perfusates
prolonged anoxia. To study this phenom- under cardiac-like pressure Waveforms (5).
enon, we developed a cytoprotective solution
and technology to selectively perfuse the PROBING FOR CELLULAR VIABILITY
porcine brain 4 hours after death (5). We began our investigations by perfusing
pig brains with a control solution, which
ONE PERSON’S TRASH IS ANOTHER’S was devoid of cytoprotective agents, start-
UNEXPECTED MODEL ing 4 hours postmortem. After 6 hours, the
Pig brains are readily discarded by the food brains began to assume a custard-like tex-
production industry. They are also large, ture, with no observable perfusion. Micro-
complex, and highly gyrified, with a com- scopically, the brain cells had disintegrated,
position similar to that of human brains making it difficult to distinguish intracellu-
(6)—important factors for simulating the lar spaces from extracellular spaces (5). PHOTO: HAROLD SHAPIRO

pathophysiology of human anoxia (7, 8). Brains perfused with the cytoprotective
By recovering these organs during the food solution, however, maintained low-resis-
tance perfusion for the entire 6 hours, in-
dicating that our perfusate prevented brain
Department of Neuroscience, Yale University,
333 Cedar Street, New Haven, CT 06510, USA. swelling. Histological and immunohistolog-
Email: zvonimir.vrselja@yale.edu ical investigation of multiple independent

46-B 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004eEssay_Vrselja.indd 46-B 9/27/19 5:47 PM

 brain regions demonstrated structurally in- After the perfusion, we microdissected ies will lay the groundwork for the develop-
tact, and morphologically normal, neurons, the hippocampus and used whole-cell patch- ment of a new experimental platform and
microglia, astrocytes, oligodendrocytes, and clamp recordings, which revealed that pyra- may ultimately help us understand how to
endothelial and ependymal cells. midal neurons were still electrochemically salvage damaged or diseased brain cells. j
Although cellular integrity was preserved, viable and demonstrating spontaneous
REF ERENCES AND NOTES
these data did not provide any evidence for synaptic activity. All evidence suggesting
1. R. W. Verwer et al., FASEB J. 16, 54 (2002).
viability. Given that our perfusate contained the presence of these cellular functions was 2. M. Onorati et al., Cell Rep. 16, 2576 (2016).
antiapoptotic agents, we hypothesized that further substantiated by the presence of a 3. B. H. Gähwiler, M. Capogna, D. Debanne, R. A. McKinney,
there would be a reduction in the activation robust global cerebral metabolism, as mea- S. M. Thompson, Trends Neurosci. 20, 471 (1997).
4. J. Noraberg et al., Curr. Drug Targets CNS Neurol. Disord.
of caspase-3, the major execution protein in sured by arteriovenous gradients (5). 4, 435 (2005).
the apoptotic pathway. We did not, however, detect a resurgence 5. Z. Vrselja et al., Nature 568, 336 (2019).
We were able to demonstrate that inter- of global electrical activity within the brain 6. D. W. Howells et al., J. Cereb. Blood Flow Metab.
30, 1412 (2010).
vention with our technology decreased apop- by electrocorticography. This lack of activity 7. P. M. Gross, N. M. Sposito, S. E. Pettersen, J. D.
tosis (5). However, again, we were left without may be due to the various antagonists pres- Fenstermacher, Blood Vessels 23, 261 (1986).
any insight into cellular functionality. ent in the perfusate, or it may simply not be 8. B. R. Ransom, S. B. Baltan, Ann. Neurol. 65, 120 (2009).
9. J. M. Lee, G. J. Zipfel, D. W. Choi, Nature 399 (suppl.), A7
To assess whether the intact cells were possible to reestablish synchronous activity (1999).
indeed functional, we began by testing vas- in the postmortem brain. Longer perfusion 10. D. M. Greer, Semin. Neurol. 26, 373 (2006).
cular dilatory functionality, which we con- studies may provide answers to this question. 11. M. A. Moskowitz, E. H. Lo, C. Iadecola, Neuron
67, 181 (2010).
firmed by administering nimodipine, a drug
12. A. Ames 3rd, R. L. Wright, M. Kowada, J. M. Thurston,
used to treat vasospasm. We observed an in- EXTENDING PERFUSION FOR LONG-TERM G. Majno, Am. J. Pathol. 52, 437 (1968).
crease in flow within the brain—exactly how EXPERIMENTATION 13. J. R. Little, F. W. Kerr, T. M. Sundt Jr., Mayo Clin. Proc.
functionally viable vascular cells react (15). Our findings indicate that cells within the 50, 264 (1975).
14. G. J. Del Zoppo et al., Stroke 17, 1254 (1986).
Next, we probed whether our technol- large mammalian brain are not as fragile 15. C. W. Haws, J. K. Gourley, D. D. Heistad, J. Pharmacol. Exp.
ogy restored glial inflammatory activity by as they were once considered. Indeed, they Ther. 225, 24 (1983).
microinjecting lipopolysaccharide, a Toll- demonstrate an observable resilience to an-
ACKNOWL EDGMENTS
like receptor 4 (TLR4) agonist and immu- oxic injury under specific conditions.
I am grateful to N. Sestan for his mentorship, to co–first
nogenic agent. We observed that glial cells Currently, we are working toward extend- author S. G. Daniele for fruitful and rewarding collaboration,
increased the release of inflammatory mole- ing perfusion time, which will help us to and to all members of the Sestan laboratory.
cules in response to the injections, suggest- better understand the dynamics of cellular
ing that glial functionality was present (5). recovery after extended anoxia. These stud- 10.1126/science.aaz3885

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 46-B

1004eEssay_Vrselja.indd 46-C 9/27/19 5:47 PM

 Movable
Objective
Where
Microscope ®

Science • For 3-photon and 2-photon

Gets
microscopy
• Objective moves 22 mm in
X, Y and Z
Social. • Objective rotates about
optical axis for imaging of
non-horizontal surfaces
and volumes
• Customizable open platform
design
• Resonant and galvo scanning
• Two or four channel detector
system with Hamamatsu PMTs
AAAS.ORG/COMMUNITY and preamplifiers
• Sutter PS-2 / PS-2 LV dual
channel PMT power supply

AAAS’ Member Community is a one-stop destination

for scientists and STEM enthusiasts alike. It’s “Where

MOM
Science Gets Social”: a community where facts ®

matter, ideas are big and there’s always a reason to

The Movable Objective Microscope® (MOM®)
come hang out, share, discuss and explore. is a multi-photon microscope capable of
imaging deep within living specimens when
combined with a Ti:Sapphire Laser. The MOM
design is unique in providing 3-dimensional
objective movement and rotation allowing
the specimen to remain stationary. Many
highly regarded imaging laboratories around
the world use the Sutter MOM, and we
constantly work with our customers to adapt
the design for their changing needs.

PHONE: +1.415.883.0128
FAX: +1.415.883.0572
EMAIL: INFO@SUTTER.COM
WWW.SUTTER.COM

1004Product.indd 47 9/25/19 1:11 PM

 SPECIAL
SP E CI A L SECTION
S ECT I ON

MORE THAN A TOOL FOR

COMMUNICATION
PHOTO: CREDIT GOES HERE AS SHOWN; CREDIT GOES HERE AS SHOWN
By Pamela J. Hines and Peter Stern

REVIEWS
Evolution of vocal learning and spoken language p. 50
The neurobiology of language beyond single-word processing p. 55
From speech and talkers to the social world: The neural processing of human spoken language p. 58
The neural basis of combinatory syntax and semantics p. 62

RELATED ITEMS
EDITORIAL p. 13 PERSPECTIVE p. 33 RESEARCH ARTICLE p. 83

48
48 00 MONTH 20XX • VOL XXX ISSUE XXXX sciencemag.org SCIENCE

1004SpecialIntropage.indd 48 9/30/19 1:17 PM

 H
ow disorienting would it be to travel to a communication in our society, it is no surprise that such a
country where you do not know the language large part of the brain is engaged in the processing of lan-
and cannot read the street signs because the guage. Syntax, grammar, and word selection define com-
letters are unfamiliar? You would not even be mon ground and di erentiate distinct human languages.
able to distinguish an exit sign from a stop How do the young learn to make sense of the cacophony?
sign. Language enables the sharing of infor- Language is central to our existence as humans. With
mation, knowledge, and predictions about language we entertain, agitate, and encourage each
the world around us. Language also gives us other; we build community. Without language we are
poetry and fiction through which to express each islands unto ourselves. Language turns us into so-
feelings and speculate about things or events that are cial beings. Some argue that the capacity for language is
not and may never become real. crucial to the runaway success of our species. Both the
Humans and animals communicate through multiple human brain and the language it enables help shape the
modalities, such as sight, odor, and touch. In this spe- world we inhabit today.
ILLUSTRATION: EIKO OJALA

cial issue, we focus on the communication of sound—

particularly language. What is language? Is human
Simplifed forms of language are useful for linguistic study. Two-word phrases
language special, or is it simply at the extreme end of an can vary the conceptual specifcity of nouns and how functional elements of
evolutionary spectrum? How does the brain produce and the language ft together. The words in the background (in English and German)
interpret language? Given the fundamental importance of are examples that have been used in research.

49

1004SpecialIntropage.indd 49 9/30/19 1:17 PM

 L A N G UA G E AN D TH E B RA I N

REVIEW my co-workers and I proposed a continuum

hypothesis, where different species have vary-
Evolution of vocal learning and spoken language ing degrees of vocal learning that evolves in
stepwise manner (fig. S1) (2, 6). This hypothesis
Erich D. Jarvis1,2 is plausible when considering large differences in
vocal learning complexity among well-established
Although language, and therefore spoken language or speech, is often considered unique to humans, the past vocal learners, and so it should not be surprising
several decades have seen a surge in nonhuman animal studies that inform us about human spoken language. that so-called vocal nonlearners vary too.
Here, I present a modern, evolution-based synthesis of these studies, from behavioral to molecular levels Such variation can be influenced by anatom-
of analyses. Among the key concepts drawn are that components of spoken language are continuous ical mechanisms of how sounds are produced,
between species, and that the vocal learning component is the most specialized and rarest and evolved by such as learned raspberry lip-smacking in chim-
brain pathway duplication from an ancient motor learning pathway. These concepts have important panzees (9) or diaphragm-induced coughing
implications for understanding brain mechanisms and disorders of spoken language. in Koko (3). In contrast, advanced vocal learners
have voluntary control of not only the oral-facial

T
articulators (lips, tongue, beak, and jaw) but
he faculty of spoken language can be Discrete versus continuum hypotheses of also the larynx (in mammals) or syrinx (in birds)
thought of as consisting of multiple com- spoken-language components (10, 11). The continuum hypothesis does not
ponent traits, some ubiquitous among Language, and therefore spoken language and negate that advanced vocal learning is conver-
species and a few specialized to rare groups some of its components, is sometimes presented gent. Further, different continuums for differ-
of species or just to humans (Fig. 1) (1). A as all or none: a species either has vocal learning ent components could explain why auditory
ubiquitous component is auditory learning, the or it does not (4). However, differences can be learning is more advanced than vocal learning
ability to learn and remember novel sound as- a matter of degree. For example, rudimentary in many species, why it is easier to listen (re-
sociations (2); for example, a dog or nonhuman vocal plasticity and learning have been found in ceptive language) than to speak (productive
primate can learn the meaning of the sound some species assumed to be vocal nonlearners, language) in a second language, and why a
“sit” or of word combinations such as “come including mice (6) and nonhuman primates nonverbal autistic child has more receptive
here, boy” or “play the flute” (3). Also ubiquitous (2, 7). Some great apes have been taught to than productive spoken language.
is vocal usage learning, the ability to learn to produce rudimentary sign language: By her
produce innate or learned sounds in unfamiliar 30s, Koko the gorilla was producing more than What anatomy makes vocal learning and
contexts (2)—for example, a dog’s or a nonhu- 1000 American Sign Language (ASL) signs, in spoken language special?
man primate’s ability to learn to bark or call word combinations that were not taught to The search for the biological substrates of spoken
when requesting food or signaling alarm. Rarer her, and recognizing ~2000 spoken words (8). language and vocal learning in humans and
is vocal production learning, or simply vocal Although none of these species demonstrate song-learning birds has led to six popular hypo-
learning, the ability to imitate sounds (2), which advanced vocal learning, syntax, or semantics, theses (Fig. 2), described below.
has thus far been found only in humans, ceta- they are not completely lacking. In prior studies,
ceans, pinnipeds, bats, and elephants among Brain size or neuron density impacts
mammals, and songbirds, parrots, and hum- spoken-language circuits
mingbirds among birds (2). Each of these groups This hypothesis proposes that a larger brain al-
has closely related species that lack vocal learn- lows more neurons for speech and vocal learning
Vocal
ing, suggesting independent evolution of the production (Fig. 2A). However, brain size is not correlated
trait (4). For other components, syntax, rules learning with vocal learning: hummingbirds, with their
governing sequences of sound, semantics, and tiny brains, can imitate complex vocalizations,
Vocal usage Speech/song
pragmatics are also present in nonhuman ani- learning production whereas chimpanzees, with their much larger
mals but are more advanced in humans (1). brains, cannot (2). Although humans have the
For example, black-capped chickadee songbirds Spoken largest brains among primates, it is a scaled-up
have songs with simple syntax for mating and Language primate brain in size and neuron density (12).
Auditory Speech/song
learned calls that signal predators and predator learning perception Other vocal learning mammals, cetaceans and
size, but they are not known to combine these elephants, have larger brains but also larger
vocalizations into longer sequences with dif- bodies (12). In contrast, two of the three vocal
ferent meanings (5) or have hierarchical syn- Semantics and Syntax learning bird species, songbirds and parrots,
tax. Here, I present a comparative synthesis pragmatics have forebrain neuron densities two times that
of spoken language as a form of learned fore- of vocal nonlearning bird species (fig. S2A).
brain sensory-motor communication, with some Perhaps the higher density in some vocal learn-
components found in most vertebrates to vary- ing birds and the scaled-up human brain accom-
ing degrees and a highly specialized advanced Fig. 1. Multicomponent view of spoken language. modated the space needed for extra neurons
vocal learning component found only in a few Spoken language, or speech, is viewed here as a of the vocal learning and spoken-language
species, with humans being the most advanced combination of seven component traits. These traits circuits, without losing older circuits and
in all components, and with all components overlap with components studied in linguistics: seman- while maintaining brain-to-body size ratios.
combined into one trait. I consider spoken tics, pragmatics, syntax, phonology, and morphology.
language and speech as equivalent. Red text indicates components that are rarest among Vocal organ with greater capacity
vertebrates. Most components could be continuous for vocalization diversity
among species (gradient in the center), with humans This hypothesis proposes that loss of air sacs
1
Laboratory of Neurogenetics of Language, The Rockefeller being the most advanced. [Diagram and components and presence of a permanently descended
University, New York, NY, USA. 2Howard Hughes Medical
Institute, Chevy Chase, MD, USA. modified from a figure produced by Tecumseh Fitch for larynx in humans (13) or additional intrinsic
Email: ejarvis@rockefeller.edu language broadly, and used with permission] syrinx muscles in songbirds (14) endowed them

50 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

with the ability to produce a greater variety of A B with connectivity similar to that of human LMC
sounds (Fig. 2B). However, subsequent studies General brain diference Vocal organ and the analogous songbird robust nucleus of
Larger brain and/or diference
found that vocal nonlearning mammals de- higher density of neurons Descended or more
the arcopallium (RA) (Fig. 3, A and B, and fig.
scend their larynx when lifting their heads that impacts vocal complex vocal organ S4). However, unlike those of humans and
upward and vocalizing. A nonhuman primate learning circuits vocal learning birds, the mouse LMC neurons
(baboon) larynx can be made in situ to produce non-VL VL non-VL VL are embedded in a region that controls non-
the majority of sounds made by the human vocal motor behaviors, does not share spe-
A2 Br A2 Br A2 Br A2 Br
larynx (fig. S3, A and B). Other nonhuman cialized gene regulation with vocal learners, is
mammals (lions, koalas, and some ungulates) not necessary for producing normal vocaliza-
have independently evolved a permanently tions, but seems necessary for modulating pitch
descended larynx (13). It is more likely that (6, 15). In birds, a rudimentary RA was found
the air sacs and descended larynx facilitated in a suboscine, a close relative of songbirds,
lower-formant frequencies, allowing an animal but not in quail, a distant relative (20). These
to acoustically exaggerate its size (13). Among findings support the continuum hypothesis of
birds, syrinx muscle complexity does not cor- vocal learning. More details on hypothesis 3 are
relate with vocal learning (fig. S3C), but it is likely provided in the supplementary text of the sup-
that complex vocal musculature allows a vocal C D plementary materials.
nonlearner to produce a greater variety of in- Forebrain pathway Direct to vocal motor
Presence versus absence neurons Direct versus indirect motor cortex–to–
nate sounds to compensate for lack of forebrain- of a forebrain vocal Direct or enhanced motor
driven vocal learning (14). Thus, vocal organ learning pathway cortex–to–brainstem vocal brainstem vocal motor neuron connection
differences cannot explain song and speech motor neuron connection This hypothesis proposes that a fundamental
diversity in vocal learning birds and humans. non-VL VL non-VL VL transition to the evolution of vocal learning
LMC/RA LMC/RA LMC/RA and spoken language was a change from (or an
Forebrain pathways for vocal learning and speech addition to) an indirect to a direct projection
This hypothesis proposes that only humans from human LMC layer 5 neurons and avian RA
and other vocal learning species have a fore- projection neurons to brainstem vocal motor
brain circuit that controls song and speech neurons (Fig. 2D and red arrows in Fig. 3),
(Fig. 2C) (4). All three song-learning bird lineages enabling fine motor control of vocalizations
share seven cerebral nuclei, which make up in humans and song-learning birds (1, 2, 18).
a posterior vocal pathway for production of In the context of the continuum hypothesis,
learned vocalizations (Fig. 3A, yellow) and mice LMC layer 5 neurons (6) and the RA-like
an anterior vocal pathway for vocal imitation region in suboscines (20) make sparse direct
E F
(Fig. 3A, red) that are not found in vocal non- Direct auditory to vocal Language module
projections (one to three innervating axons
learning species (Fig. 3C). I propose that hu- Direct or enhanced Internalization module per vocal motor neuron) compared with the
mans have analogous brain regions, which secondary auditory–to– that interacts with dense projections (up to hundreds of inner-
Broca’s cortex connection auditory and vocal circuits
include dorsal and ventral laryngeal motor vating axons per vocal motor neuron) in hu-
non-VL VL non-VL VL
cortices (dLMC and vLMC) responsible for LM
mans and song-learning birds (Fig. 2D and
speech production, and premotor LMC and A2 Br A2 Br A2 Br A2 Br fig. S4). This suggests that density of the direct
Broca’s area responsible for speech acquisi- projection may influence the degree of learned-
tion and higher-level speech functions (Fig. vocalization production. Experimentally manip-
3B) (4, 10, 15–17). The different songbird song ulating mice to express less of the repulsive axon
nuclei have cell types that may correspond to guidance receptor PlexinA1 in layer 5 neurons
the different cortical layers and the striatum levels similar in humans caused them to have
and thalamic regions of the human spoken- a denser direct projection to forelimb motor
language pathway (Fig. 4). Input into these neurons and greater forelimb manual dexter-
specialized song and speech circuits comes ity (21). Depending on the species, the avian
from auditory, somatosensory, and other path- Forebrain Vocal motor neurons Vocal organ RA song nucleus also makes direct or indirect
ways, but these other pathways are found in projections to articulatory (e.g., beak, jaw,
all vocal nonlearning species investigated to Fig. 2. Hypotheses of anatomically unique tongue) and respiratory motor neurons (fig S4A).
date (Fig. 3) (2, 4), which would explain why features to vocal learning and spoken language. The human oral-facial motor cortex (OMC)
auditory learning is more ubiquitous among Colors indicate brain subdivisions and vocal organs. between dLMC and vLMC is presumed to do
species. I propose that Wernicke’s area and its Arrows indicate neuroanatomical connections; red so as well (16). Direct innervation of tongue
network involved in speech perception were arrows, continuum hypothesis versions. (A) Larger motor neurons also exists in nonhuman pri-
present in the vertebrate lineage before being brain and/or higher density of neurons that affect vocal mates (18). This would explain why limited vocal
elaborated in humans. learning brain circuits. (B) Descended or more complex learners have more voluntary control for pro-
In the context of the continuum hypothesis, vocal organ. (C) Presence versus absence of a forebrain ducing imitated sounds using articulators like
it has been proposed that nonhuman primates vocal learning pathway. (D) Direct or enhanced motor the tongue and lips than they do for the larynx.
possess a premotor vLMC as the ancestral pre- cortex–to–brainstem vocal motor neuron connection.
cursor of human primary vLMC (Fig. 3D) (17, 18). (E) Direct or enhanced secondary auditory–to–Broca’s Connection between auditory cortex
However, premotor vLMC is not required for cortex connection. (F) Separate language model. A2, and speech premotor cortex
producing nonhuman primate vocalizations secondary auditory cortex; Br, Broca’s area; LM, This hypothesis proposes that direct connec-
(18), and a rudimentary primary vLMC may language module; LMC, laryngeal motor cortex; RA, tions between secondary auditory cortex (A2
already exist in nonhuman primates (Fig. 3D) robust nucleus of the arcopallium; non-VL, nonvocal and Wernicke’s area) and vocal premotor cortex
(19). In mice, a rudimentary LMC was found learner; VL, vocal learner. (preLMC and Broca’s area) may endow humans

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 51

 L A N G UA G E AN D TH E B RA I N

with the auditory-vocal motor integration different vocal learning lineages evolved dif- Vocal learning birds are said to only have
necessary to learn, produce, and perceive spoken ferent solutions to auditory-vocal motor inte- the externalization circuits, without hierarchical
language (Figs. 2E and 3, blue arrows) (22). gration, or that direct connections between syntax and compositional meaning and with-
A set of dorsal pathways is proposed to con- auditory and forebrain premotor areas are not out a Broca’s analog, yet others have proposed
trol auditory-vocal motor learning for speech specific to humans or vocal learners. Analysis of that some of the songbird vocal learning nu-
(to premotor cortex) and hierarchical syntax a greater range of species should bring clarity. clei HVC (letter-based name) or magnocellu-
for language (to Broca’s area), whereas a ven- lar nucleus of the anterior nidopallium (MAN)
tral set is proposed to control lexical and sem- Internalization language versus externalization is analogous to Broca’s (4). Functional mag-
antic aspects of speech and language (fig. S5A) brain circuits netic resonance imaging studies show Broca’s is
(22). But different views exist about the con- This hypothesis proposes that only humans active in language tasks regardless of modal-
nectivity differences between humans and non- have an “internalization” language brain cir- ity, increasing activity with increasing syntax
human primates (2, 22). Mice were found to cuit, historically thought of as a language load (22). In awake patients undergoing sur-
contain a direct A2 connection to their LMC- module, that processes complex algorithms gery, electrical stimulation in Broca’s, and also
M1/M2 region with respectable density (fig. like hierarchical syntax and the merging of in dLMC and vLMC, can lead to inhibition of
S5, D to F). Analogous pathways for auditory- words, which are then expressed through au- ongoing speech and/or hand movements (23);
vocal motor integration in song-learning birds ditory, speech, or limb “externalization” brain but the same LMC regions (Broca’s has not yet
are those that connect forebrain auditory re- circuits shared across species, enabling spoken, been tested) have increased activity mainly
gions to the forebrain vocal learning nuclei signed, and written language in humans (Fig. during speech-related tasks (10, 16, 24).
(Fig. 3A, blue arrow) (2). Vocal nonlearning 2F and fig. S6). A proposed internalization It should be noted that most studies like
birds are not known to have such forebrain brain region is Broca’s area, and a proposed these have not controlled for silent speech
vocal nuclei to project into. It is possible that externalization region is speech LMC (22). production. When we silently speak (i.e., inner
speech, thinking in speech), most of the brain
A Songbird B Human areas, including Broca’s, used for speech produc-
tion and perception show increased activity, as
do the laryngeal muscles, even though no sound
Central aSMA
Av Wernicke’s sulcus is produced (25). When we hear, read, or write
HVC NIf Sylvian words, activity increases in brain regions for
Shelf MO fssure
dLMC
speech production, associated with subvocal-
RA ization muscle activity of the larynx and other
Cup MAN OMC
XIIts AM aSt
A2 A1 vLMC articulators (25). Many limb gestures of ASL are
Broca’s accompanied by “mouth morpheme” movements.
Syrinx
muscles Area X Larynx As hand-control brain regions are adjacent to
DM aDLM muscles
PAG aT preLMC vocalization and oral-facial control regions (16, 17),
my interpretation of imaging results (24, 26) is
that both adjacent regions are active in ASL pro-
C Vocal birds D Primate
duction as a result of hand and oral movements.
In songbirds, electrophysiology and activity-
Central
L2 Wernicke’s dependent gene expression studies revealed that
NCL sulcus
CMM Sylvian the same brain pathways are used to learn and
fssure to produce song (Fig. 3A, red and orange) (4).
NCM
CSt These brain regions and the syrinx muscles
Ai OMC
XIIts pre aSt show singing neural firing patterns when birds
DLM A2 A1 vLMC
LMC apparently dream about singing (e.g., inner
Syrinx
DM
muscles Larynx song) (27). There is no “internalization” circuit
muscles for song syntax separate from an “externaliza-
PAG aT
tion” circuit for song production. Adjacent to
Fig. 3. Brain pathways for vocal learning and spoken language. (A) Vocal learning pathway of songbirds. song learning nuclei are nonvocal motor regions.
(B) Vocal learning and spoken-language pathway of humans. (C) Innate brainstem vocal pathway in vocal nonlearning Translating this to humans: An alternative to the
birds. (D) Vocal pathway in nonhuman primates. Comparable brain regions and connections across species are in the internalization-externalization framework is that
same color and projected on smoothed surface brain images. Orange regions and black solid arrows, posterior the spoken-language brain pathway, inclusive
vocal motor pathway. Red regions and white arrows, anterior vocal pathway. Dashed arrows, connections between the of Broca’s, could be used to learn and produce
two subpathways. Red arrows, specialized direct projection from motor cortex to brainstem vocal motor neurons in speech, whether voiced or silent during read-
vocal learners. Gray regions, innate vocal pathway. Blue regions, auditory regions. Blue arrows, auditory input to ing, writing, thinking, and signing with mouth
specialized vocal learning and spoken-language regions. Subcortical vocal regions are outlined with dashed lines. morphemes. Nonvocal and nonauditory circuits,
Orange and red regions in nonhuman primates (D) are less transparent to indicate continuum hypothesis of a e.g., forelimb and vision, may process hierar-
rudimentary forebrain vocal circuit. A subset of connections are shown for simplicity. A1, primary auditory cortex; A2, chical syntax algorithms as the adjacent speech
ADAPTED BY KELLIE HOLOSKI/SCIENCE
secondary auditory cortex; aDLM, anterior dorsolateral medial nucleus of the thalamus; Ai, intermediate arcopallium; and auditory circuits. This could explain why
Am, nucleus ambiguus; aSMA, anterior supplementary motor area; aSt, anterior striatum speech area; aT, anterior nonhuman primates have greater sign-language
thalamus speech area; Av, avalanche; CMM, caudal medial mesopallium; CSt, caudal striatum; DM, dorsal medial abilities with barely any spoken-language ability.
midbrain nucleus; HVC, a letter-based name; L2, Field L2; dLMC, dorsal laryngeal motor cortex; vLMC, ventral In this view, spoken language and sign language
laryngeal motor cortex; preLMC, premotor laryngeal motor cortex; OMC, oral motor cortex; MAN, magnocellular are the same as speech and signing, respectively.
nucleus of the nidopallium; MO, mesopallium oval nucleus; NCM, nidopallium, caudal medial part; NIf, nidopallium These hypotheses can be reconciled with neuro-
interfacial nucleus; NLC, nidopallium, lateral caudal; PAG, periaqueductal gray; RA, robust nucleus of the arcopallium; physiological recordings in Broca’s and other
XIIts, 12th vocal motor nucleus, tracheosyringeal part. [Figure is updated and modified from (4)] areas during speech and other tasks.

52 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 expression profiles supported the nuclear-to-
LMC-M1 layer 3/6
Av layered hypothesis of avian and mammalian
Controls sequencing
Auditory input cortex relationships, where the avian arcopal-
HVC
NIf
lium in which RA resides has cell types similar
Eference copy to mammalian motor cortex layer 5 neurons,
LMC-M1 Interneuron
layer 2
+
- + and the nidopallium in which HVC resides has
+
+ + cell types similar to layers 2 or 3 (15) (Fig. 4).
PreLMC
MO or Broca’s Gene expression profiles of avian song and human
10ms 10ms 10ms layer 3/6
10ms 10ms
spoken-language brain regions resembled motor
Synfre Infuences
MAN PreLMC
regions more than auditory regions (15) and di-
chain variability verged from the surrounding regions to become
or Broca’s
+ + + layer 2
++ + highly specialized. About 50 to 70 genes, many
LMC-M1 key to neural connectivity, per avian song and
layer 5 RA +
Controls acoustic human spoken-language brain region showed
structure and convergent specialized expression.
respiration Area X +
Tonic+ + Among down-regulated genes was the SLIT1
+
+
- - Spiny
Anterior
striatum
axon guidance ligand (fig. S7, A to C). SLIT1
interaction with its receptor ROBO1 prevents
Motor
neurons
+
- + RVL
aDLM
Tonic
neuron
axon connections from forming. We proposed
+ Anterior neuron
XIIts + thalamus
+ that down-regulation of SLIT1 in songbird RA
Infuences
RAm stereotopy and human LMC layer 5 neurons may produce
Larynx Syrinx
muscles muscles a permissive environment for their axons to form
Expiratory
motor neurons
dense direct projections to brainstem vocal motor
neurons with high levels of ROBO1 (fig. S7D).
This concept is supported by the PlexinA1 down-
Sound Sound regulation in cortical layer 5 and its increased
Fig. 4. Evolutionary view of vocal learning in birds translated to spoken language in humans. Shown is a connections to forelimb motor neurons men-
summary of many labs. Each oval is a songbird song/vocal or respiratory nucleus. White circles, excitatory neurons; tioned earlier (21). Mutations in the ROBO1
black circles, inhibitory neurons; plus sign, excitatory transmitter release; minus sign, inhibitory transmitter release. locus are associated with dyslexia and speech
The RA-projection neurons of HVC fire in 10-ms synfire chains, where RA translates those sequences to control vocal sound disorders (30). Mutations in FOXP2, a
motor and respiratory premotor neurons for the production of each 10 ms of sound through the syrinx (33). MAN and transcription factor that directly regulates SLIT1,
Area X inject variability and stereotopy, respectively, into both HVC and RA (4, 35). Predicted human brain regions and cause a phoneme sequence speech deficit in
neuron cell types that correspond to the vocal learning circuits in songbirds are marked with a human outline. Abbreviations humans and a similar but more rudimentary
and color-coding are the same as Fig. 3. RAm, retroambiguus; RVL, rostral nucleus of the ventral-lateral medulla. syllable sequence deficit in mice, associated
with less localized LMC layer 5 neurons (31). In
Taken together, the most strongly supported opment to innervate brainstem and spinal cord humans, partial gene duplications of the SLIT-
differences in species that have imitative song motor neurons for various muscle groups (Fig. ROBO guanosine triphosphatase 2 (SRGAP2)
and/or speech include: greater density of fore- 5B). In vocal learners, an extraneously dupli- gene (32) encode proteins that act as com-
brain neurons that may accommodate additional cated pathway (Fig. 5C) could innervate brain- petitive inhibitors of full-length SRGAP2, keep-
song and speech brain circuits (hypothesis 1); stem vocal motor neurons and then be selected ing synapses at a higher density and more
an additional or enhanced forebrain vocal motor for a vocal learning phenotype (Fig. 5D). This plastic into adulthood (fig. S2C). All these find-
learning pathway (hypothesis 3); and a novel vocal learning pathway is proposed to have ings suggest that not only convergent changes
or enhanced forebrain-to-brainstem vocal motor been duplicated at least one other time, where in the same genes between species but also on
connection (hypothesis 4). If stronger evidence the parrot inner core song system that is sim- different genes in the same genetic pathway
were found for the other hypotheses, these ilar to songbirds and hummingbirds gave rise within a species are associated with the evolu-
would be additions to hypotheses 1, 3, and 4, to the parrot shell song system unique to them tion of vocal learning and spoken language.
rather than alternatives. The question that re- (29), the human vLMC to the dLMC (17), and
mains is how convergent evolution built similar preLMC to Broca’s area (29). Testing the pathway- Vocal learning continuum hypothesis in three
brain pathways for a complex behavior. duplication hypothesis may require experimen- anatomical stages
tal tools for lineage tracing of neural stem cell I suggest that the common ancestor of verte-
Motor theory of vocal learning origin development. brates had a brainstem pathway for produc-
The finding that vocal learning and speech tion of innate vocalizations with limited vocal
pathways in birds and humans are embedded Convergent genetic changes for vocal learning plasticity, such as the Lombard effect, where
in, or adjacent to, apparent motor learning brain circuits animals increase sound production volume or
pathways (17, 28) led to the motor theory of One prediction of the motor theory of vocal pitch in noisy environments (fig. S8). In some
vocal learning origin, which posits that brain learning origin is that the vocal learning path- species, the forebrain motor learning pathway
ADAPTED BY KELLIE HOLOSKI/SCIENCE

pathways for vocal learning and speech evolved ways should share molecular and functional then duplicated and formed a vocal motor learn-
independently from surrounding motor learn- similarities with adjacent motor learning path- ing pathway with weak direct projections to
ing pathways found in all species and thus share ways but diverge in certain neural connectivity the brainstem vocal motor neurons. Thereafter,
a deep homology (28). The proposed mechanism genes, such as those that control a dense direct this forebrain vocal motor learning pathway
is brain evolution by brain pathway duplication projection to brainstem vocal motor neurons. expanded in neuron numbers causing greater
(Fig. 5) (28, 29). A motor learning pathway that My colleagues and I tested this prediction by density of neurons in the forebrain, moved out-
already receives auditory input may be repli- profiling the expression of thousands of genes side of the motor learning pathway, and gained
cated multiple times during embryonic devel- in avian and primate brains (15). The gene dense direct projections to brainstem vocal motor

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 53

 L A N G UA G E AN D TH E B RA I N

A Vocal innate pathway B Motor learning pathway C Pathway duplication D Vocal learning pathway

Cerebellum Cerebrum HVC LMAN

Av LMO

+ = RA
NIf

XII PMN
DM DLM MN XII LAreaX
Thalamus
Midbrain Syrinx
Syrinx Hindbrain Limb Limb
muscles muscles muscles muscles

Pallium/Cortex Striatum Pallidum Thalamus Midbrain Hindbrain

Fig. 5. Brain pathway duplication origin of vocal learning motor pathway. pathway in songbirds, which has similarities to the surrounding motor learning
(A) Innate vocal brainstem pathway found in all vertebrate species that vocalize, pathway in (B) connected to the innate vocal pathway in (A). Abbreviations and
shown here for birds. (B) Motor learning pathway found in all vertebrate species. arrow color-coding are the same as Fig. 3. MN, motor neuron; PMN, premotor
(C) Proposed additional forebrain motor learning pathway duplication that neurons in reticular formation; LMAN, lateral MAN; LMO, lateral MO. [Hypothesis
connects to the innate brainstem vocal pathway. (D) Resultant vocal learning based on (28, 29)]

neurons. Finally, the vocal learning pathway Some of the genes are specialized to the vocal 16. B. K. Dichter, J. D. Breshears, M. K. Leonard, E. F. Chang, Cell
then duplicated one or more times and took on learning circuit (37). I predict that humans may 174, 21–31.e9 (2018).
17. M. Belyk, S. Brown, Neurosci. Biobehav. Rev. 77, 177–193 (2017).
additional specialized gene regulation and con- also have song- and speech-driven gene regu- 18. K. Simonyan, Curr. Opin. Neurobiol. 28, 15–21 (2014).
nections, resulting in the advanced vocal learning lation in the spoken-language brain pathway, 19. C. M. Cerkevich, P. L. Strick, in Society for Neuroscience (2015).
pathways we find in parrots and in humans. in a cell type–specific manner (Figs. 3B and 4). 20. W. C. Liu, K. Wada, E. D. Jarvis, F. Nottebohm, Nat. Commun.
Neurophysiology experiments show that audi- 4, 2082 (2013).
Brain mechanisms of vocal learning and 21. Z. Gu et al., Science 357, 400–404 (2017).
tory responses in the vocal learning pathway 22. A. D. Friederici, N. Chomsky, R. C. Berwick, A. Moro,
spoken language are suppressed when songbirds sing (38). In J. J. Bolhuis, Nat. Hum. Behav. 1, 713–722 (2017).
The evolution-based findings allow us to make humans as well as nonhuman primates (mar- 23. J. D. Breshears, D. G. Southwell, E. F. Chang, Neurosurgery 85,
E496–E501 (2019).
predictive translations between species. For moset), there is also suppression in the audi-
24. A. Basilakos, K. G. Smith, P. Fillmore, J. Fridriksson,
example, the HVC-to-RA projection neurons tory cortex with vocalization (16, 39). In mice, E. Fedorenko, Cereb. Cortex 28, 1816–1830 (2018).
of songbirds fire sparsely in a synfire chain with auditory information enters the motor cortex 25. B. Alderson-Day, C. Fernyhough, Psychol. Bull. 141, 931–965 (2015).
10-ms time resolution thought to sequence sounds (Fig. S5, D to F) and, like in the songbird vocal 26. K. Emmorey et al., Neuroimage 17, 812–824 (2002).
27. B. K. Young, G. B. Mindlin, E. Arneodo, F. Goller, PeerJ 5, e4052
within and among syllables (33) (Fig. 4). RA learning pathways, is gated off in motor re- (2017).
in turn translates the forebrain signals to the gions when the mice move (40). This suggests 28. G. Feenders et al., PLOS ONE 3, e1768 (2008).
brainstem vocal and respiratory neurons to that the suppression of auditory input into the 29. M. Chakraborty, E. D. Jarvis, Philos. Trans. R. Soc. London Ser.
B 370, 20150056 (2015).
dictate acoustic structure of syllables. The HVC- vocal pathway during vocalization in vocal
30. K. Hannula-Jouppi et al., PLOS Genet. 1, e50 (2005).
to–Area X projection neurons send a corollary learners is an ancestral trait inherited from 31. J. Chabout et al., Front. Behav. Neurosci. 10, 197 (2016).
efference copy of the sequence into the striatal the adjacent motor pathway. 32. C. Charrier et al., Cell 149, 923–935 (2012).
Area X vocal nucleus of the anterior forebrain Certain mechanisms will not be shared be- 33. R. H. R. Hahnloser, A. A. Kozhevnikov, M. S. Fee, Nature 419,
65–70 (2002).
pathway (Fig. 4) (34). If the human spoken- cause of differences in avian and mammalian 34. T. W. Troyer, A. J. Doupe, J. Neurophysiol. 84, 1204–1223 (2000).
language pathway functions in a similar man- cortical organization, cell types, and peripheral 35. M. S. Fee, J. H. Goldberg, Neuroscience 198, 152–170 (2011).
ner, then I predict that: (i) human LMC layer musculature. Despite these differences, underly- 36. O. Whitney et al., Science 346, 1256780 (2014).
37. H. Horita et al., PLOS ONE 7, e42173 (2012).
2-3 neurons fire in a synfire chain (like HVC) ing principles, as outlined here, do translate 38. K. Hamaguchi, K. A. Tschida, I. Yoon, B. R. Donald, R. Mooney,
onto the LMC layer 5 neurons (like RA) to con- across different systems. Although song and eLife 3, e01833 (2014).
trol a defined millisecond time resolution for speech pathways in vocal learning birds and 39. S. J. Eliades, X. Wang, Hear. Res. 348, 98–111 (2017).
producing learned phoneme and word sequen- humans are specialized, the majority of genes, 40. D. M. Schneider, R. Mooney, Curr. Opin. Neurobiol. 33, 78–84 (2015).

ces; and (ii) some LMC neurons send an ef- neural connections, and physiology are similar AC KNOWLED GME NTS
ference copy to the anterior speech striatum to their adjacent brain pathways. I thank members of the Jarvis lab for critical discussions and
(Figs. 3B and 4). During singing, songbird MAN comments, especially C. Theofanopoulou, L. Shalmiyev, C. Vargas, and
neurons inject variable neural activity into the G. Gedman for critical reading and discussions of the manuscript. I
RE FERENCES AND NOTES thank E. Chang, M. Long, and K. Emmorey for critical discussions on
vocal motor pathway (to RA and HVC), and Area 1. W. T. Fitch, Psychon. Bull. Rev. 24, 3–33 (2017). their work on brain electrophysiology and imaging during spoken-
X in the striatum modulates or constrains that 2. C. I. Petkov, E. D. Jarvis, Front. Evol. Neurosci. 4, 12 (2012). language, sign-language, and other tasks in humans. I thank other
variability (Fig. 4) (35). Similarly, I predict that 3. M. Perlman, N. Clark, Anim. Cogn. 18, 1165–1179 (2015). members of the animal and human communication communities and
4. E. D. Jarvis, Ann. N. Y. Acad. Sci. 1016, 749–777 (2004). the brain evolution community for valuable discussions over the years
human layer 2-3 motor neurons of premotor 5. C. N. Templeton, E. Greene, K. Davis, Science 308, 1934–1937 that have helped formulate the hypotheses proposed. I apologize to
LMC or Broca’s may innervate both dLMC and (2005). those whose work I did not cite due to the need to keep citations
vLMC to inject acoustic variability in speech 6. G. Arriaga, E. D. Jarvis, Brain Lang. 124, 96–116 (2013). to a limit. Most of the additional relevant studies can be found in the
7. A. R. Lameira, Neurosci. Biobehav. Rev. 83, 429–439 (2017). papers cited. Funding: The author’s efforts are supported
production, and the anterior striatum mod-
ADAPTED BY KELLIE HOLOSKI/SCIENCE
8. F. Patterson, E. Linden, The Education of Koko (Henry Holt & by funds from the Howard Hughes Medical Institute, a NIH Director’s
ulates that variability (Figs. 3B and 4). Company, 1981). Transformative Research Award, and The Rockefeller University.
Predictions from songbirds extend to human 9. J. L. Russell, J. M. McIntyre, W. D. Hopkins, J. P. Taglialatela, Competing interests: The author declares no competing interests.
Brain Lang. 127, 520–525 (2013).
molecular and neurophysiological mechanisms.
10. K. E. Bouchard, N. Mesgarani, K. Johnson, E. F. Chang, Nature SUPPLEMENTARY MATERIALS
In songbirds, cells of the vocal learning pathway 495, 327–332 (2013).
science.sciencemag.org/content/366/6461/50/suppl/DC1
show diverse activity-dependent patterns of up- 11. T. Riede, F. Goller, Brain Lang. 115, 69–80 (2010).
Supplementary Text
or down-regulation of hundreds to thousands 12. S. Herculano-Houzel, Front. Hum. Neurosci. 3, 31 (2009).
Figs. S1 to S7
13. W. T. Fitch, Annu. Rev. Linguist. 4, 255–279 (2018).
of genes in several temporal waves after singing References (41–60)
14. S. M. Garcia et al., Curr. Biol. 27, 2677–2683.e3 (2017).
that define function of different cell types (36). 15. A. R. Pfenning et al., Science 346, 1256846 (2014). 10.1126/science.aax0287

54 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

REVIEW generalizes to other domains of cognition, such
as planning and music.
The neurobiology of language beyond The classical view and its shortcomings
single-word processing For a long time, insights into the neurobiolog-
ical basis of language were derived from the
Peter Hagoort* views of neurologists in the late 19th and early
20th century [such as Broca, Wernicke, and
In this Review, I propose a multiple-network view for the neurobiological basis of distinctly human language Lichtheim (8)], as interpreted and extended
skills. A much more complex picture of interacting brain areas emerges than in the classical neurobiological in (9). According to this classic model, the
model of language. This is because using language is more than single-word processing, and much goes human language faculty was situated in the
on beyond the information given in the acoustic or orthographic tokens that enter primary sensory cortices. left perisylvian cortex, with a strict division of
This requires the involvement of multiple networks with functionally nonoverlapping contributions. labor between the frontal and temporal
regions. Wernicke’s area in the left temporal

T
cortex was assumed to subserve the compre-
he capacity for language is a central fea- arithmetic operations. This suggests that the hension of speech, whereas Broca’s area in the
ture of the human condition. It allows us infrastructure of the human brain provides the left inferior frontal cortex (LIFC) was claimed
to communicate with our fellow citizens, child with a certain language-readiness. One to support language production. The arcuate
to accumulate knowledge, to create cul- functional feature that has been argued to fasciculus connected these two areas. Although
tural practices, and to support our thought enable human singularity in this regard is our still influential, this model proved to have
processes. Language is a complex biocultural ability to infer tree structures from sequential severe limitations and to be largely wrong
hybrid. To understand its intricate organiza- data (4–6). Tree structures, as a representa- (10). These are the key issues: (i) Broca’s and
tion and neurobiological underpinnings, we tional format introduced by Wilhelm Wundt Wernicke’s area are ill-defined and do not form
have to decompose language skills into the (7), are intricately linked to the notion of hier- natural neuroanatomical elements. Moreover,
basic building blocks and core operations. archy. This is exemplified in, among other they are further parcellated into multiple
Basic building blocks include the knowledge things, the morphological make-up of words areas with different cytoarchitectonic profiles
that has been acquired during development and the hierarchical interpretation of phrases and receptorarchitectonic fingerprints (11–13).
about the sound patterns of the one or more (Fig. 1). This propensity to compute hierarchi- (ii) Functional magnetic resonance imaging
languages a speaker commands, the meaning cal structures is not limited to language but (fMRI) and lesion studies have shown that
of its lexical items, their syntactic features the language-relevant cortex is much more
(such as noun, verb, and grammatical gender), A N extended than assumed, including large parts
the orthographic patterns (in reading), or the of the temporal cortex, part of the parietal cor-
signs in the languages of the deaf. Next to tex, and areas in the LIFC beyond Brodmann
A
these elementary linguistic units (ELUs), there areas 44 and 45. Moreover, language is less
are elementary linguistic operations (ELOs) strictly left-lateralized than once thought (Fig. 2).
that enable the retrieval of ELUs from mem- N A N (iii) Both frontal and temporal regions are
ory (for example, in word recognition), or the involved in language comprehension as well
generation of larger structures from these ele- care less ness as in language production (14, 15). (iv) The
mentary building blocks [as in morphological connectivity of the language-relevant cortex is
(de)composition in compounding or verb in- B Point to the (second [green ball]) much more extended than the classical model
flection and as in the construction of sentence assumed and certainly not restricted to the
level meaning]. In addition, the proposition arcuate fasciculus (16). And (v) the cerebellum
created by combining ELUs and ELOs has to and subcortical structures such as the thalamus
be linked to the actual or imagined situation C Click on the (second [blue ball]) and basal ganglia play an important role as well
in which it is embedded in order to establish (17), especially for the fine-tuning of timing and
its truth value (1, 2). For example, if I say sequencing in speaking. It is relevant to real-
“The editor of the journal loved the paper,” I Fig. 1. Tree-like structures that are characteristic ize that the classic view was mainly based on
have produced an impeccable sentence, but for word formation and for the interpretation single-word processing. The idea that language
one can only know whether what I said is of phrases and sentences. (A) The word has combinatorial machinery beyond single
true or false if the nouns “editor,” “journal,” “carelessness” is made up of the morphemes “care” words was lacking. This might in part explain
and “paper” can be linked to specific tokens. (N, noun), “less” (A, marker for adjective), and why the classic model substantially underrep-
As we will see below, all this is only part of the “ness” (N, marker for noun). The final morpheme resents the brain regions and fiber tracts that
complex story of language. determines its status as a noun. (B) The phrase are important for language.
Although evolutionary precursors of human “the second green ball” is interpreted by 30 out of The ability to combine words in often new
language can be identified (3), in its full capac- 31 participants as referring to the third circle and ways is a hallmark of human language. This
ity it is distinctly human. Despite its com- not the second. This is the result of a hierarchical combinatoriality is realized by dynamic inter-
plexity, most children acquire the core of this interpretation of the phrase as referring to the second of actions between Broca’s area and the adjacent
capacity in the first few years of life, without the green balls. The representation in brackets is formally cortex in the LIFC with areas in the temporal
formal instruction and well before they have equivalent to a tree structure. Participants saw the and parietal cortex. The interplay between
learned to lace their shoes or perform simple instructions without brackets. (C) In the case of an array these areas guarantees that lexical informa-
with balls and triangles, the phrase “the second blue ball” tion retrieved from memory is unified into
Max Planck Institute for Psycholinguistics, Donders Institute is again interpreted as the second of the blue balls (the coherent multiword sequences with an over-
for Brain, Cognition and Behaviour, Radboud University,
Nijmegen, Netherlands. fifth shape) instead of as the second blue item, which is a arching syntactic structure and semantic
*Corresponding author. Email: peter.hagoort@donders.ru.nl ball (the third shape) by about 99% of the participants. interpretation (18).

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 55

 L A N G UA G E AN D TH E B RA I N

Neuroplasticity architecture of the human brain on the other is dynamic interaction with temporoparietal
An additional insight is that language-relevant still far from fully understood. The degree of areas plays an important role in unifying the
cortex is somewhat variable. Recent studies individual variation in the neurobiological in- different sources of information that deter-
have shown that under certain conditions, frastructure for language is also still largely mine the interpretation of an utterance (29).
cortex outside the classical perisylvian areas unchartered territory (24). Although there is no firm evidence that pre-
can be involved. Language processing occurs diction is necessary for comprehension, it has
in the occipital cortex of congenitally blind The immediacy principle become clear that prediction contributes to
individuals (19, 20). Although the exact com- A much clearer picture has emerged for key language processing and might be relevant to
putational contributions still need to be deter- features of the processing dynamics of lan- meet the demand for speed (30–32). Very likely,
mined, there is evidence that recruitment of guage, partly as a consequence of the high lexical, semantic, and syntactic cues conspire to
the visual cortex is related to behavioral per- temporal resolution of neurophysiological predict characteristics of the next anticipated
formance in a verbal memory task (21). More- measurements. Language processing occurs word, including its syntactic and semantic make-
over, the visual cortex of congenitally blind at an amazing speed. We easily produce and up. A mismatch between contextual prediction
individuals responds to sentences with syn- understand two to five words per second. The and the output of bottom-up analysis results in
tactic movement but not to the difficulty of consequence of the need for speed is that in an immediate brain response that recruits addi-
math equations (22). This result supports the both production and comprehension, the re- tional processing resources for the sake of sal-
hypothesis that this area can contribute to trieval of the linguistic building blocks and vaging the on-line interpretation process (18).
the computation of sentence-level syntactic the combinatorial operations happen incre-
structure. Overall, these findings suggest that mentally. Furthermore, language processing Information structure
the cyto-architectonic constraints for specifica- is characterized by the “immediacy principle” The speed of language processing is also helped
tions of cognitive function leaves certain degrees (25). In comprehension, linguistic and extra- by the fact that utterances are most often part
of freedom. If input patterns to a particular linguistic information is used immediately of a longer interaction between speaker and
brain area change, this area might be recruited upon becoming available. That is, knowledge listener. In such an interaction scheme, some
for quite different functionality. On the basis about the context and the world, concomitant information is already shared between the
of these and similar findings, Bedny concludes information from other modalities (such as conversational partners. This is often referred
that “human cortices are cognitively pluri- co-speech gestures) (26, 27), and knowledge to as “topic.” The new information is what is
potent, that is, capable of assuming a wide about the speaker (28) are brought to bear added and, hence, the key component of the
range of cognitive functions. Specialization is immediately on the same fast-acting brain message. This is usually referred to as the
driven by input during development, which is system that combines the meanings of indi- “comment.” Information structure refers to
constrained by connectivity and experience” vidual words. In other words, all available the way in which the topic and comment are
(23). The mapping relation between elemen- relevant information will be used without packaged in the organization of a sentence (1).
tary building blocks and elementary operations delay to codetermine the interpretation of the Speakers often take care to mark the focus
for language on the one hand and the neural speaker’s message. In all cases, the LIFC in constituent that contains the new information.
The way this is done differs between languages.
A There is no linguistic universal for signaling
The language-relevant connectome
information structure. In some languages, syn-
tactic locations are used for marking the focus
Arcuate fasc. (long)
constituent; other languages use focus-marking
Arcuate fasc. (ant)
particles or prosodic features such as phras-
Arcuate fasc. (post) ing and accentuation. For example, in English
Inf. frontal-occipital fasc. question-answer pairs, the new or relevant in-
frontal aslant tract formation in the answer will typically be pitch
inf. logitudinal fasc. accented. After a question such as “What did
uncinate fasc. Mary buy at the market?” the answer might be
“Mary bought VEGETABLES” (accented word
in capitals). In this case, “vegetables” is the focus
constituent, providing the new information.
Information structure was found to mod-
B Sentences vs. low level baseline ulate language-relevant activations in fMRI
(33, 34) and in electrophysiological readouts
(35–37). For example, unexpected semantic and
illegal syntactic continuations resulted in size-
able N400 and P600 effects, respectively, if
they were part of the focus constituent (35, 36).
These effects were, however, strongly reduced
or even absent if the continuations were part of
the nonfocus constituent (Fig. 3). New infor-
mation marked by pitch accent activated the
domain-general attention network (33). To gain-
0.0 13.1
modulate the depth of language processing,
t
linguistic devices, such as pitch accent, that
Fig. 2. Components of the neural infrastructure for language. (A) Fiber tracts connecting language- mark the relevant information in an expres-
relevant brain regions (16). (B) Common activations for listening to sentences and sentence reading in an sion (such as the comment) seem to trigger the
fMRI study with 204 participants, compared with a low-level baseline (59). attention network into operation.

56 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 This account ties in with the A B constituent that is marked as
idea that for listeners, a full re- 2 comment or new information.
construction of the linguistic in- In addition, integrating the ut-
put is in many cases not achieved, terance content into a situation
but for most purposes, a super- model that spans multiple con-
ficial and incomplete analysis is nected utterances is required.
µV
good enough (38). According to This process appears to involve
this “good-enough” processing the right inferior frontal gyrus
account, some semantic and and the right angular gyrus (56).
grammatical details might be Last, in order to extract the in-
ignored. More recently, Ferre- -2 tended message from the coded
ira and Lowder (39) have made meaning provided by the lin-
the connection between predic- Fig. 3. Topographic distribution of N400 effects triggered by a semantically guistic utterance, the listener
tion, information structure, and unexpected compared with an expected word. (A) In the focus constituent. has to integrate linguistic and
good-enough language process- (B) The same words in a nonfocus constituent (35). nonlinguistic information. She
ing. They propose that topic in- also has to draw the required
formation is processed in a good-enough man- could be interpreted as an indirect request pragmatic inferences, which requires a contri-
ner, whereas the listener’s prediction effort is for action. One of their items combined the bution of areas crucial for mentalizing.
devoted to the comment (the new or focused utterance “It is hot here” with a picture of a Next to core areas for retrieving lexical in-
information) because “all that is needed for door. Participants interpreted this as a request formation from memory and the unification
comprehension and interpretation is an id- to open the door. However, the same utterance of the lexical building blocks in producing and
entification of the comment—the topic being combined with the picture of a desert was understanding multiword utterances, other
already available” (1). interpreted as a statement. Sentences in the brain networks are needed to realize language-
indirect-request condition activated the ToM driven interactions to their full extent. Mul-
Pragmatic inferencing network much more strongly than the same tiple brain network contributions will not be
Although areas in the perisylvian cortex, espe- sentences without the possibility to interpret language-specific but shared with other cog-
cially in the left hemisphere, are important for them as an indirect request. The recognition of nitive functions. ELUs will almost certainly be
encoding and decoding propositional content, a speech act induced by an utterance in com- domain-specific (57), but some key aspects of
this is not the whole story. The meaning of an bination with its context seems to require the the ELOs (such as unification) might well be
utterance is often strongly dependent on con- contribution of the mentalizing machinery shared with other domains, such as music and
textual information that is not actually coded instantiated in the ToM network. Further arithmetic. A much more complex picture of
in what is said (40–42). What is coded in words insights will depend on a more precise interacting brain areas emerges than in the
and linguistic constructions underdetermines account than is currently available of the com- classic model. The reason is that understanding
what is meant or intended. Extracting the putational contributions of the different nodes language is more than single-word processing,
intended meaning requires that inferences in the ToM network (53). and much goes on beyond the information giv-
are made on the basis of assumptions about en in the acoustic or orthographic tokens that
the beliefs and intentions of the interacting The multiple network view enter primary sensory cortices. This multiple-
agents and about a common understanding of One approach to the neurobiology of language brain-network view does not take away the
appropriate language use. For example, in the is to start from identifying the essence of lan- need for further detailed specifications of the
right context the utterance “It is hot here” will guage. Once the few core features of language computational contributions of the different
not only be interpreted as a statement about that make it stand out in the animal kingdom brain areas involved in language processing.
the state of affairs but first and foremost as an have been established, the question can be
implicit request to do something about it (such addressed regarding which aspects of human REFERENCES AND NOTES

as open the window or turn down the heating). brain organization enable the essence of lan- 1. P. A. M. Seuren, Language in Cognition: Language From Within
(Oxford Univ. Press, 2009), vol. 1.
The socially binding and communicative roles guage. If, for example, recursion is key (54), one 2. P. A. M. Seuren, Saussure and Sechehaye: Myth and Genius. A Study in
of language depend critically on the capacity might identify Broca’s area as the neural equi- the History of Linguistics and the Foundations of Language (Brill, 2018).
to make the right pragmatic inferences (1, 43). valent of the push-down stack subserving this 3. S. E. Fisher, Curr. Biol. 29, R65–R67 (2019).
4. A. E. Martin, Front. Psychol. 7, 120 (2016).
So far, relatively few studies have inves- distinctly human computational capacity (5, 55). 5. W. T. Fitch, Phys. Life Rev. 11, 329–364 (2014).
tigated the neural infrastructure for effective The account that I have sketched here 6. S. Dehaene, F. Meyniel, C. Wacongne, L. Wang, C. Pallier,
communication beyond the core linguistic ma- takes a different, nonessentialistic stance. Neuron 88, 2–19 (2015).
7. W. Wundt, Logik (Enke, 1880), vol. II.
chinery for word retrieval, sentential syntax, It is based on the conviction that accounting 8. W. J. M. Levelt, A History of Psycholinguistics: The Pre-Chomskyan
and semantics. Nevertheless, the picture that for the full picture of human language skills is Era. (Oxford Univ. Press, 2013).
emerges is fairly consistent (44–49). Prag- not helped by a distinction between essen- 9. N. Geschwind, Science 170, 940–944 (1970).
10. P. Tremblay, A. S. Dick, Brain Lang. 162, 60–71 (2016).
matic inferences depend on the contribution tial and nonessential aspects of speech and 11. K. Amunts, K. Zilles, Trends Cogn. Sci. 16, 418–426 (2012).
of core areas in the theory of mind (ToM) net- language. In a first step, we need to decom- 12. K. Amunts, K. Zilles, Neuron 88, 1086–1107 (2015).
work. These include the right temporoparietal pose complex language skills such as speak- 13. K. Zilles, K. Amunts, Curr. Opin. Behav. Sci. 21, 93–105 (2018).
14. L. Menenti, S. M. E. Gierhan, K. Segaert, P. Hagoort,
junction and medial prefrontal cortex, areas ing and listening into ELUs and ELOs as the Psychol. Sci. 22, 1173–1182 (2011).
typically involved in tasks that require mental key building block for encoding and decoding 15. K. Segaert, L. Menenti, K. Weber, K. M. Petersson, P. Hagoort,
Cereb. Cortex 22, 1662–1670 (2012).
state reasoning—that is, thinking about other propositional content. These are mainly sup-
16. K. Amunts, M. Catani, in The Cognitive Neurosciences,
people’s beliefs, emotions, and desires (50–52). ported by the LIFC and substantial parts of the M. S. Gazzaniga, G. R. Mangun, Eds. (MIT Press, 2015), pp. 619–628.
An example is a study in which the authors temporal cortex and parietal cortex, with a left 17. H. Barbas, M. A. García-Cabezas, B. Zikopoulos, Brain Lang.
presented participants with sentences in the hemisphere bias. Interaction with the attentional 126, 49–61 (2013).
18. P. Hagoort, Neurosci. Biobehav. Rev. 81 (Pt B), 194–204 (2017).
presence of a picture (48). In one condition, control–multiple demand system is needed 19. M. Bedny, A. Pascual-Leone, D. Dodell-Feder, E. Fedorenko,
the sentence in combination with the picture to up-regulate the processing of the utterance R. Saxe, Proc. Natl. Acad. Sci. U.S.A. 108, 4429–4434 (2011).

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 57

 L A N G UA G E AN D TH E B RA I N

20. M. Bedny, M. MacSweeney, in Human Language: From Genes REVIEW

and Brains to Behavior, P. Hagoort, Ed. (MIT Press, 2019).
21. A. Amedi, N. Raz, P. Pianka, R. Malach, E. Zohary, Nat.
Neurosci. 6, 758–766 (2003). From speech and talkers to the social world:
22. C. Lane, S. Kanjlia, A. Omaki, M. Bedny, J. Neurosci. 35,
12859–12868 (2015).
23. M. Bedny, Trends Cogn. Sci. 21, 637–648 (2017).
The neural processing of human spoken language
24. K. Mahowald, E. Fedorenko, Neuroimage 139, 74–93 (2016).
25. P. Hagoort, J. van Berkum, Philos. Trans. R. Soc. Lond. B Biol. Sophie K. Scott
Sci. 362, 801–811 (2007).
26. L. Drijvers, A. Özyürek, Brain Lang. 177-178, 7–17 (2018).
27. A. Ozyürek, R. M. Willems, S. Kita, P. Hagoort, J. Cogn.
Human speech perception is a paradigm example of the complexity of human linguistic processing;
Neurosci. 19, 605–616 (2007). however, it is also the dominant way of expressing vocal identity and is critically important for social
28. J. J. A. Van Berkum, D. van den Brink, C. M. Tesink, M. Kos, interactions. Here, I review the ways that the speech, the talker, and the social nature of speech interact
P. Hagoort, J. Cogn. Neurosci. 20, 580–591 (2008).
and how this may be computed in the human brain, using models and approaches from nonhuman
29. P. Hagoort, Curr. Opin. Neurobiol. 28, 136–141 (2014).
30. T. Brothers, T. Y. Swaab, M. J. Traxler, Cognition 136, 135–149 (2015). primate studies. I explore the extent to which domain-general approaches may be able to account for
31. G. R. Kuperberg, T. F. Jaeger, Lang. Cogn. Neurosci. 31, 32–59 (2016). some of these neural findings. Finally, I address the importance of extending these findings into a better
32. J. Rommers, A. S. Meyer, P. Praamstra, F. Huettig,
understanding of the social use of speech in conversations.
Neuropsychologia 51, 437–447 (2013).
33. L. B. Kristensen, L. Wang, K. M. Petersson, P. Hagoort,

S
Cereb. Cortex 23, 1836–1848 (2013).
34. T. M. van Leeuwen et al., Neuropsychologia 58, 64–74 (2014).
35. L. Wang, M. Bastiaansen, Y. Yang, P. Hagoort,
peech is often viewed as the auditory tions, from brief, bursting releases to lengthy,
Neuropsychologia 49, 813–820 (2011). form of a more abstract linguistic system noisy segments and from nasal sounds to
36. L. Wang, M. Bastiaansen, Y. Yang, P. Hagoort, PLOS ONE 7, and from a neuropsychological perspec- sustained vowels (8). Although all speech de-
e47917 (2012).
37. D. La Rocca, V. Masia, E. Maiorana, E. Lombardi Vallauri,
tive, the processing of this spoken lan- pends on detailed spectrotemporal processing,
P. Campisi, Neurocomputing 199, 1–15 (2016). guage has been associated with the left the precise acoustic cues that have linguistic
38. F. Ferreira, N. D. Patson, Lang. Linguist. Compass 1, 71–83 (2007). posterior temporal lobe (e.g., 1, 2). However, relevance vary somewhat from language to
39. F. Ferreira, M. W. Lowder, in Psychology of Learning and
Motivation, Vol 65, B. H. Ross, Ed. (Elsevier Academic Press
this view has been challenged, first, by models language. When confronted with a phoneme
Inc, San Diego, 2016), vol. 65, pp. 217-247. of auditory processing from the nonhuman or an acoustic cue that is not used in our own
40. S. C. Levinson, Pragmatics (Cambridge Univ. Press, 1983). primate literature arguing for multiple ana- language, we can struggle to hear this accu-
41. D. Sperber, D. Wilson, Behav. Brain Sci. 10, 697–710 (1987).
42. P. Hagoort, S. C. Levinson, in The Cognitive Neurosciences,
tomical and functional streams of processes rately and to use this linguistically (9). This
M. S. Gazzaniga, G. R. Mangun, Eds. (MIT Press, 2015). in speech and sound percep- problem comes about because
43. During a lecture, the famous physicist Paul Dirac was told by tion (2); second, by studies of of the development of pho-
someone in the audience, “I don’t understand that formula.”
Dirac continued his lecture without addressing the issue.
The chairman then interrupted and asked Dirac whether
the processing of the nonlin-
guistic information encoded
“…the neural netic processing skills in the
acquisition of speech percep-
he could reply to the question. Dirac was astonished, replying:
“Question? What question? My colleague has made an
in the speaking voice (3); and systems that tion (10). The developing brain
third, by the suggestion that learns to prioritize the audi-
assertion.” As on other occasions, Dirac failed to grasp the
intended message behind the assertion (58). we need to bear in mind the process speech tory cues that are linguistically
44. J. Bašnáková, K. Weber, K. M. Petersson, J. van Berkum,
P. Hagoort, Cereb. Cortex 24, 2572–2578 (2014).
critical social importance of
speech (4). From the first per-
develop these relevant and to downgrade the
importance of speech sounds
45. J. Bašnáková, J. van Berkum, K. Weber, P. Hagoort,
Neuropsychologia 76, 79–91 (2015). spective, the involvement of the
dorsolateral temporal lobes in
skills in social, that are less important for dis-
tinguishing the meaning of
46. A. M. Paunov, I. A. Blank, E. Fedorenko, J. Neurophysiol. 121,
1244–1265 (2019).
47. N. Spotorno, E. Koun, J. Prado, J. B. Van Der Henst,
auditory speech processing may conversational words. Indeed, the processing
rest upon perceptual networks of words in a holistic fashion
I. A. Noveck, Neuroimage 63, 25–39 (2012).
48. M. J. van Ackeren, D. Casasanto, H. Bekkering, P. Hagoort, consisting of multiple process- settings, and speech may come before more fine-
S. A. Rueschemeyer, J. Cogn. Neurosci. 24, 2237–2247 (2012).
49. F. Vanlangendonck, R. M. Willems, P. Hagoort, PLOS ONE 13,
ing pathways, comparable to
the distinct processing streams
is an overwhelmingly grained phonological skills in
development (11).
e0202943 (2018).
50. D. M. Amodio, C. D. Frith, Nat. Rev. Neurosci. 7, 268–277 (2006).
seen in the visual system. From social behavior.” However, human speech per-
51. J. Koster-Hale et al., Neuroimage 161, 9–18 (2017). the second perspective, if there ception is complex and multi-
52. R. Saxe, in Handbook of Theory of Mind, A. Leslie, T. German, is speech, there is always a stable; in the brain of a listener,
Eds. (Psychology Press, Philadelphia, PA, 2010).
talking voice (5) and there may be signifi- no one acoustic cue determines the intelligi-
53. E. Gibson, L. Bergen, S. T. Piantadosi, Proc. Natl. Acad. Sci. U.S.A.
110, 8051–8056 (2013). cant hemispheric asymmetries involved in bility of speech, and listeners will listen flexibly,
54. M. D. Hauser, N. Chomsky, W. T. Fitch, Science 298, 1569–1579 the ways that left and right temporal fields making use of relevant acoustic cues when
(2002). process these different kinds of information they are of utility (12, 13). This flexibility is
55. A. D. Friederici, Language in Our Brain. The Origins of a
Uniquely Human Capacity. (MIT Press, 2017).
(6, 7). From the third perspective, we can study necessary because we are continually con-
56. L. Menenti, K. M. Petersson, R. Scheeringa, P. Hagoort, speech in isolation, but the neural systems that fronted with new voices and accents, as well
J. Cogn. Neurosci. 21, 2358–2368 (2009). process speech develop these skills in social, as complex acoustic listening environments.
57. E. Fedorenko, M. K. Behr, N. Kanwisher, Proc. Natl. Acad. conversational settings, and speech is an over- Auditory cortical fields therefore need to adapt
Sci. U.S.A. 108, 16428–16433 (2011).
58. C. Rovelli, Reality Is Not What It Seems: The Journey to
whelmingly social behavior. in a transient way to both speech and the
Quantum Gravity (Penguin Books, 2017). listening conditions in which speech is heard.
59. J. M. Schoffelen et al., Sci. Data 6, 17 (2019). Speech: From sound to meaning Furthermore, although speech is made up of
ACKN OW LEDG MEN TS
The human speech signal is almost bewilder- sequences of speech sounds, the speech sounds
I am grateful to A. Martin, A. Meyer, A. Özyürek, W. Levelt, and
ingly complex in terms of acoustics. Speech themselves are affected by the sounds around
P. Seuren for their comments on an earlier version of this paper. sounds are made by a variety of different ac- them. Speech sounds are different depending
Writing the paper was made possible by the Nederlandse upon their position in a syllable—for most British-
Organisatie voor Wetenschappelijk Onderzoek grant Language
in Interaction, grant 024.001.006.
English accents, the “l” at the start of “leaf” is
Institute of Cognitive Neuroscience, University College
London, London, UK. very different from the “l” at the end of “bill.”
10.1126/science.aax0289 Email: sophie.scott@ucl.ac.uk Listeners are very sensitive to this contextual

58 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 information and use it to help them decode LEFT RIGHT
speech (14), and the recognition of spoken
words is based not on a sequence of abstract
phonemic categories, but more on a form-
based auditory representation (15, 16).
Speech is the auditory form of language, and
much of the aim of research has been to study
the deeper structures contained with the com- Dorsolateral
solateral
putational properties of language, with the mporal lobe
temporal
surface structure of the speech being some-
what dissociable from this higher-order in- Linguistic
guistic processing Voice processing
formation. However, the acoustic form of onation Emotion
Intonation Phonetic Emotion
speech is rich in terms of information, from ntity
Identity Lexical Syntactic
phonemic, talker, emotion, and effects through
to higher-order information such as grammat-
ical structure.
Speech is also rarely heard in silence; we Spoken
commonly listen to speech in noisy environ- comprehension
ments. Adults struggle most to understand Linguistic expertise
speech in the context of broadband noisy sounds improves talker
(such as an air-conditioning unit or a fan) (17). discrimination
However, the more “information” there is in
Talker familiarity
competing speech, the more likely it is to be improves speech
processed by the adult listener, especially if perception
it starts to compete with the attended speech
in terms of its semantic content (17). Individual talker
adaptation
Although it is conceptually simple to sep-
arate speech production and speech percep-
tion mechanisms, perception networks are
Fig. 1. Schematic representation of the left and right dorsolateral temporal lobes. Shown are the left
critical to the production of speech. Even
(yellow) and right (green) dorsolateral temporal lobes and the different candidate interactions that may occur
moderate patterns of hearing loss in devel-
between linguistic and nonlinguistic aspects of voice and language recognition processes on the basis of
opment will have an effect on the develop-
behavioral and noninvasive stimulation studies.
ment of speech skills (18); by contrast, the
absence of the ability to speak aloud is not
detrimental to the ability to learn to under- maintain the rostral-caudal structure of the studies of speech perception revealed a dom-
stand speech (19). Speech production skills, core fields, and this connectivity is preserved inant response to speech that ran forward
therefore, like many controlled motor skills, in the projections to frontal fields. There are from PAC toward the temporal pole, into the
are dependent on perceptual processing of functional differences seen within these anterior superior temporal gyrus (29–33). These
one’s own actions (20). rostral-caudal fields in nonhuman primates, left anterior STS fields are sensitive to intelligi-
One final complication is that human speech with rostral fields sensitive to different kinds ble speech no matter what the talker sounds
expresses the phenomenal complexity of hu- of conspecific vocalizations and caudal fields like (29, 31, 33), and rostral STG/STS fields show
man language, but it is also a primarily social sensitive to the spatial locations of the sounds selective responses to phonetic, syntactic, and
behavior: we very rarely speak aloud on our (26) and to somatosensory stimulation (27). semantic information (34–37). There is a highly
own, and human speech is universally the dom- Thus, perceptual processing in primates is sensitive pattern of the processing of phonemic
inant mode for social engagement (21). Con- not one unitary phenomenon, but rather one sequences with the STG, which displays the
versations are the contexts in which we learn that rests upon different perceptual networks, flexibility and adaptability that these networks
to speak (22) and are the settings for most social which can be recruited differentially depend- require to navigate a world of talkers with dif-
interactions. The study of speech thus lies at ing on the task. In the real world, actions will ferent accents and different listening environ-
the crux of auditory processing, language pro- depend on these networks working together. ments (38). This flexibility must also apply to
cessing, social processing, emotion, identity, and Quite obviously, nonhuman primates do the competing auditory environment; studies
music. How have developments in neuroscience not speak; they can make use of complex in which participants listen to speech while
helped us to model this immense complexity? vocalizations and may have similar laryngeal ignoring competing auditory maskers show
flexibility (28), but their vocalizations do not responses within rostral temporal fields that
The dorsolateral temporal lobes verge on the complexity of human speech are highly sensitive to the masking sounds
ILLUSTRATION CREDIT: V. ALTOUNIAN/SCIENCE

Speech is an auditory signal and it is thus pro- and language. Is there any value, therefore, (39–41). Unattended competing talkers’ voices
cessed in the ascending auditory pathway, to in trying to map from these nonhuman pri- are not discarded at the auditory periphery,
the primary auditory cortex (PAC) and sur- mate studies onto models of human speech but rather are processed within the same
rounding auditory association cortex, and processing? network as the attended speech. This permits
extending out laterally and down into the First, the recruitment of rostral auditory the processing of unattended speech for some
superior temporal gyrus (STG). In nonhuman fields in speech perception has been widely potentially relevant information, but also means
primates, the PAC is formed of three core reported: unlike the left posterior superior that it is competing for resources with the
fields, arranged in a caudal-rostral orientation temporal sulcus fields, which had been em- attended speech (17, 42). This must mean
(23, 24), and these then project to surrounding phasized as the core region associated with that rostral auditory fields are capable of
belt and parabelt fields (25). These projections Wernicke’s aphasia, early functional imaging the creation and representation of patterns

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 59

 L A N G UA G E AN D TH E B RA I N

of speech that are being heard at the same times of speaking, the detection of and compen- ceptual processing of phonological informa-
as the attended speech. Given the complexity of sation for these alterations are associated tion (60) and its syntactic influences (61)
auditory environments, this may simply be a with increased recruitment of caudal auditory but that does not mean that individual pho-
glimpse of the nature of the formation and fields (57). Aligning your voice with another nemes are being represented as discrete, ab-
maintenance of these rostral auditory repre- talker in unison speech (58) or deliberate stract items. STG is highly sensitive to the
sentations and our ability to switch our at- controlled production of a different vocal introduction of longer phonetic sequences,
tention between them (43). identity when speaking (i.e., trying to sound but less sensitive to the introduction of clus-
These recognition processes within rostral like someone else) also recruits these caudal ters of phonemes (62).
fields show some important hemispheric areas (59). The nature of phonemes in the STG seems
asymmetries. Right rostral temporal fields are to be more important in how they contribute
sensitive to voice-specific information in non- The nature of auditory representations to the phonemic “shape” of a sequence and
human primates (44) and are also highly sen- of speech that this form is represented at the syllable
sitive to natural pitch profiles (intonation) in The importance of phonemes in speech can level, rather than at the level of individual
speech (45, 46). Right rostral fields dominate lead to the assumption that they must form phonemes. Phonemes contribute to the se-
in the processing of talker identity in humans important representational cues in the percep- quence like facial features contribute to face:
(47), although this can also show bilateral re- tual processing of speech. However, this does their perceptual role contributes to the whole,
sponses. Humans rely strongly on pitch cues not mean that phonemes themselves, as dis- rather than to the assembled individual items.
to discriminate between talkers (48). These crete, abstract linguistic individual items, are Studies of the temporal sensitivity in STG
differences between linguistic and nonlinguistic encoded during speech comprehension before fields show a relatively slow sensitivity, peak-
processing of vocal information in the left and any lexical processing; indeed, one might well ing at time scales associated with words or
right rostral fields do not rest on basic auditory expect that the sequential dependencies be- syllables, rather than the much faster time
processing differences (6, 7), but rather seem to tween speech sounds, which are highly in- scales associated with individual phonemes (63)
reflect differences in the kind of information formative to listeners, would not be discarded (Fig. 2). Similarly, electrocorticogram (eCOG)
being processed. in the processes that lead to the comprehen- data show a sensitivity to the amplitude en-
Listeners use the accent of talkers to help sion of speech. There is evidence for early per- velope of speech sounds, which correlates
then interpret word meanings; e.g., broadly with syllable structure (64).
British-English listeners will be more Syllables are also good candidates
likely to interpret “bonnet” as mean- for linguistic universals in the orga-
ing “hat” if it is said in an American Perceptually relevant nization of spoken language (65), and
accent (49). Talkers who are familiar speech tracking at syllable structure is quite constrained.
are easier for us to understand, and the phrase level At its bare minimum, a single syllable
listeners adapt very quickly to talker- (0.6–1.3 Hz), seen in 5 1 can consist of one vowel, and the ways
2
specific quirks of speech production, the left precentral (1) that consonants can be added to the
and postcentral (2) 3
showing phoneme-specific adapta- onset and offset of the vowel varies
gyri, supramarginal
tions, but only to that speaker (50, 51). gyrus (3), and Heschl’s
across languages, although the sim-
We are also much more accurate at gyrus (not shown) with plest structure around the world is
discriminating talkers in a language the peak in the left Activation consonant-vowel or CV, not VC. Eng-
that we speak; in a language we do premotor cortex (5). area lish permits highly complex sylla-
not speak, we can struggle to tell bles with up to three consonants
talkers apart (52). These studies must before the vowel and four conso-
mean that speech and talker recog- nants after, e.g., “strengths,” which
nition brain networks may be dis- has a C1C2C3VC1C2C3C4 structure. In
tinct anatomically, but must interact Perceptually relevant Japanese, such sequences of clusters
quickly, continuously, and accurate- speech tracking of consonants would be illegal, and
ly (Fig. 1). For example, noninvasi- at the word level syllables have a much simpler CVC
vely altering function in left STG (1.8–3.0 Hz), seen in structure. It is clear that syllable
the left superior (1), 4
fields with transcranial direct cur- middle (2), and inferior
structure gives a frame that can be
rent stimulation disrupts adaptation temporal gyrus (3), roughly construed as corresponding
to the speech of talkers (53). supramarginal gyrus 1 to the onset or rhyme of the syllable,
By contrast, caudal auditory fields (4) and Heschl’s gyrus 2 and which might form the basic
are much less sensitive to the specific (not shown). The efect 3 (and language specific) context in
kind of speech and voice information peaked in the left which the phoneme sequences are
middle temporal gyrus.
being processed, and more sensitive represented. When presented with
to their sensorimotor associations. consonant clusters, Japanese listen-

ILLUSTRATION CREDIT: V. ALTOUNIAN/SCIENCE

Caudal auditory regions are reliably Fig. 2. Summary of some of the findings from Keitel et al. In Keitel et al. ers hear vowels between the conso-
activated when people move their (63), naturalistic speech was played to listeners who were scanned with nants, called epenthetic vowels, and
articulators to make a sound, or even magnetoencephalography (MEG). The authors then interrogated the cortical the perception of these epenthetic
if they mimic these actions silently activity based on the statistical properties of the speech from the original vowels is linked to experience-
(54–56). This seems to map onto a stimuli. They found that perceptually relevant speech tracking (based on dependent processing in the STG
key role for caudal auditory areas in mutual information, a measure of synchrony in MEG studies) was larger for (66). Perhaps part of the language-
the sensory guidance of speech and correct trials (relative to incorrect trials) in premotor and motor cortex for specific acoustic phonetic process-
voice production. When people change activity at the time scale of phrases (0.6 to 1.3 Hz) (top) and in the left ing that is seen within the STG may
their voice because they are coping temporal lobe for activity at the time scale of words (1.8 to 3 Hz) (bottom). form these syllable-level constraints
with altered perceptual consequences Activity is shown schematically within the dotted fields. upon speech perception.

60 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 Caudal “how” pathway scale that could be important for rhythmic
alignment in spoken language. Comparing
face-to-face conversation with face-to-face
repetition reveals greater whole-brain involve-
ment during conversation in frontal and tem-
poral areas, with an emphasis on the left
temporal pole, left temporoparietal junc-
Rostral “what” pathway “Onset” responses tion, and bilateral medial prefrontal cortex
(74). This suggests that face-to-face conversa-
tion indeed recruits different elements of the
speech perception network, including both
rostral and caudal networks, alongside other
networks recruited in face processing.
“Sustained” responses It is going to be critical to understand the
ways that the human brain is engaged in
Fig. 3. Schematic representation of the difference between slow, sustained and fast, transient neural spoken communication and the social and
responses found in rostral (yellow) and caudal (blue) pathways, respectively. These differences were emotional impact of this. Speech is more than
established by categorizing the time course of eCOG responses to all stimuli presented to participants across language and it seems that this may be more
a range of studies (68). This difference in temporal response properties is one of several new findings important to the healthy function of humans
about rostral-caudal auditory differences in responses to sounds that have been incorporated into a new and their brains than previously thought. The
domain-general theory of auditory processing (67). Shown is the difference between a yellow rostral “what” goals for further neuroscientific investigations
pathway and a blue caudal “how” pathway and includes exemplar profiles. of speech perception should start to address
the ways that we engage with speech, from
sound to social meaning and social contact.
Are we really looking at a Consistent with the suggestion of a more
speech-specific system? domain-general approach to auditory process- REFERENCES AND NOTES
To what extent will any of these networks be ing, there are distinct responses to the differ- 1. J. E. Bogen, G. M. Bogen, Ann. N. Y. Acad. Sci. 280, 834–843
specific to speech and language? The distinction ent classes of auditory objects and these lie (1976).
2. J. P. Rauschecker, S. K. Scott, Nat. Neurosci. 12, 718–724
between linguistic and nonlinguistic aspects of within rostral auditory fields (69). By contrast, (2009).
sound processing may not be fully independent. caudal fields are recruited when participants 3. R. Watson, M. Latinus, I. Charest, F. Crabbe, P. Belin, Cortex
The phonetic “tuning” into language-relevant listen to sounds that contain information 50, 125–136 (2014).
aspects of sounds that happens as we acquire a about the locations of actions and objects in 4. S. K. Scott, C. McGettigan, F. Eisner, Nat. Rev. Neurosci. 10,
295–302 (2009).
spoken language during development may also space and their proximity to the listener 5. S. K. Scott, C. McGettigan, in APA Handbook of Nonverbal
affect the perceptual processing of nonspeech (70–72). Communication, D. Matsumoto, H. Hwang, M. Frank, Eds.
sounds (9). In terms of neural activation, there (American Psychological Association, 2015), pp. 289–306.
Speech in a social world 6. C. McGettigan, S. K. Scott, Trends Cogn. Sci. 16, 269–276
may not be a complete functional dissociation
(2012).
of speech and nonspeech processing. Social engagement with speech and language 7. S. K. Scott, C. McGettigan, Brain Lang. 127, 36–45 (2013).
However, it is almost certainly not the case is seen in a further linguistic universal: spo- 8. P. Ladefoged, P. Maddieson, The Sounds of the World's
that any of these networks are limited to the ken conversation. Speech is the dominant Languages (Blackwell, 1996).
9. P. Iverson, A. Wagner, S. Rosen, J. Acoust. Soc. Am. 139,
processing of vocal information. If one limits mode for social interactions in almost all 1799–1809 (2016).
oneself to questions about speech and language, human cultures, apart from those for whom 10. J. F. Werker et al., Cognition 103, 147–162 (2007).
then one will only find results that relate to that signing is the dominant mode. As a field, we 11. M. M. Vihman, Br. J. Psychol. 108, 1–27 (2017).
12. G. A. Miller, Language and Communication (McGraw Hill,
question; however, this approach has not had have prioritized the acoustic, linguistic, and
1951).
great success in divining the computational computational aspects of speech processing 13. R. V. Shannon, F. G. Zeng, V. Kamath, J. Wygonski, M. Ekelid,
principles that might distinguish the rostral over its social relevance. Accounting for at Science 270, 303–304 (1995).
and caudal streams. We may need to reframe least some of the cortical activation seen in 14. S. Hawkins, J. Phonetics 31, 373–405 (2003).
15. J. E. Andruski, S. E. Blumstein, M. Burton, Cognition 52,
the rostral–caudal auditory distinction as a the perceptual processing of speech is that 163–187 (1994).
more general one that reflects rostral recog- outside of the laboratory, we encounter speech 16. B. McMurray, M. K. Tanenhaus, R. N. Aslin, Cognition 86,
nition processes, not one limited to linguistic in social interactions, interactions in which we B33–B42 (2002).
17. D. S. Brungart, B. D. Simpson, M. A. Ericson, K. R. Scott,
information (or, indeed, to vocal information) are commonly active participants (4, 73). A J. Acoust. Soc. Am. 110, 2527–2538 (2001).
(67). In parallel, caudal auditory networks may specific role for premotor fields in the tracking 18. K. Mogford, “Oral language acquisition in the prelinguistically
not be solely involved in coordinating speech- of the rhythm of speech was suggested as a deaf” in Language Development in Exceptional Circumstances,
D. V. M. Bishop and K. Mogford, Eds. (Churchill Livingstone,
related sensorimotor links. Rostral and caudal candidate way of aligning one’s own speech 1988), pp. 110–131.
auditory fields may be distinguished on the with other talkers in a conversation and 19. D. V. M. Bishop, in Language Development in Exceptional
ILLUSTRATION CREDIT: V. ALTOUNIAN/SCIENCE

basis of their temporal response profiles: facilitating the swift changes of talker turns Circumstances, D. V. M. Bishop and K. Mogford, Eds.
(Churchill Livingstone, 1988), pp. 220–238.
rostral fields show a slow, sustained response that characterize spoken conversations. A 20. A. A. Mattar, S. M. Nasir, M. Darainy, D. J. Ostry, Prog. Brain
to sounds, whereas caudal fields show a fast, recent study (63) looked at the sensitivity to Res. 191, 31–44 (2011).
transient response to the onsets of sounds time scales in spoken language and found 21. G. A. Miller, Review of Greenberg JL, Universals of Language,
(38, 68). The slower rostral pathway re- an important role for premotor cortex in the Contemporary Psychology 8, 417–418 (1963).
22. E. V. Clark, Lang. Learn. Dev. 14, 170–185 (2018).
sponses would be consistent with feedback processing of phrasal information in spoken 23. J. H. Kaas, T. A. Hackett, Proc. Natl. Acad. Sci. U.S.A. 97,
influences on recognition processes, whereas sequences. By contrast, STG fields were more 11793–11799 (2000).
the faster caudal responses would be consist- sensitive to the structure of syllable range se- 24. B. H. Scott et al., Cereb. Cortex 27, 809–840 (2017).
25. L. M. Romanski et al., Nat. Neurosci. 2, 1131–1136 (1999).
ent with the fast-online perceptual guidance quences (Fig. 2). This suggests that premotor 26. B. Tian, D. Reser, A. Durham, A. Kustov, J. P. Rauschecker,
of action (Fig. 3). cortex is sensitive to information at the kind of Science 292, 290–293 (2001).

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 61

 L A N G UA G E AN D TH E B RA I N

27. T. A. Hackett et al., J. Comp. Neurol. 502, 924–952 REVIEW

(2007).
28. W. T. Fitch, B. de Boer, N. Mathur, A. A. Ghazanfar, Sci. Adv. 2,
e1600723 (2016).
The neural basis of combinatory syntax
29. S. K. Scott, C. C. Blank, S. Rosen, R. J. S. Wise, Brain 123,
2400–2406 (2000).
30. C. McGettigan et al., J. Cogn. Neurosci. 24, 636–652
and semantics
(2012).
31. K. Okada et al., Cereb. Cortex 20, 2486–2495 (2010). Liina Pylkkänen1,2
32. S. Evans et al., Cereb. Cortex 24, 2350–2361 (2014).
33. S. Rosen, R. J. S. Wise, S. Chadha, E. J. Conway, S. K. Scott, Human language allows us to create an infinitude of ideas from a finite set of basic building blocks.
PLOS ONE 6, e24672 (2011).
34. Z. K. Agnew, C. McGettigan, S. K. Scott, J. Cogn. Neurosci. 23,
What is the neurobiology of this combinatory system? Research has begun to dissect the neural basis
4038–4047 (2011). of natural language syntax and semantics by analyzing the basics of meaning composition, such as
35. Z. K. Agnew, H. van de Koot, C. McGettigan, S. K. Scott, Lang. two-word phrases. This work has revealed a system of composition that involves rapidly peaking activity in
Cogn. Neurosci. 29, 1035–1045 (2014).
the left anterior temporal lobe and later engagement of the medial prefrontal cortex. Both brain regions show
36. L. Cohen, A. Jobert, D. Le Bihan, S. Dehaene, Neuroimage 23,
1256–1270 (2004). evidence of shared processing between comprehension and production, as well as between spoken and signed
37. A. D. Friederici, S. A. Kotz, S. K. Scott, J. Obleser, Hum. Brain language. Both appear to compute meaning, not syntactic structure. This Review discusses how language
Mapp. 31, 448–457 (2010).
builds meaning and lays out directions for future neurobiological research on the combinatory system.
38. H. G. Yi, M. K. Leonard, E. F. Chang, Neuron 102, 1096–1110
(2019).

W
39. S. Evans, C. McGettigan, Z. K. Agnew, S. Rosen, S. K. Scott,
J. Cogn. Neurosci. 28, 483–500 (2016). hen exposed to a familiar language, term memory traces corresponding to these
40. S. K. Scott, S. Rosen, L. Wickham, R. J. S. Wise, J. Acoust.
Soc. Am. 115, 813–821 (2004). our brains automatically compose the structures and is therefore able to evaluate
41. S. K. Scott, S. Rosen, C. P. Beaman, J. Davis, R. J. S. Wise, individual words together into larger incoming language against this knowledge.
J. Acoust. Soc. Am. 125, 1737–1743 (2009). meanings. Even without language in- This much is uncontroversial. But what are the
42. S. K. Scott, C. McGettigan, Hear. Res. 303, 58–66 (2013).
43. F. Schneider et al., Sci. Rep. 8, 17948 (2018).
put, our brains do something similar: online computations that serve to merge or
44. C. I. Petkov et al., Nat. Neurosci. 11, 367–374 (2008). We create new meanings in our thoughts and combine words during language processing,
45. J. S. Kyong et al., J. Cogn. Neurosci. 26, 1748–1763 (2014). even comprehend our own creations. This is such that the relationship between the com-
46. D. Sammler, M. H. Grosbras, A. Anwander, P. E. Bestelmeyer,
P. Belin, Curr. Biol. 25, 3079–3085 (2015).
the internal “chatter” that, for most humans, posed representations and our knowledge of
47. S. R. Mathias, K. von Kriegstein, Front. Biosci. (Schol. Ed.) 6, is hard to shut down. Although combining syntax can be evaluated? This Review focuses
92–109 (2014). meanings is instinctive and automatic, our on our current understanding of this question.
48. S. C. Creel, M. R. Bregman, Lang. Linguist. Compass 5,
190–204 (2011).
minds actually perform some rather complex
mental gymnastics while doing so. Consider Comprehension: Rapid concept composition
49. Z. G. Cai et al., Cognit. Psychol. 98, 73–101 (2017).
50. F. Eisner, J. M. McQueen, Percept. Psychophys. 67, 224–238 these seemingly simple sentences: in the left anterior temporal cortex
(2005).
Just as biologists prefer to study small ani-
51. F. Eisner, J. M. McQueen, J. Acoust. Soc. Am. 119, 1950–1953 Sally baked the black beans.
(2006). mals with fewer cells when trying to under-
Sally baked the beans black.
52. T. K. Perrachione, S. N. Del Tufo, J. D. Gabrieli, Science 333, stand living organisms mechanistically, it has
595 (2011). In the first sentence, black describes a prop- proven productive to start simple in the neuro-
53. J. Y. Choi, T. K. Perrachione, Brain Lang. 196, 104655
(2019). erty of the beans prior to the baking, a so- biology of meaning composition as well. When
54. R. J. S. Wise et al., Brain 124, 83–95 (2001). called modifier reading. In the second, the it comes to brain mechanisms of language, a
55. G. Hickok, B. Buchsbaum, C. Humphries, T. Muftuler, J. Cogn. blackness of the beans is caused by the baking, full sentence is like an elephant. A short, two-
Neurosci. 15, 673–682 (2003).
56. Z. K. Agnew, C. McGettigan, B. Banks, S. K. Scott, Neuroimage
a resultative reading. We can even make the word phrase is a more tractable representational
73, 191–199 (2013). meaning ambiguous by adding a modifier to unit, and thus research has begun to characterize
57. H. Takaso, F. Eisner, R. J. S. Wise, S. K. Scott, Journal of the adjective itself: the composition of these minimal phrases (1–10)
Speech Hearing and Language Research. 53, 226–236
(2010).
(Fig. 1). The goal is to functionally decompose a
Sally baked some beans black enough
58. K. M. Jasmin et al., J. Neurosci. 36, 4669–4680 (2016). perisylvian brain network implicated for the
to look like licorice.
59. C. McGettigan et al., J. Cogn. Neurosci. 25, 1875–1886 processing of full sentences (11).
(2013).
60. H. Blank, M. H. Davis, PLOS Biol. 14, e1002577 (2016).
In this sentence, Sally could either be baking beans When we understand language, dozens of
61. Y. Sun et al., Brain Lang. 149, 55–65 (2015). that have a licorice-like appearance (modifier processing stages are packed into a few hun-
62. C. McGettigan et al., J. Cogn. Neurosci. 23, 961–977 (2011). reading) or she could be baking beans until they dred milliseconds. For this reason, detailed
63. A. Keitel, J. Gross, C. Kayser, PLOS Biol. 16, e2004473
turned as black as licorice (resultative reading). time resolution is an important requirement
(2018).
64. Y. Oganian, E. G. Chang, BioRxiv 388280 [Preprint]. Our brains make these distinctions on the for our measurement device. This Review
9 August 2018. https://doi.org/10.1101/388280. basis of syntax. The example sentences illus- focuses on research that has used magneto-
65. L. M. Hyman, Phonology 28, 55–85 (2011). trate three structures allowed by the syntax encephalography (MEG), the only noninvasive
66. C. Jacquemot, C. Pallier, D. LeBihan, S. Dehaene, E. Dupoux,
J. Neurosci. 23, 9541–9546 (2003). of English: prenominal modification (adjective technique that offers both the millisecond
67. K. Jasmin, C. F. Lima, S. K. Scott, Nat. Rev. Neurosci. 20, before noun), resultative secondary predica- resolution required for characterization of
425–434 (2019). tion (adjective follows the noun and describes rapid language processing as well as reason-
68. L. S. Hamilton, E. Edwards, E. F. Chang, Curr. Biol. 28,
1860–1871.e4 (2018). a resultant state of the main “primary” pred- ably accurate spatial localization of the neural
69. S. Norman-Haignere, N. G. Kanwisher, J. H. McDermott, Neuron icate, the verb), and postnominal modification, currents generating the measured signals. Neu-
88, 1281–1296 (2015). which is semantically equivalent to prenomi- roscience techniques that track blood flow,
70. K. Fiehler, I. Schütz, T. Meller, L. Thaler, Multisens. Res. 28,
195–226 (2015).
nal modification but is only grammatical when such as functional magnetic resonance imag-
71. A. Callan, D. E. Callan, H. Ando, Neuroimage 66, 22–27 (2013). the modifier satisfies certain length and/or ing (fMRI) or positron emission tomography,
72. L. Ceravolo, S. Frühholz, D. Grandjean, Soc. Cogn. Affect. weight requirements. Your brain has long- have good spatial accuracy but are too slow
Neurosci. 11, 793–802 (2016). to characterize the time course of rapidly un-
73. S. Garrod, M. J. Pickering, Top. Cogn. Sci. 1, 292–304 (2009). 1
Departments of Linguistics and Psychology, New York folding language.
74. K. Jasmin et al., Brain 142, 808–822 (2019). University, New York, NY, USA. 2NYUAD Institute, New York
University Abu Dhabi, Abu Dhabi, United Arab Emirates. Although a two-word phrase is simpler than
10.1126/science.aax0288 Email: liina.pylkkanen@nyu.edu a sentence, its composition still likely engages

62 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 many different types of combinatory routines Reading phrases LATL
(Fig. 2). For example, to account for the syn- red boat (composition) vmPFC
tactic and semantic behavior of a simple phrase xtp boat
such as “black cat,” linguistics and cognitive cup, boat (list task) red boa
t

psychology hypothesize at least three types of xtp, boat

structures: (i) the syntax, in which the catego-
ries “noun” and “adjective” join to form a noun LATL LATL
phrase (12); (ii) the logical semantics, in which **
4 4 ns
the properties of blackness and catness inter-
sect to yield a representation of entities that

nAm
nAm
3 3
possess both properties (13); and (iii) the con-
ceptual structure, in which the features of the 2 2
two concepts combine (14). Because these rep-
resentations may be built simultaneously in 1 1
parallel, a problem for a mechanistic under- 0 100 200 300 400 500 184 - 255 ms
standing of composition is to determine whether vmPFC vmPFC
these possibly distinct representations disso-
ciate in neural activity. 4 4 *
ns
First, we need a characterization of what

nAm
3 3

nAm
happens in the brain when a person hears
or reads a word in a minimal combinatory
2 2
context. Most studies on minimal phrases
have used adjective–noun combinations. Re- 1
1
sults show that in English, this type of basic 0 100 200 300 400 500 331- 480 ms
composition increases activity in the left an-
terior temporal lobe (LATL) at 200 to 250 ms
LATL
after noun onset, as compared with the pro-
cessing of the same noun in a context in which vmPFC
it cannot combine with the preceding word
(1, 2). If the language has the reverse word
Producing phrases Response
order, noun before adjective, the same result
is seen with respect to the processing of the LATL
adjective (6). About 200 ms later, another acti- 15 15
vity increase often occurs in the ventromedial
prefrontal cortex (vmPFC) (Fig. 1). What aspects
Amplitude, nAm

Amplitude, nAm
of composition do these neural traces reflect? 10 10 ASL phrase (white lamp)
Syntactic effects are difficult to distinguish
from semantic effects, because in natural lan-
5 5
guage, syntactic changes usually alter the mean-
ing of the expression. Consequently, many
results on the neural basis of basic syntactic 0 0
composition are actually results on the pro- 215 - 257 ms 335 - 450 ms ASL list (green, lamp)
cessing of artificial stimuli that are intended to
lack meaning. Most commonly, such expres- vmPFC
sions are made up of nonsense words strung white lamp
15 15
together with real affixes and other grammatical
vocabulary (4, 5, 8, 15–17). Processing such
Amplitude, nAm

Amplitude, nAm

stimuli may substantially differ from the com- 10 10 English phrase

prehension of natural language.
Whereas syntax is difficult to vary while keep-
ing meaning constant, the reverse is easier: It is 5 5 green, lamp
possible to keep syntactic structures constant
while varying meaning. In fact, such manipu- 0 0
lations have ruled out a syntactic explanation 100 - 190 ms 315 - 500 ms English list
for both the left temporal and ventromedial
combinatory effects when measured with MEG. Fig. 1. Building phrases in comprehension and production. A single step of composition engages the
ADAPTED BY JOSHUA BIRD/SCIENCE

The LATL’s early (200 to 250 ms) contribution LATL and vmPFC. In comprehension (top), LATL engagement precedes vmPFC engagement (1), but in
to composition appears conceptual in nature. production (bottom) the two operate in more parallel fashion. In production, a picture-naming paradigm
This finding is consistent with those of hemo- allows for the study of different languages while controlling the physical stimulus perfectly. One can
dynamic and neuropsychological studies (14, 18). simply ask participants to name the pictures in different languages. Here, similar LATL and vmPFC effects are
Within the two-word paradigm, as used in shown for the planning of phrases in English and American Sign Language (ASL) (3). Recent research has
MEG, the clearest evidence for the conceptual addressed the functional roles of these neural signatures of composition, demonstrating that neither appears
hypothesis comes from studies that have ma- syntactic in nature and that the LATL has a distinctly conceptual profile (7, 9, 24, 25). nAm, nanoamperes.
nipulated the conceptual specificity of the Error bars indicate SE. *P < 0.05; **P < 0.01; ns, not significant.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 63

 L A N G UA G E AN D TH E B RA I N

combining lexical items. LATL ampli- LATL sensitivity to conceptual specifcity binding within a word may not have similar
tudes elicited by the second word of a redundancy. This would make single-word
phrase appear to reflect the proportion processing more fragile than basic combi-
of features contributed by the first word Lots of features natory processing. Indeed, focal brain dam-
LATL
onto the combined feature set of the whole age hardly ever results in the inability to
phrase (7, 9). Consider the phrase “Indian form simple phrases, whereas problems in
food.” Here, the modifier “Indian” restricts to single-word access and more-complex syn-
a h
the set of foods to which the noun refers: tom dis tactic processing are classic patient profiles.
all cuisines apart from Indian are excluded.
The modifier “Asian” would have a weaker Production: Planning phrases in speech
More LATL activity
effect, because “Asian food” may refer to and sign
on second word
any type of Asian food (Korean, Japanese, Although the terms “speech” and “lan-
Indian, etc.). This is the type of factor to Fewer features guage” are often used interchangeably,
which the LATL is sensitive (Fig. 2) (9). As LATL speech is only one way to externalize lan-
the feature space of the first word becomes guage, as indicated by the use of more
more specific, the magnitude of the ob- than 200 signed languages around the
e
served LATL signal increases when that e tabl i s h world. Signed languages allow for tests
word is integrated onto the second word. veg d
about whether a result reflects the fun-
The feature space of the second word damental core of human language or rel-
matters as well: the vaguer the meaning, Less LATL activity ates to the specifics of spoken languages.
on second word
the more the first word affects the joint In general, signed and spoken languages
feature set of the whole phrase and the are known to be fundamentally similar
larger the observed LATL signal on the systems (29) and engage the same broad
second word (7). Neither the syntactic λx. red (x) brain networks (30). However, under-
nor the logico-semantic system has any NP & boat (x) standing is more elusive for more-specific
representation of the conceptual feature processes. For the minimal composition
space in these terms, and thus neither sys- results discussed above, we have evidence
tem can explain these patterns. Studies for similarity between spoken and signed
also reveal that the LATL can operate in language. Detecting such similarity is diffi-
the absence of local syntactic combina- cult in comprehension, because the nature
tion, as long as the task goal is to combine Adj λx. red (x) N λy. boat (y) of the physical signal differs markedly for
meanings (2, 19). the two types of languages. At the earliest
If the LATL combines some aspects of Syntactic Logico-semantic Conceptual combination stages of perception, hand movements and
word meanings at 200 to 250 ms, those sounds engage entirely different systems,
aspects must have been retrieved from Fig. 2. Components of composition and conceptual sensitivity which makes the detection of similarity
memory by that time. This estimate is on of the LATL. Composition is thought to subdivide into syntactic, methodologically challenging. However,
the early side, given that classic and widely logico-semantic, and conceptual subroutines. The LATL shows a the early planning stages of language
replicated findings about the timing of larger signal when a more specific meaning integrates into a production could be very similar until
semantic access place it at 300 to 400 ms word from context and is therefore unlikely to reflect syntactic or the planned representations begin to en-
(20). However, today we also have evidence logico-semantic aspects of composition (7, 9). NP, noun phrase; gage the motor systems, as shown in a
that semantic access can begin much more Adj, adjective; N, noun. study that used picture naming as a shared
quickly—as early as the 100-ms mark production task for speakers of English
(21, 22). The semantic feature space may activate gions. The LATL, for example, also participates and for congenitally deaf signers of American
gradually in a few hundred milliseconds (23), by in the processing of single words. This suggests Sign Language (3).
some yet-to-be articulated mechanism. that the region not only binds features across In this picture-naming study, both speak-
The hypothesis space regarding vmPFC words but also connects features that make up ers and signers named colored objects with
function is still very open, although we do single concepts [e.g., their visual appearance, adjective–noun combinations, such as “blue
know that the vmPFC is also sensitive to se- function, or size (18)]. In MEG, activity reflecting cup.” Thus, the physical stimuli (the pictures)
mantics within syntactically parallel expres- across-words binding (composition) is more ro- were identical for the two groups. Bringing
sions. Therefore, it also is unlikely to reflect bust than activity reflecting within-word binding added value, the noncombinatory control con-
syntactic aspects of composition. The evi- (word retrieval) (7, 9), perhaps because the form- dition was lexically matched to the combina-
dence for this comes from the processing of er creates a new meaning whereas the latter ac- tory condition by presenting the colored object
expressions that involve implicit meanings tivates an existing connection. on a colored background (for example, a
triggered by semantic mismatches within well- But here is a puzzle: If the LATL is damaged, blue cup on a red background) and instruct-
formed expressions (24). Overall, the results single-word processing is impaired, not phrase ing participants to name the background
conform to a model in which the LATL serves composition (27, 28). How could this be if the color and the object during the noncombina-
ADAPTED BY JOSHUA BIRD/SCIENCE
as a rapid nonsyntactic feature combiner (25) LATL is a combinatory hub? A possible answer tory trials. Participants produced adjective–
and the vmPFC contributes to a late stage of lies in the redundant nature of the combinatory noun sequences in both conditions, but only
composition, perhaps representing the final system, as depicted in Fig. 2. If many different in one did the words form a coherent combi-
output of the entire combinatory processing subroutines carry a roughly similar function (i.e., natory representation. Replicating prior find-
stream in a region connected to broader systems building a phrase), then this system may be dif- ings for English, the study showed that both
of social cognition and episodic memory (26). ficult to break. The function of an impaired sub- signers and speakers engaged the LATL and
Note, however, that this account does not ad- routine could be compensated for by the others, vmPFC while planning the phrases describing
dress other possible contributions of these re- at least to some degree. In contrast, feature the colored objects but not while planning the

64 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 background color–plus–object descriptions. ly structural origin (39). Next, we should ask frontal gyrus (LIFG) (Fig. 3). In MEG measure-
Thus, neural reflections of composition man- whether these rhythms also track nonlin- ments, the AG peaks sharply around 170 ms
ifest for both comprehension and produc- guistic chunking or grouping. Such tracking after the onset of a visual word and shows
tion and also generalize to a different set of would suggest either that the rhythms reflect sensitivity to the number of argument slots the
articulators. something more general than combinatory word has. For example, a transitive verb elicits
language operations or that combinatory ope- higher activation than an intransitive one (41),
Syntax: Where is it hiding? rations in language share properties with replicating prior hemodynamic research (42).
Although the body of work discussed so far general chunking processes. Regardless, these Given the robustness of the signal from this
has delineated a starting point for under- rhythms may be a clue to syntactic parsing region, future work should be able to delineate
standing the brain’s mechanisms for meaning during comprehension. how this region represents the combinatory
composition, it seems to have taught us nothing potential of individual words.
about syntax. So far, each correlate of composi- Understanding the broader The LIFG has been associated with syntac-
tion has turned out to be semantically sensitive. combinatory network tic combination in several prominent models
The neuroscience-of-language field has long as- What are the most fruitful research directions (43, 44), on the basis of both hemodynamic
sumed that our brains build syntactic struc- for future progress in our understanding of and patient data. However, this region has been
ture during language processing. Today, it linguistic composition in the brain, especially relatively silent in MEG studies of composition
is reasonable to question this assumption. with regard to syntax? Results suggesting (1, 16, 35). Consequently, the timing of this
Is there a neurally implemented computation that the LATL can combine concepts, even activity is not well understood, which has lim-
that builds syntactic structure and does not when they do not syntactically combine (2, 19), ited our ability to formulate specific functional
compute any meaning—a mechanism that, indicate that, rather than trying to vary syntax hypotheses of this activity. However, a large-
upon encountering “angry bird,” composes while keeping meaning fully constant, we can sample MEG study has identified a mid-
the representation of the adjective category aim to vary syntax in ways that should not latency timing for LIFG activation during
with the representation of the noun category affect activity in the LATL. To achieve this, sentence processing, at ~300 to 450 ms after
to yield a representation of a noun phrase, we would vary syntax but not the string of word onset (11). This timing conforms to prior
with no information about the meanings of conceptually informative items. Conditions MEG results showing increased LIFG ampli-
the elements that were combined? may differ in some aspects of meaning but tudes for long-distance dependencies (as in
Despite much published literature on the would pose similar opportunities for concep- relative clauses in which the object of a verb is
neural underpinnings of syntax and a lesser tual combination. Studies following this logic expressed outside its canonical object posi-
amount on the brain basis of semantic com- have found evidence for more structure-based tion: e.g., “the ball that the dog ate”) (45). This
position, the brain basis of syntactic composi- processing in the left posterior temporal lobe result replicates prior hemodynamic literature,
tion remains mysterious (25, 31–34). (LPTL) (19, 40). MEG measurements suggest starting with the classic study of Stromswold
The state of the art can be summarized as that the timing of this activity parallels that and colleagues (46). Thus, whereas the partic-
follows: of activity in the LATL, starting at ~200 ms ipation of the LIFG in long-distance depend-
1) During the processing of natural, mean- (11, 16, 19). Thus, different parts of the tem- encies receives support from multiple methods,
ingful language, neural signals (as measured by poral lobe may simultaneously compose dif- the same is not true for basic composition. To
currently available techniques) are dominated ferent aspects of structure and meaning. understand the discrepancy, more studies are
by correlates of meaning, not structure, in both In addition to the LATL, vmPFC, and LPTL, needed in which the same design is used with
the two-word paradigm (25) and full sentences at least two other regions are thought of as both MEG and fMRI (16).
(31). This does not mean that syntax is not also part of the broader combinatory network (25): Finally, controlled laboratory experiments
computed, but it does make the isolation of the angular gyrus (AG) and the left inferior should continue to be complemented with
syntactic computations more challenging.
2) When brain activity associated with sen-
tence processing is modeled, word by word, LATL
with measures of syntactic structure, the mod- combinatory
Non-syntactic (conceptual)
els reliably predict activity in several regions
combinatory vmPFC
(35–37). However, we do not know the degree
hubs combinatory
to which these results are driven by purely
structural processing or by combinatory se-
mantic processing. For each element of addi- 0 100 200 300 400 MSEC
tional structure, an element of meaning is LIFG
usually added as well. Teasing apart structure sensitivity to long
rd PTL distance
and meaning is the fundamental challenge for wo
combinatory dependencies
the study of composition in natural language. efects
3) When syntactic phrases are presented (syntax?) AG
to listeners at consistent, predictable rates,
electrophysiological responses show power AG LIFG
PTL
ADAPTED BY JOSHUA BIRD/SCIENCE

increases that match the presentation rates of sensitivity

those structures, even in the absence of phys- to vmPFC
ical cues to structure (38). That is, our brains argument
structure LATL
notice phrases. Is the brain rhythm tracking
syntactic or semantic combinatorics? This ques-
tion is still open, but the rhythmic tracking Fig. 3. Spatiotemporal characteristics of the brain’s combinatory network. The functional profiles of the
has been replicated for an artificial grammar activities plotted above the timeline appear nonsyntactic. The regions below the timeline represent various
without meaning, pointing toward a pure- hypotheses for cortical loci of syntactic processing.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 65

 L A N G UA G E AN D TH E B RA I N

naturalistic experiments that analyze real- knowledge. In fact, many neuroimaging studies 10. A. Fyshe, G. Sudre, L. Wehbe, N. Rafidi, T. M. Mitchell,
world comprehension of language, such as use grammaticality judgments as the experi- Hum. Brain Mapp. 40, 4457–4469 (2019).
11. A. Hultén, J. M. Schoffelen, J. Uddén, N. H. L. Lam, P. Hagoort,
in stories or podcasts (35, 37). In naturalis- mental task: Participants indicate whether the Neuroimage 186, 586–594 (2019).
tic studies, neural activity is modeled, word presented expression is grammatical or not 12. N. Chomsky, Syntactic Structures (Mouton, 1957).
13. I. Heim, A. Kratzer, Semantics in Generative Grammar
by word, by regressors that represent various (4, 8). A caveat of this approach is that thinking (Blackwell, 1998), vol. 1185.
properties of the stimuli, including their com- about our syntactic knowledge may markedly 14. M. N. Coutanche, S. Solomon, S. L. Thompson-Schill,
binatory properties. Though powerful, this differ from simply using the knowledge. PsyArXiv (27 April 2019); https://doi.org/10.31234/
osf.io/9jptv.
method is only as good as the accuracy of Nevertheless, it is too early to conclude that 15. A. D. Friederici, M. Meyer, D. Y. von Cramon, Brain Lang. 74,
the employed regressors as representations syntactic composition is not part of the brain’s 289–300 (2000).
16. W. Matchin, C. Brodbeck, C. Hammerly, E. Lau, Hum. Brain
of the modeled processes. We currently have online combinatory machinery. Most of the Mapp. 40, 663–678 (2019).
a paucity of computational models of incre- necessary experiments remain to be done. For 17. C. Pallier, A.-D. Devauchelle, S. Dehaene, Proc. Natl. Acad.
mental semantic composition, as opposed to example, two-word phrases may simply be too Sci. U.S.A. 108, 2522–2527 (2011).
18. M. A. Lambon Ralph, E. Jefferies, K. Patterson, T. T. Rogers,
syntactic phrase building, and therefore this small to drive the syntactic engine. In fact, Nat. Rev. Neurosci. 18, 42–55 (2017).
body of work has not addressed the syntax- when adjective–noun combinations were 19. G. Flick, L. Pylkkänen, bioRxiv 439158 [Preprint]. 9 October 2018.
https://doi.org/10.1101/439158.
versus-semantics question. More effort should embedded in full sentences in a more intricate 20. M. Kutas, S. A. Hillyard, Science 207, 203–205 (1980).
be directed toward developing computational design, the posterior temporal cortex, associ- 21. F. Pulvermüller, R. Assadollahi, T. Elbert, Eur. J. Neurosci. 13,
models of incremental semantic composi- ated with syntax in a recent model (51), showed 201–205 (2001).
22. C. Teige et al., Cortex 103, 329–349 (2018).
tion. Today, such models can be informed a previously unidentified effect that reflected 23. J. R. Binder, R. H. Desai, Trends Cogn. Sci. 15, 527–536 (2011).
by our understanding of conceptual processing structure, not meaning (19). To test whether the 24. L. Pylkkänen, Lang. Linguist. Compass 2, 712–738 (2008).
25. L. Pylkkänen, J. R. Brennan, PsyArXiv (23 September 2019);
in the LATL—that is, we can create regres- syntactic engine needs some minimal amount https://doi.org/10.31234/osf.io/fa2xb.
sors designed to track LATL activity as it has of material before kicking into gear, studies 26. R. L. Jackson, C. J. Bajada, M. A. Lambon Ralph, L. L. Cloutman,
been observed in controlled experiments. should incrementally increase phrase size (17)— Cereb. Cortex 10.1093/cercor/bhz079 (2019).
27. J. R. Hodges, K. Patterson, S. Oxbury, E. Funnell, Brain 115,
though even then, the challenge of distinguishing 1783–1806 (1992).
Syntax as knowledge, semantics syntactic from semantic effects remains. 28. S. M. Wilson et al., J. Cogn. Neurosci. 26, 970–985 (2014).
as process? 29. W. Sandler, D. Lillo-Martin, Sign Language and Linguistic
Universals (Cambridge Univ. Press, 2006).
What if, after all of our efforts to find purely Outlook 30. M. MacSweeney, C. M. Capek, R. Campbell, B. Woll,
structural processing in the brain, we still find Our understanding of the neurobiology of com- Trends Cogn. Sci. 12, 432–440 (2008).
31. E. Fedorenko, A. Nieto-Castañon, N. Kanwisher,
nothing? Syntax in the brain is necessary to position is progressing. From characterizations Neuropsychologia 50, 499–513 (2012).
explain the fact that humans are exquisitely of how minimal phrases affect the brain, we 32. E. Lau, in Current Topics in Language, K. D. Federmeier,
D. G. Watson, Eds. (Psychology of Learning and Motivation
skilled at judging syntactic well-formedness, gain a foundation for understanding more Series, Academic Press, 2018), vol. 68, pp. 117–142.
even for sentences that have no coherent mean- complex phenomena. We will also want to 33. W. G. Matchin, Psychon. Bull. Rev. 25, 1682–1694 (2018).
ing. Chomsky made this point with his “colorless understand how the neurobiology of minimal 34. E. Fedorenko, A. Nieto-Castañón, N. Kanwisher, Brain Lang.
120, 187–207 (2012).
green ideas sleep furiously” example (12), which composition relates to simpler processing, such 35. J. R. Brennan, E. P. Stabler, S. E. Van Wagenen, W. M. Luh,
we recognize as a grammatical English sen- as associating two elements without semanti- J. T. Hale, Brain Lang. 157–158, 81–94 (2016).
36. M. J. Nelson et al., Proc. Natl. Acad. Sci. U.S.A. 114,
tence, although is it semantically incoherent. cally combining them. Computational advances E3669–E3678 (2017).
But what if we cannot find evidence that our have improved our ability to extract knowl- 37. L. Wehbe et al., PLOS ONE 9, e112575 (2014).
38. N. Ding, L. Melloni, H. Zhang, X. Tian, D. Poeppel,
brains actually build syntactic structure? edge from data. For example, application of
Nat. Neurosci. 19, 158–164 (2016).
Syntax may be something that the brain multivariate pattern analysis to time-resolved 39. H. Getz, N. Ding, E. L. Newport, D. Poeppel, Cognition 181,
knows rather than does. Perhaps the combi- electrophysiological signals allows us to inquire 135–140 (2018).
40. W. Matchin, C. H. Liao, P. Gaston, E. Lau, Neuropsychologia
natory steps, which consume energy and make when, during the processing of a stimulus, a 125, 116–128 (2019).
our neurons fire, are all semantic, and syn- particular representation is active (52). This 41. A. Williams, S. Reddigari, L. Pylkkänen, Neuropsychologia 100,
tactic processing amounts to comparing these ability enables investigations into how com- 131–143 (2017).
42. C. K. Thompson et al., J. Cogn. Neurosci. 19, 1753–1767
semantic structures to our stored knowledge position affects representations of individual (2007).
of syntax. The knowledge may have the format words (10). Computational modeling of lan- 43. A. D. Friederici, Physiol. Rev. 91, 1357–1392 (2011).
44. P. Hagoort, Trends Cogn. Sci. 9, 416–423 (2005).
of generative rules that create structure (12) or guage has also brought linguistic theory to the 45. K. Leiken, B. McElree, L. Pylkkänen, Front. Psychol. 6, 1739 (2015).
may represent the structures themselves (47). center stage of the cognitive neuroscience of 46. K. Stromswold, D. Caplan, N. Alpert, S. Rauch, Brain Lang. 52,
This type of “syntax as knowledge, semantics sentence processing, as computational model- 452–473 (1996).
47. A. E. Goldberg, Trends Cogn. Sci. 7, 219–224 (2003).
as process” model would make pure syntactic ing benefits from the broad models of sentence 48. J. Hale, in Proceedings of the Second Meeting of the North
composition unmeasurable in the incremen- structure that linguistic theory offers. Thus, an American Chapter of the Association for Computational
Linguistics on Language Technologies (Association for
tal combinatory steps that build sentences. interdisciplinary synergy begins. Although we Computational Linguistics, 2001), pp. 1–8.
However, syntactic knowledge could still be remain far from understanding how our brains 49. A. Staub, C. Clifton Jr., J. Exp. Psychol. Learn. Mem. Cogn. 32,
used to make predictions about upcoming create the meanings of natural language, the 425–436 (2006).
50. E. Lau, C. Stroud, S. Plesch, C. Phillips, Brain Lang. 98, 74–88 (2006).
language (48), and thus neural activity could path forward is becoming less and less foggy. 51. W. Matchin, G. Hickok, PsyArXiv (3 May 2019);
be modulated by the degree to which the en- https://doi.org/10.31234/osf.io/6394f.
RE FERENCES AND NOTES 52. J. R. King, S. Dehaene, Trends Cogn. Sci. 18, 203–210 (2014).
countered language matches the predictions.
1. D. K. Bemis, L. Pylkkänen, J. Neurosci. 31, 2801–2814 (2011).
These types of syntactic prediction effects have 2. D. K. Bemis, L. Pylkkänen, PLOS ONE 8, e73949 (2013). AC KNOWLED GME NTS
been widely documented both for behavioral 3. E. Blanco-Elorrieta, I. Kastner, K. Emmorey, L. Pylkkänen, I thank E. Blanco-Elorrieta, E. Fedorenko, J. Fridriksson, R. Law,
(49) and neural measures (50). In summary, on Sci. Rep. 8, 5492 (2018). J. Li, A. Marantz, A. Parrish, M. Tulling, S. Wilson, and an
4. M. Schell, E. Zaccarella, A. D. Friederici, Cortex 96, 105–120 (2017).
this hypothesis, top-down predictions could be anonymous reviewer for helpful comments. Funding: Support for
5. K. Segaert, A. Mazaheri, P. Hagoort, Eur. J. Neurosci. 48, the writing of this Review and research summarized within it was
either syntactic or semantic, but bottom-up 2651–2662 (2018). provided by grant G1001 from the NYUAD Institute, New York
composition would be entirely semantic. 6. M. Westerlund, I. Kastner, M. Al Kaabi, L. Pylkkänen, University Abu Dhabi, and grant BCS-1221723 from the National
Another way tacit syntactic knowledge could Brain Lang. 141, 124–134 (2015). Science Foundation. Competing interests: There are no
7. M. Westerlund, L. Pylkkänen, Neuropsychologia 57, 59–70 (2014). competing interests.
manifest in neural signals is if we measure 8. E. Zaccarella, A. D. Friederici, Front. Psychol. 6, 1818 (2015).
brains while they explicitly think about this 9. L. Zhang, L. Pylkkänen, Neuroimage 111, 228–240 (2015). 10.1126/science.aax0050

66 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 Publish your research
in Science Signaling

Science Signaling publishes the latest advances in regulatory biology relevant to physiology
and disease, including insights into the basic mechanisms of intracellular signaling and
intercellular regulation, host-microbe interactions, applied signaling for drug discovery
and synthetic biology, and the development of novel analysis methods.
Submit your research today. Learn more at: ScienceSignaling.org

Like us: @ScienceSignaling Follow us: @scisignal

1004Product.indd 67 9/25/19 1:11 PM

 RESEARCH
IN S CIENCE JOURNAL S
Edited by Michael Funk

The Sunda pangolin,

Manis javanica, is
critically endangered
owing to poaching
and illegal trade.

WILDLIFE TRADE

A heavy toll

T
rade in wildlife, and their parts, is well recognized for a few some sort. The impacts of trade tend to be concentrated in certain
key species, such as elephants and rhinos, but it occurs phylogenetic groups, thus the potential for long-term impact on
globally, across a wide array of species. Scheffers et al. certain lineages is substantial. This analysis allows for predic-
looked across tens of thousands of vertebrate species and tion of potential for trade where it does not yet occur, facilitating
found that one in every five species is affected by trade of proactive prevention. —SNV Science, this issue p. 71

ATOMIC PHYSICS of a few seconds. Increasing each component also has its sensory feedback through nerve
the number of tweezers should own smaller circumstellar disk. stimulation in three trans-
Building up an improve the figures of merit of Spiral filaments of dust and gas femoral amputees. Stimulation
optical clock this platform. —JS connect the small disks to the improved mobility, decreased
Arrays of optical tweezers can be Science, this issue p. 93 larger one. Material is accreting falling episodes, and increased
used to trap atoms, which can preferentially onto the star that the perception of the prosthesis
then be manipulated individually. currently has a lower mass, driv- as part of the body. Active com-
Such arrays have shown promise STAR FORMATION ing the masses toward similar plex tasks were accomplished
in quantum simulation of many- values. —KTS with reduced effort when nerve
body systems. Norcia et al. now
Gas flows and disks Science, this issue p. 90 stimulation was turned on.
demonstrate that they can also in a young binary —MM
be used as a platform for optical Many stars are in binary Sci. Transl. Med. 11, eaav8939 (2019).
NEUROPROSTHETICS
clocks. The researchers lined up systems, pairs of stars gravi-
Good vibrations

PHOTO: SUZI ESZTERHAS/MINDEN PICTURES

10 optical tweezers in a one- tationally bound to each other,
dimensional array, where each often with the two compo- The lack of sensory feedback BIODIVERSITY LOSS
tweezer held either one or zero nents having similar masses. from the leg prosthesis in lower
atoms of strontium. The atoms It remains unclear how these limb amputees is associated
Staggering decline of
were subjected to laser light systems assemble and accrete with risk of falls, low mobility, bird populations
whose frequency was tuned to a material. Alves et al. made and perception of the pros- Because birds are conspicu-
clock transition in strontium. By high-resolution observations of thesis as an external object. ous and easy to identify and
monitoring the number of atoms a young binary star that is still in Petrini et al. tested a leg neu- count, reliable records of their
in each tweezer, the researchers the process of forming. A large roprosthesis, which provided occurrence have been gathered
measured a long coherence time disk surrounds both stars, and real-time on-demand tactile over many decades in many

68 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 parts of the world. Drawing on STRUCTURAL BIOLOGY Edited by Caroline Ash
such data for North America,
The yin and yang of
IN OTHER JOURNALS and Jesse Smith
Rosenberg et al. report wide-
spread population declines of Raf inhibition
birds over the past half-century, Many human melanomas
resulting in the cumulative contain an overactive form of Leaves
loss of billions of breeding Raf kinase (B-Raf). Inhibitors developing
individuals across a wide range are effective against the mutant in sunlight or
of species and habitats. They B-Raf, but, paradoxically, they shade differ in
show that declines are not activate wild-type B-Raf, limit- size because of
restricted to rare and threat- ing their therapeutic potential. differences in
ened species—those once Kondo et al. determined the cell elongation.
considered common and wide- structure of a phosphorylated
spread are also diminished. B-Raf dimer in complex with
These results have major impli- the scaffold protein 14-3-3 by
cations for ecosystem integrity, cryo–electron microscopy.
the conservation of wildlife Although both kinases are in
more broadly, and policies the active conformation, one
associated with the protection is blocked by the C-terminal
of birds and native ecosystems tail of the other. This configu-
on which they depend. —AMS ration inhibits one active site
Science, this issue p. 120 but also stabilizes the dimer
in the active conformation.
Understanding this mechanism
3D PRINTING provides a framework for devel-
opment of inhibitors that do not
Speeding up activate wild-type Raf. —VV
submicrometer printing Science, this issue p. 109
Using light to build three-
dimensional structures with
photopolymerization is the ATMOSPHERIC SCIENCE
basis for two-photon lithogra-
phy. However, there has been
Consequences of
a trade-off between speed and nuclear war
resolution for fabricating struc- The likely catastrophic envi- PLANT SCIENCE
tures with this method. Saha et ronmental effects of global
al. optimize a new parallel print- nuclear war are well-described Light tuning of leaf size
ing methodology that relies on in the scientific literature,

T
he Arabidopsis plant, when grown under intense light,
ultrafast lasers. They show the but the effects of a more
develops thicker leaves than when grown under dim
ability to dramatically increase limited, regional nuclear war
light. Hoshino et al. parse the various processes involved.
the speed of printing while are also staggering. Toon
Cell elongation and organized divisions underpin the
maintaining submicrometer et al. combined a realistic
thickness of leaves growing in intense light. Later in
resolution. —BG war scenario involving the
development, cell expansion and increases in intercellular
Science, this issue p. 105 neighboring, nuclear-capable
spaces finalize the respective dimensions of leaves growing in
PHOTOS (LEFT TO RIGHT): VU NGUYEN AND SOURABH SAHA; DULYANUT SWDP/GETTY IMAGES

nations of India and Pakistan

sunlight or in shade. Blue-light signaling drives initial unequal
with established climate
cell elongation in sun-exposed leaves, whereas sucrose, a
models to explore plausible
photosynthetic product, promotes the late phase of leaf
global effects of such conflict.
thickening for both sun-exposed and shaded leaves. With leaf
These models suggest that, in
development linked to environmental signals, the plant’s leaf
addition to human casualties,
structure can be fine-tuned to best suit its microenviron-
an India-Pakistan nuclear war
ment. —PJH
would start fires releasing tens
Plant J. 10.1111/tpj.14467 (2019).
of trillions of grams of black
carbon into the stratosphere.
Distributed globally, this soot
would limit the amount of STELLAR ASTROPHYSICS exploded as supernovae. A sec-
sunlight that reaches Earth’s ond stellar generation formed
surface, drastically reducing
An ancient star from gas polluted by those
surface temperatures and thus missing its iron supernovae, and some of those
potentially limiting food sup- The Big Bang produced only stars are still around today.
plies globally for more than a hydrogen, helium, and traces Nordlander et al. discovered a
decade. —KH of lithium. Heavier elements star with an extremely small
3D printed pillars with submicrometer Sci. Adv. 10.1126/ were forged within the first amount of iron, much more
features sciadv.aay5478 (2019). generation of stars, which then carbon, and low abundances

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 69

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

NEUROSCIENCE

Merging motor and sensory

T
he posterior parietal cortex of the rat’s
brain is involved in a broad repertoire of
behaviors. These range from integration of
multiple sensory inputs to organizing the
corresponding muscular responses. The
function of the various microcircuit compo-
nents is incompletely understood. Mohan et al.
used several techniques to record from neurons
in the different layers of the posterior parietal
cortex in anesthetized and awake animals. The
authors found that different parts of the pro-
cessing circuits occurred in the different layers.
Touch was encoded in layers 2 to 4, whereas
exploratory whisker motion was represented
in layers 2 to 4 and layer 6, but not in layer 5.
These results may help the development of
brain-machine interfaces needed to alleviate the
effects of spinal cord injury. —PRS
J. Neurosci. 39, 7332 (2019).

Touch and corresponding whisker motion are represented

in diferent layers of the brain.

of several other elements. radiotherapy but also maintain the motivation for digging was origin in the last universal
Comparison to supernova excellent antitumor activity. realized. Although these few common ancestor. An origin of
models indicates this is an This is because T cell repro- observations were made on this branch in Archaea is likely
old second-generation star gramming is facilitated by the captive animals, they empha- given wide distribution therein
enriched by a single super- tumor microenvironment. If size that we still have much to and enzyme phylogenies
nova. The supernova ejected human tumors behave in the learn about the other animals consistent with those of the
its outer carbon-rich layers, same way, then we can expect that inhabit this planet even organisms. —MAF
but the inner core of iron more effective combinations as they approach extinction. Nat. Microbiol. 10.1038/
mostly fell back into the black of cancer treatments to be —SNV s41564-019-0534-2 (2019).
hole formed during the explo- developed. —PAK Mamm. Biol. 98, 102 (2019).
sion. —KTS Nat. Commun. 10, 3959 (2019).
ACTINIDE CHEMISTRY
Mon. Not. R. Astron. Soc. 488,
METABOLIC ORIGINS
L109 (2019).
ANIMAL BEHAVIOR
Americium in a MOF
Dawn of methylotrophy Americium is a by-product

CANCER
Warty pig sticking Methyl groups, including in nuclear reactors that
Not too long ago, it was those derived from or ending continues to emit radiation
T cells at work argued that using tools was up in methane, are managed for thousands of years. Its
Combining the three pillars of a hallmark of being human. through a dizzying array of separation from spent nuclear
cancer treatment—surgery, However, this argument has metabolic pathways. Some of fuels has therefore motivated
radiotherapy, and chemo- been dispelled by more recent these originated early in life’s copious study of its fundamen-
therapy—has greatly improved behavioral studies on other history, and their evolution tal coordination chemistry.
patient survival. With the species—first by observations paralleled and contributed to To this end, Cahill et al. have
recent development of effec- in great apes, then in spe- Earth’s chemistry. Adam et now incorporated trivalent
243
tive cancer immunotherapies, cies more distantly related to al. analyzed a set of bacterial Am ions into a metal-organic
oncologists now have a fourth humans, such as crows and and archaeal genomes, looking framework (MOF). The frame-
pillar to consider in design- dolphins. Root-Bernstein et for patterns in the presence work architecture was chosen
ing combination treatments. al. have discovered that the and divergence of enzymes on the basis of known analogs
Previous work had suggested Visayan warty pig can use in the tetrahydromethanop- prepared from chemically
that radiotherapy damages tools too. Female pigs of this terin methyl branch of the similar but more stable lantha-

PHOTO: LIFE ON WHITE/GETTY IMAGES

T cells, leading to the perception critically endangered spe- Wood–Ljungdahl pathway, nide ions. Although radioactive
that combining radiotherapy cies from the Philippines use which is involved in carbon decay progressively damaged
and immunotherapy would be sticks and bark as digging tools redox reactions. Expanding the crystals, the authors were
ineffective. Studying mouse during nest construction. The previous analyses narrowed still able to refine structural
models of tumors that are first observations were made the distance between bacte- data over the course of
already infiltrated with T cells, opportunistically and then rial and archaeal sequences, 3 months. —JSY
Arina et al. find that these T cells additional occurrences of the making horizontal transfer Angew. Chem. Int. Ed. 10.1002/
not only survive localized behavior were noticed after more plausible relative to an anie.201909988 (2019).

70 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

RESE ARCH

ALSO IN SCIENCE JOURNALS Edited by Michael Funk

PREBIOTIC CHEMISTRY INTERGALACTIC MEDIUM FOREST ECOLOGY RESPIRATORY ENZYMES

Conditions right for Glowing filaments of the Fungal influence on Hemes switch spots in a
making nucleosides cosmic web density dependence terminal oxidase
In the absence of biological Most gas in the Universe lies Tree species in highly diverse Reduction of molecular oxygen to
catalysts and metabolism, can in the intergalactic medium, tropical forests tend to exhibit water is the driving force for res-
atmospheric and geochemical where it forms into sheets and conspecific negative density piration in aerobic organisms and
processes provide the sub- filaments of the cosmic web. dependence, a phenomenon is catalyzed by several distinct
strates and conditions required Clusters of galaxies form at the whereby individuals of the integral membrane complexes.
for production of biological intersection of these filaments, same species tend to grow at These include an exclusively
molecules? Becker et al. devised fed by gas pulled along them by a distance from one another. prokaryotic enzyme, cytochrome
an abiotic synthetic scheme that gravity. Although this picture is This is understood to be a key bd–type quinol oxidase, which is
allows for accumulation of both well established by cosmological driver of species coexistence. a potential antimicrobial target.
purine and pyrimidine nucleo- simulations, it has been difficult The strength of negative density Safarian et al. determined a
side mono- and diphosphates to demonstrate observationally. dependence varies between spe- high-resolution cryo–electron
(see the Perspective by Hud and Umehata et al. mapped emission cies, but the mechanisms driving microscopy structure of this
Fialho). A key starting material from the intergalactic medium in this variation are unknown. Chen enzyme from the enteric bacte-
for this chemistry, hydroxyl- an area around galaxies that are et al. studied tree species in a rium Escherichia coli. Comparison
amine and/or hydroxylamine starting to form a cluster (see subtropical forest in China and to a homolog reveals a complete
disulfonate, can form under the Perspective by Hamden). found an important role for soil- relocation of the site of oxygen
plausible early atmospheric They found that the gas is dwelling fungi in this variation. binding and reduction caused by
conditions. Cycles between arranged into filaments, whose Elevated accumulation of patho- a change in the arrangement of
wet and dry conditions provide position and velocity correlate genic fungi leads to stronger heme cofactors and channels in
the environments necessary to with star-forming galaxies, sup- negative density dependence, the protein scaffold. This switch
complete formation of purine porting the theoretical picture. whereas elevated accumulation illustrates the diversity of struc-
and pyrimidine bases essentially —KTS of mutualistic fungi leads to ture and function in this family of
in one pot. —MAF Science, this issue p. 97; weaker negative density depen- enzymes and might reflect dif-
Science, this issue p. 76; see also p. 31 dence. —AMS ferent biochemical roles of these
see also p. 32 Science, this issue p. 124 homologs. —MAF
Science, this issue p. 100
QUANTUM PHYSICS
SONG LEARNING
A test of quantum gravity LIPID METABOLISM
CHROMATIN BIOLOGY
An imitation circuit Quantum mechanics and the
Distant cousin helps spot
Animals, including humans, rely general theory of relativity Twisting DNA to move it
heavily on imitation and social represent two pillars of modern animal enzyme A key regulatory aspect of DNA
learning, yet we know little physics, but unification of the In addition to forming the is how tightly it is wrapped
about how this process operates two theories remains an open membranes that enclose cells, around nucleosomes, the histone
in the brain. Zhao et al. used problem. Theories of quantum lipids are important signaling cylinders that help package DNA.
optogenetic manipulation of a gravity abound, but they tend to molecules. Plasmalogens, which The exposure of DNA determines
synaptic pathway connecting lack an experimental foundation. contain a vinyl ether linkage, accessibility of proteins, such as
auditory and vocal motor circuits One such proposed theory, event are an abundant group of lipids transcription factors and DNA
to implant song memories suf- formalism, predicts that a pair of in animals. How these lipids methyltransferases, which in
ficient to guide song learning entangled particles decorrelate are synthesized from precur- turn regulate gene expression
into young zebra finches (see as they pass through different sors with an alkyl ether linkage and cell identity. How can DNA
the Perspective by Clayton). regions of the gravitational well has been a mystery. Gallego- be unwound from nucleosomes
Activation of this circuit overrode of a planetary object. Xu et al. García et al. found an enzyme, to modulate gene expression?
learning from live tutors. These present results of a quantum CarF, in the social bacterium In a Perspective, Bowman and
experiments define circuits optical test of this proposal using Myxococcus xanthus that Deindl discuss recent advances
essential for social learning of the quantum satellite Micius. produces plasmalogens used in in understanding the mecha-
songs from tutors and show that Using entangled photon pairs, a signaling pathway for singlet nism by which DNA movement
such memories can be localized. one sent to the satellite and the oxygen, a marker of photooxida- around nucleosomes occurs.
—SNV other retained on Earth, they find tive stress. They then showed Nucleosome remodelers pull on
Science, this issue p. 83; no evidence for the predicted that the animal homolog could the DNA strands to influence how
see also p. 33 decorrelation effects. The results catalyze the final step in plas- the helix is twisted, which allows
may help shed light on the inter- malogen synthesis in bacterial the strands to move around the
play between quantum theory and human cells, thus resolving nucleosome. Similar mecha-
and gravity. —ISO a source for plasmalogens in nisms might also exist for other
Science, this issue p. 132 animals. —MAF protein complexes that move
Science, this issue p. 128 along DNA. —GKA
Science, this issue p. 35

70-B 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RE S E ARC H

SYSTEMS BIOLOGY LYMPHOCYTES

Protein competition Sending B cells back
drives cell fate in time
Cell survival can require B-1 cells are a subset of self-
switching mechanisms that are reactive B cells that arise in early
flexible enough to accommodate life. Precisely how and why the
environmental changes but also immune system permits the
stable for the required duration. development of self-reactive
Lord et al. created a switching B cells in neonates remains
system in bacteria based on a mystery. By studying B cell
stochastic competition between development in neonatal mice,
two proteins: one is a transcrip- Vanhee et al. uncovered the
tional repressor, and the other importance of RNA binding pro-
is an antagonist that binds the tein Lin28b in facilitating positive
repressor and locks it in an inac- selection of self-reactive B-1 cells
tive state. They show that this in neonates. They also found that
system controls switching of the ectopic expression of Lin28b
bacterium Bacillus subtilis from was sufficient to promote selec-
a motile, unicellular state to an tion of self-reactive B-1 cells in
immobile, multicellular state, adult mice. The authors propose
and that the control system is that Lin28b functions as a cell-
transferable to another distantly intrinsic switch that jumpstarts
related bacterium. Similar the generation of B cells in early
mechanisms could be more life. —AB
widely operable in biological sys- Sci. Immunol. 4, eaax4453 (2019).
tems than previously recognized.
—LBR
Science, this issue p. 116

CANCER THERAPY
Targeting a kidney cancer
People with von Hippel-Lindau
syndrome, caused by loss
of the gene VHL, are at an
increased risk of developing
clear cell renal cell carcinoma
(ccRCC). Currently, drugs
that inhibit the transcription
factor HIF-2a show some
efficacy but not in all patients.
Nicholson et al. performed a
screen to identify targets that
limited tumor growth in a VHL
double-knockout fly model. They
discovered that inhibiting the
cyclin-dependent kinases CDK4
and CDK6 together impaired
tumor growth in this model
regardless of HIF-2a depen-
dency, and they confirmed these
results in human ccRCC cells
and patient-derived xenografts
in mice. —LKF
Sci. Signal. 12, eaay0482 (2019).

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 70-C

 RESEAR CH

◥ what species are being traded and, second, the

RESEARCH ARTICLES identification of where traded species occur. In
this study, we searched the Convention on Inter-
WILDLIFE TRADE national Trade in Endangered Species of Wild
Fauna and Flora (CITES) and the International
Global wildlife trade across the tree of life Union for Conservation of Nature Red List of
Threatened Species (IUCN Red List) databases
Brett R. Scheffers1*†, Brunno F. Oliveira1,2*, Ieuan Lamb3, David P. Edwards3† to identify traded terrestrial vertebrate species
(birds, mammals, amphibians, and squamate
Wildlife trade is a multibillion dollar industry that is driving species toward extinction. Of >31,500 reptiles). Using our compiled list, we provide an
terrestrial bird, mammal, amphibian, and squamate reptile species, ~18% (N = 5579) are traded globally. evaluation of the global extent of wildlife trade
Trade is strongly phylogenetically conserved, and the hotspots of this trade are concentrated in the across the tree of life to determine whether trade
biologically diverse tropics. Using different assessment approaches, we predict that, owing to their targets specific evolutionary branches. We then
phylogenetic replacement and trait similarity to currently traded species, future trade will affect up to used species range maps to identify global hot-
3196 additional species—totaling 8775 species at risk of extinction from trade. Our assessment spots of wildlife exploitation and to determine
underscores the need for a strategic plan to combat trade with policies that are proactive rather than how those hotspots vary between trade for pets
reactive, which is especially important because species can quickly transition from being safe to being or products (i.e., medicine, luxury foods, skins).
endangered as humans continue to harvest and trade across the tree of life. Although emerging gene- and web-based tech-
niques can help to identify the precise sources

T
of traded individuals, our approach allows us to
he tree of life is being pruned by human the ivory-like casque of the helmeted hornbill identify the likely global epicenters of diversity
activities at an unprecedented rate (1). (Rhinoplax vigil) resulted in tens of thousands in traded animals.
Yet, although we understand the global of birds traded annually since around 2012 (10).
footprint of land degradation and deforest- Both species are now critically endangered (11). What species are traded?
ation and how that manifests in species Moreover, wildlife trade indirectly places sub- Trade in wildlife affects ~18% of all extant ter-
loss (2), we have a limited understanding of the stantial pressure on biodiversity through the restrial vertebrate species on Earth. Our assess-
global extent and patterns of the wildlife trade. introduction of pathogens, including the glob- ment shows that 5579 of the 31,745 vertebrate
The trade of wildlife for luxury foods and medic- ally lethal amphibian fungus Batrachochytrium species have been reported as traded, with a
inal parts and as pets is now so substantial that dendrobatidis (12), and invasive species, such higher percentage of all birds (23% of 10,278
it represents one of the most prominent drivers as the Burmese python (Python bivittatus) in species) and mammals (27% of 5420 species)
of vertebrate extinction risk globally (3). Each Florida, USA (13). globally traded than reptiles (12% of 9563
year, billions of wild plants and animals are The enormous trade in wildlife begs the ques- species) and amphibians (9% of 6484 species)
traded to meet a rapidly expanding global de- tion whether we can better protect species from (Fig. 1 and table S1). Our assessment across
mand (4, 5), a demand so insatiable that, globally, human demand, a concern that is at the fore- the CITES and IUCN databases yields a total
US$8 billion to $21 billion is reaped annually front of the wildlife trade crisis. Combating wild- that is 40 to 60% higher than prior recorded
from the illegal trade, making it one of the life trade requires, first, the identification of estimates [e.g., (3, 14, 15)]. Traded species are
world’s largest illegitimate businesses (5, 6).
The high demand for wildlife products and
pets has driven pronounced losses in enigmatic
species such as tigers, elephants, rhinos, and poi-
son dart frogs (7). Some subspecies are already
extinct [e.g., the last Javan rhino, Rhinoceros
sondaicus annamiticus, was shot for its horn
in 2010 in Vietnam (8)] or on the cusp of extinc-
tion in the wild (e.g., Bali myna, Leucopsar
rothschildi)—all because of trade. There is an
insidious aspect to this market force in that
these emblematic species only represent a tiny,
though well publicized, fraction of animal species
traded. If cultural preferences change, wildlife
trade can rapidly drive a species toward extinc-
tion. For instance, the emergence of widespread
demand in East Asia for pangolin scales and
meat has triggered major declines in some spe-
cies [e.g., Sunda pangolin (Manis javanica)] in
just two decades (9), and growing demand for
1
Department of Wildlife Ecology and Conservation, Institute of
Food and Agricultural Sciences, University of Florida,
Gainesville, FL 32611, USA. 2Department of Biology and
Environmental Sciences, Auburn University at Montgomery,
Montgomery, AL 36124, USA. 3Department of Animal and Plant Fig. 1. Wildlife trade in terrestrial vertebrates (birds, mammals, amphibians, and reptiles)
Sciences, University of Sheffield, Sheffield S10 2TN, UK. affects ~18% of species globally. Numbers in brackets are the total number of traded species.
*These authors contributed equally to this work.
†Corresponding author. Email: brett.scheffers@ufl.edu (B.R.S.); IUCN threat status codes: data deficient, DD; least concern, LC; near threatened, NT; vulnerable,
david.edwards@sheffield.ac.uk (D.P.E.) VU; endangered, EN; and critically endangered, CR.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 71

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

in higher categories of threat compared with as strong as expected under a Brownian motion table S4), and that the probability of being traded
nontraded species (especially among mam- model of evolution (fig. S2), indicating higher is positively related to body size (fig. S4). Over
mals and birds; Fig. 1 and table S2), confirm- levels of phylogenetic clustering relative to reptiles millennia, primitive human societies affected
ing wildlife trade as a driver of extinction risk. and amphibians (16). Highly traded families— large-bodied species through hunting for sub-
We found that trade occurs in 65% of all ter- those with >50% of their species traded—make sistence, which changed contemporary biogeo-
restrial vertebrate families (312 of 482 families; up more than one quarter (27%; 128 of 482 graphical patterns of animal body size (17, 18).
table S1). This pattern is evident across all ter- families) of the total families, which breaks down Our analysis shows that this pattern continues
restrial vertebrate groups considered, with mam- to 51% of mammal (N = 69), 32% of reptile (N = with modern humans through the wildlife trade.
mals and reptiles showing the highest percentage 23), 16% of bird (N = 32), and 5% of amphibian Trade also targets species that are unique
of families traded (mammals: 81%, N = 110; (N = 4) families (tables S1 and S3). and/or distinctive in traits. In our assessment
reptiles: 73%, N = 53), followed by amphibians Nonrandomness in trade across the tree of of evolutionary distinctiveness (a measure of
(55%, N = 41) and birds (55%, N = 108). Despite life implies high susceptibility for select clades, phylogenetic isolation) (19), which may yield
this broad exploitation, humans are targeting likely because of similar traits (such as voice species with unique traits (19, 20), our results
specific components of the tree of life (Fig. 2 quality, folklore, ivory, etc.). In exploring this, we suggest that, for birds—but not for mammals,
and fig. S1), as indicated by a significant phy- found that large-bodied species are more traded amphibians, or reptiles—traded species are
logenetic signal in wildlife trade for all taxa than small-bodied species, a pattern that holds more evolutionarily distinctive than non-
(fig. S2). Mammals and birds showed a signal regardless of IUCN threat category (fig. S3 and traded species (fig. S5). Furthermore, mean

Fig. 2. Wildlife trade occurs across the tree of life, but some clades are nonbold name indicates both high richness and proportion of total). The
more heavily targeted than others. Phylogeny branches for (A) birds, first outer band indicates threatened (VU, EN, CR, and DD; orange) and non-
(B) mammals, (C) amphibians, and (D) reptiles are colored to represent the threatened (LC and NT; yellow) species. DD species were considered
impact of wildlife trade up to each node (i.e., clade). Warmer colors (red) threatened because of their small geographic range size. The second outer
represent heavily traded branches (i.e., high percent of traded species). band indicates traded (red) and nontraded (pink) species. Gray concentric
The 20 highest traded families are labeled (bold name indicates high richness, circles scale a 20-million-year period.

72 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

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

family-wide evolutionary distinctiveness pre- predict that future trade will affect between 317 nomic groups (fig. S7). In Asia, Indonesia and
dicts the proportion of traded birds (fig. S6) and 3196 additional species (Fig. 3 and table Malaysia, as well as the Himalayas, are hotspots
(linear model: standardized coefficient = S5), amounting to between 101 and 826 bird, for trade (fig. S7), especially that of amphibians
0.18, P value = 0.01) but, again, not for traded 121 and 241 mammal, 9 and 268 amphibian, and mammals. Australia and Madagascar stand
mammals, amphibians, or reptiles. Humans and 86 and 1861 reptile species, with a >95 and out as the main trade hotspots for reptiles.
have long admired birds’ aesthetic attributes, >90% probability of future trade, respectively. Perhaps surprisingly, Indonesia, which is consid-
including song and plumage complexity, and As a precaution, we recommend that conserva- ered an epicenter of bird trade (21), was not
perhaps this long-standing admiration is re- tion attention be given not just to currently identified as a hotspot. Although Indonesia
flected in the bird trade. traded species but also to those species with the contains a lower diversity of traded bird species
Because we show that trade nonrandomly highest probabilities of being targeted by trade relative to some other areas (e.g., the Andes
targets species in specific clades and with specific in the future (see table S5 for the complete list and Atlantic coast of South America), birds in
traits, we were able to predict the species not of species and their probability of future trade). Indonesia are traded in very high abundance
yet (or not yet known to be) traded but at high (21). Thus, across vertebrates, some species may
risk of future trade as congeneric species become Where are the hotspots of only be collected for trade in small pockets of
rare or go extinct, or as their ranges become ac- traded species? their entire distribution range, with higher trade
cessible to hunters. On the basis of identified Although the footprint of trade spans all of volumes in certain countries, outside of protected
correlates of current trade, we provide mean- Earth’s habitable continents, we uncovered a areas, or closer to human settlements (21–23).
ingful estimates of future trade with >95 and pantropical dominance in the trade for verte- However, absent such fine-scale data for the
>90% probabilities (Fig. 3 and table S5). First, brates (Fig. 4 and fig. S7). Biogeographical majority of species and regions, our global maps
considering species in highly traded families, patterns in trade richness closely match patterns reveal the spatial idiosyncrasies in hotspots of
we predict between 5 and 48 species (i.e., 95 in species richness (Fig. 4 and table S6). South trade diversity among taxa.
and 90% probability, respectively) that are not America, Central to Southeast Africa, the Hima- Focusing on specific kinds of trade reveals
yet traded but are at high risk of being traded layas, Southeast Asia, and Australia are the that amphibians and reptiles are most com-
in the future. Second, for all nontraded species main epicenters of the wildlife trade, contain- monly traded as pets (including species traded
with available phylogenetic information (N = ing areas with the highest numbers of traded as household pets and for expositions, circuses,
29,132), we identified between 303 and 3152 species (i.e., top 5 and 25% richest cells in or zoological gardens), birds are traded both as
species at risk of future trade given their high trade) (Fig. 4 and fig. S7). pets and products (those used for commercial
phylogenetic similarity with conspecifics known Regional differences exist across taxa (Fig. 4, meat, trophy hunting, clothing, medicine, or re-
to be traded. Third, we used a phylogenetic fig. S7, and table S7). For example, in South ligious purposes), whereas mammals are pre-
logistic regression framework to identify which America, the Andes, Atlantic forest, and eastern dominately traded as products (Fig. 5 and table
species are at high risk of future trade based Amazon all contain a high diversity of traded S8). The pet trade flourishes across the tropics,
solely on their body size. Here, we found be- birds, whereas the western and central Amazon whereas species traded as products are con-
tween 11 and 35 species (all mammals) at risk contains a high diversity of traded amphibians. centrated in tropical Africa and Southeast Asia,
of future trade. Our fourth approach used evolu- Although many mammals are traded in South including the Himalayas. Although for birds
tionary distinctiveness, which did not predict America (as revealed by a large area containing and mammals there is a strong association be-
any species at risk of future trade. the top 25% of trade richness), the main hotspots tween the richness of species traded as pets and
In total, on the basis of those species with a for mammal trade are in Africa and Southeast as products, there are important geographical
probability >95 and >90% in any one of the Asia (Fig. 4). The African tropical savanna- differences in these trade types for all verte-
four assessment schemes described above, we woodland belt consists of hotspots for all taxo- brate groups (Fig. 5). For instance, the pet

Fig. 3. Predicted future traded species. Probability of a species being traded in the future based on (A) body size, (B) phylogenetic relatedness,
and (C) the proportion of species traded in respective families. Upper panels show the probability of trade across all currently nontraded species,
and lower panels reflect the probability distribution of trade around the 0.9 and 0.95 confidence intervals.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 73

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

trade of reptiles occurs mostly in Australia and Second, the IUCN Red List, the largest asses- whereas CITES indicates an additional 2029
Madagascar, whereas most amphibians are col- sor of species threat for conservation, needs to traded species omitted by IUCN (fig. S8). In
lected from the Amazon as pets and collected ensure that any evidence of trade is recorded in turn, future IUCN assessments would benefit
from Africa and Southeast Asia as products. species threat accounts, regardless of current from different analytical approaches that incor-
IUCN status. For example, we found that IUCN porate extinction risk from trade [e.g., (21, 27)],
Tackling global wildlife trade indicates 1641 traded species omitted by CITES, as well as increased communication among all
Species possessing rare phenotypes, such as
conspicuous plumage color, body shape and
size, behavior, and/or (perceived) medicinal value
tend to bring high market prices. Trade follows
a rarity-value feedback model, whereby increas-
ing rarity drives both higher demand and higher
prices for a species (22, 24), with this positive
feedback loop shown in both legal and illegal
wildlife trade. For example, in Europe, CITES-
listed pets command a higher price than non–
CITES-listed species (24). Trade also quickly
shifts to conspecifics as the availability of a
targeted species declines, which likely explains
why we uncovered a strong phylogenetic sig-
nal in the trade of all vertebrate groups (fig. S2).
For instance, as Asian pangolin species decline,
they are increasingly replaced in trade by African
pangolins, with strength of demand for African
pangolin meat and scales in Asia now high de-
spite a relative price increase of 211% versus
4.6% baseline inflation (25). On the basis of
identified morphological and phylogenetic cor-
relates of trade, we predict an increase between
5 and 57% (probabilities >95 and >90%, re-
spectively) in the total number of traded verte-
brate species (Fig. 3 and table S5), which amounts
to as many as 8775 species at risk of current
and future trade.
That trade tracks cultural [e.g., the Harry
Potter–inspired trade of owls in Asia (26)] and
economic vogue suggests that abundant spe-
cies may not be safe. Often, species are flagged
for conservation only after a severe decline is
documented [e.g., pangolins (25)]. Our study
offers two possible rectifications of this issue.
First, with the strong predictive strength of
phylogeny revealed in our analysis, we can cir-
cumvent cryptic, yet-to-come declines by flagging
species that are currently of little concern but
have a high likelihood of being traded in the
future because of their evolutionary proxim-
ity to traded species (Fig. 3 and table S5). For
instance, some highly colorful bird groups
at high risk of future trade include Tangara
tanagers (n = 46), Serinus finches (n = 35), and
Ploceus weavers (n = 37), whereas Rhinella
beaked toads (n = 55) and Rhinolophus horse-
shoe bats (n = 55) are the highest-risk amphib-
ian and mammal genera, respectively. Reptiles
yielded the largest number of species at risk
of future trade. Here, Liolaemus iguanian lizards
(n = 229), Atractus (n = 135) and Tantilla (n = 61)
colubrid ground snakes, Bothrops (n = 43) pit
vipers, and Lycodon wolf snakes (n = 48) are all Fig. 4. The geography of wildlife trade in terrestrial vertebrates. Wildlife trade richness increases with
genera at high risk of future trade. We caution, the number of species in a cell for (A) birds, (C) mammals, (E) amphibians, and (G) reptiles. (B, D, F, and H) Wildlife
however, that our identification of a species as trade richness and hotspots of wildlife trade are concentrated in tropical regions. Top 5% and top 25% indicate
potentially traded in the future does not re- areas with the largest number of traded species per cell globally. Points are color coded by the geographic realm.
veal the potential trade volume of this species. Black lines in scatterplots indicate a locally estimated scatterplot smoothing (LOESS) fit.

74 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

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

conservation groups that document and moni-

tor trade (27).
More broadly, our global assessment of wild-
life trade underscores the need for a strategic
plan to combat trade. Evolutionary history’s
ability to predict trade suggests that policies can
be proactive rather than reactive in approach.
Both online black markets and mainstream
marketplaces, such as eBay or Facebook (28),
facilitate a large volume of transactions with
few regulations to stifle trade activity. Advanced
machine-learning computer systems can be
used by vendors to monitor and stem this ac-
tivity (29, 30). Stricter penalties to merchants
of trade, as well as consumer pressure for more
sustainable and cheaper alternatives [e.g., hu-
manely harvested horn from the least-rare
rhino species (31)], may hasten the adoption
of these techniques. Our comprehensive list
of traded and at-risk species can inform these
computerized search systems.
Our global maps of trade hotspots are an
important first step in prioritization. The iden-
tification of many tropical regions as epicenters
of traded species diversity suggests that com-
bating the surge of illegal wildlife trade will
likely require action at the local community level
(32), combined with targeting key countries that
import and export wildlife (33), especially those
countries in hotspot areas that share continuous
borders (34). In many areas, hunting for wildlife
trade occurs out of sheer necessity—that is, it
occurs in impoverished areas where harvesting
wildlife to sell to middlemen represents the
only source of cash income (32). Borrowing
from other programs to halt criminal trading
of humans, arms, and drugs, wildlife trade poli-
cies would gain strength if they were linked to
transnational agreements such as the United
Nations Collaborative Programme on Reduc-
ing Emissions from Deforestation and Forest
Degradation (REDD). This may also offer eco-
nomic incentives for protection rather than
exploitation within local communities. For
instance, carbon-trading schemes could in-
crease the value of carbon in areas that are
combating wildlife trade—with the ecological
co-benefit of areas that maintain large-bodied
vertebrates yielding higher carbon stocks over
the long term (35).

Methods summary
We compiled information on traded birds,
mammals, amphibians, and squamate reptiles
using the CITES list and IUCN Red List. We
identified species traded through the IUCN
application programming interface platform
and classified each as being traded as pets
and/or products (see supplementary materials Fig. 5. Geographical patterns in wildlife trade type across birds, mammals, amphibians, and reptiles.
for details). We superimposed range maps of Pet and product trade richness for (A and B) birds, (C and D) mammals, (E and F) amphibians, and
all species in a 110 km × 110 km global grid and (G and H) reptiles. Pet trade includes species traded as household pets, for expositions, circuses, or
recorded species presence or absence in each zoological gardens. Species traded for products include those used for bush meat, trophy hunting,
cell. We determined total (pet and product) clothing, medicine, or religious purposes. Points are color coded by the geographic realm. Points
trade richness as the number of traded species occurring above the 1:1 equivalency line indicate higher levels of trade as products than as pets.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 75

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

in each cell. We defined hotspots as the upper 34. W. S. Symes, F. L. McGrath, M. Rao, L. R. Carrasco, Biol. Competing interests: None declared. Data and materials
25% and upper 5% richest cells for traded spe- Conserv. 218, 268–276 (2018). availability: All data related to this work are provided in the
35. C. A. Peres, T. Emilio, J. Schietti, S. J. M. Desmoulière, T. Levi, supplementary materials.
cies and assessed the correlation between spa- Proc. Natl. Acad. Sci. U.S.A. 113, 892–897 (2016).
tial patterns in total, traded, and threatened
SUPPLEMENTARY MATERIALS
species richness. ACKN OWLED GMEN TS
science.sciencemag.org/content/366/6461/71/suppl/DC1
We used updated time-calibrated species- We dedicate this paper to all researchers and park guards who
Materials and Methods
have lost their lives in protecting wildlife from illegal trade.
level phylogenetic trees for each vertebrate Funding: B.R.S. received financial support from the UF/IFAS Early
Figs. S1 to S8
Tables S1 to S10
group from which we obtained one maximum Career Seed Grant. Author contributions: B.R.S., B.F.O., and
References (36–61)
clade credibility tree, and we used these trees D.P.E. conceptualized the idea and methodology, I.L. and
B.F.O. collected data, B.F.O. analyzed data, B.R.S. and D.P.E. 25 September 2018; accepted 4 September 2019
in downstream analyses. We tested whether wrote the original draft, and all authors revised the work. 10.1126/science.aav5327
closely related species are traded more than
random ones using the D-statistic. We used
phylogenetic analysis of variance to test whether
traded and nontraded species differ in body size PREBIOTIC CHEMISTRY
and evolutionary distinctiveness, and phyloge-
netic logistic regression to test whether these Unified prebiotically plausible synthesis of pyrimidine
traits influence the probability of a species being
traded. We determined risk of future trade by and purine RNA ribonucleotides
(i) identifying for each nontraded species the
proportion of all species traded in their re- Sidney Becker1,2*, Jonas Feldmann1*, Stefan Wiedemann1*, Hidenori Okamura1,3, Christina Schneider1,
spective family and (ii) for each nontraded Katharina Iwan1,4, Antony Crisp1, Martin Rossa1, Tynchtyk Amatov1,5, Thomas Carell1†
species, averaging its phylogenetic distance with
the 10 closest related species that are traded. Theories about the origin of life require chemical pathways that allow formation of life’s key
building blocks under prebiotically plausible conditions. Complex molecules like RNA must have
RE FE RENCES AND N OT ES
originated from small molecules whose reactivity was guided by physico-chemical processes. RNA is
1. S. L. Pimm et al., Science 344, 1246752 (2014).
2. L. Gibson et al., Nature 478, 378–381 (2011). constructed from purine and pyrimidine nucleosides, both of which are required for accurate
3. S. L. Maxwell, R. A. Fuller, T. M. Brooks, J. E. M. Watson, Nature information transfer, and thus Darwinian evolution. Separate pathways to purines and pyrimidines
536, 143–145 (2016).
have been reported, but their concurrent syntheses remain a challenge. We report the synthesis of the
4. V. Nijman, Biodivers. Conserv. 19, 1101–1114 (2010).
5. TRAFFIC, Wildlife Trade Monitoring Network, Illegal Wildlife pyrimidine nucleosides from small molecules and ribose, driven solely by wet-dry cycles. In the
Trade; www.traffic.org/about-us/illegal-wildlife-trade/. presence of phosphate-containing minerals, 5′-mono- and diphosphates also form selectively
6. L. S. Wyler, P. A. Sheikh, “International illegal trade in wildlife:
in one-pot reactions. The pathway is compatible with purine synthesis, allowing the concurrent
Threats and U.S. policy,” CRS Report for Congress (Library of
Congress, Congressional Research Service, 2008). formation of all Watson-Crick bases.
7. E. L. Bennett, Oryx 45, 476–479 (2011).

T
8. D. S. Wilcove, X. Giam, D. P. Edwards, B. Fisher, L. P. Koh,
Trends Ecol. Evol. 28, 531–540 (2013).
9. D. Challender et al., “Sunda Pangolin: Manis javanica,” The he discovery of catalytic RNA (1) and the is an unsolved chemical problem that is under
IUCN Red List of Threatened Species (2014); www.iucnredlist. development of replicating RNA systems intensive chemical investigation (5–9). Starting
org/species/12763/45222303. (2, 3) have lent strong support to the con- from an early atmosphere mainly composed of
10. C. Beastall, C. R. Shepherd, Y. Hadiprakarsa, D. Martyr, Bird
Conserv. Int. 26, 137–146 (2016).
cept of an RNA world (4). The RNA world N2 and CO2 (10), the abiotic synthesis of life’s
11. N. J. Collar, BirdingASIA 24, 12–17 (2015). hypothesis predicts that life started with building blocks must have occurred on the
12. S. J. O’Hanlon et al., Science 360, 621–627 (2018). RNAs that were able to (self-)recognize and early Earth in aqueous environments, whose
13. R. Engeman, E. Jacobson, M. L. Avery, W. E. Meshaka Jr., Curr.
Zool. 57, 599–612 (2011).
replicate. Through a process of chemical evo- characteristics were determined by the min-
14. BirdLife International, Data zone; http://datazone.birdlife.org/home. lution, a complex RNA and later RNA–peptide erals and chemical elements from which the
15. CITES, The CITES species; www.cites.org/eng/disc/species.php. and protein world supposedly evolved, from early Earth’s crust was made (11, 12). Atmo-
16. S. A. Fritz, A. Purvis, Conserv. Biol. 24, 1042–1051 (2010).
which life ultimately emerged (4). A prerequi- spheric chemistry, impact events, and volcanic
17. L. Santini, M. González-Suárez, C. Rondinini, M. Di Marco,
Divers. Distrib. 23, 640–649 (2017). site for the RNA world is the ability to create activities must have provided the first reactive
18. G. Rapacciuolo et al., Glob. Ecol. Biogeogr. 26, 1022–1034 RNA under prebiotic conditions. This requires small molecules. These reacted in surface or
(2017). as the first elementary step the concurrent deep-sea hydrothermal vents (13–15), on min-
19. D. W. Redding, C. V. DeWolff, A. Ø. Mooers, Conserv. Biol. 24,
1052–1058 (2010). formation of pyrimidine and purine nucleo- eral surfaces (16), or in shallow ponds (17).
20. R. I. Vane-Wright, C. J. Humphries, P. H. Williams, Biol. sides in the same environment. They must Within these environments, volcanic activity,
Conserv. 55, 235–254 (1991). have condensed to form information-carrying and seasonal or day-night cycles caused fluc-
21. W. S. Symes, D. P. Edwards, J. Miettinen, F. E. Rheindt,
L. R. Carrasco, Nat. Commun. 9, 4052 (2018).
polymers able to undergo Darwinian evolu- tuations of pH and temperature. Such fluctua-
22. J. B. C. Harris et al., Conserv. Biol. 31, 394–405 (2017). tion. The question of how the pyrimidine and tions provided wet-dry conditions allowing
23. A. Benítez-López et al., Science 356, 180–183 (2017). purine nucleosides could have formed together precipitation or crystallization of chemicals
24. F. Courchamp et al., PLOS Biol. 4, e415 (2006). (18). Mixing of microenvironments may have
25. M. M. Mambeya et al., Afr. J. Ecol. 56, 601–609 (2018).
26. V. Nijman, K. A.-I. Nekaris, Glob. Ecol. Conserv. 11, 84–94 (2017). 1
opened up new reaction pathways that led to
Center for Integrated Protein Science, Department of Chemistry,
27. E. G. Frank, D. S. Wilcove, Science 363, 686–688 (2019).
LMU München, Butenandtstrasse 5-13, 81377 München, Germany.
increasing chemical complexity.
28. J. R. Harrison, D. L. Roberts, J. Hernandez-Castro, Conserv.
Biol. 30, 900–904 (2016).
2
Department of Chemistry, University of Cambridge, Lensfield Along these geophysical boundaries, two
29. J. Hernandez-Castro, D. L. Roberts, PeerJ Comput. Sci. 1, e10 Road, Cambridge CB2 1EW, UK. 3Institute for Multidisciplinary main reaction pathways have been proposed
(2015). Research for Advanced Materials, Tohoku University, 2-1-1
for the formation of purine and pyrimidine
30. E. Di Minin, C. Fink, T. Hiippala, H. Tenkanen, Conserv. Biol. 33, Katahira, Aoba-ku, Sendai, Miyagi 980-8577, Japan. 4Centre for
210–213 (2019). Translational Omics, University College London, Great Ormond nucleosides. The synthesis of the purines is
31. N. Hanley, O. Sheremet, M. Bozzola, D. C. MacMillan, Conserv. Street Institute of Child Health, 30 Guilford Street, London possible along a continuous pathway based on
Lett. 11, e12417 (2018). WC1N 1EH, UK. 5Max‐Planck‐Institut für Kohlenforschung, the reaction of formamidopyrimidine (FaPy)
32. R. Cooney et al., Conserv. Lett. 10, 367–374 (2017). Kaiser‐Wilhelm‐Platz 1, 45470 Mülheim an der Ruhr, Germany.
33. N. G. Patel et al., Proc. Natl. Acad. Sci. U.S.A. 112, 7948–7953 *These authors contributed equally to this work. precursors with ribose (6, 18). For the pyrimidines,
(2015). †Corresponding author. Email: thomas.carell@lmu.de a reaction sequence involving aminooxazoles

76 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

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

A prebiotically
plausible molecules
O pyrimidines
O
OH OH O O
H H H2N O R
OH N H2 N NH2 NH2 ribose ribose HO OH HO OH
R1 R2 HN N N
NH2OH N 5 9 HN H HO O
N HO O
1 O N N O
2: R1 = R2 = H 4 N
HCN O
O ribose N N
O 3: R1 = H, R2 = CONH2 8
10a-d H
NO2 -
6: R1 = R2 = SO3- α/β-cytidine (R = NH2), 11a/b O N O N N O
H2N NH2 unified synthesis α/β-uridine (R = OH), 12a/b H H
H H H
N N N O N N H
O N
N O H
O R1 R1
H2N N 19 O R1 R1 ribose N N
Δ N HN HN N N N
9 N N
N N N N N N
HSO3- NH2 HN N R2 N N R2
H2N N R2 H2N N R2
ribose ribose O OH O OH
H2 N R
O 21 22-25 26 27-29
20 HO OH HO OH
OH

21, 24: R1 = NH2, R2 = SMe 23, 28: R1 = OH, R2 = NH2 purines

22, 27: R1 = NH2, R2 = H 25, 29: R1 = NH2, R2 = NH2

B
Cu(II)
N2 + CH4 N HCN +
(20) 1 (13, 21)

H2 O
SO2 HSO3-
-
O 3S SO3- O
N
H 2O 2 H2O NH2OH HNCO
CO2 + N2 NO NO 2
- OH H2N NH
(17, 24) (25, 26) 6 - 2 HSO4- 2 (22) 3 OH
Fe(II) (26)
(28)
O O
+ HNCO
CO2 + H2O OH NH3 NH2
H H SO3- NH2
2 H2O H2O
NC O N O
O
atmospheric SO3- - 2 HSO4- N N
gases carbohydrates H2N NH2 7 O O NH2
5 (Z)-isomer only 4

Fig. 1. Unified synthesis of pyrimidine and purine RNA building blocks. that produces the molecules needed for the pyrimidine pathway. Reactions performed
(A) Starting from plausible prebiotic molecules, the reaction scheme depicts the route in this work are shown with green arrows, while black arrows represent well-known
toward the pyrimdines via isoxazolylurea 8 (blue background) and the purines via reactions. Formation of 4 requires reaction of 1 with hydroxylamine 2, hydroxylurea
formamidopyrimidines 22 to 25 (red background) (18). (B) Fundamental chemistry 3, or the disulfonate 6 (dark-gray box). 6 is formed from NO2− and SO2/HSO3−.

has been discovered (5). These pathways pro- block (Fig. 1A). Compound 1 is observed in metal ions (see below). Reaction of cyanoa-
vide the corresponding nucleosides under interstellar clouds and in the atmosphere of cetylene 1 with hydroxylamine 2 produced
very different and partially incompatible con- Titan (19). It has been shown to form in large 4 with 17% yield after 2 hours at pH 10.
ditions, leaving unanswered the question of quantities by electric discharge through a While hydroxylamine 2 is an accepted build-
how purines and pyrimidines could have CH4-N2 atmosphere (20) and is also a product ing block for prebiotic amino acid syntheses
formed in the same environment. Here, we of the Cu(II)-mediated reaction of HCN and (23), its potential formation on the early Earth
report a prebiotically plausible pathway to acetylene in water (Fig. 1B) (21). A recent is unclear. We therefore aimed to demonstrate
pyrimidine nucleosides that selectively pro- report suggested that molecules such as 1 are its prebiotic availability. 2 is ultimately pro-
vides the 5′-mono- and 5′-diphosphorylated plausible prebiotic starting materials which duced by reduction from NO, which is formed
nucleosides needed for RNA strand forma- could have formed in surface hydrothermal in large quantities when lightning passes
tion. By connecting the pathway with the vents in significant concentrations (13). We through moist atmospheres containing N2
reported purine route (6, 18), we establish a found that 1 reacts quickly and cleanly with and CO2 (Fig. 1B) (10). NO forms as the main
unifying reaction network that allows for the hydroxylamine 2 or hydroxylurea 3 to give product under these conditions and sponta-
simultaneous formation of both types of nu- 3-aminoisoxazole 4. The reaction of 1 with neously reacts in the presence of water to form
cleosides in the same environment and that 3 proceeds under slightly basic conditions (pH nitrite (NO2−) and nitrate (NO3−), and this
is driven by wet-dry cycles. ∼10) with 80 to 90% yield within 2 hours. 3 is leads to the assumption that both anions were
formed in almost quantitative yields from the quite abundant on the early Earth (24–26).
Results reaction of 2 with cyanate (22). Compound 4 With Fe(II) as a plausible prebiotic reductant,
Prebiotically plausible synthesis formed robustly even if we varied the tem- NO2− is converted to NH3 but not to NH2OH 2
of pyrimidine nucleosides perature (10° to 95°C), the reactant concentra- (26). Formation of the latter requires a partial
The chemistry leading to pyrimidines starts tions (10 to 100 mM), or added additional reduction. We found that this can be achieved
from cyanoacetylene 1 as the key building compounds, such as urea 5 and/or different with HSO3−, which forms from volcanic SO2

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 77

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

and water (27). NO2− and HSO3− react to 2

with up to 85% yield (Fig. 1B and fig. S1) (28).
We confirmed that this reaction gives first
hydroxylamine disulfonate 6 (Fig. 1B), which
hydrolyzes to hydroxylamine 2 and HSO4−.
We found that intermediate 6 reacts with
cyanoacetylene 1 as well (88% yield) (Fig.
1B and fig. S2) to give the stable olefin 7,
which upon hydrolysis provides again the key
intermediate 4. The overall yield of 4 via
compound 7 is 63% over these two steps. The
suggested pyrimidine intermediate 4 is there-
fore readily available from cyanoacetylene 1
upon reaction with either 2, 3, or 6 under
prebiotic conditions (Fig. 1B).
When we added urea 5 to a solution of 4,
warming (70° to 95°C) and dry-down resulted
in formation of N-isoxazolyl-urea 8 (Figs. 1A
and 2A) in a spot-to-spot reaction that is cat-
alyzed by Zn2+ or Co2+. These metal ions were
likely present on the early Earth (11, 12). In
the presence of Zn2+, compound 8 is formed
in 88% yield after 2 days at 95°C (at 70°C, the
same yield is obtained after ∼2 to 3 weeks).
With Co2+, 68% yield is achieved after 2 days
at 95°C. The reaction of 4 to 8 is in all cases
a clean process, with the only impurity being
unreacted 4 (Fig. 2A). The product 8 can be
subsequently physically enriched. Addition of
carbonated water to the dried reaction mix-
ture solubilizes 4, 5, and 8, leaving behind
the metal ions as hydroxides or carbonates.
Subsequent concentration of the supernatant
leads to spontaneous crystallization of 8 (55%).
This allowed us to obtain a crystal structure
of 8 (fig. S3). In order to simulate early Earth
chemistry, we performed a one-pot experi-
ment. We mixed 1 with 3, 5, and Zn2+ or Co2+
in a carbonate solution (pH ∼10) and obtained
compound 4 at 95°C (80 to 90%). Neutraliz-
ing the solution to pH ∼6 to 7, which may
have occurred on the early Earth as a result
of acidic rain, followed by dry-down at the
same temperature, provided compound 8 with
yields between 56% (Zn2+) and 40% (Co2+).
The continuous synthesis of the key building
block 8 was consequently achieved in a plau-
sible prebiotic setting that could have existed
in hydrothermal vents or near volcanic activ-
ity, both of which would be able to provide
elevated temperatures (fig. S3). The synthesis
is also possible at lower temperatures, but
with extended reaction times.
For the final step toward nucleosides, we
need to assume that, due to flooding or a
Fig. 2. Formation of pyrimidine nucleosides (11 and 12) from N-isoxazolylurea ribosides mixing of environments, 8 came into contact
10a and 10b. The different isomers are labeled as follows: a = a-furanosyl; b = b-furanosyl; c = with ribose 9 (Figs. 1A and 2B) or any other
a-pyranosyl; d = b-pyranosyl. (A) Formation of 4 and its conversion with urea 5 to N-isoxazolylurea 8. sugar unit, such as threose (for TNA) or glyc-
(B) Ribosylation of 8 with ribose 9 and equilibration of the reaction mixture in the presence of eraldehyde (for GNA), that was able to form a
borates gives the furanosidic isomers 10a and 10b (54%). (C) Pyrimidine nucleoside formation by backbone for a pairing system (29, 30). When
reductive N-O cleavage from the compound mixture of 10a and 10b in the presence of ammonium we mixed 8 with ribose 9 and warmed the
iron(II) sulfate hexahydrate (0.0005 equiv.). The HPLC results with detection at 260 nm show formation mixture to 95°C in the presence of boric acid,
of cytidine (C; 11a to 11d) and uridine (U; 12a and 12b). (D) Proposed catalytic cycle for the Fe2+ we observed a fast and high-yielding reaction
catalyzed reduction of the N-O bond of the isoxazole moiety. that provided the ribosylated products 10a

78 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

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

to 10d with 95% yield (fig. S4a). Other bo- Just 0.0001 equiv. of soluble Fe2+ in water is We assumed that the reaction generated the
rate minerals, such as synthetic lüneburgite sufficient for the reaction. In the absence of Fe2+, a- and b-cytidine 5′-monophosphates 13a and
{Mg3[(PO4)2|B2(OH)6]·6H2O} (31) or borax pyrimidine formation was not observed. The 13b and the 5′-diphosphorylated cytidines 14a
{Na2[B4O5(OH)4]·10H2O} (32), were also able to reduction also appears to be independent of the and 14b. Owing to hydrolysis, we also expected
catalyze this reaction with high yields (>70%) thiol source, as the products 11a to 11d and 12a some a- and b-uridine 5′-monophosphates and
(fig. S5). The major products were initially and 12b were obtained regardless of whether 5′-diphosphates 15a and 15b and 16a and 16b.
the a- and b-pyranosides (10c and 10d), which we used dithiothreitol (DTT), propanedithiol, We isolated the corresponding high-performance
dominate over the a- and b-furanosides (10a mercaptoethanol, or cysteine (fig. S6). LC (HPLC) peaks and removed the phosphate
and 10b) (fig. S4a). After heating the mixture groups enzymatically (Fig. 3, B and C). LC-MS
under slightly basic conditions at 95°C in the Selective one-pot formation of analysis showed the dephosphorylated furano-
presence of borates, the furanosides (54%; 5′-nucleoside mono- and diphosphates sides 11a and 11b and 12a and 12b with over
10a and 10b) (Fig. 2B) gradually became the The addition of naturally occurring minerals 94% in the nucleoside pool, which corresponds
dominant products (fig. S4b). Under these such as hydroxyapatite, colemanite, or (syn- to a change of the furanoside/pyranoside ratio
conditions, we also observed hydrolysis of 10a thetic) lüneburgite to the reductive pyrimidine- from an initial 4:1 to 17:1 (Fig. 3C). The formation
to 10d to 8 and 9. The accumulation of the forming reaction mixture had a strong influence of phosphorylated pyranosides 17 are only a
furanosides 10a and 10b is best explained by on the distribution of the four cytidine isomers. minor side reaction. We found no discrimina-
complexation of their cis-diols with borate (32). Synthetic lüneburgite gave a combined high tion between a- and b-anomers during the
The final step toward pyrimidine nucleosides yield of 85% (Fig. 2C). The natural furanosidic phosphorylation. The furanoside enrichment
requires reductive opening of the isoxazole b-cytidine (11b) and its a-anomer (11a) are is best explained by the presence of a primary
N-O bond, followed by tautomerization, intra- formed under these conditions with about the hydroxyl group in the furanosides which is
molecular cyclization, and water elimination same yields, together with small amounts of absent in the pyranosides. The enrichment of
in a cascade-like fashion (Fig. 2, C and D). We a- and b-uridine (12a and 12b). We found only 5′-nucleoside monophosphates and diphos-
found that this reaction occurred rapidly with small amounts of the a- and b-cytidine pyran- phates under these one-pot conditions estab-
Fe2+ in the presence of thiols (Fig. 2D) (33). osides (11c and 11d), together with the cyto- lishes a further chemical selection step that
Liquid chromatography–mass spectrometry sine base. Because synthetic lüneburgite is favors the furanosides as the components of
(LC-MS) analysis indicated that cytidine nu- known to enable nucleotide formation in the RNA. We further characterized the structures
cleosides 11a to 11d formed efficiently under presence of urea (Fig. 3A) (31), we simply added of the phosphorylated nucleosides and con-
these conditions, with the furanosidic uridine urea to the one-pot reaction mixture after py- firmed the formation of the 5′-a- and 5′-
nucleosides 12a and 12b as the corresponding rimidine formation and allowed the mixture to b-cytidine mono- and diphosphates (13a, 13b,
deamination products formed by hydrolysis evaporate to dryness at 85°C over a period of and 14a, 14b; a-/b-CMP and a-/b-CDP) (fig.
(Fig. 2C). Reductive pyrimidine formation can about 20 hours. LC-MS analysis of the reaction S8). Additional analysis allowed identification
be performed with FeS or the mineral pyrite mixture showed formation of phosphorylated of a,b-UDP 16a and 16b (fig. S9). 5′-Pyrophos-
(FeS2), and both have been discussed in the nucleosides (Fig. 3A) in a substantial 19% yield phates are the dominating species within the
context of early metabolic pathways (15, 34). relative to that for cytidine (Fig. 3B and fig. S7). diphosphorylated nucleoside mixture (fig. S8a).

Fig. 3. One-pot nucleotide formation reaction. (A) One-pot synthesis of cytidine analysis of the corresponding nucleotide peaks with UV and MS detection and iso-
and uridine 5′-mono- and 5′-diphosphates (13a and 13b to 16a and 16b) after lation of the formed nucleotides from the prebiotic reaction, followed by an enzymatic
urea addition to the reaction mixture and allowing the mixture to dry-down at removal of the phosphate groups. (C) HPLC analysis of the dephosphorylated
85°C for 20 hours. a/b represent the a- and b-anomers, respectively. (B) LC-MS product mixture showing predominant formation of a- and b-cytidine 11a and 11b.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 79

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

boiling point (228°C). Compound 4 can act as

a solvent to facilitate the formation of 21 from
the reaction of 19 with 20 under milder con-
ditions (50°C to 100°C instead of 126°C), in
contrast with results from a previous exper-
iment (18). The next step requires reduction
and formylation of 21 to the FaPy interme-
diate, but this step cannot be performed in
the presence of the isoxazole. Addition of a
water mixture eventually containing urea 5
leads to spontaneous precipitation of 21. The
supernatant containing 4 and 5 can flow away.
The water-insoluble 21, if brought into contact
with dilute formic acid and Zn (found in Earth’s
crust), reacts immediately to form the com-
pounds 22 and 24 with Zn2+ as a side product
(Fig. 5A and fig. S10a). These reaction products
are water-soluble and can potentially recom-
bine with 4 and 5. The side product Zn2+ can
then catalyze the reaction of 4 in the presence
of 5 to give N-isoxazolyl urea 8 in the presence
of 22 and 24 (Fig. 5A and fig. S10b). This leads
to the formation of the pyrimidine and purine
precursors 8, 22, and 24, which can be trans-
formed into the purine and pyrimidine nucle-
osides. In this scenario, intermediate 4 of the
pyrimidine pathway helps formation of the pu-
rine precursor 21, while Zn2+ as a side product
of the purine pathway mediates formation
of the pyrimidine precursor 8 in a mutually
synergistic way, driven by wet-dry cycles.
We combined 8 with different FaPy inter-
mediates and investigated if they reacted in a
one-pot scenario with ribose 9 to finally give
the purine and pyrimidine nucleosides. To ex-
amine this, we dissolved a mixture of 8, 22,
Fig. 4. Formation of all four Watson-Crick RNA building blocks in identical but parallel reactions.
25, ribose 9, and boric acid and warmed the
C (11b), U (12b), A (27b), and G (28b) are formed under the same conditions separately from 8, 22, and
mixture to 95°C for 14 hours, allowing for
23. HPLC results are shown with a detection at 260 nm. The nucleosides are labeled as follows: a =
slow evaporation of water. The solid material
a-furanosyl; b = b-furanosyl; c = a-pyranosyl; d = b-pyranosyl. Canonical pyrimidine and purine RNA building
was then taken up with a slightly basic solu-
blocks are labeled in blue and red, respectively.
tion containing Fe2+ (0.0005 equiv.) and DTT
(1.5 equiv.), and we allowed the mixture to
Compatible formation of pyrimidine and purine pyrimidine formation from the urea-isoxazole warm to 95°C. HPLC-MS analysis proved that
RNA nucleosides 8 would affect purine formation, we reacted these conditions simultaneously provided the
We next investigated if the prebiotically plau- 8 and the FaPy compounds 22 and 23 with purine and pyrimidine nucleosides with cyti-
sible pyrimidine and purine nucleoside path- ribose 9 under dry-down conditions. We per- dine (11a to 11d) and adenosine (27) as the
ways are compatible with each other so that formed the reaction under identical conditions main products. Diaminopurine nucleosides
they can be connected with the goal to form but in separate reaction vials (Fig. 4). Under (DA; 29), which hydrolyze to guanosine 28,
all Watson-Crick building blocks in the same these conditions, formation of all four Watson- form in this one-pot reaction as well (Fig. 5A,
environment, driven solely by wet-dry cycles. Crick nucleosides, cytidine 11, uridine 12, aden- chromatogram). We noted additional forma-
The purine synthesis (18) requires as the initial osine 27, and guanosine 28, were detected. tion of double-ribosylated adenine (rib2-A).
step a reaction of malononitrile 18 with sodium We next investigated if pyrimidines and pu- Furthermore, the nucleoside 28 was created
nitrite to give (hydroxyimino)malononitrile rines can form simultaneously in the same in this scenario when we used 23 (R1 = OH,
19. Because malononitrile 18 can be also gen- environment (Fig. 5A). For this experiment, R2 = NH2) as the starting material, but the
erated from cyanoacetylene 1, as shown by we mixed the starting materials cyanoacetylene yields were lower.
Eschenmoser (35), pyrimidines and purines 1, hydroxylurea 3, (hydroxyimino)malononitrile
can be traced back to the same chemical root 19, and amidine 20 under slightly basic con- Discussion
(Fig. 4). Compound 19 forms an organic salt ditions (pH ∼10). Analysis of the mixture indeed Ribose-based RNA and the four canonical nu-
with amidines 20 to give nitroso-pyrimidines showed formation of 4 with 86% yield, despite cleosides, A, G, C, and U, are central to modern
21 and, upon reduction and formylation, FaPys the presence of 19 and 20. It is surprising that life and to prebiotic hypotheses, such as the
(22 to 25). The latter can react with ribose 9 the N-OH functionality of compound 19 does “RNA world,” in which RNA strands repli-
to give ribosylated FaPy 26 and then purine not interfere with the formation of 4. Com- cated and evolved to give increasingly com-
nucleosides 27 to 29 (Fig. 1A) (18). To inves- pound 4 is a liquid that can enrich from a plex chemical systems (4). Whether such RNAs
tigate how the chemical conditions needed for water solution by dry-down, owing to its high were directly assembled from the canonical

80 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

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

Fig. 5. Unified chemical scenario for the formation of purine and pyrimi- driven by wet-dry cycles. In yellow, the solvent is 3-aminoisoxazole (4), which
dine nucleosides. (A) Depiction of the connected reaction pathways to can be enriched from an aqueous solution due to its high boiling point
pyrimidine and purine nucleosides, together with the HPLC analysis (260 nm) (228°C). 2-(Methylthio)-5-nitrosopyrimidine-4,6-diamine (21) is a general pre-
of the final reaction mixtures. Nucleosides are labeled as follows: a = a-furanosyl; cursor for adenosine and guanosine (18). Compounds 8, 22, and 24 are
b = b-furanosyl; c = a-pyranosyl; d = b-pyranosyl. (B) Proposed geochemical accessible in the same pot, and they can react with ribose to the RNA
scenario for the simultaneous synthesis of purine and pyrimidine nucleosides, nucleosides in a one-pot reaction.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 81

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

nucleotides (A, C, G, and U bases) or if they Our model assumes a surface environment, 13. P. B. Rimmer, O. Shorttle, Life 9, 12 (2019).
evolved from a simpler proto-RNA system is where molecules such as NO2−, HSO3−, or urea 14. W. Martin, J. Baross, D. Kelley, M. J. Russell, Nat. Rev.
Microbiol. 6, 805–814 (2008).
unclear (36). 5 could have been delivered by rain after their 15. E. Camprubi, S. F. Jordan, R. Vasiliadou, N. Lane, IUBMB Life
Here we show that a reaction network toward formation in the atmosphere (Fig. 5B) (25, 37). 69, 373–381 (2017).
the purine and pyrimidine RNA building blocks Our chemistry shows that robust reaction 16. S. A. Benner, H.-J. Kim, E. Biondi, in Prebiotic Chemistry and
Chemical Evolution of Nucleic Acids, C. Menor-Salv‡n, Ed.
can be established, starting from simple atmo- networks can be established that allow all (Springer International Publishing, Cham, 2018), pp. 31–83.
spheric or volcanic molecules. Molecular com- key intermediates to be generated efficiently 17. S. Ranjan, Z. R. Todd, P. B. Rimmer, D. D. Sasselov, A. R. Babbin,
plexity is generated by wet-dry cycles that can from relatively complex mixtures, followed by Geochem. Geophys. Geosyst. 20, 2021–2039 (2019).
18. S. Becker et al., Nat. Commun. 9, 163 (2018).
drive the chemical transformations. Therefore, their physical enrichment or separation on 19. P. Thaddeus, Philos. Trans. R. Soc. Lond. B Biol. Sci. 361,
any environment that was able to provide wet- the basis of their solubility in water. Wet-dry 1681–1687 (2006).
dry phases might have been a suitable place for cycles govern the formation of purine and 20. R. A. Sanchez, J. P. Ferris, L. E. Orgel, Science 154, 784–785 (1966).
21. B. H. Patel, C. Percivalle, D. J. Ritson, C. D. Duffy,
the origin of RNA building blocks. Our geo- pyrimidine RNA building blocks in a scenario
J. D. Sutherland, Nat. Chem. 7, 301–307 (2015).
chemical model assumes that chemistry took depicted in Fig. 5B. Of course, we will be un- 22. H. Kofod, B. Wickberg, A. Kj¾r, Acta Chem. Scand. 7, 274–279
place in several basins that were needed to able to definitively prove that the described (1953).
locally separate intermediates. We also needed scenario indeed took place on early Earth, but 23. K. B. Muchowska, S. J. Varma, J. Moran, Nature 569, 104–107
(2019).
one or two streams of water in our system to the reported chemistry shows that, under plau- 24. V. S. Airapetian, A. Glocer, G. Gronoff, E. HŽbrard, W. Danchi,
allow exchange of soluble molecules (Fig. 5B). sible prebiotic conditions, mutually synergistic Nat. Geosci. 9, 452–455 (2016).
Intermediates might precipitate upon fluctua- reaction pathways can be established in which 25. H. J. Cleaves, J. H. Chalmers, A. Lazcano, S. L. Miller, J. L. Bada,
Orig. Life Evol. Biosph. 38, 105–115 (2008).
tions of physico-chemical parameters, allow- the intermediates along one pathway help the 26. D. P. Summers, S. Chang, Nature 365, 630–633 (1993).
ing for the separation of soluble and insoluble chemistry of the other. In such a scenario, we 27. S. Ranjan, Z. R. Todd, J. D. Sutherland, D. D. Sasselov,
materials (e.g., 4 and 21). After further re- show that the key building blocks of life can be Astrobiology 18, 1023–1040 (2018).
28. G. K. Rollefson, C. F. Oldershaw, J. Am. Chem. Soc. 54,
actions, which reestablish solubility, the com- created without the need for sophisticated 977–979 (1932).
pounds can be recombined (Fig. 5B). For our isolation and purification procedures of reac- 29. G. F. Joyce, Nature 418, 214–221 (2002).
scenario we need to assume that the early tion intermediates that are common in tradi- 30. D. M. Fialho et al., Org. Biomol. Chem. 16, 1263–1271 (2018).
31. H.-J. Kim et al., Angew. Chem. Int. Ed. 55, 15816–15820 (2016).
Earth provided environmental conditions that tional organic chemistry. 32. A. Ricardo, M. A. Carrigan, A. N. Olcott, S. A. Benner, Science
fluctuated between slightly acidic (pH 3), po- The concurrent formation of pyrimidine 303, 196–196 (2004).
tentially caused by acidic rain (SO2, NOx), or and purine nucleosides in the network can be 33. M. Kijima, Y. Nambu, T. Endo, J. Org. Chem. 50, 1140–1142 (1985).
34. G. WŠchtershŠuser, Microbiol. Rev. 52, 452–484 (1988).
basic (pH 10) caused by carbonates. Even traced to just a few key starting molecules, 35. U. Trinks, A. Eschenmoser, ETH Zurich (1987); doi: 10.3929/
though most of the chemistry described here such as cyanoacetylene 1, NH3, NH2OH 2 (or ethz-a-000413538.
was performed at elevated temperatures, the the disulfonate 6), HCN, urea 5, formic acid, 36. R. Krishnamurthy, Isr. J. Chem. 55, 837–850 (2015).
37. J. Liebig, F. Wšhler, Ann. Phys. 96, 369–400 (1830).
reactions also occur at lower temperatures, and isocyanate, plus salts such as nitrites, 38. M. Preiner et al., bioRxiv 682955 (2019).
but with substantially longer reaction times. carbonates, and borates. Metals such as Zn 39. C. Bonfio et al., Nat. Catal. 1, 616–623 (2018).
We can assume that temperatures fluctuated or Fe and their ions play an important role in 40. J. S. Teichert, F. M. Kruse, O. Trapp, Angew. Chem. Int. Ed. 58,
9944–9947 (2019).
on the early Earth just like today due to day- our chemistry, consistent with their proposed 41. J. Kofoed, J.-L. Reymond, T. Darbre, Org. Biomol. Chem. 3,
night or seasonal cycles. Such fluctuations involvement in early metabolic cycles (23, 38). 1850–1855 (2005).
would certainly have brought about wet-dry In particular, iron-sulfur surfaces needed for 42. C. Meinert et al., Science 352, 208–212 (2016).
43. K. Usami, A. Okamoto, Org. Biomol. Chem. 15, 8888–8893 (2017).
cycles, akin to modern droughts and rain. The pyrimidine formation have been discussed
geophysical requirements needed for the re- as platforms for early prebiotic chemistry AC KNOWLED GME NTS
ported chemistry, including elevated temper- (15, 34, 39). The 5′-(di)phosphorylation is in- We thank J. Kampmann for x-ray diffraction measurements
atures, could have existed in geothermal fields tegrated into our pathway if phosphate min- and S. Balasubramanian for supporting S.B. during the revision
of the manuscript. Funding: Deutsche Forschungsgemeinschaft
or at surface hydrothermal vents, which are erals such as lüneburgite or struvite (figs. S11 to
(DFG) provided financial support via the programs SFB1309
plausible geological environments on early S13) are present. It remains unclear, however, (TP-A4), SFB749 (TP-A4), SPP-1784, GRK2062/1, and CA275/
Earth. how ribose or any other carbohydrate, such as 11-1, the Excellence Cluster EXC114, the European Research
Our proposed chemical pathways toward glycerol or threose, that is needed to form the Council (ERC) under the European Union's Horizon 2020
research and innovation program (grant agreement EPiR
pyrimidines and purines begin with cyano- backbone of RNA or pre-RNA could have formed 741912), and the Volkswagen Foundation (Initiative “Life”:
acetylene 1, which could have formed in surface selectively (29, 40). Sugars such as ribose can EcoRib). H.O. thanks the European Commission for a Marie
hydrothermal vents (13). Reaction of 2, 3, be produced nonselectively in a formose-like Sklodowska-Curie postdoctoral fellowship (PRENUCRNA).
Author contributions: T.C. designed and supervised research;
or 6 with 1 is the starting point for the py- reaction, which is possible in a variety of differ- S.B. helped to design the study, S.B., J.F., S.W., and H.O.
rimidines, but if 1 reacts instead with ammo- ent physico-chemical environments (32, 41–43). performed the experiments. C.S., M.R., and A.C. supported
nia, a pathway to malononitrile 18 as the the synthesis and MS quantification, K.I. performed biochemical
RE FERENCES AND NOTES studies, and T.A. helped to design the synthesis. T.C., S.B.,
precursor for purine synthesis is possible (Fig. J.F., and S.W. analyzed data. T.C. and S.B. wrote the manuscript
4) (35). Another key molecule for the synthesis 1. J. A. Doudna, T. R. Cech, Nature 418, 222–228 (2002). and designed the figures. Competing interests: The authors
2. D. P. Horning, G. F. Joyce, Proc. Natl. Acad. Sci. U.S.A. 113,
of purines and pyrimidines is NO2−, which is 9786–9791 (2016).
declare no competing interests. Data and materials
availability: The x-ray crystallographic data for isoxazoleurea
needed to nitrosate malononitrile 18 to 19 (18). 3. J. Attwater, A. Raguram, A. S. Morgunov, E. Gianni, P. Holliger,
8 are deposited in the CCDC under accession number
NO2− is also crucial for the formation of hy- eLife 7, e35255 (2018).
1889652. All other data needed to support the conclusions
4. W. Gilbert, Nature 319, 618 (1986).
droxylamine in the presence of HSO3−, which 5. M. W. Powner, B. Gerland, J. D. Sutherland, Nature 459,
of this manuscript are included in the main text and
supporting material.
is formed from volcanic SO2 (27). The concen- 239–242 (2009).
tration of NO2− that is reachable in a prebiotic 6. S. Becker et al., Science 352, 833–836 (2016). SUPPLEMENTARY MATERIALS
7. H.-J. Kim, S. A. Benner, Proc. Natl. Acad. Sci. U.S.A. 114,
setting is under debate, but it is speculated science.sciencemag.org/content/366/6461/76/suppl/DC1
11315–11320 (2017).
that the most likely place for its accumulation Materials and Methods
8. S. Stairs et al., Nat. Commun. 8, 15270 (2017).
Figs. S1 to S13
is in shallow ponds, as needed for our scenario 9. R. Saladino et al., Proc. Natl. Acad. Sci. U.S.A. 112,
References (44–51)
(17). In general, the limited stability of NO2− E2746–E2755 (2015).
10. J. F. Kasting, Science 259, 920–926 (1993). 8 March 2019; resubmitted 21 June 2019
would not be an issue, provided that it is 11. R. M. Hazen, Am. J. Sci. 313, 807–843 (2013). Accepted 21 August 2019
rapidly captured by HSO3− upon its formation. 12. E. D. Swanner et al., Earth Planet. Sci. Lett. 390, 253–263 (2014). 10.1126/science.aax2747

82 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

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

SONG LEARNING interfacialis of the nidopallium (NIf) is the

single largest source of auditory input to
Inception of memories that guide vocal learning in HVC and has also been implicated in song
learning (7, 23–26). Disruption of activity in
the songbird NIf or HVC during tutoring experiences dis-
rupts encoding of tutor song behavioral-goal
Wenchan Zhao, Francisco Garcia-Oscos, Daniel Dinh*, Todd F. Roberts† memories (7).
Neurons in NIf mark the beginning and
Animals learn many complex behaviors by emulating the behavior of more experienced individuals. This ending of song elements with sharp increases
essential, yet still poorly understood, form of learning relies on the ability to encode lasting memories and decreases in their activity during singing
of observed behaviors. We identified a vocal-motor pathway in the zebra finch where memories that (25), suggesting a potentially simple neural
guide learning of song-element durations can be implanted. Activation of synapses in this pathway mechanism for marking the duration of vocal
seeds memories that guide learning of song-element duration and can override learning from social elements in song. Moreover, HVC has been
interactions with other individuals. Genetic lesions of this circuit after memory formation, however, do not broadly implicated in the motor control of
disrupt subsequent song imitation, which suggests that these memories are stored at downstream song temporal structure (18). However, it is
synapses. Thus, activity at these sensorimotor synapses can bypass learning from auditory and social not known how behavioral-goal memories for
experience and embed memories that guide learning of song timing. timing or duration of a tutor’s song elements
are encoded in the brain of a young pupil or if

W
NIf and HVC are specifically involved. We used
e learn to emulate many social and elemental circuits capable of encoding behav- optogenetics to manipulate activity at NIf-HVC
communicative behaviors with seem- ioral-goal memories in the young male zebra synapses in juvenile birds to test whether this
ingly minimal effort. A wide range finch, a songbird that learns its adult song by manipulation could implant memories capa-
of behaviors, including those related observing, memorizing, and then slowly learn- ble of bypassing auditory and social experience
to speech and language, are initially ing to copy the singing behavior of its song from a vocal model and guide learning of song
learned by observing the behavior of teach- “tutor” during a developmental sensitive period temporal structure.
ers or other more experienced social models. (Fig. 1, A and B).
Long-term memories of observed behaviors Adult zebra finch song is well defined by its Opto-tutoring in young birds shapes the
can guide procedural learning by providing temporal and spectral features, both of which temporal structure of their adult song
internal benchmarks for evaluating the quality are learned from song tutors and can be adapt- We established methods to selectively manipulate
of future performances (1–4). Unlike episodic ively modified (16–21). Yet the manner in which NIf’s inputs to HVC in juvenile male zebra finches
memories, memories used to guide imitation— auditory signals are engaged to form memories using an axon-targeted channelrhodopsin-2
referred to here as behavioral-goal memories— of specific temporal or spectral features of tutor
are not thought to require the hippocampus, songs is still poorly understood (4). A premotor
and instead are thought to be directly encoded pallial region necessary for song production, Department of Neuroscience, UT Southwestern Medical
in cortical circuits (2, 5–15). However, the syn- HVC, has been implicated in learning from Center, Dallas, TX, USA.
*Present address: UC Davis School of Medicine, Sacramento, CA
aptic pathways encoding these memories auditory experience with a tutor (7, 8, 11, 12, 22) 95817, USA.
have yet to be identified. We sought to define (Fig. 1C). The pallial sensorimotor nucleus †Corresponding author. Email: todd.roberts@utsouthwestern.edu

A C
hatched sensory learning sensorimotor learning stable adult song
(0 dph) (~20 - 60 dph) (~40 - 90 dph) (~ 90 dph
ph onward) HVC
Av
NCM CM
NIf
Field L

RA lMAN

adult juvenile juvenile/adolescent

memorize temporal and practice song to use song to attract Ov
Uva Area X
spectral properties of tutor song imitate tutor mate
auditory DLM
B 11
motor/auditory
Hz song

0
i i i i i A B C D E F

Fig. 1. Overview of song learning and neural circuits for song. (A) Timeline nucleus avalanche; CM, caudal mesopallium; DLM, dorsolateral thalamic
for song learning in juvenile male zebra finches. (B) Spectrogram of an adult nucleus; field L, primary auditory forebrain; HVC, premotor song nucleus; lMAN,
zebra finch song; introductory notes (i) and individual elements in the song (A to lateral magnocellular nucleus of the anterior nidopallium; NCM, caudomedial
F) are noted. Scale bar, 50 ms. (C) Parasagittal schematic of auditory (blue) and nidopallium; NIf, nucleus interfacialis of the nidopallium; Ov, thalamic nucleus
song motor circuits (orange). Area X, striato-pallidal basal ganglia nucleus; Av, ovoidalis; Uva, nucleus uvaeformis; RA, robust nucleus of the arcopallium.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 83

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

A B C 75

oEPSC (pA)
HVC 50
NIf
HVC 25

10pA
0
NIf n = 10 n = 10 n = 10
20ms

2
P5
L- X
hr

hr
D Q
A
+C

-C
& DN
+
D E F

HVC
NIf

30 15
Spikes/bin

0 Time (s) 8 0 Time (s) 8

G H I 800 NIf HVC J 80

NIfAv
Av

%neurons/hem
NIfHVC/Av

#neurons/hem
HVC
VC HVC 600 60
Av NIf
NIf 400 40
?
? 200 20
NIf
0 0
n = 11 hemispheres from 7 birds
K L
HVC NIf (1.9mm) NIf (2.1mm) NIf (2.3mm)
HVC
NIf

Fig. 2. Selective manipulation of the NIf-HVC pathway. (A) Left: Schematic recordings in HVC [top, representative single trials; middle, raster plots of
of viral injections of scAAV2/9-NX-hChR2-YFP into NIf. Right: Parasagittal 20 trials; bottom, histograms of 20 trials (100-ms bins)] showing excitatory
sections through HVC and NIf showing axon terminals labeled by tracer (E) and inhibitory (F) light-evoked response. (G) Schematic illustrating
injections into NIf. Scale bars, 75 mm. (B) Light-evoked optogenetic excitatory questions regarding the NIf-Av-HVC circuit. (H) Left: Schematic of
postsynaptic currents (oEPSCs) recorded at –80 mV [20 ms, 1 Hz pulse retrograde tracer injections. Right: Parasagittal section through NIf showing
(blue trace)] from an HVC neuron, compared to the same neuron when retrograde labeling. Scale bar, 100 mm. (I and J) Quantification of NIf neurons
applying glutamate blockers (gray trace). (C) oEPSCs recorded from HVC labeled by tracer injections [(I), actual numbers of neurons labeled; (J),
neurons in response to light stimulation of NIf axon terminals (blue, gray percentages of neurons labeled]. (K) Schematic of in vivo recording in HVC
fill) are blocked by DNQX (20 mM) and DL-AP5 (100 mM) (black, no fill) and NIf. (L) Raster plots of multi-unit light-evoked response from HVC and
and are nonexistent in birds not injected with ChR2 in NIf (black, gray fill). from three sites in NIf in the same hemisphere while HVC is light-stimulated
(D) Schematic of in vivo multi-unit recordings in HVC. (E and F) Multi-unit (100-ms bins).

84 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

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

(ChR2) construct (27) delivered with a self- strong increases in activity (257/341 recording projections from NIf using anatomical and
complementary adeno-associated virus (scAAV). sites) and a small percentage exhibiting strong physiological methods (Fig. 2, H to J, and figs.
Activation of NIf axon terminals elicited mono- suppression (Fig. 2, D to F, 18/341 recording sites). S1 and S2). Approximately 70% of NIf projec-
synaptic excitatory input to HVC neurons medi- Because NIf may also relay auditory informa- tion neurons exclusively innervate HVC, ~30%
ated by AMPA/NMDA receptors (Fig. 2, A to C). tion to HVC via a second sensorimotor region exclusively innervate Av, and only <5% project
Optical excitation of NIf terminals in vivo pro- (17, 28), Avalanche (Av) (Fig. 2G), we tested the to both HVC and Av. We next made in vivo ex-
duced reliable yet complex polysynaptic re- selectivity of our optogenetic manipulations of tracellular recordings in NIf while optogeneti-
sponses in HVC, with most neurons exhibiting NIf axon terminals. We first mapped efferent cally exciting its terminals in HVC to examine

Fig. 3. Optogenetic tutoring affects learning A optogenetic tutoring

of song temporal structure. (A) Timeline for hatched ChR2 injection optical stimulation analysis of adult song
the optogenetic tutoring experiment using (0 dph) (~45 dph) (5 days) (90-120 dph)
50-ms light pulses. (B) Representative spectro-
grams from a bird tutored by a zebra finch
viral expression 50ms x 4
tutor (ZF-tutor, left), a bird reared without a song (~2 weeks)
tutor (isolate, middle), and a bird opto-tutored
by 50-ms light pulses (right). Song elements 100ms
were quantified by thresholding (yellow curves) 300 trials over
the amplitude (red curves) of sounds using 1.5 hours/day
SAP2011 (see supplementary materials). Seg-
mented song elements and their durations are B ZF-tutor isolate opto-tutor (50ms)
A B C D E A B A A A A A
labeled by solid red lines and numbers at
bottom. (C) Song element duration of birds
tutored by a zebra finch tutor (ZF-tutor, n =
73 elements, 15 birds), isolate birds (iso, n =
61 elements, 10 birds), and birds 50-ms
opto-tutored birds (opto-tutor, n = 25 elements, 137 168 80 87 148 187 303 50 42 48 43 47
7 birds). Mann-Whitney U tests: ZF-tutor duration (ms)
birds versus isolate birds (by song element C D time shift (ms)
duration, P = 2.83 × 10–5; by bird, P = 0.0025); 400 0.35 62.5ms
-25 0 25 50
ZF-tutor birds versus 50-ms opto-tutored
r
birds (by song element duration, P = 3.86 × 10–4; -tuto
opto ms)

probability density
by bird, P = 0.0031); isolate birds versus 300 (50
duration (ms)

50-ms opto-tutored birds (by song element

ate
duration, P = 4.67 × 10–9; by bird, P = 2.06 × 10–4). isol
(D) Song element duration distribution of 200
birds shown in (C). Arrowheads and numbers 87.5ms, 112.5ms
show the peak positions for each curve 137.5ms, 187.5ms
100 237.5ms, 262.5ms
(black, ZF-tutor; gray, isolate) and the corre-
sponding duration. Inset: Time shift to achieve
n = 73 n = 61 n = 25
maximum cross-correlation with normally reared 0 0 250 500
op (50m

birds is shorter for 50 ms–stimulated birds

iso
ZF

n = 7 birds, n = 15 birds, n = 10 birds

to s
-tu

(orange bar) and longer for isolate birds duration (ms)

-tu )
to

(gray bar). (E) Timeline for the optogenetic

to
r

tutoring experiment using 300-ms light pulses.

E optical stimulation analysis of adult song G 1200
(F) Spectrograms of representative songs
(5 days) (90-120 dph)
of two birds opto-tutored with 300-ms light
duration (ms)

pulses. (G) Song element durations of 800

birds opto-tutored with 300-ms light pulses
300ms
(n = 7 song elements from 4 birds) are 400
significantly longer than those of birds tutored
with 50-ms light pulses (median duration = 300 trials over
1.5 hours/day 0
323 ms). Mann-Whitney U tests: 300-ms n = 25 n=7
opto-tutored birds versus 50-ms opto-tutored opto-tutor opto-tutor
birds (by song element duration, P = 2.54 × 10–6; (50ms) (300ms)
by bird, P = 0.0061).
F opto-tutor (300ms) opto-tutor (300ms)
A A A

327 340 1052ms

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 85

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

whether terminal stimulation antidromically tutored them using light pulses designed to isolate birds (Fig. 3, B and C, and fig. S4). We
excites NIf neurons (Fig. 2, K and L, n = 3 mimic short song elements near the end of their found that the majority of 50-ms opto-tutored
hemispheres from two birds). Although opto- song sensory learning phase. Light stimulation birds produced simple songs with only one to
genetic activation of axon terminals reliably was delivered at time intervals derived from three unique elements that were repeated or
evoked postsynaptic responses in HVC, they natural song tutoring patterns while experi- trilled at a high rate (audio file S1). Song ele-
failed to drive antidromic responses in NIf mental birds were alone in acoustic chambers ment durations in opto-tutored birds clustered
(nine recording sites from three hemispheres). (Fig. 3A and fig. S3). Adult zebra finch song near 50 ms (median = 62.3 ms), whereas both
To examine whether optogenetic excitation of typically contains ~3 to 6 unique song ele- normally reared and isolate birds produced
HVC might also directly excite NIf terminals ments lasting ~100 ms each, separated by brief song elements spanning a significantly broader
innervating Av, which is located ~1.5 mm from periods of silence (73 unique song elements distribution of durations (Fig. 3D).
the end of optic fibers over HVC, we measured measured from 15 birds, average 4.9 elements To further test whether opto-tutoring im-
the depth from the surface of the brain at which per bird, median duration = 106 ms). When plants memories that guide learning of song
we could elicit excitatory responses. Compiling birds are raised without any social or audi- element duration, we next tutored juvenile
data from all of our in vivo recordings (257 re- tory exposure to a song tutor, their adult song, birds with long-duration light pulses instead
cordings exhibiting excitatory responses), we referred to as isolate song, contains ~4 to 7 song of short-duration pulses (Fig. 3E). Birds opto-
found that we could only optogenetically ex- elements that are significantly longer than nor- tutored with 300-ms excitation of NIf-HVC
cite cells within the first 500 mm from the mally reared birds with free access to a tutor synapses produced adult vocalizations with
surface of the brain, hence our optogenetic (19, 29) (Fig. 3, B and C, 61 unique song ele- one to three song elements. The duration of
manipulations were unlikely to directly excite ments measured from 10 isolate birds, aver- these song elements was significantly longer
NIf axon terminals innervating Av. Together, age 6.1 elements per bird, median duration = than for 50-ms opto-tutored birds (Fig. 3, F and
these findings indicate that we can selectively 171 ms). We should note that isolate birds in G, fig. S5, and audio file S2).
manipulate activity at NIf-HVC synapses in this context are best described as untutored To examine whether 50-ms or 300-ms opto-
juvenile birds. birds and differ from “true” isolate birds that tutored birds learned vocal parameters other
To begin testing whether manipulation of are hand-raised by people and never exposed than song element duration from manipulation
the activity of NIf axon terminals in HVC can to other conspecifics. Birds opto-tutored as of NIf-HVC axon terminals, we measured acous-
implant behavioral-goal memories, we raised juveniles with repeated 50-ms light pulses tic features typically imitated during song learn-
young males without any social or auditory produced adult vocalizations with significantly ing, including pitch, mean frequency, goodness
experience of adult song tutors, then optically shorter song elements than normally reared or of pitch, and entropy (20, 30, 31). We found that

A B
50ms opto-tutor song development 300ms opto-tutor song development
51 dph 61 dph

53 dph 65 dph

a b a b a b ab ab a1 b a1 b a1 a2 b a1 a2
66 dph 72 dph

i i a b a b a b a b a b b i i i i i i i a1 b a1 b i a2
75 dph 78 dph

i i i i a b a b a b b ab ab i i i i a1 b i a1 b i a2 b a1a2
79 dph 80 dph

i i i a b a b a b a b i i i i i a1 b i a1 b i a2
92 dph 92 dph
0.7 0.7

i i i i i A B A B AB AB 500ms i i i A B i A B i B A 500ms

Fig. 4. Optogenetic tutoring implants a memory that guides song learning. types present in the adult songs; lowercase letters denote precursors of each
(A and B) Spectrograms of representative vocalizations produced by a 50-ms song element. Red curves on spectrograms show the original sound amplitude
opto-tutored bird [(A), opto-tutored on 49 to 51 dph and 53 to 54 dph] and a without segmentation. Plots at the right of the spectrograms show the probability
300-ms opto-tutored bird [(B), opto-tutored on 56 to 57 dph and 61 to 63 dph] on density of song element durations; dotted black lines allow for comparison
different days during song development. Capital letters denote song element of song element durations from the previous stage in development.

86 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

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

opto-tutored songs did not systematically dif- tutoring (movies S1 and S2). Together, these tutored birds exhibited complex learning tra-
fer from songs of normally reared or isolate findings indicate that opto-tutoring in juve- jectories similar to those observed in normally
zebra finches (fig. S6), suggesting an acous- nile birds selectively shapes the temporal struc- tutored birds (30, 34). Opto-tutored birds showed
tic phenotype that falls between these two ture of their adult courtship song. initial changes in vocal elements within 2 to
groups. 3 days of opto-tutoring, similar to birds that
We next explored whether opto-tutored birds Opto-tutoring implants a memory that guides are first song-tutored near the end of their
used their vocalizations appropriately during song learning sensitive period for sensory learning (12, 35).
social interactions. Zebra finches use their song Opto-tutoring could shape adult song by im- Similar to normal song learning, many changes
to court female birds in a behavior commonly planting a behavioral-goal memory that guides in song elements also slowly accrued over the
referred to as directed singing (32), and they developmental learning of song element dura- month of sensorimotor learning that followed
spend extended periods of time practicing their tion. Alternatively, opto-tutoring might directly the opto-tutoring experience. Birds began to
song when alone. Likewise, opto-tutored birds imprint or entrain patterns of activity on the modulate the amplitude of the initially noisy,
practiced their song when alone and produced HVC network, thereby constraining the pro- long, unstructured subsongs in response to the
a range of other call types typically produced duction of vocalizations to those with a spe- duration of light pulses they received. Birds
by zebra finches (33). When presented with cific temporal structure. To help discern these opto-tutored with 50-ms light pulses increased
female birds, opto-tutored birds readily per- possibilities, we examined the developmental the amplitude modulation of their long vocal
formed directed singing behavior using the trajectory of song elements of our opto-tutored elements, eventually learning to produce trilled,
short or long vocal elements shaped by opto- birds (Fig. 4 and fig. S7). We found that opto- short-duration song elements and in some

A B 100 C n.s.
ZF+opto tutoring 1200

similarity to tutor (%)

ZF tutor-contingent 75

duration (ms)
optical stimulation analysis of adult song
n.s. 800
(5 days) (90-120 dph)
50

300ms 400
25
~1.5 hours/day
0
0 n = 25 n=7 n=4
ctrl iso Z F+opto-tutor

op (50m

op 00

Z F (30
to

to ms
(300ms)

(3

+o 0m
-tu s)

-tu )

pt s)
to

to

o-
r

tu
to
r
D control ZF +opto-tutor E
A B C D E F G A B C D 1200

tutor duration (ms) 800

161 180 81 107

duration 125 205 153 135 50 101 212
(ms) A B C E F G A 400

0 n = 11
pupil
op 300

iso
ZF

126 202 160 67 109 200 664ms

to ms
(

lat
-tu
-tu )

e
to
to

86% similarity 26% similarity

r
r

Fig. 5. Optogenetic tutoring overrides natural tutoring. (A) Timeline for elements from 4 birds) and ZF + opto-tutored birds (n = 4 song elements
the experiments in which birds received both zebra finch and optogenetic from three birds), P = 0.9. (D) Spectrograms of tutor-pupil songs of control
tutoring (ZF + opto tutoring). (B) Birds tutored by zebra finch tutors (black, no birds (good copy, 86% similarity) and a ZF + opto tutoring pair using 300-ms
fill) show high percent similarity to the tutor song. ZF + opto-tutored birds pulses (poor copy, 26% similarity). (E) Song element durations for birds
(blue, gray fill) do not imitate the song of their tutor and are on par with isolate opto-tutored with 300-ms pulses (blue, gray fill) and comparisons with song
birds (black, gray fill). Mann-Whitney U tests: control birds versus ZF + element durations from normally reared (black, no fill) and isolate birds (black,
opto-tutored birds, P = 0.03; isolate birds versus ZF + opto-tutored birds, gray fill). Mann-Whitney U tests: 300-ms opto-tutored birds versus isolate
P = 0.8. n.s., not significant. (C) Song element duration shows no difference for birds (by song element duration, P = 0.0010; by bird, P = 2.0568 × 10–4);
ZF+opto-tutored birds (blue, no fill) and birds only receiving opto-tutoring 300-ms opto-tutored birds versus ZF-tutored birds (by song element duration,
(blue, gray fill). Mann-Whitney U tests: 300-ms opto-tutored birds (n = 7 song P = 3.2549 × 10–6; by bird, P = 2.4684 × 10–4).

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 87

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

instances learning to produce gaps between we found that both 50-ms and 300-ms opto- from which to learn: ascending auditory infor-
these vocal elements (Fig. 4A and fig. S7). Birds tutored birds produced songs with distinct mation from a live zebra finch tutor and light-
opto-tutored with 300-ms light pulses, on the introductory notes and song motifs (Fig. 4 evoked activity at NIf-HVC synapses.
other hand, slowly learned to decrease the and fig. S7); moreover, the vocal elements of These birds failed to imitate the songs of
amplitude modulation across vocal elements, 300-ms opto-tutored birds were significantly their tutors, exhibiting levels of similarity to
leading to the gradual emergence of longer longer than those of isolate birds (Fig. 5E). their song tutor that were indistinguishable
and more harmonic vocal elements (Fig. 4B Together, these findings suggest that optoge- from the songs of isolate birds (Fig. 5B). Instead,
and fig. S7). Opto-tutored birds also crystal- netic excitation of NIf-HVC synapses implants they learned from the 300-ms optical stimula-
lized their songs starting at 85 to 90 dph behavioral-goal memories that guide learning tion, displaying song element durations simi-
(days post-hatching). Their songs before 80 dph of a bird’s courtship song. lar to those of birds that were opto-tutored but
exhibited variable vocal durations and acous- never tutored by adult zebra finches (Fig. 5, C
tic features, whereas songs after 90 dph The NIf-HVC pathway is necessary to form a and D). Grouping all birds opto-tutored with
were increasingly stereotyped, like those re- song memory 300-ms light pulses (n = 7 birds producing 11
sulting from song crystallization in normally A longstanding view is that behavioral-goal song elements), we found that their song ele-
reared birds. memories for song are encoded in auditory ment durations were significantly longer than
These results suggest that opto-tutoring im- regions presynaptic to the NIf-HVC pathway, those of isolate or normally reared birds (Fig.
plants memories that guide learning, rather such as in the caudomedial nidopallium (NCM) 5E). Therefore, even when provided with a nor-
than directly entraining a specific motor pro- and primary and secondary auditory forebrain mal song model, birds learn from opto-tutoring.
gram in young animals. However, it is also pos- (field L) (6, 36–38). Memories encoded pre- This finding suggests the possibilities that
sible that opto-tutoring simply biases or selects synaptic to HVC might be capable of guiding (i) activity at NIf-HVC synapses overrides learn-
among precursor or innate vocalizations to song learning independent of behavioral-goal ing from auditory experiences with a social
specify the production of vocal elements with memories encoded via NIf-HVC synapses. In model, or (ii) the memories that might be
certain durations in adulthood. For example, addition, auditory information entering HVC encoded presynaptically or independent of
optical stimulation of NIf axon terminals in via other routes, such as Av projections into NIf-HVC synapses are insufficient to guide
HVC could bias the 50-ms birds to only sing HVC, may also be capable of encoding these song imitation.
the short introductory notes that typically memories. To evaluate these ideas, we paired To clarify whether the pathway from NIf to
precede the bird’s normal song motif and bias opto-tutoring with normal song tutoring as HVC is necessary for a young bird to acquire a
the 300-ms birds to only sing long isolate-like juvenile birds socially interacted with live behavioral-goal memory during social expe-
vocal elements or calls. Such a scenario could tutors. Optical activation of NIf-HVC synapses riences with a tutor, we genetically lesioned
point to circuit mechanisms for how innate was contingent on the tutors’ singing behav- neurons in NIf projecting to HVC using an
vocal repertoires are selected or reinforced ior (Fig. 5A), providing the juvenile with two intersectional viral approach for cre-dependent
by activity during development. However, simultaneous potential sources of information expression of caspase3 (17, 39). NIf neurons

Fig. 6. NIf-HVC synapses are A genetic lesions NIfHVC B n.s.

100
necessary for acquisition of
genetic lesion tutoring song analysis

similarity to tutor (%)

a tutor song memory but isolate (~35 dph) (55-60 dph) (90-120 dph)
not for vocal imitation. (A) 75
Schematic of the NIf-HVC
lesion experiments. Genetic 50
lesions of NIf neurons
projecting to HVC using viral OR
expression of a cre-dependent 25
normal genetic lesion song analysis
caspase3 were performed tutoring (~35 dph) (90-120 dph)
either after birds had memo- 0
rized the song of their tutor, or lesion tutor ctrl
prior to song tutoring. Song
tutor lesion
learning outcomes were
examined when pupils reached C lesion tutor tutor lesion control
adulthood. (B) Birds with
A B C D A B C D A B C D E
NIf-HVC neurons lesioned
before tutoring (green with
dark gray fill) failed to copy the tutor
tutor songs. Controls tutored
before having their NIf-HVC
X Y Z A B/C D A B C D E
neurons lesioned (green with
light gray fill) copied the tutor
songs as well as nonlesioned pupil
birds (black with no fill). Mann-
Whitney U tests: lesioned then 37% similarity 86% similarity 89% similarity
tutored birds versus control
birds, P = 0.02; lesioned then
tutored birds versus isolate birds, P = 0.3429. (C) Spectrograms of representative tutor-pupil song comparisons shown in (B).

88 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

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

projecting to HVC were lesioned in 35- to on a specific brain circuit for their encoding 15. A. J. Peters, H. Liu, T. Komiyama, Annu. Rev. Neurosci. 40,
40-day-old isolate birds and then they were but may be more broadly distributed at latter 77–97 (2017).
16. F. Ali et al., Neuron 80, 494–506 (2013).
housed with a tutor for 5 days starting at 55 days time points when they are used to influence 17. T. F. Roberts et al., Nat. Neurosci. 20, 978–986
of age. A separate group of birds were raised behavior. (2017).
with a tutor prior to lesioning NIf neurons pro- Premotor cortical circuits involved in speech 18. M. A. Long, M. S. Fee, Nature 456, 189–194 (2008).
19. O. Fehér, H. Wang, S. Saar, P. P. Mitra, O. Tchernichovski,
jecting to HVC at 35 to 40 days of age (Fig. 6A). production are activated by listening to speech,
Nature 459, 564–568 (2009).
We found that birds with NIf-HVC lesioned even in pre-verbal infants (46), which suggests 20. D. Lipkind et al., Nature 498, 104–108 (2013).
before tutoring failed to imitate the song of that tight coupling between sensory and pre- 21. P. Ravbar, D. Lipkind, L. C. Parra, O. Tchernichovski,
their tutors (Fig. 6, B and C), exhibiting simi- motor cortical circuits exists at the earliest J. Neurosci. 32, 3422–3432 (2012).
22. P. Adret, C. D. Meliza, D. Margoliash, J. Neurophysiol. 108,
larity scores (with respect to their tutor) that stages of learning. Sensorimotor pathways 1977–1987 (2012).
were indistinguishable from those of isolate may therefore provide a general substrate 23. M. J. Coleman, R. Mooney, J. Neurosci. 24, 7251–7265
birds. It has previously been shown that non- for encoding behavioral-goal memories during (2004).
24. B. Lewandowski, A. Vyssotski, R. H. Hahnloser, M. Schmidt,
selective lesions or inactivation of NIf prior to social interactions and learning. The substrate J. Physiol. Paris 107, 178–192 (2013).
song tutoring can disrupt subsequent vocal for long-term storage of behavioral-goal mem- 25. A. L. Vyssotski, A. E. Stepien, G. B. Keller, R. H. Hahnloser,
imitation (7). Our current findings show that ories is still unknown. Previous studies have PLOS Biol. 14, e2000317 (2016).
26. H. Horita et al., PLOS ONE 7, e42173 (2012).
lesions of only the NIf-HVC pathway are suffi- suggested that tutor song memories encoded 27. L. Xiao et al., Neuron 98, 208–221.e5 (2018).
cient to disrupt song learning. However, it is in higher-order auditory regions function as 28. E. Akutagawa, M. Konishi, J. Comp. Neurol. 518, 3086–3100
not clear whether these lesions disrupt acqui- an “auditory template” used to guide vocal (2010).
29. H. Williams, K. Kilander, M. L. Sotanski, Anim. Behav. 45,
sition of tutor song memories, or whether they imitation (6, 36). Our findings indicate that 695–705 (1993).
disrupt subsequent sensorimotor learning. We behavioral-goal memories formed during social 30. O. Tchernichovski, P. P. Mitra, T. Lints, F. Nottebohm,
found that lesions to this pathway after birds interactions require sensorimotor transforma- Science 291, 2564–2569 (2001).
31. O. Tchernichovski, F. Nottebohm, C. E. Ho, B. Pesaran,
had an opportunity to learn from their tutor did tion of pertinent auditory experiences prior to
P. P. Mitra, Anim. Behav. 59, 1167–1176 (2000).
not affect the ability of juvenile birds to accu- consolidation. HVC provides feedback to the 32. E. D. Jarvis, C. Scharff, M. R. Grossman, J. A. Ramos,
rately imitate the song of their tutor (Fig. 6, B auditory system via its projection to Av, and F. Nottebohm, Neuron 21, 775–788 (1998).
and C, P = 1.0, Mann-Whitney U test), which this pathway may facilitate long-term distrib- 33. R. A. Zann, in The Zebra Finch: A Synthesis of Field and
Laboratory Studies, C. M. Perrins, Ed. (Oxford Univ. Press, 1996),
suggests that the NIf-HVC circuit plays a spe- uted encoding of behavioral-goal memories pp. 196–246.
cific role during the acquisition of behavioral- during development (17). 34. T. S. Okubo, E. L. Mackevicius, H. L. Payne, G. F. Lynch,
goal memories. Together with our opto-tutoring This research focused on how juvenile birds M. S. Fee, Nature 528, 352–357 (2015).
35. S. S. Shank, D. Margoliash, Nature 458, 73–77 (2009).
results, this indicates that the NIf-HVC circuit form memories that guide learning of song 36. J. J. Bolhuis, S. Moorman, Neurosci. Biobehav. Rev. 50, 41–55
is necessary for forming a memory used to guide temporal structure. How young birds mem- (2015).
learning of song-element duration, but that this orize other aspects of tutor song, such as the 37. P. Adret, Ann. N.Y. Acad. Sci. 1016, 303–324 (2004).
38. S. E. London, D. F. Clayton, Nat. Neurosci. 11, 579–586
circuit is dispensable once this behavioral-goal spectral and syntax structure, is not known. (2008).
memory is acquired. This suggests that at least HVC receives input from at least three pallial 39. C. F. Yang et al., Cell 153, 896–909 (2013).
certain aspects of tutor song memories used and thalamic regions aside from NIf (28). One 40. P. W. Frankland, B. Bontempi, Nat. Rev. Neurosci. 6, 119–130
to evaluate vocal performances are ultimately possibility is that unified representations of (2005).
stored downstream of synapses between NIf tutor songs only emerge through sensorimotor 41. W. B. Scoville, B. Milner, J. Neurol. Neurosurg. Psychiatry 20,
11–21 (1957).
and HVC. transformations and different inputs to HVC 42. S. Tonegawa, M. D. Morrissey, T. Kitamura, Nat. Rev. Neurosci.
carry unique streams of information, allowing 19, 485–498 (2018).
Discussion birds to form memories of temporal as well as 43. X. Liu et al., Nature 484, 381–385 (2012).
Episodic memories are initially dependent spectral and syntactical features of their tutor. 44. S. A. Josselyn, S. Köhler, P. W. Frankland, Nat. Rev. Neurosci.
16, 521–534 (2015).
on the hippocampus but are more broadly Understanding the role of other sensorimotor 45. D. Vallentin, G. Kosche, D. Lipkind, M. A. Long, Science 351,
distributed at latter time points (13, 40, 41). pathways in forming behavioral-goal memories 267–271 (2016).
Experiments in recent years have used activity- is likely to yield insights into how more com- 46. G. Dehaene-Lambertz et al., Proc. Natl. Acad. Sci. U.S.A. 103,
dependent tagging and optogenetic manipu- plex or integrated memories of vocal models 14240–14245 (2006).

lations to dissect the cellular constituents of are learned during social interactions. AC KNOWLED GME NTS
episodic memories, identifying “engram cells” We thank members of the Roberts laboratory for discussion
and how they can be manipulated to affect re- RE FERENCES AND NOTES and comments on the manuscript, and J. Holdway and
call and behavior (42–44). Our understanding 1. J. F. Gariépy et al., Front. Neurosci. 8, 58 (2014). A. Guerrero for laboratory support and animal husbandry.
2. M. Iacoboni, Curr. Opin. Neurobiol. 15, 632–637 (2005). Funding: Supported by NIH grant R01DC014364 and
of procedural memories, and of the behavioral- NSF grant IOS-1457206 (T.F.R.). Author contributions: W.Z.
3. A. J. Doupe, P. K. Kuhl, Annu. Rev. Neurosci. 22, 567–631
goal memories that guide motor imitation, has (1999). and T.F.R. designed the experiments; W.Z., F.G.-O., and
lagged far behind. Procedural memories are 4. T. F. Roberts, R. Mooney, Hear. Res. 303, 48–57 (2013). D.D. performed the experiments; all authors contributed to
5. L. R. Squire, Psychol. Rev. 99, 195–231 (1992). data analysis; and W.Z. and T.F.R. wrote the manuscript.
thought to be formed and directly represented Competing interests: Authors declare no competing interests.
6. J. J. Bolhuis, M. Gahr, Nat. Rev. Neurosci. 7, 347–357
in the circuits involved in their performance (2006). Data and materials availability: All data are available
(12, 14, 15, 45). Using optogenetic tutoring of 7. T. F. Roberts, S. M. Gobes, M. Murugan, B. P. Ölveczky, in the main text or the supplementary materials.
juvenile songbirds, we found that aspects of R. Mooney, Nat. Neurosci. 15, 1454–1459 (2012).
8. M. Tanaka, F. Sun, Y. Li, R. Mooney, Nature 563, 117–120
behavioral-goal memories can be implanted (2018). SUPPLEMENTARY MATERIALS
through manipulation of sensorimotor synaps- 9. H. Eichenbaum, Nat. Rev. Neurosci. 1, 41–50 (2000). science.sciencemag.org/content/366/6461/83/suppl/DC1
es that convey information from a social model 10. M. Iacoboni et al., Science 286, 2526–2528 (1999). Materials and Methods
11. J. F. Prather, S. Peters, S. Nowicki, R. Mooney, J. Neurosci. 30, Figs. S1 to S8
to motor circuits, but that this synaptic path- Table S1
10586–10598 (2010).
way is not necessary for evaluating vocal per- 12. T. F. Roberts, K. A. Tschida, M. E. Klein, R. Mooney, Movies S1 and S2
formances during sensorimotor learning. This Nature 463, 948–952 (2010). Audio files S1 and S2
suggests that, like episodic memories, lasting 13. J. L. McClelland, B. L. McNaughton, R. C. O’Reilly, Psychol. Rev.
102, 419–457 (1995). 18 December 2018; resubmitted 29 April 2019
memories of observed behaviors used to guide 14. J. D. Gabrieli, S. Corkin, S. F. Mickel, J. H. Growdon, Accepted 14 August 2019
imitative learning are initially dependent Behav. Neurosci. 107, 899–910 (1993). 10.1126/science.aaw4226

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 89

RES EARCH

◥ as the outcome of disk fragmentation (14).

REPORTS The Toomre parameter, which quantifies the
dynamical state of the disk (15), shows that
STAR FORMATION the [BHB2007] 11 circumbinary disk is stable
against gravitational collapse (9), so the ob-
Gas flow and accretion via spiral streamers and served spirals are unlikely to be the result of
disk fragmentation. The total length of the
circumstellar disks in a young binary protostar filaments is ~392 AU, their mean width is
~12 AU, and the mean brightness tempera-
F. O. Alves1*, P. Caselli1, J. M. Girart2,3, D. Segura-Cox1, G. A. P. Franco4, A. Schmiedeke1, B. Zhao1 ture is 14 K, from which we estimate a mean
total mass of ~19 MJup (9). This corresponds to
The majority of stars are part of gravitationally bound stellar systems, such as binaries. Observations a mean mass per unit length of e10 M⊙ pc 1 ,
of protobinary systems constrain the conditions that lead to stellar multiplicity and subsequent which is below the critical value required to
orbital evolution. We report high–angular resolution observations of the circumbinary disk around produce fragmentation in the filaments. The
[BHB2007] 11, a young binary protostar system. The two protostars are embedded in circumstellar spiral shape centered on the protobinary sys-
disks that have radii of 2 to 3 astronomical units and probably contain a few Jupiter masses. These tem indicates dynamical interaction with the
systems are surrounded by a complex structure of filaments connecting to the larger circumbinary disk. protostars. Figure 1B shows that the geomet-
We also observe accretion and radio jets associated with the protobinary system. The accretion is rical center of the circumbinary disk is located
preferentially onto the lower-mass protostar, consistent with theoretical predictions. near [BHB2007] 11A, within the angular re-
solution, suggesting that this source is the

A
primary, higher-mass object of the system. We
bout half of the stars in the solar neigh- We observed [BHB2007] 11 at 225.3 GHz suggest that the filaments are inflow streamers
borhood are in gravitationally bound (wavelength ~1.3 mm) with the Atacama Large from the extended circumbinary disk onto the
stellar systems, such as binaries (1). The Millimeter/submillimeter Array (ALMA) (9). circumstellar disks of the protobinary sys-
mean separation between components, The observations were centered on the circum- tem. We estimate a mass accretion rate from
a few tens of astronomical units (AU; binary disk of [BHB2007] 11 and reveal its the circumbinary disk into the circumstellar
1 AU = 149,597,870.7 km), is a consequence of internal structure. Figure 1 shows a multiscale disk of e1:1  10 5 M⊙ year 1 (9), which is
their formation process, thought to be frag- view of the dust distribution, extending from consistent with other protostellar sources (16).
mentation of the protostellar disk due to gra- the B59 core scale (~20,600 AU) to the sub- The rotationally supported disk may be redis-
vitational instabilities (2, 3). Observations of disk scale (~6 AU); the latter is the maximum tributing its angular momentum (through pro-
young binary systems rarely probe such small spatial resolution of our map, given the B59 cesses such as viscosity, magnetic braking, or
scales. We seek to investigate the dynamical distance of 163 pc (1 pc = 206,264.8 AU) (10). gravitational torques), causing material to fall
evolution of a binary system still embedded The dust map reveals two compact sources that into the gravitational potential wells of the
in its natal cloud, where gas from a circum- we interpret as circumstellar disks around both individual protostars. Although other factors
binary disk is expected to accrete onto each components of a protobinary system, which such as envelope accretion into the circumbi-
binary component. In the protobinary accre- has a projected separation of ~28 AU. The nary disk, misaligned disks, or eccentric bi-
tion phase, theoretical models indicate that components were also seen in previous ob- nary orbit could also produce substructure
high–angular momentum material accretes servations (11, 12) with the Karl G. Jansky Very such as filaments in the circumbinary disk
preferentially onto the less massive compan- Large Array (VLA). We designate the individ- (6, 17, 18), accretion streamers feeding the
ion, which has a higher orbital angular mo- ual components as [BHB2007] 11A (northern circumstellar disks can still occur in those
mentum than the primary, causing ultimately source) and [BHB2007] 11B (southern source). scenarios (6, 17).
an equal share of mass among the individual The dust emission seen in the ALMA data We measured kinematics by reanalyzing (9)
components (4–6). arises from circumstellar disks with radii 3.1 ± previous observations (8) of the CO (angular
[BHB2007] 11 (right ascension 17h11m23.18s, 0.6 AU ([BHB2007] 11A) and 2.1 ± 0.5 AU momentum quantum number J = 2→1) mo-
declination −27°24′31.5″, J2000 equinox) is ([BHB2007] 11B), similar to the radius of the lecular emission line in [BHB2007] 11. The
the youngest member [age 0.1 to 0.2 million asteroid belt in the Solar System. Their incli- relative positions of the molecular gas in each
years (7)] of the small cluster of young stellar nations are ~40° between the disk normals and velocity channel of the emission spectra were
objects in the Barnard 59 (B59) core (part of the line of sight. From the continuum emission, determined with positional accuracies bet-
the Pipe Nebula molecular cloud), which is the masses of the circumstellar disks are esti- ter than 0.05″ (9). The systemic velocity of
still growing mass through dust and gas ac- mated to be within an order of magnitude of a [BHB2007] 11 with respect to the local stan-
cretion (a class 0/I young stellar object). Pre- Jupiter mass (MJup), with the [BHB2007] 11A disk dard of rest is vlsr ~ 3.6 km s−1. The CO line
vious observations of this object show an slightly more massive than the [BHB2007] 11B traces primarily outflow emission, but veloc-
envelope surrounding a circumbinary disk disk (9). The dust content in the disks is a few ity components faster than the outflows are
of radius 90 AU, with prominent bipolar out- Earth masses (MEarth) (9). The larger circum- detected within the circumbinary disk (8).
flows launched near the disk edge (8). binary disk has a total mass of 0.08 ± 0.03 These high-velocity components reach up to
solar masses ðM⊙ ; where 1M⊙ e 1000 MJup e 15 km s−1. The blue- and redshifted velocities
330;000 MEarth Þ. Because this represents ~260 (vlsr < −1.5 km s−1 and vlsr > 9 km s−1, respec-
1
Center for Astrochemical Studies, Max Planck Institute for MEarth in dust, we speculate that the circum- tively) are centered on [BHB2007] 11B, with
Extraterrestrial Physics, Garching, 85748, Germany. 2Institut
stellar disks may later form rocky terrestrial the higher velocities closest to the position of
de Ciències de l’Espai, Consejo Superior de Investigaciones
Científicas, Cerdanyola del Vallès, E-08193, Catalunya, Spain. planets (13). the protostar (Fig. 2A). No high-velocity com-
3
Institut d’Estudis Espacials de Catalunya, Barcelona, The protobinary is in the center of a com- ponents are detected near [BHB2007] 11A.
E-08034, Spain. 4Departamento de Física, Instituto de plex network of dust structures distributed in The increasing velocity toward [BHB2007] 11B
Ciências Exatas, Universidade Federal de Minas Gerais, Belo
Horizonte, 30.123-970, Brazil. spiral shapes. Spirals in protostellar systems indicates acceleration of infalling gas from
*Corresponding author. Email: falves@mpe.mpg.de such as LDN 1448NB have been interpreted the circumbinary disk onto [BHB2007] 11B.

90 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

Fig. 1. Dust distribution in [BHB2007] 11 on scales from the core to the cir- levels are scaled logarithmically in (A) and (B). (C) Zoomed-in view of the circum-
cumstellar disks. (A) The core of Barnard 59 observed at 250 mm with the binary disk, revealing the dust filaments surrounding the protobinary system,
Herschel space telescope (20). (B) The [BHB2007] 11 disk and envelope observed where [BHB2007] 11A is the northern component and [BHB2007] 11B is the southern
at 1.3 mm with ALMA (8). The contours indicate intensity levels of 30, 60, 70, and and brighter component. The dashed lines are the intensity contours from (B). In (B)
450 times the map noise of 62 mJy per beam (1 Jy = 10−26 W m−2 Hz−1). Intensity and (C), the ALMA synthesized beam is shown in the bottom left corner.

The observed velocities are consistent with with the momentum rate, so the radio flux is object in the binary system. The infalling gas
the filament accretion rate estimated above. also correlated with the disk-star accretion rate onto [BHB2007] 11B suggests a hgher accre-
Figure 2B shows the CO spectrum with the (19). The radio jet–like emission along the east- tion from the circumbinary disk onto this
high-velocity channels used to compute the west direction associated with [BHB2007] 11A object than onto [BHB2007] 11A. This prefer-
CO emission peaks. (Fig. 3) suggests that this source has a higher ential accretion onto the lower-mass stellar
We used the VLA to observe the protobi- (circumstellar) disk-star accretion rate than object in a binary system matches theoretical
nary in continuum emission at 34.5 GHz (11) [BHB2007] 11B. In addition, [BHB2007] 11A predictions (4–6).
and at lower frequencies (10, 15, and 22 GHz) has an ionized mass loss rate that is a factor of In summary, the small-scale structure seen
(9). The centimeter-wavelength data are con- ~2 higher than that of [BHB2007] 11B [e6:4  in our observations includes streamers of gas
sistent with thermal free-free emission from 10 10 M⊙ year 1 and e3:7  10 10 M⊙ year 1, and dust accreting from a circumbinary disk
partially ionized collimated jets produced by respectively (9)]. The stronger radio jet and onto small circumstellar disks around the in-
angular momentum redistribution in the disk- higher ionized mass loss rate support the dividual components of a protobinary system.
star system. This type of emission correlates proposition that [BHB2007] 11A is the primary Their disks have a dust mass equivalent to a few

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 91

RES EARCH | R E P O R T S

Fig. 2. Small-scale kine-

matics in [BHB2007] 11
measured from CO
molecular line emission.
(A) High-velocity compo-
nents of the CO emission
(with respect to the ambi-
ent local standard of rest
velocity of [BHB2007] 11,
vlsr ~ 3.6 km s−1) overlaid
on the continuum map
(gray scale). Circles indi-
cate the position of CO
emission peaks in each
velocity channel (see the
color bar). CO data
between −2 and 9 km s−1
(indicated as a shaded
area in the color bar) are
not shown because most
gas close to the ambient velocity is filtered out by the interferometer and/or areas indicate the CO velocity components that exceed the reference
arises from extended emission from outflows or the envelope. Circle sizes denote H2CO spectrum associated with the rotation of the circumbinary disk (9). The
the uncertainty on the absolute position (9). (B) CO mean spectrum extracted faint dashed lines show the strong filtering of the CO line near the systemic
from the disk area showing the CO velocity channels used in (A). The shaded velocity of the source.

Fig. 3. Centimetric emission associated

with the binary system. Continuum N
emission at 0.94 cm (32.5 GHz) associated
with the [BHB2007] 11 binary system,
observed with the VLA. Contours are at E
3, 5, 10, 15, 20, and 25 times the map
noise of 12 mJy per beam; the background
image is the ALMA map from Fig. 1C.
The extended emission to the east of
[BHB2007] 11A suggests a radio jet asso-
ciated with this source. The VLA synthe-
sized beam is shown in the bottom
left corner.

Earth masses. The gas accretion is more pro- RE FERENCES AND NOTES 6. T. Matsumoto, K. Saigo, S. Takakuwa, Astrophys. J. 871, 36–52
minent in the secondary, less massive object of 1. D. Raghavan et al., Astrophys. J. Suppl. Ser. 190, 1–42 (2019).
(2010). 7. T. Y. Brooke et al., Astrophys. J. 655, 364–374 (2007).
the system, whereas the disk-star accretion in-
2. K. M. Kratter, C. D. Matzner, M. R. Krumholz, R. I. Klein, 8. F. O. Alves et al., Astron. Astrophys. 603, L3 (2017).
ferred by the ionized free-free emission is higher Astrophys. J. 708, 1585–1597 (2010). 9. Materials and methods are available as supplementary materials.
in the primary object instead. We expect that 3. L. D. G. Sigalotti, F. Cruz, R. Gabbasov, J. Klapp, 10. S. A. Dzib, L. Loinard, G. N. Ortiz-León, L. F. Rodríguez,
this two-level accretion process, circumbinary J. Ramírez-Velasquez, Astrophys. J. 857, 40 (2018). P. A. B. Galli, Astrophys. J. 867, 151–166 (2018).
4. M. R. Bate, I. A. Bonnell, Mon. Not. R. Astron. Soc. 285, 33–48 11. F. O. Alves et al., Astron. Astrophys. 616, A56 (2018).
disk to circumstellar disks, then circumstellar 12. S. A. Dzib, L. F. Rodríguez, A. T. Araudo, L. Loinard, Rev. Mex.
(1997).
disks to stars, drives the dynamics of the binary 5. M. R. Bate, I. A. Bonnell, V. Bromm, Mon. Not. R. Astron. Soc. Astron. Astrofis. 49, 345–349 (2013).
system during its mass accretion phase. 336, 705–713 (2002). 13. B. Liu, C. W. Ormel, Astron. Astrophys. 615, A138 (2018).

92 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

14. J. J. Tobin et al., Nature 538, 483–486 (2016). Radio Astronomy Observatory, a facility of the National Science of the continuum image and contributed to the continuum data
15. A. Toomre, Astrophys. J. 139, 1217–1238 (1964). Foundation operated under cooperative agreement by Associated analysis. G.A.P.F. developed the analytical approach for the mass
16. M. M. Dunham et al., in Protostars and Planets VI, H. Beuther, Universities, Inc. This research has also made use of data from the accretion rate and contributed to the manuscript preparation.
R. S. Klessen, C. P. Dullemond, T. Henning, Eds. (Univ. of Herschel Gould Belt survey (HGBS) project (http://gouldbelt- A.S. and B.Z. contributed to the theoretical interpretation of the
Arizona Press, 2014), pp. 195–218. herschel.cea.fr). The HGBS is a Herschel Key Programme jointly results. All authors contributed to the discussion and interpretation
17. P. Mösta, R. E. Taam, P. C. Duffell, Astrophys. J. 875, L21 carried out by SPIRE Specialist Astronomy Group 3 (SAG 3), of the data. Competing interests: There are no competing
(2019). scientists of several institutes in the PACS Consortium (CEA interests. Data and materials availability: The ALMA observations
18. G. M. Kennedy et al., New Astron. 3, 230–235 (2019). Saclay, INAF-IFSI Rome and INAF-Arcetri, KU Leuven, and MPIA are archived in the ALMA Science Archive http://almascience.nrao.
19. S. P. Reynolds, Astrophys. J. 304, 713–720 (1986). Heidelberg), and scientists of the Herschel Science Center (HSC). edu/aq/ under project codes 2016.1.01186.S and 2013.1.00291.S.
20. P. André et al., Astron. Astrophys. 518, L102 (2010). Funding: F.O.A., P.C., D.S.-C, A.S., and B.Z. acknowledge financial The VLA data are archived at https://science.nrao.edu/facilities/
support from the Max Planck Society. P.C. and B.Z. acknowledge vla/archive/index under VLA project code 16B-290.
ACKN OW LEDG MEN TS support of the European Research Council (ERC, project PALs
We thank the reviewers for helpful and detailed comments that 320620). J.M.G. is supported by the MINECO (Spain) grant
improved the manuscript and the ALMA and VLA staff for AYA2017-84390-C2. G.A.P.F. acknowledges support from CNPq SUPPLEMENTARY MATERIALS
performing the observations and quality assessment of the data. and FAPEMIG (Brazil). Author contributions: F.O.A. led the science.sciencemag.org/content/366/6461/90/suppl/DC1
We also thank M. Fernández López, W. Lyra, and P. Cortes for project and ALMA observing proposal, analyzed the ALMA data Materials and Methods
useful discussions on the data analysis and interpretation. ALMA is (calibration and imaging), conducted the molecular line data Figs. S1 to S3
a partnership of ESO (representing its member states), NSF (USA), analysis, and led the preparation of the manuscript. P.C. Table S1
and NINS (Japan), together with NRC (Canada), MOST and ASIAA contributed to paper preparation, molecular line interpretation, and References (21–29)
(Taiwan), and KASI (Republic of Korea), in cooperation with the derivation of dust properties. J.M.G. led the VLA observing
Republic of Chile. The Joint ALMA Observatory is operated by ESO, proposal and the VLA data analysis, including calibration, imaging, 17 December 2018; accepted 9 September 2019
AUI/NRAO, and NAOJ. The VLA is an instrument of the National and spectral index derivation. D.S.-C. performed the self-calibration 10.1126/science.aaw3491

ATOMIC PHYSICS gaps between clock interrogations, leading

to increased sensitivity to noise aliasing. The
Seconds-scale coherence on an optical clock sub–micrometer-scale optical lattices essen-
tial for such Hubbard-regime physics also
transition in a tweezer array limit the current record for atom-light co-
herence (8 s), because dephasing from atomic
Matthew A. Norcia, Aaron W. Young, William J. Eckner, Eric Oelker, Jun Ye, Adam M. Kaufman* tunneling must be balanced against deco-
herence attributable to scattering from the
Coherent control of high–quality factor optical transitions in atoms has revolutionized precision lattice (17, 18). This suggests that the next
frequency metrology. Leading optical atomic clocks rely on the interrogation of such transitions in frontier in atomic coherence will require
either single ions or ensembles of neutral atoms to stabilize a laser frequency at high precision and new tools that allow control of the atomic
accuracy. We demonstrate a platform that combines the key strengths of these two approaches, based spacing (18).
on arrays of individual strontium atoms held within optical tweezers. We report coherence times Here, we demonstrate the core capabilities
of 3.4 seconds, single-ensemble duty cycles up to 96% through repeated interrogation, and frequency for an optical-frequency metrology platform
stability of 4.7 × 10−16 (t/s)–1/2. These results establish optical tweezer arrays as a powerful tool based on arrays of individual strontium atoms
for coherent control of optical transitions for metrology and quantum information science. trapped in optical tweezers (see Fig. 1A). The
platform combines the high duty cycle and

O
microscopic control techniques of ion clocks
ptical clocks based on neutral atoms and atom shot noise, enabling frequency measure- with the scaling capacity inherent in neutral
ions achieve exceptional precision and ments with precision below the part-per-1018 atoms, and it allows for a high degree of atomic
accuracy (1–6), with applications in re- level within 1 hour (2). isolation and coherence. The single-atom oc-
lativistic geodesy (7), tests of relativity Working with large atom numbers creates cupancy readily achieved in tweezers elimi-
(8), and searches for dark matter (9). challenges associated with laser frequency noise nates perturbations associated with atomic
Achieving such performance requires balanc- aliasing and interatomic collisions, which can collisions; the low temperatures and relatively
ing competing desirable features, including a be difficult to address simultaneously. Fre- large spatial separation between tweezers can
high particle number, isolation of atoms from quency noise aliasing through the so-called suppress motional and tunneling effects. The
collisions, insensitivity to motional effects, and Dick effect arises from dead time between tweezer platform also enables rapid, state-
high–duty cycle operation (10, 11). subsequent interrogations of the atoms (13). selective, nondestructive detection (19–24) and
The leading platforms for optical clocks— Methods have been developed to overcome thus repeated clock interrogation of the same
based on trapped single ions and ensembles of dead-time effects by interleaving interroga- atoms. These experimental conditions allow
neutral atoms confined within optical lattices— tion of two independent clock ensembles in the duty cycles, stability, and atom-optical co-
take distinct approaches in the pursuit of pre- separate chambers (2, 14), and techniques herence reported in this work.
cision and accuracy. Clocks based on single for high-precision nondestructive detection Combining coherent control of the clock
ions use single-particle control and detection have been explored (15, 16). transition with the microscopic control af-
to enable high–duty cycle interrogation of an Interatomic collisions can limit both the forded by optical tweezers may also prove
isolated atom (6, 12). Optical lattice clocks, on precision and accuracy of a clock, but they useful in several subfields of quantum infor-
the other hand, typically interrogate thousands can be mitigated by isolating the atoms from mation science more broadly. Such control
of atoms in parallel to reach exceptionally low one another. This has recently been demon- is necessary for proposals for quantum gates
strated in a Fermi-degenerate optical lattice based on spin-orbital exchange interactions
clock, in which a well-defined number of atoms (25–27). Further, single-photon Rydberg tran-
occupies each lattice site (17). This approach sitions from the excited clock state allow
JILA, University of Colorado and National Institute of relies on evaporative cooling and a quantum access to many-body spin models and gate
Standards and Technology, and Department of Physics,
University of Colorado, Boulder, CO 80309, USA. phase transition in the Hubbard model ground architectures with fast time scales rela-
*Corresponding author. Email: adam.kaufman@colorado.edu state, and therefore typically requires long tive to dissipation rates, which may lead to

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 93

RES EARCH | R E P O R T S

Fig. 1. Clock transition interrogation in an optical tweezer array. interrogation cycles, each ensemble can be interrogated many times before
(A) Apparatus for interrogation of 1S0 to 3P0 “clock” transition in tweezer losing atoms. Bottom: Image averaged over many such ensembles. Inset:
arrays of strontium atoms. Using a high–numerical aperture (NA > 0.65) Narrow-line Rabi spectrum of clock transition retrieved without repeated
objective, we project tightly confining optical potentials to trap single strontium interrogation. Fits to sinc and Gaussian functions are shown in gray and black,
atoms. By tuning these traps near the so-called “magic” wavelength, we can respectively. In this case, a 1.5-s probe yields an approximately Fourier-limited
ensure that the clock transition is minimally sensitive to the local intensity Gaussian linewidth of 450 ± 20 mHz. (C) Ramsey spectroscopy in 200-photon
experienced by the atoms. (B) Repeated interrogation of clock transition. Top: recoil energy (ER) deep tweezers, showing a coherence time of 3.4 ± 0.4 s.
Image of a single ensemble of atoms loaded into tweezers with ~50% filling. The frequency of the fringes is set by the differential light shift imposed on the
After interrogating the clock transition, excitation of the 3P0 state can be inferred clock transition by the probe beam. These data were taken using the repeated
from apparent atom loss in a second image. By repumping atoms between imaging technique outlined in the text.

cooled using three-dimensional sideband cool-

ing (Fig. 1). With each run of our experiment,
individual tweezers have an approximately 50%
probability of being empty or of containing a
single atom; multiple occupancies are sup-
pressed by light-assisted collisions (33). For
the current work, we use 10 traps, generated by
applying 10 radio-frequency tones to an acousto-
optic deflector. Imaging is performed with neg-
ligible atom loss by simultaneously scattering
photons on the broad 1S0 to 1P1 transition at
461 nm and cooling on the narrow-linewidth
1
S0 to 3P1 transition at 689 nm (21, 22, 24).
We interrogate the 1S0 to 3P0 optical “clock”
transition, which is induced in our bosonic
atoms by applying a magnetic field (3, 34),
using light from a highly stable laser re-
ferenced to a crystalline optical cavity (2).
Magnetic field values of 0.14 to 2.2 mT and
probe intensities ranging from 50 mW/cm2
to 5 W/cm2 are used to generate Rabi fre-
quencies from 0.125 Hz for our narrowest-
Fig. 2. Single-site resolved coherence studies. (A) Ramsey contrast as a function of evolution time for
linewidth Rabi spectroscopy up to 17 Hz for
different trap depths, showing tweezer-induced contrast decay. Inset: 1/e Gaussian decay rates for
Ramsey spectroscopy (35). By comparing im-
Ramsey contrast versus tweezer depth, in units of the tweezer photon recoil energy ER. For deep tweezers,
ages of the atomic array taken before and after
contrast decay rate is proportional to tweezer depth. This can largely be explained by variations in the
probing, we infer the excitation to the 3P0 state
frequency of the trapping light across the tweezer array as a result of different radio-frequency tones used to
from the apparent loss of atoms in the 1S0
generate the tweezer spots, whose expected contribution is represented by the black line. The combined
ground state. Because the imaging is highly
expectation of both this effect and Raman scattering, as inferred from measured depopulation rates from
nondestructive (23, 24), we can repeat this in-
3
P0, is illustrated by the gray dashed line (18, 35). (B) Measured relative Ramsey phase shifts for individual
terrogation cycle many times before preparing
traps in 1800-ER tweezers. (C) Inferred relative frequency shifts for individual tweezers at different depths,
a new ensemble of atoms (Fig. 1B). The tweez-
with predictions based on the known sensitivity to trap detuning (37, 38).
er light has a wavelength near 813.4272 nm,
where the light shifts to 1S0 and 3P0 are nearly
improvement over analogous schemes with metrological performance (31, 32), a long- equal [the so-called “magic wavelength” (36)].
alkali species (28–30). Rydberg dressing on the standing goal for optical clocks. The use of a “magic angle” technique (21)
clock transition of microscopically controlled Our platform (21) consists of one-dimensional allows us to also achieve state-insensitive trap-
atoms also provides a clear path to the gen- arrays of neutral, bosonic 88Sr atoms that are ping on the 1S0 to 3P1 transition at this wave-
eration of entangled states with improved tightly confined within optical tweezers and length, enabling sideband cooling to average

94 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

Fig. 3. Repeated nondestructive

interrogation of the clock
transition. (A) Timing diagram for
repeated interrogation of the clock
transition for various clock interroga-
tion times (Tp) and number of
repeated interrogations (nrep): (I)
nrep = 15, Tp = 30 ms; (II) nrep = 8,
Tp = 2 s; (III) nrep = 4, Tp = 4 s. These
repetition numbers are chosen on
the basis of experimental convenience:
High repetition numbers have a more
substantial impact on duty cycle for
short interrogation times, but owing
to stochastic loading, they require
lengthy data scans for long probe
times to achieve uniform statistical
uncertainty. (B) Corresponding Rabi
spectra for sequences I to III. Sinc fits
are shown in gray, Gaussian fits in
black. Shaded regions and labels correspond to Gaussian FWHM.

phonon numbers of 0.2 along the axis of the Fig. 4. Characterization of frequency
clock interrogation. stability. The short-term stability of the
These conditions enable coherent atom- atomic clock signal relative to the interro-
light interactions on seconds-long time scales. gation laser is quantified in terms of the
Figure 1B (inset) shows the inferred excitation Allan deviation of fractional frequency
probability associated with a 1.5-s laser probe fluctuations. For Ramsey interrogation with
pulse as a function of laser frequency, with the 500 ms between pulses, we fit a fractional
intensity of the probe pulse tuned to maximize instability of 4.7 × 10−16 (t/s)–1/2 at short
transfer probability. The full width at half averaging times (red line), consistent with
maximum (FWHM) of this feature is below the expected atom shot noise limit after
500 mHz, extracted by fitting a Gaussian accounting for our measured contrast (blue
function to the excitation probability. We fur- line). The flicker noise and known drift of
ther characterize the atom-light coherence the laser are represented by the black and
through Ramsey spectroscopy, where we scan gray traces, respectively, and the predicted
the duration of the gap between two p/2 pulses combination of atom shot noise, laser noise, and laser drift by the dashed line.
from the clock laser. Because the laser is held at
the probe light–shifted resonance frequency of
the clock transition, we observe oscillations
in the final transfer fraction occurring at the and the Raman scattering from the excited light, we find that the improvements in co-
difference between this frequency and that of state (18). We use the spatial resolution of herence time begin to saturate, indicating the
the bare clock transition. The contrast of these our system to study this dephasing effect at a presence of other decoherence mechanisms
oscillations persists with a 1/e decay time of up single-atom level in Fig. 2, B and C. For the such as residual path-length fluctuations be-
to 3.4 ± 0.4 s (Fig. 1C), providing a measure of deepest tweezers used in Fig. 2A, we fit the tween the clock laser and the atoms. Further,
our atom-light coherence and a bound on our phase of the Ramsey fringes for each site in we find that the initial contrast of oscillations
atom-atom coherence. the array at different hold times. We observe decreases in shallow tweezers, which could be
We observe more rapid decay of Ramsey a relative shift in this phase between tweezers due to residual atomic motion. All spectro-
contrast when operating with deep tweezers, that increases linearly with hold time and with scopic measurements presented in this work
with a rate that depends linearly on the depth the distance from the center of the array (Fig. were performed in tweezers with depths of
of the tweezers (Fig. 2A). This can be largely 2B), from which we extract the relative frequency 200 ER unless otherwise stated.
attributed to slight variations in the optical shifts between atoms in different tweezers. These The alkaline-earth tweezer platform enables
frequencies of the tweezer light across the frequency shifts are plotted in Fig. 2C for se- nondestructive, state-selective imaging (23, 24),
different traps, stemming from the different lected depths, and they agree well with pre- which we can use to interrogate the same
radio-frequency tones applied to the acousto- dictions based on the known derivative of the ensemble of atoms many times and realize
optic deflector (adjacent tweezers are separated differential polarizability of the clock transition, high–duty cycle interrogation of the clock
by 5 MHz). Because of this frequency difference, –15.5 ± 1.1 (mHz/ER)/MHz (37, 38). To eliminate transition. This both improves the rate at
all traps cannot be operated at exactly the magic this dephasing mechanism in the future, the which statistical uncertainty can be aver-
wavelength at the same time. The predicted tweezers could be projected using a digital aged down and mitigates noise aliasing of
magnitude of this effect is illustrated by the mirror device or spatial light modulator, for the clock laser (35). Each interrogation cycle
black line in the inset of Fig. 2A; the dashed which all traps have the same optical frequency. involves 90 ms of dead time during which
gray line is an expectation that shows the For trap depths below 200 times the photon the clock transition is not being interrogated,
combined effect of this non-magic behavior recoil energy ER associated with the tweezer compared to approximately 300 ms to load

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 95

RES EARCH | R E P O R T S

new atoms into the tweezers (Fig. 3). For short tweezers for clock interrogation, ensemble effects associated with the tweezer platform,
(30 ms) interrogation pulses, we observe a loss sizes of 500 atoms or larger could plausibly as well as those unforeseen, will be critical to
probability of 0.001 (±0.001) per cycle. For be achieved, which would enable stability ex- future studies of clock accuracy.
longer cycle times (up to 8 s), we observe a ceeding the current state of the art. The ability to control and measure the po-
total loss probability consistent with 0.01 per In addition to enabling one to take full ad- sitions and states of individual atoms may
second for atoms prepared in either the ground vantage of the reduced projection noise asso- enable opportunities both for extending single-
or excited state, likely caused by collisions with ciated with large numbers of atoms, the low atom coherence and for using ensembles of
the background gas. dead times provided by our tweezer platform many atoms in previously unexplored ways.
The coherence properties of clock interro- would allow one to achieve high performance At a single-particle level, the spatial control
gation with repeated imaging are consistent with a less stable clock laser. This capability afforded by the tweezers could allow one to
with those when each set of atoms is inter- could be especially transformative for high- simultaneously suppress tunneling and light-
rogated only once. Ramsey contrast curves with performance portable optical clocks (7), where induced decoherence to push coherence times
coherence times of 3 s have been measured technical constraints limit the performance of beyond 10 s (18). When using many atoms,
using both repeated and single interrogation, the clock laser (39). In this context, our re- the microscopic control afforded by the tweez-
as shown in Figs. 1C and 2A, respectively. peated imaging technique would allow for ers, combined with interactions introduced
Figure 3 characterizes this repeated inter- relatively high duty cycles even with short using Rydberg dressing (31, 32), could facili-
rogation technique for different clock probe interrogation times, thereby greatly improving tate the creation of entangled states that can
durations (Tp) and numbers of cycles (nrep) the stability associated with both quantum pro- surpass the limitations set by atomic projec-
performed using the same ensemble of atoms. jection noise and laser noise aliasing (35). tion noise, for fundamental studies of entan-
In Fig. 3B, we use interrogation pulse dura- For use as an absolute frequency reference, glement on an optical clock transition as well
tions up to 4 s to demonstrate duty cycles as systematic shifts to the clock transition fre- as quantum-enhanced high-bandwidth sen-
high as 96% and clock spectra as narrow as quency must be carefully considered. Many of sors. Future extensions of the imaging and
350 mHz. the shifts present in the tweezer platform, par- coherent manipulation techniques demon-
We characterize the short-term stability of ticularly environmental perturbations such strated here, in which sub-ensembles of the
the atomic frequency reference by computing as blackbody radiation, are identical to those atoms are manipulated independently, could
the Allan deviation relative to the free-running present in optical lattice clocks and have been also allow for techniques that extend interro-
clock laser (Fig. 4). This measurement is per- studied in detail (10, 11). In this work, we use gation time beyond the coherence time of the
formed using single-interrogation (nrep = 1) bosonic 88Sr atoms. Although this choice is laser (14) in systems using a single vacuum
Ramsey spectroscopy with evolution times of not fundamental to our tweezer platform, it chamber.
501 ms, chosen to bias the signal to the most does require consideration of different sys- Note added in proof: During review of this
sensitive part of the Ramsey fringe. The clock tematic effects from the more commonly used manuscript, we became aware of related
laser was not actively stabilized to the atomic fermions. Shifts caused by atomic collisions work (41).
transition, which limits the total length of the are of particular concern in clocks operating
dataset to 500 s. Under these conditions, we with bosonic atoms (3) but are mitigated in
REFERENCES AND NOTES
measure a fractional frequency instability of the tweezer platform by single-atom occupancy.
1. W. F. McGrew et al., Nature 564, 87–90 (2018).
4.7 × 10−16 (t/s)–1/2 at short times. As expected, The use of the bosonic isotope requires higher
2. E. Oelker et al., Nat. Photonics 10.1038/s41566-019-0493-4
this is dominated by the atomic quantum pro- probe intensities and a relatively large applied (2019).
jection noise associated with the average atom magnetic field to reach a given Rabi frequency, 3. S. Origlia et al., Phys. Rev. A 98, 053443 (2018).
number of 4.8 per trial, whereas the long-term so second-order Zeeman and probe-induced 4. I. Ushijima, M. Takamoto, M. Das, T. Ohkubo, H. Katori,
Nat. Photonics 9, 185–189 (2015).
instability is consistent with known drifts in Stark shifts can be important. For narrowline 5. B. J. Bloom et al., Nature 506, 71–75 (2014).
the frequency of our laser (35). Rabi spectroscopy, our probe intensities and 6. C. W. Chou, D. B. Hume, J. C. J. Koelemeij, D. J. Wineland,
Currently, our demonstrated instability falls magnetic fields are similar to those of (3), T. Rosenband, Phys. Rev. Lett. 104, 070802 (2010).
between that attained with single ions (6) and where their contributions to clock inaccuracy 7. J. Grotti et al., Nat. Phys. 14, 437–441 (2018).

were bounded to the 10−17 level. Unlike con-

8. C. W. Chou, D. B. Hume, T. Rosenband, D. J. Wineland, Science
in optical lattice clocks operating with thou- 329, 1630–1633 (2010).
sands of atoms (1–4). Achieving stability com- ventional optical lattices, tweezer confinement 9. A. Arvanitaki, J. Huang, K. Van Tilburg, Phys. Rev. D 91, 015015
petitive with the most stable optical lattice relies on tightly focused beams that can have (2015).
clocks [4.8 × 10−17 (t/s)–1/2 (2)] could be achieved nonuniform polarization at the location of 10. A. D. Ludlow, M. M. Boyd, J. Ye, E. Peik, P. O. Schmidt, Rev.
Mod. Phys. 87, 637–701 (2015).
with increased atom number—for example, the atoms. Such inhomogeneous polarization 11. N. Poli, C. W. Oates, P. Gill, G. M. Tino, Riv. Nuovo Cim. 36,
by using a dark time of 2 s and approximately could complicate magic trapping in fermionic 555–624 (2013).
150 atoms. Because of our high duty cycle and isotopes, but its effect is expected to be sup- 12. P. O. Schmidt et al., Science 309, 749–752 (2005).
13. G. Dick, in Proceedings of the 34th Annual Precise Time and
low-noise laser system, Dick effect noise would pressed in bosons, whose clock states have
Time Interval Systems and Applications Meeting (ION, 1987),
contribute only modestly (16% and 6% of noise zero net angular momentum (40). Finally, in pp. 133–147; https://tycho.usno.navy.mil/ptti/1987papers/
variance for individual and 16 times–repeated our current setup there are several meters of Vol%2019_13.pdf.
interrogation, respectively). Although we dem- optical path between the atoms and the sur- 14. M. Schioppo et al., Nat. Photonics 11, 48–52 (2017).
15. G. Vallet et al., New J. Phys. 19, 083002 (2017).
onstrate coherence times exceeding 2 s in this face to which the phase of the interrogation 16. M. A. Norcia, J. K. Thompson, Phys. Rev. A 93, 023804
work, attaining this atom number would re- laser is referenced. Fluctuations in this path (2016).
quire more power than is available with our length lead to Doppler shifts, which can limit 17. S. L. Campbell et al., Science 358, 90–94 (2017).
18. R. B. Hutson, A. Goban, G. E. Marti, J. Ye, arXiv 1903.02498
current system (here, we use 13 ± 1 mW per stability and, in principle, accuracy. In anal-
[physics.atom-ph] (6 March 2019).
tweezer spot into the objective for cooling and ogy to techniques used in optical lattice clocks, 19. A. Fuhrmanek, R. Bourgain, Y. R. Sortais, A. Browaeys,
imaging). However, by performing atomic pre- this could be addressed by referencing the Phys. Rev. Lett. 106, 133003 (2011).
paration and imaging in tweezers at a wave- phase of the interrogation laser to a surface 20. M. J. Gibbons, C. D. Hamley, C.-Y. Shih, M. S. Chapman,
Phys. Rev. Lett. 106, 133002 (2011).
length with higher polarizability, smaller spot rigidly connected to the microscope objective, 21. M. A. Norcia, A. W. Young, A. M. Kaufman, Phys. Rev. X 8,
size, and higher available power (such as as this primarily defines the atoms’ locations. 041054 (2018).
515 nm) and then transferring into 813-nm Characterization of these potential systematic 22. A. Cooper et al., Phys. Rev. X 8, 041055 (2018).

96 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

23. S. Saskin, J. T. Wilson, B. Grinkemeyer, J. D. Thompson, 35. See supplementary materials. Science Foundation Physics Frontier Center at JILA (1734006),
. Phys. Rev. Lett. 122, 143002 (2019). 36. J. Ye, H. J. Kimble, H. Katori, Science 320, 1734–1738 and NIST. M.A.N. and E.O. acknowledge support from the
24. J. P. Covey, I. S. Madjarov, A. Cooper, M. Endres, (2008). NRC research associateship program. Author contributions:
Phys. Rev. Lett. 122, 173201 (2019). 37. C. Shi et al., Phys. Rev. A 92, 012516 (2015). M.A.N., A.W.Y., W.J.E., and A.M.K. designed, constructed, and
25. D. Hayes, P. S. Julienne, I. H. Deutsch, Phys. Rev. Lett. 98, 38. T. Takano, R. Mizushima, H. Katori, Appl. Phys. Express 10, operated the tweezer platform. E.O. and J.Y. designed,
070501 (2007). 072801 (2017). constructed, and operated the stable laser system. All authors
26. A. J. Daley, M. M. Boyd, J. Ye, P. Zoller, Phys. Rev. Lett. 101, 39. S. Koller et al., Phys. Rev. Lett. 118, 073601 (2017). contributed to analysis, discussion of results, and preparation of
170504 (2008). 40. A. Taichenachev, V. Yudin, C. W. Oates, Phys. Rev. A 76, the manuscript. Competing interests: The authors have no
27. G. Pagano, F. Scazza, M. Foss-Feig, Adv. Quantum Technol. 2, 023806 (2007). competing interests to declare. Data and materials availability:
1800067 (2019). 41. I. S. Madjarov et al., arXiv 1908.05619 [physics.atom-ph] The data can be accessed at Zenodo (42).
28. H. Bernien et al., Nature 551, 579–584 (2017). (15 August 2019).
29. T. Wilk et al., Phys. Rev. Lett. 104, 010502 (2010). 42. M. A. Norcia et al., Data for “Seconds-scale coherence on an SUPPLEMENTARY MATERIALS
30. L. Isenhower et al., Phys. Rev. Lett. 104, 010503 (2010). optical clock transition in a tweezer array”; doi:10.5281/ science.sciencemag.org/content/366/6461/93/suppl/DC1
31. L. I. R. Gil, R. Mukherjee, E. M. Bridge, M. P. A. Jones, T. Pohl, zenodo.3381664. Materials and Methods
Phys. Rev. Lett. 112, 103601 (2014). Supplementary Text
32. R. Kaubruegger et al., arXiv 1908.08343 [quant-ph] (22 August ACKN OWLED GMEN TS Figs. S1 to S5
2019). We thank T. Bothwell, M. Brown, J. Cline, A. Goban, R. Hutson, Table S1
33. N. Schlosser, G. Reymond, I. Protsenko, P. Grangier, Nature B. Johnston, D. Kedar, C. Kennedy, A. Ludlow, W. Milner, C. Regal,
411, 1024–1027 (2001). J. Robinson, C. Sanner, J. A. Muniz Silva, L. Sonderhouse, 16 May 2019; accepted 3 September 2019
34. A. V. Taichenachev et al., Phys. Rev. Lett. 96, 083001 J. K. Thompson, and J. Thompson for discussions and support. Published online 12 September 2019
(2006). Funding: Supported by ARO, AFOSR, DARPA, the National 10.1126/science.aay0644

INTERGALACTIC MEDIUM ground source. The low sky density of suf-

ficiently bright background sources prevents
Gas filaments of the cosmic web located study of the cosmic web on scales finer than a
few megaparsec (Mpc) (7).
around active galaxies in a protocluster Imaging the cosmic web in emission would
provide two-dimensional (2D) information.
H. Umehata1,2*, M. Fumagalli3,4,5, I. Smail3, Y. Matsuda6,7, A. M. Swinbank3, S. Cantalupo8, C. Sykes3,4, Filaments are predicted to emit the hydrogen
R. J. Ivison9,10, C. C. Steidel11, A. E. Shapley12, J. Vernet9, T. Yamada13, Y. Tamura14, M. Kubo6, Lyman-alpha (Lya) line by means of the fluo-
K. Nakanishi6,7, M. Kajisawa15, B. Hatsukade2, K. Kohno2 rescence induced by the ultraviolet background
(UVB) radiation (8). The intrinsically low in-
Cosmological simulations predict that the Universe contains a network of intergalactic gas filaments, tensity of the UVB means the expected sur-
within which galaxies form and evolve. However, the faintness of any emission from these filaments face brightness of a filament is ~2.5 × 10–20 erg
has limited tests of this prediction. We report the detection of rest-frame ultraviolet Lyman-a radiation s−1 cm−2 arcsec−2 at z ~3 (9), so direct de-
from multiple filaments extending more than one megaparsec between galaxies within the SSA22 tection of UVB-induced fluorescent emission
protocluster at a redshift of 3.1. Intense star formation and supermassive black-hole activity is occurring from IGM filaments has remained elusive (10).
within the galaxies embedded in these structures, which are the likely sources of the elevated ionizing To circumvent this limitation, one can examine
radiation powering the observed Lyman-a emission. Our observations map the gas in filamentary regions where local ionizing sources, such
structures of the type thought to fuel the growth of galaxies and black holes in massive protoclusters. as star-forming galaxies and/or active galac-
tic nuclei (AGNs), boost the local radiation

C
field and hence elevate the Lya emission to
osmological simulations of structure for- structure and for the formation and evolu- detectable levels (11). Extended (up to hun-
mation predict that the majority of gas tion of galaxies. dreds of kiloparsec) Lya nebulae have been
in the intergalactic medium (IGM) is Galaxy formation models predict that at observed around quasars, with morphologies
distributed in a cosmic web of sheets a redshift (z) of ~3, >60% of all gas in the and kinematics suggestive of cosmic web fil-
and filaments as a consequence of grav- Universe resides in filaments (5). However, aments connecting to the quasar host galaxies
itational collapse (1). The intersections of their low density makes them difficult to ob- (12–15). Similarly, using Lya-emitting galaxies
these structures become the locations at which serve in emission. Absorption spectroscopy (LAEs) or extended emission arising from the
galaxies and their supermassive black holes using background sources, such as quasars, circumgalactic medium (CGM) as tracers, sta-
(SMBHs) form and evolve (2). Streams of cool has been the primary method used to trace tistical evidence for filaments has been re-
gas falling along IGM filaments, driven by neutral hydrogen (H I) in the IGM (6, 7), which ported (16–18). These studies do not directly
gravity, are predicted to provide most of the has provided insights into the nature of the connect the cosmic web to the population of
gas required for the growth of galaxies and cosmic web (6). Nevertheless, it has not been galaxies and SMBHs on cosmological scales.
SMBHs (3, 4). Direct detection of the cosmic possible to obtain a detailed picture of these We searched for extended filamentary struc-
web in the early Universe would allow tests filaments, as information is limited to one tures using the Multi Unit Spectroscopic Ex-
of these predictions, both for the large-scale dimension along the line of sight to the back- plorer (MUSE) on the European Southern

1
RIKEN Cluster for Pioneering Research, 2-1 Hirosawa, Wako-shi, Saitama 351-0198, Japan. 2Institute of Astronomy, School of Science, The University of Tokyo, 2-21-1 Osawa, Mitaka, Tokyo 181-
0015, Japan. 3Centre for Extragalactic Astronomy, Department of Physics, Durham University, South Road, Durham DH1 3LE, UK. 4Institute for Computational Cosmology, Durham University,
South Road, Durham DH1 3LE, UK. 5Dipartimento di Fisica G. Occhialini, Università degli Studi di Milano Bicocca, Piazza della Scienza 3, 20126 Milano, Italy. 6National Astronomical Observatory
of Japan, 2-21-1 Osawa, Mitaka, Tokyo 181-8588, Japan. 7Department of Astronomy, School of Science, SOKENDAI (The Graduate University for Advanced Studies), Osawa, Mitaka, Tokyo 181-
8588, Japan. 8Department of Physics, ETH Zurich, Wolfgang-Pauli-Strasse 27, 8093 Zurich, Switzerland. 9European Southern Observatory, Karl-Schwarzschild-Str. 2, D-85748 Garching, Germany.
10
Institute for Astronomy, University of Edinburgh, Royal Observatory, Blackford Hill, Edinburgh EH9 3HJ, UK. 11Cahill Center for Astronomy and Astrophysics, California Institute of Technology,
MS 249-17, Pasadena, CA 91105, USA. 12Department of Physics and Astronomy, University of California, Los Angeles, 430 Portola Plaza, Los Angeles, CA 90095, USA. 13Institute of Space and
Aeronautical Science, Japanese Aerospace Exploration Agency, 3-1-1, Yoshinodai, Chuo-ku, Sagamihara, Kanagawa 252-5210, Japan. 14Division of Particle and Astrophysical Science, Graduate
School of Science, Nagoya University, Nagoya 464-8602, Japan. 15Research Center for Space and Cosmic Evolution, Ehime University, Bunkyo-cho 2-5, Matsuyama 790-8577, Japan.
*Corresponding author. Email: hideki.umehata@riken.jp

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 97

RES EARCH | R E P O R T S

Observatory’s Very Large Telescope (VLT). the core, which was previously mapped at with Suprime-Cam on the Subaru telescope
Our observations targeted the galaxy proto- 1.1 mm with the Atacama Large Millimeter/ (22). We identified extended Lya emission
cluster SSA22 at z = 3.09 (19), which was submillimeter Array (ALMA) [the ALMA Deep with surface brightness SLya > ~3 × 10–19 erg
already known to host 3D filamentary struc- Field in SSA22 (ADF22)] (21). To trace ex- s−1 cm−2 arcsec−2 across the observed field,
tures as traced by LAEs on a scale of 30 co- tended Lya emission in this region, we mapped visible in the optimally extracted Lya map
moving megaparsec (comoving distances the ADF22 field with a six-pointing MUSE [Fig. 2 (22)]. This map shows bright areas
remain constant over time if the two objects mosaic covering 116′′ by 169′′, equivalent to 0.9 associated with the CGM of galaxies, along
are moving with the expansion of the Universe) by 1.3 physical Mpc at z = 3.09 [Fig. 1 (22)]. with several patches of emission at low sur-
(20). At the intersection of these large-scale We searched the MUSE data cube for ex- face brightness that connect to, but are not
structures lies the protocluster core, where tended Lya emission in conjunction with a immediately associated with, individual gal-
several intensely star-forming galaxies are narrow-band image covering the expected axies in this region. Most of this low-surface-
known to lie within a 2′ by 3′ region around wavelength of redshifted Lya emission taken brightness Lya emission forms two main
filaments running in a north-south direction,
each with a total extent of >1 physical mega-
parsec in projection. The scale of this emission
far exceeds the expected size of the dark-
matter halo of even the most massive indi-
vidual galaxies at this epoch (the halo radius
is ~100 kiloparsec for a 1012.5 M⊙ halo at z~3,
where M⊙ is the mass of the Sun), so the Lya
signal likely traces a structure connecting sev-
eral galaxies. This network of filaments likely
extends beyond the region that we mapped,
because the Lya emission is detected up to
the edge of the MUSE field of view. As shown
in Fig. 3, A and B, the majority of the Lya emis-
sion is detected over a line-of-sight velocity
range of ~ 500 to ~+1000 km s−1 relative
to z = 3.09. This velocity range reflects not
only the 3D distribution of matter on large
scales, but also the gas kinematics within the
Fig. 1. Multiwavelength images of ADF22 illustrating the overdensity of galaxies and AGNs in a nar- protocluster core, which, coupled with radi-
row redshift range at z = 3.09. Each panel is centered at (a, d) = (22h17m34.0s, +00d17m00s), where a is ative transfer effects, can produce velocity
right ascension and d is declination, and 2′ by 3′ in size, with the inner dashed area showing the MUSE gradients of several hundreds of kilometers
coverage, 116′′ by 169′′ (0.9 by 1.3 Mpc at z ~3.1). North is up and east is left. (A) Pseudocolor map created per second.
from the MUSE cube (synthetized V-, R-, and i′-bands are used for the blue, green, and red channels, Observations of the SSA22 protocluster have
respectively). (B) The 1.1-mm ALMA continuum map of ADF22 (22). Identified sources at z = 3.09 are marked detected 35 Lya blobs (LABs), defined as ex-
with white circles (SMGs) and squares (AGNs); positions and redshifts are listed in table S2. (C) Pseudocolor tended Lya nebulae with sizes between several
map of the Chandra x-ray data; 2 to 8 keV (hard band), 0.5 to 8 keV (full band), and 0.5 to 2 keV (soft band) are tens and several hundreds of kiloparsec (23–25).
used for the blue, green, and red channels, respectively. Two of these LABs lie within our field of view,
each with sizes of ~40 kiloparsec when mea-
sured at a Lya surface brightness threshold of
Fig. 2. Lya emission map optimally SLya = 2.2 × 10–18 erg s−1 cm−2 arcsec−2 (24).
extracted from the MUSE observations Figure 2 shows that these two LABs are parts
and covering the same field as that in of a larger network of megaparsec-scale fila-
Fig. 1. Lya emissions largely compose two ments. Embedded in these filaments are also
groups of filamentary structures for >1 other patches of enhanced Lya emission, some
physical megaparsec. One high-surface- of which are associated with galaxies. We in-
brightness filament is visible running north terpret the previously known LABs as bright
to south on the west side of the field, knots within a wider network of gas filaments,
whereas a fainter (and hence more frag- and surmise that the fainter and more ex-
mented) structure runs north to south up tended Lya-emitting gas in these filaments
the east side of the field. Contours with has previously eluded detection because of its
solid, dashed, and dotted lines show Lya low surface brightness (18, 26).
surface brightness levels of SLya = 0.3, 1.0, To explore the link between these filaments
and 2.0 × 10–18 erg s−1 cm−2 arcsec−2, and the associated galaxy population, we mea-
respectively (these correspond to 2, 7, and sured redshifts of galaxies in this field using
13 s above the representative noise level). a multiwavelength spectroscopic dataset. The
Navy blue contours indicate the extent of deep, 1.1-mm ADF22 map enables us to iden-
the two LABs in this field (23). Positions tify submillimeter galaxies (SMGs), which are
of SMGs and x-ray–luminous AGNs at z = massive, intense starburst galaxies with large
3.09 are also shown. The large, filled black amounts of gas and dust in their interstellar
circle shows data removed around a mediums. The x-ray–luminous AGNs, which
foreground, low-redshift galaxy. host growing SMBHs, were identified from

98 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

simple assumption that the ionizing sources

typically lie 250 kpc from filaments, the re-
quired photon flux corresponds to a photon
number emission rate of Qion ~1055 s−1 to
power the whole filament. The eight x-ray
AGNs in the structure have Lx ~1044 erg s−1,
corresponding to a total rate of Qion ~1057 s−1,
whereas the 16 SMGs that are protocluster
members form stars at a rate of 160 to 1700
M⊙ year−1 and hence produce a total of Qion
~1057 s−1. This is sufficient ionizing photon
flux to power the filament emission, even un-
der the assumption that only 1% of the pho-
tons escape their host galaxies. This simple
estimate, although an approximation of the
more complex radiative transfer in this re-
gion, supports our interpretation that the
gas residing in these filamentary structures
is ionized by the photons produced by star-
forming galaxies and AGNs in the massive
protocluster core.
The volume densities of SMGs and x-ray
AGNs in this field are about three orders of
magnitude higher than the volume average
Fig. 3. Three-dimensional pictures of Lya filaments. (A) Velocity map of the Lya emission obtained
at this epoch (21). Such an overdensity of ac-
from its flux-weighted centroid in the MUSE data. Image scale and plotting symbols are the same as those
tive populations is very rare (29), and there is
in Fig. 2. Coherent velocity trends can be seen along the filament structures. (B) The 3D distribution of
little observational evidence regarding how
Lya filaments shown with blue [signal-to-noise ratio (SNR) > 2] and magenta (SNR > 5) voxels. The locations
this intense activity is fueled and sustained.
of SMGs (without detectable x-ray AGNs, orange circles), AGN-hosting SMGs (red diamonds), and
Cosmological simulations suggest that rapid
x-ray–luminous AGNs without ALMA 1-mm detections (brown hexagons) are also displayed. The Lya
infall of gas from the cosmic web in proto-
filaments and SMGs and AGNs are colocated on megaparsec scales.
clusters may lead to the formation of SMGs
(30). Although gas inflows are not directly
observations using the Chandra space tele- the ionizing radiation field that is >12 times observable in our data, the location of SMGs
scope (22, 27). Spectroscopic redshifts for these brighter than that predicted by UVB models and AGNs within the filaments supports the
populations were determined using a combi- (22). This corresponds to an ionizing photon idea that large reservoirs of gas are funneled
nation of emission lines [CO J = 3 → 2, Hb, and flux ϕ > 4 × 106 cm−2 s−1. If we assume that toward forming galaxies under the effect of
(O III) 4959 and 5008 Å lines] from ALMA data the gas is fully ionized, then the observed gravity, triggering and sustaining their star
and observations with the Multi-Object Spec- surface brightness would imply even higher formation and driving the growth and ac-
trometer for Infra-Red Exploration (MOSFIRE) photon fluxes, densities of ~6 × 10–3 cm−3 (12) tivity of their central SMBHs. Assuming a
on the Keck I telescope (22). We confirmed for gas at T ~ 104 K (where T is temperature), typical density of 6 × 10–3 cm−3 for filaments
that 16 SMGs and 8 x-ray–luminous AGNs and a typical filament width of 100 kpc (Fig. 2). with (projected) thicknesses of ~100 kiloparsec,
are protocluster members, with redshifts Fluctuations in the observed surface bright- the region imaged by our observations con-
3.085 ≤ z ≤ 3.098 (table S2). All of the SMGs ness suggest a variable density across the tains ~1012 M⊙ of gas (depending on the filling
and x-ray–luminous AGNs were distributed structure, as commonly found for bright Lya factor of the gas), which is potentially available
within the same structure (see Figs. 2 and 3), nebulae (12). Figure 2 also shows regions at to accrete onto galaxies in this region and so
closely tracking the Lya filaments both spa- much higher surface brightness, particularly fuel their continuing star formation (22).
tially (in projection) and in velocity (fig. S10). overlapping with galaxies (see also fig. S7). Our observations have uncovered a large-
A similar pattern was also evident for normal The fainter emission regions display lower scale filamentary structure in the emission
star-forming galaxies and LAEs (fig. S7). We velocity dispersions [full width at half maxi- from the core of the SSA22 protocluster. Evi-
interpret this close alignment as evidence mum (FWHM) ~150 km s−1] than the brighter dence of similar structures in other proto-
that the Lya filaments are directly linked to knots (FWHM ~730 km s−1), where the surface clusters from imaging observations (15, 25)
the population of active galaxies and SMBHs. brightness of the latter is similar to the typical suggests that this may be a general feature
Gas filaments are thought to supply (under brightness of the LABs (figs. S4 and S9). This of protoclusters in the early Universe. The
the effect of gravity) the fuel for active SMGs higher surface brightness may indicate the network of filaments in SSA22 is found to
and x-ray–luminous AGNs. presence of localized sources of ionization, connect individual galaxies across a large vol-
The filaments have Lya brightness above such as star formation or AGN activity, as ume, allowing it to power star formation and
the level expected from fluorescent emission commonly seen in LABs (28). black-hole growth in active galaxy populations
induced by the UVB (8, 11). Radiative transfer Given the large number of active galaxies in at z ~3.
calculations predict a maximum surface bright- this region, the elevated photon flux required
ness from optically thick gas of ~2.5 × 10–20 erg to power the filaments could be provided by REFERENCES AND NOTES
s−1 cm−2 arcsec−2 assuming a z~3 UVB (9). Our the galaxy population identified within our 1. J. R. Bond, L. Kofman, D. Pogosyan, Nature 380, 603–606 (1996).
observations contain emission at levels of SLya field. To test this hypothesis, we determined the 2. V. Springel et al., Nature 435, 629–636 (2005).
3. A. Dekel et al., Nature 457, 451–454 (2009).
≥ 3 × 10–19 erg s−1 cm−2 arcsec−2, which, in the number of ionizing photons provided by x-ray– 4. M. Fumagalli et al., Mon. Not. R. Astron. Soc. 418, 1796–1821
optically thick limit, requires an intensity of selected AGNs and SMGs (22). Under the (2011).

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 99

RES EARCH | R E P O R T S

5. D. Martizzi et al., Mon. Not. R. Astron. Soc. 486, 3766–3787 journal’s authorship criteria. Competing interests: The authors SUPPLEMENTARY MATERIALS
(2019). declare no competing interests. Data and materials availability: science.sciencemag.org/content/366/6461/97/suppl/DC1
6. M. Rauch, Annu. Rev. Astron. Astrophys. 36, 267–316 (1998). The MUSE data are available in the ESO archive at http://archive.eso. Materials and Methods
7. K.-G. Lee, M. White, Astrophys. J. 831, 181 (2016). org/cms.html under programs 099.A-0638 and 0101.A-0679; the Figs. S1 to S10
8. J. A. Kollmeier et al., Astrophys. J. 708, 1048–1075 (2010). ALMA data are archived at http://almascience.nrao.edu/aq/ under Tables S1 and S2
9. F. Haardt, P. Madau, Astrophys. J. 746, 125–144 (2012). project codes 2013.1.00162.S, 2015.1.00212.S, and 2016.1.00543.S; References (31–76)
10. S. G. Gallego et al., Mon. Not. R. Astron. Soc. 475, 3854–3869 and MOSFIRE observations are in the Keck Observatory Archive Data Files S1 and S2
(2018). https://www2.keck.hawaii.edu/koa under semester 2017B, program
11. S. Cantalupo, C. Porciani, S. J. Lilly, F. Miniati, Astrophys. J. ID S412. Input parameter files for our CLOUDY simulations are provided 25 January 2019; accepted 2 September 2019
628, 61–75 (2005). in data files S1 and S2. 10.1126/science.aaw5949
12. S. Cantalupo, F. Arrigoni-Battaia, J. X. Prochaska,
J. F. Hennawi, P. Madau, Nature 506, 63–66 (2014).
13. D. C. Martin et al., Nature 524, 192–195 (2015).
14. E. Borisova et al., Astrophys. J. 831, 39–57 (2016).
15. S. Kikuta, Y. Matsuda, R. Cen, C. C. Steidel, M. Yagi, RESPIRATORY ENZYMES
T. Hayashino, M. Imanishi, Y. Komiyama, R. Momose,
T. Saito, Lya view around a z=2.84 hyperluminous QSO
at a node of the cosmic web. arXiv:1904.07747
Active site rearrangement and structural divergence
[astro-ph.GA] (16 April 2019).
16. M. Fumagalli et al., Mon. Not. R. Astron. Soc. 462, 1978–1988
(2016).
in prokaryotic respiratory oxidases
17. L. Wisotzki et al., Nature 562, 229–232 (2018).
18. D. K. Erb, M. Bogosavljević, C. C. Steidel, Astrophys. J. 740,
S. Safarian1*, A. Hahn2*, D. J. Mills2, M. Radloff1, M. L. Eisinger1, A. Nikolaev3, J. Meier-Credo1,
L31–L36 (2011). F. Melin3, H. Miyoshi4, R. B. Gennis5, J. Sakamoto6, J. D. Langer1, P. Hellwig3,7,
19. C. Steidel et al., Astrophys. J. 492, 428–438 (1998). W. Kühlbrandt2†, H. Michel1†
20. Y. Matsuda et al., Astrophys. J. 634, L125–L128 (2005).
21. H. Umehata et al., Astrophys. J. 815, L8–L13 (2015).
22. See supplementary materials.
Cytochrome bd–type quinol oxidases catalyze the reduction of molecular oxygen to water in the
23. C. C. Steidel et al., Astrophys. J. 532, 170–182 (2000). respiratory chain of many human-pathogenic bacteria. They are structurally unrelated to mitochondrial
24. Y. Matsuda et al., Astron. J. 128, 569–584 (2004). cytochrome c oxidases and are therefore a prime target for the development of antimicrobial drugs. We
25. D. Christopher Martin et al., Astrophys. J. 786, 107–135 determined the structure of the Escherichia coli cytochrome bd-I oxidase by single-particle cryo–electron
(2014).
26. M. Fumagalli et al., Mon. Not. R. Astron. Soc. 471, 3686–3698
microscopy to a resolution of 2.7 angstroms. Our structure contains a previously unknown accessory
(2017). subunit CydH, the L-subfamily–specific Q-loop domain, a structural ubiquinone-8 cofactor, an active-site
27. B. D. Lehmer et al., Astrophys. J. 691, 687–695 (2009). density interpreted as dioxygen, distinct water-filled proton channels, and an oxygen-conducting
28. J. E. Geach et al., Astrophys. J. 700, 1–9 (2009). pathway. Comparison with another cytochrome bd oxidase reveals structural divergence in the family,
29. C. M. Casey, Astrophys. J. 824, 36–51 (2016).
30. D. Narayanan et al., Nature 525, 496–499 (2015).
including rearrangement of high-spin hemes and conformational adaption of a transmembrane helix to
generate a distinct oxygen-binding site.
ACKN OW LEDG MEN TS

C
We thank the reviewers for their constructive comments, which
were very helpful in improving this paper. Our data are based on
observations collected at the European Organisation for
ytochrome bd oxidases are quinol- them ideal targets for antimicrobial drug de-
Astronomical Research in the Southern Hemisphere. We thank all dependent terminal oxidases found ex- velopment (3). The cytochrome bd oxidases
ESO staff who supported us in observation preparation and clusively in prokaryotes. They catalyze are divided into two subfamilies, featuring
execution. We thank the JAO and EA-ARC staffs for preparation,
observation, and initial data reduction. ALMA is a partnership
the reduction of molecular oxygen to either a short (S) or a long (L) quinol-binding
of ESO (representing its member states), NSF (United States), and water in the respiratory chain without domain (Q-loop) (4). Most enterobacterial
NINS (Japan), together with NRC (Canada), NSC and ASIAA formation of reactive oxygen species (ROS). pathogens that cause acute infectious diseases,
(Taiwan), and KASI (Republic of Korea), in cooperation with the
This exergonic reaction is coupled to the such as Salmonella, uropathogenic E. coli
Republic of Chile. The Joint ALMA Observatory is operated by ESO,
AUI/NRAO, and NAOJ. Some of the data were obtained at the generation of an electrochemical proton gra- (UPEC), and enterohemorrhagic E. coli (EHEC),
W. M. Keck Observatory, which is operated as a scientific dient across the periplasmic membrane by rely on L-subfamily bd oxidases for their sur-
partnership among the California Institute of Technology, the vectorial release and uptake of protons (1). vival (5–7).
University of California, and the National Aeronautics and Space
Administration. We thank S. Yeh for assistance on the MOSFIRE Cytochrome bd oxidases are unrelated to the Using cryo–electron microscopy (cryo-EM), we
observations. The observatory was made possible by the generous heme-copper oxidases (HCO), including the determined the structure of the E. coli L-subfamily
financial support of the W. M. Keck Foundation. The authors cytochrome c oxidases of mitochondria (2). bd-I oxidase in lipid nanodiscs with a bound Fab
recognize and acknowledge the very important cultural role and
reverence that the summit of Mauna Kea has always had within the Cytochrome bd oxidases have a very high oxy- fragment to 2.7 Å resolution (Fig. 1A and figs.
indigenous Hawaiian community. We are most fortunate to have gen affinity and are indispensable for patho- S1 and S2). The hetero-oligomeric bd oxidase
had the opportunity to conduct observations from this mountain. genic bacteria during host infection, making consists of a pseudo-symmetric CydAB core
Funding: H.U., Y.M., B.H., Y.T., and K.K. are supported by JSPS
KAKENHI (grant nos. 17K14252, 25287043/17H04831/17KK0098,
dimer and two accessory single-transmembrane
19K03925, 17H06130, and 17H06130, respectively). Y.T. subunits, CydX and CydH (Fig. 1 and fig. S3)
1
acknowledges support from NAOJ (ALMA Scientific Research Department of Molecular Membrane Biology, Max Planck (8, 9). Subunits CydA and CydB share a common
grant no. 2018-09B). M.F., I.S., and A.M.S. acknowledge Institute of Biophysics, D-60438 Frankfurt/Main, Germany.
support from the Science and Technology Facilities Council (grant
2
Department of Structural Biology, Max Planck Institute of
architecture of two four-helix bundles and an
no. ST/P000541/1). This project has received funding from the Biophysics, D-60438 Frankfurt/Main, Germany. 3Laboratoire additional peripheral helix (Fig. 1C). The oxy-
European Research Council (ERC) under the European Union's de Bioélectrochimie et Spectroscopie, UMR 7140, Chimie de gen reduction reaction is confined to CydA,
Horizon 2020 research and innovation program (grant no. la Matière Complexe, Université de Strasbourg-CNRS, 67000
Strasbourg, France. 4Division of Applied Life Sciences,
which contains all three heme cofactors and
757535). C.S. acknowledges an STFC studentship (grant no.
ST/R504725/1). S.C. gratefully acknowledges support from the Graduate School of Agriculture, Kyoto University, Sakyo-ku, the periplasmically exposed Q-loop (10). The
Swiss National Science Foundation (grant no. PP00P2_163824). Kyoto 606-8502, Japan. 5Department of Biochemistry, low-spin heme b558 is the initial electron ac-
Author contributions: H.U. led the project and analyzed the data University of Illinois, Urbana, IL 61801, USA. 6Department of
ceptor and transfers electrons from a quinol to
(MUSE, Subaru, ALMA, Keck). M.F., I.S., S.C., and A.M.S. worked Bioscience and Bioinformatics, Kyushu Institute of
on the MUSE data reduction. C.C.S. worked on the Keck data Technology, Kawazu 680-4, Iizuka, Fukuoka-ken 820-8502, the diheme active site composed of two high-
reduction. Y.M. worked on the Subaru data reduction. M.F. and Japan. 7University of Strasbourg Institute for Advanced spin hemes (b595 and d), where molecular
C.S. performed radiative transfer calculations. R.J.I., A.E.S., J.V., Study, Strasbourg, France. oxygen is reduced to water. CydB harbors a
T.Y., Y.T., M.Ku., K.N., M.Ka., B.H., and K.K. contributed to *These authors contributed equally to this work.
interpreting the results. All authors reviewed, discussed, and †Corresponding author. Email: hartmut.michel@biophys.mpg.de structural ubiquinone-8 molecule that occu-
commented on the results and the manuscript, and met the (H.M.); werner.kuehlbrandt@biophys.mpg.de (W.K.) pies a hydrophobic pocket in a near-symmetric

100 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

Fig. 1. Cryo-EM structure of

the bd-I oxidase from E. coli.
(A) Surface representation of the
bd-I oxidase cryo-EM density
map at 2.68 Å resolution.
(B) Subunit CydH binds to a
hydrophobic cleft in CydA and
blocks oxygen accessibility to
heme b595 from the hydrophobic
lipid bilayer. (C) E. coli bd-I
contains 20 membrane-spanning
helices. (D) The Q loop is divided
into flexible QN and rigid, well-
ordered QC segments. (E) The
substrate inhibitor AD3-11
reduces deuterium exchange
rate only in the flexible QN region.
Relative differences of deuterium
uptake are mapped on the struc-
ture and the CydA amino acid
sequence. Abbreviations: A, Ala;
C, Cys; D, Asp; E, Glu; F, Phe;
G, Gly; H, His; I, Ile; K, Lys; L,
Leu; M, Met; N, Asn; P, Pro; Q,
Gln; R, Arg; S, Ser; T, Thr; V, Val;
W, Trp; Y, Tyr.

Qh1

position relative to the b-type hemes (Fig. 1C and architectures of subunits CydA, CydB, ynhF (fig. S5), which is neither part of the cyd
and fig. S4). At a distance of 3.5 nm to heme and CydX resemble the overall structure of (bd-I) nor the app (bd-II) operon. Homologs of
d, this UQ-8 is unlikely to be involved in the S-subfamily cytochrome bd oxidase from CydH are exclusively found in the proteobac-
electron transfer or oxygen reduction, and Geobacillus thermodenitrificans (G. th) (10). terial clade, where they correlate with the
is more likely to have a role in CydAB dimer The similar location of CydS in the G. th S- presence of L-subfamily bd oxidases. Two tight-
assembly. subfamily enzyme indicates an analogous ly bound lipid molecules close to CydH seal
The small noncatalytic accessory subunit structural role for CydX. the cleft between CydH and TMHs 1 and 8 of
CydX is positioned in a groove formed by A previously unknown single transmem- CydA (Fig. 1B). CydH occupies the putative
TMHs 1 and 6 of CydA (Fig. 1A and fig. S3) brane subunit CydH binds in the cleft between oxygen entry site in the bd oxidase from G. th,
where it promotes the assembly or stability transmembrane helices (TMHs) 1 and 9 of CydA where CydH is absent. The bd-I oxidase of
of the oxidase complex (8). The arrangement (Fig. 1B). CydH is encoded by the orphan gene E. coli may thus possess an alternative oxygen-

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 101

RES EARCH | R E P O R T S

conducting channel, because it is unlikely that a quinolone-type inhibitor (AD3-11) (fig. S6). to the short horizontal helix Qh1 that com-
CydH undergoes rapid rearrangement to open The Q-loop starts near the periplasmic sec- prises the highly conserved residues Lys252.A
and close an oxygen-conducting channel in a tion of TM6, where we observe a helix break and Glu257.A, critical for quinol binding and
gating-like mechanism without impairing effi- at the strictly conserved Asp239.A near pro- electron transfer (11). The entire polypeptide
cient turnover. pionate B of heme b558 (Fig. 1D and fig. S7). stretch between the helix break of TM6 and
We examined the role of the Q-loop in the Map features in the region between Glu240.A Lys252.A forms a dynamic loop, covering heme
as-isolated state of the enzyme without sub- and Asp247.A are noisy and difficult to inter- b558 that would otherwise be surface-exposed.
strate, in the presence of a synthetic sub- pret. However, we were able to trace the poly- This conformational heterogeneity extends
strate quinone (UQ-1), and in the presence of peptide chain, although with less precision, to the entire N-terminal part of the Q-loop

Fig. 2. Cofactor organization and redox reaction cycle. (A) Triangular arrangement of the heme cofactors in CydA. Heme edge-to-edge distances are indicated
by reversed parentheses. (B) Schematic overview of heme positions in E. coli and G. th. (C) Assignment of heme cofactors and their axial amino acid ligands.
Illustrated map densities are filtered to equal contour levels.

Fig. 3. Proton and oxygen pathways. (A) Interior H- and O-channel predicted by MOLE2. (B) The hydrophilic, water-filled H-channel connects the heme d oxygen
reduction site directly to the cytoplasm by the CydA pathway. A second CydB pathway is outlined by the conserved WDGNQ motif (indicated in yellow). The
hydrophobic O-channel runs parallel to the membrane plane and connects the lipid interface with heme d. Water densities are shown with transparent white surfaces.
(C) Heme d sits in an amphipathic pocket where the hydrophilic proton channel and hydrophobic oxygen pathway converge at the dioxygen binding site.

102 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

(QN) domain. The previously uncharacter- lutionarily conserved feature of cytochrome bilize the deprotonated Glu445.A. In the ferrous
ized helix-turn-helix motif of the C-terminal bd–type oxygen reductases. state, Glu445.A may become protonated and
Q-loop domain (QC) is rigid and covers the Heme b558 is found within the hydrophobic thus compensate the charge.
entire periplasmic surface of CydA (Fig. 1D membrane section adjacent to the periplas- The third cofactor, which is the site of oxygen
and fig. S7). mic QN-loop domain. The two axial ligands of binding and reduction, is a cis–heme d hydroxy-
Hydrogen/deuterium exchange mass spec- heme b558 are Met393.A and His186.A (Fig. 2C). chlorin g-spirolactone (15). Heme d is located in
trometry (HDX-MS) measurements showed The location and coordination of this heme the center of CydA near the interface to CydB
that binding of the aurachin-D–type compe- are equivalent to the G. th enzyme (12, 13). (Fig. 1C and fig. S4). Because of its well-defined
titive inhibitor AD3-11 exclusively affects Propionate A of heme b558 is coordinated by single propionate residue and characteristic
the flexible and disordered QN-loop (Fig. 1E). the conserved residues Lys252.A and Lys183.A. spirolactone group, its assignment to this lo-
No substrate-induced effect was observed Propionate B interacts with the essential cation is unambiguous (Fig. 2C and fig. S8).
in the QC-loop, implying that this insertion, Asp239.A. Heme b558 is separated by 6.7 Å from At a distance of 2.5 Å to the heme iron, the
which defines the L-subfamily, does not par- the high-spin active site (Fig. 2A). This distance invariant His19.A acts as the axial ligand of
ticipate in substrate binding. Although AD3-11 is bridged by the totally conserved Trp441.A heme d (Fig. 2B).
decreased the flexibility of the QN-loop (Fig. that may mediate electron transfer. On the opposite side of the heme along the
1E), it did not give rise to a distinct confor- In the E. coli enzyme, hemes b595 and d are His19-FeD axis, we observed a strong non-
mation that we could identify by cryo-EM interchanged with respect to the G. th oxidase peptide density that accounts for the expected
(fig. S3). The flexibility seems to be required (Fig. 2B) (4, 10, 14). In the E. coli structure, dioxygen molecule bound in the as-isolated
for the transient interaction of quinols with heme b595 is located near the periplasmic sur- state (Fig. 2C). Its slightly elongated shape is
the bd oxidase and their rapid release upon face. Detailed map features allowed us to held in a hydrophobic pocket formed by the
oxidation. Apart from their interaction with position the heme plane and its two elong- strictly conserved residues Phe140.A and Ile144.A
the flexible QN-loop, UQ-1, or AD3-11 do not af- ated propionate groups unequivocally (Fig. 2C that confine the dioxygen next to the heme d
fect the bd-I oxidase structure, making it and fig. S8). The heme b595 iron is pentaco- iron (Fig. 2B). The functionally essential Glu99.A
unlikely that additional substrate binding ordinated with Glu445.A as its axial ligand (Fig. completes the dioxygen binding site. In con-
sites exist. 2B). Both propionates and Glu445.A form a salt trast to most carboxylates in our structure, es-
The E. coli bd-I oxidase contains two b-type bridge network with the strictly conserved pecially those forming hydrogen bonds or salt
hemes and a chlorin-type heme d cofactor Arg448.A and Arg9.A. The side chain of Gln152.A bridges, the density of Glu99.A is not well de-
that are organized in a triangular geometry is positioned between the two b-type hemes fined. In the as-isolated state of the enzyme,
near the periplasmic surface of CydA (Fig. 2A). and adopts two distinct conformations (Fig. heme d is predominantly in the ferrous state,
Their arrangement is similar to the redox 2C). The two orientations most likely compen- with the charge-compensating Glu99.A pro-
center organization of our previously reported sate the different protonation states of Glu445.A tonated (1, 10). The lack of hydrogen-bonding
crystal structure of the S-subfamily homolog, in the mixed-valence state of the enzyme. In partners in this hydrophobic environment most
and this cofactor localization is thus an evo- the ferric state of heme b595, Gln152.A may sta- likely accounts for the flexibility of this residue.

Fig. 4. Structural and evolutionary diversity. (A) Superimposed dioxygen binding sites from bd oxidase of E. coli (blue and orange) and G. th (transparent beige).
Note that positions of heme d and b595 are interchanged. (B) The heme d site from E. coli and the superimposed position of b595 from G. th. (C) Sequence alignment
and conservation analysis of TMH3.A. L- and S-subfamily members are indicated as superscripts. The red arrowhead denotes the position of the Ile101 insertion.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 103

RES EARCH | R E P O R T S

The observed diversity in the high-spin heme to the dioxygen binding site of heme d (Fig. 4B). 6. J. Jones-Carson, M. Husain, L. Liu, D. J. Orlicky,
configuration is reflected in the electrochemical This channel appears to be occupied by fewer A. Vázquez-Torres, mBio 7, 02052-16 (2016).
7. S. A. Jones et al., Infect. Immun. 75, 4891–4899 (2007).
properties of the cofactors. In the G. th enzyme, water molecules, and they are spaced farther 8. J. Hoeser, S. Hong, G. Gehmann, R. B. Gennis, T. Friedrich,
the oxygen-binding heme d has the lowest mid- apart. Close to the dioxygen binding site of FEBS Lett. 588, 1537–1541 (2014).
point potential of +15 mV at pH 7 (fig. S9), heme d, the O-channel becomes hydrophilic 9. R. J. Allen et al., BMC Genomics 15, 946 (2014).
10. S. Safarian et al., Science 352, 583–586 (2016).
which would not allow for efficient electron and we observe densities that could either be 11. T. Mogi et al., Biochemistry 45, 7924–7930 (2006).
transfer and might explain the very low turn- water or dioxygen molecules. Conceivably, the 12. F. Spinner et al., Biochem. J. 308, 641–644 (1995).
over rates (16). Its structural and electrochem- O-channel might act as a pathway for direct 13. T. M. Kaysser, J. B. Ghaim, C. Georgiou, R. B. Gennis,
Biochemistry 34, 13491–13501 (1995).
ical properties suggest that it primarily acts as oxygen diffusion from the membrane interior 14. N. Azarkina et al., J. Biol. Chem. 274, 32810–32817 (1999).
an oxygen scavenger, unlike the high-turnover to the heme d reaction site. At the same time, 15. M. J. Miller, R. B. Gennis, J. Biol. Chem. 258, 9159–9165 (1983).
respiratory E. coli enzyme (17). The heme groups it may conduct protons through its connection 16. J. Sakamoto et al., Biochim. Biophys. Acta Bioenerg. 1411,
in the cytochrome bd oxidases from E. coli with the H-channel (29). 147–158 (1999).
17. A. M. Arutyunyan, J. Sakamoto, M. Inadome, Y. Kabashima,
have an uphill midpoint potential distribution The interchanged positions of hemes b595 V. B. Borisov, Biochim. Biophys. Acta Bioenerg. 1817,
(b558 < b595 < heme d), consistent with proper- and d entail an adaptation of electron, proton, 2087–2094 (2012).
ties of other L-subfamily members (18). E. coli and dioxygen pathways (10). In G. th, where 18. I. Belevich, V. B. Borisov, D. A. Bloch, A. A. Konstantinov,
M. I. Verkhovsky, Biochemistry 46, 11177–11184 (2007).
heme d has the highest midpoint potential; the oxygen-binding heme d is located close to 19. S. Jünemann, P. J. Butterworth, J. M. Wrigglesworth,
the potentials of the two heme b groups are the membrane interface, a direct oxygen entry Biochemistry 34, 14861–14867 (1995).
lower by at least 100 mV (fig. S9). site was postulated close to heme d between 20. E. P. Kavanagh, J. B. Callis, S. E. Edwards, R. K. Poole, S. Hill,
Microbiology 144, 2271–2280 (1998).
The critical step in the catalytic cycle of TMH1 and 8. In the E. coli enzyme, this site is 21. I. Belevich, V. B. Borisov, M. I. Verkhovsky, J. Biol. Chem. 282,
direct oxygen reduction is the transition from blocked by the single-transmembrane subunit 28514–28519 (2007).
the fully reduced to the oxoferryl state, where CydH (Fig. 4A). At position 101, next to the 22. V. B. Borisov, E. Forte, P. Sarti, A. Giuffrè, Biochim. Biophys.
Acta Bioenerg. 1807, 503–509 (2011).
ROS formation needs to be prevented (19–23) heme d site, the E. coli bd-I oxidase has a Leu 23. Y. Matsumoto et al., J. Biochem. 139, 779–788 (2006).
(fig. S10). In a previous proposal, the dioxygen inserted, which causes one helix turn to bulge, 24. S. A. Siletsky, F. Rappaport, R. K. Poole, V. B. Borisov,
bond is broken in a simultaneous four-electron displacing Glu99.A (Glu101 in G. th) from the PLOS ONE 11, e0155186 (2016).
transfer step, which avoids the formation of central heme iron to allow for dioxygen bind- 25. H. K. Baek, H. E. Van Wart, Biochemistry 28, 5714–5719 (1989).
26. D. Sehnal et al., J. Cheminform. 5, 39–39 (2013).
a peroxide intermediate. However, the short ing at this position (Fig. 4B). At the same time, 27. K. Yang et al., Biochemistry 46, 3270–3278 (2007).
edge-to-edge distance (6.7 Å) between the two Phe104.A is flipped and forms, together with 28. S. Yoshikawa, A. Shimada, Chem. Rev. 115, 1936–1989 (2015).
b-type hemes and the interchanged positions Ile144.A, a hydrophobic binding site for dioxy- 29. G. Hummer, J. C. Rasaiah, J. P. Noworyta, Nature 414, 188–190
(2001).
of hemes b595 and d are consistent with a se- gen. The Leu101.A insertion coincides with a dis-
quential electron transfer from heme b558 via placement of TMH9.B that forms the O-channel AC KNOWLED GME NTS
heme b595 to heme d (Fig. 2). This proposal is (Fig. 4A). The occurrence of this Leu is not We thank R. Zimmermann for outstanding technical
further supported by the midpoint potential unique for the E. coli bd-I oxidase or even assistance, S. Fritz for support in generating hybridoma cell
lines, and J. Vonck, B. Murphy, and J. Vinnemann for carefully
difference between heme b595 and d as well members of the L-subfamily (Fig. 4C), but is reading the manuscript and valuable discussions. Funding:
as electron backflow experiments (fig. S9) a strong indication for the position of the Supported by the Max Planck Society, the Deutsche
(18, 24). Earlier studies on the E. coli bd-I oxygen-binding heme in structurally unchar- Forschungsgemeinschaft (Cluster of Excellence Macromolecular
Complexes Frankfurt, SPP 2002), a Grant-in-Aid for Scientific
oxidase provide evidence for such a sequen- acterized bd oxidases.
Research (C) (16K07299 to J.S.) from the Japan Society for
tial electron transfer generating a short-lived The observed flexibility of the Q-loop poses the Promotion of Science, the University of Strasbourg, CNRS
peroxide intermediate, similar to Cpd0 in the a challenge for structure-driven design of (France), and a Nobel Laureate Fellowship of the Max Planck
catalytic cycle of peroxidases (21, 25). substrate-type inhibitors that mimic quinones Society. Author contributions: S.S. designed experiments,
purified bd oxidase and mAb20A11, prepared grids, collected
We identified two solvent-accessible areas and abolish election transfer. In case of the cryo-EM data, built the model, drafted the manuscript, and
(26) and assigned all densities found consist- E. coli bd-I oxidase, it may be possible to de- prepared figures. A.H. prepared grids, collected cryo-EM data,
ently in the three independent cryo-EM maps sign inhibitors that compete with UQ-8 for the refined the structure, drafted the manuscript, and prepared
figures. D.J.M. aligned the microscope and collected cryo-EM
to water molecules (Fig. 3). The waters define CydB binding site and interfere with the as- data. M.R. performed activity measurements and processed
a hydrophilic H-channel that starts at the cyto- sembly of a functional CydAB core dimer. An HDX-MS data. M.L.E. collected, processed and evaluated
plasmic CydAB interface and runs perpen- alternative target is the membrane-embedded HDX-MS data. J.M.-C. sequenced CydH via MS/MS. A.N. and
F.M. performed potentiometric titrations and evaluated data.
dicular to the membrane plane between the entry site of the oxygen-conducting O-channel H.M. synthesized AD3-11 and provided the compound.
intersubunit four-helix bundle formed by TMH2/ that would be targeted by hydrophobic com- R.B.G. initiated the project and supplied strains and plasmids.
3.A and TMH2/3.B until the central Asp58.B pounds to block the channel and abolish oxy- J.D.L. evaluated MS data. J.S. provided strains and samples
for potentiometric titrations. P.H. evaluated electrochemical
(Fig. 3). The conserved hydrophilic residues gen uptake. Further structural studies of bd data. W.K. and H.M. supervised the project. Competing
Ser108.A, Glu107.A, and Ser140.A continue the di- oxidases are needed to explore whether the interests: The authors declare no conflict of interest.
rect hydrophilic pathway lined by water mol- observed structural differences between E. coli Data and materials availability: Cryo-EM maps are
ecules to heme d and facilitate proton transfer and G. th are specific for their respective deposited at the Electron Microscopy Data Bank under
accession numbers EMD-4908, EMD-4911, and EMD-4916.
for either charge compensation or dioxygen bacterial clades or result from a distinct ad- The model of the cytochrome bd-I oxidase structure was
reduction (4) (Fig. 3C). Close to Asp58.B, the aptation to their ecological niche and phys- submitted to the Protein Data Bank with accession number
H-channel branches and runs along the con- iological roles. 6RKO. All other data are presented in the main text or the
supplementary materials.
served helix segment WDGNQ of TMH2.B in
the center of CydB until Trp63.B, where its diam- RE FERENCES AND NOTES
SUPPLEMENTARY MATERIALS
eter is constricted to ~2 Å (Fig. 3B) (27). It 1. I. Belevich et al., Proc. Natl. Acad. Sci. U.S.A. 102, 3657–3662
science.sciencemag.org/content/366/6461/100/suppl/DC1
(2005).
continues beyond this residue and merges with Materials and Methods
2. S. Iwata, C. Ostermeier, B. Ludwig, H. Michel, Nature 376,
the oxygen-conducting O-channel, similar to Figs. S1 to S10
660–669 (1995).
Table S1
the K-pathway in HCOs (2, 28). 3. N. P. Kalia et al., Proc. Natl. Acad. Sci. U.S.A. 114, 7426–7431
Movie S1
The O-channel starts near Trp63.B at the mem- (2017).
References (30–53)
4. V. B. Borisov, R. B. Gennis, J. Hemp, M. I. Verkhovsky, Biochim.
brane interface between TMH1 and TMH9 of Biophys. Acta Bioenerg. 1807, 1398–1413 (2011). 20 May 2019; accepted 4 September 2019
CydB and forms a direct hydrophobic pathway 5. M. Shepherd et al., Sci. Rep. 6, 35285 (2016). 10.1126/science.aay0967

104 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

3D PRINTING achieves this high intensity. Parallel printing

requires strategies like structuring the light
Scalable submicrometer additive manufacturing beam to project an arbitrary 2D pattern or
generating multiple focal spots to multiply the
Sourabh K. Saha1*†‡, Dien Wang2*, Vu H. Nguyen1*, Yina Chang2, writing speed during point scanning. How-
James S. Oakdale1, Shih-Chi Chen2‡ ever, improving the fabrication throughput
without compromising the resolution and/or
High-throughput fabrication techniques for generating arbitrarily complex three-dimensional structures the pattern complexity is challenging. For ex-
with nanoscale features are desirable across a broad range of applications. Two-photon lithography ample, some past attempts at parallelizing
(TPL)–based submicrometer additive manufacturing is a promising candidate to fill this gap. However, the TPL via projection of images from tunable
serial point-by-point writing scheme of TPL is too slow for many applications. Attempts at parallelization digital masks have failed to achieve the sub-
either do not have submicrometer resolution or cannot pattern complex structures. We overcome these micrometer depth resolution that is routinely
difficulties by spatially and temporally focusing an ultrafast laser to implement a projection-based layer-by- achievable in serial writing (19). Holographic
layer parallelization. This increases the throughput up to three orders of magnitude and expands the methods developed to generate and scan mul-
geometric design space. We demonstrate this by printing, within single-digit millisecond time scales, nanowires tiple focal spots simultaneously (20, 21) cannot
with widths smaller than 175 nanometers over an area one million times larger than the cross-sectional area. fabricate arbitrary structures, so they are
mostly suitable for printing periodic structures.

N
We developed a femtosecond projection TPL
anoscale device development requires Parallelizing TPL while maintaining the ca- (FP-TPL) technique that ensures simultaneous
scaling up nanofabrication of arbitrarily pability to print arbitrarily complex 3D patterns spatial and temporal focusing of ultrafast light
complex three-dimensional (3D) struc- is challenging because of the specific nature of to achieve parallel printing of arbitrarily complex
tures. The photopolymerization-based, femtosecond light-matter interactions. Femto- 3D structures with submicrometer resolution.
submicrometer additive manufactur- second pulsed lasers are necessary for TPL We fabricated 3D structures using a layer-
ing two-photon lithography (TPL) technique because appreciable two-photon absorption by-layer printing of patterned 2D layers rang-
generates 3D structures with features on the in photopolymers requires light intensity on ing in thickness from less than 1 mm to more
scale of 200 nm (1). The production rate is the order of ~1 TW/cm2 (5, 18). Focusing an than 4 mm. Each 2D layer was up to 165 mm ×
faster than high-resolution 2D techniques ultrafast pulsed laser beam into a small spot 165 mm in size and was written in millisecond
such as e-beam lithography (2, 3). TPL relies
on nonlinear two-photon absorption to gen-
erate features smaller than the diffraction-
limited focused light spot (4–6). This capability
allows fabrication of functional micro- and
nanoscale 3D structures, such as photonic
crystals (7, 8), mechanical metamaterials
(9–12), micromachines (13, 14), miniaturized
optics (15), flexible electronics (16), and bio-
scaffolds (17). Serially scanning a tightly
focused laser spot in a photopolymer resist to
fabricate 3D structures by overlapping the in-
dividual submicrometer volumetric pixels (i.e.,
voxels) is the most common implementation.
This serial and slow writing scheme makes
large-scale production impractical. As a result,
the adoption of TPL is mostly limited to
academic and research laboratories. We show
that a rate increase of two to three orders of
magnitude is possible without compromising
the submicrometer resolution by parallel TPL
processing based on femtosecond projection.
Our method allows access to difficult-to-explore
regions in design space, increasing both the
potential for cost-effective high-throughput
processing and the geometric complexity of Fig. 1. Femtosecond projection TPL based on spatial and temporal focusing. (A) 3D printing with
the printed objects. submicrometer resolution using layer-by-layer projection of digital masks. (B) Schematic of the fabrication setup.
2D layers are printed by projecting an image of the micromirror array onto the print plane within the photopolymer
resist. Printing can be restricted to layers of <1 mm thickness by generating strong intensity gradients through
temporal focusing of the femtosecond (fs) laser. L1 refers to the collimating lens and L2 refers to the objective lens.
1
Center for Engineered Materials and Manufacturing, Lawrence Femtosecond pulses are stretched and compressed as they pass through the optical system to implement temporal
Livermore National Laboratory, Livermore, CA, USA.
2 focusing. The micromirror array spectrally separates the different wavelengths of the femtosecond pulse and
Department of Mechanical and Automation Engineering, The
Chinese University of Hong Kong, Shatin, Hong Kong. stretches it, whereas the objective lens focuses the pulse in the time domain. The micromirror array, L1, and L2
*These authors contributed equally to this work. are set up in a 4f-like optical arrangement to ensure that the optical path lengths for all wavelengths are equal
†Present address: Woodruff School of Mechanical Engineering, between the micromirror array and the focal plane of the objective lens L2 but unequal across different wavelengths
Georgia Institute of Technology, Atlanta, GA 30332, USA.
‡Corresponding author. Email: sourabh.saha@me.gatech.edu between other points. (C) Zoomed-in schematic of temporal focusing in the focal volume of the objective lens, where
(S.K.S.); scchen@mae.cuhk.edu.hk (S.-C.C.) the shortest pulse is only achieved at the focal (build) plane. I, intensity; t, time.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 105

RES EARCH | R E P O R T S

time scales. We generated such thin 2D layers by upon hitting the DMD to generate several dis- We have achieved temporal focusing through
projecting a patterned 2D light sheet (equiv- tinct emergent beams. Because of the broad- an optical arrangement wherein the focused
alent to a planar point cloud containing more band nature of femtosecond laser light, the image plane of the objective lens is the con-
than a million points) that was focused in diffracted beams for the different wavelengths jugate plane of the digital mask. This ensures
both the space and time domains. We can pat- emerge at slightly different angles. This spec- that the optical path lengths for the different
tern the light sheet into arbitrary 2D patterns tral separation of light is accompanied by a wavelengths of the femtosecond laser emerg-
through a digital micromirror device (DMD), stretching of the pulse duration, which is a ing from the mask are identical at the focal
a standard grating-like component of digital widely exploited phenomenon in the design plane but unequal everywhere else. The full
light projectors. We harnessed the broadband of high-power ultrafast lasers (23). Unless spectrum of the laser therefore spatially over-
nature of femtosecond lasers and diffraction properly compensated, this pulse stretching laps only at the focal plane, thereby leading to
from the DMD to achieve focusing in the time may lead to a loss in the intensity gradient at the shortest pulse at the focal plane. Temporal
domain. During temporal focusing, a pre- the focal plane of the objective lens. The tem- focusing has been used for imaging and man-
stretched ultrashort light pulse is progres- poral focusing technique ensures that the pulse ufacturing applications where nontunable
sively shortened as it travels through the is compressed back to the shortest duration diffraction gratings spectrally separate light
photopolymer resist so that the shortest pulse only at the spatial focal plane of the objective (24–27). These fixed-mask techniques cannot
(the highest light intensity) is achieved only lens. This generates a strong intensity gradi- be used for additive manufacturing. We use
at the spatial focal plane. The intensity gra- ent at the spatial focal plane that restricts poly- the DMD-based digital mask as both the light
dient generated through temporal focusing merization within a thin layer centered at the structuring device and as the diffraction grat-
ensures that writing is spatially restricted to focal plane. ing. We previously demonstrated that this
the focal plane, without causing polymer-
ization above or below the focal plane (Fig. 1).
Without this temporal focusing, depth resolu-
tion is lost. Instead of forming a thin sheet, all
the material in the path of the projected beam
polymerizes to form thick extruded solid struc-
tures similar to those formed in previous paral-
lelization demonstrations (19). We consistently
and reliably printed subdiffraction 3D struc-
tures over large areas by exploiting the ultra-
thin and uniform femtosecond light sheet and
the optimized voxel aspect ratio.
We generated near-infrared femtosecond
pulses with a broad wavelength spectrum (ap-
proximately tens of nanometers) with our laser
source that provided the light to drive the
writing process. We patterned the light beam
by illuminating the digital mask, which was an
array of individually switchable micromirrors.
The intensity of light emerging from a micro-
mirror along a predetermined direction was
high when switched on and low when switched
off. We then collimated the diverging beam
that emerges from the DMD. The collimated
beam passes through the objective lens and
was focused onto a plane inside the photo-
polymer resist material. When projected onto
the image plane, the switched-on points from
the DMD had light intensities above the poly-
merization threshold, writing a pixelated image
of the digital mask in the resist. This image
contained distinct or overlapping cured voxels.
We fabricated 3D structures by moving the
focused image plane relative to the resist with
motion stages (Fig. 2). Fig. 2. Printing of complex 3D structures with submicrometer resolution via femtosecond projection
Although our FP-TPL technique appears TPL. (A to C) Millimeter-scale structure with submicrometer features supported on a U.S. penny on top of a
similar in approach to that of single-photon reflective surface. The 2.20 mm × 2.20 mm × 0.25 mm cuboid was printed in 8 min 20 s, demonstrating a 3D
projection stereolithography techniques (22), printing rate of 8.7 mm3/hour. In contrast, serial techniques would require several hours to print this cuboid.
we implemented a fundamentally different (D) A 3D micropillar printed through stacking of 2D layers, demonstrating uniformity of printing that is
projection mechanism. This difference ac- indistinguishable from that of commercial serial-scanning systems. (E and F) Spiral structures printed through
counts for the spatially coherent broadband projection of a single layer demonstrating the ability to rapidly print curvilinear structures within single-digit
laser source used in the FP-TPL system. As millisecond time scales without any stage motion. (G to J) Overhanging 3D structures printed by stitching multiple
the DMD-based digital mask is a dispersive 2D projections demonstrating the ability to print depth-resolved features. The bridge structure in (G), with 90°
element owing to its periodic micromirror overhang angles, is challenging to print using serial-scanning TPL techniques or any other technique owing to its
array structure, spatially coherent light diffracts large overhang relative to the size of the smallest feature and the submicrometer feature resolution.

106 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

temporal focusing scheme can be used for sizes by tuning the pattern on the DMD (fig. aspect ratio of ~1. We observed that the fea-
material removal processes (28, 29). S4). The nonlinear photoabsorption mecha- ture size increases with the dosage D, where
We have quantified the resolution limit of nism and the resist’s thresholding behavior D ~ tm n
exp × I in TPL. The exposure (texp) and
FP-TPL by developing a computational model ensures that features smaller than the optical the intensity (I ) are scaled with parameters
of light propagation that predicts the light in- diffraction limit can be fabricated with light m ≤ 1 and n ≥ 2. The feature size increases
tensity distribution in the focal volume. We spots of various sizes. with the dosage above a threshold dosage for
obtained the temporally focused intensity dis- We printed nanowire features thinner than polymerization. Features smaller than the op-
tribution by simulating the propagation of a the optical diffraction limit along both the tical FWHM width can be generated under
broadband femtosecond pulse through the op- lateral and the axial dimensions by varying low dosage conditions (Fig. 3). This feature
tical system (30). We found that temporal fo- the projected line thickness, the beam power, size versus dosage dependence is similar to
cusing allows for submicrometer full width at and the duration of exposure. We printed that observed in the serial single-spot TPL
half maximum (FWHM) axial thickness (figs. nanowires with lateral widths as thin as 130 process (33).
S4 and S5). The aspect ratio of the voxels we to 140 nm and with an axial height as short A peculiarity of our projection scheme is
induced by projecting DMD pixels was closer as 175 nm (Fig. 3). These features are thinner that the size of a contiguous projected feature
to 1 and smaller than the typical point spread than other parallel multiphoton printed poly- can be controlled over a wide range. This change
function of serial scanning systems. This pro- mer nanowires (19, 31, 32) and comparable in in the projected feature size alters the scaling
vides substantial advantage in terms of reso- width to serial scanning TPL systems (1). The of the lithographically generated feature size
lution and structural strength when fabricating 175 nm height of the thinnest line is smaller versus the optical FWHM (Fig. 3). For very
subdiffraction structures. We projected each than past demonstrations of both parallel and thin features (3 pixels wide), the line width
DMD pixel to an area of 151 nm × 151 nm in the serial TPL (1). Our result of a 1.25 voxel aspect stays below the optical FWHM width for a
work field and generated light spots of various ratio confirmed our model prediction of an wider range of exposure time. In contrast, line

Fig. 3. Printed nanowires demonstrating nanoscale resolution of FP-TPL. shape represent data points generated at the same power level, and all markers of a
(A) Width (along lateral direction) and (B) height (along axial direction) of specific color represent the same line width. Horizontal lines are the FWHM width
suspended nanowires printed under different conditions. Width of lines in the and height of the light intensity distributions obtained from simulations of fig. S4.
projected DMD pattern was varied from 3 to 6 pixels with a fixed period of Printing was performed with a femtosecond laser that had a center wavelength of
30 pixels. Each pixel (px) maps to 151 nm in the projected image. Labels HP, MP, 800 nm and a nominal pulse width of 35 fs and with a 60× 1.25 numerical aperture
and LP refer to high (42 nW/px), medium (39 nW/px), and low (35 nW/px) objective lens. Error bars quantify the 1 SD in a sample size of five wires printed
power levels, respectively, wherein time-averaged power was measured simultaneously in the same projection. (C and D) Scanning electron microscope
immediately before the aperture of the objective lens. All markers of a specific images of the suspended nanowire features.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 107

RES EARCH | R E P O R T S

width exceeds the optical FWHM width for reactions, but sustained photoenergy is not in-plane (lateral) printing resolution and
thicker features (6-pixel, high-power condi- needed for the reactions to proceed (36). Using are independent of the in-plane geometry
tion) even under low exposure conditions. We a 1-kHz repetition rate commercially available of the layers.
believe that this size-dependent behavior ori- ultrafast laser with a high (~1 mJ) pulse energy We compare our FP-TPL performance with
ginates from the radical-mediated photopoly- takes advantage of this underlying photo- previous TPL demonstrations (Fig. 4). The single-
merization mechanism underlying this process. polymerization mechanism to achieve massive layer volumetric processing rate of FP-TPL ex-
Our hypothesis is supported by the observation parallelization by ensuring focal spot inten- ceeds that of the existing serial techniques by
that the writing threshold exposure is lower sities of approximately terawatts per square at least three orders of magnitude while main-
for thicker projected features (fig. S10). Photo- centimeter. taining sub-500-nm features. Our 3D printing
activated radicals generated in the smaller ex- The throughput of FP-TPL is insensitive to rate exceeds that of the fastest serial system by
posed regions of the photoresist can be rapidly the number of voxels written per layer because more than 90 times for porous structures and
quenched by the oxygen and radical inhibitor of its ability to project an entire 2D layer at by more than 450 times for nonporous struc-
species diffusing in from the surrounding un- once. The area of projection, layer thickness, tures (supplementary materials, supplemen-
exposed regions. However, quenching of rad- and the time interval between projection of tary text S15). The FP-TPL method is also
icals is less likely in the central region of a subsequent layers determines the volumetric capable of patterning arbitrarily complex 3D
large illuminated area owing to depletion of throughput of FP-TPL. The area of projection structures while maintaining the submicro-
the quenchers over the entire area. A similar and the duration of exposure per projection meter feature sizes. In addition, the temporally
size-dependent thresholding behavior, which determines the areal throughput. For our sys- focused light sheet in FP-TPL enables achieving
we call the self-proximity effect, has also been tem, we can project up to ~1 million pixels per high out-of-plane (axial) resolution. Another
observed in volumetric photopolymerization exposure, and the maximum area of projection attractive feature of FP-TPL compared with
processes wherein a large volume is simulta- is 165 mm × 165 mm. We can vary the layer the serial writing technique is the ability to
neously exposed at once (34). thickness from <1 mm up to ~4 mm, and the project and print curvilinear features (Fig. 2E)
The self-proximity effect also explains the time interval between projection of subse- without requiring stage accelerations and decel-
thicker wire heights than the axial optical quent layers is 20 ms. This includes the time erations during the piecewise linear path dis-
FWHM width even for writing conditions when taken for stage motion. The duration of expo- cretization approximations. This substantially
the wire widths are smaller than the lateral sure per projection lies in the 3-ms range. The increases the throughput. The areal projection
optical FWHM width. A nonzero background voxel generation rate is ~333 million voxels/s, mechanism also enables printing of long sus-
intensity exists along the axial direction owing areal 2D printing rate is ~9 mm2/s, the single- pended bridge structures with 90° overhangs
to the projection mechanism, whereas a zero layer FP-TPL has a volumetric processing rate (Fig. 2G). Overhanging structures are challeng-
background intensity level exists in the non- between 33 and 131 mm3/hour, and the vol- ing to print using a serial-scanning technique
projected sections on the z = 0 plane (fig. S4B). umetric 3D printing rate is between 5 and because of feature drift during the long printing
This nonzero background intensity along the 20 mm3/hour. The rates are for printing con- process. We expect that the throughput, reso-
axial direction causes the wire heights to be figurations that maintain the submicrometer lution, and pattern flexibility of FP-TPL makes it
larger than the wire widths when normalized
by the respective optical FWHM widths. The
nonzero background intensity is undesirable Fig. 4. Comparison of rate
for printing of thin features but may be used to and resolution of FP-TPL
print complex features through dosage accu- with past demonstrations
mulation as is used in the computed axial lith- of TPL. S1 to S7 are
ography volumetric additive manufacturing instances of serial TPL
technique (35). In the context of FP-TPL, the implementation and P1 to
consequence of this self-proximity effect is P6 are instances of parallel
that (i) if fast writing of thin features is de- TPL. P1 and P3 are not shown
sired, one must print under low-exposure con- in the lower plot because they
ditions with moderately large projected feature have an axial (depth) resolu-
sizes, and (ii) the exposure must be varied tion higher than 10 mm.
across a projection area to achieve uniform Protocol for evaluation of the
printing of structures comprising nonuniform volumetric processing rate
feature sizes (fig. S11). is available in supplementary
During scaling of the TPL process from the text S15. FP-TPL breaks the
serial to the parallel format, the intensity at traditional resolution versus
each focal volume is similar (approximately rate trade-offs. Literature
terawatts per square centimeter). However, references are as follows:
the total energy incident on the photopolymer S1, Oakdale et al. (2); S2,
material does not scale by the number of focal Meza et al. (11); S3, Shaw et al.
spots. Instead, the actual incident beam en- (37); S4, Lu et al. (1);
ergy in the parallel format is four orders of S5, Fischer et al. (6); S6,
magnitude smaller than one would estimate by Malinauskas et al. (5); S7,
linearly scaling the energy of serial-scanning Jonušauskas et al. (38); P1,
spot by the number of focal spots (table S3). Mills et al. (19); P2, Li et al.
This discrepancy is the result of the underlying (32); P3, Yang et al. (31); P4,
chain reaction–type radical-mediated photo- Vizsnyiczai et al. (20); P5,
polymerization mechanism of TPL being sen- Yang et al. (21); and P6,
sitive to the light intensity for the initiation of Geng et al. (39).

108 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

an attractive technology to scale up the fabri- synthesized the photopolymer resists and performed SEM imaging SUPPLEMENTARY MATERIALS
cation of functional micro- and nanostructures of nanowires. S.K.S., S.-C.C., and V.H.N. analyzed the results. S.K.S. science.sciencemag.org/content/366/6461/105/suppl/DC1
prepared the manuscript with input from all coauthors, and all Materials and Methods
such as mechanical and optical metamaterials, coauthors edited the manuscript. S.K.S. supervised the work at LLNL, Supplementary Text
micro-optics, bioscaffolds, electrochemical in- and S.-C.C. supervised the work at CUHK. Competing interests: Figs. S1 to S16
terfaces, and flexible electronics—technology Patent applications related to this work have been filed at the U.S. Tables S1 to S6
Patent and Trademark Office with S.K.S. and S.-C.C. as co-inventors References (40–45)
that may play a large role in fields such as and assigned to Lawrence Livermore National Security, LLC. All
electric transportation, healthcare, clean energy Movie S1
authors declare that they have no other competing interests. Data
and water, computing, and telecommunications. and materials availability: All data are available in the manuscript or 30 April 2019; accepted 9 September 2019
the supplementary materials. 10.1126/science.aax8760
RE FE RENCES AND N OT ES
1. W.-E. Lu, X.-Z. Dong, W.-Q. Chen, Z.-S. Zhao, X.-M. Duan,
J. Mater. Chem. 21, 5650–5659 (2011).
2. J. S. Oakdale et al., Adv. Funct. Mater. 27, 1702425 (2017).
3. M. Altissimo, Biomicrofluidics 4, 026503 (2010).
STRUCTURAL BIOLOGY
4. S. Maruo, O. Nakamura, S. Kawata, Opt. Lett. 22, 132–134 (1997).
5. M. Malinauskas, A. Žukauskas, G. Bičkauskaitė, R. Gadonas,
S. Juodkazis, Opt. Express 18, 10209–10221 (2010).
Cryo-EM structure of a dimeric B-Raf:14-3-3 complex
6. J. Fischer et al., Opt. Express 21, 26244–26260 (2013).
7. S. Juodkazis, V. Mizeikis, K. K. Seet, H. Misawa, U. G. Wegst, reveals asymmetry in the active sites of
Appl. Phys. Lett. 91, 241904 (2007).
8. G. von Freymann et al., Adv. Funct. Mater. 20, 1038–1052 (2010).
9. J. Bauer et al., Adv. Mater. 29, 1701850 (2017).
B-Raf kinases
10. T. Frenzel, M. Kadic, M. Wegener, Science 358, 1072–1074
(2017). Yasushi Kondo1,2*, Jana Ognjenović3*†, Saikat Banerjee4, Deepti Karandur1,2,5, Alan Merk3†,
11. L. R. Meza, S. Das, J. R. Greer, Science 345, 1322–1326
Kayla Kulhanek4, Kathryn Wong1,2‡, Jeroen P. Roose4, Sriram Subramaniam6§, John Kuriyan1,2,5,7,8§
(2014).
12. J. Bauer, A. Schroer, R. Schwaiger, O. Kraft, Nat. Mater. 15,
438–443 (2016). Raf kinases are important cancer drug targets. Paradoxically, many B-Raf inhibitors induce the
13. T.-Y. Huang et al., Adv. Mater. 27, 6644–6650 (2015). activation of Raf kinases. Cryo–electron microscopy structural analysis of a phosphorylated B-Raf
14. S. Maruo, H. Inoue, Appl. Phys. Lett. 89, 144101 (2006).
15. T. Gissibl, S. Thiele, A. Herkommer, H. Giessen, Nat. Photonics kinase domain dimer in complex with dimeric 14-3-3, at a resolution of ~3.9 angstroms, shows an
10, 554–560 (2016). asymmetric arrangement in which one kinase is in a canonical “active” conformation. The distal
16. W. Xiong et al., Adv. Mater. 28, 2002–2009 (2016). segment of the C-terminal tail of this kinase interacts with, and blocks, the active site of the cognate
17. A. Selimis, V. Mironov, M. Farsari, Microelectron. Eng. 132,
83–89 (2015). kinase in this asymmetric arrangement. Deletion of the C-terminal segment reduces Raf activity.
18. H.-B. Sun, S. Kawata, Adv. Polym. Sci. 170, 169–273 (2004). The unexpected asymmetric quaternary architecture illustrates how the paradoxical activation
19. B. Mills, J. A. Grant-Jacob, M. Feinaeugle, R. W. Eason, of Raf by kinase inhibitors reflects an innate mechanism, with 14-3-3 facilitating inhibition of
Opt. Express 21, 14853–14858 (2013).
20. G. Vizsnyiczai, L. Kelemen, P. Ormos, Opt. Express 22,
one kinase while maintaining activity of the other. Conformational modulation of these contacts
24217–24223 (2014). may provide new opportunities for Raf inhibitor development.
21. L. Yang et al., Opt. Lasers Eng. 70, 26–32 (2015).

R
22. X. Zheng et al., Rev. Sci. Instrum. 83, 125001 (2012).
23. D. Strickland, G. Mourou, Opt. Commun. 55, 447–449 (1985). af proteins are the first kinases to be fold protein 14-3-3, which promotes dimer-
24. D. Oron, E. Tal, Y. Silberberg, Opt. Express 13, 1468–1476
(2005).
activated in the Ras/Raf/Mek/Erk ki- ization of the Raf kinase domains (6–8). B-Raf
25. G. Zhu, J. van Howe, M. Durst, W. Zipfel, C. Xu, Opt. Express 13, nase pathway, which is one of the key is the most studied of the three human Raf
2153–2159 (2005). conduits for the transmission of extra- kinases because more than 50% of human
26. D. Kim, P. T. So, Opt. Lett. 35, 1602–1604 (2010).
cellular signals to the nucleus in animal melanomas contain an activating mutation
27. J.-N. Yih et al., Opt. Lett. 39, 3134–3137 (2014).
28. C. Gu et al., Precis. Eng. 50, 198–203 (2017). cells (1). The activation of Raf is initiated by (Val600→Glu or V600E) in B-Raf (9). A break-
29. D. Wang, C. Wen, Y. Chang, W. Lin, S.-C. Chen, Precis. Eng. 52, the release of autoinhibition through the bind- through in the treatment of melanoma occurred
106–111 (2018). ing of Ras•GTP (guanosine 5′-triphosphate) with the development of small-molecule in-
30. Materials and methods are available as supplementary
materials. to the Ras-binding domain of Raf (2–5). The hibitors that are effective against the V600E
31. L. Yang et al., Opt. Commun. 331, 82–86 (2014). phosphorylation of the C-terminal tail of Raf variant of B-Raf (10, 11). Unexpectedly, these
32. Y.-C. Li et al., Opt. Express 20, 19030–19038 (2012). then results in its association with the scaf- inhibitors were found to activate signaling
33. J. Serbin et al., Opt. Lett. 28, 301–303 (2003).
34. M. Shusteff et al., Sci. Adv. 3, o5496 (2017).
through activation of wild-type Raf isoforms
35. B. E. Kelly et al., Science 363, 1075–1079 (2019). (12–14). This paradoxical activation of Raf by
36. J. B. Mueller, J. Fischer, F. Mayer, M. Kadic, M. Wegener, 1
Department of Molecular and Cell Biology, University of a wide range of inhibitors remains puzzling
Adv. Mater. 26, 6566–6571 (2014). California, Berkeley, Berkeley, CA 94720, USA. 2California and has restricted the full potential of small-
37. L. A. Shaw et al., Opt. Express 26, 13543–13548 (2018). Institute for Quantitative Biosciences, University of
38. L. Jonušauskas et al., Opt. Express 27, 15205–15221 (2019). California, Berkeley, Berkeley, CA 94720, USA. molecule therapy targeting B-Raf.
39. Q. Geng, D. Wang, P. Chen, S.-C. Chen, Nat. Commun. 10, 2179 3
Laboratory of Cell Biology, Center for Cancer Research, The paradoxical activating effects of Raf ki-
(2019). National Cancer Institute, Bethesda, MD 20814, USA.
4
nase inhibitors have their origin in the homo-
Department of Anatomy, University of California, San
ACKN OW LEDG MEN TS Francisco, San Francisco, CA 94143, USA. 5Howard and heterodimerization of Raf isoforms. Drug
We thank R. Panas and C. Spadaccini at Lawrence Livermore Hughes Medical Institute, University of California, binding to only one of the Raf proteins in the
National Laboratory (LLNL) for helpful discussions and B. Au at Berkeley, Berkeley, CA 94720, USA. 6University of British dimer results in activation of the other (12–16).
LLNL for technical support. Funding: This work was performed Columbia, Vancouver, British Columbia V6T 1Z3, Canada.
under the auspices of the U.S. Department of Energy by LLNL under 7
Department of Chemistry, University of California,
Active Raf kinase domains adopt a canonical
contract DE-AC52-07NA27344 and by The Chinese University of Berkeley, Berkeley, CA 94720, USA. 8Divisions of structure, which is defined by an open con-
Hong Kong (CUHK) under subcontracts. (Document release Molecular Biophysics and Integrated Bioimaging, formation of the activation loop and inward
LLNL-JRNL-770497). Funding was provided by LLNL’s LDRD project Lawrence Berkeley National Laboratory, Berkeley, CA
16-ERD-047. The work at CUHK was partially supported by HKSAR 94720, USA.
rotation of helix aC (identified in Fig. 1). We
Research Grants Council (RGC), General Research Fund (GRF) *These authors contributed equally to this work. refer to the active conformation as the “ON-
14209081 and 14206517. Author contributions: S.K.S. and S.-C.C. †Present address: Frederick National Laboratory for Cancer state,” and it is stabilized through side-to-side
Research, Frederick, MD 21701, USA. ‡Present address: Keck
conceived of and designed the study. D.W., V.H.N., and S.K.S. set up
dimerization of the kinase domains (16, 17).
the FP-TPL systems and performed the FP-TPL experiments. V.H.N. School of Medicine of USC, Los Angeles, CA 90033, USA.
designed and developed the electronic hardware and control system §Corresponding author. Email: kuriyan@berkeley.edu (J.K.); Inactive Raf kinase domains do not dimerize,
at LLNL. S.K.S. and Y.C. performed the optics simulations. J.S.O. sriram.subramaniam@ubc.ca (S.S.) either because of increased dynamics or because

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 109

RES EARCH | R E P O R T S

of adoption of a catalytically unproductive only one kinase ready for catalysis. The asym- adenosine 5′-triphosphate (ATP) in the struc-
“OFF-state” conformation, which is com- metry in the kinase dimer is due to a close tures of active kinases (fig. S5D). The wedge
mon to many kinases and referred to as the but asymmetric interaction between the B- formed by the distal tail segment of B-Raf OUT
Cdk/Src OFF-state conformation (16, 18, 19). Raf kinase dimer and the 14-3-3 dimer, med- prevents closure of the N-lobe of B-RafIN over
Many B-Raf inhibitors destabilize this OFF- iated by the C-tails of B-Raf (Fig. 2A). The the C-lobe and in this way favors maintenance
state and, surprisingly, promote activation of kinase-proximal segment of the C-tail (“prox- of the ON-state conformation of B-RafIN (fig. S7)
the partner kinase in a dimer. At subsaturating imal tail segment”) (Fig. 1A) makes specific and, as a consequence, is expected to stabilize
concentrations of these inhibitors, inhibitor- interactions with 14-3-3 (fig. S5) and is fol- the B-Raf kinase dimer.
bound kinase domains can dimerize with lowed by Ser729, which is phosphorylated and As a first step toward validating the inter-
partner kinases that do not have the inhib- docks on 14-3-3 in the canonical manner actions seen in our model, we used Ba/F3 pro-
itor bound, stabilizing the ON-state in the (23). These interactions constrain each kinase B cells that rely on interleukin-3 (IL-3) for
inhibitor-free partner (12–14). By contrast, domain to be near the corresponding 14-3-3 growth (25, 26). In agreement with a critical
certain sulfonamide inhibitors that favor the subunit. role for IL3/B-Raf signaling, we were unable
adoption of the OFF-state conformation re- Although both the kinase dimer and the 14- to expand single Ba/F3 clones in which B-Raf
sult in destabilization of the side-to-side Raf 3-3 dimer are twofold symmetric, their sym- was deleted by means of CRISPR/Cas9. We
dimers and do not exhibit paradoxical kinase metry axes are not aligned, which results in therefore applied heterologous expression of
activity (16). the active site of one B-Raf kinase domain B-Raf variants fused to a fluorescent protein
The striking ability of kinase inhibitors to (denoted B-RafIN) (Figs. 1A and 2A, cyan) [enhanced green fluorescent protein (EGFP)]
activate Raf leads to questions of fundamen- being closer to 14-3-3 than that of the other coupled to dual-parameter flow cytometry,
tal interest in cell signaling and the develop- (B-Raf OUT) (Figs. 1A and 2A, magenta). This simultaneously measuring levels of phospho–
ment of effective cancer therapeutics. To help unexpected asymmetric disposition of the two extracellular signal–regulated kinase (pERK),
answer these questions, we have determined kinases with respect to 14-3-3 is a consequence a downstream product of Raf activation, and
a structure of Raf in an activated state in com- of an important difference in the paths of the GFP levels in Ba/F3 cells (fig. S8). Because
plex with 14-3-3, without kinase inhibitors B-Raf C-tails as they exit their respective 14-3-3 Ba/F3 cells express endogenous Raf proteins,
bound to it. We purified full-length human binding sites. The two B-Raf C-tails run in we restricted the measurement of pERK levels
B-Raf in complex with the e and z isoforms of opposite directions, so that the tail of one to cells that express a high level of the EGFP-
insect cell 14-3-3 (supplementary materials, kinase domain points toward the other ki- B-Raf fusion protein. Under these conditions,
materials and methods). The B-Raf protein nase domain (Figs. 1A and 2A). The active we assume that B-Raf dimers predominantly
is produced through intein ligation of the site of B-RafIN is positioned close to the 14-3-3 contain either two molecules of the heterol-
N-terminal regulatory segments of B-Raf, which surface and captures the kinase-distal seg- ogous fusion protein or one molecule each
is expressed in bacteria, to the C-terminal seg- ment of the C-tail of B-Raf OUT (“distal tail of the fusion protein and endogenous Raf
ment, which is produced in insect cells. The segment”) as it exits from the 14-3-3 binding kinases. Heterologous overexpression of wild-
purified B-Raf protein is phosphorylated at site (Figs. 1 and 2 and fig. S5C). By contrast, type EGFP-B-Raf resulted in similar pERK
multiple sites, including the activation loop the active site of B-Raf OUT points away from levels compared with those observed for GFP-
and the C-terminal 14-3-3 binding element the 14-3-3 surface and is too far away for a negative cells, both at baseline and after IL-3
(table S1). The N-terminal segment of B-Raf reciprocal interaction with the distal tail seg- stimulation, indicating that regulation of
contains an autoinhibitory phosphorylation ment of B-Raf IN. B-Raf signaling to ERK is intact (fig. S9).
site (Ser365) that is unphosphorylated in the For the C-tail of B-Raf OUT, continuous den- By contrast, EGFP-B-Raf V600E resulted in
B-Raf samples because the N-terminal seg- sity is seen extending from Leu733 in the 14-3- uniformly spontaneous phospho-ERK induc-
ment is produced in bacteria. Thus, the complex 3 binding element, toward the active site of tion in nonstimulated cells relative to the un-
we have produced appears to have bypassed B-Raf IN (fig. S6). We observed strong density transfected control, as expected (Fig. 3, A
the necessity of release of autoinhibition by for the backbone of a seven-residue segment and B) (20).
interaction with Ras•GTP, is active, and can of the distal tail segment (residues Asp742 to Using this platform, we first investigated the
phosphorylate mitogen-activated protein ki- Cys748), within which the side chains of Phe743, importance of the positioning of the C-tails
nase (MAPK) kinase 1 (MEK1) (fig. S1). In Leu745 , and Tyr746 can be resolved clearly of B-Raf onto 14-3-3. Expression of EGFP-B-
the cryo–electron microscopy (cryo-EM) struc- (Fig. 2B and fig. S6). The distal tail segment Raf S729A, a variant lacking the phosphoryl-
ture of the complex that we determined at forms a helical turn between residues 743 ation site that triggers binding to 14-3-3,
an overall resolution of 3.9 Å (Fig. 1A and and 747, exiting the active site of B-Raf IN in a resulted in impaired pERK induction upon
figs. S2 and S3), the B-Raf kinase domain direction that is orthogonal to the direction IL-3 stimulation (Fig. 3C), as expected (27).
dimer arrangement corresponds closely to the of entry. In this manner, the distal tail seg- The cryo-EM structure reveals that Leu721 in
ON-state dimer seen in many crystal struc- ment of B-Raf OUT makes an L-shaped wedge both proximal tail segments is sandwiched
tures of isolated Raf kinase domains (Fig. within the active site of B-Raf IN, with marked between the kinase domain and the 14-3-3
1B and fig. S4) (20, 21). The N-terminal reg- resemblance to the manner in which some surface (fig. S5), and a L721G mutation (in
ulatory segments are not visualized in the kinase inhibitors are bound at this site (fig. which leucine at position 721 is replaced by
cryo-EM map, presumably because the re- S5D). The hydroxyl group of the side chain of glycine) is expected to disrupt these inter-
lease of autoinhibition leads to disorder in Tyr746 in the distal tail segment of B-Raf OUT actions. Expression of EGFP-B-Raf L721G resulted
these segments. forms hydrogen bonds with the backbone of in lower levels of pERK relative to that of
Both of the kinase domains in the dimer the hinge connecting the N- and C-lobes of the control (Fig. 3D). We also made two variants
are in the ON-state conformation, as shown B-Raf IN kinase domain (Fig. 2C). The forma- of B-Raf in which glycine and serine residues
by comparison with the structure of B-Raf in tion of such hydrogen bonds is a characteristic were inserted into the proximal tail segment
complex with MEK (fig. S4) (22), but the ac- feature of the interaction of small-molecule to increase flexibility and to alter the length
tive site of one kinase is blocked by insertion inhibitors with protein kinases (24). The tyro- of the proximal tail segment, which is highly
of the C-terminal tail (C-tail) of the other sine side chain occupies the space that is conserved in B-Raf (fig. S10). In one variant,
(Fig. 1A). Thus, the dimer is asymmetric, with normally occupied by the adenine group of the sequence GSGSGS was inserted after the

110 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

last residue in the kinase domain (Arg719), and pERK relative to that of control, at baseline distal tail segment. These mutations did not
in the second variant, the same six residues and after IL-3 stimulation (Fig. 3, E and F). lead to a reduction in activity in this cellular
were inserted in the middle of the proximal Thus, precise positioning of the B-Raf dimer assay but, instead, led either to no change or
tail, between residues Lys723 and Ile724 (de- onto the 14-3-3 dimer is essential for optimal to a slight increase in pERK levels (Fig. 3G
noted Ins-1 and Ins-2, respectively, in Fig. 3A pERK signaling. and fig. S11, A to D). These results suggest
and fig. S5). Expression of the Ins-1 and Ins-2 We expressed EGFP-tagged variants of Raf that the distal tail segment could also play an
variants of B-Raf resulted in lower levels of that contain mutations that alter or delete the autoinhibitory role, as has been proposed

A activation loop B-Raf OUT

helix αC B-Raf OUT
N-lobe B-Raf IN B-Raf IN
active site
B-Raf IN proximal tail
of B-Raf OUT
N-lobe segment
B-Raf IN
C-lobe 180˚ B-Raf
IN
active site
14-3-3 of B-Raf IN

pS729 pS729

C-tail
of B-Raf OUT B-Raf OUT
B-Raf OUT C-tail of
distal tail 14-3-3
distal tail B-Raf IN 14-3-3 B-Raf OUT
segment B-Raf IN 14-3-3
segment proximal tail
binding element segment
S729
726 733
B-Raf IN N kinase C
449 720 S729 766
726 733
B-Raf OUT N kinase C
449 720 proximal tail 14-3-3 distal tail segment 749 766
segment binding element (only in B-Raf OUT)

B B-Raf IN proximal tail

B-Raf IN 14-3-3 segment helix αI
binding element

B-Raf IN
C-lobe
B-Raf OUT
N-lobe helix αC
activation
helix αC loop B-Raf OUT
distal tail
activation segment
loop B-Raf OUT
distal tail
helix αC
segment

helix αC

B-Raf OUT 14-3-3 B-Raf OUT B-Raf IN

B-Raf IN
C-lobe
N-lobe helix αI B-Raf OUT B-Raf OUT 14-3-3
proximal tail binding element
segment 14-3-3

Fig. 1. Structure of the B-Raf:14-3-3 complex. (A) (Left) Schematic diagram segments of the C-tails. The kinase domains and the C-tails are not to scale, which is
of the structure. (Middle and right) Two orthogonal views of the molecular indicated by the breaks. Dashed lines indicate regions for which there is no
surface of the cryo-EM model. The two B-Raf kinases in the dimer are shown in interpretable density. (B) A view of the B-Raf:14-3-3 complex, looking down the
cyan and magenta, respectively, and the 14-3-3 dimer is shown in gray. B-RafIN twofold symmetry axis of the B-Raf kinase dimer. The B-Raf kinase dimer seen in the
(cyan) is positioned closer to 14-3-3 than is B-RafOUT (magenta). (Bottom) cryo-EM structure closely resembles the dimeric structure of ON-state B-Raf bound
Schematic diagrams denoting the boundaries of the kinase domains and the C-tails to MEK (PDB ID: 4MNE) (fig. S4) (22). Helix aG of B-RafOUT has weaker density
of B-Raf that are included in the structural model, and the terms used to identify compared with the rest of the complex and is shown in gray.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 111

RES EARCH | R E P O R T S

for the C-tail of C-Raf (28). Only one distal (B-Raf-WT) and one without the N-terminal phosphorylate MEK1 robustly, and for both,
tail segment in a dimer (that of B-Raf OUT) is regulatory domain, corresponding to the cryo- the deletion of the distal tail segment resulted
required for activation, and endogenous B-Raf EM model (B-Raf-DN). For these two constructs, in substantially lower levels of MEK1 phos-
can provide this function, prompting our con- we either retained the distal tail segment or phorylation, whereas the binding of DDTS
servative interpretation of the results of these deleted it, generating two additional constructs mutants to 14-3-3 was unaffected (Fig. 3, H
cellular signaling assays. (B-Raf-DDTS and B-Raf-DNDDTS) (Fig. 3A). We and I, and fig. S11, E and F). The reduction
To further investigate the importance of the purified the B-Raf proteins corresponding to of MEK1 phosphorylation by B-Raf-DNDDTS
distal tail segment for B-Raf activity, we car- these constructs in complex with endogenous shows that interactions made by the distal
ried out phosphorylation assays with purified mammalian 14-3-3 proteins (Fig. 3H) and used tail segment are crucial, even in a stripped-
proteins. We prepared two sets of N-terminally them to phosphorylate the Raf substrate MEK1, down construct that lacks all of the regu-
Flag-tagged B-Raf constructs for mammalian with phospho-MEK1 detected with Western latory domains. The interpretation of these
cell expression, one based on full-length B-Raf blotting (Fig. 3I). Both B-Raf-WT and B-Raf-DN mutational data is complicated, however, by

Fig. 2. Interaction of the distal A B-Raf OUT

B-Raf OUT
tail segment of B-RafOUT with B-Raf IN active site
the active site of B-RafIN. active site
(A) Orthogonal views of the
cryo-EM structure of the B-Raf IN B-Raf IN
B-Raf:14-3-3 complex. On the
left, the C-tail of B-RafOUT 180˚
(magenta) is seen bound to
14-3-3 (gray), and the distal tail B-Raf OUT B-Raf OUT
proximal tail
segment enters the active site
segment
of B-RafIN (cyan). (B) View of the
ATP-binding site of B-RafIN
(cyan), with cryo-EM density B-Raf OUT
B-Raf OUT
shown in gray. Residues in the 14-3-3 binding B-Raf IN14-3-3
distal tail
distal tail segment of B-RafOUT element binding element 14-3-3
pS729 14-3-3 segment pS729
(magenta) are identified with
asterisks. (C) The hydrophobic B C
side chains of the C-spine of B-Raf IN
B-RafIN are shown as yellow B-Raf IN
spheres, with two side chains
of the B-RafOUT distal tail segment A481 Y746*
(magenta) completing the W531 V471
C-spine. (D) Comparison of the W531
structure of the B-Raf:14-3-3 F743*
complex with that of the CDK2:
Cyclin A:p27Kip1 complex (PDB ID: C532 F743*
1JSU) (33) and the autoinhibited L745*
Y746* pS729
form of twitchin kinase (PDB ID: F583
1KOB) (34). In the B-Raf:14-3-3
complex, the distal tail segment of L745*
B-Raf OUT B-Raf OUT B-RafOUT
B-RafOUT (magenta) enters
distal tail segment 14-3-3 binding distal tail
the ATP-binding site of B-RafIN segment
element
(cyan). In the CDK2:Cyclin A:
p27Kip1 complex, the p27Kip1
inhibitor (magenta) enters D
the ATP-binding site of CDK2 B-Raf OUT C-terminal tail
(cyan). In twitchin kinase, the B-Raf OUT p27 Kip1 of twitchin
distal tail kinase
C-terminal tail of the kinase
segment
(magenta) enters the ATP-binding
site. Selected hydrophobic side
chains in the inhibitory segments
are shown as spheres.

Cyclin A

B-Raf IN CDK2
Kip1
B-Raf:14-3-3 complex CDK2:Cyclin A:p27 complex Twitchin kinase

112 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

a possible autoinhibitory role for the distal We used molecular dynamics simulations to We initiated two sets of trajectories starting
tail segment. In addition, the distal tail seg- test whether engagement of the distal tail seg- from the cryo-EM model. In one set, an intact
ment contains two highly conserved phos- ment of B-Raf OUT in the active site of B-Raf IN B-Raf:14-3-3 complex, with the distal tail seg-
phorylation sites (Ser750 and Thr753), whose is required to maintain the asymmetric inter- ment of B-Raf OUT inserted into the active
mutation could alter B-Raf activity (29). action between the B-Raf dimer and 14-3-3. site of B-Raf IN, was used to generate three

A disordered N-terminal region

distal tail segment
deleted
B-Raf IN
B-Raf OUT distal tail segment

DTS: distal tail segment deleted

B-Raf OUT
(stop codon after R735)

S729A disruption of
V600E 14-3-3 binding
cancer mutation
Ins-2: GSGSGS inserted after K723
Ins-1: GSGSGS
B-Raf OUT proximal tail segment B-Raf- N DTS:14-3-3 complex
inserted after R719 L721G

B * * ns
C ** * * D
WT WT *** ** * WT
V600E S729A L721G

0 min 5 min 10 min 0 min 5 min 10 min 0 min 5 min 10 min

IL3 stimulation IL3 stimulation IL3 stimulation

E F G ** ns ns
** ** * WT ** * ** WT WT
Ins-1 Ins-2 DTS

0 min 5 min 10 min 0 min 5 min 10 min 0 min 5 min 10 min

IL3 stimulation IL3 stimulation IL3 stimulation

H (kDa) M 1 2 3 4 I
250
150 + B-Raf-WT + B-Raf- DTS
100 B-Raf-WT Time (min) 0 1 3 10 30 90 0 1 3 10 30 90

75 B-Raf- DTS WB: -pMEK

Coomassie
50
+ B-Raf- N + B-Raf- N DTS
37
B-Raf- N Time (min) 0 1 3 10 30 90 0 1 3 10 30 90
B-Raf- N DTS WB: -pMEK
14-3-3
25 Coomassie

Fig. 3. Mutational analysis of B-Raf. (A) (Left) Schematic diagram of the cryo- this. The statistical significance of each measurement is indicated by ns (not
EM structure, indicating the B-Raf variants that were analyzed. (Right) Schematic significant), P > 0.05; *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001. (H) SDS–
diagram of the B-Raf-DNDDTS:14-3-3 complex, which lacks the N-terminal region polyacrylamide gel electrophoresis gel analysis of B-Raf constructs purified from
and the distal tail segment (dotted circle). (B to G) Relative levels of pERK for human embryonic kidney–293T cells. M, Precision Plus Protein Unstained
cells expressing B-Raf variants. Mean values for relative pERK levels and Standards (Bio-Rad); 1, B-Raf-WT; 2, B-Raf-DDTS; 3, B-Raf-DN; 4, B-Raf-
standard deviations were plotted from three flow cytometry experiments (the DNDDTS. (I) Western blot analysis of MEK1 phosphorylation by B-Raf constructs
complete histograms for pERK levels in the experiments are shown in fig. S9). with the N-terminal regulatory region present and without the N-terminal
For each experiment, the pERK level for unstimulated wild-type EGFP-B-Raf regulatory region. Coomassie brilliant blue staining of the membrane shows the
transfected cells at 0 min was set to 1, and all other values were normalized to total amount of MEK1 protein loaded to each lane on the gel.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 113

RES EARCH | R E P O R T S

trajectories, each for 500 ns. In the other set, association with 14-3-3 seen in the cryo-EM on Arg509, are disrupted within ~350 ns (Fig.
the distal tail segment was deleted, and three model (Fig. 4A and fig. S12B). These simu- 4B and fig. S13). Within each kinase, we did
trajectories, each for 500 ns, were generated lations indicate that the interaction made by not see conformational transitions that corre-
(supplementary materials, materials and meth- the distal tail segment of B-RafOUT and the spond to marked departure from the ON-state,
ods). For simulations with the distal tail seg- kinase domain of B-RafIN is the principal de- such as a reorientation of helix aC or refold-
ment of B-Raf OUT left intact, the interaction terminant of the asymmetric orientation of ing of the activation loop. Nevertheless, the
between this segment and the active site of B-Raf and the 14-3-3 proteins. disruption of the interactions pinning down
B-RafIN was stable in the three independent The aspect of the Raf kinase dimer that is helix aC point to the release of a brake on
simulations, supporting our interpretation critical for maintenance of the ON-state is a such transitions, which presumably require
of the density for the distal tail segment of tight interaction between the N-lobes of the much longer simulations to be sampled.
B-Raf OUT (fig. S12A). The asymmetric organiza- two kinases (17). Key interactions are made The last 15 residues in each 14-3-3 molecule
tion of the complex was maintained throughout by the side chains of Arg509 in each kinase, are not visible in the cryo-EM map and in-
each of these simulations (fig. S12B). A differ- which form two hydrogen bonds each with clude several negatively charged residues
ent result was obtained in the simulations in the last helical turn of helix aC in the other (fig. S14). The cryo-EM structure positions pos-
which the distal tail segment of B-Raf OUT was kinase, a structural element that plays an itively charged regions in the C-lobes of the
deleted; the quaternary structure was disrup- important regulatory role (Fig. 4B) (17). In kinase domains in the general vicinity of the
ted rapidly, within a few nanoseconds in each one of the simulations with the distal tail C-terminal tails of 14-3-3. The molecular dy-
simulation, and the N-lobe of the kinase do- segment deleted, the reciprocal interactions be- namics simulations show the formation of
main of B-RafIN moved away from the close tween the N-lobes of the two kinases, centered transient ion pairs between the 14-3-3 tails

A B-Raf OUT

B-Raf IN

B-Raf OUT
distal tail
segment

14-3-3
distal tail segment present distal tail segment absent
initial structure (500 ns) (6 ns)

B B-Raf OUT B-Raf IN C C-tail of 14-3-3

B-Raf IN protomer 2

W450* W450* 14-3-3

R509* R509 protomer 2

B-Raf OUT
14-3-3
B-Raf OUT B-Raf IN

R509 R509*
W450 W450
14-3-3 protomer 1
initial structure structure obtained from simulation C-tail of 14-3-3
without distal tail segment protomer 1

Fig. 4. Molecular dynamics simulations of the B-Raf:14-3-3 complex. initial structure and (right) the structure after 500 ns of simulation. (C) Interactions
(A) Instantaneous structures from two representative simulations are shown. (Left) between the C-terminal tails of 14-3-3 and the B-Raf kinase domains. Shown here
Initial structure. (Middle) Structure after 500 ns, for one of the simulations with is a superposition of the backbone structures of the 14-3-3 tails (yellow) for
the distal tail segment intact. (Right) Structure after 6 ns, for one of the simulations three simulations with the distal tail segment of B-RafOUT intact, sampled every
with the distal tail segment deleted. Orange dashed circles indicate a region of close nanosecond over 500 ns. The 14-3-3 tails cluster around the C-lobes of the two
contact between B-RafIN and 14-3-3 in the initial structure. (B) Disruption of the B-Raf kinase domains. This occurs because of electrostatic complementarity, with
B-Raf dimer interface in one of the simulations in which the distal tail segment each instantaneous structure forming two to three ion pairs between each tail
of B-RafOUT was deleted. The interface between the kinases is shown for (left) the segment and the adjacent kinase domain (fig. S15).

114 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

and the C-lobes of the kinase domains, re- may still be able to form such an interaction 16. N. Thevakumaran et al., Nat. Struct. Mol. Biol. 22, 37–43
sulting in persistent interactions over the (fig. S10). (2015).
17. T. Rajakulendran, M. Sahmi, M. Lefrançois, F. Sicheri,
course of each of the simulations. This sug- These examples of molecular mimicry em- M. Therrien, Nature 461, 542–545 (2009).
gests that the 14-3-3 tails may provide addi- phasize that the distal tail segment of B-Raf OUT 18. N. Jura et al., Mol. Cell 42, 9–22 (2011).
tional stabilization to the complex with Raf is actually an inhibitor of B-Raf IN, and it is the 19. X. Zhang, J. Gureasko, K. Shen, P. A. Cole, J. Kuriyan, Cell 125,
1137–1149 (2006).
kinase domains (Fig. 4C and fig. S15). asymmetry of the B-Raf:14-3-3 complex that
20. P. T. C. Wan et al., Cell 116, 855–867 (2004).
ATP-bound kinases in the ON-state contain leaves one of the two active sites in the dimer 21. A. J. King et al., Cancer Res. 66, 11100–11105 (2006).
a stack of hydrophobic side chains termed the open and ready for catalysis (Figs. 1 and 2A). 22. J. R. Haling et al., Cancer Cell 26, 402–413 (2014).
catalytic spine (C-spine), which pack on either The comparison also shows that the inter- 23. M. B. Yaffe et al., Cell 91, 961–971 (1997).
side of the adenine group of ATP (15, 30). action of pSer729 with 14-3-3, coupled with 24. Q. Wang, J. A. Zorn, J. Kuriyan, Methods Enzymol. 548, 23–67
(2014).
Small-molecule inhibitors of Raf that result in dimer formation, forces each distal tail seg- 25. R. Palacios, M. Steinmetz, Cell 41, 727–734 (1985).
paradoxical activation substitute for the ade- ment of B-Raf to be positioned away from the 26. S. Whittaker et al., Sci. Transl. Med. 2, 35ra41 (2010).
nine group of ATP and stabilize the active- active site of its own kinase domain, prevent- 27. T. Brummer et al., Oncogene 25, 6262–6276 (2006).
like configuration of the C-spine. Our findings ing cis-autoinhibition. In B-Raf OUT, 20 resi- 28. A. S. Dhillon et al., Cell. Signal. 21, 1645–1651 (2009).
29. D. A. Ritt, D. M. Monson, S. I. Specht, D. K. Morrison,
are consistent with an asymmetric activation dues separate the end of helix aI and the Mol. Cell. Biol. 30, 806–819 (2010).
model for B-Raf proposed earlier, in which part of the distal tail segment that is inserted 30. S. S. Taylor, A. P. Kornev, Trends Biochem. Sci. 36, 65–77
stabilization of the C-spine through mutation into the active site of B-RafIN. Thus, it is (2011).
31. J. Hu et al., Cell 154, 1036–1046 (2013).
results in activation, even though ATP bind- plausible that in monomeric forms of Raf,
32. J. Hu et al., Mol. Cell. Biol. 35, 264–276 (2015).
ing is blocked (31, 32). The distal tail segment the C-tail acts as a cis-autoinhibitor, as the 33. A. A. Russo, P. D. Jeffrey, A. K. Patten, J. Massagué,
of B-Raf OUT converts B-Raf IN into an Òacti- corresponding segment does in twitchin ki- N. P. Pavletich, Nature 382, 325–331 (1996).
vatorÓ of B-Raf OUT by stabilizing the C-spine nase (Fig. 2D). 34. B. Kobe et al., EMBO J. 15, 6810–6821 (1996).
35. A. A. Wylie et al., Nature 543, 733–737 (2017).
during the innate activation process (Fig. 2C). The structure of the B-Raf:14-3-3 complex
36. B. Nagar et al., Cell 112, 859–871 (2003).
From this, we infer that this natural mecha- identifies the importance of the interaction
nism of activation through stabilization of an between the proximal tail segment of B-Raf
asymmetric kinase dimer is mimicked by mu- and 14-3-3 for activation. The geometry of AC KNOWLED GME NTS
tations that introduce bulky side chains into this interaction depends on the conforma- We thank X. Cao for insect cell culture and J. Paul for helpful
the active site, or by some small-molecule tion of the C-terminal helix aI of the kinase discussions and help with experiments. This work used the
Vincent J. Proteomics/Mass Spectrometry Laboratory at
inhibitors. domain, suggesting that small molecules the University of California, Berkeley, supported in part by
The way in which the distal tail segment of that bind to this region of the kinase and NIH S10 Instrumentation Grant S10RR025622. This research
one B-Raf kinase inserts into the active site of alter the disposition of aI could interfere was in part supported by the National Cancer Institute’s
National Cryo-EM Facility at the Frederick National Laboratory
the cognate kinase in the dimer has parallels with B-Raf activity. That such an approach for Cancer Research under contract HSSN261200800001E.
to inhibition mechanisms of other kinases. For might be feasible is suggested by the suc- Molecular graphics and analyses were performed with University
example, there is a striking similarity between cessful development of drugs that inhibit Abl of California, San Francisco (UCSF) ChimeraX, developed by
the Resource for Biocomputing, Visualization, and Informatics
the binding of the distal tail segment to B- by altering the conformation of helix aI (35). at UCSF, with support from NIH R01-GM129325 and
RafIN and the way the cell-cycle inhibitor Abl is regulated differently than B-Raf (36), P41-GM103311. This work used the Extreme Science and
p27Kip1 blocks the active site of cyclin-dependent but in both cases, an interaction that is cri- Engineering Discovery Environment (XSEDE), which is
supported by National Science Foundation grant ACI-1548562.
kinase 2 (CDK2) (Fig. 2D) (33). A segment of tical for regulation depends on the confor- The authors also acknowledge the Texas Advanced
the p27Kip1 protein inhibits CDK2 by entering mation of helix aI. Our discovery of a natural Computing Center (TACC) at The University of Texas at
the ATP-binding site near the hinge region, counterpart to the paradoxical activation of Austin for providing grid resources that have contributed to
the research results reported within this paper. Funding:
as does the B-Raf OUT distal tail segment. With- Raf by inhibitors suggests new avenues to This work was supported in part by a grant from the NIH
in the active site, p27Kip1 forms a single helical explore for inhibition of the Ras-MAP kinase (P01-AI091580, to J.P.R. and J.K.), a UCSF IRACDA grant
turn, which is capped by the side chain of the pathway. (K12GM081266, to S.B.), an award from the Aqueduct
Foundation, VGH & UBC Hospital Foundation (to S.S.) and a
active site lysine residue. In the B-Raf com-
Canada Excellence Research Chair Award (to S.S.). Author
plex, the corresponding residue (Lys483) is contributions: Y.K., J.O., S.S., and J.K. conceived the study.
positioned similarly. The helical turn of p27Kip1 RE FERENCES AND NOTES Y.K. and K.W. prepared reagents. Y.K., J.O., and A.M.
presents the side chain of Tyr88 for hydrogen 1. H. Lavoie, M. Therrien, Nat. Rev. Mol. Cell Biol. 16, 281–298 performed EM experiments. S.B., K.K., and J.P.R. designed
(2015). and performed cell-based assays. D.K. performed molecular
bonding with the kinase hinge, just as is seen 2. X. F. Zhang et al., Nature 364, 308–313 (1993). dynamics simulations. J.P.R., S.S., and J.K. supervised the
in the B-Raf OUT distal tail segment. Although 3. A. B. Vojtek, S. M. Hollenberg, J. A. Cooper, Cell 74, 205–214 project. All authors commented on the manuscript. Competing
the details are different, Tyr88 of p27Kip1 is (1993). interests: J.P.R. is a cofounder and scientific advisor of Seal
4. L. Van Aelst, M. Barr, S. Marcus, A. Polverino, M. Wigler, Biosciences and is on the scientific advisory committee for
flanked by hydrophobic side chains that fill Proc. Natl. Acad. Sci. U.S.A. 90, 6213–6217 (1993). the Mark Foundation for Cancer Research. J.K. is a cofounder
the space normally occupied by the adenine 5. S. A. Moodie, B. M. Willumsen, M. J. Weber, A. Wolfman, of Nurix Therapeutics and is on the scientific advisory boards
group of ATP. In twitchin kinase, an auto- Science 260, 1658–1661 (1993). of Carmot and Revolution Medicine. Data and materials
6. E. Freed, M. Symons, S. G. Macdonald, F. McCormick, availability: Accession nos. for the B-Raf:14-3-3 structure
inhibitory segment follows the last helix in R. Ruggieri, Science 265, 1713–1716 (1994). is as follows: EMD-20708 (density maps; Electron Microscopy
the kinase domain (aI), and the distal portion 7. D. K. Morrison, G. Heidecker, U. R. Rapp, T. D. Copeland, Data Bank) and PDB-6UAN (coordinates of atomic models;
of this segment enters the ATP-binding site J. Biol. Chem. 268, 17309–17316 (1993). Protein Data Bank).
8. H. Fu et al., Science 266, 126–129 (1994).
while adopting a helical structure (Fig. 2D) 9. H. Davies et al., Nature 417, 949–954 (2002).
(34). In twitchin kinase, Ile345 replaces the 10. K. T. Flaherty et al., N. Engl. J. Med. 363, 809–819 SUPPLEMENTARY MATERIALS
tyrosine residues that form hydrogen bonds (2010). science.sciencemag.org/content/366/6461/109/suppl/DC1
11. B. Agianian, E. Gavathiotis, J. Med. Chem. 61, 5775–5793 Materials and Methods
with the kinase hinge regions in the B-Raf:14-
(2018). Figs. S1 to S17
3-3 and CDK2:Cyclin A:p27Kip1 complexes. 12. S. J. Heidorn et al., Cell 140, 209–221 (2010). Tables S1 and S2
These variations may explain the absence of a 13. G. Hatzivassiliou et al., Nature 464, 431–435 (2010). References (37–62)
corresponding tyrosine residue in the distal 14. P. I. Poulikakos, C. Zhang, G. Bollag, K. M. Shokat, N. Rosen,
Nature 464, 427–430 (2010). 23 May 2019; accepted 11 September 2019
tail segments of A-Raf and C-Raf, which have 15. A. S. Shaw, A. P. Kornev, J. Hu, L. G. Ahuja, S. S. Taylor, Published online 19 September 2019
other hydrophobic residues in this region but Mol. Cell. Biol. 34, 1538–1546 (2014). 10.1126/science.aay0543

SCIENCE sciencemag.org 4 OCTOBER 2019 ¥ VOL 366 ISSUE 6461 115

RES EARCH | R E P O R T S

SYSTEMS BIOLOGY assumed that the target gene is only expressed

in cells with no free SinR. This is a truly min-
Stochastic antagonism between two proteins imal model in which the component that is
less abundant at any given time is entirely
governs a bacterial cell fate switch sequestered into inert complexes—essentially
a dynamic version of taking the difference be-
Nathan D. Lord1*†, Thomas M. Norman1,2*‡, Ruoshi Yuan1, Somenath Bakshi1§, tween two random variables. Because the total
Richard Losick2||, Johan Paulsson1|| amounts of SinI and SinR vary randomly, the
pool of free SinR alternates between periods of
Cell fate decision circuits must be variable enough for genetically identical cells to adopt a multitude of zero and nonzero abundances, and the target
fates, yet ensure that these states are distinct, stably maintained, and coordinated with neighboring gene alternates between being expressed at a
cells. A long-standing view is that this is achieved by regulatory networks involving self-stabilizing constant rate or not at all (Fig. 1B).
feedback loops that convert small differences into long-lived cell types. We combined regulatory mutants This simple model recapitulates not only the
and in vivo reconstitution with theory for stochastic processes to show that the marquee features expected ultrasensitivity (fig. S2A), but also the
of a cell fate switch in Bacillus subtilis—discrete states, multigenerational inheritance, and timing of distinct statistical properties of state switching:
commitments—can instead be explained by simple stochastic competition between two constitutively Bouts in a SinR-dominant (i.e., “motile”) state
produced proteins that form an inactive complex. Such antagonistic interactions are commonplace persist for tens or hundreds of generations and
in cells and could provide powerful mechanisms for cell fate determination more broadly. are interrupted by narrowly timed pulses of
SinR target expression at memoryless (i.e.,

C
exponentially distributed) intervals (Fig. 1, C
ell fate decisions reflect a balance be- reminiscent of the programs suggested for and D). The model also quantitatively captures
tween stochasticity and determinism: decision-making in higher organisms (3, 5, 10). the nontrivial shape of the autocorrelation
Distinct gene expression programs can Many control principles are thus consistent function observed for chaining reporter ex-
be faithfully maintained through gen- with the known dynamics and biochemistry. pression in B. subtilis strains lacking slrR
erations and coordinated between neigh- Two components required for switching (fig. S3A). Furthermore, similar qualitative
boring cells, yet different fates are often are SinR, a transcriptional repressor of genes features were preserved in an array of related
selected probabilistically from similar initial associated with the multicellular lifestyle, models that included progressively more mech-
cell states (1–5). The transition between motile- and its antagonist, SinI, which locks SinR anistic details (SM, mathematical derivations
unicellular and sessile-multicellular lifestyles into an inert complex (Fig. 1A, left panel). A sections 4 and 7).
in Bacillus subtilis is an elegant model sys- second antagonist, SlrR, whose synthesis is An exceedingly simple stochastic model
tem for studying such effects (6). Phenotypic repressed by SinR, extends and sharpens the without feedback or classical bistability thus
characterization shows that the two states can duration of the chained state but has no ef- captures the unusual kinetic principles ob-
persist for tens or even hundreds of gener- fect on the initiation process (7). Similar se- served for this cell fate switching system. How-
ations, but with markedly different temporal questration motifs arise in many biological ever, model fits are rarely unique in biology.
behavior (fig. S1). Motile cells initiate the for- processes (11–16), and mathematical models We therefore returned to experiments to test
mation of chains independently and without have shown that they can produce ultrasen- additional predictions. Specifically, the model
memory of the time spent in past states (Fig. 1E sitive switches where small changes in pro- predicts that the fraction of time spent in each
and fig. S1), whereas chains are maintained for duction rates create large changes in the state should be ultrasensitive to changes in
a narrowly distributed time window to ensure concentrations of the free components, as well either SinI or SinR production rates, but cor-
that each community reaches a critical size be- as large fluctuations in individual cells. We relate weakly with absolute abundances of
fore synchronously disassembling (Fig. 1F and therefore speculated that complex formation either component. We therefore replaced sinI
fig. S1). Switching occurs without the need for between SinI and SinR could convert small with an IPTG-inducible construct (isopropyl-
external cues, and key features of the dynamics relative differences from noisy gene expres- b-D-1-thiogalactopyranoside, an inducer of
are genetically separable, as knocking out core sion into alternating periods of high and low the lac promoter) and indeed observed ultra-
circuit components abolishes some aspects levels of free SinR, but it was unclear if this sensitive shifts in cell state as we titrated sinI
of dynamics but leaves others quantitatively model could explain the long-lived cell types expression (fig. S2B). Next, we replaced sinI
intact (7). However, the kinetic mechanisms and distinct switching dynamics observed, es- and sinR with functional HALO-tagged ver-
of switching remain unclear. Numerous com- pecially because deterministic descriptions sions (enzymatic tags for fluorescent visualiza-
ponents have been implicated in control, sug- predict a single, globally stable steady state tion of proteins, fig. S4), enabling us to observe
gesting an extensive regulatory network for all parameters (11). endogenous protein levels. As predicted, these
involving multiple transcriptional repressors, We therefore formulated a biochemically only weakly correlated with a reporter for the
master regulators, and feedback loops (8, 9), motivated but exceedingly simple stochastic chained state (Fig. 2B). The slight positive cor-
kinetic model in which SinI and SinR are relation between SinR and the chaining re-
produced constitutively and form a bimolecular porter may seem counterintuitive, but arises
1
Department of Systems Biology, Harvard Medical School, complex, and free SinR represses a downstream in models accounting for the expression of
Boston, MA 02115, USA. 2Department of Molecular and
Cellular Biology, Harvard University, Cambridge, MA
target gene. All components were assumed sinR both from a dedicated constitutive pro-
02138, USA. to be eliminated with the same decay rate to moter and from a second, bicistronic promoter
*These authors contributed equally to this work. Present address: mimic dilution due to cell growth [Fig. 1A shared with sinI (19, 20) (fig. S5).
Department of Molecular and Cellular Biology, Harvard University,
Cambridge, MA 02138, USA. àPresent address: Department of
and supplementary materials (SM), mathe- The model further predicts that other modes
Cellular and Molecular Pharmacology, University of California, matical derivations section 2]. Because the of SinI or SinR regulation, such as expression
San Francisco, San Francisco, CA 94158, USA. §Present address: affinity between SinI and SinR is extremely control or SlrR feedback, are dispensable for
Department of Engineering, University of Cambridge Cambridge
high (17), we first considered the case of im- switching—constant expression of each com-
CB2 1PZ, UK.
||Corresponding author. Email: johan_paulsson@hms.harvard. mediate and irreversible complex formation, ponent should suffice. We therefore deleted
edu (J.P.); losick@mcb.harvard.edu (R.L) and, because SinR is a strong repressor (18), we slrR and uncoupled sinI from known modes

116 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

of regulation using the IPTG-inducible con- other factors in the cell are necessary to drive monitor the circuit output (fig. S9). The re-
struct (Fig. 2A). By following gene expression the dynamics of SinI and SinR, and that the constituted circuit recapitulated the unusual
in hundreds of individual bacteria under con- components we identify simply track those dy- qualitative features of the native switch:
stant inducer concentrations for hundreds namics. For example, the switching was hy- multigenerational pulses of gene expression
of consecutive generations in a microfluidic pothesized to reflect bistability in sD-driven punctuated by long periods of inactivity, with
device (7, 21) (Fig. 2G, fig. S6, and movie S1), gene expression (9), which in principle could pulse frequencies and sizes that correlated
we found that the reduced circuit exhibited control the balance between SinI and SinR. with the production rate of SinI (Fig. 2, H
similar qualitative state switching behavior Collectively ruling out alternative mechanisms and I; movie S2; and fig. S10). It also reca-
(as measured by the chaining reporter), with is challenging because yet-to-be-discovered pitulated many quantitative principles, like
the switching rate set by the amount of factors crucial for function in principle could ultrasensitive changes in the fraction of cells
inducer (Fig. 2C). The modified circuit even exist anywhere in the extended reaction net- in the reporter-positive state as a function of
reproduced the unusual quantitative princi- work. This problem can be addressed by bio- SinI abundance (fig. S2C); memoryless entry
ples of the native switch (SM, mathematical chemical reconstitutions in test tubes, as were into the SinI dominant state (Fig. 2J and
derivations section 1): Residence times in the done for the cyanobacterial circadian clock fig. S11); subexponential distributions of res-
reporter-negative state were exponentially dis- (22) or dynamic instability in microtubules (23). idence times in the reporter-positive state
tributed, whereas pulses of reporter expres- However, the stochastic competition mecha- (Fig. 2K); a nontrivial characteristic shape
sion were precisely timed (Fig. 2, D and E). nism referred to above differs from these sys- of the autocorrelation function for the SinR
Gene expression pulses initiated with identi- tems in that it requires constant turnover in the target gene (figs. S3C and S12 and SM, math-
cal statistics when SlrR feedback was reintro- amounts of the two proteins from cell divi- ematical derivations section 8); and the stereo-
duced (fig. S7), demonstrating that SinI-SinR sion, as well as low-copy expression noise. typed shapes of reporter pulses, regardless of
competition can set the frequency of entry into We therefore reconstituted the SinI-SinR their amplitude (fig. S13). All these features
the chained state independently of feedback. circuit in Escherichia coli, which diverged were independent of cell age and fluctuations
Further, any putative regulation of sinR was from B. subtilis a billion years ago and lacks in cell physiology (fig. S14) and, to our knowl-
dispensable, as parallel experiments with an the chaining pathway. Apart from the sinI and edge, are exceptional for E. coli, whether con-
IPTG-inducible sinR construct led to similar sinR open-reading frames, the circuit was sidering the published record (21, 24, 25)
reporter pulsing, but with higher inducer con- built entirely from synthetic parts (Fig. 2F). or our own observations of transcriptional
centrations mapping onto lower frequencies We paired an IPTG-inducible sinI construct reporters under the same conditions (see
of pulsing, as expected (fig. S8B). with an expression-matched constitutive sinR fig. S12 for an example).
This degree of model validation is extensive, construct and engineered a SinR-repressible Many key dynamic features of the native
but still does not rule out the possibility that green fluorescent protein (GFP) reporter to swimming-chaining decision are thus explained

Fig. 1. A simple stochastic competition model can quantitatively explain increases from the upper to lower panel (lI/lR = 0.6, 0.7, and 0.8). The peak-
switching. (A) (Left) Logic of the SinI-SinR regulatory motif. SinI neutralizes SinR calling pipeline is discussed in detail in figs. S18 and S19. (C) Distribution of
by sequestering it away from target promoters and into an inert complex interpeak times for the simulated stochastic competition system. Data plotted in
(shaded). When unbound by SinI, SinR represses the chaining regulon. (Right) A blue, green, and orange were drawn from the simulations of matching color in (B).
simple model implementing stochastic competition. SinI and SinR (I and R) are (Inset) Log-transformed cumulative distributions of interpeak times. Under this
produced with constant probabilities lI and lR and form complexes with rate cRI. transformation, an exponential distribution yields a straight line. (D) Comparison
R represses the synthesis of a target, Z, and all components are presumed stable distributions of Z expression peak duration. (E and F), Quantitative properties
with average lifetime t. I and R are assumed to be produced one at a time for the of the native B. subtilis swimming-chaining decision. All data are derived from an
simulations in this figure (i.e., b = 1). Formal treatment of the model can be found in slrR mutant reporter strain (TMN-1158, DslrR Phag-mKate2 PtapA-cfp hagA233V)
the mathematical derivations section of the SM. (B) Realizations of stochastic used in our previous study (7). (E) Distribution of interpeak times. (Inset) Log-
competition. Component lifetimes were set to one generation to capture dilution transformed cumulative distribution of interpeak times. (F) Distribution of chaining
due to cell growth, and complex formation was rapid relative to dilution. lI/lR reporter peak duration.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 117

RES EARCH | R E P O R T S

by the simplest possible race between two tially greater production of SinI, which in- is indeed expected when a random burst of
constitutively produced proteins. The sec- creases pulse frequency because the initiation SinI expression is followed by a combination
ond SinR antagonist SlrR is not required rate is highly sensitive to SinI abundance of zero- and first-order removal through ti-
for any of these features: It is not needed to (Fig. 2, D and J; fig. S2; and SM, mathemat- tration with SinR (produced at a constant
create the two distinct states; it has no effect ical derivations sections 2 and 3). SlrR feed- rate) and dilution caused by cell growth,
on the entry into the chained state in either back should make it possible to uncouple chain respectively. In brief, dilution ensures that
the native or reduced systems; and it is not initiation from duration (26) (Fig. 3A, inset). To even widely variable bursts of SinI result in
crucial to maintain the chained state, as sim- test this hypothesis, we added the slrR gene similar durations of SinI dominance; a burst
ply modifying the constitutive rates of sinI to the reconstituted E. coli circuit under the of twice the size on average takes just one
and sinR expression can achieve this effect. control of a second synthetic SinR-repressible more generation to dilute away (Fig. 3D, left
This illustrates a key point when analyzing promoter (materials and methods, “Plasmid panel). However, dilution of the last few mol-
gene regulatory networks: Knockout pheno- and strain construction”). Reporter pulses ecules is associated with large noise (27).
types can point to a gene’s function, but to could indeed be made much longer in the Zero-order SinI removal, which starts to dom-
understand its role in control, it is often nec- presence of SlrR (Fig. 3, A and B, and movie inate over first-order dilution at low abun-
essary to consider compensatory changes S3) while leaving the memoryless initiation dances, avoids this problem by ensuring that
elsewhere in the system. For example, if a process quantitatively intact (fig. S15). The the SinI removal rate does not slow down at
knockout strain can recover native dynam- slrR system thus allows the cell to generate the end of the removal process. Thus, the
ics with slight changes in other expression large pulses of gene expression even in a combination of two processes that are noto-
rates—as we observe in this study—the gene parameter regime that creates only small, rious for randomizing intracellular concen-
of interest may play some other nontrivial role. rare bouts of SinI dominance. trations—expression bursts and zero-order
Indeed, without slrR, stochastic competi- We finally turned to the question of how titration effects—can create an exceptional
tion suffers a key limitation: Because an initial simple stochastic competition can create sub- degree of memory, timing, and coordina-
excess of SinI is approximately halved at each exponential timing for the SinI-dominant tion. Our experimental data indeed sug-
generation, higher stability requires exponen- state without feedback control. Such timing gest that periods of SinI dominance in the

Fig. 2. Stochastic competition drives cell fate switching. (A) Genetic logic of (F) Genetic logic of the reconstituted circuit in E. coli. (G) Schematic of the mother
the reduced B. subtilis circuit. System output was monitored by using a chaining machine device. Magnifying glass indicates that only “mother cells” at the
reporter gene (PtapA-cfp). (B) Single-cell SinI-HALO (TMN-1227, DsinI DslrR ylnf/ end of each cell channel were tracked. (H) Kymograph of the reconstitution strain
yloA::PsinI -sinI-HALO PtapA-cfp) and SinR-HALO (TMN-1229, DsinR DslrR ywrK:: (NDL-423, DclpXP PrpsOMod-sinR Plac-sinI PsynthR1-GFP) cells in a single lane of
PsinR--sinR-HALO PtapA-cfp) expression distributions. (C) Example traces of chaining the device as they exhibit a pulse of SinR reporter expression (orange). Frames
reporter expression in the reduced B. subtilis strain (TMN-1159, DsinI DslrR were taken 8 min apart. (I) Traces of synthetic reporter expression in the
Pspank-sinI PtapA-cfp Phag-mKate2 hag233V) grown in 7.5 mM (blue), 10 mM (green) reconstitution strain grown in 90 mM (blue), 100 mM (green), and 110 mM (orange)
and 12.5 mM (orange) IPTG. Measurements of Pspank promoter activity (fig. S2) suggest IPTG. Measurements of Plac promoter activity (fig. S2) suggest that growth
that growth in 7.5 mM, 10 mM and 12.5 mM IPTG lead to relative SinI expression in 90, 100, and 110 mM IPTG lead to relative SinI expression levels of 0.6, 0.8, and
levels of 0.4, 0.7, and 1.0, respectively. (D and E) Quantitative properties of the 1.0, respectively. (J and K) Quantitative properties of the reconstituted SinI-SinR
reduced B. subtilis system. Curve colors are matched with the data in (C). system in E. coli. Curve colors are matched with the data in (I). (J) Distribution of
(D) Distribution of interpeak times. (Inset) Log-transformed cumulative distribution interpeak times. (Inset) Log-transformed cumulative distributions of interpeak
of interpeak times. (E) Distributions of chaining reporter peak durations. times. (K) Distributions of synthetic reporter peak durations.

118 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

reconstituted system exhibit substantial tim- minutes and mimic them on a scale of hours suggested for cell fate decisions, which cannot
ing (fig. S16). or days at the level of whole-cell physiology. easily avoid exponentially distributed residence
Simple modifications of this scheme could Because the duration is determined in ad- times in each state.
create a wide range of timing dynamics. If SinI vance, this “as above, so below” principle in We find that much of the dynamical behav-
bursts were primarily removed by complex which microscopic and macroscopic features ior of the B. subtilis swimming-chaining de-
formation with SinR, the initial pool of free mirror one another can anticipate the time cision can be explained in quantitative detail
SinI would decrease linearly in time owing of the next “random” switching event and pre- by the stochastic fluctuations inherent to a
to the constant production of SinR (SM, math- pare accordingly. Mathematical analysis fur- simple antagonistic protein-protein inter-
ematical derivations section 4). The duration ther showed that adding open-loop expression action. The wide array of cell types in higher
of SinI dominance would then be determined cascades to such mechanisms can extend organisms makes metazoan cell fate deter-
by the initial net burst size (Fig. 3, C and D, the average residence times to tens or even mination appear to be very complex at first
right panel), which is set by molecular prop- hundreds of generations, without sacrificing glance. However, gene expression noise and
erties such as mRNA lifetimes. Cells could timing and coordination (see fig. S17 for de- complex formation are ubiquitous in biolog-
thus take the timing of individual molecular tails). This provides an attractive alternative to ical circuits, and a few antagonistic modules
events occurring on a time scale of seconds or the types of positive feedback loops typically operating in parallel could generate large
numbers of distinct states. Stochastic compe-
tition may thus be a driver of cell fate decision-
making more broadly, but because the requisite
Fig. 3. Tuning commit- interactions are so common and do not stand
ment to the ON state. out in qualitative genetic experiments as, for
(A) Scatter plot relating example, positive feedback loops do, this ex-
the peak duration and planation may have been hidden in plain sight.
height in the reconstitu-
REFERENCES AND NOTES
tion strain (NDL-423,
1. H. Maamar, A. Raj, D. Dubnau, Science 317, 526–529
gray points; 90, 100, and (2007).
110 mM IPTG conditions) 2. R. Losick, C. Desplan, Science 320, 65–68 (2008).
3. M. Acar, A. Becskei, A. van Oudenaarden, Nature 435, 228–232
and the SlrR+ feedback
(2005).
strain (NDL-419; identical 4. G. M. Süel, J. Garcia-Ojalvo, L. M. Liberman, M. B. Elowitz,
to NDL-423 but Nature 440, 545–550 (2006).
5. J. E. Ferrell Jr., E. M. Machleder, Science 280, 895–898
containing a SinR-
(1998).
controlled slrR allele, blue 6. D. B. Kearns, R. Losick, Genes Dev. 19, 3083–3094
points; 60, 80, and 100 mM (2005).
7. T. M. Norman, N. D. Lord, J. Paulsson, R. Losick, Nature 503,
IPTG conditions). Bulk
481–486 (2013).
trends are indicated with 8. H. Vlamakis, Y. Chai, P. Beauregard, R. Losick, R. Kolter,
sliding mean curves. Nat. Rev. Microbiol. 11, 157–168 (2013).
(Inset) Genetic logic of 9. L. M. Cozy et al., Mol. Microbiol. 83, 1210–1228 (2012).
10. R. Ahrends et al., Science 344, 1384–1389 (2014).
the reconstituted circuit. 11. J. Paulsson, M. Ehrenberg, Q. Rev. Biophys. 34, 1–59 (2001).
(B) Example traces of the 12. J. Elf, J. Paulsson, O. G. Berg, M. Ehrenberg, Biophys. J. 84,
reconstitution strain 154–170 (2003).
13. N. E. Buchler, F. R. Cross, Mol. Syst. Biol. 5, 272 (2009).
(NDL-423, gray) and 14. D. Sprinzak et al., Nature 465, 86–90 (2010).
SlrR+ feedback strain 15. J. Park et al., Cell Syst. 6, 216–229.e15 (2018).
(NDL-419, blue) grown in 16. E.Rotem et al., Proc. Natl. Acad. Sci. U.S.A. 107, 12541–12546
(2010).
90 and 60 mM IPTG, 17. J. A. Newman, C. Rodrigues, R. J. Lewis, J. Biol. Chem. 288,
respectively. (C) Sche- 10766–10778 (2013).
matic of SinI bursting in 18. D. B. Kearns, F. Chu, S. S. Branda, R. Kolter, R. Losick,
Mol. Microbiol. 55, 739–749 (2005).
stochastic competition.
19. Y. Chai, T. Norman, R. Kolter, R. Losick, EMBO J. 30,
Abundances of free SinI 1402–1413 (2011).
and SinR (dark blue and 20. N. K. Gaur, K. Cabane, I. Smith, J. Bacteriol. 170, 1046–1053
(1988).
dark orange, respec-
21. P. Wang et al., Curr. Biol. 20, 1099–1103 (2010).
tively) and bound SinI 22. M. Nakajima et al., Science 308, 414–415 (2005).
and SinR (light blue and 23. T. Mitchison, M. Kirschner, Nature 312, 237–242 (1984).
light orange, respec- 24. D. W. Austin et al., Nature 439, 608–611 (2006).
25. M. J. Dunlop, R. S. Cox 3rd, J. H. Levine, R. M. Murray,
tively) are plotted for a M. B. Elowitz, Nat. Genet. 40, 1493–1498 (2008).
single simulated trace. 26. Y. Chai, T. Norman, R. Kolter, R. Losick, Genes Dev. 24,
Total SinI and SinR levels 754–765 (2010).
27. L. Potvin-Trottier, N. D. Lord, G. Vinnicombe, J. Paulsson,
are indicated by the Nature 538, 514–517 (2016).
stacked height of the free and bound curves. Rare, large bursts of SinI (“production burst”) titrate all free SinR,
AC KNOWLED GME NTS
leading to a pool of free SinI (the “net burst”) that then decays away due to a combination of dilution and
We thank C. Saenz, V. Lien, S. Hickman, and J. Deng for help
titration by newly produced SinR. (D) Timing behaviors of stochastic competition. (Left) In the strong with microfluidic device design and fabrication. Microfluidic devices
dilution regime, bursty production produces timed periods of SinI dominance. SinI dominance durations were were fabricated at the Harvard medical School Microfluidics Facility
simulated (main panel) for processes in which SinI is produced in geometrically distributed bursts (inset) of and at the Center for Nanoscale Systems (CNS). This paper is
dedicated to the memory of Gary Barker, an exceptional scientist,
average size 10 (blue), 50 (green), and 100 (orange) molecules. (Right) In the weak dilution regime, SinI bursts
educator, and mentor. Funding: This work was supported by grants
are faithfully copied into commitment times. Dominance durations (main panel) were simulated for processes in from the NIH to R.L. (GM18568) and J.P. (GM095784). S.B. and R.Y.
which SinI mRNA has exponential (green), gamma (blue), and bimodally distributed lifetimes (orange, inset). were supported by DARPA grant HR0011-16-2-0049 and NSF

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 119

RES EARCH | R E P O R T S

awards 1517372 and 1615487 to J.P. CNS is a member of the N.D.L., T.M.N., R.L, and J.P. conceived the study, analyzed the Materials and Methods
National Nanotechnology Infrastructure Network (NNIN), which is results, and wrote the manuscript. Competing interests: None Mathematical Derivations
supported by the National Science Foundation under NSF award declared. Data and materials availability: Original data, analysis Movies S1 to S4
no. ECS-0335765. Author contributions: N.D.L. and T.M.N. code, and strains are available upon request. Reference (28)
performed experiments and simulations, designed and fabricated
microfluidic devices, and built analysis pipelines. R.Y. and J.P. SUPPLEMENTARY MATERIALS
formulated and analyzed mathematical models and performed science.sciencemag.org/content/366/6461/116/suppl/DC1 20 December 2018; accepted 12 September 2019
simulations. S.B. performed mother machine imaging experiments. Figs. S1 to S21 10.1126/science.aaw4506

BIODIVERSITY LOSS birds across almost all biomes, a reduction of

Decline of the North American avifauna

29% (95% CIs = 27–30%) since 1970 (Fig. 1 and
Table 1). Analysis of NEXRAD data indicates a
similarly steep decline in nocturnal passage of
Kenneth V. Rosenberg1,2*, Adriaan M. Dokter1, Peter J. Blancher3, John R. Sauer4, Adam C. Smith5, migratory biomass, a reduction of 13.6 ± 9.1%
Paul A. Smith3, Jessica C. Stanton6, Arvind Panjabi7, Laura Helft1, Michael Parr2, Peter P. Marra8† since 2007 (Fig. 2A). Reduction in biomass
passage occurred across the eastern United
Species extinctions have defined the global biodiversity crisis, but extinction begins with loss in abundance States (Fig. 2, C and D), where migration is
of individuals that can result in compositional and functional changes of ecosystems. Using multiple and dominated by large numbers of temperate-
independent monitoring networks, we report population losses across much of the North American avifauna and boreal-breeding songbirds; we observed
over 48 years, including once-common species and from most biomes. Integration of range-wide population no consistent trend in the Central or Pacific
trajectories and size estimates indicates a net loss approaching 3 billion birds, or 29% of 1970 abundance. flyway regions (Fig. 2, B to D, and table S5).
A continent-wide weather radar network also reveals a similarly steep decline in biomass passage of migrating Two completely different and independent
birds over a recent 10-year period. This loss of bird abundance signals an urgent need to address threats to monitoring techniques thus signal major pop-
avert future avifaunal collapse and associated loss of ecosystem integrity, function, and services. ulation loss across the continental avifauna.
Species exhibiting declines (57%, 303 out of

S
529 species) on the basis of long-term survey
lowing the loss of biodiversity is one of effective long-term monitoring of population data span diverse ecological and taxonomic
the defining environmental challenges of sizes and trends, data that are rarely available groups. Across breeding biomes, grassland birds
the 21st century (1–5). Habitat loss, cli- for most taxa. showed the largest magnitude of total popu-
mate change, unregulated harvest, and Birds are excellent indicators of environ- lation loss since 1970—more than 700 million
other forms of human-caused mortality mental health and ecosystem integrity (16, 17), breeding individuals across 31 species—and
(6, 7) have contributed to a thousandfold in- and our ability to monitor many species over the largest proportional loss (53%); 74% of
crease in global extinctions in the Anthropocene vast spatial scales far exceeds that of any other grassland species are declining. (Fig. 1 and
compared to the presumed prehuman back- animal group. We evaluated population change Table 1). All forest biomes experienced large
ground rate, with profound effects on ecosystem for 529 species of birds in the continental avian loss, with a cumulative reduction of more
functioning and services (8). The overwhelm- United States and Canada (76% of breeding than 1 billion birds. Wetland birds represent
ing focus on species extinctions, however, has species), drawing from multiple standardized the only biome to show an overall net gain
underestimated the extent and consequences bird-monitoring datasets, some of which pro- in numbers (13%), led by a 56% increase in
of biotic change, by ignoring the loss of abun- vide close to 50 years of population data. We waterfowl populations (Fig. 3 and Table 1).
dance within still-common species and in ag- integrated range-wide estimates of popula- Unexpectedly, we also found a large net loss
gregate across large species assemblages (2, 9). tion size and 48-year population trajectories, (63%) across 10 introduced species (Fig. 3, D
Declines in abundance can degrade ecosystem along with their associated uncertainty, to and E, and Table 1).
integrity, reducing vital ecological, evolution- quantify net change in numbers of birds across A total of 419 native migratory species ex-
ary, economic, and social services that orga- the avifauna over recent decades (18). We also perienced a net loss of 2.5 billion individuals,
nisms provide to their environment (8, 10–15). used a network of 143 weather radars (NEXRAD) whereas 100 native resident species showed a
Given the current pace of global environmen- across the contiguous United States to estimate small net increase (26 million). Species over-
tal change, quantifying change in species abun- long-term changes in nocturnal migratory pas- wintering in temperate regions experienced the
dances is essential to assess ecosystem impacts. sage of avian biomass through the airspace largest net reduction in abundance (1.4 billion),
Evaluating the magnitude of declines requires in spring from 2007 to 2017. The continuous but proportional loss was greatest among spe-
operation and broad coverage of NEXRAD cies overwintering in coastal regions (42%),
provide an automated and standardized mon- southwestern aridlands (42%), and South
1
Cornell Lab of Ornithology, Cornell University, Ithaca, NY 14850, itoring tool with unrivaled temporal and spa- America (40%) (Table 1 and fig. S1). Shorebirds,
USA. 2American Bird Conservancy, Washington, DC 20008,
USA. 3National Wildlife Research Centre, Environment and
tial extent (19). Radar measures cumulative most of which migrate long distances to winter
Climate Change Canada, Ottawa, ON K1A 0H3, Canada. passage across all nocturnally migrating spe- along coasts throughout the hemisphere, are
4
Patuxent Wildlife Research Center, United States Geological cies, many of which breed in areas north of experiencing consistent, steep population
Survey, Laurel, MD 20708-4017, USA. 5Canadian Wildlife
Service, Environment and Climate Change Canada, Ottawa, ON
the contiguous United States that are poorly loss (37%).
K1A 0H3, Canada. 6Upper Midwest Environmental Sciences monitored by avian surveys. Radar thus ex- More than 90% of the total cumulative loss
Center, United States Geological Survey, La Crosse, WI, USA.
7
pands the area and the proportion of the can be attributed to 12 bird families (Fig. 3A),
Bird Conservancy of the Rockies, Fort Collins, CO 80521, USA. migratory avifauna that is sampled relative to
8 including sparrows, warblers, blackbirds, and
Migratory Bird Center, Smithsonian Conservation Biology
Institute, National Zoological Park, P.O. Box 37012 MRC 5503, ground surveys. finches. Of 67 bird families surveyed, 38 showed
Washington, DC 20013-7012, USA. Results from long-term surveys, accounting a net loss in total abundance, whereas 29 showed
*Corresponding author. Email: kvr2@cornell.edu for both increasing and declining species, re- gains (Fig. 3B), indicating recent changes in
†Present address: Department of Biology and McCourt School of
Public Policy, Georgetown University, 37th and O Streets NW, veal a net loss in total abundance of 2.9 billion avifaunal composition (table S2). Although not
Washington, DC 20057, USA. [95% credible interval (CI) = 2.7–3.1 billion] optimized for species-level analysis, our model

120 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

Fig. 1. Net population change in North American birds. (A) By integrating Bird Conservation Regions and land cover classification (18). (B) Net
population size estimates and trajectories for 529 species (18), we show loss of abundance occurred across all major breeding biomes
a net loss of 2.9 billion breeding birds across the continental avifauna except wetlands (see Table 1). (C) Proportional net population change
since 1970. Gray shading represents the 95% credible interval (CI) around relative to 1970, ±95% CI. (D) Proportion of species declining in
total estimated loss. Map shows color-coded breeding biomes based on each biome.

loss

Fig. 2. NEXRAD radar monitoring of nocturnal bird migration across the 1 July), estimated for the period 2007–2017. Darker red colors indicate higher
contiguous United States. (A) Annual change in biomass passage for the declines and loss of biomass passage, whereas blue colors indicate biomass
full continental United States (black) and (B) the Pacific (green), Central increase. Circle size indicates trend significance, with closed circles being
(brown), Mississippi (yellow), and Atlantic (blue) flyways [borders indicated in significant at a 95% confidence level. Only areas outside gray shading have a
(C)], with percentage of total biomass passage (migration traffic) for each spatially consistent trend signal separated from background variability.
flyway indicated; declines are significant only for the full United States and (D) Ten-year cumulative loss in biomass passage, estimated as the product of
the Mississippi and Atlantic flyways (tables S3 to S5). (C) Single-site trends in a spatially explicit (generalized additive model) trend, times the surface of
seasonal biomass passage at 143 NEXRAD stations in spring (1 March to average cumulative spring biomass passage.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 121

RES EARCH | R E P O R T S

Fig. 3. Gains and losses across the North American avifauna over the past total gain and total loss yields a net loss of 2.9 billion birds across the entire
half-century. (A) Bird families were categorized as having a net loss (red) or avifauna. (C) For each individually represented family in (B) and (C), proportional
gain (blue). Total loss of 3.2 billion birds occurred across 38 families; each family population change within that family is shown. See table S2 for statistics on
with losses greater than 50 million individuals is shown as a proportion of each individual family. (D) Percentage population change among introduced
total loss, including two introduced families (gray). Swallows, nightjars, and and each of four management groups (18). A representative species from
swifts together show loss within the aerial insectivore guild. (B) Twenty-nine each group is shown (top to bottom, house sparrow, Passer domesticus;
families show a total gain of 250 million individual birds; the five families with sanderling, Calidris alba; western meadowlark, Sturnella neglecta; green heron,
gains greater than 15 million individuals are shown as a proportion of total Butorides virescens; and snow goose, Anser caerulescens). (E) Proportion of
gain. Four families of raptors are shown as a single group. Note that combining species with declining trends.

indicates that 19 widespread and abundant (Fig. 3). Notably, our population loss estimates their population reductions and possible ex-
landbirds (including two introduced species) are conservative because we estimated loss tinctions will have severe direct and indirect
each experienced population reductions of only in breeding populations. The total loss and consequences (10, 22). Population declines can
>50 million birds (data S1). Abundant species impact on communities and ecosystems could be reversed, as evidenced by the exceptional
also contribute strongly to the migratory pas- be even higher outside the breeding season if recovery of waterfowl populations under adapt-
sage detected by radar (19), and radar-derived we consider the amplifying effect of “missing” ive harvest management (23) and the associ-
trends provide a fully independent estimate of reproductive output from these lost breeders. ated allocation of billions of dollars devoted to
widespread declines of migratory birds. Extinction of the passenger pigeon (Ectopistes wetland protection and restoration, providing
Our study documents a long-developing migratorius), once likely the most numerous a model for proactive conservation in other
but overlooked biodiversity crisis in North bird on the planet, provides a poignant re- widespread native habitats such as grasslands.
America—the cumulative loss of nearly 3 billion minder that even abundant species can go Steep declines in North American bird pop-
birds across the avifauna. Population loss is extinct rapidly. Systematic monitoring and ulations parallel patterns of avian declines
not restricted to rare and threatened species, attention paid to population declines could emerging globally (14, 15, 22, 24). In particu-
but includes many widespread and common have alerted society to its pending extinction lar, depletion of native grassland bird pop-
species that may be disproportionately influ- (20). Today, monitoring data suggest that ulations in North America, driven by habitat
ential components of food webs and ecosystem avian declines will likely continue without loss and more toxic pesticide use in both breed-
function. Furthermore, losses among habi- targeted conservation action, triggering addi- ing and wintering areas (25), mirrors loss of
tat generalists and even introduced species tional endangered species listings at tremen- farmland birds throughout Europe and else-
indicate that declining species are not replaced dous financial and social cost. Moreover, where (15). Even declines among introduced
by species that fare well in human-altered because birds provide numerous benefits to species match similar declines within these
landscapes. Increases among waterfowl and ecosystems (e.g., seed dispersal, pollination, same species’ native ranges (26). Agricultural
a few other groups (e.g., raptors recovering pest control) and economies [47 million people intensification and urbanization have been
after the banning of DDT) are insufficient to spend U.S.$9.3 billion per year through bird- similarly linked to declines in insect diversity
offset large losses among abundant species related activities in the United States (21)], and biomass (27), with cascading impacts on

122 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

Table 1. Net change in abundance across the North American avifauna, 1970–2017. Species are grouped into native and introduced species, management
groups (landbirds, shorebirds, waterbirds, waterfowl), major breeding biomes, and nonbreeding biomes [see data S1 in (18) for assignments and definitions of
groups and biomes]. Net change in abundance is expressed in millions of breeding individuals, with upper and lower bounds of each 95% credible interval (CI)
shown. Percentage of species in each group with negative trend trajectories is also noted. Values in bold indicate declines and loss; those in italics indicate gains.

Net abundance Percent change

change (millions) and 95% CIs and 95% CIs Proportion species
Species group No. of species
in decline
Change LC95 UC95 Change LC95 UC95
Species summary
............................................................................................................................................................................................................................................................................................................................................
All N. Am. species 529 –2,911.9 –3,097.5 –2,732.9 –28.8% –30.2% –27.3% 57.3%
............................................................................................................................................................................................................................................................................................................................................
All native species 519 –2,521.0 –2,698.5 –2,347.6 –26.5% –28.0% –24.9% 57.4%
............................................................................................................................................................................................................................................................................................................................................
Introduced species 10 –391.6 –442.3 –336.6 –62.9% –66.5% –56.4% 50.0%
............................................................................................................................................................................................................................................................................................................................................
Native migratory species 419 –2,547.7 –2,723.7 –2,374.5 –28.3% –29.8% –26.7% 58.2%
............................................................................................................................................................................................................................................................................................................................................
Native resident species 100 26.3 7.3 46.9 5.3% 1.4% 9.6% 54.0%
............................................................................................................................................................................................................................................................................................................................................
Landbirds 357 –2,516.5 –2,692.2 –2,346.0 –27.1% –28.6% –25.5% 58.8%
............................................................................................................................................................................................................................................................................................................................................
Shorebirds 44 –17.1 –21.8 –12.6 –37.4% –45.0% –28.8% 68.2%
............................................................................................................................................................................................................................................................................................................................................
Waterbirds 77 –22.5 –37.8 –6.3 –21.5% –33.1% –6.2% 51.9%
............................................................................................................................................................................................................................................................................................................................................
Waterfowl 41 34.8 24.5 48.3 56.0% 37.9% 79.4% 43.9%
............................................................................................................................................................................................................................................................................................................................................
Aerial insectivores 26 –156.8 –183.8 –127.0 –31.8% –36.4% –26.1% 73.1%
............................................................................................................................................................................................................................................................................................................................................
Breeding biome
............................................................................................................................................................................................................................................................................................................................................
Grassland 31 –717.5 –763.9 –673.3 –53.3% –55.1% –51.5% 74.2%
............................................................................................................................................................................................................................................................................................................................................
Boreal forest 34 –500.7 –627.1 –381.0 –33.1% –38.9% –26.9% 50.0%
............................................................................................................................................................................................................................................................................................................................................
Forest generalist 40 –482.2 –552.5 –413.4 –18.1% –20.4% –15.8% 40.0%
............................................................................................................................................................................................................................................................................................................................................
Habitat generalist 38 –417.3 –462.1 –371.3 –23.1% –25.4% –20.7% 60.5%
............................................................................................................................................................................................................................................................................................................................................
Eastern forest 63 –166.7 –185.8 –147.7 –17.4% –19.2% –15.6% 63.5%
............................................................................................................................................................................................................................................................................................................................................
Western forest 67 –139.7 –163.8 –116.1 –29.5% –32.8% –26.0% 64.2%
............................................................................................................................................................................................................................................................................................................................................
Arctic tundra 51 –79.9 –131.2 –0.7 –23.4% –37.5% –0.2% 56.5%
............................................................................................................................................................................................................................................................................................................................................
Aridlands 62 –35.6 –49.7 –17.0 –17.0% –23.0% –8.1% 56.5%
............................................................................................................................................................................................................................................................................................................................................
Coasts 38 –6.1 –18.9 8.5 –15.0% –39.4% 21.9% 50.0%
............................................................................................................................................................................................................................................................................................................................................
Wetlands 95 20.6 8.3 35.3 13.0% 5.1% 23.0% 47.4%
............................................................................................................................................................................................................................................................................................................................................
Nonbreeding biome
............................................................................................................................................................................................................................................................................................................................................
Temperate N. America 192 –1,413.0 –1,521.5 –1,292.3 –27.4% –29.3% –25.3% 55.2%
............................................................................................................................................................................................................................................................................................................................................
South America 41 –537.4 –651.1 –432.6 –40.1% –45.2% –34.6% 75.6%
............................................................................................................................................................................................................................................................................................................................................
Southwestern aridlands 50 –238.1 –261.2 –215.6 –41.9% –44.5% –39.2% 74.0%
............................................................................................................................................................................................................................................................................................................................................
Mexico–Central America 76 –155.3 –187.8 –122.0 –15.5% –18.3% –12.6% 52.6%
............................................................................................................................................................................................................................................................................................................................................
Widespread neotropical 22 –126.0 –171.2 –86.1 –26.8% –33.4% –19.3% 45.5%
............................................................................................................................................................................................................................................................................................................................................
Widespread 60 –31.6 –63.1 1.6 –3.7% –7.4% 0.2% 43.3%
............................................................................................................................................................................................................................................................................................................................................
Marine 26 –16.3 –29.7 –1.2 –30.8% –49.1% –2.5% 61.5%
............................................................................................................................................................................................................................................................................................................................................
Coastal 44 –11.0 –14.9 –6.7 –42.0% –51.8% –26.7% 68.2%
............................................................................................................................................................................................................................................................................................................................................
Caribbean 8 –6.0 1.4 –15.7 12.1% –2.8% 31.7% 25.0%
............................................................................................................................................................................................................................................................................................................................................

birds and other consumers (24, 28, 29). Given research to identify limiting factors must be REFERENCES AND NOTES
that birds are one of the best monitored ani- coupled with effective policies and societal 1. M. C. Urban, Science 348, 571–573 (2015).
mal groups, birds may also foreshadow a much change that emphasize reducing threats to 2. R. Dirzo et al., Science 345, 401–406 (2014).
3. S. L. Pimm et al., Science 344, 1246752 (2014).
larger problem, indicating similar or greater breeding and nonbreeding habitats and min-
4. A. D. Barnosky et al., Nature 471, 51–57 (2011).
losses in other taxonomic groups (28, 30). imizing avoidable anthropogenic mortality 5. W. Steffen, J. Crutzen, J. R. McNeill, Ambio 36, 614–621 (2007).
Pervasiveness of avian loss across biomes year-round. Endangered species legislation 6. S. R. Loss, T. Will, P. P. Marra, Annu. Rev. Ecol. Evol. Syst. 46,
and bird families suggests multiple and inter- and international treaties, such as the 1916 99–120 (2015).
acting threats. Isolating spatiotemporal limiting Migratory Bird Treaty between Canada and 7. A. M. Calvert et al., Avian Conserv. Ecol. 8, art11 (2013).
8. D. U. Hooper et al., Nature 486, 105–108 (2012).
factors for individual species and populations the United States, have prevented extinctions
9. G. Ceballos, P. R. Ehrlich, R. Dirzo, Proc. Natl. Acad. Sci. U.S.A.
will require additional study, however, because and promoted recovery of once-depleted bird 114, E6089–E6096 (2017).
migratory species with complex life histories species. History shows that conservation action 10. C. J. Whelan, Ç. H. Şekercioğlu, D. G. Wenny, J. Ornithol. 156
are in contact with many threats throughout and legislation work. Our results signal an (S1), 227–238 (2015).
11. M. Galetti et al., Science 340, 1086–1090 (2013).
their annual cycles. A focus on breeding sea- urgent need to address the ongoing threats 12. G. C. Daily, Ed., Nature’s Services: Societal Dependence
son biology hampers our ability to understand of habitat loss, agricultural intensification, on Natural Ecosystems (Island Press, Washington,
how seasonal interactions drive population coastal disturbance, and direct anthropogenic DC, 1997).
change (31), although recent continent-wide mortality, all exacerbated by climate change, 13. S. Bauer, B. J. Hoye, Science 344, 1242552 (2014).
14. K. J. Gaston, R. A. Fuller, Trends Ecol. Evol. 23, 14–19
analyses affirm the importance of events during to avert continued biodiversity loss and po- (2008).
the nonbreeding season (19, 32). Targeted tential collapse of the continental avifauna. 15. R. Inger et al., Ecol. Lett. 18, 28–36 (2015).

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 123

RES EARCH | R E P O R T S

16. M. L. Morrison, in Current Ornithology, R. F. Johnston, Ed. FOREST ECOLOGY

(Springer US, Boston, MA, 1986; https://link.springer.com/10.
1007/978-1-4615-6784-4_10), pp. 429–451.
17. J. Burger, M. Gochfeld, EcoHealth 1, 263–274 (2004).
18. See supplementary materials.
Differential soil fungus accumulation and density
19. A. M. Dokter et al., Nat. Ecol. Evol. 2, 1603–1609
(2018). dependence of trees in a subtropical forest
20. J. C. Stanton, Biol. Conserv. 180, 11–20 (2014).
21. U.S. Department of the Interior, U.S. Fish and Wildlife Service,
Lei Chen1, Nathan G. Swenson2, Niuniu Ji3, Xiangcheng Mi1, Haibao Ren1,
and U.S. Department of Commerce, U.S. Census Bureau,
“National Survey of Fishing, Hunting, and Wildlife-Associated Liangdong Guo3, Keping Ma1*
Recreation” (2016).
22. C. H. Sekercioğlu, G. C. Daily, P. R. Ehrlich, Proc. Natl. Acad.
Sci. U.S.A. 101, 18042–18047 (2004).
The mechanisms underlying interspecific variation in conspecific negative density dependence
23. J. D. Nichols, M. C. Runge, F. A. Johnson, B. K. Williams, (CNDD) are poorly understood. Using a multilevel modeling approach, we combined long-term
J. Ornithol. 148 , 343–349 (2007). seedling demographic data from a subtropical forest plot with soil fungal community data by
24. C. A. Hallmann, R. P. B. Foppen, C. A. M. van Turnhout,
means of DNA sequencing to address the feedback of various guilds of soil fungi on the
H. de Kroon, E. Jongejans, Nature 511, 341–343 (2014).
25. R. L. Stanton, C. A. Morrissey, R. G. Clark, Agric. Ecosyst. density dependence of trees. We show that mycorrhizal type mediates tree neighborhood
Environ. 254, 244–254 (2018). interactions at the community level, and much of the interspecific variation in CNDD is explained
26. J. De Laet, J. D. Summers-Smith, J. Ornithol. 148 , 275–278
by how tree species differ in their fungal density accumulation rates as they grow. Species with
(2007).
27. F. Sánchez-Bayo, K. A. G. Wyckhuys, Biol. Conserv. 232, 8–27 higher accumulation rates of pathogenic fungi suffered more from CNDD, whereas species with
(2019). lower CNDD had higher accumulation rates of ectomycorrhizal fungi, suggesting that mutualistic
28. B. C. Lister, A. Garcia, Proc. Natl. Acad. Sci. U.S.A. 115, and pathogenic fungi play important but opposing roles.
E10397–E10406 (2018).
29. D. L. Narango, D. W. Tallamy, P. P. Marra, Proc. Natl. Acad. Sci.

C
U.S.A. 115, 11549–11554 (2018).
30. R. E. A. Almond, M. Grooten, “Living Planet Report - 2018: onspecific negative density dependence trients and protection from pathogens, re-
Aiming Higher” (WWF, Gland, Switzerland, 2018). (CNDD) is thought to explain species co- sulting in monodominance through positive
31. P. P. Marra, E. B. Cohen, S. R. Loss, J. E. Rutter, C. M. Tonra,
Biol. Lett. 11, 20150552 (2015).
existence in diverse tree communities plant-soil feedback (15). Saprotrophic fungi
32. F. A. La Sorte et al., Glob. Change Biol. 23, 5284–5296 (1, 2). Coexistence theories suggest that are major agents of litter decomposition and
(2017). specialist natural enemies accumulate nutrient cycling and might also have an ef-
33. A. C. Smith, AdamCSmithCWS/
Estimating_Change_in_NorthAmerican_Birds, Zenodo (2019);
near dense patches of their hosts and attack fect on species coexistence through altering
https://doi.org/10.5281/zenodo.3218403. seeds and seedlings of the same species, ulti- nutrient availability (16, 17). Recent work in-
34. A. M. Dokter, L. Veen, J. H. Spaaks, adokter/vol2bird: vol2bird, mately lending an advantage to locally rare vestigating the strength of negative density
Version 0.4.0, Zenodo (2019); https://doi.org/10.5281/
zenodo.3369999.
species (3, 4). Experimental applications of dependence in North American trees has
35. A. M. Dokter, S. Van Hoey, P. Desmet, adokter/bioRad: bioRad, fungi (5, 6) or fungicides and soil sterilization found that tree mycorrhizal association could
Version 0.4.0, Zenodo (2019); https://doi.org/10.5281/ (7, 8) and reciprocal transplants (9, 10) pro- explain plant responses to their conspecific
zenodo.3370005.
vide compelling evidence for the diversity- neighbors, where ectomycorrhizal (EcM) spe-
ACKN OW LEDG MEN TS promoting effects of pathogens on local tree cies experience weaker CNDD than do arbus-
This paper is a contribution of The Partners in Flight International neighborhoods. Tree species vary greatly cular mycorrhizal (AM) species, potentially
Science Committee and the American Ornithologist Society in how they respond to conspecific density because, compared with AM fungi, EcM fungi
Conservation Committee, and the study benefited from many
discussions with these groups. S. Bessinger, J. Fitzpatrick, S. Loss,
within a community (11, 12), yet few studies act as superior protectors from antagonists
T. Scott Sillett, W. Hochachka, D. Fink, S. Kelling, V. Ruiz-Gutierrez, have focused on the underlying mechanisms (18). Thus, a current research challenge is to
O. Robinson, E. Miller, A. Rodewald, and three anonymous that lead to such variation among species in understand how different functional types of
reviewers made suggestions to improve the paper. J. Ditner and
M. Strimas-Mackey helped with figures and graphics. T. Meehan
CNDD and have inspired much debate. Key fungi mediate conspecific neighbor effects and
provided an analysis of trends from National Audubon’s Christmas studies of correlations between the strength whether feedback with soil fungi contributes
Bird Count. We thank the hundreds of volunteer citizen-scientists of CNDD and species abundance in tropical to a better mechanistic understanding of var-
who contributed to long-term bird-monitoring programs in North
forests suggest that at finer spatial scales iation among tree species in their patterns
America and the institutions that manage these programs. Photos
in Fig. 3 are from Macaulay Library, Cornell Lab of Ornithology. (<10 to 20 m), species that are more abundant of CNDD.
Funding: NSF LTREB DEB1242584 to P.P.M.; AWS Cloud Credits actually suffer less from CNDD than rarer Here, we present a study combining long-
for Research to A.M.D.; NSF ABI Innovation DBI-1661259. Author species do (i.e., a rare-species disadvantage) term seedling survival measurements and an
contributions: All authors conceived of the idea for the paper;
A.C.S., P.J.B., A.M.D., J.R.S., P.A.S., and J.C.S. conducted analyses; (9, 12). assay of the fungal community to address the
K.V.R., A.M.D., and P.P.M. primarily wrote the paper, although all Soil fungi are one of the most diverse groups feedback of various guilds of soil fungi on the
authors contributed to the final manuscript. Competing interests: of microorganisms in the world and can in- CNDD of tree species in a 24-ha Gutianshan
M.P. is president, and a member of the board of directors, of the
American Bird Conservancy. All remaining authors declare no
teract intimately with plants as pathogens, subtropical forest dynamics plot. Species in
competing interests. Data and materials availability: All data and symbiotic mutualists, and decomposers (13). the plot can be classified into three different
software are archived and available on Zenodo (33–35) and will The accumulation of fungal pathogens could groups: AM, EcM, and ericoid mycorrhizal
be published in future versions of the Avian Conservation
Assessment Database (http://pif.birdconservancy.org/ACAD/).
promote and maintain tree community diver- (ErM) plants, and more than half (56.4%) of
sity by reducing the recruitment and sur- the total basal area was contributed by EcM
SUPPLEMENTARY MATERIALS vival of dominant species (14), whereas fungal trees. We test two hypotheses with respect to
science.sciencemag.org/content/366/6461/120/suppl/DC1 mutualists can benefit plants by offering nu- tree-fungus associations and their roles in
Materials and Methods driving interspecific variation in the strength
Figs. S1 to S7
Tables S1 to S5
of CNDD. First, we hypothesize that species
Databases S1 and S2 1
State Key Laboratory of Vegetation and Environmental
that accumulate pathogenic fungi through
References (36–101) Change, Institute of Botany, Chinese Academy of Sciences, tree ontogeny more quickly are likely to suf-
Beijing 100093, China. 2Department of Biology, University of fer more negative demographic impacts when
20 November 2018; resubmitted 23 May 2019 Maryland, College Park, MD 20742, USA. 3State Key
Accepted 5 September 2019
conspecific densities are higher. Second, we
Laboratory of Mycology, Institute of Microbiology, Chinese
Published online 19 September 2019 Academy of Sciences, Beijing 100101, China. hypothesize that the variation in functional
10.1126/science.aaw1313 *Corresponding author. Email: kpma@ibcas.ac.cn types of soil fungi (pathogenic, AM, EcM, and

124 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

Fig. 1. Tree size effects on the abundance and richness of each fungal black lines) are all significant compared with no relationship at significance
functional group. The GLMM fits the log odds of (A) fungal sequence and level a = 0.05, and dashed black lines show 95% bootstrapped credible
(B) OTU richness for plant pathogens, EcM fungi, and saprotrophs by a linear intervals. Gray lines represent the predicted species-specific responses if
function of log-transformed tree DBH, whereas a linear mixed model fits tree size effects varied significantly across species (LRT test, P < 0.05).
the tree size effects on (C) AM fungal OTU richness. Overall curves (solid There were 319 individual trees from 34 species included in these models.

saprotrophic) accumulated through tree on- tocol (23). For 28 of the 34 plant species that resulted in 141 OTUs after excluding the non-
togeny is itself related to variation among spe- had a sufficient number of seedlings, two-level AM fungal OTUs, ranging from 21 to 65 per
cies in the strength of CNDD. mixed models were then used to test whether sample. These OTUs covered all four AM fun-
To test these hypotheses, we first collected species-level variation in density-dependent gal orders (27.0% to Archaeosporales, 12.1%
soil from the rooting zone of 322 individual trees seedling survival can be predicted by the rate to Diversisporales, 54.6% to Glomerales, and
from 34 species (23 AM, 8 EcM, and 3 ErM) at which each tree species accumulated vari- 5.0% to Paraglomerales), and Glomerales con-
once at the end of the seedling census in the ous guilds of soil fungi. Lastly, we also tested tributed 85.0% of sequences (fig. S2). Com-
same plot. High-throughput DNA sequenc- whether fungi were host specific by tallying pared with tropical forests in northern South
ing was used to measure the soil fungal taxa the number of tree species on which each taxon America (25), the proportion of EcM fungal
associated with each focal tree. Each taxon was found, as host specificity for soil fungi in taxa was much higher (20.8% versus 8.0%), and
was identified as a pathogenic, AM, EcM, or plant rhizospheres was an essential compo- plant pathogens were relatively less abundant
saprotrophic fungi on the basis of the most nent in generating differential fungal effects and diverse than in lowland AM-dominated
updated knowledge of functional annotations on plants. tropical forests.
of fungal genera (19). We quantified the accu- We used ITS1 and 18S barcoding primers to Most fungal taxa were associated with sev-
mulation rate of fungal species in each tree sequence the total fungal community and AM eral tree hosts, and only 609 of the taxa detected
species by characterizing the abundance and fungal community, respectively, at a 97% se- by using ITS1 sequencing and 18 of the taxa
richness of each fungal guild of conspecific quence similarity level. Clustering at a cutoff detected by using 18S sequencing were found
individuals in different tree size classes {i.e., sequence similarity of 97% is often considered from a single host species (fig. S3). Of the major
using individuals with varying diameters at a compromise between natural intraspecific functional groups, pathogens were the most
1.3 m above ground [diameter at breast height and interspecific sequence variation and ran- specialized guild on the basis of their having
(DBH)]} (19). We used this approach because dom sequencing errors (24). We obtained 6261 the highest mean species specialization indices,
larger trees within a species are generally operational taxonomic units (OTUs) affiliated followed by saprotrophs, EcM, and then AM
older than smaller individuals in environments with the ITS1 region, of which 70.7% could be fungi (fig. S4). Similarity in a fungal commu-
that are relatively homogeneous spatially and assigned to different fungal functional groups. nity between any pair of hosts was negatively
temporally (20) and because our study orga- The largest group was saprotrophs, which ac- but weakly correlated with phylogenetic dis-
nisms are long lived, making repeat sampling counted for 2705 (43.2%) of the taxa, followed tance (Mantel test, P > 0.1). Adonis permutation
through the life span of an individual not fea- by EcM fungi (1304; 20.8%) and then plant analysis revealed that host phylogeny explained
sible. Second, we used a neighborhood model- pathogens (242; 3.9%) (fig. S1). However, EcM only a small but statistically significant portion
ing approach (21, 22) to measure conspecific fungi contributed to the largest proportion of the variation (<2.7%) in fungal communities
neighborhood effects on seedlings from natu- of sequences (48.4%), reflecting the domi- after accounting for the effect of the host spe-
ral seed-fall recruitment in nine censuses from nance of EcM tree species in the plot. For the cies (<3.6%) (table S1). However, it remains pos-
2006 to 2014, following a standard census pro- 18S primer system, the AM fungal dataset sible that the impacts of plant-fungus interactions

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 125

RES EARCH | R E P O R T S

will follow a strong phylogenetic signal, as re- Pathogen accumulation rate also depended on the positive seedling response to conspecific
vealed by inoculation tests (26, 27). mycorrhizal association of tree species. For saplings reflects higher seedling survival in
A wide array of antagonists is thought to example, AM tree species accumulated path- preferred habitats (2).
accumulate in both diversity and density over ogen sequences faster through ontogeny than Turning to the fungal guild level, seedling
tree ontogeny and can help inhibit the estab- did EcM tree species (fig. S5), implying greater survival responses to the neighborhood varia-
lishment of conspecifics near larger adults (1). pathogen pressure under an adult AM tree bles varied considerably among mycorrhizal
We quantified the fungus accumulation rate by than under an adult EcM tree. types, and density-dependent mortality was
examining how the sequences and OTU rich- To evaluate the strength of CNDD, we most common for AM seedlings. Compared
ness of different fungal guilds affiliated with modeled annual seedling survival from 2006 with EcM and ErM species, AM seedlings near
focal trees change with focal tree size (Fig. 1). to 2014 as a function of the density and iden- other conspecific seedlings and adult neigh-
In an average species, larger trees had a sig- tity of neighbors using generalized linear mixed bors were more negatively affected than those
nificantly higher proportion of pathogens and models (GLMMs) for all species combined and near heterospecific seedling and conspecific
saprotrophs in both the number of sequences for seedlings belonging to different mycorrhizal sapling neighbors (Fig. 2, A and B, and table
and the number of OTUs. Conversely, AM fungal types. At the community level, we found dis- S3). By dividing heterospecific neighbors into
OTU richness and the proportion of EcM fungi tinct effects between conspecific and hetero- AM and EcM species, AM seedlings benefited
in sequences and OTUs decreased with focal specific neighbors. Seedling survival significantly (i.e., higher survival rates) from the presence
tree size. As size class increased, there was declined with the density of conspecific seedling of both heterospecific AM and EcM seedling
significant variation among tree species in and adult neighbors (Fig. 2A), confirming the neighbors, and EcM seedling survival was sig-
the rate at which the pathogens, EcM fungi, prevalence of CNDD (28). However, both het- nificantly enhanced by the density of hetero-
and saprotrophs accumulated (Fig. 1, A and B). erospecific seedling and conspecific sapling specific AM seedlings and EcM adults (Fig. 2,
Although AM fungi appear ubiquitous, the neighbors had significant positive effects on C and D), supporting the safe-site hypothesis
effect of tree size on AM fungal OTU richness seedling survival (Fig. 2, A and B). These are (30). That is, heterospecific neighbors may
varied little among species, likely because of in line with previous findings that seedling improve survival by offering AM and EcM tree
their low host specificity (Fig. 1C). Tree species survival was enhanced by increased heter- seedlings a site safe from natural enemies,
with lower EcM fungus accumulation rates of ospecific crowding as a result of fewer en- with the exception of ErM species (table S3).
sequences generally had higher saprotroph counters with species-specific natural enemies A recent study of spatial associations of AM
or pathogen accumulation rates (table S2). (i.e., species herd protection) (22, 29), whereas and EcM trees predicted that opposing mycor-
rhizal associations had significant inhibition
(31). However, we only found that EcM seed-
ling survival was significantly inhibited by the
occurrence of AM saplings and adults in the
vicinity, whereas AM seedling survival was
largely enhanced by EcM neighbors (Fig. 2).
Our results, therefore, implicate mycorrhizal
association of species as important drivers of
tree neighborhood interactions at the com-
munity level (18, 31, 32).
For 28 species, sufficient seedlings were
available to evaluate the contributions of var-
ious guilds of soil fungi accumulated through
tree ontogeny to interspecific variation in
CNDD. Seedling response to neighbors varied
widely among species [likelihood ratio test
(LRT) for the random effect of neighborhood ×
species interaction, P < 0.001]. The strength
of the negative impact of conspecific seedling
neighbors on survival was negatively correlated
with the rate at which trees accumulated path-
ogen sequences as they grew, but it was posi-
tively correlated with the rate at which EcM
fungi accumulated (Fig. 3, A and B). The effect
of neighboring conspecific adults, however,
correlated only with EcM sequence accumu-
lation, not pathogen accumulation (Fig. 3, C
and D). These results suggest that exposure to
EcM fungi protects seedlings from pathogen
damage (33, 34) and that mutualists, as well
as pathogens, are important players in driving
Fig. 2. Neighborhood effects (odds ratios ± SE) on seedling survival for all, AM, and EcM interspecific variation in CNDD. Additionally,
tree species. (A to D) Seedling density in a 1-m2 seedling plot and sapling and adult tree basal EcM and saprotrophic fungus accumulation
area in a 10-m radius were included in GLMMs, and each of them were standardized before rates were significantly related to interspecific
entering the model. Odds ratios >1 indicate a positive effect on seedling survival, whereas variation in conspecific positive sapling effects
values <1 indicate a negative effect. Estimated coefficients and additional results for ErM (table S4). Most species experiencing signifi-
species are presented in table S3. cant positive conspecific sapling effects (fig. S6)

126 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

have strong habitat associations, as revealed by

a previous study in this same forest (33), and
soil fungi may play an essential role in nutrient
exploitation during early seedling develop-
ment (16, 17). However, consistent with a recent
study (18), we found that mycorrhizal type ex-
plained little of the variation among species in
the strength of density dependence (table S5).
The presence of EcM fungi in the soil and/or
on the root tips of EcM plants with a fungal
mantle may be considered important for only
the EcM plants in a neighborhood. However,
we found that both AM and EcM tree seed-
lings benefit from the accumulation of EcM
fungi and that the accumulation rates of EcM
fungi in some AM tree species over ontogeny
are even higher than those of EcM species
(Fig. 3 and fig. S6). The accumulation of EcM
fungi in AM tree species through ontogeny
(e.g., Acer cordatum, Eurya muricata, and
Loropetalum chinensis) may arise because of
an overlapping spatial distribution with EcM
species (fig. S7). Although AM species cannot
directly affect the EcM fungal composition in
the soil, they may alter soil conditions [physi-
cal, (bio)chemical, or biological], which may
indirectly influence the EcM fungal commu-
nity. Recent empirical studies provide strong
evidence that an EcM fungal community could
be indirectly affected by non-EM plants (34, 35),
and our results suggest that, if such impacts
occur, they may decrease the impacts of CNDD
on non-EcM tree species. However, more ex-
perimental tests using controlled inoculations
would be required to test this possibility and Fig. 3. Relationships between CNDD and pathogenic and EcM fungus accumulation rates over tree
to further elucidate the general role of changes ontogeny. (A to D) The overall relationships fitted by two-level mixed models suggest that increasing
in the soil fungal composition on plant com- pathogenic fungal density caused lower seedling survival, but increased EcM fungal density favored
munities (36). seedling survival by means of density-dependent effects. AM tree species are in green, EcM tree species
In AM-dominant tropical forests, where are in red, and ErM tree species are in blue (n = 28 tree species). The black solid and dashed lines
many of the concepts of CNDD were originally indicate that the fitted regressions were significant at P < 0.05 or not statistically significant, respectively.
developed, previous work has indicated that Parameter estimates are available in table S4.
common tree species suffered less from CNDD
than rare species, potentially because of a higher
resistance to pathogens (9, 12). Nonetheless, we (38). Combined, these results provide an extra groups simultaneously, despite increasing
found that tree species abundance was not a dimension to the Janzen-Connell hypothesis recognition of the contrasting roles of mutu-
good predictor of neighborhood effects (i.e., that pathogen accumulation rates may play a alistic and pathogenic fungi (14, 41, 42). Our
CNDD) (table S4). This discrepancy may arise key role in driving the strength of CNDD, but community-level analyses, relying on stan-
because AM fungi offer less protection from EcM fungi may overrule them (39). We specu- dardized long-term seedling census datasets
antagonists than EcM fungi do (37). Trees may late that locating communities along the con- and sequenced natural populations of fungal
maintain an extramatrical mycelial network tinuum from pathogen driven to EcM driven functional categories, represent a step toward
from which EcM fungi could decrease patho- could help elucidate the mechanism underly- a better understanding of the feedback of
gen colonization and damage to seedlings of ing the latitudinal gradients in CNDD and its various guilds of soil-inhabiting fungi on the
rare species in EcM-dominant forests. Our relationship with species abundance observed density dependence of trees. Our results high-
findings that heterospecific adult EcM trees in natural forests (38, 40). light the critical role of EcM fungi on con-
enhanced EcM seedling survival and that AM The negative response of seeds and seed- specific neighbor effects, and we suggest that
seedlings responded positively, or less nega- lings to conspecific density is widely observed these fungi might contribute to the dynamics
tively, to adjacent EcM neighbors provide fur- in forest trees (28) and likely plays a major role and assembly of tree communities more than
ther evidence supporting this inference (Fig. 2). in maintaining diversity patterns at local com- has been appreciated. Ultimately, models of tree
Our work is congruent with the results of study munity and landscape scales (14, 23, 38, 40). diversity should incorporate the role of both
on the strength of CNDD at the global scale, However, until now, studies have not examined plant pathogens and mutualists (39, 41, 42).
which shows that the CNDD-species abun- whether species’ differential responses to soil
dance relationship varies systematically among fungi can cause variation among tree species REFERENCES AND NOTES
sites, from positive in the tropics to neutral (or in the magnitude of CNDD and rarely have 1. J. Terborgh, Am. Nat. 179, 303–314 (2012).
negative) in subtropical and temperate forests examined the roles of different functional 2. J. S. Wright, Oecologia 130, 1–14 (2002).

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 127

RES EARCH | R E P O R T S

3. D. H. Janzen, Am. Nat. 104, 501–528 (1970). L.C., X.M., and H.R. conducted fieldwork, and L.C., L.G., SUPPLEMENTARY MATERIALS
4. J. H. Connell, in Dynamics of Populations, P. J. den Boer, and N.J. collected soil samples and carried out the molecular science.sciencemag.org/content/366/6461/124/suppl/DC1
G. R. Gradwell, Eds. (Centre for Agricultural Publishing and analysis. L.C., K.M., and N.G.S. performed statistical analysis Materials and Methods
Documentation, 1971), pp. 298–312. and wrote the draft of the manuscript. All authors contributed to Figs. S1 to S8
5. A. Packer, K. Clay, Nature 404, 278–281 (2000). revisions and gave final approval for publication. Competing Tables S1 to S6
6. M. Konno, S. Iwamoto, K. Seiwa, J. Ecol. 99, 1394–1401 interests: We declare no conflicts of interest. Data and References (44–76)
(2011). materials availability: The DNA sequencing datasets have
7. R. Bagchi et al., Ecol. Lett. 13, 1262–1269 (2010). been deposited in (43).The full census dataset used in this 9 May 2018; resubmitted 27 June 2019
8. M. Liang et al., Ecol. Lett. 19, 1448–1456 (2016). study is available at the CTFS-ForestGEO database portal: Accepted 22 August 2019
9. S. A. Mangan et al., Nature 466, 752–755 (2010). http://ctfs.si.edu/datarequest/. 10.1126/science.aau1361
10. S. McCarthy-Neumann, R. K. Kobe, J. Ecol. 98, 396–407
(2010).
11. D. J. Johnson, R. Condit, S. P. Hubbell, L. S. Comita, Proc. R.
Soc. B 284, 20172210 (2017).
12. L. S. Comita, H. C. Muller-Landau, S. Aguilar, S. P. Hubbell,
LIPID METABOLISM
Science 329, 330–332 (2010).
13. M. Blackwell, Am. J. Bot. 98, 426–438 (2011).
14. R. Bagchi et al., Nature 506, 85–88 (2014).
15. K. L. McGuire, Ecology 88, 567–574 (2007).
A bacterial light response reveals an orphan
16. W. H. van der Putten, M. A. Bradford, E. Pernilla Brinkman,
T. F. J. van de Voorde, G. F. Veen, Funct. Ecol. 30, 1109–1121
desaturase for human plasmalogen synthesis
(2016).
17. T. Miki, M. Ushio, S. Fukui, M. Kondoh, Proc. Natl. Acad. Sci. U.S.A. Aránzazu Gallego-García1*, Antonio J. Monera-Girona1*, Elena Pajares-Martínez1*,
107, 14251–14256 (2010).
Eva Bastida-Martínez1, Ricardo Pérez-Castaño1, Antonio A. Iniesta1, Marta Fontes1,
18. J. A. Bennett et al., Science 355, 181–184 (2017).
19. See the supplementary materials and methods. S. Padmanabhan2†, Montserrat Elías-Arnanz1†
20. L. Chen et al., Ecography 41, 1114–1123 (2018).
21. M. Uriarte et al., Ecol. Lett. 13, 1503–1514 (2010). Plasmalogens are glycerophospholipids with a hallmark sn-1 vinyl ether bond. These lipids
22. H. A. Peters, Ecol. Lett. 6, 757–765 (2003).
23. K. E. Harms, S. J. Wright, O. Calderón, A. Hernández, are found in animals and some bacteria and have proposed membrane organization, signaling,
E. A. Herre, Nature 404, 493–495 (2000). and antioxidant roles. We discovered the plasmanylethanolamine desaturase activity that
24. L. Tedersoo, R. Drenkhan, S. Anslan, C. Morales-Rodriguez, is essential for vinyl ether bond formation in a bacterial enzyme, CarF, which is a homolog
M. Cleary, Mol. Ecol. Resour. 19, 47–76 (2019).
25. L. Tedersoo et al., Science 346, 1256688 (2014). of the human enzyme TMEM189. CarF mediates light-induced carotenogenesis in Myxococcus
26. X. Liu et al., Ecol. Lett. 15, 111–118 (2012). xanthus, and plasmalogens participate in sensing photooxidative stress through singlet
27. G. S. Gilbert, C. O. Webb, Proc. Natl. Acad. Sci. U.S.A. 104, oxygen. TMEM189 and other animal homologs could functionally replace CarF in M. xanthus,
4979–4983 (2007).
and knockout of TMEM189 in a human cell line eliminated plasmalogens. Discovery of the
28. L. S. Comita et al., J. Ecol. 102, 845–856 (2014).
29. C. Wills, R. Condit, R. B. Foster, S. P. Hubbell, Proc. Natl. Acad. human plasmanylethanolamine desaturase will spur further study of plasmalogen biogenesis,
Sci. U.S.A. 94, 1252–1257 (1997). functions, and roles in disease.
30. D. D. Bloom, N. R. Lovejoy, Proc. R. Soc. B 281, 20132081
(2014).

P
31. D. J. Johnson, K. Clay, R. P. Phillips, Oecologia 186, 195–204
(2018). lasmalogens are glycerophospholipids correlates with various human disorders, in-
32. F. P. Teste et al., Science 355, 173–176 (2017). with a hydrocarbon chain linked by a cluding cancer and Alzheimer’s disease (1, 2, 11).
33. K. Tian, L. Chen, X. Mi, K. Ma, J. Chen, China Sci. Bull. vinyl ether bond at the glycerol sn-1 po- Mammalian plasmalogen biosynthesis involves
(Chinese Version) 58, 3561–3569 (2013).
34. C. Gao et al., Mol. Ecol. 22, 3403–3414 (2013).
sition (fig. S1, A and B) found in animals several steps (fig. S2A) (12). A key late step in
35. T. Taniguchi, N. Kanzaki, S. Tamai, N. Yamanaka, K. Futai, and some anaerobic bacteria but not in the endoplasmic reticulum (ER) converts an
New Phytol. 173, 322–334 (2007). plants, fungi, or most aerobic bacteria, except alkyl ether phosphatidylethanolamine (AEPE)
36. K. Kadowaki et al., Commun. Biol. 1, 196 (2018). myxobacteria (1–3). In mammals, plasmalo- or plasmanylethanolamine to the vinyl ether
37. E. Laliberté, H. Lambers, T. I. Burgess, S. J. Wright,
New Phytol. 206, 507–521 (2015).
gens are most abundant in the brain, heart, and (alkenyl) phosphatidylethanolamine (VEPE) or
38. J. A. LaManna et al., Science 356, 1389–1392 leukocytes and occur in nearly all subcellular plasmenylethanolamine (fig. S2A). This oxygen-
(2017). membranes (1–3). Lipid rafts—membrane micro- dependent step was described almost 50 years
39. W. H. van der Putten, Science 355, 134–135 (2017). domain platforms proposed to recruit and ago (13), but the plasmanylethanolamine desa-
40. D. J. Johnson, W. T. Beaulieu, J. D. Bever, K. Clay, Science 336,
904–907 (2012).
activate signaling proteins (4)—are enriched turase involved is unidentified: an orphan en-
41. I. A. Dickie, S. A. Schnitzer, P. B. Reich, S. E. Hobbie, Ecol. Lett. in plasmalogens (5, 6). The plasmalogen vinyl zyme (14, 15). In the obligately aerobic Myxococcus
8, 1191–1200 (2005). ether bond confers physical-chemical attri- xanthus and related myxobacteria, ether lipid
42. N. Christian, E. A. Herre, L. C. Mejia, K. Clay, Proc. R. Soc. B butes that affect membrane fluidity, signaling, synthesis requires the multifunctional enzyme
284, 20170641 (2017).
43. L. Chen, Soil fungal communities Gutianshan (GTS), Version and function (1, 2) and is susceptible to cleav- ElbD and, in an alternative minor pathway,
2, Figshare (2019); https://doi.org/10.6084/m9.figshare. age by reactive oxygen species (ROS) such as MXAN_1676 (fig. S2B) (3, 16, 17). ElbD was
8152922.v2. singlet oxygen (1O2), yielding products that speculated to also direct vinyl ether bond for-
ACKN OW LEDG MEN TS
may act as second messengers (7–10). Anti- mation (16), but we have discovered that this
We acknowledge the hard work of the hundreds of field oxidative and signaling mechanisms have thus desaturase activity corresponds to another pro-
assistants who were involved in the collection of tree census and been linked to plasmalogens, whose deficiency tein, CarF, and that it is functionally conserved
seedling datasets in the Gutianshan forest dynamics plot. We in animals, including humans.
especially thank R. Condit for statistical advice and comments
on the manuscript and C. Gao for suggestions on molecular
In M. xanthus, CarF indirectly signals the
analysis. Funding: The study was financially supported by the
1
Departamento de Genética y Microbiología, Área de light response triggered by 1O2 produced upon
National Key Research and Development Program of China Genética (Unidad Asociada al Instituto de Química Física photoexcitation of protoporphyrin IX (PPIX)
(2017YFA0605103), the Strategic Priority Research Program “Rocasolano,” Consejo Superior de Investigaciones
of the Chinese Academy of Sciences (XDB31030000), the Científicas), Facultad de Biología, Universidad de Murcia,
to inactivate a membrane-associated anti-s
NSFC (31270495), and the Youth Innovation Promotion Murcia 30100, Spain. 2Instituto de Química Física factor, CarR, and release its cognate s factor,
Association of CAS (2013058). Collaboration and manuscript “Rocasolano,” Consejo Superior de Investigaciones CarQ (18–21). This enables transcription of
preparation were facilitated by NSF US-China Dimensions Científicas, 28006 Madrid, Spain.
genes for the biosynthesis of carotenoids,
of Biodiversity grants (DEB-1046113 and DEB-1241136) to *These authors contributed equally to this work.
Stuart Davies of the Smithsonian Institution and N.G.S. †Corresponding author. Email: melias@um.es (M.E.-A.); padhu@ which quench 1O2 and other ROS, and pro-
Author contributions: L.C. and K.M. designed the study. iqfr.csic (S.P.) vokes a yellow-to-red color change (the Car+

128 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

phenotype). In another pathway, light directly verifying that CarF with a C-terminal Strep cated CarF, which is crucial in the response
inactivates a B12-based photoreceptor that re- tag is functional in M. xanthus, we expressed to light, to MxVEPE synthesis, we investigated
presses genes for carotenogenesis in the dark it in Escherichia coli and affinity-purified it how in-frame gene deletions of elbD (DelbD),
(22–24). CarF sequence homologs exist in from the detergent-solubilized membrane MXAN_1676 (D1676), or both affected the light
myxobacteria and few other bacteria (mostly fraction (fig. S5). Purified CarF, which con- response. Single DelbD- and D1676-deletion
Leptospiraceae and Alphaproteobacteria); in tained two equivalents of iron (supplementary mutants acquired in the light a reddish color,
invertebrate and vertebrate animal species, materials, materials and methods), tended to which appeared somewhat less intense for
including humans, in which they are denoted slowly form higher-order oligomers possibly the DelbD mutant than for the D1676 mutant
TMEM189 or Kua and have unknown functions detrimental to activity (fig. S5). Consequently, or the wild-type strain, whereas the double
(19–21, 25); and in plants, with Arabidopsis the DcarF cell extract was incubated with DelbDD1676 mutant was Car–, like the DcarF
thaliana having three (At1, At2, and At3) (fig. freshly purified CarF under aerobic condi- strain (Fig. 2B). Consistent with this, activity
S3 and table S1). From phylogenetic analyses, tions, followed by FAME analysis. Besides in the light of a lacZ gene fused to the pro-
myxobacterial and Leptospira CarF homologs molecular oxygen, the desaturase activity moter of the carotenogenic gene cluster (carB::
appear more related to those in animals, and that generates the plasmalogen vinyl ether lacZ reporter) was, relative to the wild type,
those from Alphaproteobacteria appear more bond requires reduced nicotinamide adenine ~60% less for the DelbD mutant and ~20%
related to ones in plants (fig. S3A). CarF has dinucleotide phosphate (NADPH) and an elec- less for the D1676 mutant but at low basal
12 histidines, all of which are cytoplasmic on tron transport chain involving cytochrome b5 levels (no photoinduction) in the DelbDD1676
the basis of its experimentally determined four- (27), as in aerobic fatty acid desaturation (28). mutant, as in the DcarF strain (Fig. 2B). To
transmembrane helix topology (Fig. 1A), with Hence, we included NADPH and assumed correlate these data with ether lipid and
five shown to be important for function thus that the extract contains the other necessary MxVEPE contents, and to infer the step in
far (19, 20). Animal CarF homologs conserve 11 components except CarF. Under these condi- plasmalogen biosynthesis at which CarF acts,
of these histidines, plants conserve eight, and tions, we detected an i15:0-DMA peak, albeit we performed FAME analysis of each strain.
bacteria conserve eight or more (fig. S3 and small (Fig. 2A), lending further support to Both i15:0-DMA and iso15:0-O-alkylglycerol
table S1). To assess the functional importance the inference that MxVEPE formation re- bis-trimethylsilyl ether (i15:0-OAG-bisTMS,
of each His in vivo, we conditionally expressed, quires CarF. indicative ion m/z = 205, derived from alkyl
using a vanillate-inducible system, single His- Single gene disruptions in M. xanthus im- ether lipids) (fig. S4) were absent in the
to-Ala N-terminal FLAG-tagged CarF variants plicated ElbD and, to a minor level, MXAN_1676 DelbDD1676 strain, but consistent with pre-
in the Car– carF-deleted (DcarF) M. xanthus in ether lipid biosynthesis and ElbD in the vious findings using single disruption mutants
strain. FLAG-tagged CarF is functional because formation of a neutral alkyl (not vinyl) ether (16), both species were present at low levels in
cells that express it were Car+, as were variants lipid that signals fruiting body development the DelbD strain and closer to wild-type levels
H183A and H190A (in which histidine at and sporulation and is synthesized only upon in the D1676 strain (Fig. 2C and fig. S6). By
positions 183 and 190 was replaced by alanine) starvation (16, 17). Because our data impli- contrast, i15:0-OAG, yet not i15:0-DMA, was
and that mutated in the nonconserved His218,
whereas the remaining nine His mutants were
all Car– (Fig. 1B). Nine histidines are therefore Fig. 1. M. xanthus CarF and
important in CarF function. All nine are con- histidine mutational analy-
served in CarF homologs from other myxobac- sis. (A) Experimentally
teria, Leptospirales, and animals, but one of determined membrane
these, His113, differs in plant (where it is often topology of CarF (20)
an Arg) and other bacterial homologs (fig. S3 and location of its
and table S1). The HxxxH and HxxHH pattern 12 histidines. Numbers indi-
of the functionally crucial His164/His168 and cate residues in the CarF
His 191 /His 194 /His 195 , respectively, occurs sequence. Histidines
in membrane-associated diiron fatty acid conserved in all homologs
desaturases and hydroxylases of otherwise low are shown as dots, those
overall sequence similarity to CarF (19–21, 25). conserved only in some
Furthermore, At3 is a chloroplast desaturase homologs are shown as tri-
(FAD4) that generates an unusual trans double angles, and both are colored
bond in the sn‐2 acyl carbon chain (fig. S1C) (26). according to their mutation
We therefore investigated whether M. xanthus to alanine affects (red) or
CarF is a desaturase. not (green) CarF function.
Fatty acid methyl ester (FAME) analysis The blue square is for a
revealed complete absence of a peak in the nonconserved histidine.
DcarF strain (Fig. 2A) that in the wild-type (B) Mutational analysis of
M. xanthus corresponds to iso15:0-dimethylacetal histidines in CarF. Each cell
[i15:0-DMA, indicative ion mass/charge ratio spot corresponds to the
(m/z) = 75] derived from its plasmalogen, M. xanthus DcarF strain
MxVEPE (fig. S4) (16, 17). This i15:0-DMA peak expressing wild-type (WT)
was restored in the DcarF strain that expresses CarF or the indicated His-to-
a vanillate-inducible carF gene (Fig. 2A), indi- Ala mutant version. Cells
cating that CarF is essential for the production that express a nonfunctional CarF variant do not undergo the yellow-to-red color change when exposed to
of MxVEPE. We also examined whether the light (supplementary materials, materials and methods). (Bottom) A Western blot of M. xanthus cell extracts
i15:0-DMA peak reappeared when DcarF cell expressing each CarF variant probed by using anti-FLAG and anti-RNAP b antibodies (as loading control),
extracts were incubated with pure CarF. After with sizes (in kilodaltons) indicated on the left.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 129

RES EARCH | R E P O R T S

Fig. 2. CarF mediates plasmalogen synthesis in M. xanthus. (A) FAME GC-MS chromatogram sections of total lipid extracts from M. xanthus strains. i15:0-DMA
peak level is shown (in percent) relative to fixed amount of 17:0 internal standard. (B) Light-induced colony color assay and light-inducible carB::lacZ reporter
probe assay [mean and standard errors, n = 3 biological replicates; WT levels set to 100%]. (C) Relative abundance of i15:0-DMA and i15:0-OAG-bisTMS (fig. S4) for
indicated M. xanthus strains (mean and standard errors, n = 3 biological replicates; normalized to fixed amount of internal 17:0 standard and WT levels set to 100%).

Fig. 3. Plasmalogens mediate signaling of M. xanthus light-induced strain). (E) Light-induced colony-color assay with the DcarF strain expressing CarF
carotenogenesis, and animal but not plant CarF homologs enable or CarF247 (as controls) and animal (Hs, Mm, Dr1, Dr2, Ce, or Dm) or plant
plasmalogen synthesis and light-induced carotenogenesis in M. xanthus. (At1, At2, or At3) CarF homologs, all N-terminally FLAG-tagged. Western blots of
(A) HsVEPE1 and HsAEPE1 chemical structures. (B and C) Rescue of light-induced the corresponding M. xanthus cell extracts are shown below, with sizes (in
carotenogenesis upon feeding M. xanthus strains with (B) HsVEPE1 or (C) HsAEPE1. kilodaltons) indicated on the left. (F) Light-inducible carB::lacZ reporter assay
(D) Detection of 18:0-OAG-bisTMS and 18:0-DMA in indicated M. xanthus strains, in strains used in (E). (G) Abundance of i15:0-DMA relative to internal 17:0
with 17:0 as internal standard (mean and standard errors, n = 3 biological standard in strains used in (E). Values in (F) and (G) are mean and standard
replicates, relative to 18:0-OAG-bisTMS set to 100% in the DcarFDelbDD1676 errors (n = 3 biological replicates) relative to 100% for the wild type.

observed in the DcarF strain (Fig. 2C). Ex- ogen formation, we asked whether the light both DcarF strains remained Car– (Fig. 3C).
pressing elbD or MXAN_1676 in trans from an response depends directly on plasmalogens Moreover, FAME analysis detected 18:0-DMA
inducible promoter in the Car– DelbDD1676 by supplying them exogenously to the above and 18:0-OAG (neither natural to M. xanthus)
strain caused reappearance of ether lipids and mutant strains. We used three commercially in the HsAEPE1-fed DelbDD1676 mutant but
plasmalogens and of reddish colony color in available plasmalogens typically found in human only 18:0-OAG in the DcarFDelbDD1676 strain
the light (fig. S7), ruling out polar effects from cells, with an 18(Plasm) chain at the sn-1 position (Fig. 3D). Accordingly, LC-MS/MS detected
the gene deletions. Consistent with the gas and an 18:1 (HsVEPE1), a 20:4 w-6 (HsVEPE2), HsAEPE1 and HsVEPE1 in the double mutant
chromatography–mass spectrometry (GC-MS) or a 22:6 w-3 (HsVEPE3) chain at the sn-2 ester but only HsAEPE1 in the triple mutant (fig.
results, direct assessment by means of liquid position (Fig. 3A and fig. S9). These human S10). This result confirms HsAEPE1 uptake
chromatography–MS (LC-MS)/MS revealed loss plasmalogens therefore differ from M. xanthus and its conversion, only if CarF is available,
of MxVEPE and MxAEPE in the DelbDD1676 MxVEPE, which has i15 chains at both posi- to HsVEPE1. Thus, CarF can produce plasmal-
strains but only of MxVEPE in the DcarF mu- tions (fig. S1B) (16, 17). Yet, all three rescued ogen from the corresponding human alkyl
tant (fig. S8). Thus, whereas ElbD or MXAN_1676 the Car+ phenotype when fed to the DcarF ether lipid, despite the latter having sn-1 and
enable biosynthesis of alkyl ether lipids, their and DelbDD1676 strains, as well as to the triple sn-2 moieties that differ from those in the likely
conversion to the vinyl ether form requires DcarFDelbDD1676 mutant strain (Fig. 3B and natural precursor in M. xanthus (fig. S4).
CarF. In other words, CarF appears to be the fig. S9). When fed with HsAEPE1, which is Various human plasmalogens, distinct from
ether lipid desaturase in plasmalogen synthesis. HsVEPE1 but with an sn-1 alkyl instead of vinyl the endogenous one, can mediate signaling in
Because CarF is implicated in both M. xanthus ether bond (Fig. 3A), the CarF-containing M. xanthus light-induced carotenogenesis, and
light-induced carotenogenesis and plasmal- DelbDD1676 strain became Car + , whereas CarF can convert a human alkyl ether lipid to

130 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

Fig. 4. Analysis of human TMEM189. (A) Mutational analysis

of histidines in human TMEM189. Each spot corresponds to
a DcarF M. xanthus strain expressing, from a vanillate-inducible
promoter, TMEM189 (WT), or the indicated His-to-Ala mutant
version, all N-terminally FLAG-tagged. Mutation of the
nonconserved His222 is also shown. Western blots of
the corresponding M. xanthus cell extracts are shown
below, with sizes (in kilodaltons) indicated on the left.
(B) Levels in the parental and TMEM189-KO human
HAP1 cells of ether lipids and plasmalogens measured
as OAG-bisTMS and DMA, respectively, by means of
FAME GC-MS analysis (mean and standard errors relative
to total, n = 3 biological replicates). (C) Localization of
human TMEM189 with an N-terminal fusion to enhanced
green fluorescent protein in transfected human HeLa cells
monitored with confocal microscopy. Scale bar, 10 mm.
ER-Tracker and Hoechst 33258 were used to image the ER
and nuclear DNA, respectively.

its vinyl ether form. We therefore investigated ether lipid precursor, despite it being differ- in the 1O2 response triggered by photoexcita-
whether human and other eukaryotic homologs, ent from their natural substrates, to generate tion of PPIX mentioned earlier (21). 1O2 is
as well as two bacterial ones (neither from myxo- MxVEPE. known to provoke cleavage of the plasmalogen
bacteria), can functionally replace CarF in the Given the above results, we further examined at its vinyl ether bond, yielding as breakdown
M. xanthus light response and plasmalogen human TMEM189. Eleven of the 12 histidines products 2-monoacylglycerophosphoethanol-
biosynthesis. Plasmids for vanillate-inducible in CarF are conserved in TMEM189 (fig. S3B), amine (lyso-PE) and a fatty aldehyde (fig. S15)
expression in M. xanthus of codon-optimized and single mutations of nine of these to Ala im- (7–10). Plasmalogen cleavage by 1O2 to lyso-PE
N-terminal FLAG-tagged homologs from paired function in M. xanthus (Fig. 4A), just as likely perturbs local membrane structure and
human, mouse, fly, worm, zebrafish (two), with the equivalent mutations in CarF (Fig. 1B), properties, and adduct formation with the
A. thaliana (three), the animal pathogen thus providing further evidence that the two fatty aldehyde can render membrane proteins
Leptospira interrogans, and the alphapro- proteins are closely related. Hence, we tested inactive (1, 2, 7). Either or both conceivably
teobacterial plant symbiont Bradyrhizobium whether TMEM189 is the plasmanylethanol- underlie the light-1O2-plasmalogen signaling
diazoefficiens were constructed, and each was amine desaturase in human plasmalogen syn- mechanism that leads to CarR inactivation,
introduced into the DcarF strain. Among cells thesis by using cell lines. We obtained a HAP1 CarQ release, and induction of carotenogene-
transformed with the eukaryotic homologs, human cell line with a CRISPR/Cas9 TMEM189 sis in M. xanthus (fig. S16), although a less
those expressing animal homologs (41 to 46% knockout (KO) to compare its ether lipid likely mechanism without plasmalogen break-
identity, 56 to 59% similarity to CarF) (table content with that of the parental cells by means age cannot be excluded.
S1) were all Car+, like those expressing FLAG- of FAME analysis. Plasmalogens were detected
tagged or untagged CarF or its version (CarF247) as 16:0-, 18:0-, and 18:1-DMA peaks in the pa- REFERENCES AND NOTES
lacking the C-terminal 34-residue segment ab- rental cells but were absent in their TMEM189- 1. N. E. Braverman, A. B. Moser, Biochim. Biophys. Acta 1822,
sent in many homologs (Fig. 3E and figs. S3B KO derivative, whereas their alkyl ether lipid 1442–1452 (2012).
and S11A). Cells that express the A. thaliana precursors were detected as the corresponding 2. J. M. Dean, I. J. Lodhi, Protein Cell 9, 196–206 (2018).
3. H. Goldfine, FEBS Lett. 591, 2714–2719 (2017).
homologs (31 to 33% identity to CarF and OAG-bisTMS derivatives in both cell lines (Fig. 4. E. Sezgin, I. Levental, S. Mayor, C. Eggeling, Nat. Rev. Mol. Cell
which do not conserve the functionally crucial 4B and fig. S13). Moreover, plasmalogens re- Biol. 18, 361–374 (2017).
H113 of CarF) (fig. S3B and table S1) remained appeared in the TMEM189-KO cells upon tran- 5. L. J. Pike, X. Han, K. N. Chung, R. W. Gross, Biochemistry 41,
2075–2088 (2002).
Car–, even though these proteins were present sient expression of EGFP-TMEM189 (fig. S14), 6. C. Rodemer et al., Hum. Mol. Genet. 12, 1881–1895 (2003).
at levels comparable with or greater than and this functional EGFP-fused protein local- 7. C. M. Jenkins et al., J. Biol. Chem. 293, 8693–8709 (2018).
those of CarF247 or the fly homolog (Fig. 3E). ized in HeLa cells to the ER (Fig. 4C), the 8. O. H. Morand, R. A. Zoeller, C. R. Raetz, J. Biol. Chem. 263,
11597–11606 (1988).
Consistently, light-induced carB::lacZ reporter proposed site for vinyl ether bond generation 9. R. A. Zoeller, O. H. Morand, C. R. Raetz, J. Biol. Chem. 263,
activity was greater than basal levels in DcarF in plasmalogen biosynthesis (fig. S2A). 11590–11596 (1988).
strains that express animal, but not A. thaliana, Our study reveals TMEM189 as the long- 10. S. Stadelmann-Ingrand, S. Favreliere, B. Fauconneau, G. Mauco,
C. Tallineau, Free Radic. Biol. Med. 31, 1263–1271 (2001).
homologs (Fig. 3F); and in FAME analysis, the sought desaturase for animal plasmalogen 11. M. C. F. Messias, G. C. Mecatti, D. G. Priolli,
i15:0-DMA peak reappeared only in cells that biosynthesis and that its plant homologs lack P. de Oliveira Carvalho, Lipids Health Dis. 17, 41 (2018).
express the animal homologs (Fig. 3G and fig. this activity. Identification of the animal plas- 12. N. Nagan, R. A. Zoeller, Prog. Lipid Res. 40, 199–229 (2001).
13. R. L. Wykle, M. L. Blank, B. Malone, F. Snyder, J. Biol. Chem.
S11, B and C). Of the two bacterial homologs manylethanolamine desaturase allows direct
247, 5442–5447 (1972).
tested, only the one more related to animal studies of the nexus between plasmalogens, 14. T. Harayama, H. Riezman, Nat. Rev. Mol. Cell Biol. 19, 281–296
homologs, that from L. interrogans (43% iden- their cellular functions, and diverse pathol- (2018).
tity to CarF), could functionally replace CarF in ogies, which have thus far relied on targeting 15. K. Watschinger, E. R. Werner, Biochimie 95, 59–65 (2013).
16. W. Lorenzen, T. Ahrendt, K. A. Bozhüyük, H. B. Bode,
M. xanthus (fig. S12). Thus, animal TMEM189 genes involved earlier in the biosynthetic path- Nat. Chem. Biol. 10, 425–427 (2014).
proteins can use the endogenous M. xanthus way. We also establish plasmalogens as crucial 17. M. W. Ring et al., J. Biol. Chem. 281, 36691–36700 (2006).

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 131

RES EARCH | R E P O R T S

18. M. Elías-Arnanz, S. Padmanabhan, F. J. Murillo, Curr. ACKN OWLED GMEN TS authors. A.J.M.-G., E.P.-M., A.G.-G., E.B.-M., R.P.-C, M.F., and A.A.I.
Opin. Microbiol. 14, 128–135 (2011). We thank C. L. Drennan (MIT) for critical comments on the performed the experiments. All authors contributed to data analysis.
19. M. Fontes, L. Galbis-Martínez, F. J. Murillo, Mol. Microbiol. 47, manuscript; D. González-Silvera and P. L. Valero for advice on lipid A.G.-G., M.F., S.P., and M.E.-A. supervised research. S.P. and M.E.-A.
561–571 (2003). analysis; J. A. Madrid and V. López-Egea for technical assistance; acquired funding, wrote the original draft, and reviewed and edited
20. L. Galbis-Martínez, M. Galbis-Martínez, F. J. Murillo, M. Fontes, personnel at the instrumentation, sequencing, tissue culture, it with input from all authors. Competing interests: The authors
Microbiology 154, 895–904 (2008). and confocal microscopy facilities (all at the University of Murcia); declare no competing interests. Data and materials availability: All
21. M. Galbis-Martínez, S. Padmanabhan, F. J. Murillo, and J. Abellón-Ruiz (Newcastle University) for the exchange on data are available in the main text or the supplementary materials,
M. Elías-Arnanz, J. Bacteriol. 194, 1427–1436 (2012). membrane protein purification. Funding: This work was supported and strains and plasmids are available upon request.
22. M. Jost et al., Nature 526, 536–541 (2015). by grants BFU2015-67968-C2-1P and PGC2018-094635-B-C21
23. J. M. Ortiz-Guerrero, M. C. Polanco, F. J. Murillo, (to M.E.-A.) and BFU2015-67968-C2-2P and PGC2018-094635-B- SUPPLEMENTARY MATERIALS
S. Padmanabhan, M. Elías-Arnanz, Proc. Natl. Acad. Sci. U.S.A. C22 (to S.P.) from the Agencia Estatal de Investigación (AEI)–
108, 7565–7570 (2011). science.sciencemag.org/content/366/6461/128/suppl/DC1
Spain and the European Regional Development Fund (FEDER), by
24. S. Padmanabhan, M. Jost, C. L. Drennan, M. Elías-Arnanz, Materials and Methods
grants 19429/PI/14 and 20992/PI/18 (to M.E.-A.) from Fundación
Annu. Rev. Biochem. 86, 485–514 (2017). Figs. S1 to S16
Séneca (Murcia)-Spain, and Ph.D. fellowship contracts from the
25. T. M. Thomson et al., Genome Res. 10, 1743–1756 (2000). Tables S1 to S3
Ministerio de Educación y Cultura-Spain (to E.P.-M., A.J.M.-G., and
26. J. Gao et al., Plant J. 60, 832–839 (2009). References (29–40)
E.B.-M.) and Ministerio de Economía y Competitividad-Spain
27. M. L. Blank, F. Snyder, Methods Enzymol. 209, 390–396 (1992). (to R.P.-C.). Author contributions: M.E.-A. conceived the study. 29 May 2019; accepted 4 September 2019
28. H. Goldfine, Prog. Lipid Res. 49, 493–498 (2010). M.E.-A., S.P., and A.G.-G. designed experiments, with input from other 10.1126/science.aay1436

QUANTUM PHYSICS zation. However, these models violate a basic

tenet of relativity by allowing signaling faster
Satellite testing of a gravitationally induced quantum than the speed of light. This happens because
the characteristic quantum feature of entan-
decoherence model glement allows acausal effects within the com-
pact region containing the CTCs to spread into
Ping Xu1,2*, Yiqiu Ma3*, Ji-Gang Ren1,2*, Hai-Lin Yong1,2, Timothy C. Ralph4, Sheng-Kai Liao1,2, the surrounding spacetime (21, 24).
Juan Yin1,2, Wei-Yue Liu1,2, Wen-Qi Cai1,2, Xuan Han1,2, Hui-Nan Wu1,2, Wei-Yang Wang1,2, Feng-Zhi Li1,2, An alternative approach that avoids this
Meng Yang1,2, Feng-Li Lin5, Li Li1,2, Nai-Le Liu1,2, Yu-Ao Chen1,2, Chao-Yang Lu1,2, Yanbei Chen3, problem argues that inconsistent evolutions of
Jingyun Fan1,2†, Cheng-Zhi Peng1,2†, Jian-Wei Pan1,2† quantum states in spacetimes with CTCs could
be avoided for arbitrary initial conditions by
Quantum mechanics and the general theory of relativity are two pillars of modern physics. However, a taking the density operator describing the local
coherent unified framework of the two theories remains an open problem. Attempts to quantize general quantum state as the fundamental object that
relativity have led to many rival models of quantum gravity, which, however, generally lack experimental is required to be self-consistent within the CTC
foundations. We report a quantum optical experimental test of event formalism of quantum fields, a theory region (17). This has the effect that any entan-
that attempts to present a coherent description of quantum fields in exotic spacetimes containing closed glement between systems traversing the CTC
timelike curves and ordinary spacetime. We experimentally test a prediction of the theory with the and systems that do not is erased, thus pre-
quantum satellite Micius that a pair of time-energy–entangled particles probabilistically decorrelate venting acausal effects from spreading outside.
passing through different regions of the gravitational potential of Earth. Our measurement results are Such a feature could arise in quantum field
consistent with the standard quantum theory and hence do not support the prediction of event formalism. theory, where it was suggested that this be-
havior could be understood as a differential

Q
decay of the commutator between field modes
uantum mechanics and the general is particularly interesting because it presents propagating in different metrics. A speculative
theory of relativity describe our world a coherent description of quantum fields in new theory was established on the basis of this
from completely separate perspectives. exotic spacetimes containing closed timelike observation: event formalism (11). As a non-
Although highly expected, a coherent curves (CTCs) and ordinary spacetime. The linear extension of ordinary quantum theory,
interface between quantum theory and exotic spacetimes are interesting both because event formalism attempts to present a coher-
the theory of general relativity remains elusive they are a fundamental feature of general rela- ent description of quantum fields for both
(1–3). In response to this situation, new ideas tivity and because such spacetime structures ordinary spacetime and nonhyperbolic space-
have been conceived, particularly from the could be formed owing to quantum fluctua- times containing CTCs (11, 12, 25). Event for-
quantum optical approach, to test fundamen- tion of spacetime itself—that is, originating at malism makes predictions that can be tested
tals of the interplay between quantum theory a deeper level from quantum gravity (14). in the gravitational wells of planetary objects.
and the gravity theory (4–13). Among these The discussion of quantum mechanics in The mode operator in event formalism is
approaches, event formalism of quantum fields exotic spacetimes with CTCs that violate cau- given as (11)
sality has attracted attention (15–19). Although
^ ðt; xÞ ¼ ∫dkgðkÞeikðx ∫dW JðWÞeiWðt
many proposals for solving this problem have tÞ
1
Hefei, National Laboratory for Physical Sciences at Microscale a tÞ d
^ k;W
a
and Department of Modern Physics, University of Science and been discussed, the approach that follows most
Technology of China, Hefei 230026, P. R. China. 2Shanghai directly from standard quantum field theory ð1Þ
Branch, CAS Center for Excellence and Synergetic Innovation
Center in Quantum Information and Quantum Physics,
is the path-integral approach, where coherent,
University of Science and Technology of China, Shanghai action-weighted sums over all single-valued where g(k) is the photon spectral distribu-
201315, P. R. China. 3Theoretical Astrophysics, California (i.e., self-consistent) histories are evaluated
Institute of Technology, Pasadena, CA 91125, USA. 4Centre for tion; for simplicity, we work in (1 + 1) spacetime
Quantum Computation and Communication Technology, School
(20, 21). For exotic spacetimes, a nonstandard coordinates. An extra degree-of-freedom, W,
of Mathematics and Physics, The University of Queensland, renormalization is typically required to pre- is introduced with a distribution of J(W). t is
St. Lucia, Queensland 4072, Australia. 5Department of Physics, serve the probabilistic interpretation of the
National Taiwan Normal University, Taipei 116, China. given as
theory. Equivalently, this approach corresponds
*These authors contributed equally to this work.
to teleportation models of CTCs (22, 23), where td
t ¼ ∫t ds
†Corresponding author. E-mail: fanjy@ustc.edu.cn (J.F.);
pcz@ustc.edu.cn (C.-Z.P.); pan@ustc.edu.cn (J.-W.P.). postselection plays the role of the renormali- ð2Þ

132 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

Fig. 1. Schematics of experimental test of event

formalism in Earth’s gravitational field. (A) The
ground station sends both entangled single photons
and faint coherent laser pulses to the satellite
Micius. (B) Preparation of entangled photon pairs and
faint coherent laser pulses at the ground station.
Note that all beam splitters (BSs) have a reflectivity
of ~2%. We use the reflected laser pulses at
780 nm to prepare faint coherent laser pulses,
upconvert the rest in an LiB3O5 (LBO) crystal to
390 nm, and then down-convert the 390-nm photons
in a 1-mm BiB3O6 (BIBO) crystal into polarization-
entangled photon pairs (780 nm) (27). The pair
of photons exit separately into paths 1 and 2. The
faint coherent laser pulses and entangled single-
photon pulses in path 1 are combined, coupled
into a single-mode optical fiber, and directed
to the transmitter to be launched to the satellite.
(C) Single photons received by the satellite are
passed through polarization analysis and detected by
single-photon detectors (SPDs). SPDs are labeled by 1G, 2GR, 2GT, 3G, 1ST, and 1SR. Refer to (27) for high-bandwidth, high-precision stabilization of the ground
station-satellite optical link and synchronization. HWP, half-wave plate; QWP, quarter-wave plate; IF, interference (band-pass) filter; PBS, polarizing beam splitter; NDF,
neutral density filter; POL, polarizer.

where ds is the propagation time across an Fig. 2. Satellite

incremental local reference frame, td is the passes. The
photon detection time in asymptotically flat altitude angle
global coordinates, and t can be interpreted as (q) of satellite
the photon emission time. with respect
It can be shown that in event formalism, the to its flying
mode operators at different spacetime locations time for
along the photon trajectory commute to be five different
compatible with exotic spacetime-containing satellite passes
CTCs. However, they follow the same commu- labeled by dates
tation relation as standard quantum theory in (for data in
flat spacetime (11, 25). This new formalism of Fig. 4A), where
quantum fields predicts that a pair of time- we arbitrarily
energy–entangled photons may decohere if they set the time for
pass through different regions of a curved space- the altitude angle of the satellite to be q = 5° as the origin of the (horizontal) time axis. We choose
time. Suppose that a pair of time-energy– to collect experimental data for q ∈ [40°,60°] (gray-shaded area), in which the photon loss is
entangled photons are generated at a ground relatively small, and we have more satellite passes for better statistics. Note that the satellite passes
station. One photon of the pair is detected at on 10/21/2017 and 11/21/2017 (Beijing time, UTC/GMT +8) happen to be on the opposite sides (west
the ground station and its entangled twin is sent side and east side) and are symmetric with respect to the ground station; the altitude angles of the
to and detected at a satellite orbiting around two passes basically coincide.
Earth. Event formalism predicts that in this
setting, the initially time-energy–entangled
pair of photons probabilistically decorrelate in that in standard quantum theory, which is giv- differ by an amount that is significant with
time, which is different from the predictions of en by D≈expð 0:5D2t =dt2 Þ (11, 13), where dt is respect to the photon pulse length.
standard quantum theory (12, 13). At the same the photon coherence time and Dt comes from Micius was launched recently to a sun-
time, the properties of single photons remain the effect of curved spacetime as derived in synchronous orbit with an altitude of ~500 km.
unchanged. For the pairs of photons remain- (13, 26). Event formalism also predicts that in the The end-to-end loss of the free space optical
ing time-energy entangled, their correlation same setting, the correlation properties of faint link, from light launched by the transmitter
properties are given by the standard quantum coherent laser pulses is completely described by antenna at the ground station to light detection
theory; for the pairs of photons decorrelated the standard quantum theory. We present an at the satellite, was determined to be <50 dB on
in time, the observed two-photon coincidence experimental study of these predictions that is average at the wavelengths of interest, which
events spread out. Observationally, the deco- based on the quantum satellite Micius (27). A was sufficient for the demonstration of quan-
herence effect predicted by the event formal- feasibility study of such experiments was re- tum teleportation between ground station and
ism will be the sum of these two effects. The cently presented (13). The basic requirement for satellite, with fidelity surpassing the classical
probability of losing the time-energy entan- the predicted decorrelation effect is that the limit (27). This established quantum optical
glement, P, is characterized by the decorre- difference between the locally measured prop- platform offers an excellent opportunity to ex-
lation factor, D, with D = 1 – P. This can be agation time along the photon trajectory in perimentally examine theoretical proposals
quantified by the ratio between the two-photon the gravitational well, t, and the propagation on the interplay between quantum theory and
coincidence probability in event formalism and time as measured by a clock in flat space, td, gravity, such as event formalism.

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 133

RES EARCH | R E P O R T S

to 60° (data collection range in the experiment,

Fig. 2); the numbers of two-photon coincidence
events and background events that are func-
tions of q are pass dependent, i.e., they may
vary significantly as the satellite takes differ-
ent passes. However, the decorrelation factor,
D(q), is only a function of the altitude angle, q,
i.e., pass independent, in our experimental
configuration. To reduce the statistical error in
our data analysis, in the experiment associated
with one satellite pass, we group the two-
photon coincidence events for every 5° (Dq = 5°)
of altitude angle as one data point, Cexp,EPR(q)
Fig. 3. Spacetime diagram for the detection of a pair of entangled photons in the experiment. A pair and Cexp,COH(q), with 40° < q < 60°. For each
of entangled photons are created at the ground station (event S). Whereas one photon is detected at of these data points, we estimate the num-
the ground station (event M2), its twin is sent to the satellite, where it is detected (event M1). (A) Counting ber of two-photon coincidence events on
all latency, the wavefront from finishing event M2 arrives earlier than the earliest time of event M1 by the basis of standard quantum theory with
dt ~ 18 ns. The two events are not space-like separated (26). (B) With the insertion of s 1-km fiber in path 2, CSQT,EPR(q) = h2SEPR(q), where SEPR(q) is the
the wavefront from the earliest time of finishing event M2 arrives later than the latest time of finishing number of the single-photon detection events
event M1 by dt ~ 5 ms. The two events are space-like separated. The line-of-sight distance between satellite that are due to the entangled photon pair
and ground station varies between 550 and 740 km. (Figure is not drawn to scale.) source at the satellite and h2 = 34.4 (2)% is the
efficiency of detecting entangled photons in
path 2 (Fig. 1B). The two-photon coincidence
events that are due to faint coherent laser
pulses are given by CSQT,COH(q) = SCOH(q)S3tp/
tDq, where SCOH(q) and S3 are the numbers of
single-photon detection events that are due to
faint coherent laser pulses at the satellite and
in path 3, respectively, tp = 12.5 ns is the pulse-
to-pulse separation (26), and tDq is the time to
collect data for Dq = 5°. We then obtain the
decorrelation factor in the ith satellite pass as
DEPR(q,i) = Cexp,EPR(q,i)/CSQT,EPR(q,i) for the
entangled photon pair source and as DCOH(q,i) =
Cexp,COH(q,i)/CSQT,COH(q,i) for the faint coher-
ent laser source, respectively. For a total number
of N satellite passes, we obtain X
the decor-
relation factors as
X
D EPR ðqÞ ¼ i
DEPR ðq; iÞ=N
and DCOH ðqÞ ¼ i DCOH ðq; iÞ=N, respectively.
Note that we attenuate the coherent laser
pulses to make the two-photon coincidence
rates similar to that for the entangled photon
pair source.
We conducted the experiment with and
without fulfilling the no-signaling condition to
account for certain quantum collapse models
Fig. 4. Experimental estimation of decorrelation factors. Decorrelation factors are plotted as a function (28, 29). The no-signaling condition is realized
of the satellite’s altitude angle (q) for without and with fulfilling the nonsignaling condition in (A) averaged by adding a 1-km optical fiber in path 2 at the
over five satellite passes and (B) averaged over four satellite passes, respectively. Open circles: experimental ground station such that the detection event of
results with entangled photon pair source; filled circles: experimental results with coherent laser source; a single photon in path 2 is separated space-
smooth line: predictions of event formalism. Error bars are 1 SD estimated over the relevant numbers like from the detection event of its entangled
of satellite passes. Inset: magnified view of the predictions of event formalism. twin at the satellite (Fig. 3) (26). The measure-
ment results for both spacetime settings are
plotted in Fig. 4, A and B (26), respectively.
The schematics of the experiment are de- tum satellite. We denote the measured two- Given our experimental condition with dt ~
picted in Fig. 1. The system was originally devel- photon coincidence event as Cexp,EPR(q). For 0.07 mm (≈0.2 ps) (26) and satellite altitude
oped to teleport single-photon quantum states comparison, we also recorded two-photon co- of ~500 km, event formalism predicts decor-
from the ground station at Ngari, Tibet, China incidence events Cexp,COH(q) with pulsed coher- relation effects, D(q) < 10–6 for 40° < q < 60°
(latitude: 32°19′33.07′′ N; longitude: 80°1′34.18′′ ent laser light source in the same setting. Here, (smooth lines). In contrast to these predic-
E; and altitude: 5047 m) to Micius (27). In this q is the altitude angle of satellite with respect tions, we observe that the experimentally mea-
work, we generated entangled photon pairs to the ground station. Micius orbits Earth by sured decorrelation factors, DEPR(q) (Fig. 4,
at the ground station and examined the time a different pass in each night. For different filled circles) and DCOH(q) (Fig. 4, open circles),
correlation between one photon at the ground passes, it takes a different amount of time for are flat in the same range of q and scatter
station and its twin distributed to the quan- the altitude angle of satellite varying from 40° around 1 (standing for no decorrelation) for all

134 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

 RES EARCH | R E P O R T S

conditions in the study, i.e., including entan- resolution to the order of the photon coherence 23. S. Lloyd et al., Phys. Rev. Lett. 106, 040403 (2011).
gled photon pair source and faint coherent time. This will not only present a conclusive 24. T. C. Ralph, T. G. Downes, Contemp. Phys. 53, 1–16
(2012).
laser source and with or without satisfying the experimental verification of these decoherence 25. J. L. Pienaar, C. R. Myers, T. C. Ralph, Phys. Rev. A 84, 062316
no-signaling condition. We thus conclude that effects, but will also provide more insightful (2011).
our experimental results are consistent with knowledge about a number of interesting 26. For more details, see supplementary materials.
the descriptions of standard quantum theory gravity-related models, including event for- 27. J.-G. Ren et al., Nature 549, 70–73 (2017).
28. A. Kent, Phys. Rev. A 72, 012107 (2005).
and do not support the predictions of event malism, gravitational dilation, and broaden- 29. A. Kent, Phys. Rev. A 72, 012108 (2005).
formalism. ing (30, 31). 30. D. E. Bruschi, T. C. Ralph, I. Fuentes, T. Jennewein,
It is a generic feature of many speculative M. Razavi, Phys. Rev. D Part. Fields Gravit. Cosmol. 90,
RE FERENCES AND NOTES 045041 (2014).
theories attempting to modify quantum mech- 31. D. Rideout et al., Class. Quantum Gravity 29, 224011
1. S. Carlip, Rep. Prog. Phys. 64, 885–942 (2001).
anics to be more consistent with general rela- 2. C. Kiefer, Ann. Phys. 15, 129–148 (2005). (2012).
tivity in that decoherence of entanglement 3. C. Rovelli, Quantum Gravity (Cambridge Univ. Press, 2004).
4. W. Marshall, C. Simon, R. Penrose, D. Bouwmeester, AC KNOWLED GME NTS
appears under conditions where standard
Phys. Rev. Lett. 91, 130401 (2003). We thank colleagues at the National Space Science Center,
quantum theory does not predict decoherence. 5. I. Pikovski, M. R. Vanner, M. Aspelmeyer, M. S. Kim, C. Brukner, Shanghai Institute of Technical Physics, China Xi’an Satellite
Here, we have tested the specific predictions Nat. Phys. 8, 393–397 (2012). Control Center, National Astronomical Observatories, and Ngari
of event formalism and find they are not 6. H. Yang, H. Miao, D. S. Lee, B. Helou, Y. Chen, Phys. Rev. Lett. Observatory for their management and coordination. Funding: This
110, 170401 (2013). work was supported by the Strategic Priority Research Program
supported; however, this does not necessarily
7. A. Chou et al., Class. Quantum Grav. 34, 065005 (2017). on Space Science of the Chinese Academy of Sciences and
rule out other approaches. Indeed, even within 8. A. Belenchia et al., Phys. Rev. D 98, 126009 (2018). the National Natural Science Foundation of China. Author
event formalism, there is some ambiguity as 9. D. Carney, P. C. E. Stamp, J. M. Taylor, Class. Quantum Grav. contributions: J.F., C.-Z.P., and J.-W.P. conceived the research.
to the scale of the effect. If, instead of taking 36, 034001 (2019). P.X., J.-G.R., J.F., C.-Z.P., and J.-W.P. designed the experiment.
10. H. Miao, D. Marytnov, H. Yang, Quantum correlation of light P.X., J.-G.R., H.-L.Y., X.H., H.-N.W., W.-Y.W., S.-K.L., J.Y., W.-Q.C.,
the global time, td, as defined by a clock in mediated by gravity. arXiv:1901.05827 [quant-ph] L.L., N.-L.L., Y.-A.C., and C.-Y.L. conducted the experiment.
asymptotic flat space (as assumed here and in (17 January 2019). H.-L.Y., W.-Y.L., W.-Q.C., F.-Z.L., and M.Y. developed the software.
the original proposal), one uses a clock local 11. T. C. Ralph, G. J. Milburn, T. Downes, Phys. Rev. A 79, 022121 Y.M., T.R., F.L., Y.C., and J.F. performed the theoretical study.
(2009). P.X., Y.M., T.C.R., Y.C., J.F., C.-Z.P., and J.-W.P. analyzed the data
to the detector as the global reference, then a 12. T. C. Ralph, J. Pienaar, New J. Phys. 16, 085008 (2014). and prepared the manuscript. All authors contributed to data
weaker decoherence effect is predicted, with 13. S. K. Joshi et al., New J. Phys. 20, 063016 (2018). collection, discussed the results, and reviewed the manuscript.
D(q) between 0.96 and 0.98 for the current 14. J. Friedman et al., Phys. Rev. D Part. Fields 42, 1915–1930 J.-W.P. supervised the whole project. Competing interests: The
(1990). authors declare no competing interests. Data and materials
experimental configuration. An experimental availability: All data are available in the manuscript or the
15. M. S. Morris, K. S. Thorne, Am. J. Phys. 56, 395–412 (1988).
study of this weaker decoherence effect with 16. H. D. Politzer, Phys. Rev. D Part. Fields 46, 4470–4476 supplementary materials. Requests for materials should be
sufficient statistical confidence requires a sig- (1992). addressed to J.F., C.-Z.P., or J.-W.P.
nificant number of satellite passes. This is far 17. D. Deutsch, Phys. Rev. D Part. Fields 44, 3197–3217 (1991).
SUPPLEMENTARY MATERIALS
beyond the lifetime of Micius and thus we 18. J. B. Hartle, Phys. Rev. D Part. Fields 49, 6543–6555 (1994).
19. S. W. Hawking, Phys. Rev. D Part. Fields 52, 5681–5686 science.sciencemag.org/content/366/6461/132/suppl/DC1
plan to perform such a test in a future ex- (1995). Materials and Methods
periment (26). Moreover, considering that the 20. F. Echeverria, G. Klinkhammer, K. S. Thorne, Phys. Rev. D Part. Fields Figs. S1 to S5
44, 1077–1099 (1991). Tables S1 to S3
two-photon coincidence time window (3 ns) is References (32–37)
21. H. D. Politzer, Phys. Rev. D Part. Fields 49, 3981–3989
a few thousand times bigger than the photon
(1994). 1 July 2019; accepted 6 September 2019
coherence time (0.2 ps) in this experiment, 22. D. T. Pegg, Quantum Mechanics and the Time Travel Paradox Published online 19 September 2019
future experiments may improve the temporal (Consiglio Nazionale delle Richerche, 2001). 10.1126/science.aay5820

SCIENCE sciencemag.org 4 OCTOBER 2019 • VOL 366 ISSUE 6461 135

 LIFE SCIENCE TECHNOLOGIES Produced by the Science/AAAS Custom Publishing Office

new pro d uc ts
Colloidal Gold
Automated Liquid Handler -0"

 
  

Hamilton Company introduces the compact  
  


 


-
Microlab Prep automated liquid handler as an 
"








$5


ideal entry-level solution for those transition- 
"



ing from manual pipetting in 96- and 384-well nanoparticles without compromising on quality, and pass on the savings
microplates and other sample vessels. In addi- to you. We can supply any of our colloidal gold in quantities to suit your
           
$2   
of two independent pipetting channels, the
  
# # "
handler features a high-speed, multiprobe $/

&% "'% ")% "*% 
 -
  

       - 

$2
 

ing both pipetting technology types for the  





 

 .6!,&%

 

 # 

      

 (%%1!$

"

enhance laboratory productivity and eliminate repetitive use injuries,
42+%%&

3."
 ## 
&

' 
  


  


 - consistency and reliability.

   

% $ 
 
   - Expedeon
ed systems, it uses air-displacement pipetting technology, eliminating )  '&""$ !$#$
  
          




frequency as compared to liquid-displacement pipettors. High pipet-
ting accuracy, precision, and consistency are ensured throughout the Automation-Compatible Culture Media Plate
    "%! %  
  Compatible with all available automated specimen processors, our
errors and variability that occur during manual pipetting.   - +

 . )-
Hamilton *  ,
 
   

For info: 800-648-5950 
   % 
   -
www.hamiltoncompany.com/automated-liquid-handling      
- +

 
a sustainable assay-plate design that reduces the resin materials used
    
 

      
Rubella Virus-Like Particles its advanced durability. These plates support the latest automated and
.* $%   ,



  
/)+  manual methods for the analysis of microbiological samples for clinical,
  
    .    food safety testing, and pharmaceutical laboratories. The SmartPlate
,


/)+   
           

 
 
 

  , 
    ease-of-transport for several sample plates in a single movement. This
,




 '&(!#"






 improves manual handling around the laboratory and between facilities,
    
/)+
         


 (
researchers achieve consistent and accurate results throughout   - +
  

   
  
-
immunoassay development and scale-up manufacturing. The Native

   

     
  
$%  
  ,


   

 
 and incubators.
&)-$      
   

 
 

The Native Antigen Company For info: 866-356-0354

0  ,'' # $*)( (+(%&# 
thenativeantigencompany.com
Cryo-EM Sample Preparation System
Compound Libraries ''%#
 
 
 
 

/ 15  
 
     

     !$'
  
  
     new system facilitates consistent application of samples to high-quality
tool compounds and of assembling relevant sets of focused compounds   !$

 

 



 
 !3
   5.4//2 6/11.   is a technically demanding, largely manual process. Chameleon will allow
1
   
      
   - researchers to more easily and accurately determine the structure of
 !/ 

  ,$ 

  

    

 & 





     " -%  


   

  !$ 
documentation set that highlights activity descriptions, plate positions, protein structures that were previously impenetrable using traditional

     
 

- &

  
 

  

 







   

       "
  
  
 -
compounds.
 
    
  
 
 
 
Enzo Life Sciences  


  



0  ,*## +'% #'&#     
  
  


!
!   !$
   

TTP Labtech

   
 

www.ttplabtech.com/products/sample-preparation/chameleon

Electronically submit your new product description or product literature information! Go to www.sciencemag.org/about/new-products-section for more information.

Newly offered instrumentation, apparatus, and laboratory materials of interest to researchers in all disciplines in academic, industrial, and governmental organizations are featured in this
space. Emphasis is given to purpose, chief characteristics, and availability of products and materials. Endorsement by Science or AAAS of any products or materials mentioned is not
implied. Additional information may be obtained from the manufacturer or supplier.

136 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org/custom-publishing SCIENCE

1004Product.indd 136 9/25/19 1:11 PM

 online @sciencecareers.org

1004Recruitment_SC.indd 137 10/1/19 11:54 AM

 ADVERTISEMENT

1004Recruitment_SC.indd 138 10/1/19 12:11 PM

 ADVERTISEMENT

1004Recruitment_SC.indd 139 10/1/19 12:11 PM

 FOCUS ON CAREER ADVERTISING FEATURE: Produced by the Science/AAAS Custom Publishing Office

neuro sc i enc e

Alzheimer’s research reset

After some costly and disappointing drug trial failures, the a worldwide collaboration that connects researchers studying
field welcomes a funding surge, tools for tracking disease, and the unique population of people at high risk for early-onset
interdisciplinary collaborations to tackle one of science’s most disease. “There’s never been a better time for moving into
stubborn puzzles. By Kendall Powell Alzheimer’s research.”

I
n the last five years, as several large clinical trials testing Multifaceted disease, many approaches
drugs for Alzheimer’s disease failed, the field came to a Worldwide, 47 million people have Alzheimer’s disease or
stark conclusion: These approaches did nothing to slow related dementias, and that number is predicted to double
down—let alone reverse—the course of the disease once in the next 20 years. Aging is by far the biggest risk factor
patients already exhibited symptoms of early dementia. for developing Alzheimer’s—if everyone lived to be 85, one
“We think now that the disease develops over 25 years or in two people would develop dementia. The lion’s share of
so,” says Eric McDade, cognitive neurologist at Washington Alzheimer’s research and drug discovery to date has focused
University in St. Louis School of Medicine, and a coinvestigator on misfolded amyloid and tau proteins, which aggregate
on an 11-year observational study of 500 patients who have to form plaques (amyloid) and tangles (tau) in the brain.
inherited mutations that put them at risk for early-onset forms But the body’s attempt to clear the sticky proteins might
of Alzheimer’s disease. also be contributing to or causing the neurodegeneration.
The failed trials, along with the dawning realization that the Drug trials have almost exclusively sought to use antibodies
disease unfolds over decades, have put the entire field on a targeted toward these two proteins to try to attack and clear
reset—to develop and test interventions that can be used much the misfolded forms or mop up soluble forms, or to inhibit
earlier, to discover new targets beyond misfolded amyloid enzymes responsible for generating the miscreant peptides.
and tau proteins, and to fund large, interdisciplinary, big data Ricardo Dolmetsch, global head of neuroscience for
collaborations. Novartis in Cambridge, Massachusetts, paints a rosier view of
Advances in understanding the role of neuroinflammation, the field’s failures as representing big leaps in understanding

PHOTO: © LIGHTSPRING/SHUTTERSTOCK.COM
new biomarkers and research tools, and an influx of research Alzheimer’s. “We’ve learned a lot from the trials, and our
funding mean it’s a good time to make a career move into capacity to measure disease progress is getting much better,”
the Alzheimer’s field. As history has shown, there won’t be he says. “There are a lot of really exciting experiments to do
any easy answers, and advances will come only through large now.”
collaborations that require a hefty dose of teamwork and Because those experiments run the gamut from nearly every
heaping amounts of perseverance. imaginable corner of biomedical science, researchers with any
“Most people who have thought about the role of amyloid in biology or clinical background, or even training in engineering
Alzheimer’s know that we’ll also need to identify different ways and computational science, can find a way to move into
of intervening,” says McDade, also associate director of the Alzheimer’s work. Expertise in neuroscience and protein
Dominantly Inherited Alzheimer Network Trials Unit (DIAN-TU), folding alone are no longer ideal, or even absolutely required.
cont.>
Upcoming features
Top Employers Survey—October 25 Diversity/Inclusion—January 24 Postdocs—August 21

140 sciencecareers.org SCIENCE

1004Recruitment_SC.indd 140 10/1/19 11:54 AM

 ADVERTISING FEATURE: Produced by the Science/AAAS Custom Publishing Office FOCUS ON CAREER
neuro s c i enc e

factor for Alzheimer’s is age,” he notes, which is, unfortunately,

not modifable. “But we often say that what’s good for the heart
“There’s never been a is good for the brain.”
Another booming area of Alzheimer’s research is the

better time for moving into

development of biomarkers and diagnostic tests to monitor
disease presence and progression. Radioactive positron
emission tomography (PET) tracers enable physicians to image

Alzheimer’s research.” and measure amyloid and tau proteins in the brains of living
patients. Other biomarkers can be measured precisely from
collecting cerebral spinal fuid. However, both types of tests

— Eric McDade are invasive, and PET scans are expensive.

“We need a blood test like [the one] we have for cholesterol
that can be done in any doctor’s offce quickly and
inexpensively,” says Friedman. To spur such development,
“It’s very exciting in terms of careers, because we need ADDF has teamed up with Bill Gates and other philanthropists
people doing biochemical analysis and mass spectroscopy, to fund the Diagnostics Accelerator program. Friedman says
geneticists, cell biologists, pathologists, and those running the program anticipates awarding about $10 million in the
animal studies,” says Paul Aisen, director of University of frst round of funding. So far it has funded projects for blood
Southern California’s Alzheimer’s Therapeutic Research tests and tests that detect amyloid or vascular changes in the
Institute (ATRI) in San Diego. retina, and is reviewing applications for digital tests that use a
smartphone or tablet to monitor disease signs or symptoms.
Projects in progress
At the Alzheimer’s Drug Discovery Foundation (ADDF), a Big data requires big money
nonproft charity based in New York City, diversity is reFected Dedicated Alzheimer’s research funding levels have never
in the research projects being funded there. “We’re really been higher. “There has been a dramatic boost in funding
seeing every type of different approach to Alzheimer’s disease during the last few years,” says Laurie Ryan, chief of the
and related dementias,” says Lauren Friedman, director of Dementias of Aging Branch of the National Institute on Aging
scientifc affairs at ADDF. The areas being explored include the (NIA) in Bethesda, Maryland, adding that the U.S. National
vascular system, epigenetics, neuroprotection, synaptic health, Institutes of Health (NIH) funding specifcally for Alzheimer’s
immunity and inFammation, and metabolic dysfunction, among and related dementias research has blossomed from around
others. USD 440 million in 2011 to USD 2.3 billion this year.
NeuroinFammation and proteostasis, or the management of In 2012, the United States launched the National Plan to
proteins within cells, are trending areas of research in ADDF’s Address Alzheimer’s Disease and set a goal of treating and
portfolio. Researchers investigating proteostasis would like to preventing Alzheimer’s and related dementias by 2025. In
fnd ways to boost the cell’s “garbage disposal” systems, which support of the National Plan, NIH developed a research
identify misfolded, clumping proteins and chew them up for framework to enable the development of therapies for
recycling. individuals at all stages of the disease. Suzana Pentanceska,
Like other companies, Novartis is exploring targets in director of the Offce for Strategic Development and
inFammation and the immune system. Dolmetsch notes that Partnerships in NIA’s Division of Neuroscience, says
the feld has just begun to sort through all of the activated the increased funding and new framework bring many
inFammatory cells that show up in Alzheimer’s patients’ opportunities for Alzheimer’s researchers in both academia
brains and to characterize their various states of activity and and small businesses—ranging from basic, discovery research
exhaustion. Exhausted microglia, the custodians of the brain, to all stages of therapy development.
for example, start pumping out cytokines to call up more One of those opportunities, the Accelerating Medicines
cells (T cells, B cells, macrophages), which leads to chronic Partnership Alzheimer’s Disease Project, or (AMP)-AD, is a
inFammation. public–private partnership that brings together six NIA-funded
Such an inFammation might be fne for a temporary academic teams, four biopharmaceutical companies, and
response to an injury in, say, the fnger. But as a chronic state, fve nonproft organizations. Co-led by Ryan and Petanceska,
“it doesn’t work so well in your brain,” Dolmetsch says. For the fve-year, USD 185 million collaboration is focused on the
neuroinFammation projects, Novartis looks for scientists discovery of new therapies and biomarkers for Alzheimer’s
who have training with overlaps between neuroscience, by harnessing the power of big data and enabling rapid data
immunology, and bioinformatics. sharing.
The George & Anne Ryan Institute for Neuroscience at NIA requires prepublication release of data to the wider
the University of Rhode Island in Kingston investigates research community through the AMP-AD Knowledge Portal
underexplored factors in brain health, including the roles of and the associated web-based interactive platform Agora,
vasculature, immunology, and neuroinFammation as well as where biologists can peruse 500 predicted target genes that
lifestyle and environment in Alzheimer’s disease and other one or more teams have identifed to date as well as the data
neurodegenerative disorders. Behavioral neuroscientist John backing up those predictions.
Robinson works on rodent models of how exercise might play a There’s no embargo on the secondary use of the data by
role in modifying or preventing Alzheimer’s. “The biggest risk other laboratories, says Petanceska, who leads the cont.>

SCIENCE sciencecareers.org 141

1004Recruitment_SC.indd 141 10/1/19 11:54 AM

 FOCUS ON CAREER ADVERTISING FEATURE: Produced by the Science/AAAS Custom Publishing Office

neuro sc i enc e

Featured participants
McDade notes that there’s a shortage of people with
AbbVie Dominantly Inherited Alzheimer M.D./Ph.D.s who have the medical training necessary to do
www.abbvie.com Network Trials Unit (DIAN-TU) clinical research, run clinical trials, and develop appropriate
dian.wustl.edu diagnostic tests. He encourages any early-career researchers
Accelerating Medicines with an interest in the feld to fnd collaborators. “It’s extremely
Partnership Alzheimer’s Disease Eisai bullish—this is the time to come into the feld.”
Project (AMP-AD) www.eisai.com
fnih.org/what-we-do/programs/
amp-ad George & Anne Ryan Institute for Wanted: Diverse skills
Neuroscience With the emergence of so many new avenues of
Alzheimer’s Drug Discovery www.uri.edu/uri-tv/george-anne- Alzheimer’s studies, researchers can come from a wide
Foundation (ADDF) ryan-institute-for-neuroscience variety of biomedical backgrounds, experts say. Mike
www.alzdiscovery.org Gold, vice president of neuroscience development at the
National Institute on Aging biopharmaceutical frm AbbVie in North Chicago, Illinois,
Alzheimer’s Therapeutic Research www.nia.nih.gov notes that his company and others working on solutions for
Institute (ATRI)
Novartis Alzheimer’s want to recruit from biology backgrounds in
keck.usc.edu/atri
www.novartis.com molecular genetics, systems biology, electrophysiology, cell
biology, bioinformatics, and beyond. “Do you have an interest
in radiochemistry to develop new diagnostic ligands? Are
target discovery arm of the AMP-AD. “It’s empowering young you someone who wants to develop animal models to look at
investigators to tap into resources that would take years for a protein transition?” he asks. “We have a place for you.”
single lab to generate,” she says. But the interdisciplinary and “collaboratory” nature of these
Similarly, ADDF has a partnership with Harrington Discovery projects requires that early-career researchers also “frm
Institute to help academic scientists spin their good ideas into up their soft skills,” says Petanceska. Young investigators
drugs that could be tested in the clinic. Grant awardees, who working in large consortia who become versed in the various
receive up to USD 600,000 for two years, will be connected to languages of computational biologists, clinicians, or drug
industry consultants to help shepherd their research through discovery experts will be better placed to contribute to
drug discovery and preclinical phases to develop potential projects and eventually lead teams. Friedman agrees that
therapies. researchers must have strong communications skills. Any
Another innovative NIA translational program, the scientist should be able to boil down the complexity of their
Alzheimer’s Clinical Trials Consortium (ACTC), was launched project into why it’s important for funders, investors, other
in 2017 with nearly USD 70 million in support for fve years, scientists, and the public in a honed elevator pitch.
to accelerate studies of Alzheimer’s and related dementia “Collaboration has been essential, and it’s going to be even
therapies. The ACTC helps researchers with new therapeutics more so,” says Robinson. He advises graduate students and
connect to leading Alzheimer’s clinical experts at 35 sites postdocs to practice the art of collaboration early. Neriman
around the United States. The program solicits ideas frst, and Botas, Eisai’s executive vice president of human resources and
then, for those approved, collaboratively develops those ideas corporate communications in Woodcliff Lake, New Jersey, says
into a clinical trial NIA R01 grant application. that one way to practice collaboration is to have a “beginner’s
This effort also promotes open science with data-sharing mindset” when you are in a meeting. “That means you focus
requirements. For pivotal clinical trials, participants in the more on what you don’t know, no matter how much of an
consortium are required to make their patient screening expert you are, [and try] to ask questions and be tolerant and
and prerandomization data available within 12 months of respectful when seeking input from others.”
enrollment completion and to make their postrandomization Similarly, Novartis, recruiters look for emotional resilience in
data and biosamples available upon trial completion, workers. These are key qualities for a feld where most things
regulatory approval, or within 18 months, whichever fail and often fail big, says Dolmetsch.
comes frst. Making the data available not only honors trial Botas encourages candidates to be prepared for “behavioral
participants, says Ryan, but gives the feld much-needed interviewing”—that is, to have anecdotes and examples ready
information to solve an incredibly complex problem. “It’s a that show how they persisted to overcome a problem in
heterogeneous disease and we need as many eyes on the their graduate or postdoctoral work, or how they worked
data and as many minds focused on solving the problem collaboratively in a team or drew on others’ expertise to get
as possible.” things done.
Pharmaceutical company Eisai brought Aricept,one of the Gold notes that despite the above-average dose of scientifc
frst Alzheimer’s drugs, to market; Eisai also has two anti- frustration faced by Alzheimer's researchers, there is a bit of a
amyloid compounds in Phase III clinical trials and is investing trade-off for them. “You get to work alongside extraordinary
in more research as well. In July, Eisai opened the Center people fghting the battle,” he says. “We need creative
for Genetics Guided Dementia Discovery, dedicated to scientists willing to bring different ideas to the table and try
exploratory immunodementia research. The facility, located radically different approaches. That takes people with courage
in Cambridge, Massachusetts, will host 130 employees, and a sense of adventure.”
including new hires, interns, and postdocs, and an incubator
for Cambridge-based startup companies in neuroscience Kendall Powell is a freelance science writer based in Lafayette, Colorado.
therapeutics.

142 sciencecareers.org SCIENCE

1004Recruitment_SC.indd 142 10/1/19 11:54 AM

 PRIZES

online @sciencecareers.org
myIDP:
Call for Nominations:
A career plan customized Scolnick Prize in
for you, by you. Neuroscience
The McGovern Institute for Brain Research
is accepting nominations for the 17th annual
Edward M. Scolnick Prize in Neuroscience.
The $150,000 prize recognizes an
outstanding discovery or significant
advance in the feld of neuroscience. The
prize ceremony, which includes a public
lecture at MIT, will take place in Spring
2020. A dinner for the recipient and invited
guests follows the prize lecture.

Nomination Deadline: December 15, 2019

Nomination procedures:
For your career in science, there’s only one Candidates for the award must be
nominated by individuals af2liated with
universities, hospitals, medicals schools,
or research institutes, with a background
in neuroscience. Self-nomination is not

focus on neuroscience
Features in myIDP include: permitted. Each nomination should include:
• A biosketch or CV of the nominee.
 Exercises to help you examine your skills, interests, and values. • A letter of nomination with a summary
and analysis of the major contributions of
 A list of 20 scientifc career paths with a prediction of which the nominee to the 2eld of neuroscience.
ones best ft your skills and interests. • Up to two representative reprints will be
accepted.
 A tool for setting strategic goals for the coming year, with
Selection Procedure:
optional reminders to keep you on track. • Members of the selection committee
and faculty af2liated with MIT are not
 Articles and resources to guide you through the process. eligible.
• Announcement of the award recipient
 Options to save materials online and print them for further will be made in January 2020.
review and discussion. • Recipient must attend all events to be
awarded the prize.
 Ability to select which portion of your IDP you wish to share with
advisors, mentors, or others. Past Scolnick Prize Recipients:
2019: Richard Huganir, Johns Hopkins
 A certifcate of completion for users that fnish myIDP. University; 2018: David J. Anderson,
HHMI, Caltech; 2017: Catherine Dulac,
HHMI, Harvard University; 2016: Cornelia
Bargmann, HHMI, The Rockefeller
University; 2015: Charles Gilbert, The
Visit the website and start planning today! Rockefeller University; 2014: Huda Zoghbi,
HHMI, Baylor University; 2013: Thomas
myIDP.sciencecareers.org Jessell, HHMI, Columbia University;
2012: Roger Nicoll, UCSF; 2011: Bruce
McEwen, The Rockefeller University;
2010: Lily and Yuh-Nung Jan, UCSF;
2009: Jeremy Nathans, Johns Hopkins
University; 2008: Michael Davis, Emory
University; 2007: David Julius, UCSF;
In partnership with:
2006: Michael Greenberg, Children’s
Hospital/HMS; 2005: Judith Rapoport,
NIH; 2004: Masakazu Konishi, CalTech

Send nomination packet to:

gwolf@mit.edu

For more information:

mcgovern.mit.edu/scolnick

1004Recruitment_SC.indd 143 10/1/19 11:54 AM

 ways that Science Careers
can help advance your career

1. Register for a free online account on

ScienceCareers.org.

2. Search thousands of job postings and fnd

your perfect job.

3. Sign up to receive e-mail alerts about job

postings that match your criteria.

4. Upload your resume into our database and

connect with employers.

5. Watch one of our many webinars on

diferent career topics such as job searching,
networking, and more.

6. Download our career booklets, including

Career Basics, Careers Beyond the Bench,
and Developing Your Skills.

7. Complete an interactive, personalized career

plan at “my IDP.”

8. Visit our Employer Profles to learn more

about prospective employers.

9. Research graduate program information and

fnd a program right for you.

10. Read relevant career advice articles from our

library of thousands.

Visit ScienceCareers.org
today — all resources are free

SCIENCECAREERS.ORG

1004Recruitment_SC.indd 144 10/1/19 11:55 AM

 online @sciencecareers.org
Independent Research
Fellowships Leading to Tenured
Faculty Positions at the
John Innes Centre
The John Innes Centre (JIC), Norwich, UK is a world leading centre
of excellence in plant and microbial sciences based on the Norwich
Confused about Research Park. We are inviting applications from outstanding researchers
who either hold, or wish to apply for Independent Research Fellowships
your next career move? [such as a UKRI Future Leaders Fellowship (https://www.ukri.org/funding/
funding-opportunities/future-leaders-fellowships/), a Royal Society
University Research Fellowship (http://royalsociety.org/grants/schemes/
Download Free Career university-research/), or a BBSRC David Phillips Fellowship (https://
bbsrc.ukri.org/funding/filter /david-phillips/)]. Shortlisted candidates will
Advice Booklets! be invited to give a seminar at a Fellows Conference at the JIC on 03
February 2020. At the conference, you will also be able to discuss your
ScienceCareers.org/booklets proposals, the development of your group and your future career plans
in depth with JIC faculty.
After the conference, we will select and mentor outstanding candidates
in writing Fellowship applications and/or offer the opportunity to move
existing Fellowships to the JIC. Candidates who win Fellowships
will be considered for transfer onto tenure-track at 3 years, and if
transferred, for tenure at 5 years. Considerable additional resources
will be provided to Fellows by the Centre. For further information,
contact mark.buttner@jic.ac.uk
Further details and particulars can be found at https://www.jic.ac.uk/
vacancies/
To apply: please e-mail a 2-page summary of your research plan, a
copy of your CV and arrange for three letters of recommendation
to be e-mailed to fellows@jic.ac.uk by Friday 22 November 2019.
Before applying please read our Privacy Notice.
The John Innes Centre is a registered charity
(No223852) grant-aided by the Biotechnology and
Biological Sciences Research Council and is an
Equal Opportunities Employer and supports flexible
working.

YOU’D SMILE
TOO IF YOU
JUST ADVANCED
YOUR CAREER.
Find your next job at ScienceCareers.org

There’s scientific proof that when you’re happy

with what you do, you’re better at what you do. Access
career opportunities, see who’s hiring and take
advantage of our proprietary career-search tools.
Get tailored job alerts, post your resume and manage
your applications all in one place: sciencecareers.org

1004Recruitment_SC.indd 145 10/1/19 11:54 AM

 WORKING LIFE
By Adriana L. Romero-Olivares

” Review with care

T
his paper requires significant editing, as it is not written in sound English and cannot be accepted
in its current form.” “This sentence does not make any sense.” “The authors need a native English-
speaking co-author to thoroughly revise the grammar of this manuscript.” My heart sank when I
received this feedback from a reviewer for the first paper I had ever submitted as a first author.
The reviewer didn’t say a thing about the underlying science—but isn’t that what peer review is
supposed to evaluate? Regardless, the reviewer recommended that our manuscript be rejected.
This happened about 10 years ago. But as a nonnative English speaker who needs to publish in English,
I still think about the experience today.

I expected my paper to fare well. I Reviewers, you don’t need to

knew the science was strong. I had be rude. Here are three principles
the resources and backup data to for providing constructive reviews,
make whatever scientific changes even when you think an article is
the reviewers suggested. I also knew poorly written.
that my English skills weren’t great. First and foremost, do not make
But I had a native English-speaking assumptions about the quality of a
co-author and an adviser who was paper based on the authors’ names
proficient in English, and they both and affiliations. This form of bias is
thoroughly revised the manuscript. rooted in racism.
When I received those comments Second, you are a reviewer—not
from the reviewer, I was crushed. I an editor. Focus on the research.
felt our work wasn’t being judged If the English is so poor you can-
fairly. Had the reviewer assumed not review the paper or provide
that, because my co-authors and feedback on the science, tell the
I were all affiliated with the same editor so that they can decide how
Mexican research institute, none to proceed.
of us was proficient in English? My “You are not a martyr or the Third, if the paper is not written
co-author told me I shouldn’t take
the comments personally, and that
savior of people who did not grow in sound English, it is OK to cor-
rect grammatical errors and help
they did not reflect on the quality up speaking English.” improve the writing. But remem-
of my work, but that didn’t make ber that you are not a martyr or the
me feel any better. How could I succeed in academia if I savior of people who did not grow up speaking English. Be
could not manage to publish a paper in English? kind. For example, you can write, “I cannot understand
Luckily, the other two reviewers made useful suggestions what the authors are trying to communicate here.” You
about the science, and the editor invited us to submit a re- can also suggest editing help from someone with “full pro-
vised version. I modified the science based on the helpful fessional proficiency in English.” Don’t demand “a native
reviewers’ feedback and made the grammar and style cor- English speaker”; that is not synonymous with being a
rections demanded by the rude reviewer, and my first first- good writer.
author paper was accepted. As reviewers, we help ensure that high-quality sci-
Since then, no reviewers have had a problem with my ence gets published. We can—and should—do this with-
English—maybe because I moved to the United States and out forgetting that behind every paper are people who

ILLUSTRATION: ROBERT NEUBECKER

started publishing with co-authors who have American- worked hard to put together a manuscript reporting
sounding names. I also put a lot of work into improving their cherished research. As scientists, we know that writing
my language skills, including participating in writing work- is hard. Keep in mind that writing in a second language is
shops and reading books about writing in English. even harder. j
But many of my colleagues in Mexico tell me stories
of dealing with reviewers who are harsh and patronizing Adriana L. Romero-Olivares is a postdoc at the University
about their English. And I know that other researchers of New Hampshire in Durham. Send your career story to
around the world experience similar insults. SciCareerEditor@aaas.org.

146 4 OCTOBER 2019 • VOL 366 ISSUE 6461 sciencemag.org SCIENCE

1004WorkingLife.indd 146 9/27/19 6:13 PM

 Antibodies Recognize Antigens...
How will you get recognized for your
immunology research?

Science Immunology publishes original, peer-reviewed, science-based

research articles that report critical advances in all areas of immunological
research, including important new tools and techniques.
Submit your research today. Learn more at: ScienceImmunology.org

1004Product.indd 147 9/25/19 1:11 PM

 Get A Great Assay
Go from good to great by following
our ELISA guide

DOW NLOA D
Learn more at rndsystems.com/elisa The ELISA Guide

Global bio-techne.com info@bio-techne.com TEL +1 612 379 2956 USA TEL 800 343 7475 Canada TEL 855 668 8722
Europe | Middle East | Africa TEL +44 (0)1235 529449 China TEL +86 (21) 52380373
For research use or manufacturing purposes only. Trademarks and registered trademarks are the property of their respective owners.

1004Product.indd 148 9/25/19 1:11 PM