Science20190712 DL Artificial Muscles

Can biodiversity survive on Boosting CAR–T cells for Promoting reproducibility
oil palm plantations? p. 112 solid tumors pp. 119 & 162 with confidential data p. 127
$15
12 JULY 2019
sciencemag.org
ARTIFICIAL
MUSCLES
Heating prestretched
fbers triggers contraction
pp. 125, 145, 150, & 155
Since the first Deep Brain Stimula-
tion initiative of Tsinghua Univeristy
in 2000, PINS Medical has graduallyy
established a multinational corpora a-
tion with headquarter based in Beijiing
and international business cente er in
Singapore. As an innovative high-tech ch enter
enter-
prise with focus on neuromodulation, a variety of
clinical products have been developed to date, which include stimulators for deep brain, vagus nerve, spinal cord and
sacral nerve stimulation therapies. PINS Medical devotes itself to providing cutting-edge treatments for patients who
suffer from neurological disorders such as Parkinson’s Disease, Epilepsy, Chronic Pain and OAB, etc.
As part of the “National Engineering Laboratory for Neuromodulation”, PINS Medical works in close cooperation with
Tsinghua University and the numerous affiliated clinical centers, becoming a center of attraction for a wide range of
professional talents in areas of clinical research, innovative R&D and business management. Since 2008, PINS Medi-
cal has developed rapidly in becoming a leading brand in neuromodulation within the Chinese market, due to the
success of its creative research platform that efficiency links basic research, R&D of novel products, clinical testing and
market entry.
With an outstanding reputation as a high-tech healthcare corporation, PINS Medical has a primary mission for providing
innovative, high-quality products and services for patients to improve quality of life. PINS, which stands for Programma-
ble Implanted Neuromodulation Stimulator, is also an abbreviation of “Patient Is No.1 alwayS”. This clearly presents the
goal of PINS Medical for “restoring hope”, not simply as an innovation company but also across society to citizens.
Looking into the future with the continuous rise in incidence of neuropsychiatric disorders and increased social burden
across the globe, PINS Medical along with local governments, research centers, companies and top academic scien-
tists, are now developing and promoting innovative therapies worldwide.
0712Product.indd 98 7/2/19 9:00 AM

CONTENTS
1 2 J U LY 2 0 1 9 • V O LU M E 3 6 5 • I S S U E 6 4 4 9
112
The palm oil boom has consumed orangutan habitat. But some ecologists think better farming practices could help preserve biodiversity.
NEWS 111 WAS OUR SPECIES IN EUROPE 210,000

YEARS AGO?
Skepticism greets startling conclusion
120 MAPPING GLOBAL PROTEIN CONTACTS
Genome-scale coevolution experiments
expand bacterial protein interactomes
from skull fragment found in Greek By S. Vajda and A. Emili
IN BRIEF cave By L. Wade ▶ REPORT P. 185
104 News at a glance 122 STROMAL DIRECTIVES CAN

FEATURES
CONTROL CANCER
IN DEPTH 112 MAKING PEACE WITH OIL PALM Could reprogramming the pretumor
106 POLIO ERADICATION CAMPAIGN Some scientists are helping make microenvironment transform cancer
LOSES GROUND palm plantations greener. Others say care? By T. D. Tlsty and P. Gascard
Surging cases in Pakistan and Africa that doing so legitimizes a destructive
have dashed hopes of defeating the virus industry By D. Rochmyaningsih 124 SUPRAMOLECULAR CAGES
this year By L. Roberts TRAP PESKY ANIONS
Carbon-hydrogen bonds in a cryptand
107 CALIFORNIA’S STEM CELL RESEARCH
FUND DRIES UP INSIGHTS cage boost chloride entrapment to
ultrahigh affinity By K. Bowman-James
▶ REPORT P. 159
Researchers hope a planned ballot
initiative will renew funding in 2020
PERSPECTIVES 125 STRONGER ARTIFICIAL MUSCLES,
By J. Kaiser
116 SEABIRD CLUES TO ECOSYSTEM WITH A TWIST
Tailored fiber-shaped actuators with
108 SOLAR PLUS BATTERIES IS NOW HEALTH
twisted and coiled designs have high
CHEAPER THAN FOSSIL POWER Seabird monitoring provides essential
energy densities By S. Tawfick and Y. Tang
Falling prices help utilities start to information on the state of marine
▶ REPORTS PP. 145, 150, & 155
decarbonize By R. F. Service ecosystems By E. Velarde et al.
▶ PODCAST
POLICY FORUM
109 GUT MICROBES MAY HELP
MALNOURISHED CHILDREN 118 PROTECTING FETAL DEVELOPMENT 127 CERTIFY REPRODUCIBILITY
Commonplace foods restore bacteria Control of cell fusion in the placenta WITH CONFIDENTIAL DATA
that can improve health By E. Pennisi affects the balance between health and A trusted third party certifies that
▶ RESEARCH ARTICLES PP. 139 & 140 harm By P. Kellam and R. A. Weiss results reproduce By C. Pérignon et al.
▶ REPORT P. 176
110 PSYCHOLOGIST AIMS TO STUDY BOOKS ET AL.
PHOTO: ERIK MEIJAARD
DIVERSE MINDS, NOT WEIRDOS 119 BOOSTING ENGINEERED T CELLS 129 THWARTING AND EXPLOITING
Daniel Haun wants his fellow Vaccines augment the antitumor NAZI SCIENCE
psychologists to study people activity of cellular therapy A thrilling tale of wartime derring-do
outside of rich, Western societies By N. Singh and C. H. June meets a richly researched story of postwar
By K. Kupferschmidt ▶ REPORT P. 162 intellectual exploitation By C. Hall
SCIENCE sciencemag.org 12 JULY 2019 • VOL 365 ISSUE 6449 99
0712TOC.indd 99 7/9/19 5:07 PM

CONTENTS
RESEARCH ARTICLES 176 REPRODUCTION

MICROBIOTA IFITM proteins inhibit placental
139 Effects of microbiota-directed syncytiotrophoblast formation and
foods in gnotobiotic animals and promote fetal demise J. Buchrieser et al.
▶ PERSPECTIVE P. 118
undernourished children
J. L. Gehrig et al.
RESEARCH ARTICLE SUMMARY; FOR FULL
180 NEUROSCIENCE
Posterior parietal cortex plays a causal
TEXT:dx.doi.org/10.1126/science.aau4732
role in perceptual and categorical
140 A sparse covarying unit that decisions Y. Zhou and D. J. Freedman
describes healthy and impaired
human gut microbiota development 185 PROTEIN NETWORKS
A. S. Raman et al. Protein interaction networks revealed
RESEARCH ARTICLE SUMMARY; FOR FULL by proteome coevolution Q. Cong et al.
TEXT:dx.doi.org/10.1126/science.aau4735 ▶ PERSPECTIVE P. 120
▶ NEWS STORY P. 109
141 NEUROSCIENCE DEPARTMENTS

A sense of space in postrhinal cortex 101 EDITORIAL
P. A. LaChance et al. Balancing science and security
RESEARCH ARTICLE SUMMARY; FOR FULL TEXT: By Mary Sue Coleman
109, 139, & 140 dx.doi.org/10.1126/science.aax4192
194 WORKING LIFE
REPORTS A father’s odyssey By Kazumasa
130 AMERICAN ADVENTURE AND Z. Tanaka
142 SURFACE CHEMISTRY
AUTONOMOUS CARS Molecular structure elucidation with
What will become of U.S. car culture charge-state control S. Fatayer et al.
in the age of self-driving vehicles?
By L. Vinsel ARTIFICIAL MUSCLES
145 Strain-programmable fiber-based
LETTERS artificial muscle M. Kanik et al.
133 CONSERVATION: BEYOND 150 Sheath-run artificial muscles
POPULATION GROWTH J. Mu et al.
By T. Weldemichel et al.
155 Shape memory nanocomposite
133 RESPONSE fibers for untethered high-energy
OBERT NEUBECKER
By J. O. Ogutu et al. microengines J. Yuan et al.
▶ PERSPECTIVE P. 125
CREDITS: (PHOTO) INTERNATIONAL CENTRE FOR DIARRHOEAL DISEASE RESEARCH, BANGLADESH; (ILLUSTRATION) ROBERT
134 POLAND’S CONFLICTING
ENVIRONMENTAL LAWS 159 HOST-GUEST CHEMISTRY
By G. Grzywaczewski and I. Kitowski Chloride capture using a C–H
ON THE COVER
hydrogen-bonding cage Y. Liu et al.
134 TECHNICAL COMMENT ABSTRACTS ▶ PERSPECTIVE P. 124 Upon heating, a
strain-programmable
162 IMMUNOTHERAPY fiber-based artificial
RESEARCH Enhanced CAR–T cell activity

against solid tumors by vaccine
boosting through the chimeric receptor
muscle composed of
thermoplastic and elas-
tomer layers actuates a
miniature printed arm
L. Ma et al.
and lifts a dumbbell 650
IN BRIEF ▶ PERSPECTIVE P. 119
times its own weight.
135 From Science and other journals These fiber-based muscles, fabricated at
168 INORGANIC CHEMISTRY lengths of hundreds of meters, are highly
Characterization of hydrogen-substituted tunable in terms of power and lateral
REVIEW
silylium ions in the condensed phase dimensions and thus may offer opportuni-
138 ANTHROPOLOGY Q. Wu et al. ties in robotics and prosthetics.
Late Pleistocene exploration and See pages 125, 145, 150, and 155.
settlement of the Americas by modern 173 CATTLE DOMESTICATION Photo: Ken Richardson
humans M. R. Waters Ancient cattle genomics, origins, and
REVIEW SUMMARY; FOR FULL TEXT: rapid turnover in the Fertile Crescent New Products .............................................190
dx.doi.org/10.1126/science.aat5447 M. P. Verdugo et al. Science Careers ..........................................191
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.
100 12 JULY 2019 • VOL 365 ISSUE 6449 sciencemag.org SCIENCE
0712TOC.indd 100 7/9/19 5:07 PM

EDITORIAL
Balancing science and security
F
ederal elected officials and members of the Unit- mation security controls, NSDD 189 has ensured the
ed States intelligence community have expressed widespread, public, and open dissemination of research
concern about the security of the nation’s scien- results. Maintaining such access is essential to scientific
tific and technological information, articulating progress as well as to national and economic security.
worries about China, Russia, and Iran. The fears The core principles underlying NSDD 189 are now
include potential academic espionage, theft of threatened. Legislative proposals, such as that intro-
intellectual property, and threats to academic induced recently in Congress by Sen. J. Hawley (R-MO),
tegrity. As federal policy-makers respond, it is critical would impose new limitations on who can work on, Mary Sue Coleman
that they work with the scientific community to bal- and what information can be shared about, unclassi-
is president of
ance securing strategically important information with fied research projects deemed by government bureau-
the Association
maintaining the free flow of fundamental scientific crats to be “sensitive”—a category that actually does
of American
knowledge and international talent necessary for scien- not exist under current rules. If enacted, this proposal
tific progress. History is a guide to striking this balance. would negatively affect universities’ ability to engage Universities,
During the Cold War, the U.S. security community in scientific research on behalf of the U.S. government. Washington, DC,
raised similar concerns about the Soviet Union. In A more effective approach to address the current USA. marysuec@
response, the Association of Ameri- security concerns is contained in aau.edu
can Universities worked with the the Securing American Science and
Department of Defense in the early
1980s to establish a forum for De-
“Indiscriminate Technology Act, introduced in May
by Rep. R. M. Sherrill (D-NJ) and
fense and university officials. As a
result, the academic and security restrictions... Rep. A. Gonzalez (R-OH). This leg-
islation, now part of a larger bill,
communities held important discus-
sions on how to identify and secure could do establishes an interagency working
group under the existing authority
research that warranted special pro-
tections while simultaneously en- irreparable harm to granted to the White House Office
of Science and Technology Policy’s
suring that such measures did not National Science and Technology
unnecessarily restrict what could be the U.S. scientific Council. The working group would
published in scientific journals or coordinate activities across dispa-
limit the ability of universities to tap
foreign scientific talent.
enterprise.” rate federal agencies to ensure that,
in accordance with NSDD 189, exist-
Building on those discussions, in ing security controls such as clas-
1982 the U.S. National Academy of Sciences released sification are properly employed to protect national
Scientific Communication and National Security. security while not limiting the free flow of scientific
Citing this report, President Reagan issued National information. Additionally, the legislation establishes a
Security Decision Directive 189 (NSDD 189) in 1985. new National Academy of Sciences, Engineering, and
NSDD 189 states that to the maximum extent possible, Medicine roundtable to facilitate ongoing dialogue
the products of basic and applied research funded between the university and scientific community and
by the federal government should be published and federal officials about this crucial balance.
widely disseminated, and that classification should be For American science to advance, basic and applied
used in those limited circumstances when controlling research must be openly and widely shared. At the same
scientific information is necessary to protect national time, the United States must continue to benefit—as
security. NSDD 189 was reaffirmed in 2001 by Secre- it has for decades—from the world’s best and bright-
tary of State Condoleezza Rice during the George W. est scholars coming to the country to study and work.
Bush administration. Indiscriminate restrictions on either could do irreparable
PHOTO: ASSOCIATION OF AMERICAN UNIVERSITIES
By establishing that government strongly protects a harm to the U.S. scientific enterprise.
narrow set of key technologies when imposing infor- –Mary Sue Coleman
10.1126/science.aay5856
0712Editorial.indd 101 7/9/19 4:34 PM

Advertorial
Best of both worlds:

Next-generation sequencing in frozen and FFPE tissue
Clinical tissue samples come in many forms, and some are easier to When dealing with clinical samples, DNA sequencing can be
analyze by nucleic acid sequencing than others. There are two popular applied in many ways. In the case of rare genetic diseases and types of
forms of sample preservation: fresh-frozen and formalin-fixed paraffin- hereditary cancer, comparing sequences of a person with the disease
embedded (FFPE). In short, it’s easier to sequence fresh-frozen samples and a family member without it can identify the disease-related
than FFPE ones, but FFPE tissues can be stored for longer and at ambient mutation. A well-known example involves the BRCA (BReast CAncer)
temperatures. New technology, however, accurately analyzes the nucleic genes and female breast cancer. “In the case of sporadic cancers, the
acids in both kinds of samples. oncologist wants to identify variants present in the cancer cells—and
not in the normal cells of a person—which lead to defects in one or
Scientists started fixing samples in formalin in the late 1800s. As a several genes,” Robbe says. Performing WGS/WES on normal and
result, there are now millions of extant FFPE samples around the world, cancerous cells can provide valuable data. “This information will guide
which represent an invaluable collection of potential information about the clinician in the diagnosis—in subtyping and grading the cancer
disease—especially cancer, because virtually every kind of tumor can or identifying the tissue of origin of a metastasis—or in treatment, by
IMAGES: © SHUTTERSTOCK (TOP) AND DRS. ROBBE AND MESSINA (BOTTOM TWO RIGHT).
be found in this sample format. Despite this gigantic clinical resource, giving a drug that will take advantage of these mutated genes.”
much of it has long remained unreachable through sequencing These sequencing applications can be challenging. If a fresh-frozen
analysis, because FFPE samples of cancer tissue include only small sample includes a large proportion of healthy cells relative to tumor
amounts of low-quality nucleic acids (1). cells, sequencing it in bulk can
Until just a few years ago, microgram quantities of purified DNA miss crucial mutations. Archiving
or RNA were required for next-generation sequencing (NGS) profiling. samples through an FFPE process
However, that has changed. “Recent advances have enabled the preserves the tissue morphology,
preparation of high-quality NGS libraries from just tens of nanograms but degrades the DNA and
of material, including degraded samples with fragment sizes of less RNA, increasing the difficulty
than 100 base pairs,” says Kevin Meldrum, vice president of product of the sequencing analysis
development at Illumina, a biotechnology company headquartered in and reducing its accuracy. To
San Diego, California. “Genomic profiling can now be utilized across determine the quality of DNA
a broader set of samples and on a more routine basis,” he says, and it samples, Robbe uses Illumina’s
Pauline Robbe, (left), and Dave Messina, (right),
can be applied equally effectively to both fresh-frozen and FFPE clinical InfiniumTM FFPE QC Kit. For both use various kits from Illumina to aid in
samples, both with traditional mechanical fragmentation, and more damaged DNA, Illumina’s their research.
recently with advances in enzymatic fragmentation, such as Illumina’s TruSightTM Oncology UMI
NexteraTM Flex for Enrichment kit. Reagents use unique molecular modifiers (UMIs) that can be combined
with computational analysis to improve sequencing accuracy.
Viewing variants Another issue can arise when researchers are working with RNA and
Using whole-genome sequencing (WGS), scientists can sequence want to study only messenger RNA (mRNA)—ribosomal and transfer RNA
the entire genome of an organism; or with whole-exome sequencing can contaminate their sample. In such cases, Meldrum suggests using
(WES), they can choose to sequence only the nucleic acids that are Illumina’s TruSeq™ RNA Exome Kit. “It allows researchers to interrogate
transcribed to RNA. “The key clinical applications of WGS and WES are mRNA using a capture approach that avoids contamination by other
the discovery of pathogenic variants,” says Pauline Robbe, NGS expert at RNA species,” he explains. “This method is also compatible with nucleic
the University of Oxford in the United Kingdom. acids derived from fragmented FFPE samples.”
0712Product.indd 102 7/2/19 9:00 AM

Advertorial
Decoding the immune system fresh-frozen and FFPE samples. The

To understand disease and how the body fights it, scientists group concluded that single-cell profiling
are intently studying the immune system. At Cofactor Genomics, can be used to “dissect the genetics of
they precisely characterize the immune cells in FFPE samples. “This histologically or phenotypically
is informative for understanding cancer,” says chief distinct cancer components, and
operating officer Dave Messina. “For example, trace their evolutionary history.”
immune characterization of a sample can
be used to predict a tumor’s response to From heterogeneity to
chemotherapy.” harmony
To characterize the immune The ability to work with FFPE
cells in tumor samples, workers at samples will dramatically increase
Cofactor Genomics first use RNA the amount of clinical sequencing
sequencing (RNA-Seq) to acquire information produced—which
hundreds of signals of pure will in turn create a series of data
immune-cell populations, then challenges. NGS data, for instance,
place them into computational drives three levels of analysis.
models they call Health Expression Primary analysis consists of identifying
Models. “We’re taking advantage the bases in the raw read of a sequence, a
of modern computational techniques to function built into each sequencer.
integrate lots of signals into a robust model of Secondary analysis differentiates the bases in
Researchers can now conduct next-generation
something as complex as immune cells,” Messina sequencing on single cells from fresh-frozen and
an analyzed sample from a reference sequence, and
explains. “We pair that with molecular technology formalin-fixed paraffin-embedded (FFPE) tissue. identifies genomic variants. Various algorithms—such
to get reproducible readouts even from challenging as DRAGEN (Dynamic Read Analysis for Genomics)™,
clinical samples.” Using this multidimensional approach enables developed by Edico Genome, which is now owned by Illumina—exist
Cofactor to quantitatively measure immune cells in complex mixtures for this process. “DRAGENTM accelerates this analysis with hardware and
with high sensitivity. software, reducing whole-genome variant calling from hours to 15 or
Scientists can take advantage of this technology through Cofactor 20 minutes,” Meldrum says.
ImmunoPrismTM, an immune-profiling kit that works with Illumina Finally, tertiary analysis interprets the significance of the bases and
sequencing platforms. “In particular, this combination provides variants in a sequence. “There are references and databases that you
a clear readout of what immune cells are doing in the tumor can ping to see if the variants are important,” Meldrum explains.
microenvironment,” says Messina, who notes that ImmunoPrismTM can To make the most of this collection of heterogenous data,
also be used with fresh-frozen or FFPE samples. In addition, Cofactor companies must harmonize their data formats. “There are a couple of
Genomics offers research-use and clinical-use versions of the platform, community forums underway,” Meldrum says. For example, Illumina is
and the latter have been approved under the company’s CAP-CLIA collaborating with the Broad Institute in Cambridge, Massachusetts, to
license. use its Genome Analysis Toolkit (GATK). “We’d like to make sure that our
algorithms are compatible with GATK,” he points out.
Hunting down heterogeneity There are many good reasons to make the effort needed to drive
To deal more precisely with cancer, both in diagnosis and treatment, this kind of teamwork. As Meldrum says, “These advancements enable
scientists must study tumors more carefully. “Historically, people cancer researchers to utilize a broader range of NGS-based profiling
did bulk testing—taking a general tissue sample and processing it,” techniques on precious clinical and archival samples, moving beyond
Meldrum says. “Now, we have techniques to isolate single cells from targeted assays that generate data of variable quality.”
tissues and profile those cells individually for a higher-resolution view
at a cellular level to identify mutations.”
There’s good reason to delve into cancer at the single-cell level.
“A substantial proportion of tumors consist of genotypically distinct
IMAGE: © SHUTTERSTOCK
subpopulations of cancer cells,” according to Jorge Reis-Filho and References

colleagues at the Memorial Sloan Kettering Cancer Center, in New 1. P. G. Patel et al., PLOS ONE 12, e0179732 (2017).
York City, New York. “This intratumor genetic heterogeneity poses 2. L. G. Martelotto et al., Nat. Med. 23, 376–385 (2017).
a substantial challenge for the implementation of precision
medicine” (2). Produced by the Science/AAAS
Custom Publishing Office
Reis-Filho’s team used Illumina library preparation and multiplex
sequencing to analyze FFPE breast-cancer samples. Their study
produced similar results when interrogating gene copy number in
0712Product.indd 103 7/2/19 9:00 AM

NEWS
It’s an insult to the science, of course, but it’s also an
“ insult to the people who need this information.
”
Joel Clement, a former U.S. Department of the Interior climate staffer, in E&E News, on reports
that the U.S. Geological Survey scrubbed mentions of climate change from press releases.
Hungary’s academy loses control

IN BRIEF | Hungary’s populist govern-
G OV E R N A N C E
ment last week enacted a controversial
Edited by Jeffrey Brainard law that lets it exert more control over the
country’s Academy of Sciences. A new orga-
nization called the Eötvös Loránd Research
Network will take over the academy’s
15 research institutions and their bud-
gets and buildings. The network’s board
will consist of six members named by
the Academy of Sciences and six by the
Ministry of Innovation, with the deciding
vote cast by the head of the committee,
who will be appointed by the minister of
innovation. The government says the law
will promote efficiency; critics, such as
the European Federation of Academies of
Sciences and Humanities, say it severely
restricts academic freedom and is part of
efforts by nationalist Prime Minister Viktor
Orban’s government to silence critics.
#MeToo activist out at Vanderbilt

| BethAnn McLaughlin, the
WO R K P L AC E
controversial neuroscientist who founded
the advocacy organization #MeTooSTEM
last year and has become a prominent
ASTROPHYSICS advocate for better policing of sexual
harassment in science, left Vanderbilt
Crab nebula packs a photon punch University Medical Center in Nashville on
8 July, “by mutual agreement” with the
institution, she says. McLaughlin was
denied tenure in 2017 but appealed, saying
C
hinese and Japanese scientists last week described the highest
her tenure process had been tainted by
energy photons ever observed: gamma rays with energies up to
retaliation for her testimony in a sexual
450 trillion electron volts. They traced the particles of light to the harassment case. A faculty committee
Crab nebula (above), the remnant of a stellar explosion observed in February declined to reverse that
by Chinese astronomers nearly 1000 years ago, and the powerful tenure denial, and Vanderbilt University
PHOTO: NASA/ESA/J. HESTER AND A. LOLL/ARIZONA STATE UNIVERSITY

pulsar, a rotating neutron star, that now sits at the nebula’s heart. Chancellor Nicholas Zeppos accepted its
decision. In a statement, Vanderbilt said,
Researchers spotted the photons using the Tibet ASgamma array of “We share a strong commitment to the
nearly 600 detectors, 4300 meters above sea level on the Tibetan Plateau success of women in [science, technology,
in China. When gamma rays strike Earth’s atmosphere, they create cas- engineering, and math], and we wish Dr.
cades of electrons and other subatomic particles that hit the detectors; McLaughlin well in her future endeavors.”
astrophysicists can study the particle showers to reconstruct the rays’
energies and trajectories. Scientists believe the new observations sup- Health institute head steps down
port models of a process called inverse Compton scattering, in which #METOO | The director of the Arnhold
electrons whipped to extremely high energies by a pulsar’s magnetic field Institute for Global Health at the Icahn
School of Medicine at Mount Sinai in
smash into photons in the cosmic microwave background, hurling them New York City is leaving that position
through the galaxy. The 90 researchers from two dozen institutions re- after current and former employees sued
port their findings in a paper accepted at Physical Review Letters. him and the school in April alleging age
0712NewsInBrief.indd 104 7/9/19 5:28 PM

and gender discrimination. The direc- will be “integrated” under NRF. “This would
tor, Prabhjot Singh, “has chosen to step be adequately supplemented with additional Alzheimer’s grant dispute ends
down,” wrote Dennis Charney, the school’s funds,” Sitharaman said. India spends about | The University of
L E G A L A F FA I R S
dean, in a 3 July email. Singh will remain $15 billion annually on R&D, 0.67% of its Southern California (USC) in Los Angeles
on the medical school’s faculty, Charney gross domestic product. has agreed to pay the University of
wrote; one of the plaintiffs, Natasha California, San Diego (UCSD), $50 million
Anushri Anandaraja, who directs the to settle a fractious legal dispute that began
medical school’s Office of Well-Being and Austria nears glyphosate ban when a physician who headed a major study
Resilience, called that “unacceptable.” | Lawmakers in Austria
AG R I C U LT U R E of Alzheimer’s disease moved institutions.
voted last week to forbid all use of In the 2015 lawsuit, UCSD accused recently
glyphosate, the world’s most widely used departed faculty member Paul Aisen and his
India mulls science foundation weed killer, citing controversial claims then-new employer, USC, of commandeer-
FUNDING | Indian leaders plan to set up a that it may cause cancer (Science, 24 May, ing data from the UCSD-based Alzheimer’s
new National Research Foundation (NRF) p. 717). If the law passes the upper house Disease Cooperative Study (ADCS) and
to oversee research both in science and the as expected, the country would be the misrepresenting his claim to $55 million in
humanities. The plan, presented as part of first European Union member to ban the federal funding for the project. The study
the first budget of Prime Minister Narendra herbicide. It’s unclear whether the ban can continues at UCSD and its collaborating
Modi’s new government last week, needs go into effect; in 2017, the European Union institutions, though some sponsors of trials
approval from India’s Parliament, and reauthorized use of the herbicide through previously in the ADCS network shifted
details remain sketchy. It’s unclear what 2022 for all member countries, a policy their contracts to a new center at USC. In
NRF’s budget will be and how its funds will that many observers say takes precedence a statement about the settlement, USC said
be distributed. Now, about a dozen Indian over national laws. Many farmers and the way Aisen and his team transferred
ministries and departments fund research; some conservation groups say glyphosate is their research and staff “did not align with
Minister of Finance Nirmala Sitharaman crucial to farming techniques that conserve the standards of ethics and integrity which
said in Parliament on 5 July that these funds soil, such as minimal plowing. USC expects of all its faculty.”
BY THE NUMBERS
24%
Share of nearly 200 U.S. and
Canadian biology conferences
that have posted codes of
conduct. Only half of the codes
mentioned sexual misconduct
(Proceedings of the
National Academy of Sciences).
A 7.1-magnitude earthquake on 5 July

displaced desert floor near Ridgecrest, California.
10°C
Increase over average
temperature in parts of Europe
during late June. It was the
SEISMOLOGY
hottest June on record in Europe
Experts ponder alert trigger after California earthquakes and worldwide, driven by climate
change (Copernicus
T
wo earthquakes last week, centered in Southern California’s Mojave Desert—the Climate Change Service).
largest to hit the region in 20 years—rattled residents’ nerves and raised questions
about a warning system. Neither quake caused deaths or significant damage. Still,
PHOTO: ROBYN BECK/CONTRIBUTOR/GETTY IMAGES
91%
many residents of Los Angeles, California, some 150 kilometers away, felt the trem-
ors and expressed surprise that they had not been alerted by the U.S. Geological
Survey’s (USGS’s) new earthquake warning system, ShakeAlert. It can send a notifica-
tion through a cellphone app to people far from an earthquake’s epicenter at least a few Proportion of U.S. survey
seconds before the quake’s vibrations reach them (Science, 2 November 2018, p. 514). respondents who say
But USGS said the ShakeAlert system had worked as intended, creating an alert 6.9 sec- U.S. scientific achievements
onds after the first quake began. It did not relay the alert to Los Angeles residents—the make them proud of their
only members of the public with access to ShakeAlert—because the quakes were too country—more than seven
distant to cause damaging shaking there. In response to users’ confusion, city and USGS other aspects of government
officials have said they may lower the shaking intensity at which alerts are delivered. and society, including the
military (Gallup).
0712NewsInBrief.indd 105 7/9/19 5:28 PM

A relative mourns
a female polio
vaccinator killed by
IN DEP TH gunmen in Pakistan
in January 2018.
GLOBAL HEALTH
Polio eradication campaign loses ground

Surging cases in Pakistan and Africa have dashed hopes of defeating the virus this year
By Leslie Roberts low, but only about one in every 200 people says virologist Mark Pallansch of the U.S.
infected with the virus develops paralysis, Centers for Disease Control and Prevention
T
he global initiative to eradicate po- meaning thousands have been infected. The in Atlanta, one of the partner agencies in
lio is badly stuck, battling the virus virus is circulating widely: A strain from the initiative. Deadly attacks on polio vac-
on two fronts. New figures show the Karachi, Pakistan, has popped up in Iran. cinators and their police escorts are on the
wild polio virus remains entrenched Adding to the worries, the spike occurred rise. Still, if Pakistan’s government makes
in Afghanistan and in Pakistan, its in the low season, when viral transmission eradication a national priority and puts
other holdout, where cases are surg- subsides—a harbinger of worse to come in money behind it, Sutter says, “There is a
ing. In Africa, meanwhile, the vaccine itself the second half of the year. fighting chance” of success.
is spawning virulent strains. The leaders of Pakistan and Afghanistan are considered In Afghanistan, ongoing conflict has kept
the world’s biggest public health program one epidemiologic block, with the virus vaccinators out of broad swaths of the coun-
are now admitting that success is not just flowing freely across the border. The main try. The Taliban has banned polio vaccina-
around the corner—and intensively debating problem in both countries is that the mas- tion in some places; in others, local leaders
how to break the impasse. sive vaccination campaigns held every few have prohibited door-to-door polio drives.
“The biggest problem for me for a long months are still not reaching every child. One of their concerns is that outsiders
time was recognizing that we truly have In Pakistan, national elections in July might gather information that could enable
a problem, and business as usual will 2018 distracted government officials, and the United States to target drone strikes,
not get us to the finish line,” says Roland the quality of the vaccination campaigns some in the program say. “Polio eradica-
Sutter, who leads polio research at the World slipped. Meanwhile, the poor, neglected tion is collateral damage to the peace talks,”
Health Organization (WHO) in Geneva, communities where the polio virus lurks are says WHO’s Michel Zaffran, who heads the
Switzerland, where the polio eradication ef- increasingly refusing the vaccine. Lacking global initiative. Afghanistan has not yet
fort is based. “The rose-tinted glasses are off,” running water, sanitation, and basic health seen a spike in cases this year, but one is
adds longtime program spokesperson Oliver services, people face far more immediate inevitable if vaccinators can’t reach large
Rosenbauer of WHO. “Now, really tough health threats than the now-slim chance numbers of children, says WHO’s Chris
questions are openly being asked—questions of contracting polio, and they don’t un- Maher, who has been leading eradication
that even 12 months ago no one asked.” derstand why vaccinators arrive with only operations in the region. Still, the program
The program, which has spent $16 bil- polio drops. Since the government began has dealt with conflict before, says Jay
PHOTO: AP PHOTO/ARSHAD BUTT
lion over 30 years, had planned to eradicate to jail parents who do not comply several Wenger of the Bill & Melinda Gates Foun-
polio from Pakistan and Afghanistan this years ago, the opposition has gone under- dation in Seattle, Washington, another part-
year (only the latest of many deadlines). In- ground, with parents hiding their children ner in the eradication effort: “In Syria and
stead, almost four times as many cases have or using fake finger markings to pretend Nigeria we found ways to get good enough
occurred there so far this year than in the they have been vaccinated. vaccine coverage to stop the virus.”
same period in 2018 (see graphic, p. 107). At Exacerbating the situation is a vitriolic In Africa, the wild polio virus appears to
51, the total number of cases may still seem disinformation campaign on social media, be gone, but the vaccine-derived viruses cir-
0712NewsInDepth.indd 106 7/9/19 5:29 PM

NEWS
culating there are just as dangerous. These detected since 2018, type 2 virus has spread SCIENCE FUNDING
strains arise when the weakened live virus from the north to the densely populated port
used in the oral polio vaccine (OPV) mutates
and regains its virulence. In rare instances,
where a population’s immunity is low, they
city of Lagos; it has also entered neighbor-
ing Niger. The Democratic Republic of the
Congo has seen 26 cases. And the situation
California’s
can spread just like the wild virus. Last year,
vaccine-derived viruses paralyzed 105 chil-
is deteriorating, a key WHO committee con-
cluded on 29 May. stem cell
research fund
dren worldwide; the wild virus just 33. Meanwhile, the program has already used
To prevent outbreaks of vaccine-derived nearly 260 million doses of mOPV2. “We are
virus, WHO has declared that once the down to less than 10 million doses for the
wild virus is gone, countries must stop all
use of OPV. As a first step, in April 2016 all
countries switched from the trivalent ver-
whole planet, and that is not enough,” says
Pallansch, who chairs a committee advis-
ing WHO’s director-general on the vaccine’s
dries up
sion of OPV—which covers all three types of use. “No one thought it was possible that we Researchers hope a planned
polio virus—to a bivalent one, which lacks would use that amount.” And as a result of
the type 2 component. (Wild type 2 virus is the 2016 vaccine switch, an increasing num-
ballot initiative will renew
the only one that has been eradicated.) ber of children lack immunity to the type funding in 2020
The program’s scientific advisers knew 2 virus, setting the stage for an explosive
some vaccine-derived type 2 virus would outbreak. That puts the program in a bind.
linger in the first few years after the switch, “We have no choice but to keep using” the By Jocelyn Kaiser
sparking outbreaks. But modeling sug- monovalent vaccine, Zaffran says. “It is all
S
gested the program could quickly squelch we’ve got. We have to live with the risk until tem cell scientists in California who
them—without starting new ones—with the we have a technical solution.” have benefited from a $3 billion state
judicious use of a new live vaccine, mono- Two are on the horizon. A novel OPV2, research agency created in 2004, at
valent OPV2 (mOPV2), which is effective genetically engineered to reduce its chances the height of federal limits on work-
against only type 2. It’s akin to fighting of reversion dramatically, has passed a ing with cells from human embryos,
fire with fire; the gamble was that mOPV2 phase I clinical trial, supported by the Gates have long known that it would even-
would not spawn new outbreaks of its Foundation. “It looks as good as it can now,” tually run out of money. That reality set
own. (An alternative exists, the killed or Wenger says. The earliest it could possibly in last month, when the California Insti-
inactivated polio vaccine, which can’t revert be available for use, however, is 2020. Fur- tute for Regenerative Medicine (CIRM) in
but simply isn’t powerful enough to quash ther out is a new inactivated vaccine power- Oakland announced it is no longer taking
an outbreak.) ful enough to end outbreaks. “The race is grant applications.
The switch worked, except in Africa, where on,” Sutter says, “and it is very hard for me Ongoing payments for approved projects
type 2 vaccine-derived outbreaks have been to predict which will win.” continue, but scientists are already tighten-
more frequent and much harder to stop than Nor will he predict when polio will be gone ing their belts for a funding gap. They are
the models projected; they are now smolder- for good. Optimistic projections, he says, are also contemplating the end of a boom in
ing in seven countries. By using mOPV2, “We “just setting the program up for failure.” j stem cell research in the state. California’s
have now created more new emergences of voters may be asked to renew CIRM with an-
the virus than we have stopped,” Pallansch Leslie Roberts is a former deputy news other bond initiative next year, “but there’s
says. In Nigeria, where 43 cases have been editor at Science. no guarantee,” says Arnold Kriegstein, who
heads a stem cell center at the University
of California (UC), San Francisco, and has
Still no end in sight received CIRM funding in the past.
Longtime CIRM grantee Jeanne Loring,
Polio cases caused by the wild virus, which is still circulating in Afghanistan and Pakistan, are up compared with
the same period last year. Meanwhile, vaccine-derived outbreaks pose a continuing problem, primarily in Africa. who retired in June from the Scripps Re-
search Institute in San Diego, California,
and runs a biotech startup to advance one
Cases: January–10 July 2019 of her projects, says the agency has made
51 25 the state the “center of the stem cell uni-
verse. It would be tragic to unravel [that in-
Cases: January–10 July 2018 frastructure] now. But the funding in 2004
13 28 was so dependent on the politics and inter-
China 1 est at the time, and I don’t know if those
circumstances can be replicated.”
Pakistan 53 That year 59% of California voters ap-
Niger 11 Afghanistan 31 proved CIRM, which had been placed on
Nigeria 43 the ballot as a response to restrictions im-
Somalia 15
Ethiopia 1
posed by then-President George W. Bush’s
MAP: A. CUADRA/SCIENCE
administration on the use of federal fund-

DRC 26 ing for studies of stem cells derived from
Indonesia 1
Angola 2 human embryos. At the time, the National
Mozambique 1 Papua New Guinea 26
Institutes of Health (NIH) could only fund
work on a small number of preexisting hu-
Polio cases (2018 and 2019 combined): Wild Vaccine-derived man embryonic cell lines. (Former Presi-

NEWS | I N D E P T H
dent Barack Obama’s administration later

lifted those restrictions.)
CIRM initially expected to focus on human
embryonic stem cells, but later expanded its
remit to more specialized adult stem cells
such as those that form blood or the increas-
ingly popular induced pluripotent stem cells,
created by reprogramming adult cells to an A large-scale
embryolike state. CIRM’s money led to the solar farm
creation of major stem cell centers in Cali- in Southern
fornia and lured several biotech companies California.
to set up shop in the state. Although CIRM
supported infrastructure, basic research, and
training early on, in the past 3 years it has ENERGY
poured most of its remaining $759 million
into clinical trials—a total of 55 of which are
ongoing or completed to date—as the agency
faced pressure to produce the medical treat-
Solar plus batteries is now
ments its supporters were initially promised.
In a memo to its board released on cheaper than fossil power
20 June, CIRM said it had received applica-
tions totaling $88 million in its latest funding Falling prices help utilities start to decarbonize
call but had only $33 million left to distrib-
ute. The agency announced the next day that By Robert F. Service more than 7000 global storage projects by
it was taking no new grant applications as Bloomberg New Energy Finance reported
T
of 28 June, aside from a sickle cell disease his month, officials in Los Angeles, that the cost of utility-scale lithium-ion bat-
program jointly funded with NIH. “There is California, are expected to approve teries had fallen by 76% since 2012, and by
no money available for new projects,” CIRM a deal that would make solar power 35% in just the past 18 months, to $187 per
communications director Kevin McCormack cheaper than ever while also address- MWh. Another market watch firm, Navigant,
wrote in a 1 July blog post. ing its chief flaw: It works only when predicts a further halving by 2030, to a price
Some researchers who explore the ba- the sun shines. The deal calls for a well below what 8minute has committed to.
sic science of stem cells had already been huge solar farm backed up by one of the Large-scale battery storage generally
looking for other funding sources as CIRM world’s largest batteries. It would provide relies on lithium-ion batteries—scaled-up
began to emphasize clinical work and their 7% of the city’s electricity beginning in 2023 versions of the devices that power laptops
support wound down. But others, espe- at a cost of 1.997 cents per kilowatt hour and most electric vehicles. But Jane Long,
cially those planning clinical trials, will be (kWh) for the solar power and 1.3 cents per an engineer and energy policy expert who
hit hard. “It’s going to be a huge impact on kWh for the battery. That’s cheaper than recently retired from Lawrence Livermore
my lab and many others if they end,” says any power generated with fossil fuel. National Laboratory in California, says bat-
April Pyle of UC Los Angeles (UCLA), whose “Goodnight #naturalgas, goodnight #coal, teries are only part of the energy storage an-
11-person group works on using muscle goodnight #nuclear,” Mark Jacobson, an at- swer, because they typically provide power
stem cells to treat muscular dystrophy. Her mospheric scientist at Stanford University in for only a few hours. “You also need to man-
last CIRM grant ends in March 2020 and Palo Alto, California, tweeted after news of age for long periods of cloudy weather, or
although she also has some NIH funding, the deal surfaced late last month. “Because winter conditions,” she says.
it does not support the animal testing and of growing economies of scale, prices for re- Local commitments to switch to 100%
other studies needed to move her work to- newables and batteries keep coming down,” renewables are also propelling the rush
ward a clinical trial. adds Jacobson, who has advised countries toward grid-scale batteries. By Jacobson’s
Future clinical work will face “at best sig- around the world on how to shift to 100% count, 54 countries and eight U.S. states
nificant delays, and many projects to iden- renewable electricity. As if on cue, last week have required a transition to 100% renew-
tify new therapies will stop” if the agency a major U.S. coal company—West Virginia– able electricity. In 2010, California passed
doesn’t continue, says gene therapy re- based Revelation Energy LLC—filed for a mandate that the state’s utilities install
searcher Donald Kohn, who heads several bankruptcy, the second in as many weeks. electricity storage equivalent to 2% of their
such trials at UCLA. The new solar plus storage effort will peak electricity demand by 2024.
CIRM’s efforts to raise $200 million in be built in Kern County in California by Although the Los Angeles project may
bridge funding from private sources have 8minute Solar Energy. The project is ex- seem cheap, the costs of a fully renewable–
been unsuccessful to date. Now, CIRM boost- pected to create a 400-megawatt solar array, powered grid would add up. Last month,
ers are looking to a $5.5 billion bond initiative generating roughly 876,000 megawatt hours the energy research firm Wood Mackenzie
that real estate developer Robert Klein, who (MWh) of electricity annually, enough to estimated the cost to decarbonize the U.S.
PHOTO: 8MINUTE SOLAR ENERGY
led the original push to create the agency, power more than 65,000 homes during day- grid alone would be $4.5 trillion, about half
hopes to add to the November 2020 ballot. light hours. Its 800-MWh battery will store of which would go to installing 900 billion
If approved, “We would hope there would electricity for after the sun sets, reducing watts, or 900 gigawatts (GW), of batteries and
be very little gap” in funding, McCormack the need for natural gas–fired generators. other energy storage technologies. (Today,
says. But if the voters reject the initiative, Precipitous price declines have already the world’s battery storage capacity is just
he expects CIRM’s staff to dwindle and the driven a shift toward renewables backed 5.5 GW.) But as other cities follow the exam-
agency to fold by about 2023. j by battery storage. In March, an analysis of ple of Los Angeles, that figure is sure to fall. j

BIOMEDICINE
Gut microbes may help malnourished children

Commonplace foods restore bacteria that can improve health
By Elizabeth Pennisi maturation, Arjun Raman, a postdoc in find foods that could “encourage catchup
the Gordon lab, analyzed data from fecal growth in the microbiota,” Gordon says. Milk
E
ven after starving children get enough samples that Ahmed’s team had collected powder and rice, standard components of
to eat again, they often fail to grow. monthly from 50 healthy infants in Bangla- food aid, did little to foster maturation, but
Their brains don’t develop properly, desh as they grew. Raman identified 15 types chickpea, banana, and soy and peanut flours
and they remain susceptible to dis- of bacteria that increased and decreased in helped the microbiomes mature.
eases, even many years later. Two concert as the microbiome matured. He and The researchers then fed mice and piglets
studies, on pp. 139 and 140, now sug- the team saw the same pattern in healthy supplements that combined all four foods
gest fostering the right gut microbes may children from Peru and India and even in and saw that the animals’ microbiomes ma-
help these children recover. The work also germ-free piglets given the bacteria and tured, and their growth improved. “This
pinpoints combinations of foods that nur- eating the same food as the Bangladeshi study points to the importance and utility
ture the beneficial microbes. infants and children, they report on p. 140. of thoughtfully selected nutrients to support
Most of the experiments were in animals, Fostering or suppressing those microbes key members of a microbiota,” Relman says.
but a small group of malnourished children could be key to helping children recover As a final proof of principle, Ahmed,
given those foods also showed signs of Gordon, and their colleagues com-
improvement. “This is an outstanding pared their supplements to standard
and extremely comprehensive study,” recovery fare in about 60 malnourished
says Honorine Ward, a microbiologist Bangladeshi children for 1 month. That
and global health expert at Tufts Uni- was too little time to assess long-term
versity School of Medicine in Boston. physical recovery, but long enough to
Tailoring food aid to foster the micro- see effects on the molecular signatures
biome “could be a key to new strategies in the blood. Only the four-food com-
for improving global public health and bination sharply improved those sig-
human potential,” adds David Relman, natures, they report on p. 139. It also
a microbiologist at Stanford University improved the 15 bacteria Raman’s team
in Palo Alto, California. linked to maturation.
Tahmeed Ahmed, director of nutri- The results suggest a way to improve
tion research at the International Cen- nutrition even in well-fed children, in
tre for Diarrhoeal Disease Research, whom a poor diet can lead to an in-
Bangladesh, in Dhaka, has tried for creased risk of obesity, diabetes, and
30 years to help malnourished children other diseases as adults. “If care-givers
recover better. About a decade ago, he know the best foods to feed their chil-
was intrigued by work by Jeffrey Gordon dren to support the development of their
at Washington University in St. Louis, gut microbiota, we could potentially
Missouri, linking certain gut microbes prevent undernutrition and immature
to obesity. The two scientists wondered microbiota to begin with,” Gehrig says.
whether the microbiome—the set of Relman and others are more cau-
microbes living in and on the human tious, particularly because it’s not clear
PHOTO: INTERNATIONAL CENTRE FOR DIARRHOEAL DISEASE RESEARCH, BANGLADESH
body—might also play a role in obesity’s the mature microbiomes will persist
opposite number, malnutrition. in malnourished infants who ate the
Together, their teams reported in A malnourished child in Bangladesh will get special food supplements. Chris Damman, a senior
2014 that the gut microbiome nor- supplements to help recover. program officer at the Bill & Melinda
mally “matures” as an infant grows Gates Foundation in Seattle, Wash-
into a toddler. They also noticed that it re- from malnutrition, the researchers thought. ington, which funded the work, says, “I
mains immature in severely malnourished To test that idea, they needed a way to moni- wouldn’t say this is ‘the future’ or ‘the cure-
children, dominated by bacteria found in tor recovery. Looking for a molecular signa- all.’ But it’s certainly an exciting lead and
younger healthy children. Two years later, ture of healthy growth, a graduate student one that we think has promise.”
the researchers put mature and immature in Gordon’s lab, Jeanette Gehrig, and col- Back in Bangladesh, Ahmed is coordinat-
microbiomes from children into mice raised leagues monitored more than 1000 proteins ing an effort to provide the microbe-boosting
without microbes. Animals given the im- and metabolites, such as fatty and amino diet and others to a larger group of malnour-
mature microbiomes put on less muscle, acids, for ones that change as healthy chil- ished children. The team will follow the
had weaker bones, and had impaired me- dren grow and malnourished ones recover. children for 3 months, long enough to see
tabolisms, suggesting a mature microbiome Gehrig and the rest of the team then tested whether the beneficial effects translate into
might be needed for proper development. the effects of various foods on mice and piglets healthy development. He’s hopeful, he says,
To pinpoint which among the micro- whose microbiomes had been transplanted that “this can be a game-changer in the treat-
biome’s hundreds of strains are linked to from malnourished infants. They hoped to ment of malnutrition.” j

NEWS | I N D E P T H
PSYCHOLOGY
Psychologist aims to study

diverse minds, not WEIRDos
Daniel Haun wants his fellow psychologists to study
people outside of rich, Western societies
By Kai Kupferschmidt, in Leipzig, Germany
Nine years ago, a research team argued that psychology studies relied to an alarming rate
on people from WEIRD societies: Western, educated, industrialized, rich, and democratic.
But research was already showing that their members—WEIRDos, as they were dubbed—
are outliers in some traits, tending to view themselves positively and to be susceptible to
certain optical illusions. WEIRDos “may represent the worst population on which to base
our understanding of Homo sapiens,” the authors wrote in Behavioral and Brain Sciences.
In 2017, psychologist Daniel Haun reported that little had changed: In 2015, about 92%
of all papers in his specialty of developmental psychology featured participants from
English-speaking countries and non–English-speaking Europe—a percentage matching
that of 2008. As the new director of the comparative cultural psychology department at A #Akhoe Hai//om man from Namibia can more
the Max Planck Institute for Evolutionary Anthropology here, Haun hopes to help change accurately size up circles than most people from
that. This interview has been edited for brevity and clarity. Western countries.
Q: How do other societies differ from of the time, no matter who contributed Gray from the Max Planck Institute for
WEIRD ones? how much. Everybody has this emotional the Science of Human History in Jena,
A: In so many ways. For instance, some gut response to being treated “unfairly.” [Germany,] has set up a research project on
share more. In the #Akhoe Hai//om But depending on where you grew up, the Vanuatu, a southern Pacific Island state with
community in Namibia, who were hunter- gut feeling you develop might be com- only about a quarter of a million inhabitants.
gatherers until three generations ago, pletely different. But they speak more than 100 languages.
everything that is shareable in principle So it’s an incredible place to look at the cog-
belongs as much to you as it belongs to Q: Why do psychologists still study nitive mechanisms behind cultural diversity.
me. I could tell you to give me “my” shoes, mostly WEIRDos? We plan to join forces with him there.
and the fact that you’re currently wearing A: Many still assume that what we find
them does not matter. The natural conse- in one population can be generalized Q: Should there be a coordinated effort
quence is that everybody has about similar to others, so why study other cultures? to replicate some of the big findings of
amounts of everything. Cross-cultural work is difficult, tedious, psychology in diverse societies?
We also find differences in the perception and slow. In an environment where publi- A: Yes. Psychology is going through a very
of geometric objects. For example, in the cation frequency is rewarded, this is a very interesting transition, where psychologists
Ebbinghaus illusion, a circle surrounded difficult strategy. started noticing that a lot of published
by much bigger circles would look smaller material isn’t easily replicated. And even
to you than a circle of the same size sur- Q: How do you plan to change this? if a study is perfectly replicable within one
PHOTO: DANIEL HAUN/MAX PLANCK INSTITUTE FOR EVOLUTIONARY ANTHROPOLOGY

rounded by much smaller circles. The A: One of the great things [at Max Planck] is community, there is added value in trying
#Akhoe Hai//om fall for that illusion much that there is a long-term funding perspec- to replicate it across communities. As
less than German participants do. tive, which means that we can try and traditional ways of life disappear at an accel-
invest in a network of cross-cultural field erating pace, there is a very precious time
Q: What might psychology be getting wrong sites. [We will] set up long-term infrastruc- window now to study this diversity.
by studying only WEIRDos? ture with local research assistants and
A: We assume that concepts true in our researchers as well as scientists from here, Q: Neo-Nazis and other racists still use
own cultural context are true generally. who collect data year-round. The idea is to cross-cultural differences to argue that
For example, Marie Schaefer, a former create functioning developmental psycho- certain groups are superior. How do you
member of my group, led a study on fair- logy research stations in multiple places avoid this danger?
ness norms. Let’s say a friend and I go to around the world. A: By confronting it head on. Scientists can
the beach and look for shells. I spend a lot be careful in interpreting their data and
of time searching and find a lot. He spends Q: Where? engage in the debate. I don’t think racism
his time laying on the beach. If we divide A: I am planning five stations, and we have goes away if we avoid the fact that there
up the shells, I get a lot more than he will, two as a starting point. [One is] in northern is variation as well as similarity across
and that’s fair, right? That’s what German Namibia with the #Akhoe Hai//om, with humans. And the drivers of variation might
children mostly do. But #Akhoe Hai//om whom we have worked for more than give us some answers about fundamental
children distribute the goods equally most a decade now. And a colleague, Russell questions of who we are and how we work. j

PALEOANTHROPOLOGY
Was our species in Europe 210,000 years ago?

Skepticism greets startling conclusion from skull fragment found in Greek cave
By Lizzie Wade was likely their original shape. The result serves enough of the cranium to demonstrate
showed the face of a typical Neanderthal, jut- that it is definitively Homo sapiens,” Delson
I
n the late 1970s, anthropologists ex- ting from the skull and complete with pro- says. But not everyone agrees. “It’s plausible,”
ploring a cave on the rugged coast of truding brow ridges. The ratio of uranium to says Susan Antón, a paleoanthropologist at
southern Greece found two mysterious its decay products in the bones revealed an New York University in New York City. “But
hominin skull fossils. Time had left age of about 170,000 years old. for me it’s not a slam dunk.”
them fragmented and distorted, and the For Apidima 1, Harvati and her team cre- In ancient humans, the shape of the back
jumbled stratigraphy of the cave made ated a mirror image of the fossil and stitched of the skull doesn’t always predict the shape
them hard to date. For decades, the fossils the two together to see the full shape of the of the face, she says. The Jebel Irhoud skull,
sat on a shelf, their identity unknown. Now, back of the skull. It was short and round, for example, has an archaic, elongated back
a state-of-the-art analysis of their shape like the skulls of H. sapiens, and lacked a but a distinctly modern face. Zollikofer adds
together with new dates suggest one skull ridge and furrow that Neanderthal skulls that the Neanderthal lineage may encom-
might represent our own species, living in typically have at the back. “You couldn’t pass more anatomical variations than re-
Greece more than 200,000 years ago. The bend [the Apidima 1 reconstruction] into searchers yet realize—perhaps including a
findings, reported in Nature this week, a classic Neanderthal cranium,” agrees short, round skull. “It highlights the scarcity
would make this the oldest known Homo Christoph Zollikofer, a paleoanthropolo- of our knowledge,” he says. In fact, Marie-
sapiens fossil found in Europe, by at least gist at the University of Zurich in Switzer- Antoinette de Lumley, a paleoanthropo-
150,000 years. land who wasn’t involved in the research. logist at CNRS, the French national research
If so, H. sapiens’s first forays out of its Harvati and her team concluded that the agency in Paris, has recently argued that both
African cradle likely happened earlier and skull most likely belonged to H. sapiens. skulls are actually ancestors of Neanderthals.
extended much farther than most paleo-
anthropologists thought, into territory dom-
inated by Neanderthals, our extinct cousins.
“And then [H. sapiens] disappeared” from
Europe, says Eric Delson, a paleoanthropo-
logist at the City University of New York in
New York City, until a later wave success-
fully spread across the continent about
50,000 years ago. But because the evidence
is no more than a piece from the back of the
skull, some researchers aren’t sure the fossil
can be definitively identified as H. sapiens.
And others question the old date. Investigators scanned a fragment from the back of an ancient hominin’s skull (right) and digitally reconstructed
Katerina Harvati, a paleoanthropologist it (left and center), revealing the rounded skull of Homo sapiens, rather than an elongated Neanderthal skull.
at the University of Tübingen in Germany,
has long suspected that southeast Europe The finding startled Harvati herself— Israel Hershkovitz, a paleoanthropologist
was a hot spot for ancient humans. Not especially because the uranium dating of Api- at Tel Aviv University in Israel who found the
only is the region “at the crossroads of three dima 1 put its age at 210,000 years old. That fossils in Misliya Cave, thinks that because
continents”—Africa, Asia, and Europe—but makes it at least 15,000 years older than the members of H. sapiens were in the Middle
IMAGES: KATERINA HARVATI AND EBERHARD KARLS/UNIVERSITY OF TÜBINGEN
it enjoyed a relatively mild climate when next oldest fossil of our species found outside East about 200,000 years ago, they could
other parts of Europe were covered by gla- of Africa, in Misliya Cave in Israel (Science, have made an early excursion to southern
ciers, she says. So she was thrilled to receive 26 January 2018, p. 456). It’s about 100,000 Europe, too. Harvati points out that some
permission to study the fossils, which are years younger than the oldest known H. sa- Neanderthal genomes preserve a trace of an
named for the cave. The first individual, piens fossils in the world, from Jebel Irhoud interbreeding event with H. sapiens that took
Apidima 1, is represented by the skull piece. in Morocco. But Warren Sharp, a uranium place before 200,000 years ago, a sign that
The second, Apidima 2, is more complete dating expert at the University of California, our ancestors must have entered Neander-
and includes the face. Berkeley, points out that the Apidima 1 sam- thal territory early, before vanishing again.
The Apidima 1 skull fragment was more ples actually returned dates ranging from Perhaps “they didn’t like the climate, or
complete on one side than the other, and more than 300,000 years old to less than didn’t like the fauna to eat, or didn’t like hav-
Apidima 2’s skull and face were distorted. 40,000 years old. “It’s not a well-behaved ing Neanderthals around, and pulled back,”
So Harvati began by figuring out what they sample,” he says. “You have this huge spread Delson says.
originally looked like. She and her team of apparent ages, and you don’t know if any But Hershkovitz isn’t convinced Apidima
scanned both fossils with x-rays and created of them are any good.” 1 represents those ancient pioneers. Without
3D reconstructions. They digitally broke Researchers are also divided on whether more complete fossils and confirmation of
Apidima 2 into 66 bone fragments and Apidima 1 convincingly represents a member their dates by other techniques, he says, “The
painstakingly reassembled them into what of our species. “[Apidima 1] pretty clearly pre- evidence is very weak.” j

FEATURES
MAKING PEACE WITH OIL PALM

Some scientists are helping make palm plantations greener.
Others say that doing so legitimizes a destructive industry
By Dyna Rochmyaningsih, in Libo on Sumatra, Indonesia
C
rickets were chirping one clear important pests of oil palm trees. He soon only in a small circle around each tree. As a
morning in April when Anak Agung found more insect killers in the palm grove: result, many tall ferns and shrubs were grow-
Aryawan walked under the canopy a Nephila spider, known for its big, elaborate ing beneath the canopy, creating a home for
of a quarter-century-old oil palm web, and the bright yellow Cosmolestes, an- insects, spiders, and snakes.
plantation here. Suddenly Agung, an other species of assassin bug. Many Indonesian planters would abhor PHOTO: ARROWHEAD FILMS
agroecologist, stopped. “Look, that’s Agung works for SMARTRI, an oil palm this semiwilderness, worrying the under-
a Sycanus!” He pointed at a black in- research institute here owned by Sinar Mas, story would compete with oil palm trees for
sect perched on a fern in the forest one of Indonesia’s largest business conglom- water and nutrients. Agung sees it differently.
understory. Known as an assassin bug, Syca- erates. The study plot he was visiting was Allowing a luxuriant understory to grow in
nus uses its mouthpart to stab its insect prey, managed without herbicides or insecticides; plantations can protect insects and some
including the fire caterpillar, one of the most plantation workers weeded it by hand, and small mammals, such as the leopard cat—and

NEWS
On a plantation in West Kalimantan in Indonesia, oil Policies to stem the tide have not worked
palms have replaced all but small patches of forest. very well. In 2011, the Indonesian govern-
ment issued a moratorium on deforesta-
ing yield. Finding a way to protect species tion, and in 2016 it halted the draining and
while satisfying the world’s demand for clearing of peatlands for plantations. In Sep-
palm oil is “a vital conservation priority of tember 2018, President Joko Widodo also
the modern era,” Edgar Turner, a conserva- stopped issuing new oil palm permits, which
tion scientist at the University of Cambridge for now has stalled deforestation in the prov-
in the United Kingdom who heads BEFTA, ince of Papua—another biodiversity mecca—
wrote in a 2011 paper. where 1800 km2 have been cleared so far.
Some critics call the approach naïve. By ac- Yet banning palm oil would not end bio-
cepting industry funding—and using its giant diversity loss, according to a 2018 report by
plantations as laboratories—scientists risk the International Union for Conservation
losing their independence, they say, and they of Nature (IUCN); it would only displace
legitimize the companies’ business by giv- it to other parts of the globe and possibly
ing it a veneer of sustainability. “They take a worsen it. One hectare of tropical land can
pragmatic approach in the face of a desperate produce 4 tons of oil annually, at least four
situation,” says David Gaveau of the Center times the yield of 1 hectare of rapeseed, sun-
for International Forestry Research in Bogor, flowers, or soybeans planted in temperate
Indonesia, a vocal critic of the oil palm indus- regions. Unlike those crops, the oil palm is
try. “If funding and prestige that comes with a tree that can live up to 25 years—enough
access to large data sets lures scientists to a for a diverse ecosystem to thrive in a planta-
particular direction while ignoring the big el- tion, if growers allow it. “It’s actually a good
ephant in the room, then this is problematic,” crop for conservation, but it just happens
says Maria Brockhaus, a forest politics expert to grow in the most biodiverse parts of the
at the University of Helsinki. world,” Turner says.
But scientists working with oil palm com- Instead of urging a ban, the IUCN report
panies say they don’t feel constrained scientif- calls for reining in deforestation and dis-
ically, and they welcome the money. “It’s hard couraging the use of unsustainable palm
to find long-term funding for research,” says oil. (Some 19% of the global output is certi-
Matthew Struebig, a conservation scientist at fied as “sustainable” by the Roundtable on
the University of Kent in the United Kingdom Sustainable Palm Oil, which includes thou-
who has consulted for two plantations owned sands of growers, traders, and manufactur-
by Wilmar International, the world’s largest ing companies as well as groups such as the
palm oil trader. Moreover, the demand for World Wildlife Fund and Oxfam. To qualify,
vegetable oil will only grow, and palm oil is a company needs to show it’s not contribut-
the most efficient way to produce it, Turner ing to deforestation and is treating its work-
says. Biodiversity loss is a “complete tragedy,” ers well, among other things.)
but “we need to feed the world,” he says. “We One of the lead authors of the IUCN report
need to produce crops that are very produc- is Dutch ecologist Erik Meijaard, who runs
ultimately benefit the oil palm trees as well. tive … in the smallest area possible. And oil his own consultancy company from Brunei.
Sycanus and other predators control pests, palm is the best.” Meijaard is a prominent figure in Indonesian
for example, and other invertebrates improve conservation science. In 1997, he discovered
the soil and pollinate the palms. PALM OIL IS USED in a staggering number of an orangutan population in a small patch
Oil palm (Elaeis guineensis) is one of the consumer products, from fast food, chocolate of forest in North Sumatra; a decade later,
most controversial crops today, because the spread, and cereals to toothpaste and dog he and others found that genetic and mor-
plantations often replace tropical rainforests chow. It is also a source of biodiesel. Some phological characteristics set it apart as a
rich in biodiversity, depriving iconic species 90% of the global supply comes from Indo- separate species, now named the Tapanuli
such as the orangutan of their habitats. Vast nesia and Malaysia, where plantations cover orangutan, after the place they inhabit.
swaths of Indonesia and Malaysia are given 17 million hectares, almost half the area of Meijaard feels palm oil is unjustly vilified.
over to the crop. But Agung and a grow- Germany. Growing demand is pushing the There are many other threats to Indonesia’s
ing number of other scientists say it’s time industry into Africa and South America. biodiversity, he notes. “We always assume
to work with oil palm companies—some Gaveau, a landscape ecologist, has tracked that the forest would have remained if oil
of them in the crosshairs of environmental the spread of plantations on Borneo—home palm plantations had not been developed,
activists—to make the best of a bad situation. of Indonesia’s largest rainforest—in images but my lesson from Indonesia is that forests
Researchers have accepted industry from NASA’s Landsat and data on oil palm that are unmanaged are ultimately cut down,
funding to study habitat fragmentation and concessions to produce the Borneo Atlas, a either legally or illegally.”
advised oil palm companies on how to best new online platform. In the past 2 decades, he Meijaard is pragmatic when it comes to
manage the surviving wildlife in their con- found, big oil palm companies have cleared collaborating with palm oil companies. In
cessions. And at SMARTRI, a long-term eco- 24,000 square kilometers (km2) of Borneo’s 2011, he decided to accept an offer to ad-
logical experiment called Biodiversity and forests—almost five times the area of Bali. vise ANJ Agri on how to manage a patch of
Ecosystem Function in Tropical Agriculture (Pulpwood production, smallholder farming, forest within the company’s concession in
(BEFTA) is testing whether the plantations mining, dam construction, and other devel- West Kalimantan—an Indonesian province
can host more biodiversity without affect- opment consumed another 36,000 km2.) on Borneo—that is home to 150 orangutans.

NEWS | F E AT U R E S
That raised eyebrows among some other like a community forest located nearby.” SMARTRI IS RUN BY Jean-Pierre Caliman, a
conservationists. According to Gaveau’s Meijaard stresses, “There are plenty of French agronomist with a passion for oil
Borneo Atlas, the company cleared 38 km2 companies I wouldn’t take money from.” palms who moved here from Africa in 1993.
of primary orangutan habitat in West Kali- When he was invited in 2018 for a “dialogue” Employed by Sinar Mas, he leads a team of
mantan in 2012, a year after the Indonesian by PT North Sumatera Hydro Energy, a 81 Indonesian scientists and has a $10 mil-
moratorium on deforestation began. And company building a hydroelectric dam that lion annual budget. Until recently, its the
Greenpeace included ANJ Agri on a black- threatens the home of the Tapanuli orang- institute’s research focused on increasing
list for clearing forests without permission utan, his answer was resolute. “There is yield or reducing cost; biodiversity wasn’t
from local indigenous people in South So- nothing to discuss or have a dialogue over, really on their radar. Now, SMARTRI scien-
rong in West Papua; it also says the com- apart from the total cancellation of the proj- tists are studying the carbon dynamics in
pany’s private police beat a Papuan man ect,” he wrote to the company. plantations and putting a price tag on the
during a 2017 demonstration. Yet Gaveau says it’s hard for scientists ecosystem services provided by hundreds
Meijaard says that in Kalimantan, the to know exactly what the companies they of species living among the oil palms. “How
company only cleared second-growth for- work with are doing. In 2018, for instance, much money do we have to spend if a spe-
ests, which have less biodiversity than pri- Greenpeace accused Wilmar of disguising its cies disappears?” is a key question now,
mary rainforest; he declined to comment de facto ownership of plantations to avoid Caliman says.
on the South Sorong accusations. (An ANJ accountability and buying palm oil from The Sumatran barn owl (Tyto alba), for
Agri spokesperson in Jakarta says an in- 18 companies that had cleared forests, de- instance, is a scourge of the rats that eat the
depth investigation showed the beating spite Wilmar’s 2013 adoption of a “no defor- fruits of the oil palm tree, lowering yield.
didn’t happen.) In any case, Meijaard says, estation” policy. (In a press release, Wilmar Without owls, Caliman says, plantation
the company is serious about protecting the denied some of the allegations, but the com- managers would need to buy rodenticides
orangutans in its concession. “They worked pany did adopt a plan to better monitor its worth up to $4 per hectare annually; to lure
hard in getting rid of rampant illegal logging suppliers 3 months later.) “We cannot trust the birds, Sinar Mas has installed 26,000 ar-
and hunting, and invested heavily in forest the companies blindly,” Gaveau says. “They tificial nest boxes on Sumatra.
fire prevention,” he says. “Without the com- will seek loopholes to cheat the system for SMARTRI’s major effort to find a place
pany, that piece of forest would be gone, just their advantage whenever they can.” for biodiversity in palm plantations began in
2011, when William Foster, a Cambridge in-
sect ecologist and Turner’s Ph.D. supervisor,
A more natural plantation asked Caliman whether he wanted to collabo-
A large-scale ecological experiment on a plantation in Indonesia tested three different understory treatments. rate on a long-term ecological study. Caliman
It suggests reducing herbicide use can lead to a more diverse understory without affecting yield. embraced the idea, which became BEFTA.
Turner got some of his ideas for BEFTA while
Eliminating the understory Normal understory complexity doing research in Sabah in Malaysia. “From
All understory vegetation is removed using herbicides— The standard practice at this plantation: Herbicides walking around plantations [there], it’s quite
a practice used by many growers, who worry the plants are sprayed in a circle around each tree and on paths, clear that plantations with a higher level of
will compete for water and nutrients with oil palm trees. and some woody vegetation is removed manually.
understory have a high level of biodiversity,
and we would like to find out what impact
that had on functioning and yield,” he says.
Sinar Mas provides BEFTA with funding—
Turner and Caliman declined to say how
much—and Turner was given 18 research
plots, each measuring 150 by 150 meters, at
the plantation. On six of them, researchers
used herbicides to remove all of the under-
story, as well as ferns living on palm trees,
which many plantation owners do. In six
Enhanced understory complexity others, they used standard Sinar Mas prac-
No herbicides are used. Plants near the trees are removed by hand, leaving luxuriant tice, which is to spray herbicides only on
understory where invertebrates thrive, controlling pests. paths and in a circle around each tree to
1 Negligible give plantation workers access, while leaving
impact on yield most of the understory alone. On the last six
2 Increased soil plots—including the one where Agung found
biodiversity his Sycanus—they used no herbicides at all;
3 No decline workers manually removed plants around
in soil fertility 1 the trees and on paths.
4 Pesticide Data collection finished last year, and
cost down, 5 some results have come out. In a paper pub-
labor cost up lished in December 2018, the team reported GRAPHIC: V. ALTOUNIAN/SCIENCE
5 Nest boxes that abandoning herbicides improved the

attract barn condition of the soil and increased the diver-
owls, which sity of soil macrofauna, such as earwigs (Der-
prey on rats. 4
maptera) and millipedes (Diplopoda), which
2 break down leaf litter, making nutrients
3 available to other species. Another paper re-

Anak Agung Aryawan inspects a tree planted in 2018 as part of a study seeking to restore riverbanks in plantations to a more natural state.
ported that the plantation is home to 69 spe- The Indonesian government, not industry, AFTER A LONG DRIVE through the planta-
cies of dragonfly, including five never before should fund oil palm research, Brockhaus tion, Agung got out of his Land Cruiser and
spotted on Sumatra. And the herbicide-free says: “The country has the responsibility to walked toward a narrow stream. The area
plots had soil nutrient levels just as high as ensure independent and critical research looked very different; there were no palm
the chemically treated ones, suggesting wor- to serve the interest of the wider society trees in sight, only young forest trees and
ries about competition from the understory and not in favor of selected interests.” The wild shrubs. They formed part of a new long-
are unfounded. As-yet-unpublished findings government could also study social and eco- term experiment by the Cambridge team,
show that even the most ecofriendly regimen nomic aspects of the industry, adds Hariadi launched in 2018 and called Riparian Ecosys-
had a negligible impact on yields, Caliman Kartodihardjo, a forest policy expert at Bo- tem Restoration in Tropical Agriculture.
says. He and Turner are optimistic they can gor Agricultural University. Revenues pri- Under a 2015 government policy, compa-
persuade Sinar Mas management to adopt marily benefit the country’s ruling elite and nies can’t plant new oil palms in 50-meter-
the strategy widely. a small number of tycoons, Kartodihardjo wide ribbons along rivers in their plantations;
says; meanwhile, millions of plantation the idea is to start to give these zones back to
ALTHOUGH ENVIRONMENTALISTS are wary of workers toil, often on low wages, in isolated nature. How best to restore them to a more
such collaborations, some oil palm experts places with conflicts over land use and few natural state is not clear. The team is now
dismiss the results. Agus Eko Prasetyo, an educational opportunities. “This is some- testing four different strategies. In one plot,
expert on plant protection at the Indone- thing that needs to be solved,” he says. all palm trees were cleared and replaced with
sian Oil Palm Research Institute in Medan, But government funding for research is six native tree species. Some were struggling.
says scientists have known since the 1980s scarce; many Indonesian scientists can only A young red meranti (Shorea leprosula), an
that most ferns and shrubs don’t decrease dream of the budget SMARTRI has. The In- icon of the lowland tropical forest on Suma-
palm oil yields. But he says planters do need donesian Oil Palm Estate Fund, a govern- tra and in Kalimantan, was dying.
to control certain species—especially woody ment body that collects taxes on palm oil But Agung’s face lit up in another plot,
plants whose roots suck up more water and exports, also funds some research, but most where the researchers had left the oil palm
nutrients—and doing so manually rather of it is in agronomy and postharvesting pro- trees in place and had planted native trees
than with herbicides will drive up cost. “I cessing, not on biodiversity or social and between them. Here, a red meranti was thriv-
bet Sinar Mas won’t adopt” the ecofriendly economic issues. ing; maybe it needed to be shaded by the
regime, Prasetyo says. Turner has no qualms about the collabo- aging palm trees early in life, Agung specu-
Others say that in working with large ration with Sinar Mas: “For large experi- lated. Another native forest tree, Peronema
companies, ecologists are missing worse mental trials, you need a lot of resources,” canescens, which the team had planted last
PHOTO: DEDY SUTISNA/RIAU IMAGES
offenders: smallholder farmers who own he says. Struebig sees the overall balance as year, was already taller than Agung. He
40% of Indonesia’s oil palm plantations positive as well. Working with companies stood next to it to take a selfie, ignoring
and may be less informed about biodiver- gives researchers access to sites and data a weaver ant crawling on his neck. “I just
sity and oil palm management. If research- and helps build trust between science and can’t wait to see the plot in a few years,”
ers “focus on large companies to ‘help’ the industry, he says. “To me, working with he said. “It’s going to look like a forest.” j
them do a little bit less harm to nature, the industry will lead to bigger improve-
then who is going to study the unsustain- ments in sustainability than working with- Dyna Rochmyaningsih is a journalist
able practices?” Brockhaus asks. out,” he says. based in Deli Serdang, Indonesia.

INSIGHTS
PERSPECTIVES
MARINE ECOLOGY
Seabird clues to ecosystem health

Seabird monitoring provides essential information on the state of marine ecosystems
By Enriqueta Velarde1, Daniel W. These insights provide valuable information a failed breeding in years of low food avail-
Anderson2, Exequiel Ezcurra3 for sustainable ecosystem management. Co- ability has a smaller negative impact on their
ordinated efforts to gather standardized sea- overall fitness (4).
F
or millennia, humans have used seabird bird data will be essential for monitoring the Seabirds are sentinels in two ways. First,
sightings and behavior as indicators of health of the global ocean. they can serve as biomonitors of ecosystem-
conditions of the marine environment. Seabirds have evolved a set of adaptations scale changes, such as the presence of or-
Seabirds are highly visible and nest in to their environment, such as being able to ganic pollutants or heavy metals in their
large colonies in normally constant loca- fly long distances, locate and capture prey tissues and marine litter such as plastics
tions, allowing efficient data gathering. underwater, and find nesting sites safe from and microplastics in their stomachs (5).
Different species can provide information on predators and near highly productive marine Second, they can be quantitative indicators
different parts of the food chain. Individuals regions, where they obtain food. Seabirds of ecosystem components such as fish abun-
are often easier to observe and capture than have comparatively long life spans (20 to 60+ dance. Many instances of seabird breeding
other marine organisms, allowing behavioral, years), reach sexual maturity late (between 2 failures or population declines have pre-
anatomical, physiological, demographic, and and 10 years of age), reproduce annually, and saged fisheries collapses and, through their
genetic information to be gathered (1). In re- have small clutch sizes (one to three eggs) and diet composition and distribution shifts,
cent decades, seabird breeding and feeding long periods of chick development (50 to 350 they provide reliable signals of many fish PHOTO: OCTAVIO ABURTO/ILCP
observations have revealed the connection days) (4). These traits likely evolved because
between sea surface temperatures in upwell- of the large effort required in delivering food 1
Instituto de Ciencias Marinas y Pesquerías, Universidad
ing regions and seabird reproductive success, to offspring from the open ocean, which can Veracruzana, Boca del Río, Veracruz 94290, México.
2
as well as the frequency with which warm negatively affect the chances of adults’ sur- Department of Wildlife, Fish, and Conservation Biology,
University of California, Davis, CA 95616, USA. 3Department of
oceanographic anomalies occur and the evo- vival to the next reproduction. Because sea- Botany and Plant Sciences, University of California, Riverside,
lution of seabird life history strategies (2, 3). birds are long-lived and reproduce annually, CA 92521, USA. Email: enriqueta_velarde@yahoo.com.mx
0712Perspectives.indd 116 7/3/19 2:21 PM

Elegant terns (Thalasseus elegans) are precise tive proposed for better ocean management, plankton diversity, and seabird parameters,
predictors of the fishery catch per unit effort of requires deep knowledge of complex ocean are in development (13), but internationally
the Pacific sardine in the Gulf of California. dynamics, including the impact of fishing coordinated efforts to gather and use such
activity on ecosystem processes, to attain data are in their infancy. Such coordinated
stocks (1, 2, 6). These correlations are used sustainable stocks. Unfortunately, this ap- efforts would also help in the development
to understand the dynamics of the ocean proach is difficult to implement in ecosys- of ocean observation programs, which can
environment (6). Excessive fishing can also tems with limited data, as is the case in most use seabird data and oceanographic variables
contribute to seabird distribution shifts and of the world’s oceans. The use of indicator to generate ocean health standards and best
population decline when seabirds and fish- species such as seabirds that react quickly to practices (11, 13, 14). For example, Moore and
ers compete for the same resource (1, 7). ecosystem changes and provide early warn- Kuletz (11) have shown how marine bird and
Seabird diet reflects the abundance of prey ings of unsustainable stock harvest may be mammal research can be included in proto-
species within their foraging range (usually more appropriate (1, 6). cols of ocean observatories to develop marine
a radius of hundreds of kilometers). This al- Another application of seabirds as senti- ecosystem models.
lows monitoring of the fish populations on nels is to inform about the effects of grow- The Oslo and Paris Convention in the
which they feed, which may be relevant to ing variability in ocean conditions due to North Sea uses the northern fulmar (Ful-
fisheries. However, the relationship between climate change, which causes changes in the marus glacialis) as an indicator species for
seabird diet and prey abundance is often abundance and distribution of forage fish (9). acceptable ecological quality. The maximum
nonlinear and threshold-driven, and the Seabirds are used directly to inform on the acceptable level is that fewer than 10% of
functional response of seabirds’ food choice state of fish stocks (1, 6), or indirectly through fulmars may have more than 0.1 g of plastic
to fluctuations in prey abundance must be changes in critical habitat linked with in their stomachs. Currently, more than 60%
known. For example, three seabird species in changes in marine productivity and interac- of fulmars exceed this amount, and only re-
the Gulf of California track sardine populations in the food web. These studies reveal mote Arctic locations approach the Oslo and
tions through their diet, showing a strong re- the capability of sentinel species to adjust to Paris Convention standards (5). Other large-
lationship with industrial fisheries’ catch per large-scale changes, a dynamic that needs to scale, long-term international programs for
unit effort (CPUE) (6). be better understood to determine the pos- the observation of marine ecosystems and
Seabirds are affected by changes at a range sible results of ocean management strategies. biodiversity, such as the Commission for
of time scales. Changes in the environment Data loggers (small devices deployed on the Conservation of Antarctic Marine Liv-
on a scale of weeks or months may be re- wild animals to record or transmit environ- ing Resources (14), the Poseidon system (15),
flected in reproductive performance. Because mental and physiological data) can play a key the Distributed Biological Observatory, the
of their longevity and annual reproduction, role in this effort by providing information Global Ocean Observing System, and others
seabirds have relatively stable populations such as foraging area, water temperature, or (11, 13), are identifying essential ocean vari-
on a year-to-year basis, so longer-term vari- diving depth (8). For example, Harwood et al. ables to monitor the global ocean and its
ability in environmental conditions can be (10) have shown that population size, date of biological resources with standardized inter-
detected from fluctuations in population size arrival at breeding areas, body condition, and national procedures (13).
(7), changes in distributional ranges, or re- breeding success are affected by environmen- It is unclear how quickly different seabird
mote sensing of tagged individuals. tal variables. Because seabirds adjust behav- species will be able to adapt to rapidly chang-
Different seabird parameters can be used iorally to environmental changes, monitoring ing environmental conditions, but, as senti-
in a complementary way and combined with of foraging activity and prey taken to nest- nels of the health of the global ocean, they
environmental parameters to provide useful lings may reveal high-productivity areas or are clearly providing a warning of the drastic
information about marine ecosystem func- switches in diet to demersal prey when water effects that this fast pace of changes is hav-
tioning (2). For example, to understand the temperature increases (10, 11). ing on marine species worldwide. We have a
impact of oceanographic phenomena on eco- Recent demographic history studies by chance to use the knowledge provided by sea-
system components, Bost et al. (8) used long- Ruiz et al. (12) have shown that the coupled birds to make the right decisions to prevent
term datasets (~18 years) on king penguins, trophic relationship between seabirds and the loss of biodiversity and at the same time
including nesting population size, breeding forage fish likely evolved over hundreds of develop sustainable ways to harvest it. j
success, diet composition, food consumption, thousands of years. They found that Heer-
REF ERENCES AND NOTES
and foraging distance, depth, time, and suc- mann’s gulls (Larus heermanni) nesting in
1. J. F. Piatt, W. J. Sydeman, Mar. Ecol. Prog. Ser. 352, 199
cess, together with environmental variables the Gulf of California had two population (2007).
such as sea surface temperature, thermo- expansions that roughly coincided with 2. G. R. W. Humphries et al., PICES Press 23, 18 (2015).
cline depth, and distance to polar front. They population expansions of their two main fish 3. E. Velarde, E. Ezcurra, Condor 120, 388 (2018).
4. E. A. Schreiber, J. Burger, Eds., Biology of Marine Birds
found that under conditions of subtropical prey, the Pacific sardine (Sardinops sagax, (CRC Press, 2001).
anomalies, the birds could find their prey ~317,000 to 218,000 years before the pres- 5. B. D. Hardesty, T. P. Good, C. Wilcox, Ocean Coast. Manage.
only by diving deeper and farther from their ent) and the northern anchovy (Engraulis 115, 4 (2015).
6. E. Velarde, E. Ezcurra, D. W. Anderson, J. Mar. Syst. 146, 82
colony, resulting in a reduction in breeding mordax, ~92,000 to 30,000 years before the (2015).
success and colony size. present), in the same region. 7. D. Anderson et al., Sci. Mar. 43, 1 (2017).
Most fish stocks have been seriously de- A better understanding of regime changes 8. C. A. Bost et al., Nat. Commun. 6, 8220 (2015).
9. W. Cai et al., Nature 564, 201 (2018).
pleted, lowering ocean biodiversity and affecting fish body condition, fish behavior, 10. L. A. Harwood et al., Prog. Oceanogr. 136, 263 (2015).
dangerously threatening the structure and fish and seabird migrations will be cru- 11. S. E. Moore, K. J. Kuletz, Deep Sea Res. Part II Top. Stud.
and resilience of marine ecosystems. The cial for developing sustainable fisheries (13). Oceanogr. 162, 211 (2019).
12. E. A. Ruiz, E. Velarde, A. Aguilar, Auk 134, 308 (2017).
practice of monitoring and managing fish This will require a globally coordinated effort
13. P. Miloslavich et al., Glob. Change Biol. 24, 2416 (2018).
populations through single-stock models to study environmental changes relevant to 14. W. J. Sydeman et al., Ecol. Indic. 78, 458 (2017).
has failed repeatedly, leading to the present marine species. Many local or regional pro- 15. S. K. Lowerre-Barbieri, I. A. Catalán, A. Frugård Opdal, C.
critical state of global marine ecosystems. grams for gathering diverse ocean variables, Jørgensen, ICES J. Mar. Sci. 76, 467 (2019).
Ecosystem-based management, an alterna- such as mangrove cover, fish abundance, 10.1126/science.aaw9999

INSIGHTS | P E R S P E C T I V E S
DEVELOPMENT
Protecting fetal development

Control of cell fusion in the placenta affects the balance between health and harm
By Paul Kellam1 and Robin A. Weiss2 stem cells, tissue stem cells, and, as now plasm through a fusion-induced pore. For
shown by Buchrieser et al., the placental enveloped viruses that enter cells through
T
he interferon-induced transmem- trophoblast. The role of IFITM1–3 in pla- the plasma membrane, IFITM1 has the
brane (IFITM) protein family in- cental integrity, however, is likely to be equivalent activity (9). Buchrieser et al.
cludes members that protect multiple complex. When embryonic stem cells prolif- found that most of the activity in inhib-
cell types from infection by many erate and differentiate, IFITM expression is iting trophoblast cell fusion is mediated
viruses by preventing the fusion of down-regulated (4). Furthermore, IFITM3 by IFITM3, which may seem inconsistent
virus envelopes with host cell mem- is pleiotropic: It prevents excessive pro- with the known locations of IFITMs. How-
branes (1). The syncytiotrophoblast at the duction of cytokines such as interleukin-6 ever, both IFITM2 and IFITM3 traffic to
maternal-fetal interface of the placenta (IL-6) during murine cytomegalovirus in- endosomes via the plasma membrane (9).
develops through cell-cell fusion of cyto- fection (5), enhances cellular proliferation IFITM1–3 can work cooperatively and ac-
trophoblast cells by a mechanism akin to in tumor cells (6, 7), and controls inflam- celerate the rate of trafficking endosomes
virus-cell fusion (see the figure). This cell mation in the gut, which is associated with containing virus particles to lysosomes for
fusion is mediated by syncytins. These are decreased colon tumorigenesis (8). destruction (10).
membrane glycoproteins encoded by an- Human IFITM1 is expressed at the Different types of retrovirus have given
cient endogenous retroviral envelope genes, plasma membrane (9), whereas IFITM2 rise to syncytins among diverse types of
which were acquired and have become re- and IFITM3 become localized to intracel- placental mammals, and indeed they pre-
purposed for placental development (2). lular membranes, mainly of endosomes. ceded the evolution of a full placenta (2, 11).
On page 176 of this issue, Buchrieser et al. For enveloped viruses that enter cells via Buchrieser et al. show that human syncy-
(3) show that interferon stimulation causes the acidic endosomes, IFITM3 prevents tins 1 and 2 and mouse syncytins A and B
IFITM expression in trophoblast cells, completion of virus membrane and endo- are subject to inhibition by IFITMs and that
which blocks syncytin-mediated cell fusion, some membrane fusion, thereby inhibiting induction of type 1 interferon expression
and that interferon stimulation also leads virus nucleocapsid entry into the cyto- by the synthetic double-stranded RNA mol-
to placental dysfunction and ecule polyinosinic:polycytidylic
fetal resorption in wild-type acid (poly-I:C) in the mouse
mice, but not in mice lacking Cell fusion in the placental villus leads to loss of developing fe-
Ifitm genes. These findings have In the first trimester, the villous syncytiotrophoblast is situated at the fetal- tuses by stimulating IFITM1–3
implications for understanding maternal interface and is derived from the fusion of cytotrophoblasts mediated by overexpression. They note that
how infections and other com- syncytins. Interferon-induced up-regulation of interferon-induced transmembrane a similar mechanism likely
plications during pregnancy (IFITM) proteins suppresses the amount of cell fusion. contributes to the pathology of
can lead to miscarriage. transplacental infections such
Interferon-stimulated genes as Zika virus and the TORCH
(ISGs) fall into two broad Syncytiotrophoblast [toxoplasmosis, others (includ-
groups. The first is perma- ing syphilis, listeriosis, vari-
nently on guard duty, where Fetal capillary Cytotrophoblast cella, parvovirus B19), rubella,
the protein is constitutively Mesenchymal
cytomegalovirus, and herpes
expressed at a basal level but stroma simplex virus] constellation
is up-regulated during infec- of congenital pathogens. They
tion by interferon, which is a also point out that an abnor-
cytokine produced by virus- mal syncytiotrophoblast is ob-
infected cells that makes neigh- served in several complications
boring cells refractory to in- of human pregnancy, including
fection. The IFITMs belong to intrauterine growth retarda-
this group. The expression of tion, preeclampsia, lupus, and
a second group of ISGs is more Down syndrome (possibly be-
stringently controlled by inter- Extravillous cause chromosome 21 encodes
ferons, presumably due to acute trophoblast an interferon receptor). It seems
toxicity of these proteins. A cru- unlikely that all such infections
GRAPHIC: KELLIE HOLOSKI/SCIENCE

cial function of IFITM1–3 is to and syndromes result from
protect stem cells from infec- IFITM interference of syncytio-
tion, including early embryonic trophoblast formation, and it
will be interesting to determine
1
Department of Medicine, Division of Infectious whether deletion of individual
Diseases, Imperial College London, London, UK. IFITMs, especially IFITM3,
2
Division of Infection and Immunity, University
College London, London, UK. Email: p.kellam@ Fusion of cytotrophoblasts mimics the phenotype of com-
imperial.ac.uk; r.weiss@ucl.ac.uk to form syncytiotrophoblast pound Ifitm-ablated mice.

The degree of IFITM3 expression thus IMMUNOTHERAPY
requires a delicate balancing act because
a high level will provide better protec-
tion from infection, whereas a low level
will help to protect fetal development.
Boosting engineered T cells
This tension between protection from Vaccines augment the antitumor activity of cellular therapy
infection and impairment of normal fe-
tal development may help to explain the
relatively high frequency of alleles that re- By Nathan Singh1 and Carl H. June2 and that delivery to mice of anti-FITC–engi-
duce IFITM3 protein expression in differ- neered CAR–T cells followed by immuniza-
A
ent human populations (1). This was first fter decades of work, researchers tion with amph-FITC significantly improved
observed (12) in the 2009 H1N1 influenza have finally begun to see broadly re- T cell proliferation in vivo. Notably, this effect
pandemic, in which a single-nucleotide producible success of engineered T relied on costimulatory signals delivered to T
polymorphism (SNP) of IFITM3 was asso- cells in the treatment of cancer. Chi- cells by APCs expressing amph-FITC, identify-
ciated with a much higher risk of severe meric antigen receptors (CARs) are ing the need for CAR-independent costimula-
flu. There are two distinct noncoding SNPs synthetic molecules that combine the tion in this system. Expanding these findings
in IFITM3, in East Asian and in European antigen specificity of monoclonal antibod- to a tumor-antigen model, the authors used a
and African American populations, respec- ies with the signaling of the T cell receptor CAR targeting a splice variant of epidermal
tively (13, 14). Both of these SNPs reduce (TCR) to direct patient-derived (autologous) growth factor receptor, EGFRvIII, which is
constitutive expression of IFITM3 and T cells to seek out and destroy cancer cells. commonly expressed in glioma, in combi-
therefore increase the virulence of influ- T cells engineered to express CARs targeting nation with a vaccine containing an amphi-
enza A infection. The association of these the B cell antigen CD19 can induce durable philic polymer linked to the EGFRvIII target
two SNPs may be important markers for an remissions in many patients with refractory antigen. In mice with EGFRvIII+ gliomas,
increased risk of pregnancy complications B cell neoplasms (1–3), and two CAR–T cell vaccination resulted in improved CAR–T cell
upon maternal infection, which could be products have recently been approved by the proliferation and survival, and improved in-
easily tested noninvasively in the mother. U.S. Food and Drug Administration to treat B filtration of activated CAR–T cells into tumor
The high frequency of SNPs that reduce cell leukemia and lymphoma. Despite these sites compared to CAR–T cell delivery alone.
IFITM3 expression (1) could be explained successes in hematological cancers, CAR–T A potential limitation of the EGFRvIII
if there were counterselection to maintain cell activity against solid tumors has been strategy is that the CAR targets a linear epi-
low or no constitutive expression of IFITM3, limited. On page 162, Ma et al. (4) describe tope in the EGFRvIII protein, which is more
despite the risk of enduring more severe a platform that uses a vaccine-boosting strat- readily targeted than the more common con-
symptoms in viral infections. Therefore, egy to improve the efficacy of CAR–T cells to formational epitopes of cancer-specific anti-
the need to form a healthy and robust pla- target solid tumors. gens. To develop a generalized therapeutic
centa may provide the balancing selection, Cellular immunotherapy of solid tumors strategy, Ma et al. constructed a dual-targeted
just as ablation of the Ifitm genes in mice presents several distinct barriers: After infu- “tandem CAR,” composed of a CAR targeted
by Buchrieser et al. resulted in resistance to sion, engineered T cells must traffic to sites to a tumor antigen linked to a CAR target-
interferon-induced placental abnormalities of tumor residence, infiltrate a highly disor- ing FITC. In mouse models of melanoma and
and fetal readsorption. Whether these al- ganized tumor architecture, and survive in breast cancer, they demonstrated that deliv-
leles affect IFITM3 expression in embryonic a hostile tumor microenvironment. Several ery of these tandem CARs followed by vacci-
tissues and the fetal environment during in- studies have shown that despite success- nation with amph-FITC leads to significantly
fection remains to be more fully elucidated ful trafficking and infiltration, CAR–T cells improved antitumor activity, obviating the
and might indicate ways of reducing the quickly become dysfunctional after encoun- need to have a vaccine individualized for each
risk of miscarriage. Overall, it appears that tering tumors (5, 6). Ma et al. devised a clever target antigen and independent of whether
one of the most powerful natural selection strategy to boost CAR–T cell proliferation and the epitope is linear or conformational.
pressures, that of fecundity, trumps defense survival with vaccination (see the figure). The Vaccine-based strategies have long been
against infection. j authors previously developed a membrane- explored as a method to improve antitumor
integrating phospholipid polymer that could immunity, and several recent studies have
REFERENCES AND NOTES
be linked to small molecules or peptides, re- combined vaccination with CAR–T cells to
1. X. Zhao et al., Front. Microbiol. 9, 3228 (2019).
2. C. Lavialle et al., Philos. Trans. R. Soc. Lond. B Biol. Sci. 368,
sulting in expression of a desired target anti- treat various solid tumors in mice and pa-
20120507 (2013). gen on the cell surface after “immunization” tients (8–10). These approaches all follow
3. J. Buchrieser et al., Science 365, 176 (2019). with this molecule (7). By binding to albumin the same paradigm: selecting (or engineer-
4. X. Wu et al., Cell 172, 423 (2018). after injection, these amphiphilic polymers ing) T cells with known TCR specificity and
5. D. Zhang et al., Thorac. Cancer 8, 337 (2017).
6. J. Min et al., FEBS Open Bio. 8, 1299 (2018).
are directed to lymph nodes, where they are engineering these to express CARs, thus gen-
7. Z. Alteber et al., Immunol. Cell Biol. 96, 284 (2018). preferentially displayed on resident antigen- erating a T cell product with dual specificity.
8. S. E. Smith et al., J. Virol. 93, e2003 (2019). presenting cells (APCs), which prevents their These studies have demonstrated feasibility
9. J. S. Spence et al., Nat. Chem. Biol. 15, 259 (2019). loss in systemic circulation. but with limited antitumor efficacy.
10. M. A. Stacey et al., J. Clin. Invest. 127, 1463 (2017).
Ma et al. demonstrated that these amphi- Previous clinical trials may provide clues
11. G. Cornelis et al., Proc. Natl. Acad. Sci. U.S.A. 112, E487
(2015). philic polymers linked to fluorescein isothio- about the limited activity of these strategies.
12. A. R. Everitt et al., Nature 484, 519 (2012). cyanate (FITC) as the antigen (amph-FITC) Persistent TCR stimulation of CAR–T cell
13. S. S. Prabhu, T. T. Chakraborty, N. Kumar, I. Banerjee, Gene are stably expressed on the surface of APCs, products can lead to “terminal” differentia-
674, 70 (2018).
tion, a state of irreversible T cell dysfunction
14. E. K. Allen et al., Nat. Med. 23, 975 (2017).
1
(11). Furthermore, strategies to disrupt TCR
Washington University School of Medicine, St. Louis, MO,
USA. 2Center for Cellular Immunotherapies, University of genes in CAR–T cells, both as a means to
10.1126/science. aay2054 Pennsylvania, Philadelphia, PA, USA. Email: cjune@upenn.edu prevent off-target TCR-driven T cell activ-

ity that causes side effects, and to improve to antigen-expressing lymph node tissues. COMPUTATIONAL BIOLOGY
CAR-driven antitumor efficacy, are being ex- Paradoxically, this suggests that CAR–T cells
plored (12). The conceptual innovation of the
approach taken by Ma et al. is to bypass the
major histocompatibility complexes (MHCs)
can be stimulated by APCs without killing
them, while the CAR–T cells retain the abil-
ity to kill antigen-expressing tumor cells. If
Mapping
that present antigens to TCRs needed for tra-
ditional vaccine responses, while preserving
confirmed, this flexibility has not previously
been demonstrated by CAR–T cells, which global protein
the immune stimulation provided by vac-
cination. Another conceptual innovation of
this approach is that it uses the CAR not only
are MHC independent and thus may not be
subject to the same regulatory signals as en-
dogenous T cells. Additionally, vaccine boost-
contacts
for tumor targeting but also as a machine to
enhance T cell activity, demonstrating that
ing of CAR–T cells resulted in activation of
endogenous T cells and the development of
Genome-scale coevolution
this chimeric molecule may have multifunc- CAR-independent immune memory to other experiments expand bacterial
tionality. Exploring how these synthetic CAR
molecules can be used more efficiently, effec-
tumor antigens. Understanding how adop-
tively transferred CAR–T cells interact with
protein interactomes
tively, and creatively will open doors to new the endogenous immune system to promote
therapeutic platforms. such epitope spreading is important, and By Sandor Vajda and Andrew Emili
T
he divergence of orthologous protein
Vaccination strategy to boost antitumor efficacy sequences across evolutionary lineages
In mice, the amph-ligand vaccine results in stimulation and proliferation of engineered chimeric antigen can be used to pinpoint possible con-
receptor (CAR)–T cells in lymph nodes. These then migrate to solid tumors, where they have improved tacts at specific amino acids by exploit-
proliferation and enhanced efficacy. ing the tendency for compensating
mutations to coevolve at interacting
Endogenous positions within proteins (1). This coevolu-
T cell tionary approach has galvanized the vast
Amph-
ligand improvement in protein-structure prediction
Inactive
Active
over the past two decades (2). It has also
Albumin CAR–Tcells been used to locate contact points between
CAR–T cell
Tumor
Tandem cell
pairs of interacting proteins (3), which can
CAR serve as distance restraints for high-quality
CAR
models of multiprotein complexes by struc-
tural docking (4). On page 185 of this issue,
Cong et al. (5) use this method to explore
Lymph potential interactions among all Escherichia
node APC coli and Mycobacterium tuberculosis proteins
and thereby enhance knowledge of bacterial
protein interaction networks (interactomes).
Because macromolecular complexes drive
1 A compound expressing the 2 In the lymph node, stimulation of 3 Trafcking of CAR–T cells into most biological processes, elucidating the un-
target of the CAR, amph-ligand, CAR–T cells by amph-ligand expressing solid tumors results in antitumor derlying networks of physically interacting
is injected into recipients. Upon antigen-presenting cells (APCs) results activity, enhanced CAR–T cell proteins is key to understanding the molecu-
binding to albumin, the vaccine in CAR–T cell activation, proliferation, survival, and activation of lar machinery of a cell. Protein-protein inter-
migrates to lymph nodes. and diferentiation. endogenous antitumor immunity.
actions have been investigated traditionally
by labor-intensive experimental methods, ap-
The principal limitation of the study of Ma strategies that combine CAR–T cells with plied separately to a small set of potentially
et al. is the unknown ability of this strategy vaccines may open a critical window of coop- interacting protein targets. Starting with mi-
to boost CAR–T cells in humans. The primary eration between synthetic and natural anti- crobes, scientists began to construct larger
toxicity of CAR–T cells has been cytokine re- cancer immunity. j networks by performing large-scale analyses
lease syndrome, a systemic inflammatory dis- based on yeast two-hybrid (Y2H) screens (6)
order resulting from CAR–T cell activation, or affinity purification of protein complexes
1. S. J. Schuster et al., N. Engl. J. Med. 380, 45 (2019).
which is more exaggerated in humans than 2. S. L. Maude et al., N. Engl. J. Med. 378, 439 (2018). coupled to mass spectrometry identification
in mice. Their innovative approach avoids 3. J. H. Park et al., N. Engl. J. Med. 378, 449 (2018). (APMS) (7). However, such high-throughput
systemic expression of the surrogate CAR tar- 4. L. Ma et al., Science 365, 162 (2019). approaches can miss certain interactions or
5. M. Philip et al., Nature 545, 452 (2017).
get (i.e., FITC) on vital cells or organs such as 6. D. M. O’Rourke et al., Sci. Transl. Med. 9, eaaa0984 (2017). yield spurious ones and are limited to orga-
the brain or liver. However, whether polymer- 7. H. Liu et al., Nature 507, 519 (2014). nisms in which molecular techniques such
antigen expression will be limited to APCs in 8. C. Rossig et al., Leukemia 31, 1087 (2017). as gene manipulation are possible (8). To im-
GRAPHIC: A. KITTERMAN/SCIENCE
9. Y. Akahori et al., Blood 132, 1134 (2018).
humans, particularly after extensive chemo- 10. C. Y. Slaney et al., Clin. Cancer Res. 23, 2478 (2017).
prove the coverage and reliability of micro-
therapy and/or radiotherapy, which may alter 11. C. R. Cruz et al., Blood 122, 2965 (2013). bial interaction networks, researchers have
constitutive antigen presentation, remains to 12. J. Eyquem et al., Nature 543, 113 (2017). tried to incorporate knowledge of functional
be learned. ACKNOWL EDGMENTS
A notable finding from the work of Ma N.S. and C.H.J. are inventors on patents in the field of CAR–T cells. Departments of Biomedical Engineering, Biochemistry,
et al. is that boosting CAR–T cell effector and Biology, Boston University, Boston, MA 02215, USA.
function did not result in detectable injury 10.1126/science.aax6331 Email: aemili@bu.edu

relatedness, such as the closeness of bacterial et al. performed a prescreen based on a less based screen outperformed existing experi-
genes (membership in operons) or similari- computationally demanding analysis of resi- mental methods. One potential caveat is that
ties in phylogenetic profiles, but such integra- due-residue correlations. Still, close to a mil- APMS studies report co-complexes, rather
tive scoring approaches can lead to bias (9). lion potentially interacting pairs remained than binary interactions.
Given the need to enhance the scope and to be scrutinized with both direct coupling By exploiting coevolution, Cong et al.
quality of protein interaction networks, the analysis (10) and GREMLIN to rank higher- boosted interactome coverage while lever-
use of coevolutionary information on a ge- likelihood candidate protein pairs based on aging existing bacterial protein–interaction
nome scale is a game-changing addition; it the number of predicted residue couplings. data. In this way, their approach is comple-
is also a tremendous challenge. It is difficult The authors selected the top 21,818 protein mentary to experimental ones. In E. coli, only
to find true evolutionary covariation between pairs from the ranked list and built protein- 24.7% of the strongly coevolving 1618 pairs
residues for a single protein because one protein complexes by computational model- are new or unexpected. In fact, 936 interac-
must minimize the effect of transitive (false- ing, using the predicted contacts as distance tions were reported previously, homologous
positive) correlations; these can be observed, restraints for docking. templates in the Protein Data Bank (PDB) ex-
for example, when two amino acid residues Restricting considerations to complexes ist for a further 126 pairs, and 156 pairs have
contact the same third residue but not each that displayed the predicted contacts within genomic associations. These results increase
other. Transitive correlation can be removed the putative protein-protein interface re- confidence in the coevolution approach.
by global statistical approaches involving di- duced the number of protein pairs to 804. The results for M. tuberculosis are more
From protein pairs to complex interaction networks

With the method shown below, coevolutionary analyses of amino acid contacts in E. coli protein pairs can unearth new interprotein contacts that serve to expand
protein interaction networks. The multiple sequence alignment shown in the “Analyze” step is for a hypothetical E. coli protein and its orthologs (blue rectangles) and
a potentially interacting partner from the same organisms (gray rectangles). Red boxes indicate interprotein contacts predicted by GREMLIN.
Find pairs Analyze Build protein complexes Chart

Map of all E. coli protein pairs Construct multisequence alignments with Use predicted amino acid contacts Add validated interactions (red) to an
orthologs; use coevolutionary analyses to (in red, blue, magenta, and yellow) E. coli interaction network with
Proteins Proteins Lnd coupled amino acid as restraints previously known contacts (black)
------
---------------
--------
---------
---
-----
---------
-------
----
rect coupling analysis (10), pseudo-likelihood This is a substantial gain relative to existing informative. Of the 667 predicted pairs, only
optimization (11), or machine learning (12). experimental datasets (14) but falls far short 203 (30.4%) are supported by homologous
Cong et al. used GREMLIN, a pseudo- of the putative bacterial interactome. Under- templates in the PDB or were reported pre-
likelihood method they offer as a public representation was especially pronounced for viously. Thus, the most notable advances of
server (13). As with most evolutionary tools, less widely conserved assemblies and multi- Cong et al. may be the potential to map the
GREMLIN starts with multiple sequence protein complexes. As with previous data- binary protein interfaces and global inter-
alignment of all available orthologs for a filtering strategies, Cong et al.’s multistep action networks of bacterial pathogens and
target protein (see the figure). It is based algorithm eliminated potential interactions study how the core protein interactomes
on constructing a parametric model that that would have been considered meaning- have evolved across microbial species. Adapt-
generates the observed sequences with the ful in later steps of the search. To reduce the ing this approach to eukaryotic networks rep-
highest probability. However, estimating number of false negatives, the authors intro- resents a formidable future challenge. j
the parameters of an exact model is com- duced, into the later stages of their analysis,
putationally intractable; hence, GREMLIN protein pairs reported in previous experi-
1. D. S. Marks et al., Nat. Biotechnol. 30, 1072 (2012).
optimizes a pseudo-likelihood function mental studies or known to be expressed by 2. R. F. Service, Science 10.1126/science.aaw2747(2018).
to reduce unlikely couplings early in the the same operon, thus arriving at 1618 pairs. 3. O. Lichtarge et al., J. Mol. Biol. 257, 342 (1996).
4. S. Ovchinnikov et al., eLife 3, e02030 (2014).
search and requires sequences of ortho- False-positive predictions can also be sub- 5. Q. Cong et al., Science 365, 185 (2019).
logs across a large number of diverse orga- stantial. In fact, paralogs that do not interact 6. B. Schwikowski et al., Nat. Biotechnol. 18, 1257 (2000).
nisms. Cong et al. adapted their approach to can show strong coevolution. The authors 7. A. Kumar, M. Snyder, Nature 415, 123 (2002).
8. J. S. Bader et al., Nat. Biotechnol. 22, 78 (2004).
one used to determine interprotein contacts, attempted to minimize this uncertainty by 9. L. J. Lu, et al., Genome Res. 15, 945 (2005).
which requires large joint alignments of ho- eliminating proteins that show nonspe- 10. F. Morcos et al., Proc. Natl. Acad. Sci. U.S.A. 108, E1293
(2011).
mologous protein pairs known to interact (4). cific coevolution with many other proteins 11. H. Kamisetty et al., Proc. Natl. Acad. Sci. U.S.A. 110, 15674
The innovation of Cong et al. is their inves- and also benchmarked their results against (2013).
tigation of coevolution on a whole-proteome previously published gold standards. They 12. S. Wang et al., PLOS Comput. Biol. 13, e1005324 (2017).
13. GREMLIN, http://gremlin.bakerlab.org.
GRAPHIC: N. CARY/SCIENCE
scale using orthologs mapped across more compared coevolution-based interaction 14. M. Babu et al., Nat. Biotechnol. 36, 103 (2018).
than 40,000 known bacterial genomes. To predictions with those inferred from high-
ACKNOWL EDGMENTS
understand the challenge, consider that E. throughput Y2H and APMS studies relative
A.E. acknowledges financial start-up support from Boston
coli lab strains express ~4262 proteins, with to structure-based benchmarks derived from University; S.V. was supported by grants NIH R35 GM118078 and
more than 9 million potentially interacting x-ray and cryo–electron microscopy analyses NSF DBI 1759472.
protein pairs. To lessen the complexity, Cong of E. coli protein assemblies. The coevolution- 10.1126/science.aay1440

CANCER
Stromal directives can control cancer

Could reprogramming the pretumor microenvironment transform cancer care?
By Thea D. Tlsty and Philippe Gascard of dysplasia, which in turn occasionally de- ence of a metaplastic reaction at the site of
velops into malignancy, suggesting that this injury. Increased cellular plasticity and the
T
here is ample evidence that inflam- process is error prone. reexpression of pathways associated with em-
matory processes and signaling play The progression of Barrett esophagus to bryogenesis and fetal development (4, 5) have
a critical role in the progression of EAC is rare (1 in 1000 Barrett esophagus been observed in the context of the metaplas-
most cancers (1), especially as potent cases). However, epidemiological, physi- tic reaction and other adaptive responses to
initiators at sites of chronic injury ological, molecular, and histological lines injury. Could such distinct alterations play a
(2). In these chronic inflammation– of evidence strongly support the origin of role in the development of aggressive tumor
associated cancers (CIACs), tissue injury is most EACs from the cellular field of Barrett phenotypes associated with CIACs?
caused by sustained or recurrent infection, esophagus. Patients diagnosed with Barrett Cellular plasticity can be enhanced and po-
irritation, or trauma. This distinct class of esophagus have a 40-fold increased lifetime tentially stabilized by nonmutational events
malignancies, which includes acid reflux– risk of developing EAC and a 0.1 to 0.3% risk in the chronic inflammatory microenviron-
associated esophageal adenocarcinoma of progression to EAC per year. Critically, the ment. For instance, the cytokine milieu gen-
(EAC), smoking-associated lung squamous 5-year survival rate for patients with EAC erated by chronic inflammation promotes
cell carcinoma, Helicobacter pylori infec- is 15%. Unlike symptomatic patients, many epigenetic changes within epithelial cells that
tion–associated gastric adenocarcinoma, patients with acid reflux and associated Bar- are often associated with cell-fate plasticity.
and inflammatory bowel disease–associated rett esophagus are asymptomatic and are Chronic inflammation was shown to drive
colon adenocarcinoma, is responsible for 20 therefore not followed clinically through en- the epigenetic silencing of the gene encod-
to 25% of all cancer deaths worldwide, kill- doscopic surveillance to detect EAC. Indeed, ing the tumor suppressor p16 in a chronic
ing more than 2 million people annually. 85% of patients with EAC first present in the inflammation–associated lung cancer mouse
Compared with other tumor types, CIACs clinic with late-stage, metastatic, unresect- model (6). Previous work has demonstrated
tend to have a poorer prognosis, attribut- able, and highly drug-resistant cancers with that repression of p16 expression not only re-
able to a high propensity for metastasis limited treatment options. Identifying the laxes cell cycle checkpoint response, which is
and drug resistance. The underlying mech- rare subset of Barrett esophagus patients important for maintaining genomic stability,
anisms by which chronic inflammation who will progress to EAC is an unmet clinical but also increases the expression of chroma-
promotes the development of such lethal challenge, as is the search for effective thera- tin remodeling proteins and, consequently,
cancers are still poorly understood. peutics to treat patients at various stages of activates epigenetic plasticity (4). Loss of p16
EAC is a typical example of a CIAC. The EAC progression. To improve the diagnosis activity also reduces control on centrosome
clinical course of this disease progresses and treatment of these diseases, the follow- number (which is important for accurate
through stages characterized by increased ing question must be answered: How does cell division) and activates a transcriptional
alterations of the tissue landscape: metapla- chronic injury and the ensuing chronic in- stress response supporting multiple ma-
sia, dysplasia, and carcinoma. These stages flammation lead to tumorigenesis? lignant traits (7). Thus, cells in which p16
are shared by other CIACs and may suggest Inflammation is not fundamentally a activity is silenced through DNA hypermeth-
a common biological mechanism linking pathogenic process. The inflammatory re- ylation acquire a multitude of phenotypes
chronic inflammation to carcinogenesis (3) sponse is an essential component of the that poise the cells for cell fate plasticity as
(see the figure). In 10 to 15% of patients with host immune reaction to microbial infec- well as malignant potential. Epigenetic si-
repeated acid reflux, the persistent injury tion and tissue injury. Inflammation is usu- lencing of p16 expression is common at the
and associated chronic inflammation leads ally self-limiting; however, the inflammatory metaplastic stage in a substantial subset of
to the activation of an adaptive metaplastic response is poorly adapted to resolving per- CIACs, including injury-associated squamous
response, defined as the replacement of one sistent injurious stimuli. When injury per- cell carcinoma of the lung, H. pylori infec-
tissue type by a different tissue type not nor- sists, cycles of tissue destruction perpetuate tion–associated gastric cancer, chronic pan-
mally present at that location. Metaplasia in the inflammatory response and establish a creatitis–associated pancreatic cancer, and
the injured esophagus, called Barrett esopha- chronic inflammatory state associated with acid reflux–associated EAC.
gus, is characterized by the replacement of an unresolved DNA damage response, secre- It is postulated that, in addition to promot-
the native squamous esophageal epithelium, tion of proinflammatory cytokines, and reac- ing cellular plasticity, chronic inflammation
most often with columnar intestinal and tive oxygen species (ROS). Consequently, the also disrupts homeostatic epithelial-mesen-
gastric epithelium. Barrett esophagus is not prevailing paradigm for conceptualizing the chymal interactions and alters contextual
inherently premalignant, as the resultant tis- relationship between chronic inflammation cues that direct cellular communities to
sue is well suited to protecting the esophagus and cancer initiation focuses on an increase adopt new compositional states, which can be
from acid exposure owing to the presence of in the epithelial mutation rate and genomic either protective (metaplasia) or potentially
intestinal goblet cells, which naturally elabo- instability in the perturbed environment, as pathogenic (malignancy) (8, 9). Developmen-
rate acid-neutralizing mucus. However, meta- well as the activation of oncogenic signal- tal biology has taught us that tissue organi-
plasia is associated with a higher frequency ing pathways by inflammatory mediators. zation and cell identity are coordinated by
However, as with Barrett esophagus, an in- epithelial-mesenchymal interactions, which
Department of Pathology, University of California, San Francisco, triguing feature of many CIACs that is not include soluble factors, cell-cell interactions,
San Francisco, CA 94143, USA. Email: thea.tlsty@ucsf.edu often observed in non-CIACs is the pres- cell shape, cell–extracellular matrix (ECM)

interactions, as well as mechanical and physi- active ECM fragments; and decreased depo- ized ECM proteins from these mesenchymes,
cal forces (8, 9). sition of decorin, matrilin, and nidogens 1 lose malignant properties and acquire more-
Recent work demonstrates that epithelial- and 2 (9). Notably, the increased expression differentiated phenotypes (15). Therefore,
mesenchymal interactions continue to dic- of ECM proteins and their posttranslational just as there are mesenchymal signals that
tate tissue structure and function even in cross-linking increase tissue stiffness and al- can induce and maintain malignant epithe-
adult organisms. For example, implantation ter mechanotransduction in inflamed tissue lial states, there are also mesenchymal sig-
of neural stem cells or human embryonic (12). Thus, epithelial cells activated to a plas- nals that can induce and maintain epithelial
stem cells into mouse mammary fat pads tic state and bereft of appropriate contextual differentiation states.
reprograms the cells into milk-producing cues may be at the highest risk of following a Understanding the factors influencing the
mammary epithelium (10). Additionally, in malignant cell state trajectory. fidelity of the metaplastic response is critical
both the developmental and adult contexts, The idea that signaling from the stroma, to identifying the mechanisms underlying
much of the instructive information for cel- as well as the instructive information con- the genesis of CIACs and addressing their
lular identity is contained within the ECM, a tained within the ECM, may underlie tissue clinical consequences. Identifying common
network of proteins that constitutes the non- pathologies, including the development of contextual elements (such as ECM proteins)
cellular component of the stroma. This is evi- malignancy, is not new (8, 9). This concept favoring the maintenance of an adaptive pro-
denced by experiments demonstrating that is supported by many experimental studies. tective metaplastic state that can no longer
decellularized ECM can recapitulate many For example, the ectopic expression of matrix progress to malignancy may support the de-
velopment of different approaches for early
diagnosis, risk stratification, prevention,
Chronic inflammation–associated cancer pathogenesis and therapeutic intervention to decrease
The pathogenic evolution of chronic inflammation–associated cancers (CIACs) involves metaplasia, dysplasia, the clinical burden of CIACs. This potential
and then cancer. During this process, multiple changes take place within the tissue microenvironment, including new class of therapy for epithelial cancers
changes in stromal cells (immune cells, endothelial cells, and fibroblasts) and the extracellular matrix (ECM). would restore homeostatic tissue interac-
tions that normally prevent cancers from
Normal Metaplasia Dysplasia Cancer
forming or advancing but that are often lost
Chronic infammation Cell plasticity Altered signaling Chromosomal alterations in the context of chronic inflammation. The
•Injury •Radiation Genetic alterations
•Pathogens •Chemicals severity and duration of chronic inflamma-
tion largely dictates the risk of future tumor
formation, providing a considerable window
for monitoring and intervention. A strategy
Apoptosis Fibroblast combining current therapies that target cell
and necrosis remodeling Fibrosis
intrinsic oncogenic pathways with newly de-
ECM veloped therapies aimed at restoring tissue
remodeling homeostasis could be a viable complement
Immune cell to overcome failures. These failures may re-
Fibroblasts infltration Cytokines sult from focusing on therapeutic killing of
Cytokines Carcinoma-associated
Reactive fbroblasts mutant cancer cells without integrating the
oxygen restoration of microenvironmental interac-
species
tions that naturally participate in repressing
Endothelial Protumorigenic malignant phenotypes. j
cells vasculature
1. K. Taniguchi, M. Karin, Nat. Rev. Immunol. 18, 309 (2018).
of the instructive features of mesenchyme metalloproteinase 3 (MMP3) results in ECM 2. A. Mantovani et al., Nature 454, 436 (2008).
in guiding cell fate and state (9). For exam- reorganization and is sufficient to induce 3. M. Barbera, R. C. Fitzgerald, Surg. Oncol. Clin. N. Am. 18,
ple, decellularized ECM from rat mammary malignant transformation of the epithelium, 393 (2009).
4. H. Li et al., Nature 460, 1136 (2009).
glands in either virgin or lactating states is suggesting that epithelial cells lacking nor- 5. S. G. Willet et al., EMBO J. 37, e98311 (2018).
sufficient to reprogram mammary epithelial mative mesenchymal cues are more likely 6. D. Blanco et al., Neoplasia 9, 840 (2007).
cells to recapitulate the physiological state of to adopt a malignant fate (13). Moreover, 7. H. Berman et al., Cold Spring Harb. Symp. Quant. Biol. 70,
317 (2005).
the tissue from which the ECM was derived allowing carcinoma-associated fibroblasts 8. D. E. Ingber, Differentiation 70, 547 (2002).
(11). These studies inform us that the stroma, to exchange signals with nontumorigenic 9. C. M. Nelson, M. J. Bissell, Annu. Rev. Cell Dev. Biol. 22, 287
and the ECM in particular, is dominant and epithelial cells redirects the epithelial cells to (2006).
10. B. W. Booth et al., Proc. Natl. Acad. Sci. U.S.A. 105, 14891
dynamic in dictating development and cel- malignancy (14). (2008).
lular differentiation. The composition and Previous studies have shown that nor- 11. P. Schedin et al., Mol. Carcinog. 41, 207 (2004).
structure of the ECM are also therefore likely malizing cues from an embryonic milieu or 12. J. M. Northcott et al., Front. Cell Dev. Biol. 6, 17 (2018).
13. M. D. Sternlicht et al., Cell 98, 137 (1999).
critical to cellular decisions between a meta- a healthy stromal environment can force
14. A. F. Olumi et al., Cancer Res. 59, 5002 (1999).
plastic or a malignant path. aggressive cancer cells to lose tumorigenic 15. A. G. Bischof et al., Integr. Biol. 5, 1045 (2013).
The properties of the ECM are determined properties and exhibit benign and differen-
GRAPHIC: JOSHUA BIRD/SCIENCE
ACKNOWL EDGMENTS
by complex interactions between fibroblasts tiated characteristics. For example, aggres-
T.D.T. and P.G. thank J. Caruso and D. Pan for extensive discus-
and other mesenchymal cells, as well as im- sive teratocarcinoma cells introduced into
sions, the Cancer Research UK (CRUK) Grand Challenge
mune and epithelial cells. As in cancer, the the environment of a blastocyst (a stage STORMing Cancer team for critical remarks, and S. Huang for
ECM is extensively remodeled during chronic of embryonic development) contribute to help with the figure. T.D.T. acknowledges support from National
inflammation, which includes the increased functional tissues in newborn mice (9). Like- Cancer Institute grant R35 CA197694 and funding of the
STORMing Cancer team by a CRUK Grand Challenge grant.
deposition of the ECM components collagen, wise, malignant breast cancer cells placed
periostin, and fibronectin; liberation of bio- on embryonic mesenchymes, or decellular- 10.1126/science.aaw2368

HOST-GUEST CHEMISTRY
Supramolecular cages trap pesky anions

Carbon-hydrogen bonds in a cryptand cage boost chloride entrapment to ultrahigh affinity
By Kristin Bowman-James Liu et al. (2) report the design of a bicyclic bicyclic hosts, which they named cryptands,
cryptand that specifically recognizes Cl– containing polyether rather than hydrocar-
T
he design of cryptands—organic mol- ions with high affinity. Their results build bon chains (4). These metal ion hosts can
ecules that can capture simple inor- on a long history of increasingly complex selectively bind and solubilize alkali and al-
ganic and organic ions from solution cryptand cages specifically designed to tar- kaline earth ions, and other insoluble metal
through multiple weak interactions— get anions (see the figure). salts in organic solvents, enabling their use
can be quite challenging, especially Rather than rely on the covalent bonds as phase transfer catalysts. Cryptands with
when the task is to capture a specific found between atoms in molecules, supra- a variety of bridges and sizes are now very
ion with high affinity. In many instances, molecular chemistry leverages multiple useful molecular containers not only for
targeted recognition is necessary to remove weaker interactions, including hydrogen metal ions but for other cations, as well as
an unwanted species from some site because bonds, electrostatic attractions, and p-p anions and neutral molecules.
of its overabundance, to remove contamina- stacking forces. Supramolecular anion co- Anion coordination has become a vibrant
tion, or even to capture it for its economic ordination often involves hydrogen bonds, field of supramolecular chemistry, with ap-
plications that include anion
sensing, transport, anion-
A timeline of chloride ion cages based catalysis, materials
Cryptand hosts for encapsulated chloride ions have developed from the first reported example of a cryptand anion host in 1968 (1) chemistry, and separations
to the triazole-based cryptand reported by Liu et al. (6). Only several key cryptands have been selected, including widely used amines (1, 5). Selective targeting of
(1 to 3), an amide (4), a mixed calixpyrrole-amide (5), and a triazole cryptand that uses C-H bonds as hydrogen-bond donors (6). halides is especially chal-
lenging because of their
spherical charge distribu-
tion, so control of cavity size
becomes a critical factor in
NH+ +
host design. For example,
HN
the cavity size of Lehn’s “bis-
tren” cryptand is structurally
well suited for Cl– (structure
1 1968 2 1984 3 2000 2) but a bit large for F– (6).
A tinier octaaza cryptand of
Lehn and co-workers bound
F– very tightly (7) but was
deemed too small to bind
larger halides. This supposi-
tion was dispelled when the
crystal structure of encapsu-
lated Cl– ion was reported
(structure 3) (8).
The use of other hydrogen
bond–donating functional
4 2003 5 2014 6 2019 groups was then explored to
further tune anion selectiv-
Carbon Chlorine Nitrogen Oxygen Hydrogen ity and affinity. A hexaamide
cryptand (9) similar in size to
value (e.g., gold from seawater) (1). Nega- but the cryptand cage reported by Liu et al. Lehn’s bis-tren cryptand encapsulates F– with
tively charged ions, especially smaller an- is notable not only in that nontraditional high affinity and remains intact in solution.
ions such as Cl–, are especially troublesome. C-H groups are used as hydrogen-bond do- The same cryptand’s affinity for Cl– is high
Greater energies are generally needed to peel nors, but also that exceptionally effective enough for it to scavenge the small amounts
away their more tightly held hydration shells binding of Cl– is achieved. of HCl frequently present in chloroform, de-
GRAPHIC: KELLIE HOLOSKI/SCIENCE

relative to larger anions of the same charge, Supramolecular anion coordination spite a seemingly nonoptimal structural fit
such as I–, which has a more diffuse charge can be traced back to a largely overlooked (structure 4). By combining different func-
cloud. Likewise, anions possess higher free seminal paper published in 1968 by Park tional groups in the same cryptand, Ko et al.
energies of hydration than cations of similar and Simmons (3), who reported that bicy- obtained a strapped calix[4]pyrrole bridged
size and charge. On page 159 of this issue, clic diaza katapinands could encapsulate by a pyridine dicarboxamide–containing
Cl– through in,in-orientations of their two chain (structure 5) (10). This mixed-group
Department of Chemistry, University of Kansas, Lawrence, KS protonated amines (structure 1). Dietrich, cryptand promoted coupled Cl– and Na+
66045, USA. Email: kbjames@ku.edu Lehn, and Sauvage soon reported similar transport in cells.

Chloride affinities of these earlier ARTIFICIAL MUSCLES
cryptands (3, 6–9) are not, however, as high
as for the triazole cryptand of Liu et al., even
taking into account the large solvent effect
on binding variabilities. Only a few reported
Stronger artificial muscles,
receptors have affinities exceeding the 109
M–1 necessary to overcome the strong hydra-
with a twist
tion energy of Cl– in water. However, ligand
design has reached a point where high bind-
Tailored fiber-shaped actuators with twisted and coiled
ing affinities, in addition to other desirable designs have high energy densities
qualities such as on-off switches and sensors,
are possible with supramolecular chemistry.
A key attribute of Liu et al.’s host, in addition By Sameh Tawfick1,2 and Yichao Tang1 uli in the form of electricity, heat, or chemis-
to its bicyclic nature (structure 6), is that try transform the material’s microstructure
H
the bridging CH donors (triazoles) impart a eat engines have long been used for and generate large motion (strain) and forces
more rigid cavity than many other cryptands transportation, but electric motors, (stress) (see the second figure). These actua-
reported previously. This design builds on which deliver clean mechanical ac- tors are made from lightweight polymers, so
work reported by Flood and co-workers that tuation and come in a wide range they deliver energy densities more than 50
used the triazolo CH donor group in chelat- of sizes, have enabled the spread of times that of skeletal muscles and can lift
ing foldamers (11) that could bind and release automated motion used, for example, more than 1000-fold their own weight. All
Cl–, and a triazolo macrocycle (12) that in pumps, compressors, fans, and escalators. have similar twisted or coiled shapes and can
could bind Cl– effectively, if not as strongly, Nonetheless, natural muscles, their biologi- be actuated by heating. Mu et al. and Kanik
as the cryptand. cal counterparts, are far more ubiquitous. et al. also used spontaneously coiled fibers
The new class of cryptands possesses Human bodies have more than 600 muscles that contract in the axial direction when
incredibly high attomolar affinities for Cl– that drive functions such as heartbeat, fa- stimulated. Contraction is also used by skel-
extraction into dichloromethane and poten- cial expressions, and locomotion. There are etal muscles because it eliminates the risk of
tially could inhibit Cl–-induced corrosion many opportunities for expanding the use buckling encountered if the muscles were to
of iron. Indeed, the many cryptand design of actuators that mimic muscles by directly do work by pushing against a weight.
possibilities now allow for fine-tuning the using electric, thermal, or chemical energy Mu et al. tailored the cross section of
complementarity of the cryptand cavities to to generate motion and enable more perva- polymer yarns by selectively forming an ac-
targeted guests, ensuring expansion of the sive automation. In this issue, on pages 150, tive sheath that encloses a passive core to
use of these supramolecular cages in future 155, and 145, Mu et al. (1), Yuan et al. (2), boost the performance of the muscles. The
applications. Targeted recognition of mo- and Kanik et al. (3), respectively, describe outer sheath expands in response to ex-
lecular and ionic species impacts the vast new types of fiber-shaped artificial muscles ternal stimuli. They used several different
arena of chemical fields, such as biology, that exploit advantages derived from the polymers as sheath materials to respond
medicine, agriculture, engineering, and ma- mechanics of twisted and coiled geometries. to various stimuli. For example PEO-SO3 [a
terials technology, and in general the global These fiber-shaped actuators exploit tai- blend of poly(ethylene oxide) and a copoly-
environment. The results of Liu et al. are an lored materials composition and architecture mer of tetrafluoroethylene and sulfonyl fluo-
excellent example of how design concepts in within the fiber diameter (see the first figure) ride vinyl ether] responds to ethanol vapor.
the field of supramolecular chemistry have to achieve record performance. External stim- Polyurethane responds to heat, and carbon
been refined in the past 50 years. Many
more exciting advances in anion coordina-
tion and the entire area of supramolecular Composition and architecture of artificial muscle fibers
chemistry are yet to come. j Three designs for coiled and/or twisted actuators are illustrated schematically. The dashed box shows
the region presented as exploded views below, where the circles are cross-sectional views.
1. J. Kemsley, Chem. Eng. News 94, 18 (2016).
2. Y. Liu, W. Zhao, C.-H. Chen, A. H. Flood, Science 365, 159 Stimuli-responsive
(2019). actuation
3. C. H. Park, H. E. Simmons, J. Am. Chem. Soc. 90, 2431 In all three designs, the fber
(1968). undergoes twist to store
4. B. Dietrich, J. M. Lehn, J. P. Sauvage, Tetrahedron Lett. 10, energy that enables motion.
2889 (1969).
5. N. Busschaert, C. Caltagirone, W. Van Rossom, P. A. Gale,
Chem. Rev. 115, 8038 (2015).
6. B. Dietrich, J. Guilhem, J.-M. Lehn, C. Pascard, E. Sonveaux,
Helv. Chim. Acta 67, 91 (1984). Design diferences
7. B. Dietrich, J.-M. Lehn, J. Guilhem, C. Pascard, Tetrahedron The design of Kanik et al. (left)
Lett. 30, 4125 (1989). uses a Janus fber that has
8. M. A. Hossain, J. M. Llinares, C. A. Miller, L. Seib, K. thermoplastic polymer on one
GRAPHIC: V. ALTOUNIAN/SCIENCE
Bowman-James, Chem. Commun. (Camb.) 2000, 2269 side and an elastomer on the
(2000). other side. The design of Yuan
9. S. O. Kang, J. M. Llinares, D. Powell, D. VanderVelde, K. et al. (middle) uses a shape-
Bowman-James, J. Am. Chem. Soc. 125, 10152 (2003). memory polymer reinforced
10. S.-K. Ko et al., Nat. Chem. 6, 885 (2014). with graphene oxide platelets.
11. S. Lee, Y. Hua, A. H. Flood, J. Org. Chem. 79, 8383 (2014). The design of Mu et al. (right)
12. Y. Li, A. H. Flood, Angew. Chem. Int. Ed. 47, 2649 (2008). uses a stimuli-responsive
sheath surrounding a twisted
10.1126/science.aax9369 yarn core.

nanotube (CNT) sheaths swell in response ometry of the graphene oxide platelets is chanical behavior between the PE and COCe
to electrochemical charges. critical to this enhanced performance as causes the fiber to contract when thermally
The inner core provides restoring stiff- they undergo substantial bending, folding, stimulated and produces maximum work
ness and directs the twisting motion of the and twisting within the fiber to store a large capacity of 7.42 kJ/kg by heating to merely
fiber. Mu et al. demonstrated the use of amount of elastic energy during the twist- ~40°C. The process to make these fibers
several yarns as a passive core, including programming step. Moreover, the graphene could scale to industrial production.
twisted CNTs, polyacrylonitrile, silk, and oxide platelets offer critical strength and The mechanism leading to the extremely
bamboo. The fiber composition is carefully toughness benefits that allow more me- energy-dense actuation in these three reports
tailored to boost the performance compared chanical energy to be stored within the fiber is still unclear. The key factor is the elastic
with existing uniform fishing-line muscles before it fails by fracture. These enhance- energy stored during the twisting and coil-
(4) by placing the stimuli-responsive mate- ments can perhaps be tracked to prior work ing process that is reminiscent of mechani-
rial only on the outside sheath, which leads by Poulin and co-workers on the exceptional cal spring batteries, in which the stimulated
to faster responses because the diffusion toughness of nanocomposite PVA fibers (5). swelling releases some of the elastic energy
path is shorter. The swelling forces are more The fabrication process uses scalable stored during winding (6). The amount of
effective when they act on the outermost nanoparticle dispersion and wet-spinning energy that can be stored is related to the
surface of the yarn, where the local fiber techniques. This thermally driven nano- toughness of the fibers—the energy to fail-
bias is maximum. This material distribution composite SMP fiber delivers a record work ure. This heterogeneous composition of each
simultaneously reduces the mass density of density of ~2.8 kJ/kg and can be manually of the three fibers enhances fracture tough-
the muscle by keeping the core yarn unfilled reprogrammed and used at least 10 times ness. The fraction of energy released upon
to increase the energy density. As a result, before failure. This actuator can potentially stimulation is also a function of the amount
these new muscles achieve 2.12 kJ/kg for be coupled to an elastic yarn-like core to of swelling of these fibers and the changes in
sorption swelling–powered PEO-SO3 sheath their elastic modulus during swelling.
muscles and 1.33 kJ/kg for polyurethane These three studies allow a glimpse
sheath muscles actuated electrothermally. Artificial muscle metrics of the future automation opportunities
The tailored fiber designs by Mu et al. Performance of actuators and artificial muscles with these materials. Mu et al. demon-
explore untapped opportunities to boost shows the trade-off between actuation stress (force strated textiles made of these fibers that
the performance of artificial muscles while generated) and strain (amount of deformation that can expand and vary their porosity when
also reducing their cost. The fiber core can results). Twisted artificial muscles have at least stimulated. The textile can be designed to
be essentially any commercial yarn that twofold lower mass density than that of other metal respond to various stimulations such as
has reasonable high torsional stiffness and and ceramic technologies. glucose sensing, which could enable bio-
strength-to-weight ratio, such as bamboo or medical applications. Yuan et al. demon-
nylon 6 fibers, and the stimuli-responsive 103 strated the use of their fiber to propel a
material is only a thin sheath-like coating on 102
small boat, and one can see these used in
the outside. Only a thin layer of relatively ex- Twisted muscle the future to propel microswimmers, es-
Actuation strain (%)
Pneumatic
pensive CNTs wrapped around a nylon yarn 101 pecially given the similarity to the flagella
was needed to enable electrochemical actua- DEA of bacteria. Kanik et al. used a nanowire
100 IPMC SMA
tion. This actuation is potentially more en- mesh sprayed on the fibers to measure
ergy efficient than thermal actuation, which Magnetostrictive their contraction by means of a piezoresis-
10–1
is limited by the Carnot cycle. tive effect, which enabled feedback control
Yuan et al. also rationally designed high- 10–2 Piezoelectric to precisely actuate these muscles.
performance microengines from twisted Although all of these three fiber actuators
nanocomposite fibers. They synthesized 10–3 are extremely energy dense, their energy ef-
10–2 10–1 100 101 102 103
the fibers from polyvinyl alcohol (PVA) as a Actuation stress (MPa) ficiency is typically <6%, which is quite low.
matrix filled with dispersed graphene oxide DEA, dielectric elastomers; SMA, shape-memory alloys;
Even with such low efficiency, they could be
platelets. PVA is a shape-memory polymer IPMC, ionic polymer–metal composites. used to propel small robots, in drug-delivery
(SMP) that can be programmed at high tem- systems, or to morph textures that do not
perature to adopt a certain shape—here, provide a restoring torque that allows it to require large mechanical work. Like natu-
a highly twisted configuration—and then actuate repeatedly like an artificial muscle. ral skeletal systems, they will use hinged
thermally quenched while the twisted shape Kanik et al. produced an artificial muscle and articulated mechanisms to facilitate
is fixed. Upon reheating to just above the fiber with a Janus-like architecture by me- complex kinematics. In the long term, these
programming temperature, the twisted fiber chanical drawing of large heterogeneous muscles could be very suitable for environ-
recovered its straight shape by rapidly un- polymer ingots (3). One side of the fiber mentally responsive morphing architectures
twisting while delivering mechanical work. consists of a cyclic olefin copolymer elasto- and kinetic materials, where new high-en-
The performance of these fiber-shaped mer (COCe), and another side is polyethyl- ergy efficiency designs are needed. j
engines is greatly enhanced by the use of ene (PE). This Janus cross section enables REF ERENCES AND NOTES
5% weight graphene oxide platelets, which the self-coiling of these muscles driven by 1. J. Mu et al., Science 365, 150 (2019).
provide several benefits. First, they increase residual stresses developed during cold 2. J. Yuan et al., Science 365, 155 (2019).
GRAPHIC: V. ALTOUNIAN/SCIENCE
the stiffness so that the fiber delivers higher drawing. PE deforms during drawing, elon- 3. M. Kanik et al., Science 365, 145 (2019).
4. M. D. Lima et al., Science 338, 928 (2012).
torque and work upon untwisting. The ge- gating at the same rate of the external draw- 5. P. Miaudet et al., Science 318, 1294 (2007).
ing, whereas COCe elastomer only partially 6. C. Lamuta, S. Messelot, S. Tawfick, Smart Mater. Struct. 27,
055018 (2018).
1
Department of Mechanical Science and Engineering, extends while storing some elastic energy
University of Illinois Urbana-Champaign, Urbana, Il 61801, during the drawing process. When released ACKNOWL EDGMENTS
USA. 2The Beckman Institute for Advanced Science and
from the drawing process, the fiber coils S.T. acknowledges partial support by NSF grant 1825758.
Technology, University of Illinois Urbana-Champaign, Urbana,
Il 61801, USA. Email: tawfick@illinois.edu spontaneously. The mismatch in thermome- 10.1126/science.aax7304

(PPoPP) conferences. Alternatively, the re-
producibility review can be outsourced to a
third party, as in the partnership between
the American Journal of Political Science
(AJPS) and the Odum Institute (9).
Yet, despite the proliferation of such ef-
forts, current computational research does
not always comply with the most basic
principles of reproducibility (10, 11). Use of
confidential data are often mentioned as a
major impediment (12). Since the end of the
1970s, with the development of computers
and a growing concern about privacy pro-
tection, legal frameworks regarding access
to sensitive personal information have been
reinforced in most countries. Though ini-
tially only direct identification (e.g., name,
address, and social security number) was
considered for defining confidentiality, the
perimeter has been gradually enlarged to
P OLICY FORUM indirect identification that points to the use
of multiple variables that potentially lead to
a risk of identification (6). Given the exten-
sion of the legal spectrum of confidential
REPRODUCIBILITY data, a growing fraction of data used in sci-
ence can fall into this category.
Certify reproducibility One possible approach to engaging in re-

producible research with confidential data
is to generate synthetic data by applying
with confidential data an information-preserving but anonymiz-
ing transformation to the initial data (13).
By contrast, all Public Library of Science
A trusted third party certifies that results reproduce (PLOS) journals request that whenever data
cannot be accessed by other researchers, a
By Christophe Pérignon1,2, Kamel confidential-data producers and which can public dataset that can be used to validate
Gadouche3, Christophe Hurlin2,4, guarantee that the code and the data used the original conclusions must be provided.
Roxane Silberman3,5, Eric Debonnel3 by a researcher indeed produce the results An alternative approach relies on improv-
reported in a scientific paper. ing the accessibility of restricted data for
M
any government data, such as sen- researchers—for example, the research pass-
sitive information on individuals’ NATURAL TENSION port proposed by the Inter-university Con-
taxes, income, employment, or In general, research is said to be reproduc- sortium for Political and Social Research, or
health, are available only to accred- ible if the researchers provide all the re- the DataTags framework developed within
ited users within a secure comput- sources such as computer code, data, and Dataverse and the multidisciplinary Privacy
ing environment. Though they can documentation required to obtain pub- Tools Project.
be cumbersome to access, such microdata lished results (3, 4). Reproducibility has the The natural tension between confidenti-
can allow researchers to pursue questions potential to serve as a minimum standard ality and reproducibility could be alleviated
that could not be addressed with only public for judging scientific claims (5, 6). Recent by using a third-party certification agency
data (1). However, researchers using confi- promising initiatives to facilitate reproduc- to formally test whether the results in the
dential data are inexorably challenged with ible research include new research environ- tables and figures of a given scientific ar-
regard to research reproducibility (2). Em- ments (e.g., Code Ocean, Popper convention, ticle can be reproduced from the computer
pirical results cannot be easily reproduced Whole Tale), workflow systems (e.g., Kepler, code and the confidential data used by the
by peers and journal referees, as access to Kurator), and dissemination platforms or researcher. A first attempt to implement
the underpinning data are restricted. We data repositories (e.g., DataHub, Dataverse, such an external certification process is
describe an approach that allows research- openICPSR, IEEE DataPort, Mendeley). The under way in France. Explicitly designed to
ILLUSTRATION: DAVIDE BONAZZI/SALZMANART
ers who analyze confidential data to signal journal BioStatistics introduced a process deal with confidential data, the reproduc-
the reproducibility of their research. It rein which an editorial board member aims ibility assessment, like the original analy-
lies on a certification process conducted to reproduce the results in a new submis- sis, is conducted within a restricted-access
by a specialized agency accredited by the sion using the code and data provided by data environment.
the authors (7). Other examples of internal
1
HEC Paris, 78350 Jouy-en-Josas, France. 2Certification Agency reproducibility assessments include the Ap- CERTIFICATION AND TRUST
for Scientific Code and Data (cascad), 78350 Jouy-en-Josas, plications and Case Studies of the Journal of In France, the Centre d’Accès Sécurisé aux
France. 3Centre d’Accès Securisé aux Données (CASD), 91120 the American Statistical Association (8) or Données (CASD) is a public research infra-
Palaiseau, France. 4University of Orléans, LEO, 45067 Orléans,
France. 5French National Center for Scientific Research the artifact evaluation process of the Prin- structure that allows users to access and
(CNRS), 75016 Paris, France. Email: perignon@hec.fr ciples and Practice of Parallel Programming work with government confidential data
0712PolicyForum.indd 127 7/8/19 12:38 PM

INSIGHTS | P O L I C Y F O RU M
under secured conditions. This center cur- pares the output with the results displayed process is supposed to be completed within
rently provides access to data from the in the tables and figures of the paper, and 2 weeks; (v) ease replication and robustness
French Statistical Institute and the French lists any potential discrepancies in an execu- tests. Once certification is completed, the
Ministries for Finance, Justice, Education, tion report. In practice, such discrepancies computer code and detailed information
Labor, and Agriculture, as well as Social Se- can arise because of typos in the manu- about the source datasets (metadata) can be
curity contributions and health data. script, numerical convergence issues, or dif- publicly posted on the Zenodo archive and
The application process for researchers ferences in software package versions. This used to facilitate additional replication and
to get access to the CASD datasets takes execution report is transferred to a cascad robustness analyses.
around 6 months and requires a presen- reproducibility editor, a senior researcher Undoubtedly, the biggest challenge for
tation of the research project before the specialized in the author’s research field, any new certification service is to build
French Statistical Secrecy Committee, which who ultimately decides on the reproduc- trust. To build trust and increase credibil-
gathers data producers. Once the accredita- ibility of the article. Lastly, a reproducibility ity, cascad implements a transparent and
tion is granted, CASD creates a virtual ma- certificate is sent to the author and is stored detailed certification process. For each cer-
chine allowing the researcher to access the in the cascad database. tification, all the actions, interactions, and
specific source datasets required for the An example of a study recently certified problems that occurred during the process
project, as well as the required statistical by cascad is one that proposes a direct mea- are recorded. Furthermore, all operations
software. Remote access to the virtual ma- sure of tax-filing inefficiency in French co- carried out within the virtual machine by
chine is made possible thanks to a specific habiting couples (14). The analysis relies on the reviewer are recorded and traceable.
piece of hardware provided by CASD, which the Echantillon Démographique Permanent, Once the reviewing process is over, the envi-
includes a fingerprint biometric reader. an administrative dataset only available ronment is closed, sealed, and archived. The
Since the inception of CASD in 2010, the through CASD that combines information recorded operations and output can be ref-
question of allowing journal referees to get from birth, death, and marriage registers; erenced externally and shared with journal
access to data used by researchers has been electoral registers; censuses; tax returns; editors after proper accreditation.
heavily debated. However, both the legal and pay slips. All the tables and figures of Trust by data producers makes the pro-
framework and the technical restrictions the paper have been reproduced by a cascad cess feasible, and trust by academic journals
made intermittent and short-period access reproducibility reviewer from the source makes the process useful and worthwhile.
for referees difficult. datasets, and Python scripts provided by Cascad has the trust of data producers at
The Certification Agency for Scientific the authors [see certificate (15)]. CASD, who want their data to be useful to
Code and Data (cascad, www.cascad.tech) is While complying with strict data confi- society and accessible for reproducibility.
a not-for-profit certification agency created dentiality rules, the cascad-CASD partner- Now cascad needs to convince researchers
by academics (all coauthors of this paper) ship offers several advantages: (i) signal and journals to value its certificates. Over-
with the support of the French National research reproducibility when data are con- all, the experience with cascad thus far sug-
Centre for Scientific Research, foundations, fidential. The author can transfer the repro- gests that preserving confidentiality and
universities, and local governments. During ducibility certificate to an academic journal privacy does not necessarily have to lead to
initial meetings between cascad and CASD when submitting a new manuscript, similar opaque and nonreproducible research. j
teams, they quickly understood the mutual to the reproducibility badges introduced by
benefit of joining forces to design a repro- the Association for Computing Machinery;
1. L. Einav, J. Levin, Science 346, 1243089 (2014).
ducibility certification process for confiden- (ii) outsource reproducibility review. Exter- 2. C. Lagoze, L. Vilhuber, Chance 30, 68 (2017).
tial data. Thanks to this partnership, cascad nal certification enriches the peer review 3. J. B. Buckheit, D. L. Donoho, in Wavelets and Statistics,
was granted a permanent accreditation by process, but this extra step is outsourced A. Antoniadis, G. Oppenheim, Eds., Lecture Notes in
Statistics (Springer, New York, 1995), vol. 103, pp. 55–81.
the French Statistical Secrecy Committee by academic journals, as in the AJPS-Odum 4. V. Stodden et al., Science 354, 1240 (2016).
to all 280 datasets available on CASD. This partnership. The staff of the certification 5. G. Christensen, E. Miguel, J. Econ. Lit. 56, 920 (2018).
first-of-its-kind accreditation was motivated agency is specialized and has more time 6. Y.-A. de Montjoye et al., Sci. Data 5, 180286 (2018).
7. R. D. Peng, Science 334, 1226 (2011).
by the fact that cascad was providing a solu- than editorial teams at academic journals; 8. M. Fuentes, Amstat News (2016); http://magazine.
tion to the long-recognized problem of the (iii) provide economies of scale to the re- amstat.org/blog/2016/07/01/jasa-reproducible16/.
lack of reproducibility of research based on search community. This model connects a 9. T.-M. Christian, S. Lafferty-Hess, W. G. Jacoby, T. Carsey,
Int. J. Digit. Curation 13, 114 (2018).
confidential data. Also key for the approval data-provision organization (CASD, a single 10. A. C. Chang, P. Li, Am. Econ. Rev. 107, 60 (2017).
was that the whole certification process re- entry point to a large number of data pro- 11. V. Stodden, J. Seiler, Z. Ma, Proc. Natl. Acad. Sci. U.S.A. 115,
mains within the CASD environment and ducers) and a reproducibility certification 2584 (2018).
12. J. K. Harris et al., PLOS ONE 13, e0202447 (2018).
that no data can ever be downloaded. organization (cascad, a single entry point to
13. B. E. Shepherd, M. Blevins Peratikos, P. F. Rebeiro, S. N.
When an author requests a cascad cer- a large number of journals and researchers). Duda, C. C. McGowan, Am. J. Epidemiol. 186, 387 (2017).
tification for a paper, he or she needs to The cascad-CASD model is a generalization 14. O. Bargain, D. Echevin, N. Moreau, A. Pacifico, working
provide the paper, the computer code used of the standard reproducibility process in paper (2019); https://doi.org/10.5281/zenodo.3241310.
15. https://doi.org/10.5281/zenodo.3256633.
in the analysis, and any additional informa- which a single researcher goes through the
tion (software version, readme files, etc.) whole reproducibility process on his or her ACKNOWL EDGMENTS
required to reproduce the results. Then, a own, obtaining similar data and redoing the We are grateful to the anonymous referees and to conference
reproducibility reviewer, who is a full-time analysis; (iv) speed up reproducibility re- participants at the 2018 Conference of European Statistics
Stakeholders (Bamberg, Germany) and 2019 Advances in
cascad employee specialized in the software view. Any skeptical researcher can still seek Social Sciences using Administrative and Survey Data (Paris,
used by the author, accesses a CASD virtual to reproduce the work himself or herself, but France) for their comments. We thank the French Statistical
machine that is a clone of the one used by unlike researchers who need to go through a Secrecy Committee and its president Jean-Eric Schoettl, the
French National Research Agency (ANR-10-EQPX-17), the
the author. It includes a copy of the source 6-month application process, cascad review- French National Center for Scientific Research (CNRS), the
datasets and of the author’s computer code, ers benefit from a fast-track process (2 days) region Centre-Val de Loire, and the HEC Paris Foundation for
as well as all software required to run the to access any data necessary to conduct the their support.
code. The reviewer executes the code, com- reproducibility assessment. The certification 10.1126/science.aaw2825
0712PolicyForum.indd 128 7/3/19 2:13 PM

Ducipsap erspelit ut faccat as nobit vitiunt et
B O OKS et al . magniam volorro rercili quost, sandit is quasint
HISTORY OF SCIENCE
Thwarting
and exploiting
Nazi science
A thrilling tale of wartime
derring-do meets a richly
researched story of postwar
intellectual exploitation
By Charlie Hall
I
n December 1944, former Major League
Baseball (MLB) catcher Moe Berg sat in MLB catcher Moe Berg was
a freezing Zurich lecture theater listening tasked with assassinating
to a talk by Germany’s leading nuclear Germany’s leading nuclear
physicist, Werner Heisenberg. Berg was physicist, if necessary.
not there merely as an interested audi-
ence member; he had been tasked by the U.S.
Office of Strategic Services—forerunner to the have turned out had the Germans developed Third Reich in early 1945, he received a
CIA—with determining how close the Nazis it before the Allies. The story Kean tells is of checklist that, among other things, “recom-
were to developing an atomic bomb and, if the efforts made by the Allies to prevent this mended he update his will and pay up his
necessary, assassinating Heisenberg with the from happening, from the aforementioned life insurance.” Kean imagines Goudsmit
pistol concealed within his coat. Fortunately assassination plot to the endeavors of the thinking that he “might as well call his wife
for Heisenberg, Berg’s assessment was that so-called “Alsos” teams, which raced through right now and tell her he was a goner.” The
the Nazis still had some way to go, and the Europe after D-Day, seizing (and occasionally line between historical fact and Kean’s fer-
oblivious physicist survived his Swiss speak- destroying) Nazi atomic research facilities tile imagination is not always as clear as it
ing engagement. This unlikely tale is one of and equipment and arresting any German could be, however, and the book’s lack of
many that appear in Sam Kean’s The Bastard physicists they could find. references and relatively succinct bibliogra-
Brigade, which documents the efforts made Kean’s delivery is breezy, light, and of- phy do little to help the reader assess the
by the Allies to sabotage the Nazi atomic ten humorous (which can make the occa- authenticity of the story’s details.
bomb project during the Second World War. sional references to the Holocaust or Nazi The Bastard Brigade is an eminently read-
As that story suggests, the book is a grip- atrocities feel jarring), and his passion for able book, and its fast pace and wartime
ping read. Alongside Berg, a polyglot stu- history and for physics comes across on al- thriller tropes certainly commend it as an
dent of physics and classics once described most every page. He also has a knack for undemanding vacation read. It should, how-
as “the brainiest guy in baseball,” the book is conjuring up imagined conversations, and ever, be treated as an intriguing gateway to
populated by a cast of interesting and often even private thoughts and motivations, for a little-known piece of history and not as an
eccentric characters: Joe Kennedy Jr., the his characters, which breathe life into these authoritative account on the subject.
older brother of the future president; Irène historical accounts. He writes, for example, When the war ended, it was clear that the
Joliot-Curie, the Nobel-winning chemist and that when Samuel Goudsmit was prepar- Nazi nuclear program had been but a shadow
daughter of Marie Curie; and Samuel Goud- ing for his first voyage into the crumbling of its Anglo-American equivalent, the Man-
smit, a Jewish Dutch-American physicist who hattan Project, but in other fields, such as
led the investigations into the Nazi atomic rocketry or submarine technology, Nazi sci-
PHOTO: BETTMANN/CONTRIBUTOR/GETTY IMAGES
bomb while also attempting to save his par- ence had outstripped that of the victorious
ents from a concentration camp. Allies. This is where Douglas O’Reagan’s Tak-
The fascinating stories are delivered with ing Nazi Technology picks up, exploring the
the narrative aplomb of an accomplished efforts of the United States, Britain, France,
novelist. Indeed, in the Author’s Note, Kean and the Soviet Union to exploit the scientific
admits that when planning a book, he looks spoils of the defeated Third Reich and how
for “rip-roarin’ stories first and foremost,” they changed the nature of international
and he has certainly found one. technology transfer in the process.
The Bastard Brigade Taking Nazi Technology
The Nazi atomic bomb remains a fascinat- Douglas M. O’Reagan O’Reagan’s book is a work that is utterly
Sam Kean
ing historical idea, not least when we consider Little, Brown, 2019. Johns Hopkins University rooted in scholarly rigor and a close engage-
how different the war, and the world, might 460 pp. Press, 2019. 294 pp. ment with both primary source material
0712Books.indd 129 7/3/19 2:20 PM

INSIGHTS | B O O K S
across three countries and extensive existing development, intellectual property law, and
literature. As such, it provides a wide-ranging the relationships among science, business,
view of the scientific and technological ex- and the state never stagnate.
ploitation carried out by all four of the pow- Nonetheless, there are places where an
ers that occupied Germany in 1945, without anecdote of the sort on which Kean relies so
losing depth, nuance, or historical context. heavily would illuminate the ramifications
The story it tells is at least as fascinating as of the policies under discussion or showcase
that in The Bastard Brigade, but Taking Nazi some of the absurdities that the process of
Technology also interrogates the very concept exploitation so frequently produced. It is, af-
of international technology transfer and asks ter all, a remarkable historical phenomenon,
important questions about the processing of unprecedented in scale and boldness. This
huge quantities of information and the value can sometimes be forgotten when reading
of so-called “know-how.” As O’Reagan points O’Reagan’s account.
out, this is a story that has not been widely It seems at first that these two books have
told before, and where it has been, its telling relatively little in common. The events dis-
has generally been uneven, speculative, sen- cussed in The Bastard Brigade take place
sationalized, or all three. mostly during the Second World War; the key
By 1945, after the advent of ballistic mis- players are spies, geniuses, and eccentrics;
siles, jet engines, and the atomic bomb, it was and the tale is recounted with the breathless
axiomatic that technology would be central excitement of a particularly lively work of
to any future war effort. The victorious Al- fiction. Taking Nazi Technology, meanwhile,
lies sought a shortcut to scientific supremacy gives an account primarily rooted in the
through the exploitation of German tech and postwar years, in which the protagonists are
expertise, and this soon expanded to include civil servants and technical experts, and all
civilian topics as well as military ones. Labo- is couched firmly within the relevant histori-
ratories and factories were examined; docu- cal context. That said, in reading both works,
ments and equipment were confiscated; and some commonalities emerge.
scientists and technicians were detained, in- In a sense, they can be seen to present ENGINEERING
terrogated, and sometimes recruited. two sides of the same topic: Kean writes of
Perhaps the greatest asset and innovation
of Taking Nazi Technology is its coverage of
the schemes and policies of all four occupy-
dastardly Nazis and Allied derring-do, of con-
spiracies and plots and bizarre adventures,
and although he explores these traits in rela-
American
ing powers, addressing each in turn before
bringing them together to explore broader
tion to wartime atomic espionage, they can
be found just as readily in postwar techno- adventure and
trends. The Americans, we learn, were able to
hire Wernher von Braun, the man behind the
V-2 rocket, and his team and later put them
logical exploitation. O’Reagan explains the
policies and plans that underpinned these
dramatic tales and fits them into the broader
autonomous
to work on the U.S. space program; the Brit-
ish produced hundreds of reports on every
aspect of German science and industry and
historical concepts to which they relate. And
again, although his focus is on exploitation,
the central idea of the state’s mobilization of,
cars
sold them to buyers across the world; the and interactions with, science is equally rel- What will become of U.S.
French sent students to work in German lab- evant to the atomic case.
oratories to imbibe skills and experience and By reading the two works in tandem, it is
car culture in the age
to spy on their new colleagues; and the So- possible to get the benefit of both the bigger of self-driving vehicles?
viets forcibly relocated some 2500 scientific picture and the more astonishing anecdotes
workers (and their families) from their zone associated with this important moment in By Lee Vinsel
of Germany to the USSR proper in one major history. As a result, I feel confident in recom-
I
deportation in October 1946. The mass dias- mending both books, even though it would n his new book, Are We There Yet?, Dan
pora of German expertise and know-how is be fair to surmise that each is targeted at a Albert reveals how automobiles came to
one of the great untold stories of the postwar slightly different audience. be seen as a technology of freedom in
era, and as O’Reagan explains, its ramifica- The Bastard Brigade is perfect as a first America and how the ability to experi-
tions are still being felt today. foray into this period, and I defy any reader ence the world became enmeshed with
Although Taking Nazi Technology is not to be drawn into the world of unlikely personal identity. On one level, it is a gen-
PHOTO: BETTMANN/CONTRIBUTOR/GETTY IMAGES

certainly a more scholarly and more sober spies and Nazi Nobel Prize winners that eral romp through the history of the automo-
account than The Bastard Brigade, it is nei- Kean paints so vividly and infuses with such bile in the United States—and a good one at
ther dull nor inaccessible. It has been writ- energy. However, for many, this will be little that. But on a deeper level, Albert uses this
ten in such a way as to allow readers with more than a historical amuse-bouche that en- history to reflect on the near future, when
very little prior knowledge to engage with courages further reading. Here, Taking Nazi many predict that autonomous vehicles will
the narrative, and its discussions of policy Technology makes for the perfect second wipe away “car culture.”
course—meatier and more substantial but When the automobile industry first
still broad and accessible enough to suit all emerged in the 1890s in Europe and the
The reviewer is at the School of History, University of Kent, but the lightest appetites. j United States, cars were playthings for the
Canterbury, UK, and the author of British Exploitation of
German Science and Technology, 1943–1949 (Routledge, rich. Families such as the Astors, the Drexels,
2019). Email: ch515@kent.ac.uk 10.1126/science.aax7325 and the Vanderbilts amassed collections of
0712Books.indd 130 7/8/19 12:39 PM

Patrons of the 1939 New York World’s Fair
view a scale model of GM’s vision for
the future of automobile infrastructure.
European vehicles that cost many times the tionship with cars came not from consumer Are We There Yet?
average American’s annual income. desire alone but was created by powerful ac- The American
The rise of mass production, especially at tors, including automobile executives and Automobile Past,
the Ford Motor Company, changed every- government planners. In one of the book’s Present, and Driverless
thing. Henry Ford envisioned a car for aver- most original chapters, “The hidden his- Dan Albert
Norton, 2019. 400 pp.
age people, and the talented and experienced tory of the superhighways that transformed
men who worked under him reorganized America,” he provocatively shows that inter-
manufacturing, driving prices ever down- states “are not the product of Republican
ward. General Motors, under the guidance of free-market ideology but of its opposite: stat- cles (AVs) promise that this technology is on
its president, Alfred P. Sloan, took Ford’s les- ist central planning.” He locates the birth the near horizon and that it will make cars
sons even further by introducing consumer of the interstate highway vision squarely in much safer and likely decrease automobile
credit and perfecting body styling and annual the New Deal, when the designer Norman ownership. After a man died in an accident
model changes. The result, Albert writes, was Bel Geddes and other pro-planning types at- while using the autopilot feature in his Tesla
an “American automobile that made each tended a “no black tie—very informal” stag car, Elon Musk warned journalists away from
driver feel unique.” dinner at Franklin Roosevelt’s White House writing negative stories about accidents in-
Although Albert occasionally waxes nos- to build support for an ambitious national volving the technology. “If, in writing some
talgic, he is not naïve. He emphasizes that highway system. article that’s negative, you . . . dissuade people
automakers have vociferously fought off Whether revisiting 1950s visions of self- from using [AVs], you’re killing people.”
safety and pollution regulation and describes driving vehicles or 1970s fantasies of a At times, Albert seems more confident
the dangers drivers pose to others and how post-automobile society, hardly any of the about the inevitability of autonomous ve-
they fill the air with deadly gases and burn conversations we’re having today are specific hicles than I and many other auto-watchers
through finite resources. In admirably writ- to this moment. In this sense, the book’s title are. But his core concern seems to be what
ten chapters, he recounts the long history of has another layer of meaning: technofutur- this will mean for the emotional landscape of
attempts to control car culture’s dark sides, ists have been promising self-driving auto- automobiles. He writes, “When we embrace
from the racist public safety campaigns that mobiles and the death of car culture since the driverless cars, we will surrender our Ameri-
identified immigrants and African Americans mid-20th century. Are we there yet? can automobile as an adventure machine, as
as potentially unsafe drivers in the 1930s to Albert dedicates the book’s third and fi- a tool of self-expression, and the wellspring of
the work of Ralph Nader and other safety ad- nal part to exploring this question. Millen- our wealth and our defense.”
vocates, who pushed for collision science in nials appear to be less keen than previous Albert still gets a lump in his throat when
the 1950s and 1960s. generations to own cars or even get drivers’ he thinks about the death of the 1985 Saab
Albert also shows that our emotional rela- licenses, and climate change makes mass 900S that carried him across the country as
automobile use look downright irrespon- a young man (he dedicates the book to the
The reviewer is at the Department of Science, Technology, sible. According to the U.S. Environmental “honor and loving memory of” the cars he’s
and Society, Virginia Polytechnic Institute and State University, Protection Agency, transportation creates owned). “When a driverless car dies, don’t ex-
Blacksburg, VA 24061, USA, and the author of Moving about 30% of U.S. greenhouse gas emissions, pect anyone to shed a tear,” he concludes. j
Violations: Automobiles, Experts, and Regulations in the
United States (Johns Hopkins University Press, 2019). with most coming from cars and trucks.
Email: leev@vt.edu Moreover, boosters of autonomous vehi- 10.1126/science.aax8083
0712Books.indd 131 7/8/19 12:39 PM

Members and friends of
AAAS are invited to join
AAAS Travels on fascinating
trips to all 7 contintents!
WILD ALASKA
Where
& Denali Extension
Science
Gets
Mt. Denali
Social.
June 1- 6, 2020
June 26 - July 1, 2020
July 16-21, 2020
Discover Alaska on an
exciting expedition to see the
region’s iconic wildlife, epic
landscapes, and local culture.
AAAS.ORG/COMMUNITY
Hike, kayak, raft, and
paddleboard. Travel on board
National Geographic Sea Bird.
From $4,420 pp + air.
With optional 7-day Denali
extension following each cruise
$3,990 per person + air.
WILD GALÁPAGOS
ESCAPE
the
r on re!
nal ai partu T
ter 0 de a gos R
e in 2 p AAAS’ Member Community is a one-stop destination
Fre 26, 20 /Gala
ov. quil
N aya
Gu for scientists and STEM enthusiasts alike. It’s “Where
Science Gets Social”: a community where facts

August 6-13, 2020
Nov. 26 - Dec. 3, 2020 matter, ideas are big and there’s always a reason to
Galápagos is a life-list must! It is
legendary for wildlife that has come hang out, share, discuss and explore.
never developed an instinctive
flight response to humans,
making for incredible up-close
encounters and fabulous photos.
Travel aboard the National
Geographic Islander, equipped
with all the tools needed for
exploring. From $5,800 pp + air.
For a detailed brochure,
please call (800) 252-4910
All prices are per person twin share + air
BETCHART EXPEDITIONS Inc.

17050 Montebello Rd
Cupertino, California 95014
Email: AAASInfo@betchartexpeditions.com
www.betchartexpeditions.com
0712Product.indd 132 7/2/19 9:00 AM

LET TERS
Human activity has increased

along the borders of the Maasai
Mara National Reserve in Kenya.
Edited by Jennifer Sills into “upgraded” buffer zones and “down- 5. A. Mittal, E. Fraser, “Losing the Serengeti: The Maasai land
that was to run forever” (The Oakland Institute, Oakland
graded” village lands, leaving pastoralists Institute, 2018).
Conservation: Beyond with reduced landholdings and leading to
mounting pressures on remaining grazing
6. L. E. Bartels, in Land Use Competition: Ecological,
Economic, and Social Perspectives, J. Niewöhner et al.,
Eds. (Springer, 2016), pp. 149–164.
population growth areas. When the land area available to local
people shrinks because of dispossessions
7. M. Ngoitiko, M. Sinandei, P. Meitaya, F. Nelson, in
Community Rights, Conservation, and Contested Land:
The Politics of Natural Resource Governance in Africa,
In their Research Article “Cross-boundary and evictions implemented to expand pro- F. Nelson, Ed. (Earthscan, London, 2010), chap. 12, pp.
human impacts compromise the Serengeti- tected areas, more human activity becomes 269–289.
Mara ecosystem” (29 March, p. 1424), M. P. necessary in the remaining areas bordering 8. G. Bois, Past Pres. Soc. 79, 60 (1978).
9. W. Adams, Oryx 36, 213 (2002).
Veldhuis et al. argue that human population protected land.
growth in nearby areas, and the result- Veldhuis et al.’s myopic focus on popula- 10.1126/science.aax6056
ing increased human activity, is squeezing tion growth reproduces a neo-Malthusian
wildlife into existing protected areas in a explanation (8, 9) of a bygone era. Such Response
way that might lead to decline in wildlife explanations may invite the immediate
numbers throughout the ecosystem. As a attention of the general public and policy- Weldemichel et al. dismiss our argument
solution, they suggest extending the space makers due to the simplicity and sense of that human population growth drives
under protection by incorporating wildlife urgency that they communicate. However, mounting pressures around protected
migration corridors and dispersal spaces effective conservation measures demand the areas and instead propose that these
into the core protected area, thereby implic- recognition of historical and empirical com- patterns are driven through land dispos-
itly heightening restrictions on human use. plexity and the recognition and inclusion of session by authorities for conservation,
However, Veldhuis et al.’s attribution of prob- local communities’ concerns about environ- causing concerns about environmental
lems to population growth is misleading. mental justice. justice. However, population growth and
The increased human activity on the borders Teklehaymanot Weldemichel1*, the resulting increased livestock and land
of protected areas has resulted from social, Tor. A. Benjaminsen2, Connor Joseph use changes are the more likely cause of
economic, and political variables. Cavanagh2, Haakon Lein1 the trends we observed.
1
In Kenya, the rapid expansion of new Department of Geography, Norwegian University of The establishment of Mara conservan-
Science and Technology, 7491 Trondheim, Norway.
forms of conservancies has come at the 2
Department of International Environment and
cies in Kenya since 2004 [discussed in
PHOTO: STUART BLACK/ALAMY STOCK PHOTO
expense of pastoralists’ communal lands, Development Studies, Faculty of Landscape and our Research Article and in (1)] cannot be
squeezing local people into ever-smaller and Society, Norwegian University of Life Sciences, the main cause of the observed changes
NO-1432 Ås, Norway.
more marginal areas (1–3). The expansion *Corresponding author. Email: weldemichel@ntnu.no because, as our Research Article makes
of these conservancies has precipitated clear, the onset of the Mara wildlife declines
conflicts and led to widespread fencing of REF ERENCES AND NOTES predates the conservancies by about 30
remaining open areas around Maasai Mara 1. B. Butt, Hum. Ecol. 39, 289 (2011). years. In other parts of Kenya, increased
2. B. Butt, Humanity Int. J. Hum. Rights Humanit. Dev. 7, 91
(2, 3). In Tanzania, authorities have violently (2016). fencing of private lands, which also predates
forced pastoralists out of historical grazing 3. M. Løvschal et al., Sci. Rep. 7, 1 (2017). conservancies, is better explained by human
4. K. Homewood, P. Kristjanson, P. C. Trench, Eds., Staying
spaces in Loliondo to establish buffer zones Maasai? Livelihoods, Conservation and Development in population growth, increasing competition
(4–7). Pastoral lands are therefore divided East African Rangelands (Springer, London, 2009). for grazing areas, and land-use change (2, 3).

INSIGHTS | L E T T E R S
Owners of private land choose to estab-

lish wildlife conservancies (4, 5) because
Poland’s conflicting REF ERENCES AND NOTES
1. Act of 16 April 2004 on Nature Protection, Polish Journal of
Law Dz. U. No. 92 (2004); http://prawo.sejm.gov.pl/isap.
they are a viable land-use alternative in
drylands (1, 6).
environmental laws nsf/download.xsp/WDU20040920880/O/D20040880.
pdf [in Polish].
Our Research Article shows that, along Poland’s protected lands, and the spe- 2. Act of 9 June 2011 on Geology and Mining, Polish Journal
with the increased human population, cies that depend on them, are subject to of Law Dz. U. No. 163 (2011); http://prawo.sejm.gov.
total livestock numbers have increased conflicting laws that undermine their pl/isap.nsf/download.xsp/WDU20111630981/U/
D20110981Lj.pdf [in Polish].
by 54% in the Mara area, including inside sustainability. Poland’s conservation laws 3. Polesie National Park, Rules and Directions for
conservancies, matching Kenya-wide (1) apply only to land above the surface, Visitors (2019); http://www2.poleskipn.pl/index.php/
trends (2, 7). Conservancies cover 16% whereas mining laws pertain to areas zasady-korzystania-z-parku.
of the Mara area we studied, whereas underground (2). Poland must resolve this 4. BirdLife International, Acrocephalus paludicola (The
IUCN Red List of Threatened Species, 2017).
agriculture, which is expanding into drier contradiction by amending its regulations
5. G. Grzywaczewski, I. Kitowski, Oryx 52, 14 (2018).
areas (8), increased from 4.7% in 1984 to to give conservation laws priority. 6. J. T. A. Verhoeven, Ecol. Eng. 66, 6 (2014).
26.7% in 2018 in the same area (as shown One ecosystem threatened by these 7. T. J. Chmielewski, S. Chmielewski, Problemy Ekologii
in table S3 of our Research Article). contradictory regulations is Polesie National Krajobrazu XXVI, 121 (2010) [in Polish].
Increased livestock numbers, settlements, Park (3), located in Poland’s Lublin prov- 8. A. Błaszczak, J. Stochlak, Przegląd Geologiczny 32, 343
(1984) [in Polish].
and agricultural conversion, all of which ince, which is home to primeval marshes 9. P. Whittington, J. S. Price, Hydrol. Proc. 27, 1845 (2013).
are direct consequences of human popu- that serve as breeding habitats for vulner- 10. F. Tanneberger, J. Kubacka, Eds., The Aquatic Warbler
lation growth (9, 10), thus far outweigh able species such as the aquatic warbler Conservation Handbook [Brandenburg State Office for
the effect of partial livestock restrictions (Acrocephalus paludicola) (4), Europe’s rarest Environment (LfU), Potsdam, 2018].
11. Minister of Environment, “The decision for the discovery
in conservancies (11). We consistently migratory bird (5). After decades of degrada- of hard coal deposits in the area of ‘Sawin II’” (2018);
found these patterns across the entire tion, only about 20% of the primeval marshes http://geoportal.pgi.gov.pl/surowce/mapy_koncesyjne
ecosystem spanning two countries, mul- that existed 100 years ago remain (6). Most [in Polish].
tiple ethnic groups, and different types of of the bogs were destroyed by drainage and 12. M. Kuchler, G. Bridge, Energ. Res. Soc. Sci. 41, 136 (2018).
protection status. peat mining between the 1960s and 1980s 10.1126/science.aax5830
The heart of the problem is that (7). In the 1980s, the first hard coal mine
current conservation paradigms were was established (8), which led to additional
designed when the human population TECHNICAL COMMENT ABSTRACTS
in East Africa was a tenth of the cur-
rent size, and the current institutions Comment on “Long-term measles-
responsible for managing the coexistence induced immunomodulation increases
of people and wildlife have not evolved overall childhood infectious disease
accordingly (2, 8). It is an important mortality”
political and societal responsibility to Niket Thakkar and Kevin A. McCarthy
ensure that this new reality does not Mina et al. (Reports, 8 May 2015, p. 694)
increase inequality and marginalization used population-level statistical analysis to
of socioeconomically or politically weaker argue that measles infection results in a 2-
community members. Denying the to 3-year immunomodulation, implicating
importance of human population growth measles in substantially more child mortal-
in Africa as the ultimate driver of change ity than previously thought. We show, us-
only blurs discussions of environmental ing both simulation and data from Iceland,
justice and is dangerously shortsighted. that the statistical approach used may be
Joseph O. Ogutu1, Michiel P. Veldhuis2, confounded by the 2-year periodicity of
Thomas A. Morrison3, J. Grant C. Poland’s rare aquatic warbler is threatened by the measles incidence in the areas studied.
Hopcraft3, Han Olff2 degradation of its breeding habitats. Full text: dx.doi.org/10.1126/science.aax5552
1
University of Hohenheim, 70599 Stuttgart,
Germany. 2University of Groningen, 9747AG drainage of lowland bogs (9, 10), further
Groningen, Netherlands. 3University of Glasgow, Response to Comment on “Long-term
Glasgow G12 8QQ, UK. endangering the birds that depend on them.
*Corresponding author. Despite a network of international and measles-induced immunomodulation
Email: m.p.veldhuis@gmail.com national nature protection areas (1), Poland increases overall childhood infectious
continues to invest in coal mining (11, 12). disease mortality”
1. C. Bedelian, J. O. Ogutu, Pastor. Pol. Pract. 7, 1 (2017). To convince Poland’s government to Michael J. Mina, Bryan T. Grenfell, C.
2. J. O. Ogutu et al., PLOS One 11, e0163249 (2016). prioritize conservation and biodiversity over Jessica E. Metcalf
3. H. Olff, J. G. C. Hopcraft, in Serengeti III: Human Impacts destructive corporate activities, scientists Thakkar and McCarthy suggest that peri-
on Ecosystem Dynamics, A. R. E. Sinclair, C. Packer, S.
PHOTO: TOMASZ KUBIS/ALAMY STOCK PHOTO

Mduma, J. Fryxell, Eds. (University of Chicago Press, should petition the minister of the environ- odicity in measles incidence artifactually
2008), pp. 95–122. ment, and citizens groups should work drives our estimates of a 2- to 3-year
4. P. M. Osano et al., Nat. Res. For. 37, 242 (2013). together to persuade legislators to make duration of measles “immune-amnesia.”
5. D. Western, J. Waithaka, J. Kamanga, Parks 21, 51 (2015).
6. B. F. Allan et al., Front. Ecol. Environ. 15, 328 (2017).
environmentally responsible changes. We show that periodicity has a negligible
7. J. O. Ogutu et al., Open Conserv. Biol. J. 7, 11 (2013). Grzegorz Grzywaczewski1* and Ignacy efect relative to the immunological signal
8. H. Daly, Sci. Am. 293, 100 (2005). we detect, and demonstrate that immune-
Kitowski2
9. R. H. Lamprey, R. S. Reid, J. Biogeogr. 31, 997 (2004).
10. J. M. Mukeka, J. O. Ogutu, E. Kanga, E. Røskaft, Glob. Ecol.
1
University of Life Sciences in Lublin, 20-950 Lublin, amnesia is largely undetectable in small
Poland. 2State School of Higher Education in Chelm, populations with large fluctuations in mor-
Conserv. 18, e00620 (2019).
22-100 Chelm, Poland.
11. M. Y. Said et al., J. Nat. Conserv. 34, 151 (2016). tality of the type they use for illustration.
*Corresponding author.
10.1126/science.aay3049 Email: grzegorz.grzywaczewski@up.lublin.pl Full text: dx.doi.org/10.1126/science.aax6498

RESEARCH
A simplified route
to silylium ions
ons
Wu et al., p. 168
IN S CIENCE JOURNAL S Edited by Stella Hurtley
PSYCHOLOGY
How did we evolve pointing gestures?
I
n nearly every human culture, young children learn to point as a form of social communi-
cation. But how did this behavior develop? O’Madagain et al. report that when we point,
our fingers are not merely oriented in the direction of a specific object but are positioned
as though we are aiming to touch it. The authors performed a series of experiments
designed to disentangle the mechanisms involved. From infancy to adulthood, the angle
at which your finger points is less of an indication of an object’s location but is more aligned
with how you would make hand contact with an object that is out of reach. —KSL
Sci. Adv. 10.1126/sciadv.aav2558 (2019).
Pointing gestures emerged

from attempted touch.
SURFACE CHEMISTRY molecules resulting from charge- an elastomer and a semicrystal- be recovered on heating. Twisting
state changes. —PDS line polymer where the difference mechanical deformation was
Visualizing molecular
CREDITS: (GRAPHIC) Q. WU ET AL.; (PHOTO) AFLO CO., LTD./ALAMY STOCK PHOTO
Science, this issue p. 142 in thermal expansion enabled applied to the fibers above the
charging thermally actuated artificial glass transition temperature and
High-resolution atomic force muscles. Iterative cold stretch- then stored via rapid quenching.
microscopy (AFM) has been ARTIFICIAL MUSCLES ing of clad fibers could be used —MSL
used to control and image the to tailor the dimensions and
charge state of organic molecules
Getting the most mechanical response, making it
Science, this issue p. 145, p. 150,
p. 155; see also p. 125
adsorbed on multilayer sodium out of muscles simple to produce hundreds of
chloride films. Fatayer et al. Materials that convert electri- meters of coiled fibers. Mu et al.
biased an AFM probe tip with a cal, chemical, or thermal energy describe carbon nanotube yarns HOST-GUEST CHEMISTRY
voltage to charge and discharge into a shape change can be used in which the volume-changing
molecules such as azobenzene to form artificial muscles. Such material is placed as a sheath
A C–H bonding trap
and porphine from cations to materials include bimetallic outside the twisted or coiled fiber. for chloride
anions. Subsequent imaging with strips or host-guest materials This configuration can double the Part of the reason salt dissolves
carbon monoxide–functional- or coiled fibers or yarns (see work capacity of tensile muscles. so well in water is that the
ized tips revealed changes in the Perspective by Tawfick and Yuan et al. produced polymer polarized O–H bonds attract
the conformation, bond order, Tang). Kanik et al. developed a fiber torsional actuators with the the negatively charged chlo-
and aromaticity of the organic polymer bimorph structure from ability to store energy that could ride ions. It has therefore been

RESEARCH | I N S C I E N C E J OU R N A L S
common to include O–H or N–H the Perspective by Singh and Edited by Caroline Ash
bonds in molecular receptors June). Injected “amph-ligand” IN OTHER JOURNALS and Jesse Smith
designed to capture anions vaccines promoted synthetic
such as chloride. Liu et al. report antigen presentation and led to
the surprising finding that suf- CAR–T cell activation, expan-
ficiently polarized C–H bonds sion, and increased tumor
can work even better (see killing. The system could poten-
the Perspective by Bowman- tially be applied to boost any
James). They designed a cage CAR–T cell. —PNK
with triazole rings that directed Science, this issue p. 162;
C–H bonds inward to encapsu- see also p. 119
late chloride with a remarkable
attomolar affinity in dichloro-
methane. —JSY CATTLE DOMESTICATION
Science, this issue p. 159;
see also p. 124
How cow genomes
have moo-ved
Cattle were domesticated
NEUROSCIENCE ~10,000 years ago, but analy-
sis of modern breeds has not
Perceptual decision- elucidated their origins. Verdugo
making in primates et al. performed genome-
Past investigations of the wide analysis of 67 ancient
posterior parietal cortex (PPC) Near Eastern Bos taurus DNA
in perceptual decisions tested samples. Several populations of
only its contribution to motor ancient aurochs were progeni-
aspects of decisions. However, tors of domestic cows. These
Zhou and Freedman tested genetic lineages mixed ~4000
the primate PPC’s role in both years ago in a region around the
sensory and motor aspects of Indus Valley. Interestingly, mito-
decisions. Inactivation of the chondrial analysis indicated that
lateral intraparietal area strongly genetic material likely derived
impaired sensory processing from arid-adapted Bos indicus
aspects of decision-making, (zebu) bulls was introduced by
more so than motor aspects. introgression. —LMZ MILKY WAY
Lateral intraparietal area neu- Science, this issue p. 173
rons targeted for inactivation Mapping dust in three dimensions
were highly correlated with the
T
he interstellar medium contains micrometer-sized solid
monkeys’ trial-by-trial decisions ALLERGY grains, referred to as dust, that absorb and scatter
about stimuli in the neurons’
receptive fields. Thus, the pos-
A shocking contributor background starlight. Lallement et al. have combined
data from the Gaia spacecraft with ground-based
terior parietal cortex is indeed in anaphylaxis photometry to determine the distances to and dust
involved in decision-making but Classic anaphylaxis is thought absorption toward millions of stars in the Milky Way. Applying
plays a more sensory role than to depend on immunoglobu- an adaptive inversion method, they use these observations to
predicted. —PRS lin E (IgE) engagement of Fc produce a three-dimensional map of interstellar dust over a
Science, this issue p. 180 receptors on granulocytes, large part of our Galaxy. Features visible in their maps include
but allergen-specific IgE is not the Galactic plane, parts of the spiral arms, and voids where
always present. To resolve this
IMMUNOTHERAPY the dust has been evaporated. —KTS
conundrum, Jönsson et al.
A boost for CAR–T cells Astron. Astrophys. 625, A135 (2019).
IMAGE: R. LALLEMENT ET AL., ASTRON. ASTROPHYS. 625, A135 (2019)

studied samples from people
Chimeric antigen receptor who experienced anaphylaxis
(CAR)–T cell immunotherapy after anesthesia. Drug-specific False-color map of dust density in the Milky Way, looking along the Galactic
has been highly successful for IgG complexes could trigger plane. The dashed line encloses the region of best resolution.
treating certain blood cancers. neutrophils ex vivo, and anaphy-
Yet this approach has been a lactic subjects had increased
challenge for solid tumors, in neutrophil extracellular traps. MEDICINE and ulcerative colitis, can be
part because it is difficult to Not all of the subjects had distinguished by endoscopy and
target functional engineered T detectable antidrug IgE. Thus,
Blood and guts, imaging, but these methods are
cells to the tumor site. Ma et al. alternative pathways may be at but in a good way invasive and costly, particularly
designed a vaccine strategy to play, and knowledge of these Inflammatory bowel diseases for longitudinal monitoring of
improve the efficacy of CAR–T pathways could aid the develop- (IBDs) are chronic, sometimes disease progression and thera-
cells by restimulating the CAR ment of tests to reduce risks in debilitating, disorders of the peutic response. Rubin et al.
directly within the native lymph anesthesia. —LP immune system. The two developed a noninvasive, blood-
node microenvironment (see Sci. Transl. Med. 11, eaat1479 (2019). major forms, Crohn’s disease based assay for classification of

human IBDs that monitors the BIOMATERIALS in their surrounding condi- difficult to discern. This is
subtypes and functions of cir- tions, including the presence especially true if there has been
culating leukocytes that traffic
Caging cells in of toxins and pathogens larger incomplete lineage sorting or
to the gut in response to break- functional shells than 5 nanometers, and can be hybridization among species.
down of the intestinal barrier. In Unprotected cells are sensitive recovered via chelation of the One clade for which it has been
a pilot study of 68 individuals, to changes in their environment, shell. Depending on the choice of difficult to disentangle species
the authors detected immune including temperature, pH, and nanoparticle, the caged cells can relationships is the cat family.
signatures that differentiated osmotic pressure. Zhu et al. exhibit magnetic, conductive, or Examining the genomes of 27
between the two forms of IBD, have developed a facile route fluorescent properties. —MSL cat species, Li et al. showed
identified the disease state to encapsulating mammalian Adv. Mater. 31, 1900545 (2019). that the phylogenetic signal is
(flare versus remission), and cells inside functional nanopar- not constant within genomes.
provided other information ticles. A range of nanoparticles, However, relationships among
EVOLUTION
about interpatient variability in encompassing several types cat clades can be identified in
disease behavior. —PAK of metal-organic frameworks, Cats in trees regions of low recombination,
Nat. Commun. 10, 2686 (2019). mesoporous silica, and iron Studies of phylogenetic such as the X chromosome.
oxide, were mixed with interpar- relationships among species These may explain the
ticle ligands to form a shell or have been moving toward discrepancy between the fossil
NEURODEVELOPMENT
exoskeleton around HeLa cells. whole-genome analysis. record and molecular estimates
Yin and yang of hair cells The rapid complexation of the Even with full genomes, the of species diversification in
The sensory hair cells housed particles prevented uptake via evolutionary history among many clades. —LMZ
in the inner ear develop endocytotic pathways. The cells some species, especially those Mol. Biol. Evol.
sequentially in organized rows. are then protected from changes that diverged rapidly, can be 10.1093/molbev/msz139 (2019).
The progenitor cells exit the
cell cycle in order and then
differentiate in the opposite Humans selected for
order. Prajapati-DiNubila et al. dogs’ expressive eyebrows.
show that the cytokine activin
A and its antagonist follistatin
regulate expression of the
transcription factor ATOH1 and
thus of hair cell differentiation.
Activin A activated expression
of ATOH1, whereas follistatin
reduced its expression.
Similarly, follistatin signaling
regulated the number of inner
hair cells. —PJH
eLife 10.7554/eLife.47613 (2019).
INORGANIC CHEMISTRY
Silanone synthesis
Carbon dioxide and silicon diox-
ide are almost nothing alike,
despite the apparent similarity
of their chemical formulas. The
reason is that whereas carbon
forms discrete multiple bonds
to oxygen, silicon tends to form
networks of single bonds in COMPANION ANIMALS
an extended solid. Kobayashi
et al. now report that, with the How puppy dog eyes evolved
help of rather bulky aryl groups,
D
ogs are adorable. We are often amazed at just how humanlike their facial expressions can
they synthesized a silanone: a
be when they gaze into our eyes. Kaminski et al. provide an evolutionary explanation for
molecule with a Si=O double
how “puppy dog eyes” came to be, by comparing the facial muscles of dogs with their
bond analogous to a carbon-
closest living relatives, wolves. As dogs became domesticated thousands of years ago and
based ketone. The compound
began to form bonds with humans, both their anatomy and behavior developed as a result
was stable at room tem-
of selection for favorable traits. Dogs (but not wolves) harbor the levator anguli oculi medialis
perature and characterized by
muscle, which is responsible for raising the inner eyebrow. This allows dogs to produce eyebrow
PHOTO: SARAH BICKEL
crystallography but dimerized

movements that are more frequent and expressive than those of wolves, in turn giving dogs an
head-to-tail in solution upon
evolutionary advantage with humans, who instinctively respond more positively. —PNK
heating. —JSY
Angew. Chem. Int. Ed. Proc. Natl. Acad. Sci. U.S.A. 10.1073/pnas.1820653116 (2019).
10.1002/anie.201905198 (2019).

RESEARCH
ALSO IN SCIENCE JOURNALS Edited by Stella Hurtley
ANTHROPOLOGY put their metabolic and growth weak coordination to bromine demise (see the Perspective by
profiles on a healthier trajec- substituents of the carboranes. Kellam and Weiss). The results
How humans colonized tory. —CA —JSY provide a molecular explanation
the Americas Science, this issue p. 139, p. 140 Science, this issue p. 168 for the placental dysfunctions
The arrival and spread of observed in interferon-mediated
humans across the American disorders such as intrauterine
PROTEIN NETWORKS
continent is a research topic NEUROSCIENCE growth retardation, TORCH
of abiding interest. Numerous
From local space to Predicting protein pairs (toxoplasmosis, other, rubella,
archaeological finds in recent Biological function is driven by cytomegalovirus, and herpes)
years have led to a reap- higher-order maps interaction between proteins. infections, and some forms of
praisal of the timing of the first Successful movement depends High-throughput experimental preeclampsia. —BAP
occupations, before the Clovis on an accurate sense of one’s techniques have provided large Science, this issue p. 176;
culture of 13,000 years ago. location within a particular datasets of protein interactions see also p. 118
Genetic research—especially environment. Neuroscientists in several organisms; however,
genomic research over the past distinguish self-centered and much combinatorial space
5 years—also points to probable world-centered navigation and remains uncharted. Cong et al. OCEANS
earlier dates for the founder have been searching for a brain predict protein interfaces by
populations that spread from region where all ingredients identifying coevolving residues
Know the seabirds,
Beringia ~15,000 years ago. of navigation come together. in aligned protein sequences know the ocean
Waters reviews these research As rats foraged in an open (see the Perspective by Vajda Seabird observations can
advances and provides sign- field, LaChance et al. recorded and Emili). In comparison with provide a wealth of information
posts to the promise of future activity from single neurons in gold-standard and negative on ocean ecosystem health. In a
genomic studies for enriching an area called the postrhinal control sets, they show that Perspective, Velarde et al. explain
our knowledge of the ances- cortex. The authors found a the accuracy is higher than for how measurements of pollut-
tral history of humans in the population of cells that trans- proteome-wide two-hybrid and ants in seabird tissue can serve
Americas. —AMS form an animal’s immediate mass spectrometry screens. The as indicators of ocean pollution
Science, this issue p. 138 sensory perception of its envi- approach predicts 1618 protein levels. Furthermore, data on
ronment into a spatial map. This interactions in Escherichia coli, seabird breeding and feeding
map is markedly different from 682 of which were unantici- behaviors provide information
MICROBIOTA the high-level representations pated, and 911 interacting pairs about changes in the abundance
observed in hippocampal place in Mycobacterium tuberculosis, and distribution of their prey
Malnutrition and cells or entorhinal grid cells, but most of which had not been as a result of fishing or climate
dietary repair it is very flexible and is likely to previously described. With an change. Local and regional
Childhood malnutrition is provide the necessary building expected false-positive rate of programs are building on these
accompanied by growth stunt- blocks for creating higher-level between 10 and 20%, the pre- findings, but coordinated efforts
ing and immaturity of the gut representations. —PRS dicted interactions and networks to gather standardized data will
microbiota. Even after therapeu- Science, this issue p. 141 provide an excellent starting be needed to ensure that seabird
tic intervention with standard point for further study. —VV observations are used to their full
commercial complementary Science, this issue p. 185; potential. —JFU
foods, children may fail to thrive. INORGANIC CHEMISTRY see also p. 120 Science, this issue p. 116
Gehrig et al. and Raman et al.
monitored metabolic param-
An acidic route to
eters in healthy Bangladeshi silicon cations REPRODUCTION CANCER
children and those recovering The simplest silicon cation,
from severe acute malnutri- with a central Si atom bonded
Placenta formation and Chronic inflammation as
tion. The authors investigated to just three hydrogens, has fetal demise a cancer driver
the interactions between long eluded bulk synthesis. Wu A critical step of placental Immune responses toward
therapeutic diet, microbiota et al. now report a straightfor- development is the fusion of cancer cells are largely consid-
development, and growth recov- ward route to this molecule trophoblast cells into a multi- ered beneficial owing to their
ery. Diets were then designed by reacting a carborane acid nucleated syncytiotrophoblast anticancer effects. However, an
using pig and mouse models with phenyl silane, producing layer. Trophoblast fusion is emerging paradigm is that of
to nudge the microbiota into a benzene and the silylium carbo- mediated by syncytins, encoded chronic inflammation–associated
mature post-weaning state that rane ion pair. A similar protocol by endogenous retrovirus– cancers (CIACs) caused by infec-
might be expected to support offered efficient syntheses of derived envelope glycoproteins. tion or tissue injury that cannot
the growth of a child. These primary and secondary silyl Buchrieser et al. report that inter- be resolved. These cancers are
were first tested in mice inocu- cations through acidic cleavage feron-induced transmembrane often aggressive with high mor-
lated with age-characteristic of Si–phenyl or Si–H bonds. (IFITM) proteins inhibit syncytin- tality rates. In a Perspective, Tlsty
gut microbiota. The designed All three products, charac- mediated syncytiotrophoblast and Gascard discuss the distinct
diets entrained maturation of terized crystallographically formation, restricting placental pathogenesis of CIACs and how
the children’s microbiota and and in solution, manifested development and triggering fetal they might be treated more
137-B 12 JULY 2019 • VOL 365 ISSUE 6449 sciencemag.org SCIENCE

RESEARCH
effectively by therapeutically
normalizing the tumor microen-
vironment in addition to targeting
the tumor cells. —GKA
Science, this issue p. 122
T CELLS
Human resident memory
T cells hit the road
Skin and other barrier tissues
are home to long-lived tissue-
resident memory T cells (TRM)
that function as sentries capable
of rapidly responding to previ-
ously encountered antigens.
Klicznik et al. investigated the
relationship between human
skin CD4+ TRM and human blood
CD4+ memory T cells express-
ing skin-homing markers by
comparing immunopheno-
types, gene expression, and
T cell receptor sequences.
Shared phenotype, function,
and clonotypes between blood
and skin CLA+CD103+CD4+
T cells indicated that blood
CD4+CLA+CD103+ T cells were
previously skin resident. Analysis
of immunodeficient mice bearing
human skin xenografts revealed
that human skin CD4+ TRM can
exit the skin, reenter the circula-
tion, and home to secondary
human skin sites. —IW
Sci. Immunol. 4, eaav8995 (2019).
CELL BIOLOGY
Remembering the past for
CXCR4 signaling
Activation of the chemokine
receptor CXCR4 by its ligand
CXCL12 enables cancer cells to
disseminate and metastasize.
Spinosa et al. analyzed signaling
responses mediated by CXCR4
at the single-cell level. They
found that exposing cells to dif-
ferent growth-promoting stimuli
before CXCL12 exposure biased
CXCR4 signaling through either
of its downstream kinases, Akt
or ERK. Furthermore, kinase
inhibitors that are used to treat
cancer increased the likelihood
of CXCR4 activation of Akt and/
or ERK. Thus, such drugs may
inadvertently promote prometa-
static CXCR4 signaling. —WW
Sci. Signal. 12, eaaw4204 (2019).
SCIENCE sciencemag.org 12 JULY 2019 • VOL 365 ISSUE 6449 137-C

R ES E A RC H
◥ sites show that people were present and suc-

REVIEW SUMMARY cessfully occupying different areas of North and
South America between ~15.5 and ~14 ka ago,
thereby leading the way to a new understand-
ANTHROPOLOGY
ing of the first Americans.
In the last 15 years, genetic information from
Late Pleistocene exploration contemporary Indigenous Americans and the
remains of ancient individuals from Asia and
and settlement of the Americas the Americas has transformed our understand-
ON OUR WEBSITE
◥
ing of the ancestry of the
first Americans. Genetic
by modern humans Read the full article
at http://dx.doi.
studies first concentrated
on the analysis of mito-
org/10.1126/ chondrial DNA, but in the
Michael R. Waters
science.aat5447 last decade, technological
..................................................
breakthroughs have per-
BACKGROUND: North and South America radiocarbon date early sites and the belief that mitted the reconstruction of prehistoric ge-
were the last continents populated by modern the ~13,000-year-old Clovis lanceolate fluted nomes. These genomic studies have conclusively
humans. The timing of their arrival, the routes projectile points associated with mammoth re- shown that the first Americans were the result
they took, their homeland of origin, and how mains represented the first people to enter the of ancestral east Asian and northern Eurasian
they explored and settled diverse environ- continent. This view began to change with the admixture. This founder population made its
ments filled with now-extinct animals have discovery of artifacts dating ~14.2 thousand years way to eastern Beringia and after additional
been debated for over a century. Addressing (ka) ago at the Monte Verde site in southern population splits traveled south of the conti-
these questions is key to understanding the Chile. This discovery signaled that people must nental ice sheets covering Canada sometime
development of later prehistoric and contem- have been in the Americas before Clovis and between ~17.5 and ~14.6 ka ago. These genetic
porary Indigenous cultures. that early sites should be present in other parts results agree with the emerging late Pleistocene
of the Americas. Initially, many sites proposed archaeological record.
ADVANCES: The study of the first Americans to predate Clovis did not stand up to scrutiny,
made slow but steady progress during the having issues with geological context, dating, or OUTLOOK: The key to learning more about
20th century. The first half of the century even the archaeological evidence itself. How- the first Americans is investigating archaeo-
brought the realization that people had entered ever, the last 30 years have seen an increasing logical sites with solid geological contexts that
the Americas at the end of the Pleistocene. The number of sites providing evidence of early are accurately dated. Only rigorously investi-
second half of the century brought the ability to occupation that cannot be dismissed. These gated sites using the best practices of archeol-
ogy, geoarchaeology, and geochronology will
provide the primary and pivotal data to inter-
pret the past. Analysis of biomolecules, includ-
ing DNA, proteins, and lipids from these sites,
will enhance environmental reconstructions
and archaeological interpretations. This will
require time and patience because building
archaeological knowledge is a slow process.
Genetics is a powerful new tool that has al-
ready broadly deciphered the origins and popu-
lation history of the first Americans. Although
the general outline of the ancestry of the In-
digenous American genome will likely remain
unchanged moving forward, recent genetic
studies show even greater genetic complexity
ART BY GREGORY A. HARLIN, NATIONAL GEOGRAPHIC IMAGE COLLECTION

during the peopling process, especially once
people were south of the ice sheets, and this
story will surely change dramatically and
quickly with the generation of additional
genomes. The ancestral history of the earliest
peoples in the Americas will be realized as
genetic knowledge from living populations
and ancient individuals is combined with
archaeological, geological, ethnographic, and
oral records. This will require scientists and
Indigenous peoples working as partners to
0 5cm uncover the past.
▪
Page-Ladson site, Florida, ~14,550 years ago. Page-Ladson is the oldest radiocarbon-dated
Email: mwaters@tamu.edu
site in North America with artifacts of the first Americans, including a bifacial knife (inset), Cite this article as M. R. Waters, Science 365, eaat5447
found among the bones of extinct animals. (2019). DOI: 10.1126/science.aat5447

R ES E A RC H
◥ ancestral Indigenous Americans, but both popu-

REVIEW lations maintained gene flow between them
until at least ∼11.5 ka ago, which suggests their
close geographic proximity (11). This branching
ANTHROPOLOGY took place in either eastern Eurasia or Beringia
(11). If the split occurred in eastern Eurasia, then
Late Pleistocene exploration these two lineages would have moved together
or sequentially into eastern Beringia as weakly
structured populations (Fig. 1, A and B), main-
and settlement of the Americas taining gene flow between themselves but not
with Asian and Siberian populations. Alterna-
by modern humans tively, the ancestral Indigenous American popu-

lation could have entered eastern Beringia and
then AB emerged (Fig. 1, C and D), ensuring
Michael R. Waters gene flow between them but isolation from
Asian or Siberian populations.
North and South America were the last continents to be explored and settled by modern Sometime between ∼17.5 and ∼14.6 ka ago,
humans at the end of the Pleistocene. Genetic data, derived from contemporary populations groups from the ancestral Indigenous American
and ancient individuals, show that the first Americans originated from Asia and after population split into two branches: Northern
several population splits moved south of the continental ice sheets that covered Canada Native Americans (NNA) and Southern Native
sometime between ~17.5 and ~14.6 thousand years (ka) ago. Archaeological evidence Americans (SNA) (9, 11). The location of the
shows that geographically dispersed populations lived successfully, using biface, blade, divergence of the NNA and SNA branches from
and osseous technologies, in multiple places in North and South America between the ancestral Indigenous American population
~15.5 and ~14 ka ago. Regional archaeological complexes emerged by at least ~13 ka ago most likely occurred either while the groups were
in North America and ~12.9 ka ago in South America. Current genetic and archaeological migrating south from Beringia or after they had
data do not support an earlier (pre–17.5 ka ago) occupation of the Americas. entered unglaciated North America (Fig. 1, A
and C) (9, 11, 13). This is based on the fact that
T
he discovery of Folsom projectile points and Y-chromosome DNA lineages gave the first AB do not belong to either the NNA or SNA
with extinct bison and Clovis artifacts with genetic insights into Indigenous American pop- branches and are equally related to both, and
mammoth remains in New Mexico, in the ulation history (6). These studies demonstrated because there was no gene flow between AB
first half of the 20th century, established that the ancestors of all contemporary Indige- and the SNA and NNA populations. Alterna-
that people had entered the Americas at nous people had descended from only five mater- tively, the two branches may have diverged in
the end of the Pleistocene (1). Since then, more nal lineages (haplogroups A, B, C, D, and X) and eastern Beringia and then these groups migrated
Clovis sites were found and investigated, radio- two paternal lineages (haplogroups C and Q). south, but this would have required strong popu-
carbon dating placed these sites between ∼13 These lineages also showed that the founding lation structure for thousands of years to pre-
and ∼12.7 thousand years (ka) ago, and Clovis population came from Asia and experienced vent gene flow among the ancestral Indigenous
became accepted as the oldest occupation in the a severe genetic bottleneck, in which a small Americans, AB, and the NNA and SNA groups
Americas. For decades, archaeological sites pro- number of people with limited genetic diversity while in eastern Beringia (Fig. 1, C and D).
posed to predate ~13 ka ago were rejected be- gave rise to all Indigenous people who occupied The Americas were populated by members
cause they lacked artifacts, geological context, the continent before European arrival. Further, of the SNA and NNA branches. These branches
or secure dates—or had a combination of these mtDNA analyses suggested that the source pop- emerged sometime between ~17.5 and ~14.6 ka
problems (2). However, over the past 30 years, ulation from which the first Americans were ago, placing a maximum limiting age on the
archaeological investigations in both North and derived had been isolated from Asian lineages, peopling of the unglaciated lands south of the ice
South America revealed occupations predating most likely in eastern Beringia, before they dis- sheets. Analysis of mitogenomes places the arrival
Clovis that could not be dismissed (1, 3–5). In persed south. After this “Beringian Standstill” of humans into unglaciated America at ~16 ka
tandem with these archaeological discoveries, (6), a small group fissioned from this isolated ago (14), and Y-chromosome estimates place their
genetic studies of contemporary Indigenous source population, traveled south of the conti- arrival sometime between ∼19.5 and ∼15.2 ka ago
Americans and prehistoric individuals provided nental ice sheets that covered most of Canada, (15). Genetic analysis of the evolutionary history
new perspectives on the origin and population and explored and successfully populated North, of dogs, which accompanied the first Americans
history of the first Americans (6). Together, ar- Central, and South America. on their journey from Eurasia to the Americas,
chaeology and genetics are telling a coherent, Analysis of the genomes of contemporary provides additional insights about the timing of
but complex, story of the first people to enter, Indigenous populations and ancient human re- the arrival of the first Americans (16). The dogs
explore, and settle the Americas. mains has built on this framework to provide a that traveled with the first Americans originated
deeper understanding of the first American an- in Siberia and split from Siberian dogs some-
Genetic history of the first Americans cestry (6–13). These genomes show that the ances- time between ∼17.65 and ~13.7 ka ago. These
Genetic studies of contemporary Indigenous tral Indigenous American population emerged in “precontact” American dogs were south of the ice
people and ancient individuals from Asia and Eurasia, descending from a single founding group sheets by sometime between ∼16.5 and ∼13 ka ago.
the Americas reveal an outline of the ancestry that split from ancestral East Asians ∼36 ka ago, Once south of the ice sheets, the NNA branch
of the first humans to settle the Americas, pro- but maintained a high level of gene flow with East became geographically restricted to northern
viding age estimates for the timing of popula- Asians until at least ∼25 ka ago. This ancestral North America, whereas most of unglaciated
tion contact, divergence, and migration. Studies population also received gene flow from ancient North and South America was peopled by multi-
of contemporary mitochondrial DNA (mtDNA) Siberian populations with northern Eurasian an- ple groups of the SNA branch (Fig. 1E) (9, 12, 17).
cestry (Mal’ta) until ~25 to ~20 ka ago. Afterward, The earliest SNA individuals, Anzick-1 (12.85 ±
the ancestral Indigenous American population 0.05 ka), Spirit Cave (10.95 ± 0.2 ka), and Lagoa
Center for the Study of the First Americans, Department of
Anthropology, Texas A&M University, College Station, TX
became isolated from external gene flow. Santa (10.4 ± 0.1 ka), have a close genetic rela-
77843, USA. Sometime between ∼22 and ∼18.1 ka ago, tionship and form a clade (12). Analysis shows
Email: mwaters@tamu.edu Ancient Beringians (AB) branched from the that the common ancestor of Anzick-1 and Spirit
Waters, Science 365, eaat5447 (2019) 12 July 2019 1 of 9

R ES E A RC H | R E V IE W
NNA - SNA Split

South of Beringia East Beringia
22-18 ka 22-18 ka
AB AB Anzick
ANE 17.5-14.6ka ANE 17.5-14.6ka Rocky
Mtns. ASO
East East Spirit Cave
NNA Dispersal
Asians Asians
Asia
NNA & SNA - AB Split
NNA NNA
Kennewick
SNA SNA
Mixe
A B
SNA Dispersal
22-18 ka 22-18 ka
AB AB
ANE ANE 17.5-14.6ka An
17.5-14.6ka de
East Beringia
s
East East
Asians Asians
Lagoa
NNA NNA Santa
SNA Uncertain SNA

migration route NNA & SNA
Gene flow Admixture
C D E
Fig. 1. Maps showing the genetic ancestry of the Indigenous American genome. (A to D) General population history and the possible locations
where population splits and interactions took place. (E) Human dispersal, divergence, and contacts within the Americas and the location of
important prehistoric genomes. ANE, Ancient Northern Eurasians; ASO, Ancient Southwestern Ontario populations (Algonquians). [Modified
from (11) with permission]
Cave diverged from the common ancestor of Andaman Islanders (6, 8, 12, 20). This signal the context of Late Pleistocene geographic bar-
Lagoa Santa and contemporary Central American appears to be derived from an extinct ancient riers to human migration and the archaeological
Mixe sometime between ∼14.9 and 13.2 ka ago, ancestor of both groups (Population Y), but evidence left by the earliest Americans.
and that the Lagoa Santa population diverged does not represent a migration of a group of
from Mixe sometime between ∼14.8 and 12.8 ka Australasian ancestors to the Americas (20). Late Pleistocene archaeology
ago, suggesting that the movement of people The contemporary Mixe population carries a dis- of Beringia
from North to South America took hundreds or a tinctive genetic legacy from an outgroup called Yana RHS, in the Siberian Arctic, is the oldest
few thousand years. Population expansion after “Unsampled Population A,” which is neither AB, archaeological site in western Beringia (Fig. 2A)
the initial entry into the Americas is documented NNA, nor SNA and split from the ancestral Indig- (24). This ∼32,000-year-old site contains an elab-
by the rapid radiation of Y-chromosome haplo- enous American population sometime between orate osseous technology with utilitarian and
group Q-M848 within Q-M3 sometime between ∼30 and ~22 ka ago in Beringia and mixed with symbolic artifacts, along with a simple lithic
∼16.9 and ∼13.2 ka ago (15). Analysis of genetic the Mixe population during the early Holocene flake-core technology. Although Yana is impor-
variation across South America shows that when (12). The ∼5600-year-old Big Bar remains from tant to the peopling of Siberia, genomic analysis
early hunter-gatherers reached the northern edge British Columbia represent a previously undetected of human teeth (25) reveals that the people of
of South America, they advanced southward along outgroup that split from the ancestral Indige- Yana were not directly involved in the peopling
two main routes: the Atlantic and Pacific coasts nous American population in Beringia after AB of the Americas.
(12, 17, 18). Contemporary and ancient mito- diverged, but before the NNA–SNA split (12). Two sites suggest an early human presence
genomes from the western Andes detected sub- These findings show that the Late Pleistocene in eastern Beringia. From Lake E5 in northern
haplogroups that originated in South America Beringian population was not homogeneous and Alaska (Fig. 2A), human fecal biomarkers found
between ∼15.7 and ~13.5 ka ago after initial entry even suggests genetic structure between groups in lacustrine sediments suggest human occupa-
into this region (19). Geographic clustering of in- in Beringia, perhaps because they were widely tion of the region since ~32 ka ago (26). Cutmarks
dividual male sublineages of haplogroup Q-M848 dispersed. on 15 animal bones dated from ∼24 to ∼15 ka ago
in South America suggests that population struc- Genetic studies conclusively show that the first at Bluefish Caves in the Yukon are believed to be
ture emerged between ∼13.9 and ∼10.8 ka ago Americans did not originate from Europe (9, 21) the result of human activity (Fig. 2A) (27). The
(15). In North America, gene flow between the as posited by the Solutrean hypothesis (22, 23). absence of stone tools, alternative natural tapho-
NNA and SNA groups occurred before ∼9 ka ago, Genetic evidence also does not support a success- nomic explanations for the bone modification,
as documented in the genomes of Kennewick ful occupation of the Americas before ~17.5 ka ago and site formation issues render the evidence
man and ancient Algonquians (Fig. 1E) (10, 12, 13). (11). Although genetic studies have painted a from these sites equivocal (28).
The peopling of South America is further com- broad outline of the genetic ancestry of the first The first unequivocal evidence of humans in
plicated by the later entry of SNA groups without Americans, these studies do not provide a clear eastern Beringia appears at Swan Point in cen-
Anzick-1 lineage (17). picture of where population events occurred and tral Alaska (Fig. 2A). Here, Yubetsu-style wedge-
Ancient genomes also reveal several poorly provide only broad estimates for the timing of shaped microblade cores were used to make
understood population connections. The Lagoa these events. In addition, genetic populations small blades that were inset into osseous pro-
Santa individuals and some contemporary Ama- do not equate with archaeologically defined jectile points 14.15 ± 0.15 ka ago (Fig. 2E) (29).
zonian tribes share a subtle genetic connection cultures and artifact complexes. Genetically This microblade technology is derived from the
with Indigenous New Guineans, Australians, and derived interpretations must be understood in Siberian Diuktai culture of central Siberia, which

the movement of people from Beringia to un-
70°
70°
60°
°
60°
80
80
°
Yana RHS glaciated areas to the south (Figs. 2A and 3).
Nakita Lake This changed as temperatures rose at the end
Berelekh of the Pleistocene, causing ice margins to melt
Laurentide
Diuktai Cave Raven
Tuluaq Hill
Lake E5 Ice Sheet and create an inland “ice-free” corridor and a
Bluff Pacific coastal corridor along which humans
Mesa Dry Creek
could travel. Our knowledge about the opening
Upward Sun River
of these corridors is incomplete, but current
Swan Point Bluefish evidence provides a rough picture of the timing
of their development.
°
Caves
50
50
°
Ushki Co Charlie Lake The ice-free corridor was open and animals
Ice rdill Cave were traversing this passageway by ~13 ka ago.
Serpentine Sh eran
Hot Springs ee
t
The presence of bison, from a genetically distinct
Alexander
Archipelago Anzick population that developed north of the ice sheets
40 Shuká Káa during the LGM, in the central corridor at 13.15 ±
° Haida
40
° 0.15 ka ago and in Edmonton at ~13 ka ago shows
Sanak Island Gwai
Kodiak Island that the entire corridor was open by this time
Triquet and
Calvert Islands (36). Osseous artifacts made of elk at the Anzick
Vancouver site in Montana indicated that elk were present
Island Arlington
A 0 1000 km Channel
Islands
Springs in the northern plains by 12.85 ± 0.05 ka ago
(9, 37). Further, shortly after ~13 ka ago, Lake
Agassiz was draining northward through the
corridor and into the Mackenzie River valley
(38). When the entire 2000-km-long corridor
initially opened remains uncertain. Cosmogenic
10
Be dating of glacial erratics along a 500-km
length of the southernmost portion of the inte-
rior corridor shows that the Laurentide ice sheet
B C D E rapidly decoupled from the Cordilleran ice sheet
0 5 cm by 14.9 ± 0.9 ka ago (39), placing a maximum lim-
iting age on the opening of the corridor (Fig. 3).
Luminescence ages on sand dunes occupying
Fig. 2. Map of Beringia and artifacts. (A) Map showing Beringia (brown), Laurentide and Cordilleran recently deglaciated areas show subsequent
ice sheets (white), the ice-free corridor between the ice sheets, the Pacific coastal route, and the rapid retreat of the Laurentide ice margin in the
location of key archaeological and geological sites. Diuktai Cave lies in the core area of the Diuktai central portion of the corridor (40), coincident
culture. Alaskan fluted points were found in dated geological contexts at Serpentine Hot Springs with Bølling-Allerød warming from ~14.6 to
and Raven Bluff. The oldest Sluiceway points are from Tuluaq Hill; Mesa points were defined at the ~12.9 ka ago (Fig. 3). Age estimates for the open-
Mesa site. Dry Creek is the type site for the Nenana complex. (B) Base of fluted projectile point. ing of the corridor based on the analysis of lake
(C and D) Projectile points of the Nenana complex. (E) Reconstruction of osseous projectile point with sediments in the Glacial Lake Peace region (41)
inset microblades. provide underestimates for the opening of the
corridor (42). Erroneous radiocarbon ages from
the central portion of the corridor (43) provide
dates from ∼18 to ∼12.6 ka ago (29, 30) and is complexes, which together define the Northern overestimates for the opening of the corridor
found is western Beringia (31). Paleoindian tradition (33). Sluiceway projectile (42). The oldest human presence in the central
Starting ~13.8 ka ago, an assemblage charac- points are large, lanceolate forms with convex segment of the corridor is documented at Charlie
terized by small triangular or teardrop-shaped bases that first appear ∼13 ka ago at Tuluaq Hill, Lake Cave, where stone tools, including a fluted
Chindadn bifacial projectile points (Fig. 2C and Alaska (Fig. 2A), and continued to be made until point, are associated with bison radiocarbon
2D), blade and flake cores, gravers, and unifacial ∼11 ka ago. Mesa projectile points are thick, dated to 12.35 ± 0.5 ka ago (Fig. 2A) (44).
tools, but without microblades and wedge- lanceolate points with a shallow concave base The opening of a passage along the Pacific
shaped microblade cores (30, 31), is present in that date from ∼12.4 to ∼10.9 ka ago. Lanceolate coast is tied to the recessional history of the
both western and eastern Beringia. This tech- fluted projectile points have deep concave bases Cordilleran ice sheet that exposed the continen-
nology appears in Siberia at Berelekh at 13.8 ± with multiple basal flutes and date from ∼12.4 to tal shelf and extant islands. Paleotemperature
0.2 ka ago and at Nakita Lake at 13.7 ± 0.1 ka ago ∼12.0 ka ago (Fig. 2B) at the Serpentine Hot proxies from marine sediment cores in the Gulf
(Fig. 2A) (31, 32). In Alaska, this assemblage Springs and Raven Bluff sites in Alaska (Fig. 2A) of Alaska and western British Columbia show
defines the Nenana complex, which ranges from (34). No counterparts to these complexes occur that the Cordilleran ice margin began retreat-
at least ~13.5 to ∼12.7 ka ago (30). Sites with in western Beringia. Instead, at ∼13.1 ka ago, ing by ~17 ka ago (45). In the Gulf of Alaska,
microblade technology, burins, and lanceolate small-stemmed projectile points were being made radiocarbon ages from terrestrial records show
bifacial points reappear ∼12.5 ka ago. These in Kamchatka at the Ushki Lake site (35). The that Sanak Island was ice free by 15.65 ± 0.25 ka
Denali complex sites are found over large areas Alaskan fluted points and Mesa points repre- ago and Kodiak Island by 15.95 ± 0.15 ka ago
of eastern Beringia and with counterparts in sent the later movement of this technology into (Fig. 2A) (46). Cosmogenic 10Be ages show that the
western Beringia (11, 30) and are genetically Alaska from the North American plains (33, 34). western margin of the islands of the Alexander
associated with the AB population at the Upward Archipelago were free of ice by 17.0 ± 0.7 ka ago
Sun site in Alaska (11). Routes to the south: Moving from (Figs. 2A and 3) (45). Radiocarbon ages on car-
Across the northern portion of eastern Beringia, Beringia to the unglaciated Americas nivore and ringed seal bones from Shuká Káa
the earliest site assemblages are dominated by The Laurentide and Cordilleran ice sheets reached Cave on Prince of Wales Island suggest that ter-
distinctive lanceolate bifacial projectile points their maximum extent during the Last Glacial restrial and marine ecosystems were reestab-
that define the Sluiceway, Mesa, and fluted-point Maximum (LGM, ~26 to ~19 ka ago) and blocked lished by 17.2 ± 0.6 ka ago (45). Along the British

140° W 135° W 130° W 125° W 120° W

from eolian sediments at the Wally’s Beach site
in Alberta, Canada. Core and flake tools are asso-
ciated with these carcasses, which date to 13.3 ±
0.02 ka ago (57).
Along the Gulf of Mexico, the Page-Ladson site
60° N 12.7 ± 0.2 ka is buried under 4 m of sediment and is sub-
15.9 ± 0.6 ka
merged in a midchannel sinkhole along a segment
13.8 ± 0.16 ka of the Aucilla River in Florida (58, 59). Here, lithic
14.25 ± 0.4 ka
artifacts, including a biface (Fig. 4), are associated
with a human-modified mastodon tusk. Seventy-
one radiocarbon dates show that these artifacts
and modified tusk are ~14,550 years old (59).
60° N
In the northwest continental United States,
five 14,150 ± 50-year-old human coprolites, yield-
17.2 ± 0.6 ka 11.6 ± 0.45 ka
13.9 ± 1.2 ka ing mtDNA belonging to Indigenous American
11.7 ± 0.9 ka haplogroups A and B, were recovered from well-
55° N
stratified and dated deposits at Paisley Caves in
17.0 ± 0.7 ka
Oregon (60, 61). Associated with the coprolites
12.6 ± 0.05 ka
are stone tools and debitage. At the Manis site
13.15 ± 0.15 ka 12.9 ± 0.8 ka in Washington, a disarticulated mastodon skele-
14.0 ± 1.0 ka
ton was found in pond sediments. The tip of a
14.2 ± 0.3 ka 12.9 ± 0.6 ka 55° N
17.5 ± 0.1 ka 14.3 ± 0.5 ka bone projectile point is embedded into a rib of
14.3 ± 0.4 ka this animal and dates to 13.77 ± 0.02 ka ago (62).
13.04 ± 0.03 ka
Along Buttermilk Creek in central Texas at the
18.1 ± 0.2 ka Debra L. Friedkin and Gault sites, stone tools oc-
17.7 ± 0.3 ka cur beneath layers with Late Prehistoric, Archaic,
50° N Late Paleoindian, Folsom, and Clovis artifacts
(63–66). At the Friedkin site, these early artifacts
include blades, bladelets, scrapers, bifacial dis-
50° N
coidal cores, snap-fracture tools, retouched flakes,
14.9 ± 0.9 ka expedient tools, ground hematite, 11 complete
0 500 km
13.3 ka
and fragmentary lanceolate stemmed projectile
Be10 age
points, and a triangular lanceolate projectile
OSL age
point with a basally thinned concave base (Fig. 4),
C age
14
125° W 120° W 115° W 110° W
along with ∼100,000 pieces of debitage. This
assemblage occurs in deposits dated between
∼15.5 and ∼13.5 ka ago by 71 optically stimulated
Fig. 3. Map of key dated localities along the inland and coastal corridors. Inland corridor is
luminescence (OSL) ages (63, 64). At the Gault
shown in red and the coastal corridor in black. [Modified from (42) with permission]
site, five stemmed and two concave base projec-
tile points (Fig. 4) were dated using the OSL
method to ∼16 ka ago (65, 66), along with bifaces,
Columbia coast, the continental shelf between (51, 52). The earliest radiocarbon-dated sites along blades, blade cores, scrapers, gravers, and other
Haida Gwaii and the mainland was ice free and the coastal corridor include ~12,600-year-old hu- tools and ∼150,000 pieces of debitage. Points sim-
vegetated by 17.5 ± 0.1 ka ago (Fig. 2A) (47). man footprints and stone artifacts from EjTa-4 ilar to those from central Texas were excavated
Cosmogenic 10Be ages on erratics, bedrock, and on Calvert Island in British Columbia (Fig. 2A), from lacustrine deposits associated with mammoth
moraines show that ice receded from different and evidence of ~12,600 to ~12,500-year-old bear skeletons at the Santa Isabel Iztapan I and II sites
parts of Calvert Island by 18.1 ± 0.2 and 17.7 ± hunting at K1 Cave and Gaadu Din 1 Cave on in Mexico, which are bracketed by ~14,500- and
0.3 ka ago (Fig. 3) (48). Haida Gwaii (Fig. 2A) (53, 54). ~10,800-year-old tephras (Fig. 4) (67, 68).
These studies indicate that a coastal route, Along the Pacific coast of South America, at
free of barriers and biologically productive, was Late Pleistocene archaeology south of Monte Verde II in southern Chile, structural
minimally available by ∼16 ka ago (45, 48). It is the ice sheets foundations, hearths, wooden tools, lithic arti-
hypothesized that a coastal kelp forest existed from In North and South America, a number of sites facts including bipointed El Jobo projectile
Asia to the Americas during the Late Pleistocene dating between ∼15.5 and ~13.3 ka ago provide points (Fig. 4), bolo stones, medical and edible
and that people moved along this resource-rich evidence of the first human presence south of the plants, and animal bones and hides were found
“kelp highway” (49) using watercraft, but no evi- ice sheets (Fig. 4). At these sites, artifacts are on a discrete buried surface (69). Radiocarbon
dence of Late Pleistocene boat use has yet been found in undisturbed geological contexts that ages from hearths within two of the structures
found in the archaeological record of the Americas. are well dated. date to 14.2 ± 0.1 ka ago (70). At the Huaca Prieta
The earliest inferential evidence of the use of boats Near the southern margin of the Laurentide site in Peru, 42 artifacts including debitage and
comes from the 12,800 ± 100-year-old Arlington ice sheet, at the Hebior site in Wisconsin, four edge-retouched flakes and cobbles were buried
Springs human remains on the Channel Islands lithic artifacts, including two bifaces, were found within multiple discrete layers of alluvium dated
in California (Fig. 2A) (50), because a watercraft among the disarticulated bones of a mammoth from 14.15 ± 0.05 to 13.35 ± 0.05 ka ago (71).
would have been necessary to cross the ∼8 km of in pond clays dating to 14.85 ± 0.15 ka ago (55). The evidence from most of these sites has been
ocean separating the island and the mainland. Also in Wisconsin, the disarticulated remains of criticized (28, 72, 73). However, the questions raised
Alternatively, people may have traversed the another mammoth are associated with two lithic about each site have been addressed with new
northern Pacific coast on foot, subsisting on salmo- artifacts in pond clays at the Schaefer site that data to provide secure evidence that people were
nids along their travels south and using simple date to 14.65 ± 0.15 ka ago (56). Seven butchered in the Americas by ∼15 ka ago. Compelling but
watercraft to traverse waterways and other obstacles horses and one butchered camel were recovered equivocal evidence of early occupation comes

foreshafts for the hafting of stone points, and

other tools.
Clovis artifacts are found exclusively south
of the continental ice sheets and do not occur in
Asia or Beringia (Fig. 5) (77). The densest con-
centration of Clovis artifacts lies east of the
Mississippi River, but these artifacts also occur
in high frequencies west of the Mississippi River
Manis
to the eastern edge of the Rocky Mountains and
13.8 ka Hebior 14.85 ka south into northern Mexico (77–79). Securely
Schaefer 14.65 ka radiocarbon-dated Clovis sites range from ∼13 to
Paisley Caves
∼12.7 ka ago (37, 79). Two sites, Aubrey in Texas
14.15 ka Wally’s
(80) and El Fin del Mundo in Mexico (81), may
Beach
indicate that Clovis extends to ∼13.3 to ~13.4 ka
13.3 ka
ago; however, the three radiocarbon ages from
Rocky
ck Meadowcroft these sites are problematic (4, 79). If accurate,
Mtns.
ns Miles Point & though, these sites would indicate that the oldest
Parson’s Island Clovis sites occur in the southernmost portion of
Cactus Hill the Clovis range. West of the Rocky Mountains,
Clovis artifacts are sparse and have been dated
only in Arizona to 12.75 ± 0.05 ka ago (79). More
abundant in this region are “western-fluted”
points, which are morphologically distinct from
Clovis, have not been dated, and can cooccur
Page Ladson
with points of the Western Stemmed Tradition
14.55 ka
(82–84). Clovis artifacts are absent from the
Gault & Friedkin Pacific coast (77, 82, 83) and are also not found
15.5 ka in Central or South America (78, 85).
In the Intermountain West, the Western
Stemmed Tradition, characterized by lanceolate
points with basal stems, dominates the Paleo-
indian record (Fig. 5) (82, 83). The lithic tech-
An nology associated with the Western Stemmed
de Tradition is distinct and appears not to have
s
Huaca Prieta been derived from Clovis (64, 82–84, 86). The
14.15 ka earliest directly dated stemmed points occur
at Paisley Caves in Oregon and are minimally
dated ∼13 to ∼12.7 ka ago (86). No points are
associated with the ∼14,200-year-old occupation
Santa Isabel Iztapan, >13 ka at Paisley Caves (60, 61), but the lithic technol-
0 5 cm ogy represented by the oldest artifacts compares
Monte Verde favorably to the Western Stemmed Tradition. At
14.2 ka Cooper’s Ferry in Idaho, stemmed points are
Extent of Glaciation dated to ∼13.2 ka ago and possibly earlier (87).
16,000 ka At Bonneville Estates Rockshelter in Nevada,
14,000 ka Arroyo Seco 2
13,000 ka charcoal from a hearth in the deepest deposits
yielded ages of 12.85 ± 0.05 ka ago (88). No diag-
nostic artifacts were found associated with this
Fig. 4. Map of key ~15,500- to ~13,300-year-old archaeological sites in the Americas. hearth, but the technology represented by the
Diagnostic projectile points and bifaces associated with sites are shown. Rocky Mountains and debitage is more consistent with the Western
Andes Mountains are shown in purple. [Modified from (64, 102) with permission] Stemmed Tradition than with Clovis. Together,
this evidence shows that the Western Stemmed
Tradition is contemporaneous with and perhaps
from Meadowcroft Rockshelter in Pennsylvania Starting ∼13 ka ago, the first regional archae- older than Clovis.
and from Arroyo Seco 2 in Argentina (Fig. 4). ological complexes in North America emerged, In the southern cone of South America, dis-
At Meadowcroft Rockshelter, ~700 artifacts from the Clovis complex and Western Stemmed Point tinctive Fishtail projectile points occur in sites
Stratum IIa, including the lanceolate Miller pro- Tradition (Fig. 5). Clovis is identified by its dis- dated between ∼12.8 and ~12.2 ka ago (Fig. 5)
jectile point, may date between ~15 and ~14 ka tinctive biface, blade, and osseous technologies (89–92). In the deepest layers at Cerro Tres Tatas,
ago (74), but the site remains equivocal because (77). The primary trajectory of biface manufac- (90), Casa del Minero (4), and Piedra Museo (90),
of concerns about the early radiocarbon ages ture is the production of lanceolate, concave base, Argentina, lithic assemblages with cores, modified
from the site (28, 72, 73, 75). Horse and sloth fluted projectile points (Fig. 5), but also large flake tools, bifacial knives, scrapers, and choppers
remains associated with lithic artifacts at Arroyo ovate bifaces that were used as knives or cores. without projectile points date to 12.85 ± 0.05 ka
Seco 2 in Argentina suggest repeated episodes Blades were made from prepared cores and ago. At Santa Julia, Chile, a lithic assemblage that
of megafauna processing from ~14 to ~13 ka ago used without modification or were made into included a stemmed bifacial preform dates to
(76), but more information about the geological end scrapers, knives, gravers, and other tools. 12.9 ± 0.08 ka ago (93). These early sites indicate a
context and site formation processes is needed to Osseous technology includes the use of antler, human presence coeval with Clovis and may in-
make a full evaluation. bone, and ivory to make projectile points, needles, dicate that Fishtail production began ~12.9 ka ago.

~17,100-year-old Cactus Hill site in Virginia (96)

and the ~31,000- to ~25,000-year-old Miles Point
site (97) and the >14,500-year-old Parson’s Island
site (75) on and near the Delmarva Peninsula
(Fig. 4). At each site, artifacts occur in geologic
contexts that may predate Clovis, but issues re-
lated to site formation and geochronology are
unresolved (5, 28, 72, 73).
Several paleontological sites in North and
South America dating from ~130 to ~13 ka ago
with mammoth, mastodon, bison, or sloth re-
mains are suggested to be archaeological sites
Cooper’s Ferry
Anzick (e.g., 5, 28, 72, 95, 98–100). Stone tools are absent
Paisley Caves Lange- from these localities, and the evidence of human
Colby Ferguson
Bonneville Estates La Prele activity is based entirely on bone breakage pat-
Rockshelter Dent terns, interpretation of surface marks on bones,
Jake Bluff Sheriden Shawnee-Minisink
sin
nk
Cave and the spatial arrangement of the bones. These
Domebo Cactus Hill
sites are equivocal because of the absence of stone
Aubrey
Blackwater
tools and because cut marks, spiral fractures, and
Draw percussion marks on bone can be created by
Lehner
Lehn natural processes (28, 72). Some of these local-
Murra
Murray ities may indeed be archaeological sites (72), but
S in
Spr
Springs
these sites will remain equivocal until a secure
way is found to identify human interaction with
El Fin del Mundo Clovis carcasses where stone tools are absent.
Discussion
Archaeological and genetic evidence, indepen-
dently derived using distinctly different methods,
Western
Western Stemmed converges to tell a complementary story of the
Tradition first people who explored and settled the Americas
0 5 cm
at the end of the Pleistocene (Fig. 6). This evi-
dence, much of it obtained in the last few decades,
has upended long-held beliefs about the Late
Pleistocene peopling of the Americas. The archae-
ological evidence shows that geographically dis-
persed populations lived successfully and used
Santa Julia biface, blade, and osseous technologies in mul-
tiple places in North America by ~15.5 ka ago,
with the earliest artifacts appearing in South
Tigre America by ~14.2 ka ago, documenting the ini-
Cerro El Sombrero A-1 tial arrival and movement of people across the
Fishtail Cerro La China 1 & 2 continents of the Western Hemisphere (Fig. 6).
Ambrigo Los Pinos In agreement, the genetic evidence indicates that
Cueva del Medio Amalia Site 2 people were south of the continental ice sheets
Fell’s Cave Piedra Museo
Casa del Minero sometime between ~17.5 and ~14.6 ka ago and
Cerro Tres Tetas also shows that there is biological continuity be-
tween the first Americans and all Indigenous
people who followed (Fig. 6).
Fig. 5. Map of key 13,000- to 12,700-year-old archaeological sites in the Americas. The solid
Genetic studies clearly show that eastern Asia
shaded areas in North America indicate the distribution of Clovis. Colors indicate high (brown),
was the homeland of the first Americans. It is
moderate (orange), and low (yellow) densities of Clovis artifacts. All radiocarbon-dated
there that we must look for the origins of the
Clovis sites are labeled and shown with red dots. The proposed early Clovis sites, Aubrey
blade, biface, and osseous technologies docu-
and El Fin del Mundo, are shown by purple triangles. Hatched region designates the
mented in the ~15.5 to ~14 ka assemblages of
geographic extent of the Western Stemmed Tradition. Key Western Stemmed Tradition
the Americas. Although the Siberian Upper Paleo-
sites are indicated by green squares and are labeled. Dark green region in South America
lithic archaeological record shows clear linkages
designates the highest density of Fishtail projectile points, which occur in lower frequencies
to later assemblages in eastern Beringia (30),
in the light green shaded areas. Key Late Pleistocene sites and all dated Fishtail point sites
Siberian linkages to the Late Pleistocene assem-
are designated by yellow diamonds. Rocky Mountains and Andes Mountains are shown in purple.
blages south of the ice sheets are less clear.
Also shown are typical Clovis, Western Stemmed Tradition, and Fishtail projectile points.
Stronger connections to the earliest assemblages
of North America may be found in other parts of
Asia, such as Hokkaido, with its diverse Upper
Fishtail points also occur in other parts of South the LGM (5). The evidence from most of these sites Paleolithic assemblages (101). Furthermore, the
America and into Central America (78, 85). is problematic, with issues related to geologic con- known eastern Beringian assemblages are youn-
A number of sites from North and South America text, geochronology, or the absence of definitive ger than the earliest sites south of the ice sheets
are proposed to date from ~50 to ~17 ka ago, sug- human-made artifacts (1, 2, 5, 28, 72, 73, 94, 95). and may be more related to the settlement of
gesting that humans entered the Americas before The most promising older sites are the ~18,400- to eastern Beringia and unrelated to the earliest

Fig. 6. Relationship Corridors North South Debra L. Friedkin and Gault sites in Texas (64, 66).
Inland Coastal Genetics America America Stemmed points are present in the earliest assem-
among geological,
blages of South America with the ~14,200-year-old
Western
archaeological, and
Fishtail
Clovis
Stem
Western
Open
genetic data for
stemmed El Jobo points at Monte Verde, Chile,
the first Americans.
followed by stemmed Fishtail points. The Fishtail
Diagram shows time 13 ka complex of South America is minimally dated to
estimates for the
~12.8 ka ago and may date back to ~12.9 ka ago.
opening of the inland
7 It is undetermined whether this type was inde-
and coastal routes
pendently invented in South America from the
earlier biface technology or if Fishtail points are
Early South
Americans
from Beringia to
the Americas, the
descended from one or both North American
6
point traditions.
Early North Americans

maximum and
minimum estimated
14 ka The archaeological and genetic evidence shows
9
Open
5 that the peopling of the Americas was a complex
time range when 8
SNA and NNA popula-
process that we are only beginning to under-
tions entered the
stand. For the rest of this century, we need to
4 find and excavate sites of the first Americans in
Americas, and the 3
early archaeological
Beringia and across the Americas. Datable Late
2 Pleistocene sites will be difficult to find be-
record of North and
15 ka cause of issues of site preservation and visibility.
South America.
Erosional processes have removed volumes of
SNA & NNA groups south of ice sheets

Black dots indicate
the earliest archaeo-
Late Pleistocene sediments from many locations
and with it any potentially early sites. Deep burial
Time
logical sites in North 1

and South America: 1,
hampers finding early sites, and sea-level rise has
Debra L. Friedkin and
submerged the early archaeological record on the
Gault; 2, Hebior; 3,
continental shelves. However, the known Late
16 ka Pleistocene sites show that there are places where
Schaefer; 4, Page-
Ladson; 5, Paisley Caves;
this record is preserved and accessible. When found,
6, Manis; 7, Wally’s
excavated, and properly dated, the archaeological
Beach; 8, Monte Verde;
data from these sites will provide the key empirical
and 9, Huaca Prieta.
evidence needed to learn more about the first
people to enter and settle the Americas. These sites
will also yield the remains of ancient Americans.
17 ka The genomes from these individuals, especially
those tied to the archaeological record, will better
define the movement of people across the land-
scape as they settled the Americas.
Ice blocks
Finally, we must always remember that we

routes
are investigating the ancestors of contemporary

18 ka Indigenous peoples and as such, we should strive
to include Indigenous Americans in our studies
as partners in our quest to uncover their past.
Collaboration between scientists and Indigenous
occupations south of the ice sheets (52). This The archaeological evidence from North peoples will enrich our understanding of the
suggests that older sites should occur in Alaska, America shows that regionally distinct assem- story of the first Americans.
as suggested by the fecal biomarkers from Lake blages, Clovis and the Western Stemmed Tradi-
E5 (26), or that the oldest sites are submerged in tion, first appeared by ~13 ka ago as people REFERENCES AND NOTES
central Beringia or will be found in the uplifted adapted to rapidly changing climates and major 1. D. J. Meltzer, First Peoples in a New World (Univ. of California
Press, 2009).
areas of coastal Alaska and British Columbia floral and faunal reorganizations (Fig. 6). The 2. C. V. Haynes Jr., The earliest americans. Science 166,
(53, 54). origin and chronological and technological rela- 709–715 (1969). doi: 10.1126/science.166.3906.709;
Whereas both corridors into the Americas tionships between these complexes are unclear. pmid: 17776753
require more investigation, current evidence Artifacts of Clovis and the Western Stemmed 3. T. Goebel, M. R. Waters, D. H. O’Rourke, The late Pleistocene
dispersal of modern humans in the Americas. Science
favors the Pacific coast as the route taken by the Tradition are technologically distinct, and it 319, 1497–1502 (2008). doi: 10.1126/science.1153569
first Americans (Fig. 6). This is based entirely on would be difficult to derive one from the other pmid: 18339930
chronological information showing the early (82–84, 86). Evidence from sites in the Inter- 4. M. R. Waters, T. W. Stafford Jr., “The first Americans: a
opening of this corridor by ~16 ka ago, but ar- mountain West suggest that stemmed points review of the evidence for the Late-Pleistocene peopling of
the Americas” in Paleoamerican Odyssey, K. E. Graf,
chaeological evidence older than ~12.6 ka ago is may predate Clovis (86, 87). Sites in Texas show C. V. Ketron, M. R. Waters, Eds. (Texas A&M Univ. Press,
absent. The ice-free corridor was definitely open that stemmed points have deep time depth in 2014), pp. 541–560.
by ~13 ka ago and bison and other animals were North America and may indeed be the earliest 5. D. B. Madsen, A framework for the initial occupation of the
passing through it. The corridor may have been point style in North America brought by the Americas. PaleoAmerica 1, 217–250 (2015). doi: 10.1179/
2055557115Y.0000000006
open by ~14 ka ago, but if it was, it is difficult to first migrants (64, 66). There is little doubt that 6. P. Skoglund, D. Reich, A genomic view of the peopling of the
explain why bison waited 1000 years to traverse Clovis originated south of the continental ice Americas. Curr. Opin. Genet. Dev. 41, 27–35 (2016).
the corridor. More chronological information is sheets by ~13 ka ago and we should look for doi: 10.1016/j.gde.2016.06.016; pmid: 27507099
needed from the interior corridor to determine the origins of Clovis in the biface, blade, and 7. M. Raghavan et al., Upper Palaeolithic Siberian genome
reveals dual ancestry of Native Americans. Nature 505,
when it unequivocally opened, and a search osseous technologies that make up the ~15,500- 87–91 (2014). doi: 10.1038/nature12736; pmid: 24256729
for early sites should be undertaken along both to ~14,000-year-old North American assem- 8. M. Raghavan et al., POPULATION GENETICS. Genomic
corridors. blages. There are hints of this transition at the evidence for the Pleistocene and recent population history of

Native Americans. Science 349, aab3884 (2015). Yana-Indighirka lowland, Arctic Siberia. Quat. Int. 406, 52. M. Q. Sutton, The “fishing link”: Salmonids and the initial
doi: 10.1126/science.aab3884; pmid: 26198033 202–217 (2016). doi: 10.1016/j.quaint.2015.12.039 peopling of the Americas. PaleoAmerica 3, 231–259 (2017).
9. M. Rasmussen et al., The genome of a Late Pleistocene 32. V. V. Pitulko, A. Basilyan, E. Y. Pavlova, The Berelekh doi: 10.1080/20555563.2017.1331084
human from a Clovis burial site in western Montana. mammoth “graveyard”: New chronological and stratigraphic 53. D. Fedje, Q. Mackie, T. Lacourse, D. McLaren, Younger Dryas
Nature 506, 225–229 (2014). doi: 10.1038/nature13025; data from the 2009 field season. Geoarchaeology 29, environment and archaeology of the Northwest Coast of
pmid: 24522598 277–299 (2014). doi: 10.1002/gea.21483 North America. Quat. Int. 242, 452–462 (2011). doi: 10.1016/
10. M. Rasmussen et al., The ancestry and affiliations of 33. H. L. Smith, J. T. Rasic, T. Goebel, “Biface traditions of j.quaint.2011.03.042
Kennewick Man. Nature 523, 455–458 (2015). doi: 10.1038/ northern Alaska and their role in the peopling of the Americas” 54. D. McLaren et al., Terminal Pleistocene epoch human
nature14625; pmid: 26087396 in Paleoamerican Odyssey, K. E. Graf, C. V. Ketron, footprints from the Pacific coast of Canada. PLOS ONE 13,
11. J. V. Moreno-Mayar et al., Terminal Pleistocene Alaskan M. R. Waters, Eds. (Texas A&M Univ. Press, 2014), pp. 105–123. e0193522 (2018). doi: 10.1371/journal.pone.0193522;
genome reveals first founding population of Native 34. T. Goebel et al., Serpentine Hot Springs, Alaska: Results of pmid: 29590165
Americans. Nature 553, 203–207 (2018). doi: 10.1038/ excavations and implications for the age and significance of 55. D. F. Overstreet, “Late-glacial ice-marginal adaptation in
nature25173; pmid: 29323294 northern fluted points. J. Archaeol. Sci. 40, 4222–4233 southeastern Wisconsin” in Paleoamerican Origins: Beyond
12. J. V. Moreno-Mayar et al., Early human dispersals within the (2013). doi: 10.1016/j.jas.2013.05.027 Clovis, R. Bonnichsen, B. T. Lepper, D. Stanford, M. R. Waters,
Americas. Science 362, eaav2621 (2018). doi: 10.1126/ 35. T. Goebel, M. R. Waters, M. Dikova, The archaeology of Ushki Eds. (Texas A&M Univ. Press, 2005), pp. 183–195.
science.aav2621 Lake, Kamchatka, and the Pleistocene peopling of the 56. D. J. Joyce, “Pre-Clovis megafauna butchery sites in the
13. C. L. Scheib et al., Ancient human parallel lineages within Americas. Science 301, 501–505 (2003). doi: 10.1126/ western Great Lakes region, USA” in Paleoamerican Odyssey,
North America contributed to a coastal expansion. science.1086555; pmid: 12881567 K. E. Graf, C. V. Ketron, M. R. Waters, Eds. (Texas A&M Univ.
Science 360, 1024–1027 (2018). doi: 10.1126/science.aar6851; 36. P. D. Heintzman et al., Bison phylogeography constrains Press, 2014), pp. 467–483.
pmid: 29853687 dispersal and viability of the Ice Free Corridor in western 57. M. R. Waters, T. W. Stafford Jr., B. Kooyman, L. V. Hills,
14. B. Llamas et al., Ancient mitochondrial DNA provides Canada. Proc. Natl. Acad. Sci. U.S.A. 113, 8057–8063 (2016). Late Pleistocene horse and camel hunting at the southern
high-resolution time scale of the peopling of the Americas. doi: 10.1073/pnas.1601077113; pmid: 27274051 margin of the ice-free corridor: Reassessing the age
Sci. Adv. 2, e1501385 (2016). doi: 10.1126/sciadv.1501385; 37. L. Becerra-Valdivia et al., Reassessing the chronology of the of Wally’s Beach, Canada. Proc. Natl. Acad. Sci. U.S.A. 112,
pmid: 27051878 archaeological site of Anzick. Proc. Natl. Acad. Sci. U.S.A. 4263–4267 (2015). doi: 10.1073/pnas.1420650112;
15. T. Pinotti et al., Y chromosome sequences reveal a short 115, 7000–7003 (2018). doi: 10.1073/pnas.1803624115; pmid: 25831543
Beringian Standstill, rapid expansion, and early population pmid: 29915063 58. J. S. Dunbar, “Paleoindian archaeology” in First Floridians
structure of Native American founders. Curr. Biol. 29, 38. J. B. Murton, M. D. Bateman, S. R. Dallimore, J. T. Teller, and Last Mastodons: The Page-Ladson Site in the Aucilla River,
149–157.e3 (2019). doi: 10.1016/j.cub.2018.11.029; Z. Yang, Identification of Younger Dryas outburst flood path S. D. Webb, Ed. (Springer, 2006), pp. 403–438.
pmid: 30581024 from Lake Agassiz to the Arctic Ocean. Nature 464, 740–743 59. J. J. Halligan et al., Pre-Clovis occupation 14,550 years ago at
16. M. Ní Leathlobhair et al., The evolutionary history of dogs in (2010). doi: 10.1038/nature08954; pmid: 20360738 the Page-Ladson site, Florida, and the peopling of the
the Americas. Science 361, 81–85 (2018). doi: 10.1126/ 39. M. Margold et al., Beryllium-10 dating of the Foothills Erratic Americas. Sci. Adv. 2, e1600375 (2016). doi: 10.1126/
science.aao4776; pmid: 29976825 Train in Alberta, Canada, indicates detachment of the sciadv.1600375; pmid: 27386553
17. C. Posth et al., Reconstructing the deep population history of Laurentide Ice Sheet from the Rocky Mountains at ~15 ka. 60. M. T. P. Gilbert et al., DNA from pre-Clovis human coprolites
Central and South America. Cell 175, 1185–1197.e22 (2018). Quaternary Research (2019). doi: 10.1017/qua.2019.10 in Oregon, North America. Science 320, 786–789 (2008).
doi: 10.1016/j.cell.2018.10.027; pmid: 30415837 40. K. Munyikwa, T. M. Rittenour, J. K. Feathers, Temporal doi: 10.1126/science.1154116; pmid: 18388261
18. A. Gómez-Carballa et al., The peopling of South America constraints for the Late Wisconsinan deglaciation of western 61. D. L. Jenkins et al., “Geochronology, archaeological context,
and the trans-Andean gene flow of the first settlers. Canada using eolian dune luminescence chronologies from and DNA at the Paisley Caves” in Paleoamerican Odyssey,
Genome Res. 28, 767–779 (2018). doi: 10.1101/gr.234674.118; Alberta. Palaeogeogr. Palaeoclimatol. Palaeoecol. 470, K. E. Graf, C. V. Ketron, M. R. Waters, Eds. (Texas A&M Univ.
pmid: 29735605 147–165 (2017). doi: 10.1016/j.palaeo.2016.12.034 Press, 2014), pp. 485–510.
19. S. Brandini et al., The Paleo-Indian entry into South America 41. M. W. Pedersen et al., Postglacial viability and colonization in 62. M. R. Waters et al., Pre-Clovis mastodon hunting 13,800 years
according to mitogenomes. Mol. Biol. Evol. 35, 299–311 North America’s ice-free corridor. Nature 537, 45–49 (2016). ago at the Manis site, Washington. Science 334, 351–353
(2018). doi: 10.1093/molbev/msx267; pmid: 29099937 doi: 10.1038/nature19085; pmid: 27509852 (2011). doi: 10.1126/science.1207663; pmid: 22021854
20. P. Skoglund et al., Genetic evidence for two founding 42. D. Froese, J. M. Young, S. L. Norris, M. Margold, Availability 63. M. R. Waters et al., The Buttermilk Creek complex and the
populations of the Americas. Nature 525, 104–108 (2015). and viability of the Ice-Free corridor and Pacific Coast routes origins of Clovis at the Debra L. Friedkin site, Texas.
doi: 10.1038/nature14895; pmid: 26196601 for the peopling of the Americas. SAA Archaeol. Rec. 19, Science 331, 1599–1603 (2011). doi: 10.1126/science.1201855;
21. J. A. Raff, D. A. Bolnick, Does mitochondrial haplogroup X 27–33 (2019). pmid: 21436451
indicate ancient trans-Atlantic migration to the Americas? A 43. B. A. Potter et al., Early colonization of Beringia and northern 64. M. R. Waters et al., Pre-Clovis projectile points at the
critical re-evaluation. PaleoAmerica 1, 297–304 (2015). North America: Chronology, routes, and adaptive strategies. Debra L. Friedkin site, Texas-Implications for the Late
doi: 10.1179/2055556315Z.00000000040 Quat. Int. 444, 36–55 (2017). doi: 10.1016/ Pleistocene peopling of the Americas. Sci. Adv. 4, eaat4505
22. D. J. Stanford, B. A. Bradley, Across Atlantic Ice (Univ. of j.quaint.2017.02.034 (2018). doi: 10.1126/sciadv.aat4505; pmid: 30397643
California Press, 2012). 44. J. C. Driver et al., Stratigraphy, radiocarbon dating and culture 65. K. Rodrigues et al., OSL ages of the Clovis, Late Paleoindian,
23. S. Oppenheimer, B. Bradley, D. Stanford, Solutrean hypothesis: history of Charlie Lake Cave, British Columbia. Arctic 49, and Archaic components at Area 15 of the Gault site,
Genetics, the mammoth in the room. World Archaeol. 46, 265–277 (1996). doi: 10.14430/arctic1202 central Texas, U.S.A. J. Archaeol. Sci. 7, 94–103 (2016).
752–774 (2014). doi: 10.1080/00438243.2014.966273 45. A. J. Lesnek, J. P. Briner, C. Lindqvist, J. F. Baichtal, 66. T. J. Williams et al., Evidence of an early projectile point
24. V. V. Pitulko et al., The Yana RHS site: Humans in the Arctic T. H. Heaton, Deglaciation of the Pacific coastal corridor technology in North America at the Gault Site, Texas, USA.
before the last glacial maximum. Science 303, 52–56 (2004). directly preceded the human colonization of the Americas. Sci. Adv. 4, eaar5954 (2018). doi: 10.1126/sciadv.aar5954;
doi: 10.1126/science.1085219; pmid: 14704419 Sci. Adv. 4, eaar5040 (2018). doi: 10.1126/sciadv.aar5040; pmid: 30009257
25. M. Sikora et al., The population history of northeastern pmid: 29854947 67. L. A. A. De Anda, M. Maldonado-Koerdell, Association of
Siberia since the Pleistocene. Nature (2019). doi: 10.1038/ 46. N. Misarti et al., Early retreat of the Alaskan Peninsula Glacier artifacts with Mammoth in the Valley of Mexico. Am. Antiq.
s41586-019-1279-z; pmid: 31168093 Complex and the implications for coastal migration of first 18, 332–340 (1953). doi: 10.2307/277101
26. R. S. Vachula et al., Evidence of ice age humans in eastern Americans. Quat. Sci. Rev. 48, 1–6 (2012). doi: 10.1016/ 68. S. Gonzalez et al., Paleoindian sites from the Basin of
Beringia suggests early migration to North America. j.quascirev.2012.05.014 Mexico: Evidence from stratigraphy, tephrochronology and
Quat. Sci. Rev. 205, 35–44 (2019). doi: 10.1016/ 47. T. Lacourse, R. W. Mathewes, D. W. Fedje, Late-glacial dating. Quat. Int. 363, 4–19 (2015). doi: 10.1016/
j.quascirev.2018.12.003 vegetation dynamics of the Queen Charlotte Islands and j.quaint.2014.03.015
27. L. Bourgeon, A. Burke, T. Higham, Earliest human presence in adjacent continental shelf, British Columbia, Canada. 69. T. D. Dillehay, Monte Verde: A Late Pleistocene Settlement in
North America dated to the Late Glacial Meximum: New Palaeogeogr. Palaeoclimatol. Palaeoecol. 226, 36–57 (2005). Chile, Volume 2: The Archaeological Context and
radiocarbon dates from Bluefish Caves, Canada. PLOS ONE doi: 10.1016/j.palaeo.2005.05.003 Interpretation (Smithsonian Institution, 1997).
12, e0169486 (2017). doi: 10.1371/journal.pone.0169486; 48. C. M. Darvill, B. Menounos, B. M. Goehring, O. B. Lian, 70. T. D. Dillehay et al., Monte Verde: Seaweed, food, medicine,
pmid: 28060931 M. W. Caffee, Retreat of the western Cordilleran Ice sheet and the peopling of South America. Science 320, 784–786
28. G. Haynes, The Early Settlement of North America margin during the late deglaciation. Geophys. Res. Lett. 45, (2008). doi: 10.1126/science.1156533; pmid: 18467586
(Cambridge Univ. Press, 2002). 9710–9720 (2018). doi: 10.1029/2018GL079419 71. T. D. Dillehay, Ed., Where the Land Meets the Sea (Univ. of
29. Y. A. Gómez Coutouly, C. E. Holmes, The microblade industry 49. J. M. Erlandson et al., The kelp highway hypothesis: Marine Texas Press, 2017).
from Swan Point cultural zone 4b: Technological and ecology, the coastal migration theory, and the peopling 72. G. Haynes, The millennium before Clovis. PaleoAmerica 1,
cultural implications from the earliest human occupation in of the Americas. J. Island Coast. Archaeol. 2, 161–174 (2007). 134–162 (2015). doi: 10.1179/2055556315Z.00000000016
Alaska. Am. Antiq. 83, 735–752 (2018). doi: 10.1017/ doi: 10.1080/15564890701628612 73. S. J. Fiedel, “Is that all there is? The weak case for pre-Clovis
aaq.2018.38 50. J. R. Johnson, T. W. Stafford Jr., H. O. Ajie, D. P. Morris, occupation in eastern North America” in In the Eastern
30. K. E. Graf, I. Buvit, Human Dispersal from Siberian to “Arlington Springs revisited” in Proceedings of the Fifth Fluted Point Tradition, J. A. M. Gingerich, Ed. (Univ. of Utah
Beringia: Assessing a Beringian Standstill in Light of the California Islands Symposium, D. Browne, K. Mitchell, Press, 2013), pp. 333–354.
Archaeological Evidence. Curr. Anthropol. 58, 583–603 H. Chaney, Eds. (Department of the Interior, Pacific OCS 74. J. M. Adovasio, D. R. Pedler, “Pre-Clovis sites and their
(2017). doi: 10.1086/693388 Region, 1999), pp. 541–544. implications for human occupation before the Last Glacial
31. V. V. Pitulko, E. Y. Pavlova, A. E. Basilyan, Mass 51. M. Q. Sutton, Paleoindian colonization by boat? Refining Maximum” in Entering America: Northeast Asia and Beringia
accumulations of mammoth (mammoth ‘graveyards’) the coastal model. PaleoAmerica 4, 325–339 (2019). Before the Last Glacial Maximum, D. M. Madsen, Ed. (Univ. of
with indications of past human activity in the northern doi: 10.1080/20555563.2019.1565750 Utah Press, 2004), pp. 139–158.

75. J. C. Lothrop, D. L. Lowery, A. E. Spiess, C. J. Ellis, Early PaleoAmerica 3, 203–230 (2017). doi: 10.1080/ 96. J. M. McAvoy, L. D. McAvoy, Nottoway River Survey Part II:
human settlement of northeastern North America. 20555563.2017.1328953 Cactus Hill and Other Excavated Sites (Dietz, Nottoway River
PaleoAmerica 2, 192–251 (2016). doi: 10.1080/ 86. D. L. Jenkins et al., Clovis age Western Stemmed projectile Survey Research Report no. 5, 2015).
20555563.2016.1212178 points and human coprolites at the Paisley Caves. 97. D. L. Lowery, M. A. O’Neal, J. S. Wah, D. P. Wagner,
76. G. G. Politis, M. A. Gutiérrez, D. J. Rafuse, A. Blasi, The arrival Science 337, 223–228 (2012). doi: 10.1126/science.1218443; D. J. Stanford, Late Pleistocene upland stratigraphy of the
of Homo sapiens into the Southern Cone at 14,000 years ago. pmid: 22798611 western Delmarva Peninsula, USA. Quat. Sci. Rev. 29,
PLOS ONE 11, e0162870 (2016). doi: 10.1371/journal. 87. L. G. Davis, A. J. Nyers, S. C. Willis, Context, provenance and 1472–1480 (2010). doi: 10.1016/j.quascirev.2010.03.007
pone.0162870; pmid: 27683248 technology of a Western Stemmed Tradition artifact cache 98. S. R. Holen et al., A 130,000-year-old archaeological site in
77. B. A. Bradley, M. B. Collins, A. Hemmings, Clovis Technology from the Cooper’s Ferry site, Idaho. Am. Antiq. 79, 596–615 southern California, USA. Nature 544, 479–483 (2017).
(International Monographs in Prehistory, no. 17, 2010). (2014). doi: 10.7183/0002-7316.79.4.596 doi: 10.1038/nature22065; pmid: 28447646
78. G. Acosta-Ochoa, P. Pérez-Martínex, X. Ulloa-Montemayor, 88. T. Goebel, J. L. Keene, “Are Great Basin stemmed points as 99. S. R. Holen, K. Holen, “The Mammoth Steppe hypothesis: The
“The Clovis-like and Fishtail occupations of southern Mexico old as Clovis in the Intermountain West?” in Archaeology Middle Wisconsin (Oxygen Isotope Stage 3) peopling of North
and Central America” in People & Culture in Ice Age in the Great Basin and Southwest, N. J. Parezo, J. C. Janetski, America” in Paleoamerican Odyssey, K. E. Graf, C. V. Ketron,
Americas, R. Suárez, C. F. Ardelean, Eds. (Univ. of Utah Press, Eds. (Univ. of Utah Press, 2014), pp. 35–60. M. R. Waters, Eds. (Texas A&M Univ. Press, 2014), pp. 429–444.
2019), pp. 93–107. 89. C. Weitzel, N. Mazzia, N. Flegenheimer, Assessing Fishtail 100. S. M. Kenady, M. C. Wilson, R. F. Schalk, R. R. Mierendorf,
79. M. R. Waters, T. W. Stafford Jr., Redefining the age of Clovis: points distribution in the Southern Cone. Quat. Int. 473, Late Pleistocene butchered Bison antiquus from Ayer Pond,
Implications for the peopling of the Americas. Science 161–172 (2018). doi: 10.1016/j.quaint.2018.01.005 Orcas Island, Pacific Northwest: Age confirmation and
315, 1122–1126 (2007). doi: 10.1126/science.1137166; 90. L. Prates, G. Politis, J. Steele, Radiocarbon chronology of the taphonomy. Quat. Int. 233, 130–141 (2011). doi: 10.1016/
pmid: 17322060 early human occupation of Argentina. Quat. Int. 301, 104–122 j.quaint.2010.04.013
80. C. R. Ferring, The Archaeology and Paleoecology of the (2013). doi: 10.1016/j.quaint.2013.03.011 101. M. Izuho et al., New AMS dates from the Shukubai-Kaso site
Aubrey Clovis Site (41DN479), Denton County, Texas 91. R. Suárez, G. Piñeiro, F. Barceló, Living on the river edge: The (loc. Sankakuyama), Hokkaido (Japan): Refining the
(U.S. Army Corps of Engineers, 2001). Tigre site (K-87) new data and implications for the initial chronology of small flake-based assemblages during the
81. G. Sanchez et al., Human (Clovis)-gomphothere (Cuvieronius colonization of the Uruguay River Basin. Quat. Int. 473, Early Upper Paleolithic in the Paleo-Sakhalin-Hokkaido-Kurile
sp.) association ~ 13,390 calibrated yBP in Sonora, Mexico. 242–260 (2018). doi: 10.1016/j.quaint.2017.08.024 Peninsula. PaleoAmerica 4, 134–150 (2018). doi: 10.1080/
Proc. Natl. Acad. Sci. U.S.A. 111, 10972–10977 (2014). 92. M. Waters, T. Amorosi, T. W. Stafford Jr., Redating Fell’s 20555563.2018.1457392
doi: 10.1073/pnas.1404546111; pmid: 25024193 Cave, Chile and the chronological placement of the 102. M. R. Waters, Early exploration and settlement of North
82. C. Beck, G. T. Jones, Clovis and Western Stemmed: Fishtail projectile point. Am. Antiq. 80, 376–386 (2015). America during the Late Pleistocene. SAA Archaeol. Rec. 19,
Population migration and the meeting of two technologies in doi: 10.7183/0002-7316.80.2.376 34–39 (2019).
the Intermountain West. Am. Antiq. 75, 81–116 (2010). 93. D. Jackson, C. Méndez, R. Seguel, A. Maldonado, G. Vargas,
doi: 10.7183/0002-7316.75.1.81 Initial occupation of the Pacific coast of Chile during Late
83. C. Beck, G. T. Jones, “Complexities of the colonization Pleistocene times. Curr. Anthropol. 48, 725–731 (2007). AC KNOWLED GME NTS
process: a view from the North American West” in doi: 10.1086/520965 T. Stafford, J. Raff, D. Carlson, and A. Linderholm provided useful
Paleoamerican Odyssey, K. E. Graf, C. V. Ketron, M. R. Waters, 94. J. W. Tune, M. R. Waters, K. A. Schmalle, L. R. G. DeSantis, comments. J. Lynch provided pictures of Beringian artifacts.
Eds. (Texas A&M Univ. Press, 2014), pp. 273–291. G. D. Kamenov, Assessing the proposed pre-Last Glacial Figures were prepared by J. Keene. Funding: This work, including
84. L. G. Davis, S. C. Willis, S. J. Macfarlan, “Lithic technology, Maximum human occupation of North America at Coats-Hines- figure preparation, was supported by funds from the Chair in First
cultural transmission, and the nature of the Far Western Litchy, Tennessee, and other sites. Quat. Sci. Rev. 186, American Studies and the North Star Archaeological Research
Paleoarchaic/Paleoindian co-tradition” in Meeting at the 47–59 (2018). doi: 10.1016/j.quascirev.2018.02.018 Program, Center for the Study of the First Americans, Texas A&M
Margins, D. Rhode, Ed. (Univ. of Utah Press, 2012), pp. 47–64. 95. G. G. Politis, L. Prates, “Clocking the arrival of Homo University. Competing interests: The author declares no
85. G. A. Pearson, Bridging the gap: An updated overview of sapiens in the Southern Cone of South America” in New competing interests.
Clovis across Middle America and its techno-cultural relation Perspectives on the Peopling of the Americas, K. Harvati,
with fluted point assemblages from South America. J. H. Reyes-Centeno, Eds. (Kerns, 2018), pp. 79–106. 10.1126/science.aat5447

Science Webinars help you keep pace with
emerging scientific fields!
Stay informed about scientific breakthroughs and discoveries.
Gain insights into current research from top scientists.
Take the opportunity to ask questions during live broadcasts.
Get alerts about upcoming free webinars.
Sign up at: webinar.sciencemag.org/stayinformed
SJAD001.indd 1 3/1/19 11:06 AM

R ES E A RC H
◥ provided a readout of their biological features

RESEARCH ARTICLE SUMMARY as they transitioned from SAM to a state of
persistent moderate acute malnutrition (MAM)
with accompanying persistent microbiota
MICROBIOTA
immaturity. Significant correlations were iden-
tified between levels of plasma proteins, an-
Effects of microbiota-directed thropometry, plasma metabolites, and the
representation of bacteria in their microbiota.
foods in gnotobiotic animals and Gnotobiotic mice were subsequently colonized

with a defined consortium of bacterial strains
that represent various phases of microbiota
undernourished children development in healthy Bangladeshi children.
Administration of different combinations of
Jeanette L. Gehrig*, Siddarth Venkatesh*, Hao-Wei Chang*, Matthew C. Hibberd, Bangladeshi complementary food ingredients
Vanderlene L. Kung, Jiye Cheng, Robert Y. Chen, Sathish Subramanian, to colonized mice and germ-free controls re-
Carrie A. Cowardin, Martin F. Meier, David O’Donnell, Michael Talcott, Larry D. Spears, vealed diet-dependent increases in the abun-
Clay F. Semenkovich, Bernard Henrissat, Richard J. Giannone, Robert L. Hettich, dance and changes in the metabolic activities
Olga Ilkayeva, Michael Muehlbauer, Christopher B. Newgard, Christopher Sawyer, ◥
of targeted weaning-phase
ON OUR WEBSITE
Richard D. Head, Dmitry A. Rodionov, Aleksandr A. Arzamasov, Semen A. Leyn, strains as well as diet- and
Andrei L. Osterman, Md Iqbal Hossain, Munirul Islam, Nuzhat Choudhury, Read the full article colonization-dependent
Shafiqul Alam Sarker, Sayeeda Huq, Imteaz Mahmud, Ishita Mostafa, Mustafa Mahfuz, at http://dx.doi. augmentation of growth-
org/10.1126/ promoting host signaling
Michael J. Barratt, Tahmeed Ahmed, Jeffrey I. Gordon†
science.aau4732 pathways. Host and micro-
..................................................
bial effects of microbiota-
INTRODUCTION: There is a dimension to post- repair restores healthy growth requires iden- directed complementary food (MDCF) prototypes
natal human development that involves assem- tification of microbial targets that are not only were subsequently examined in gnotobiotic
bly of microbial communities in different body biomarkers of community assembly but also mice colonized with immature microbiota from
habitats, including the gut. Children with acute mediators of various aspects of growth. Iden- children with post-SAM MAM and in gnoto-
malnutrition have impaired development of their tifying ingredients in complementary foods, biotic piglets colonized with a defined consor-
gut microbiota, leaving them with communities consumed during the transition from exclusive tium of targeted age- and growth-discriminatory
that appear younger (more immature) than those milk feeding to a fully weaned state, that in- taxa. A randomized, double-blind study of stan-
of chronologically age-matched healthy individu- crease the representation and expressed bene- dard therapy versus various MDCF prototypes
als. Current therapeutic foods given to children ficial functions of growth-promoting bacterial emerging from these preclinical models, con-
with acute malnutrition have not been formu- taxa in the developing microbiota could pro- ducted in Bangladeshi children with MAM,
lated based on knowledge of how they affect the vide an effective, affordable, culturally accept- identified a lead MDCF that increased levels of
developmental biology of the gut microbiota. able, and sustainable approach to treatment. biomarkers and mediators of growth, bone for-
Moreover, they are largely ineffective in amelio- mation, neurodevelopment, and immune func-
rating the long-term sequelae of malnutrition RESULTS: Metabolomic and proteomic analy- tion toward a state resembling healthy children.
that include persistent stunting, neurodevelop- ses of serially collected plasma samples were Using an approach inspired by statistical meth-
mental abnormalities, and immune dysfunction. combined with metagenomic analyses of se- ods applied to financial markets, we show in the
rially collected fecal samples from Bangladeshi accompanying paper by Raman et al. that this
RATIONALE: Repairing microbiota immatu- children with severe acute malnutrition (SAM) lead MDCF was most effective in repairing the
rity and determining the degree to which such treated with standard therapy. The results microbiota.
CONCLUSION: These findings demonstrate

the translatability of results obtained from pre-
clinical gnotobiotic animal models to humans,
directly support the hypothesis that healthy
microbiota development is causally linked to
healthy growth, illustrate an approach for treat-
ing childhood undernutrition, and with the
capacity to deliberately reconfigure immature
microbiota, suggest a means to decipher how
elements of the gut microbial community op-
erate to regulate various host systems involved
in healthy growth.
▪
The list of author affiliations is available in the full article online.
*These authors contributed equally to this work.
†Corresponding author. Email: jgordon@wustl.edu
This is an open-access article distributed under the terms
of the Creative Commons Attribution license (http://
creativecommons.org/licenses/by/4.0), which permits
Overview of therapeutic food discovery and testing. The approach used for integrating unrestricted use, distribution, and reproduction in any
medium, provided the original work is properly cited.
preclinical gnotobiotic animal models with human studies to understand the contributions of Cite this article as J. L. Gehrig et al., Science 365, eaau4732
perturbed gut microbiota development to childhood malnutrition and to identify MDCFs. (2019). DOI: 10.1126/science.aau4732

R ES E A RC H
◥ ties from healthy and stunted or underweight

RESEARCH ARTICLE 6- and 18-month-old Malawian children were
transplanted into groups of recently weaned
germ-free mice fed a diet representative of that
MICROBIOTA consumed by the human donor population. The
results disclosed that compared with mice col-
Effects of microbiota-directed onized with normally maturing microbiota from

the healthy donors, animals harboring imma-
ture microbiota exhibited reduced rates of lean
foods in gnotobiotic animals and body mass gain, alterations in bone growth, and
metabolic abnormalities (4). These studies pro-
undernourished children vided preclinical evidence for a causal link be-

tween microbiota immaturity and undernutrition;
they also revealed that a subset of the age-
Jeanette L. Gehrig1,2*, Siddarth Venkatesh1,2*, Hao-Wei Chang1,2*, discriminatory strains is growth-discriminatory.
Matthew C. Hibberd1,2, Vanderlene L. Kung1,2,3, Jiye Cheng1,2, Robert Y. Chen1,2, Moreover, a cultured consortium of these age-
Sathish Subramanian1,2†, Carrie A. Cowardin1,2, Martin F. Meier1,2, David O’Donnell1,2, and growth-discriminatory taxa ameliorated
Michael Talcott4, Larry D. Spears5, Clay F. Semenkovich5, Bernard Henrissat6,7, the impaired growth phenotype transmitted to
Richard J. Giannone8, Robert L. Hettich8, Olga Ilkayeva9,10, Michael Muehlbauer9,10, recipient gnotobiotic mice by an immature mi-
Christopher B. Newgard9,10,11,12, Christopher Sawyer13,14, Richard D. Head13,14, crobiota (4).
Dmitry A. Rodionov15,16, Aleksandr A. Arzamasov15,16, Semen A. Leyn15,16, One question arising from these observations
Andrei L. Osterman16, Md Iqbal Hossain17, Munirul Islam17, Nuzhat Choudhury17, is, how do we design optimal foods that steer a
Shafiqul Alam Sarker17, Sayeeda Huq17, Imteaz Mahmud17, Ishita Mostafa17, microbiota into an age-appropriate and healthy
state? Breastfeeding plays a major role in reduc-
Mustafa Mahfuz17, Michael J. Barratt1,2, Tahmeed Ahmed17, Jeffrey I. Gordon1,2‡
ing childhood malnutrition. As such, WHO and
To examine the contributions of impaired gut microbial community development to
the United Nations Children’s Fund (UNICEF)
childhood undernutrition, we combined metabolomic and proteomic analyses of plasma
recommend exclusive breastfeeding for the first
samples with metagenomic analyses of fecal samples to characterize the biological state
6 months of postnatal life and continued breast-
of Bangladeshi children with severe acute malnutrition (SAM) as they transitioned, after
feeding after the introduction of complementary
standard treatment, to moderate acute malnutrition (MAM) with persistent microbiota
foods up to 24 months of age (5). Suboptimal
immaturity. Host and microbial effects of microbiota-directed complementary food
complementary feeding practices are important
(MDCF) prototypes targeting weaning-phase bacterial taxa underrepresented in SAM and
contributors to malnutrition in children less than
MAM microbiota were characterized in gnotobiotic mice and gnotobiotic piglets colonized
2 years of age (6). However, current complemen-
with age- and growth-discriminatory bacteria. A randomized, double-blind controlled
tary feeding recommendations are not based on
feeding study identified a lead MDCF that changes the abundances of targeted bacteria
knowledge of how foods affect the develop-
and increases plasma biomarkers and mediators of growth, bone formation,
mental biology of the gut microbiota during the
neurodevelopment, and immune function in children with MAM.
weaning process. Together, these observations
raise the question: Do certain complementary
E
food ingredients or combinations of ingredients
vidence is accumulating that disruption Bangladeshi children with severe acute mal- have the ability to selectively increase the rep-
of “normal” gut community (microbiota) nutrition [SAM; defined as a weight-for-height resentation and expressed beneficial functions of
development may contribute to the path- z-score (WHZ) >3 standard deviations below the age- and growth-discriminatory strains deficient
ogenesis of undernutrition. Using culture- median for a World Health Organization (WHO) in SAM- or MAM-associated microbiota? If the
independent surveys, bacterial membership reference healthy growth cohort (3)]. The results answer is yes, then prescribed feeding of these
has been defined in fecal samples collected month- revealed gut communities that resembled those ingredients could help “repair” or prevent de-
ly during the first 2 postnatal years from healthy of healthy children who were chronologically velopment of microbiota immaturity in chil-
members of a birth cohort living in an urban younger. This microbiota “immaturity” was more dren, with potentially long-lived, health-promoting
slum (Mirpur) in Dhaka, Bangladesh (1, 2). Ap- pronounced in children with SAM as compared effects.
plying machine learning [Random Forests (RF)] with those with moderate acute malnutrition Here, we describe a process for identifying
to the resulting 16S ribosomal DNA (rDNA) data- (MAM; WHZ score between –2 and –3) and was microbiota-directed complementary foods (MDCFs)
set yielded a “sparse” model composed of the most not repaired in a clinical study that tested the designed to treat children with acute malnutri-
age-discriminatory bacterial strains; changes in effects of two therapeutic foods (2). tion. We first characterized gut microbial com-
the relative abundances of these organisms de- Impaired microbiota development has also munity and host responses over the course of
scribed a program of normal microbiota develop- been documented in undernourished Malawian 12 months in Bangladeshi children who were
ment (2). This RF-derived model was subsequently children (4). To examine the functional impor- treated for SAM with one of three conventional
used to characterize fecal samples collected from tance of this impairment, microbial communi- therapeutic foods. Measurement of the levels of
1
Edison Family Center for Genome Sciences and Systems Biology, Washington University School of Medicine, St. Louis, MO 63110, USA. 2Center for Gut Microbiome and Nutrition Research,
Washington University School of Medicine, St. Louis, MO 63110, USA. 3Department of Pathology and Immunology, Washington University School of Medicine, St. Louis, MO 63110, USA. 4Division
of Comparative Medicine, Washington University, St. Louis, MO 63110, USA. 5Department of Medicine, Washington University School of Medicine, St. Louis, MO 63110, USA. 6Architecture et
Fonction des Macromolécules Biologiques, Centre National de la Recherche Scientifique and Aix-Marseille Université, 13288 Marseille cedex 9, France. 7Department of Biological Sciences, King
Abdulaziz University, Jeddah, Saudi Arabia. 8Chemical Sciences Division, Oak Ridge National Laboratory, Oak Ridge, TN 37830, USA. 9Sarah W. Stedman Nutrition and Metabolism Center, Duke
University Medical Center, Durham, NC 27710, USA. 10Duke Molecular Physiology Institute, Duke University Medical Center, Durham, NC 27710, USA. 11Department of Pharmacology and Cancer
Biology, Duke University Medical Center, Durham, NC 27710, USA. 12Department of Medicine, Duke University Medical Center, Durham, NC 27710, USA. 13Department of Genetics, Washington
University School of Medicine, St. Louis, MO 63110, USA. 14Genome Technology Access Center, Washington University School of Medicine, St. Louis, MO 63110, USA. 15A. A. Kharkevich Institute
for Information Transmission Problems, Russian Academy of Sciences, Moscow 127994, Russia. 16Infectious and Inflammatory Disease Center, Sanford Burnham Prebys Medical Discovery
Institute, La Jolla, CA 92037, USA. 17International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b), Dhaka 1212, Bangladesh.
*These authors contributed equally to this work. †Present address: Department of Medicine, Massachusetts General Hospital, Boston, MA 02114, USA.
àCorresponding author. Email: jgordon@wustl.edu
Gehrig et al., Science 365, eaau4732 (2019) 12 July 2019 1 of 12

R ES E A RC H | R E S EA R C H A R T I C LE
1305 human plasma proteins—including regu- charge, and 6 months after discharge for tar- plasma leptin (Fig. 2A), and weight gain (table
lators and effectors of physiologic, metabolic, and geted mass spectrometry (MS)–based metabolic S1B). However, 6 months after treatment, multi-
immune functions—combined with mass spec- profiling; a sufficient quantity of blood was ob- ple plasma amino acids and their metabolites
trometric profiling of plasma metabolites and tained from eight children at all three time points had declined to levels comparable with those at
culture-independent analyses of serially col- for aptamer-based proteomics analysis (8–10). admission, whereas fatty acids and fatty acid–
lected fecal samples provided a “readout” of the Of these children, 44% had MAM at 12 months derived metabolites remained at similar concen-
biological features of these children as they proof followup. None of the therapeutic foods pro- trations to those observed at discharge (Fig. 2, A
gressed from SAM to a state of incomplete recov- duced a significant effect on their severe stunt- to C). Insulin-like growth factor 1 (IGF-1) levels
ery (post-SAM MAM) with persistent microbiota ing [height-for-age z-score (HAZ)] (Fig. 1B and did not change significantly during this period
immaturity. This readout included correlations table S1B). (Fig. 2A), potentially explaining the absence of a
between plasma proteins, anthropometry, plas- signature of pronounced lipolysis that had been
ma metabolites, and the representation of age- Metabolic phenotypes observed at enrollment. Although the suppression
discriminatory members of their microbiota. We Targeted MS of plasma samples obtained at en- of lipolysis at 6 months after discharge suggests a
then screened complementary foods in gnotobi- rollment revealed high levels of ketones, non- sustained effect of nutritional resuscitation, the
otic mice colonized with a consortium of bacte- esterified fatty acids (NEFA), and medium to long fall in essential amino acids and the lower level
rial strains that had been cultured from children even-chain acylcarnitines (Fig. 2, A and B, and of IGF-1 compared with that found in similarly
living in Mirpur to identify ingredients that table S2), which is consistent with the known aged healthy children from the same community
promote the representation of constituent age- acute malnutrition-induced lipolytic response (44.5 versus 69.4 ng/mL; P = 0.02, t test) may
discriminatory strains that are underrepresented that raises circulating fatty acids and activates contribute to the observed failure to achieve catch-
in the setting of acute malnutrition. Subsequently, fatty acid oxidation (11). By discharge, this meta- up growth.
a representative microbiota from a child with bolic feature had normalized, whereas levels of a
post-SAM MAM was transplanted into gnoto- number of amino acids had increased signifi- The plasma proteome
biotic mice. Recipient animals were fed a diet cantly, including the gluconeogenic amino acid Significant correlations between levels of plasma
resembling that consumed by children in Mirpur alanine; the branched-chain amino acids leucine, proteins, anthropometric indices, plasma metab-
but supplemented with ingredients identified in isoleucine, and valine; plus products of branched- olites, and host signaling pathways regulating
the screen, in order to establish whether one or chain amino acid metabolism [C3 (propionyl)- key facets of growth are described in the supple-
more of these MDCF formulations could repair carnitine and their ketoacids] (Fig. 2, A to C). mentary text, results (table S3, A and B)—for
a microbiota from a subject who had already These findings suggest that the increased protein example, components of the growth hormone
received conventional therapy. Lead formu- provided by the therapeutic foods prompted a (GH)–IGF axis, including soluble growth hor-
lations were subsequently tested in gnotobiotic switch from fatty acid to amino acid oxidation, mone receptor (also known as growth hormone
piglets colonized with a defined consortium of leading to repletion of fat depots, increases in binding protein), multiple IGF binding proteins
age- and growth-discriminatory strains to test
their biological effects in a host species that is
physiologically and metabolically more similar
to humans than mice. Last, three MDCF proto-
types were administered to children with MAM,
and their effects on the microbiota and host bio-
logical state were determined.
Effects of conventional therapeutic

foods on the biological state
of children with SAM
A total of 343 Bangladeshi children aged 6 to
36 months with SAM were enrolled in a multi-
center, randomized, double-blind “noninferiority”
study designed to compare two locally produced
therapeutic foods (supplementary materials, mate-
rials and methods) with a commercially available,
ready-to-use therapeutic food (RUTF) (7) (study
design is provided in Fig. 1A, and the composi-
tions of these therapeutic foods are available in
table S1A). Children received standard manage-
ment for SAM during the acute stabilization phase
of in-hospital treatment, including a short course
of antibiotics. Eligible children were then ran-
domized to one of the three therapeutic food
arms (~200 kcal/kg/day, mean duration 16.1 ±
10.3 days) (table S1B). Children were discharged
after meeting criteria described in the supple-
mentary materials, materials and methods. In a Fig. 1. Longitudinal study of Bangladeshi children with SAM treated with therapeutic foods.
subset of 54 children, fecal samples were collected (A) Study design. (B) Anthropometry and MAZ scores. Gray bars represent the three time points
at enrollment [age 15.2 ± 5.1 months (mean ± SD)] at which blood samples were collected. (C) Summary of MAZ scores for children with SAM
before randomization, twice during treatment (WHZ < –3; n = 96 fecal samples) and subsequently (post-SAM) MAM (WHZ > –3 and <–2; n =
with a therapeutic food, and at regular intervals 151 fecal samples), plus healthy children aged 6 to 24 months living in the same area in which the
up to 12 months after discharge (Fig. 1A; clinical SAM study was conducted (n = 450 fecal samples). Mean values for WHZ, WAZ, HAZ, and
metadata is available in table S1B). Blood samples MAZ ± SEM are plotted on the x axes of (B) and (C). ****P < 0.0001 (one-way ANOVA followed
(plasma) were also obtained at enrollment, dis- by Tukey’s multiple comparisons test).

Fig. 2. Metabolic features of children with SAM before and after treatment. (A to C) Levels of (A) standard clinical metabolites and selected
hormones, (B) acylcarnitines, and (C) amino acids and ketoacids in plasma collected from children at enrollment (Fig. 1A, B1 blood sample),
discharge (Fig. 1A, B2 sample), and 6 months after discharge (Fig. 1A, B3 sample). Abbreviations for branched chain ketoacids in (C) are KIC,
a-ketoisocaproate; KIV, a-ketoisovalerate; and KMV, a-keto-b-methylvalerate. Mean values ± SEM are plotted. *P < 0.05; **P < 0.01; ***P < 0.001;
****P < 0.0001 (paired t test followed by FDR correction).
(IGFBPs), and regulators of IGFBP turnover (the and WAZ [Pearson correlation coefficient (r) = cohort with consistently healthy anthropome-
metalloprotease pappalysin-1 and its inhibitor 0.16, P = 0.0004; r = 0.13, P = 0.003; and r = 0.10, try and 15 of the 54 children enrolled in the
stanniocalcin-1). P = 0.02, respectively]. The MAZ score was not SAM study (table S5B); these 15 children were
different at discharge but improved 1 month later selected according to their age (12 to 18 months)
The gut microbiome (P = 0.0051 versus admission, Mann-Whitney and that we had corresponding plasma metab-
A sparse RF-derived model of normal gut micro- test). This improvement could reflect increased olomic and proteomic datasets for at least two of
biota development comprising 30 bacterial taxa dietary diversity, reduced antibiotic usage (table the three time points sampled. The abundances
[operational taxonomic units (OTUs)] and ob- S1B), and/or other factors associated with return- of microbial genes that mapped to pathways in
tained from 25 healthy members of a birth cohort ing to the home environment. MAZ-scores did the microbial communities SEED (mcSEED) data-
living in Mirpur (table S4, A to C) was applied not change significantly thereafter (Fig. 1B). base (12)—related to metabolism of amino acids,
to bacterial V4-16S rDNA datasets generated A number of the age-discriminatory strains carbohydrates, fermentation products, and B
from fecal samples serially collected from the were significantly correlated with anthropo- vitamins and related cofactors—were first de-
children in the SAM study (n = 539 samples). This metric indices as well as with plasma proteins fined in healthy children sampled monthly from
model allowed us to define microbiota-for-age involved in biological processes that mediate birth to 2 years of age. A set of age-discriminatory
z-scores (MAZ) as a function of treatment arm growth. We also identified significant negative metabolic pathways (mcSEED “subsystems” or
and time [9.3 ± 3.7 samples/child (mean ± SD)]. correlations between these taxa and mediators pathway modules) was identified. The resulting
The MAZ score measures the deviation in de- of systemic inflammation and anorexia/cachexia. sparse RF-derived model (fig. S1, A and B, and
velopment of a child’s microbiota from that of Bifidobacterium longum (OTU 559527) had the materials and methods) allowed us to assign a
chronologically age-matched reference healthy greatest number of significant correlations (114) state of development (functional age or “matu-
children on the basis of the representation of [table S3C; further discussion is available in sup- rity”) to the fecal microbiomes of the 15 children
the ensemble of age-discriminatory strains con- plementary text, results]. treated for SAM. Relative functional maturity was
tained in the RF-derived model (2). Significant The effects of the therapeutic food interven- significantly correlated with MAZ, WHZ, and WAZ
microbiota immaturity was apparent in the SAM tions on the representation of metabolic path- scores during the course of the trial (Pearson r
and post-SAM MAM groups (Fig. 1C and table ways in the gut microbiome were defined by and P values are MAZ, r = 0.55, P < 0.0001; WHZ,
S5A). Moreover, MAZ-scores in this SAM cohort shotgun sequencing of 331 fecal DNA samples r = 0.30, P = 0.0011; WAZ, r = 0.23, P = 0.013).
were significantly correlated with WHZ, HAZ, obtained from 30 members of the Mirpur birth At enrollment, and just before administration of

therapeutic foods, children with SAM had more immature SAM-associated microbiota, we col- cantly higher relative abundances of a number
immature microbiomes [one-way analysis of var- onized 5-week-old, germ-free C57Bl/6J mice of weaning-phase age-discriminatory taxa, in-
iance (ANOVA) P = 0.0002; Dunnett’s multiple with the consortium of cultured, sequenced bac- cluding Faecalibacterium prausnitzii, Dorea
comparisons test for healthy versus SAM adjusted terial strains. After colonization, an 8-week pe- longicatena, and B. luti (P < 0.01; Mann-Whitney
P values at the two time points, 0.027 and 0.0001, riod of diet “oscillations” was initiated (fig. S2B). test) (Fig. 3A and table S9B). This prototype did
respectively]. There was a statistically significant We incorporated 12 complementary food ingre- not promote the fitness of the SAM donor-
improvement in functional maturity from initia- dients commonly consumed in Mirpur (6) into derived strains, with the exception of Escherichia
tion of therapeutic food treatment to discharge, 14 different diets using a random sampling strat- fergusonii.
and at 1 and 6 months after discharge (Tukey’s egy (table S8, A to E, and materials and meth- We used targeted MS to quantify cecal levels
multiple comparisons test; adjusted P values = ods). The composition of these complementary of carbohydrates, short-chain fatty acids, plus
0.039, 0.0028, and 0.025, respectively). However, food combinations (CFCs) and their order of ad- amino acids and their catabolites (table S10, A
this improvement was not sustained at later time ministration to mice were based on considerations to D). Germ-free animals served as reference
points (fig. S1D). Comparing the relative abun- described in the legend to fig. S2, B and C. controls to define levels of cecal nutrients that,
dances of the 30 most age-discriminatory path- Spearman’s rank correlation coefficients were by inference, would be available for bacterial
ways at six time points revealed that the SAM calculated between the relative abundances of utilization in the different diet contexts. There
microbiome had significantly reduced represen- the 14 bacterial strains that colonized mice and were several noteworthy findings: (i) Levels of
tation of (i) amino acid metabolic pathways, levels of complementary food ingredients in the butyrate and succinate were significantly higher in
including those involved in isoleucine, leucine, 14 CFCs tested (fig. S2D and table S9A). Chickpea colonized animals consuming MDCF compared
valine biosynthesis, and uptake; (ii) several car- and banana had strong positive correlations with MS/KH (Fig. 3B and table S10B). (ii) There
bohydrate utilization pathways (arabinose and with the greatest number of strains representing were no statistically significant diet-associated dif-
arabinosides, rhamnose and rhamnogalacturonan, weaning-phase age-discriminatory OTUs. Tilapia ferences in levels of any of the amino acids mea-
and sialic acid); and (iii) multiple pathways had a narrower range of significant positive ef- sured in germ-free animals, but when compared
involved in B-vitamin metabolism, including fects (fig. S2D). Chickpea, banana, and tilapia with their colonized MS/KH–fed counterparts,
“niacin/NADP (nicotinamide adenine dinucleo- also had significant negative correlations with colonized MDCF-consuming animals had sig-
tide phosphate) biosynthesis” (fig. S1E and table levels of the preweaning, milk-adapted B. longum nificantly elevated cecal levels of six amino acids
S5C). The observed underrepresentation of age- subsp. infantis isolate. A sobering observation was classified as essential in humans (the three
discriminatory OTUs and metabolic pathways that a number of complementary food ingre- branched-chain amino acids leucine, isoleucine,
in the gut communities of children with post- dients typically represented in diets consumed and valine plus phenylalanine and tryptophan)
SAM MAM provided the rationale for developing by 18-month-old children living in Mirpur had (Fig. 3C and table S10C). And (iii) two tryptophan-
a pipeline to test complementary food ingredients significant negative correlations with six of the derived microbial metabolites that play important
for their ability to repair this immaturity. weaning-phase age-discriminatory strains, in- roles in suppressing inflammation and are neuro-
cluding rice, milk powder, potato, spinach, and protective, 3-hydroxyanthranillic acid (3-HAA)
Screening complementary sweet pumpkin (fig. S2D). Rice gruel with milk is and indole-3-lactic acid (16–21), were signifi-
food ingredients the most common first complementary food cantly elevated in colonized animals fed MDCF
Nine age-discriminatory bacterial strains were given to Bangladeshi children (14). Moreover, compared with their MS/KH–treated counter-
cultured from the fecal microbiota of three egg, which is included in a number of regimens parts (table S10D).
healthy children, aged 6 to 23 months, who lived for nutritional rehabilitation of children with acute Findings from RNA-sequencing (RNA-seq)
in Mirpur, and genomes of these isolates were malnutrition (15), was negatively correlated with analysis of the transcriptional responses of com-
sequenced (table S6, A and C). Seven of the the abundance of two weaning-phase strains, munity members to the two diets based on Kyoto
nine isolates had V4-16S rDNA sequences that Dorea formicigenerans and Blautia luti. Encyclopedia of Genes and Genomes (KEGG)–
corresponded to age-discriminatory OTUs whose and mcSEED-derived annotations of the 40,735
representation is associated with the period of com- Testing an initial MDCF prototype predicted protein-coding genes present in con-
plementary food consumption (“weaning-phase” Khichuri-halwa (KH) is a therapeutic food com- sortium members are described in tables S9C
OTUs) (fig. S2A), whereas two, Bifidobacterium monly administered together with milk-suji (MS) and S11, A to C, and supplementary text, results,
longum subsp. infantis and Bifidobacterium to Mirpur children with SAM. A previous study and in silico predictions of their ability to produce,
breve, are most prominent during the period of documented the inability of this intervention use, and/or share nutrients are provided in table
exclusive, predominant milk feeding (fig. S2A) to repair gut microbiota immaturity (2). We pre- S6, D and E. For example, community-level anal-
(13). OTUs representing seven of the nine cul- pared a diet that mimicked MS and KH (MS/KH) ysis revealed specific members manifested MDCF-
tured strains were significantly depleted in the (table S8, D and E); 7 of its 16 ingredients are associated increases in expression of genes involved
fecal microbiota of Bangladeshi children with commonly consumed complementary foods that in biosynthesis of the essential amino acids, in-
SAM before treatment (table S7 and fig. S3). Seven had little, if any, effect on the representation of cluding branched-chain amino acids (Ruminococcus
additional age-discriminatory strains were cul- weaning-phase age-discriminatory strains (rice, obeum and Ruminococcus torques), and generation
tured from the immature fecal microbiota of a red lentils, potato, pumpkin, spinach, whole-wheat of aromatic amino acid metabolites (R. obeum,
24-month-old child with SAM enrolled in the same flour, and powdered milk) (fig. S2D). The effects R. torques, and F. prausnitzii) (table S11C, ii).
study as the subcohort shown in Fig. 1 (table S6, of MS/KH on members of the 14-member con-
A and C). Together, the consortium of 16 strains sortium and the host were compared with those Host effects
represented OTUs that directly matched 65.6 ± produced by an initial MDCF prototype that con- Serum levels of IGF-1 were significantly higher in
22.8% (mean ± SD) of V4-16S rDNA sequencing tained chickpeas, banana, and tilapia (table S9B). colonized mice that consumed the initial MDCF
reads generated from 1039 fecal samples col- Five-week-old germ-free C57Bl/6J mice colonized prototype compared with those that consumed
lected from 53 healthy members of the Mirpur with the consortium were monotonously fed either MS/KH. This effect was diet- and colonization-
birth cohort during their first 2 postnatal years, of the two diets ad libitum for 25 days. dependent, with germ-free animals exhibiting
and 74.2 ± 25.2% of the reads produced from significantly lower levels of IGF-1 in both diet
fecal samples obtained from 38 children with Microbial community responses contexts (Fig. 3D). Serum insulin levels were also
SAM (table S7). Community profiling by means of short read higher in colonized animals that consumed MDCF
To identify complementary foods that selec- shotgun sequencing (COPRO-seq) of cecal DNA compared with MS/KH [800.7 ± 302.9 ng/mL
tively increase the representation of weaning- revealed that compared with MS/KH, consump- (mean ± SD) versus 518.7 ± 135.1 ng/mL, respec-
phase age-discriminatory strains deficient in tion of the MDCF prototype resulted in signifi- tively; P = 0.06; unpaired t test].

Fig. 3. Comparison of microbial community and host effects of an initial proteins involved in IGF-1 signaling and IGF-1 production. Levels of phospho-
MDCF prototype versus MS/KH. Separate groups of germ-free mice or ani- rylated proteins were normalized to the total amount of the corresponding
mals colonized with the defined consortium of 14 bacterial strains were fed the nonphosphorylated protein or to glyceraldehyde-3-phosphate dehydrogenase
two diets, monotonously, for 25 days, after which time they were euthanized, and (GAPDH). (F) Effect of diet and colonization status on the cortical thickness of
cecal contents were analyzed. (A) The relative abundances of strains in the cecal femoral bone. (G) Effects of diet in colonized gnotobiotic mice on branched-
microbiota of colonized mice. Mean values ± SD shown. (B and C) Diet- and chain amino acids in serum and acylcarnitines in muscle and liver. [C5-OH/C3
colonization-dependent effects on (B) cecal levels of short chain fatty acids and are isobars that are not resolved through flow injection MS/MS. C5-OH is a
(C) essential amino acids plus the tryptophan metabolite, indole 3-lactic acid. mix of 3-hydroxy-2-methylbutyryl carnitine (derived from the classical
Each dot represents a sample from a mouse in the indicated treatment group. isoleucine catabolic intermediate 3-hydroxy-2-methylbutyryl CoA) and
Mean values ± SD are shown. ***P < 0.001; ****P < 0.0001 [2-way ANOVA fol- 3-hydroxyisovaleryl carnitine (a noncanonical leucine metabolite)]. For (D)
lowed by Tukey’s multiple comparisons test for (A) to (C)]. (D) Diet- and colonization- to (G), mean values ± SD are shown. ns, not significant. *P < 0.05;
dependent effects on serum IGF-1 levels. (E) Effects of diet on levels of liver **P < 0.01; ****P < 0.0001 for (D) to (G) (Mann-Whitney test).

IGF-1 binding to its receptor tyrosine kinase, commonly consumed plant-based sources of Eighteen mcSEED pathway modules involved
IGF-1R, affects a variety of signal transduction protein. To identify alternatives to tilapia, we in amino acid metabolism were significantly
pathways, including one involving the serine/ selected an additional 16 plant-derived comple- up-regulated in the cecal microbiomes of mice
threonine kinase Akt/PKB, phosphatidylinositol-3 mentary food ingredients with varied levels and consuming Mirpur(PCSB) or Mirpur(P) com-
kinase (PI3K), and the mammalian target of quality of protein (25) that are culturally ac- pared with those consuming Mirpur-18, with
rapamycin (mTOR). Absorption of several amino ceptable, affordable, and readily available in the most up-regulated being “isoleucine, leucine,
acids from the gut—notably, branched-chain Bangladesh (fig. S4A and table S13, A and B). valine biosynthesis” [other age-discriminatory
amino acids and tryptophan—leads to activation Their effects were tested in gnotobiotic mice mcSEED pathway modules that showed signifi-
of mTOR (22). Colonized animals fed MDCF had colonized with a defined, expanded consortium cantly lower abundances in the fecal microbiomes
significantly higher levels of hepatic phosphoSer473- of 18 age- and growth-discriminatory bacterial of Bangladeshi children with SAM and whose
Akt, which is consistent with activation of Akt strains (table S6A). We generated 48 mouse diets expression was increased by Mirpur(PCSB) or
by IGF-1 signaling through the PI-3K pathway by supplementing a prototypic base diet repre- Mirpur(P) in gnotobiotic mice are provided in
(Fig. 3E). Levels of phospho–AMPK (5′ adeno- sentative of that consumed by 18-month-old chil- Fig. 4B and fig. S1E]. Serum levels of a product of
sine monophosphate-activated protein kinase) dren living in Mirpur (Mirpur-18), with each of branched-chain amino acid metabolism, C5:1-
were not significantly affected by diet (Fig. 3E), the individual ingredients incorporated at three acylcarnitine, were significantly higher in mice
suggesting that Akt phosphorylation is not caused different concentrations (fig. S4A and table S13A). consuming Mirpur(PCSB) compared with un-
indirectly by altered hepatic energy status. Phos- The results revealed that in this defined commu- supplemented Mirpur-18 (0.148 ± 0.015 versus
phorylation of hepatic Jak 2 (Tyr1007/1008) and nity context, peanut flour had the greatest effect 0.086 ± 0.0098 mM, respectively; P = 0.014, un-
mTOR (Ser2448), which are involved in IGF-1 on the largest number of targeted weaning-phase paired t test). Findings from mass spectrometric
production, was significantly increased in colon- age-discriminatory taxa, followed by chickpea analysis of cecal contents, isolation, and compara-
ized mice consuming MDCF (Fig. 3E), whereas flour (fig. S4B and table S13C). Soy flour, which tive genomic analysis of an F. prausnitzii strain
phosphorylation of STAT5, also implicated in promoted the representation of two of these prominently represented in the transplanted
IGF-1 production, was not significantly altered. taxa, had the second-highest percentage pro- community, and characterization of the in vivo
Previous studies of adult germ-free mice re- tein after peanut flour (fig. S4A), and its protein transcriptional responses of this strain to the
ported increases in serum IGF-1 after their col- quality was among the highest of the ingre- different diets, are described in table S15F and
onization with gut microbiota from conventionally dients tested (table S13B). On the basis of these supplementary text results.
raised mice; increased IGF-1 levels were also as- observations, we chose soy and peanut flours as
sociated with increased bone formation (23, 24). replacements for tilapia in subsequent MDCF Gut mucosal barrier function
Micro-computed tomography (mCT) of mouse formulations. Epithelium and overlying mucus from the prox-
femurs revealed a significant increase in femoral We reasoned that by transplanting a repre- imal, middle, and distal thirds of the small intestine
cortical bone area in MDCF-fed animals; the sentative immature intact microbiota into young, were recovered with laser capture microdissec-
effect was both diet- and microbiota-dependent germ-free mice, we could investigate whether gut tion (LCM) (Fig. 4D). Listed in table S15C are the
(Fig. 3F). health (defined by relative abundances of com- 30 most abundant OTUs identified by means of
We used targeted MS to quantify levels of munity members, expression of microbial genes V4-16S rDNA analysis of LCM mucosal DNA ob-
amino acids, acyl–coenzyme As (acylCoAs), acyl- in mcSEED metabolic pathways, and biomark- tained from the different small intestinal seg-
carnitines, and organic acids in serum, liver, ers and mediators of gut barrier function) was ments within a given diet group and between
and gastrocnemius muscle (table S12). Products improved by supplementing the Mirpur-18 diet similarly positioned segments across the differ-
of nonoxidative metabolism of glucose and pyr- with one or more complementary food ingredients ent diet treatments. For example, Mirpur(PCSB)
uvate (lactate from glycolysis and alanine from that target weaning-phase age-discriminatory taxa. produced a statistically significant increase in
transamination of pyruvate, respectively) were Fifteen fecal samples from 12 different children, the relative abundance of F. prausnitzii in the
significantly lower in mice fed MDCF compared obtained during or after treatment for SAM, were proximal two-thirds of the small intestine, with-
with mice fed MS/KH; this was true for alanine screened in gnotobiotic mice to identify samples out significantly affecting the proportional rep-
in serum, skeletal muscle, and liver and for lac- containing the greatest number of transmissi- resentation of a milk-associated age-discriminatory
tate in liver (table S12, A to C and H). Oxidative ble weaning-phase age-discriminatory taxa and Bifidobacteria OTU (Fig. 4, C and D).
metabolism of glucose is associated with nutri- to assess their response to supplementation of Gene expression was characterized in the
tionally replete, anabolic conditions. These find- Mirpur-18 (table S14A). We selected a sample jejunal mucosa (Fig. 4D, SI-2 segment) recovered
ings are consistent with the observed elevations obtained from a donor (PS.064) who had post- by LCM from mice belonging to all six treat-
of the anabolic hormone IGF-1 in MDCF-fed com- SAM MAM; in addition to the successful transmis- ment groups. Significant differences in expres-
pared with MS/KH–fed mice. MDCF-fed mice sion of targeted taxa, 88.7 ± 1.3% (mean ± SD) of sion were categorized based on enriched Gene
had significantly higher circulating levels of the recipient animals’ gut communities consisted Ontology (GO) terms for “Molecular Function.”
branched-chain amino acids than those of their of OTUs that were detected at >0.1% relative In colonized mice, Mirpur(P) and Mirpur(PCSB)
MS/KH–fed counterparts (Fig. 3G and table S12, abundance in the donor sample (table S14B). significantly up-regulated genes assigned to
A to C). Skeletal muscle C5 carnitine and the Three groups of mice were colonized with this “cadherin binding” (GO: 0045296) and “cell ad-
closely related metabolite C5-OH/C3 carnitine microbiota and monotonously fed one of three hesion molecule binding” (GO: 0050839) com-
were significantly higher in animals consuming diets: unsupplemented Mirpur-18, Mirpur-18 sup- pared with Mirpur-18 (table S17A). The diet effect
MDCF (Fig. 3G and table S12F). In liver, C3 and C5 plemented with peanut flour [Mirpur(P)], or was colonization-dependent; there were no signif-
acylcarnitines were significantly lower in MDCF- Mirpur-18 supplemented with four of the lead icant differences in expression of these genes or
treated mice (Fig. 3G and table S12E), suggesting ingredients [Mirpur(PCSB), with peanut flour, these GO categories in germ-free mice consum-
that the more nutritionally replete state asso- chickpea flour, soy flour, and banana] (Fig. 4A ing supplemented versus unsupplemented diets
ciated with MDCF may act to limit branched- and table S15A). Three control groups were main- (table S17). (Further discussion is available in the
chain amino acid oxidation in this tissue. tained as germ-free; each group was fed one of the supplementary text, results, and histochemical
three diets. and immunohistochemical analyses of tissue
Testing additional MDCF prototypes We characterized the effects of diet supple- sections prepared along the length of the small
in gnotobiotic mice mentation on cecal and serum levels of metab- intestines of these mice are provided in fig. S6).
Incorporating tilapia into MDCF prototypes poses olites as well as on expression of genes in various On the basis of its effects on microbial commu-
several problems: Its organoleptic properties microbial metabolic pathways (tables S15, B, D, nity organismal composition, gene expression,
are not desirable, and its cost is higher than and E, and S16 and supplementary text, results). and gut barrier function, we deemed Mipur-18

Fig. 4. Effects of Mirpur-18 diet supplementation on a post-SAM mcSEED subsystem/pathway module; all P values are FDR-adjusted).
MAM donor microbiota transplanted into gnotobiotic mice. (A) (C) Effects of supplementing Mirpur-18 with one or all four complementary
Experimental design. dpg, days post gavage of the donor microbiota; food ingredients on the relative abundances of a weaning-phase– and a
Mirpur(P), Mirpur-18 supplemented with peanut flour; Mirpur(PCSB), milk-phase–associated taxon in feces obtained at dpg 21 (one-way ANOVA
Mirpur-18 supplemented with peanut flour, chickpea flour, soy flour, and followed by Tukey’s multiple comparisons test). (D) Relative abundances
banana. (B) Expression of microbial mcSEED metabolic pathway/modules of the two taxa in mucosa harvested by means of LCM from the proximal,
in the ceca of gnotobiotic mouse recipients of the post-SAM MAM donor middle, and distal thirds of the small intestine. (Right) Schematic of
gut community as a function of diet treatment. *P < 0.05; **P < 0.001; locations in the small intestine where LCM was performed. The same
***P < 0.0001 (statistical comparisons indicate results of gene set color code for diets is used in (A) to (D). *P < 0.05; **P < 0.01; ****P <
enrichment analysis expression on a per-gene basis across the indicated 0.0001 (Mann-Whitney test).
supplemented with the four lead complementary content (with equivalent representation of amino sumed ad libitum for the remainder of the ex-
foods [Mirpur(PCSB)] superior to that supple- acids), and also met current ready-to-use thera- periment (n = 4 piglets/treatment arm) (Fig.
mented with just peanut flour [Mirpur(P)]. peutic food guidelines for children with respect to 5A). Animals were euthanized on day 31 after a
macro- and micronutrient content (table S18A) (28). 6-hour fast.
Characterizing MDCF prototypes Four-day-old germ-free piglets fed a sow milk– Piglets fed MDCF(PCSB) exhibited signifi-
in gnotobiotic piglets based formula were colonized with a 14-member cantly greater weight gain than those receiving
We examined the effects of MDCF prototypes consortium of bacterial strains that consisted MDCF(CS) (Fig. 5B). Micro-computed tomogra-
in a second host species whose physiology and of the same nine Bangladeshi age-discriminatory phy of their femurs revealed that they also had
metabolism are more similar to that of humans. strains used for the diet oscillation experiments significantly greater cortical bone volume (Fig.
Gnotobiotic piglets provide an attractive model described in fig. S2, plus five weaning-phase age- 5C). COPRO-seq analysis disclosed that pig-
for these purposes; piglets manifest rapid growth discriminatory strains cultured from Malawian lets fed MDCF(PCSB) had significantly higher
rates in the weeks after birth (26), and methods children (table S6B). In an earlier study, sev- relative abundances of C. symbiosum, R. gnavus,
for conducting experiments with gnotobiotic eral members of this consortium (B. longum, D. formicigenerans, R. torques, and Bacteroides
piglets have been described (27). F. prausnitzii, Clostridium, Ruminococcus gnavus, fragilis in their cecum and distal colon compared
On the basis of the results from the gnotobiotic and D. formicigenerans) had been classified as with those of piglets consuming MDCF(CS) (Fig.
mouse studies, we designed two MDCF proto- growth-discriminatory by means of a RF-based 5D and table S18B); all are weaning-phase age-
types. One prototype was formulated to be anal- analysis of their representation in gnotobiotic discriminatory strains, and the former three
ogous to Mirpur-18, which contains milk powder; mouse recipients of healthy and undernourished were, as noted above, also defined as growth-
this prototype was supplemented with peanut donor microbiota and the animals’ weight and discriminatory. Conversely, the relative abundances
flour, chickpea flour, soy flour, and banana lean body mass gain phenotypes (4). After ga- of three members of Bifidobacteria (including
[MDCF(PCSB)]. The other diet lacked milk powder vage, the two groups of piglets were weaned over two milk-associated age-discriminatory strains,
and was supplemented with just chickpea flour the course of 10 days (supplementary materials, B. breve and B. longum subsp. infantis) were
and soy flour [MDCF(CS)]. The two MDCFs were materials and methods) onto one or the other significantly higher in the ceca and distal co-
isocaloric, matched in lipid levels and total protein irradiated MDCF prototypes, which they con- lons of piglets fed MDCF(CS) (table S18B). These

findings led us to conclude that MDCF(PCSB) formulations (MDCF-1, -2, and -3) were designed the four treatment groups (14 to 17 children per
promoted a more weaning-phase–like (mature) to be similar in protein energy ratio and fat en- group) and received 4 weeks of twice-daily
community configuration than MDCF(CS). (ge- ergy ratio and provide 250 kcal/day (divided over feeding under supervision at the study center,
nome annotations, microbial RNA-seq, and tar- two servings). MDCF-2 contained all four lead preceded and followed by 2 weeks of observa-
geted MS analyses of cecal metabolites are ingredients (chickpea flour, soy flour, peanut tion and sample collection. Mothers were en-
provided in tables S18, C and D, and S19A and flour, and banana) at higher concentrations than couraged to continue their normal breastfeeding
supplementary text, results). in MDCF-1. MDCF-3 contained two lead ingre- pactices throughout the study (Fig. 6A). There
The effects of the two MDCF prototypes on dients (chickpea flour and soy flour). A rice- and were no significant differences in the mean daily
host biology were defined by means of MS- lentil-based ready-to-use supplementary food amount of each MDCF or RUSF consumed per
based serum metabolomic and proteomic analy- (RUSF), included as a control arm, lacked all child or in the mean incidence of morbidity
ses (tables S19 and S20). Notable findings four ingredients but was otherwise similar in across the four treatment groups (table S21, B to
included significant increases in levels of tryp- energy density, protein energy ratio, fat energy D). All three MDCFs and the RUSF control im-
tophan, methionine, and C3-acylcarnitine with ratio, and macro- and micronutrient content to proved WHZ scores [–1.9 ± 0.5 (mean ± SD) at
MDCF(PCSB) as well as changes produced in the those of the MDCFs (Fig. 6A). Milk powder was the completion of intervention compared with
serum proteome that are shared with children included in MDCF-1 and RUSF. All formulations –2.2 ± 0.4 at the start of intervention; n = 63
in the SAM trial (Fig. 5, E and F, and supple- were supplemented with a micronutrient mixture children, P = 2.06 × 10−11 for all groups com-
mentary text, results). designed to provide 70% of the recommended bined, paired t test]. There were no statistically
daily allowances for 12- to 18-month-old children. significant differences between the four inter-
Testing MDCFs in Bangladeshi The formulations were produced locally and ventions in the change in WHZ (P = 0.31, one-
children with MAM tested for organoleptic acceptability before ini- way ANOVA). Despite the small group size and
To assess the degree to which results obtained tiating the trial (table S21A). the short length of the study, there were signif-
from the gnotobiotic mouse and piglet models Children from Mirpur with MAM and no prior icant differences in treatment effects on another
translate to humans, we performed a pilot ran- history of SAM were enrolled (mean age at en- anthropometric indicator, with MDCF-2 produc-
domized, double-blind controlled feeding study rollment, 15.2 ± 2.1 months, mean WHZ –2.3 ± ing a significantly greater increase in mid-upper
of the effects of three MDCF formulations. The 0.3). Participants were randomized into one of arm circumference (MUAC) than MDCF-3 (one-
way ANOVA, P = 0.022; with Tukey’s multiple
comparisons test, P = 0.017) (table S21E).
Effects on biological state

To contextualize the biological effects of the di-
etary interventions, we performed quantitative
proteomics (SOMAscan) on plasma collected
from 21 12- to 24-month-old Mirpur children
with healthy growth phenotypes (mean age, 19.2 ±
5.1 months; WHZ, 0.08 ± 0.58; HAZ, –0.41 ± 0.56,
WAZ, –0.12 ± 0.60) and 30 children with SAM
before treatment (Fig. 1A, B1 sample; WHZ < –3;
mean age 15.2 ± 5.1 months) [metadata associ-
ated with the healthy, SAM, and MAM (MDCF
trial) cohorts are provided in table S22]. We rank-
ordered all detected proteins according to fold
differences in their abundances in plasma collected
from healthy children compared with children
with untreated SAM. The top 50 most differentially
abundant proteins (P < 10−7; R package “limma”)
that were significantly higher in healthy children
were designated “healthy growth-discriminatory,”
whereas the top 50 differentially abundant pro-
teins that were significantly higher in children
with SAM were designated “SAM-discriminatory”
(table S23A). We next compared the mean dif-
ference for each protein in the pre- versus post-
intervention plasma samples for all children in
each MDCF/RUSF treatment group. Proteins were
then ranked according to the fold differences
of the pre- versus post-treatment levels in each
of the four groups (table S23B), and these treat-
ment effects were mapped onto the 50 most
Fig. 5. Effects of two different MDCF prototypes in gnotobiotic piglets. (A) Experimental healthy growth-discriminatory and 50 most SAM-
design. (B) Weight gain in piglets weaned onto isocaloric MDCF prototypes containing either peanut discriminatory proteins. Strikingly, MDCF-2 elicited
flour, chickpea flour, soy flour, and banana [MDCF(PCSB)] or chickpea and soy flours [MDCF(CS)]. a biological response characterized by a marked
(C) mCT of femoral bone obtained at euthanasia. (D) Effects of the MDCFs on the relative abun- shift in the plasma proteome toward that of healthy
dances of community members in cecal and distal colonic contents. (E) Examples of serum proteins children and away from that of children with SAM;
with significantly different post-treatment levels between the two diet groups. (F) Effect of diet MDCF-2 increased the abundance of proteins that
on serum C3 acylcarnitine levels. Mean values ± SD are plotted. *P < 0.05; **P < 0.01; ***P < 0.005, are higher in plasma from healthy children and
****P < 0.001 [two-way ANOVA in (B), unpaired t test in (C), (D), and (F)]. The color code reduced the levels of proteins elevated in SAM
provided in (B) also applies to (C), (D), and (F). plasma samples (Fig. 6B).

product elevated in SAM plasma that reduces

appetite and negatively correlated with HAZ,
was also decreased by MDCF-2.
We identified GO terms that were enriched
among the group of treatment-responsive pro-
teins and ranked them according to the P value
of their enrichment (table S23C). Proteins be-
longing to GO terms significantly higher in healthy
compared with SAM plasma samples were deemed
“healthy growth-discriminatory,” whereas those
that were significantly higher in SAM were deemed
“SAM-discriminatory” (fold-difference >30%; FDR-
adjusted P value <0.05). This analysis revealed
multiple healthy growth-discriminatory proteins
associated with GO processes “osteoblast differ-
entiation” and “ossification” that were increased
by supplementation with MDCF-2 (Fig. 7A and
table S23C). Examples include key markers or
mediators of osteoblast differentiation [osteo-
pontin (SPP1), bone sialoprotein 2 (IBSP), and
bone morphogenetic protein 7 (BMP7)] as well as
matrix metalloproteases (MMP-2 and MMP-13)
involved in terminal differentiation of osteoblasts
into osteocytes and bone mineralization.
A number of plasma proteins categorized un-
der the GO process “CNS development,” includ-
ing those involved in axon guidance and neuronal
differentiation, were also affected by MDCF-2 sup-
plementation. Levels of the SAM-discriminatory
semaphorin SEMA3A, a potent inhibitor of axonal
growth, decreased with this treatment, where-
as healthy growth-discriminatory semaphorins
(SEMA5A, SEMA6A, and SEMA6B) increased
(Fig. 7B). Other healthy growth-discriminatory
proteins whose abundances increased with
MDCF-2 included receptors for neurotrophin
(NTRK2 and NTRK3) plus various axonal guidance
proteins [netrin (UNC5D), ephrin A5 (EFNA5),
roundabout homolog 2 (ROBO2), and SLIT and
NTRK-like protein 5 (SLITRK5)] (Fig. 7B). Ex-
pression of a number of neurotrophic proteins
belonging to these families has been reported
to be influenced by nutrient availability in pri-
mates (29).
Fig. 6. Comparing the effects of MDCF formulations on the health status of Bangladeshi Compared with healthy children, the plasma
children with MAM. (A) Study design and composition of diets. (B) Quantitative proteomic proteome of children with SAM was charac-
analysis of the average fold-change, per treatment group, in the abundances of the 50 plasma terized by elevated levels of acute phase proteins
proteins most discriminatory for healthy growth and the 50 plasma proteins most discriminatory for [such as C-reactive protein (CRP) or interleukin-
SAM (protein abundance is column-normalized across treatment groups). (C) Average fold-change 6 (IL-6)] and inflammatory mediators, including
in abundances of plasma proteins that significantly positively or negatively correlate with HAZ several agonists and components of the nuclear
[absolute value of Pearson correlation > 0.25, FDR-corrected P value < 0.05; abundance is column- factor–kB (NF-kB) signaling pathway (Fig. 7C).
normalized as in (B)]. These components include the pro-inflammatory
cytokines IL-1b, tumor necrosis factor–a (TNF-a),
and CD40L, plus ubiquitin-conjugating enzyme
Aggregating proteomic datasets from the com- IGF-1 binding protein IGFBP-3, growth hormone E2 N (UBE2N), which is involved in induction
bined cohort of 113 children with SAM, MAM, receptor (GHR), and leptin (LEP) (Fig. 6C). Growth of NF-kB– and mitogen-activated protein kinase
and healthy growth phenotypes for whom plas- differentiation factor 15 (GDF15) was reduced (MAPK)–responsive inflammatory genes (30).
ma samples were available, we identified a total after 4 weeks of dietary supplementation with MDCF-2 supplementation was associated with
of 27 plasma proteins that were significantly MDCF-2 (Fig. 6C). This transforming growth reductions in the levels of all of these SAM-
positively correlated with HAZ and 57 plasma factor–b superfamily member, which was nega- associated proteins (Fig. 7C).
proteins that were significantly negatively cor- tively correlated with HAZ, has been implicated
related with HAZ [absolute value of Pearson in the anorexia and muscle wasting associated Effects on the microbiota
correlation > 0.25, false discovery rate (FDR)– with cancer and with chronic heart failure in Our analysis of fecal microbiota samples re-
corrected P value < 0.05]. Among the treatments, children; it was elevated in children with SAM vealed no significant change in the representa-
MDCF-2 was distinctive in its ability to increase and positively correlated with their lipolytic tion of enteropathogens within and across the
the abundances of a broad range of proteins pos- biomarkers NEFA and ketones (supplementary four treatment groups (fig. S7A and table S21F).
itively correlated with HAZ, including the major text, results). Peptide YY, an enteroendocrine cell MDCF-2–induced changes in biological state

were accompanied by increases in the relative tions. Interpretation of this finding was con- community to a mature state similar to that
abundances of several weaning-phase taxa, in- founded by unexpectedly high baseline microbiota characteristic of healthy Bangladeshi children.
cluding OTUs assigned to F. prausnitzii (OTU maturity scores in this group of children with
851865) and a Clostridiales sp. (OTU 338992) MAM [MAZ, –0.01 ± 1.12 (mean ± SD)] (table Conclusions
that are closely related to taxa ranked first and S22) compared with a small, previously char- We have integrated preclinical gnotobiotic ani-
second in feature importance in the sparse acterized Mirpur cohort with untreated MAM mal models with human studies to understand
Bangladeshi RF-derived model of gut microbiota (2). Hence, we developed an additional measure the contributions of perturbed gut microbiota
maturation (fig. S7, B and C). MDCF-2 supple- of microbiota repair (31). This involved a statis- development to childhood undernutrition and
mentation was associated with a significant tical analysis of covariance among bacterial taxa to identify new microbiota-directed therapeu-
decrease in B. longum (OTU 559527) (fig. S7B), in the fecal microbiota of anthropometrically tic approaches. We identified a set of proteins
which is ranked third in feature importance in healthy members of a Mirpur birth cohort who that distinguish the plasma proteomes of healthy
the RF-derived model and discriminatory for a had been sampled monthly over a 5-year period. children from those with SAM. Using these data,
young, milk-oriented microbiota. None of the Using approaches developed in the fields of we have developed a supplemental food proto-
other members of the 30 OTU model showed econophysics and protein evolution to charac- type, MDCF-2, that shifted the plasma proteome
significant changes. By contrast, MDCF-1 did terize the underlying organization of interacting of children with MAM toward that of healthy
not produce significant increases in any of the systems with seemingly intractable complexity, individuals, including proteins involved in linear
taxa in the model. The other two formulations such as financial markets, we found that the gut growth, bone development, neurodevelopment,
were each associated with a significant change community in healthy children could be decom- and immune function. MDCF-2 is a tool for in-
in just one member [an increase in the relative posed into a sparse unit of 15 covarying bacterial vestigating, in larger studies across different
abundance of an early age-discriminatory OTU taxa termed an “ecogroup” (31). These ecogroup populations with varying degrees of undernu-
(Streptococcus; ranked 30th) with MDCF-3 sup- taxa include a number of age-discriminatory trition, how repair of gut microbiota immaturity
plementation, and a decrease in another OTU strains in the Bangladeshi RF-derived model affects various facets of child growth.
(Enterococcus faecalis; ranked 29th) with RUSF (such as B. longum, F. prausnitzii, and Prevotella
supplementation] (table S4B). copri). We used the ecogroup to show that in Overview of methods
MAZ scores were not significantly different addition to its effects on host biological state, Human studies
between groups at enrollment, nor were they MDCF-2 was also the most effective of the four Children aged 6 to 59 months with SAM (n =
significantly improved by any of the formula- treatments in reconfiguring the gut bacterial 343 participants) were enrolled in a study en-
titled “Development and field testing of ready-
to-use therapeutic foods (RUTF) made of local
ingredients in Bangladesh for the treatment of
children with severe acute malnutrition.” The
study was approved by the Ethical Review Com-
mittee at icddr,b (ClinicalTrials.gov identifier
NCT01889329). Written informed consent was
obtained from their parent or guardian. A sub-
set of 54 children were included in a substudy
that involved regular fecal sampling and three
blood draws for up to 1 year after discharge.
Children aged 12 to 18 months with MAM who
were no longer exclusively breastfed (n = 63
participants) were enrolled in a double-blind,
randomized, four-group, parallel assignment
interventional trial study (ClinicalTrials.gov iden-
tifier NCT03084731) approved by the Ethical
Review Committee at icddr,b. Fecal and plasma
samples were collected as described in the sup-
plementary materials, materials and methods,
and stored at –80°C. Samples were shipped to
Washington University with associated clinical
metadata and maintained in a dedicated bio-
specimen repository with approval from the
Washington University Human Research Pro-
tection Office.
Analysis of plasma samples

Methods for targeted MS-based metabolomics
are described in the supplementary materials.
Fig. 7. The effects of different MDCF formulations on biomarkers and mediators of bone and The SOMAscan 1.3K Proteomic Assay plasma/
CNS development, plus NF-kB signaling. (A to C) Average fold-change (normalized across treat- serum kit (SomaLogic, Boulder, Colorado,) was
ment groups) in the abundances of plasma proteins belonging to GO categories related to (A) bone, used to measure 1305 proteins. The R package
(B) CNS development, and (C) agonists and components of the NF-ĸB signaling pathway. Proteins in “limma” (Bioconductor) was used to analyze
the GO category that were significantly higher in the plasma of healthy compared with SAM children differential protein abundances (32). Spearman
(fold-difference >30%; FDR-adjusted P value < 0.05) are labeled “healthy growth-discriminatory,” whereas correlation analyses were performed between
those higher in SAM compared with healthy children (fold-difference >30%; FDR-adjusted P value measured proteins and anthropometric scores,
< 0.05) are labeled “SAM-discriminatory.” Levels of multiple “healthy growth-discriminatory” proteins plasma metabolites, and the abundances of bac-
associated with (A) GO processes “osteoblast differentiation” and “ossification”, and (B) the GO process terial taxa in fecal samples. Plasma proteins in
“CNS development” are enhanced by MDCF-2 treatment while (C) NF-kB signaling is suppressed. children with healthy growth phenotypes or SAM

(before treatment) were rank-ordered according the effects MDCF prototypes—including (i) design Metab. 99, 2128–2137 (2014). doi: 10.1210/jc.2013-4018;
to the fold-difference in their levels between and preparation of diets; (ii) culturing of age- pmid: 24606092
12. R. Overbeek et al., The SEED and the Rapid Annotation of
these two groups. The top 50 most differentially discriminatory and SAM-associated bacterial microbial genomes using Subsystems Technology (RAST).
abundant proteins in healthy compared with SAM strains; (iii) shotgun sequencing of DNA isolated Nucleic Acids Res. 42, D206–D214 (2014). doi: 10.1093/nar/
were designated as healthy growth-discriminatory from serially collected fecal samples; (iv) micro- gkt1226; pmid: 24293654
proteins, and the top 50 most differentially bial RNA-seq of cecal contents; (v) targeted MS of 13. D. A. Sela et al., The genome sequence of Bifidobacterium
longum subsp. infantis reveals adaptations for milk utilization
abundant in SAM compared with healthy were cecal contents, liver, gastrocnemius muscle, and within the infant microbiome. Proc. Natl. Acad. Sci. U.S.A.
designated as SAM-discriminatory proteins. serum samples for measurement of amino acids, 105, 18964–18969 (2008). doi: 10.1073/pnas.0809584105;
The average fold-change for these healthy growth- acylcarnitines, organic acids, and acylCoAs; (vi) pmid: 19033196
and SAM-discriminatory proteins was then cal- Western blot analysis of IGF-1 pathway com- 14. T. Ahmed et al., Nutrition of children and women in Bangladesh:
Trends and directions for the future. J. Health Popul. Nutr. 30,
culated for each treatment arm in the MDCF trial ponents in liver; (vii) mCT of bone; and (viii) 1–11 (2012). doi: 10.3329/jhpn.v30i1.11268; pmid: 22524113
(before versus after MDCF/RUSF treatment) and characterizing the effects of a transplanted fecal 15. L. L. Iannotti et al., Eggs in early complementary feeding
normalized to the mean fold-change across all microbiome from a donor with post-SAM MAM and child growth: A randomized controlled trial. Pediatrics
four arms. Limma was used to calculate statisti- in recipient gnotobiotic mice as a function of diet 140, e20163459 (2017). doi: 10.1542/peds.2016-3459;
pmid: 28588101
cal significance. All 1305 proteins were mapped treatment by histochemical and immunohisto- 16. W. R. Russell et al., Major phenylpropanoid-derived metabolites
to all GO “Biological Processes” in the GO data- chemical analysis of their intestinal segments, in the human gut can arise from microbial fermentation of
base (www.geneontology.org). SetRank, a gene LCM of their small intestinal epithelium, and protein. Mol. Nutr. Food Res. 57, 523–535 (2013). doi: 10.1002/
set enrichment analysis (GSEA) algorithm (33), RNA-seq analysis of gene expression in LCM mnfr.201200594; pmid: 23349065
17. D. Krause et al., The tryptophan metabolite 3-hydroxyanthranilic
was used to identify GO Biological Processes mucosa—are all described in the supplementary acid plays anti-inflammatory and neuroprotective roles
that were significantly enriched for proteins materials. during inflammation: Role of hemeoxygenase-1. Am. J. Pathol.
that exhibited changes in abundance from be- 179, 1360–1372 (2011). doi: 10.1016/j.ajpath.2011.05.048;
fore to after treatment with MDCF/RUSF. Enrich- Gnotobiotic piglets pmid: 21855684
18. L. Cervantes-Barragan et al., Lactobacillus reuteri induces gut
ment was calculated by using the setRankAnalysis Experiments were performed under the supervi- intraepithelial CD4+CD8aa+ T cells. Science 357, 806–810
function in the SetRank R library (parameters sion of a veterinarian by using protocols approved (2017). doi: 10.1126/science.aah5825; pmid: 28775213
were use.ranks = TRUE; setPCutoff = 0.01; and by the Washington University Animal Studies 19. G. Das et al., An important regulatory role for CD4+CD8 a a
fdrCutoff = 0.05). The average fold-change for Committee and that followed American Veter- T cells in the intestinal epithelial layer in the prevention of
inflammatory bowel disease. Proc. Natl. Acad. Sci. U.S.A.
each protein in the statistically significant Bi- inary Medical Association guidelines for eutha- 100, 5324–5329 (2003). doi: 10.1073/pnas.0831037100;
ological Process category was calculated for each nasia. The protocol for generating germ-free pmid: 12695566
treatment arm and normalized to the mean fold- piglets; preparing diets, feeding, colonization, 20. H. Cheroutre, M. M. Husain, CD4 CTL: Living up to the
change across all four arms. We defined proteins and husbandry of piglets; measurement of weight challenge. Semin. Immunol. 25, 273–281 (2013). doi: 10.1016/
j.smim.2013.10.022; pmid: 24246226
within the GO Biological Process as “healthy gain; mCT of femurs; and liquid chromatography– 21. T. Sujino et al., Tissue adaptation of regulatory and intraepithelial
growth-discriminatory” if they were at least 30% MS (LC-MS)/MS–based serum proteomics are all CD4+ T cells controls gut inflammation. Science 352, 1581–1586
more abundant in healthy individuals compared described in the supplementary materials. (2016). doi: 10.1126/science.aaf3892; pmid: 27256884
with those with SAM and “SAM-discriminatory” 22. R. A. Saxton, D. M. Sabatini, mTOR signaling in growth,
metabolism, and disease. Cell 169, 361–371 (2017).
if they were at least 30% more abundant in chil- doi: 10.1016/j.cell.2017.03.035; pmid: 28388417
RE FERENCES AND NOTES
dren with SAM compared with those who were 23. J. Yan et al., Gut microbiota induce IGF-1 and promote bone
1. T. Ahmed et al., The MAL-ED cohort study in Mirpur,
classified as healthy. formation and growth. Proc. Natl. Acad. Sci. U.S.A. 113,
Bangladesh. Clin. Infect. Dis. 59, S280–S286 (2014).
E7554–E7563 (2016). doi: 10.1073/pnas.1607235113;
doi: 10.1093/cid/ciu458; pmid: 25305298
Characterizing human fecal microbial 2. S. Subramanian et al., Persistent gut microbiota immaturity in
pmid: 27821775
communities malnourished Bangladeshi children. Nature 510, 417–421
24. M. Schwarzer et al., Lactobacillus plantarum strain maintains
growth of infant mice during chronic undernutrition.
(2014). doi: 10.1038/nature13421; pmid: 24896187
Methods for V4-16S rRNA gene sequencing and Science 351, 854–857 (2016). doi: 10.1126/science.aad8588;
3. World Health Organization Department of Nutrition for Health
data analysis, calculation of MAZ scores and pmid: 26912894
and Development, WHO child growth standards. Length/
25. D. J. Millward, Amino acid scoring patterns for protein quality
functional microbiome maturity, and quantifi- height-for-age, weight-for-age, weight-for-length, weight-for-
assessment. Br. J. Nutr. 108 (Suppl 2), S31–S43 (2012).
cation of enteropathogen burden by means of height and body mass index-for-age: methods and
doi: 10.1017/S0007114512002462; pmid: 23107544
development (2000); www.who.int/childgrowth/en
multiplex quantitative polymerase chain reac- 4. L. V. Blanton et al., Gut bacteria that prevent growth
26. P. L. Altman, D. S. Dittmer, Growth, Including Reproduction
tion (qPCR) are described in the supplemen- and Morphological Development (Federation of American
impairments transmitted by microbiota from malnourished
Societies for Experimental Biology, 1962).
tary materials. children. Science 351, aad3311 (2016). doi: 10.1126/science.
aad3311; pmid: 26912898 27. M. R. Charbonneau et al., Sialylated milk oligosaccharides
promote microbiota-dependent growth in models of infant
Animal studies 5. World Health Organization (WHO), Infant and Young Child
undernutrition. Cell 164, 859–871 (2016). doi: 10.1016/
Gnotobiotic mice feeding; fact sheet no. 342, 1–5 (WHO, 2016).
j.cell.2016.01.024; pmid: 26898329
6. L. Manikam et al., A systematic review of complementary
All mouse experiments were performed by using feeding practices in South Asian infants and young children: 28. U.S. Department of Agriculture (USDA), “Commercial Item
The Bangladesh perspective. BMC Nutr. 3, 56 (2017). Description: Ready-to-Use Therapeutic Food (RUTF)”
protocols approved by the Washington Uni- A-A-20363B (USDA, 2012).
doi: 10.1186/s40795-017-0176-9
versity Animal Studies Committee. Mice were 7. H. Sandige, M. J. Ndekha, A. Briend, P. Ashorn, M. J. Manary, 29. I. Antonow-Schlorke et al., Vulnerability of the fetal primate
housed in plastic flexible film gnotobiotic iso- Home-based treatment of malnourished Malawian children brain to moderate reduction in maternal global nutrient
lators under a 12-hour light cycle. Male germ-free with locally produced or imported ready-to-use food. J. Pediatr. availability. Proc. Natl. Acad. Sci. U.S.A. 108, 3011–3016 (2011).
Gastroenterol. Nutr. 39, 141–146 (2004). doi: 10.1097/ doi: 10.1073/pnas.1009838108; pmid: 21252306
C57BL/6 mice were initially weaned onto an auto-
00005176-200408000-00003; pmid: 15269617 30. K. Taniguchi, M. Karin, NF-kB, inflammation, immunity and
claved, low-fat, high-plant polysaccharide chow 8. L. Gold, J. J. Walker, S. K. Wilcox, S. Williams, Advances in cancer: Coming of age. Nat. Rev. Immunol. 18, 309–324
that was administered ad libitum (diet 2018S, human proteomics at high scale with the SOMAscan (2018). doi: 10.1038/nri.2017.142; pmid: 29379212
Envigo). Animals were maintained on this diet proteomics platform. Nat. Biotechnol. 29, 543–549 (2012). 31. A. S. Raman et al., A sparse covarying unit that describes
until 3 days before the beginning of experiments pmid: 22155539 healthy and impaired human gut microbiota development.
9. B. Lollo, F. Steele, L. Gold, Beyond antibodies: New affinity Science 365, eaau4735 (2019).
involving tests of the effects of complementary reagents to unlock the proteome. Proteomics 14, 638–644 32. M. E. Ritchie et al., limma powers differential expression
food ingredients. Defined consortia of sequenced (2014). doi: 10.1002/pmic.201300187; pmid: 24395722 analyses for RNA-sequencing and microarray studies.
bacterial strains cultured from Bangladeshi chil- 10. J. Candia et al., Assessment of variability in the SOMAscan Nucleic Acids Res. 43, e47 (2015). doi: 10.1093/nar/gkv007;
dren, or intact uncultured microbiota from do- assay. Sci. Rep. 7, 14248 (2017). doi: 10.1038/s41598-017- pmid: 25605792
14755-5; pmid: 29079756 33. C. Simillion, R. Liechti, H. E. L. Lischer, V. Ioannidis,
nors with post-SAM MAM, were introduced by 11. S. Bartz et al., Severe acute malnutrition in childhood: R. Bruggmann, Avoiding the pitfalls of gene set enrichment
means of gavage into recipient mice at 5 weeks of Hormonal and metabolic status at presentation, response to analysis with SetRank. BMC Bioinformatics 18, 151 (2017).
age. Methods for identifying and characterizing treatment, and predictors of mortality. J. Clin. Endocrinol. doi: 10.1186/s12859-017-1571-6; pmid: 28259142

ACKN OW LEDG MEN TS distribution, and reproduction in any medium, provided the original R.D.H. and M.J.B. produced the quantitative proteomic datasets
We are grateful to the families of members of the human studies work is properly cited. To view a copy of this license, visit http:// from plasma samples with DNA aptamer-based arrays, and R.Y.C.,
described in this work for their participation and assistance. We creativecommons.org/licenses/by/4.0. This license does not apply M.J.B., J.L.G., and M.C.H. analyzed the data. C.S. and M.C.H.
are indebted to the staff and health care workers at icddr,b for to figures/photos/artwork or other content included in the article performed and analyzed qPCR assays for enteropathogens. J.L.G.,
their contributions to the recruitment and enrollment of mothers that is credited to a third party; obtain authorization from the S.V., H.-W.C., M.J.B., and J.I.G. wrote the paper with assistance
as well as the collection of biospecimens and data from their rights holder before using such material. Author contributions: provided by co-authors. Competing interests: J.I.G. is a cofounder
offspring. We thank M. Karlsson, M. Meier, S. Wagoner, S. Deng, T.A., I.H., M.I., N.C., S.H., I.Ma., M.Ma., S.A.S., and I.Mo. were of Matatu, a company characterizing the role of diet-by-microbiota
J. Serugo, and J. Hoisington-López for superb technical assistance; responsible for the design and conduct of the human studies plus interactions in animal health. L.D.S. is currently a scientific sales
K. Ahsan for assistance with maintaining the biospecimen collection of biospecimens and clinical metadata. M.J.B. established representative at STEMCELL Technologies. Data and materials
repository and associated database; J. Guruge for help with and maintained biospecimen repositories and associated databases availability: V4-16S rDNA sequences in raw format prior to
anaerobic microbiology; O. Delannoy-Bruno for assistance with the of de-identified metadata. J.L.G. and M.F.M. generated the 16S rDNA post-processing and data analysis, COPRO-seq, microbial RNA-seq,
gnotobiotic piglet experiment; Mars, Inc. for their assistance with and shotgun sequencing datasets from human fecal samples, and and proteomics datasets, plus shotgun sequencing datasets produced
manufacturing the MDCF(PCSB) and MDCF(CS) diets; A. Lutz J.L.G. and M.C.H. analyzed the data. S.V., J.L.G., M.J.B., and J.I.G. from human fecal DNA, cecal contents of gnotobiotic mice with a
and J. Yu (Genome Technology Access Center at Washington designed the gnotobiotic mouse studies. H.-W.C., M.J.B., and J.I.G. post-SAM MAM human donor community and cultured bacterial
University) for their contributions to generating SOMAscan designed the gnotobiotic piglet experiments. J.L.G., S.V., and strains, have been deposited at the European Nucleotide Archive
datasets; R. Olson and other members of RAST/SEED development S.S. cultured bacterial strains. J.L.G., S.V., H.-W.C., and M.C.H. (ENA) under study accession no. PRJEB26419. SOMAscan-generated
team at the Argonne National Laboratory for support with the sequenced and assembled the genomes of bacterial strains used in human plasma proteomic datasets have been deposited in the
mcSEED-based genome analysis and subsystem curation; and gnotobiotic animal experiments. D.A.R., A.A.A., S.A.L., and A.L.O. Gene Expression Omnibus (GEO) database under accession no.
D. Leib for developing the computer program to quantify CT scan performed in silico metabolic reconstructions with the genomes of GSE119641. All raw mass spectra for quantification of serum proteins
data obtained from the femurs of gnotobiotic piglets. Funding: the cultured strains. B.H. provided updated CAZyme annotations. in gnotobiotic piglets have been deposited in the MassIVE and
This work was supported by the Bill & Melinda Gates Foundation as S.V. and J.L.G. performed gnotobiotic mouse experiments with ProteomeXchange data repositories under accession nos.
part of the Breast Milk, gut Microbiome, and Immunity (BMMI) cultured bacterial strains and intact uncultured communities, MSV000082286 (MassIVE) and PXD009534 (ProteomeXchange),
Project. As members of Washington University’s Medical Scientist respectively. H.-W.C., D.O.D, and M.T. conducted the gnotobiotic with data files available at ftp://massive.ucsd.edu/MSV000082286.
Training Program, R.Y.C. and S.S. received support from NIH piglet experiments. S.V. and H.-W.C. generated COPRO-Seq Fecal and plasma specimens from the SAM and MDCF studies used
grant GM007200. mCT of femoral bone was performed using datasets. S.V., J.L.G., M.C.H., and H.-W.C. produced microbial for the analyses described in this study were provided to Washington
resources provided by the Washington University Musculoskeletal RNA-seq datasets. V.L.K. performed laser capture microdissection, University under materials transfer agreements with icddr,b.
Research Center (NIH P30 AR057235). Histochemical and V4-16S rDNA analysis of intestinal mucosa-associated bacterial
immunohistochemical processing of tissue sections was performed community composition, RNA-seq–based characterization of SUPPLEMENTARY MATERIALS
at the Washington University Digestive Diseases Research Core mouse small intestinal mucosal gene expression, and histochemical
science.sciencemag.org/content/365/6449/eaau4732/suppl/DC1
Center, funded by NIH P30 DK052574. Plasma proteomic datasets assays of intestinal morphometry. J.C., S.V., J.L.G., H.-W.C., M.Mu.,
Materials and Methods
were generated by the Genome Technology Access Center at O.I., and C.B.N. conducted metabolomic analyses of mouse,
Supplementary Text
Washington University School of Medicine, which is supported in piglet and human biospecimens. C.A.C. performed mCT of femurs;
Figs. S1 to S7
part by NIH Grants P30 CA91842 and UL1TR002345. D.A.R., measured serum IGF-1 levels in gnotobiotic mice; and quantified
References (34–124)
A.A.A., and S.A.L. were supported in part by the Russian Science leptin, IGF-1, and insulin in plasma obtained from children in the SAM
Tables S1 to S23
Foundation (grants 14-14-00289 and 19-14-00305). J.I.G. is the trial. H.-W.C. generated microcomputed tomographic datasets from
recipient of a Thought Leader Award from Agilent Technologies. piglet femurs. L.D.S. and C.F.S. characterized levels of IGF-1 pathway 13 June 2018; resubmitted 24 April 2019
This work is licensed under a Creative Commons Attribution 4.0 components in the livers of gnotobiotic mice. R.J.G. and R.L.H. Accepted 7 June 2019
International (CC BY 4.0) license, which permits unrestricted use, were responsible for MS-based proteomics of piglet serum samples. 10.1126/science.aau4732

R ES E A RC H
◥ RESULTS: A statistical workflow was devel-

RESEARCH ARTICLE SUMMARY oped to identify conserved bacterial taxon-
taxon covariance in the gut communities of
healthy members of a Bangladeshi birth co-
MICROBIOTA
hort who provided fecal samples monthly from
postnatal months 1 to 60. The results revealed
A sparse covarying unit that an “ecogroup” of 15 bacterial taxa that together
exhibited consistent covariation by 20 months
describes healthy and impaired of age and beyond. Ecogroup taxa also described
gut microbiota development in healthy mem-
bers of birth cohorts residing in Bangladesh,
human gut microbiota development India, and Peru to an extent comparable to
what is achieved when
◥
Arjun S. Raman, Jeanette L. Gehrig, Siddarth Venkatesh, Hao-Wei Chang, Matthew C. Hibberd, ON OUR WEBSITE considering all detected
Sathish Subramanian, Gagandeep Kang, Pascal O. Bessong, Aldo A.M. Lima, Read the full article
bacterial taxa; this finding
Margaret N. Kosek, William A. Petri Jr., Dmitry A. Rodionov, Aleksandr A. Arzamasov, at http://dx.doi. suggests that the ecogroup
Semen A. Leyn, Andrei L. Osterman, Sayeeda Huq, Ishita Mostafa, Munirul Islam, org/10.1126/ network is a conserved gen-
Mustafa Mahfuz, Rashidul Haque, Tahmeed Ahmed, Michael J. Barratt, Jeffrey I. Gordon* science.aau4735 eral feature of microbiota
..................................................
organization. Moreover,
INTRODUCTION: Ecosystems such as the hu- scribing the form of a microbiota—one that the ecogroup provided a framework for char-
man gut microbiota are typically described by takes into consideration the abundances as acterizing the state of perturbed microbiota de-
a “parts list” with enumeration of component well as the interactions between members. velopment in Bangladeshi children with severe
members. Accordingly, the abundances of com- acute malnutrition (SAM) and moderate acute
munity components are commonly used as a RATIONALE: Borrowing from the fields of malnutrition (MAM), as well as a quantitative
metric for relating its configuration to features econophysics and protein evolution, where metric for defining the efficacy of standard ver-
of its habitat and to the biological state of the identification of conserved covariation has sus microbiota-directed therapeutic foods in re-
host. Although this approach has provided provided insights about the organization of configuring their gut communities toward a
much insight, the structure and function of complex dynamic systems, we searched for fea- state seen in age-matched healthy children liv-
biological systems are emergent, arising from tures amidst the seemingly intractable complex- ing in the same locale. These results highlight
the collective action of constituent parts rather ity of human gut microbial communities that the importance of the ecogroup as a descriptor,
than each part acting in isolation. This char- could serve as a framework for understanding both for fundamental and practical uses. A con-
acteristic demands a different approach to de- how they assemble and function. sortium of cultured ecogroup taxa, introduced
into gnotobiotic piglets, reenacted changes in
Ecogroup as a concise Enterobacteriaceae their relative abundances that were observed in
description of microbiota E. hallii human communities as the animals transitioned
S. thermophilus
Ruminococcaceae from exclusive milk feeding to a fully weaned
form. (Top) Network diagram Dialister F. prausnitzii (514940)
of covarying taxa where node C.mitsuokai F. prausnitzii state consuming a prototypic Bangladeshi diet.
(851865) This pattern of change correlated with the rep-
(taxon) color indicates Blautia
Bifdobacterium
ecogroup (green) or non- L.garvieae
resentation of a sparse set of metabolic path-
B.bifdum
ecogroup (gray), node size B.longum
ways in the genomes of these organisms and, in
E.coli the fully weaned state, with their expression.
indicates number of mutually
covarying taxa, and connection S.gallolyticus
R.torques S.ventriculi
between nodes indicates Prevotella CONCLUSION: The ecogroup represents a sim-
Clostridiales
covariance between two taxa. P.copri E.rectale plified feature of community organization and
(Bottom) Measuring the (840914) P.copri components that could play key roles in commu-
L.ruminis
representation of ecogroup (588929) nity assembly and function. As the gut microbiota
Ecogroup taxa
taxa reveals that children with Independently E.faecalis constantly faces environmental challenges, “em-
Prevotella
SAM treated with standard varying taxa bedding” a sparse network of covarying taxa in a
Prevotellaceae
therapeutic foods have an larger framework of independently varying orga-
ecogroup profile similar to that nisms could represent an elegant architectural
MAM Healthy
of children with untreated MDCF3
MAM solution developed by nature to maintain robust-
0.02 MAM RUSF MAM ness while enabling adaptation. The approach
MAM, indicating persistent
MDCF1 MDCF2
perturbations in their gut 0.01
used to identify and characterize the sparse net-
MAM work of covarying ecogroup taxa is, in principle,
▪
community relative to healthy untreated
PC3 (15%)
children. In contrast, children 0 generalizable to a wide variety of ecosystems.

with MAM treated with a SAM 6 mo post-discharge
-0.01
therapeutic food designed to SAM at discharge SAM 12 mo post-discharge
-0.02 The list of author affiliations is available in the full article online.
target the microbiota (MDCF-2) SAM 1 mo post-discharge
*Corresponding author. Email: jgordon@wustl.edu
have an ecogroup profile that This is an open-access article distributed under the terms of
-0.03
overlaps nearly entirely with 0.02
the Creative Commons Attribution license (http://creative-
that of healthy children. commons.org/licenses/by/4.0/), which permits unrestricted
0 0 use, distribution, and reproduction in any medium, provided
PC2 SAM untreated -0.01
(25 -0.02 -0.02 the original work is properly cited.
%) -0.03 31%) Cite this article as A. S. Raman et al., Science 365, eaau4735
-0.04 -0.04 PC1 ( (2019). DOI: 10.1126/science.aau4735

R ES E A RC H
◥ In this spirit, we have developed a computa-

RESEARCH ARTICLE tional workflow to calculate temporally conserved
covariance of gut bacterial taxa over time in
members of a healthy Bangladeshi birth cohort
MICROBIOTA sampled monthly for the first five postnatal years.
The results revealed a network of 15 covarying
bacteria that we term an “ecogroup.” Ecogroup
A sparse covarying unit that taxa not only describe healthy gut microbial de-
velopment in children residing in Bangladesh
describes healthy and impaired as well as several other low- and middle-income
countries; they also distinguish the microbiota
human gut microbiota development of Bangladeshi children with untreated moder-

ate and severe acute malnutrition and the degree
to which these communities are reconfigured
Arjun S. Raman1,2, Jeanette L. Gehrig1,2, Siddarth Venkatesh1,2, Hao-Wei Chang1,2, toward a healthy state in response to several
Matthew C. Hibberd1,2, Sathish Subramanian1,2*, Gagandeep Kang3, Pascal O. Bessong4, therapeutic food interventions. Colonizing germ-
Aldo A.M. Lima5, Margaret N. Kosek6,7†, William A. Petri Jr.8, Dmitry A. Rodionov9,10, free piglets with a consortium of ecogroup taxa
Aleksandr A. Arzamasov9,10, Semen A. Leyn9,10, Andrei L. Osterman10, Sayeeda Huq11, and following them during the transition from
Ishita Mostafa11, Munirul Islam11, Mustafa Mahfuz11, Rashidul Haque11, exclusive milk feeding through weaning onto a
Tahmeed Ahmed11, Michael J. Barratt1,2, Jeffrey I. Gordon1,2‡ representative diet consumed by Bangladeshi
children recapitulates features of healthy com-
Characterizing the organization of the human gut microbiota is a formidable challenge munity development and reveals microbial ge-
given the number of possible interactions between its components. Using a statistical nomic features and expressed metabolic attributes
approach initially applied to financial markets, we measured temporally conserved important for fitness during succession.
covariance among bacterial taxa in the microbiota of healthy members of a Bangladeshi
Identifying the ecogroup
birth cohort sampled from 1 to 60 months of age. The results revealed an “ecogroup”
of 15 covarying bacterial taxa that provide a concise description of microbiota Thirty-six members of a birth cohort with con-
development in healthy children from this and other low-income countries, and a means for sistently healthy anthropometric scores living
monitoring community repair in undernourished children treated with therapeutic foods. within the Mirpur district of Dhaka, Bangladesh,
Features of ecogroup population dynamics were recapitulated in gnotobiotic piglets as underwent monthly fecal sampling for the first
they transitioned from exclusive milk feeding to a fully weaned state consuming a 60 months of postnatal life [height-for-age Z
representative Bangladeshi diet. score (HAZ), –0.92 ± 1.19 (mean ± SD); weight-
for-height Z score (WHZ), –0.48 ± 1.33; n =
I
1961 fecal samples, 55 ± 4 samples collected
nnumerable studies of the functioning of given species, that can engage in higher-order per individual; table S1]. In Bangladesh, the
biological systems have underscored the interactions with other community members. median duration of breastfeeding is 4 months,
importance of characterizing interactions Using a conservative species count of 100, the whereas the weaning process is long, with a
between their component parts (1–5). De- number of terms needed to mathematically rep- median of 25 months (14). Samples collected
fining microbial communities in this way resent all possible species-species interactions less frequently, or only after 36 months, from
can present a seemingly intractable challenge (pairwise and higher-order) is ~1030. A central 19 other children from Mirpur were also in-
(1–3, 6). For example, the gastrointestinal tract question is how biologically important inter- cluded in our analysis (HAZ, –0.58 ± 1.12; WHZ,
of a healthy adult human harbors multiple actions between component members can be –0.25 ± 0.96; n = 25.7 ± 10.5 samples per child).
species, with multiple strain-level variants of a identified so as to reduce the number of fea- Amplicons generated from variable region 4 (V4)
tures necessary for characterization of microbial of bacterial 16S rRNA genes present in these
1
Edison Family Center for Genome Sciences and Systems
community properties, such as assembly during 2455 fecal samples were sequenced, and the
Biology, Washington University School of Medicine, St. Louis, the postnatal period, or temporal responses to resulting reads were assigned to operational
MO 63110, USA. 2Center for Gut Microbiome and Nutrition various perturbations. taxonomic units with ≥97% nucleotide sequence
Research, Washington University School of Medicine, Co-occurrence analysis has been used to de- identity (97%ID OTUs) (15, 16) (fig. S1). In total,
St. Louis, MO 63110, USA. 3Translational Health Science and
Technology Institute, Faridabad, Haryana, India. 4HIV/AIDS
scribe community organization but is limited 118 97%ID OTUs were represented at a relative
and Global Health Research Programme, Department of in its ability to describe interactions between fractional abundance of at least 0.001 (0.1%)
Microbiology, University of Venda, Thohoyandou 0950, South microbes (7, 8). Recently developed approaches in at least two of the samples collected over
Africa. 5Center for Global Health, Department of Physiology have focused on defining microbe-microbe inter- the 60-month period.
and Pharmacology, Clinical Research Unit and Institute of
Biomedicine, School of Medicine, Federal University of Ceará,
actions using cross-sectional data (9, 10), although An initial broad description of microbiota
Fortaleza, CE 60430270, Brazil. 6Department of International these methods were not explicitly designed to development in this cohort was obtained by
Health, Bloomberg School of Public Health, Johns Hopkins address the temporal conservation of these in- applying unweighted and weighted UniFrac
University, Baltimore, MD 21205, USA. 7AB PRISMA, Ramirez teractions in, for example, longitudinal studies. to compute overall phylogenetic dissimilarity
Hurtado 622, Iquitos, Peru. 8Departments of Medicine,
Microbiology, and Pathology, University of Virginia School of
Therefore, we turned to approaches developed between gut communities from the 36 children
Medicine, Charlottesville, VA 22908, USA. 9A. A. Kharkevich in the fields of econophysics and protein evo- sampled monthly from 1 to 60 months and 49
Institute for Information Transmission Problems, Russian lution. Applying the concept of statistical co- fecal samples collected in a previous study from
Academy of Sciences, Moscow 127994, Russia. 10Infectious variance coupled with analytical techniques 12 unrelated adults, aged 23 to 41 years, living
and Inflammatory Disease Center, Sanford Burnham Prebys
Medical Discovery Institute, La Jolla, CA 92037, USA.
of matrix decomposition has identified co- in Mirpur (17). This metric indicated that the
11
International Centre for Diarrhoeal Disease Research, fluctuating economic sectors and cooperative mean “infant/child-to-adult” distance decreases
Bangladesh, Dhaka 1212, Bangladesh. amino acid networks of functional relevance to “adult-to-adult” by 3 years of age (fig. S2, A
*Present address: Department of Medicine, Massachusetts General (11–13). The underlying presumption is that co- and B). Alpha diversity also increased to adult-
Hospital, Boston, MA 02114, USA. †Present address: Department
of Medicine, University of Virginia School of Medicine, Charlottesville,
variation that is conserved is covariation that like levels during this time period (fig. S2, C and
VA 22908, USA. may be informative about the organization of D). As an additional description of community
‡Corresponding author. Email: jgordon@wustl.edu complex dynamic systems. development, we used the 16S rDNA dataset to
Raman et al., Science 365, eaau4735 (2019) 12 July 2019 1 of 11

construct a sparse Random Forests (RF)–derived month (Fig. 1B and table S2B). PCA performed Three different matrices were created where
model comprising age-discriminatory taxa (fig. S3, on this matrix revealed that PC1 encompassed each row was a fecal sample collected from an
A to E). Microbiota “age” can be computed by 80% of the data variance (Fig. 1C; see supple- individual at a particular month in the healthy
noting the fractional abundances of these age- mentary text for a sensitivity analysis of the Bangladeshi cohort and columns were either (i)
discriminatory taxa in a given sample obtained workflow). A group of 15 covarying taxa repre- all 118 taxa, (ii) the 15 ecogroup taxa, or (iii) the
at a given time point (14). Applying the RF- sented the top 20% of all taxa projections along remaining 103 non-ecogroup taxa. PCA was
generated model disclosed a high degree of PC1 (Fig. 1C; see table S3 for different thresh- performed on the rows of these matrices; fecal
correlation between microbiota age and chron- old cutoffs). They include OTUs assigned to samples were plotted on the first three principal
ologic age (R2 = 0.8) (fig. S3C). Although these Bifidobacterium longum, another member of components. The left panel of Fig. 2A shows the
approaches provide measures of community Bifidobacterium, Faecalibacterium prausnitzii, results obtained when considering the fractional
development, they do not characterize inter- a member of Clostridiales, Prevotella copri, representation of all 118 taxa in fecal samples
actions between community members during Streptococcus thermophilus, and Lactobacillus collected at postnatal months 4, 10, and 20.
this process. ruminis, all of which are age-discriminatory There is substantial interpersonal variation in
Principal components analysis (PCA) applied bacterial strains identified from RF-based anal- gut community structure at postnatal month 1,
to taxa present in monthly fecal samples offers ysis of bacterial 16S rDNA datasets generated as evidenced by the broad distribution along
a way to mathematically characterize gut micro- from healthy members of this Bangladeshi co- PC1, but this variation converges by month 4
biota organization by defining principal com- hort (fig. S3D). (Fig. 2A, fig. S7, A and B, and table S2D). There-
ponents (eigenvectors). The result of PCA is a The results of PCA performed on data gen- after, changes in the structure of the fecal micro-
ranked list of principal components (principal erated from 478 samples collected from children biota are depicted by right-to-left movement
component spectrum or “eigenspectrum”) where sampled at postnatal months 50 to 60 provide along PC1, with minor variance observed along
each principal component carries a percent- an illustration of statistical covariation between PC2 and PC3. Minimal movement along PC1 is
age of data variance. Tracking the principal these taxa: PC1 reveals that B. longum (OTU observed after month 20 (fig. S7C), consistent
component spectrum through time offers a 559527) and L. ruminis (OTU 1107027) positively with the results of iPCA in fig. S4, B and C. (No-
description of the evolving temporal organiza- covary with one another across samples and neg- tably, children in this cohort had completed
tion of the gut microbiota. The approach we atively covary with two P. copri strains (OTUs weaning by month 23; see fig. S8 for a description
used, iterative PCA (iPCA), is described in fig. S4A. 840914 and 588929); PC2 documents how two of the nature and timing of their dietary tran-
For each month, we created a matrix where the F. prausnitzii OTUs (514940 and 851865) posi- sitions.) Ecogroup taxa recapitulate the variance
rows were fecal samples and the columns com- tively covary with each other and negatively depicted by PC1, PC2, and PC3. Moreover, the
prised the 118 taxa described above. In the ex- covary with S. gallolyticus (OTU 349024) and ecogroup taxa capture (i) the significant inter-
ample shown, time point 1 considers monthly E. coli (OTU 1111294); PC3 discloses that the personal variation observed at postnatal month 1,
fractional abundance data from month 1 and a two P. copri OTUs negatively covary with the (ii) the subsequent convergence to a B. longum–
reference time point. The dissimilarity between two F. prausnitzii OTUs (Fig. 1D). predominant microbiota at postnatal month 4,
the two time points is reflected in the primary Figure 1E provides a graphical depiction of and (iii) temporal changes noted at postnatal
principal component (PC1). The system is con- this network of covarying taxa. Each green node months 10 and 20 (Fig. 2, A and B, fig. S7, A and B,
sidered to be “stable” at the time point where represents one of the 15 OTUs that manifest a middle panels, and table S2E). In contrast, the
adding subsequent months’ data negligibly con- high degree of conserved covariance between remaining 103 non-ecogroup taxa provide a
tributes to variance; mathematically, this is when months 20 and 60. Two nodes are connected by less informative representation of developmen-
the eigenvalue of PC1 reaches an asymptote. an edge if their temporally averaged covariance tal changes in the microbiota, as exemplified by
i; j
We performed iPCA on sequentially joined value (hCbin it from Fig. 1B) is within the top 20% the fact that PC1, PC2, and PC3 each capture
monthly data with month 36 taken as a refer- of all such values. Node size is proportional to ≤10% of the variance (Fig. 2A and fig. S7, A and
ence (fig. S4B). Month 36 was chosen on the basis the number of connections (edges) present. The B, right panels). The importance of taxa with
of the results of phylogenetic dissimilarity and green nodes collectively covary with one another. low average fractional abundances and large
diversity measurements presented in fig. S2 [note In contrast, gray nodes depict taxa that covary standard deviations, such as P. copri (Fig. 2B,
that previous cross-sectional studies using these with green nodes but not with one another inset), is often overlooked when they are con-
metrics had also indicated that an adult-like con- (Fig. 1E). The green nodes constitute an “insu- sidered in isolation. However, analysis of taxon-
figuration was achieved by this time point; e.g., lated” ecostructure; its members exhibit signif- taxon covariation can reveal relationships between
(18)]. iPCA revealed that month 20 and beyond icant intragroup covariation (fig. S6 and table member species, as illustrated by P. copri and
signify a time period of minimal structural var- S2C). We chose the term “ecogroup” to reflect B. longum (Fig. 1D, blue box).
iation in the gut microbiota (fig. S4B). This con- the conserved collective statistical covariation To determine the extent to which the eco-
clusion was supported by using the very last of this sparse network of 15 organisms. group is a generalizable descriptor of the micro-
time point in the 5-year longitudinal study as biota in infants and children with healthy growth
the reference (fig. S4C). Therefore, we were able Microbiota development in other phenotypes, we turned to the MAL-ED network
to design a workflow to compute reproducible birth cohorts of study sites located in low- and middle-income
covariance (covariance conserved across time in We asked whether components of the ecogroup countries (20, 21). Fecal samples had been col-
a mature community assemblage, as opposed to provide a concise description of postnatal devel- lected monthly for the first two postnatal years,
transient covariance that may occur during com- opment of the microbiota in healthy members allowing sparse 30-taxon RF-generated models of
munity assembly) using months 20 to 60 without of the Bangladeshi cohort and, if so, whether normal community development to be generated
having to make any a priori assumptions about changes in the representation of these taxa fol- from members of birth cohorts residing in Loreto,
the importance of any taxon. For each month low a pattern that is shared across other healthy Peru (periurban area) and Vellore, India (urban
spanning postnatal months 20 to 60, we calcu- birth cohorts representing distinct geographic area) (supplementary text, fig. S9, and table S4).
lated the covariance between the 118 taxa over all locales and anthropologic features (20). Moreover, Our ability to identify a network of covarying
individuals to generate monthly taxon-taxon co- we postulated that if ecogroup taxa are informa- taxa in the Mirpur cohort depended on a high-
variance matrices (19) (see Fig. 1A, fig. S5, and tive biomarkers of normal community develop- resolution time-series study that extended well
table S2A). The matrices were averaged to a ment, these taxa might be useful for characterizing beyond the month at which the microbiota was
i; j
single taxon-taxon matrix ( hCbin it ) that repre- impaired development and/or the extent to which determined to be “stable” (month 20). This du-
sented a definition of consistent covariance where community repair is achieved as a function of ration of sampling did not occur at these other
i and j are bacterial taxa and t designates the various therapeutic interventions (21). MAL-ED sites, obviating our ability to identify

Fig. 1. Defining a sparse, consistently covarying network of bacterial covary with each other, hCi;j bin it = 0. The matrix shown illustrates sparse
taxa (“ecogroup”) in healthy Bangladeshi children. (A) Workflow. temporally conserved coupling, with many taxa showing no consistent
Left: 16S rDNA sequencing of fecal microbiota samples collected monthly covariance (hCi;j
bin it ≈ 0; white pixels) but a few exhibiting a high degree
from healthy members of the birth cohort from postnatal months of conserved covariance (hCi;j bin it ≥ 0.5; deep red pixels). (C) Eigende-
20 to 60. For each month, a matrix is created where rows are taxa and composition of temporally conserved covariance matrix. Note that 80% of
columns are fecal samples of individuals. Center: Taxon-taxon covariance the data variance in hCi;j bin it can be represented by a single principal
matrices for each month are calculated. Right: Monthly taxon-taxon component. The histogram shows projections of taxa along PC1; data are
covariance matrices are normalized relative to the maximum monthly fit to a generalized extreme value distribution (red line). Applying a 20%
covariance value. If a normalized monthly covariance value for a given (i, j) threshold to this distribution identifies 15 taxa that reproducibly covary
taxon-taxon pair is within the top or bottom 10% of all monthly covariance over time. (D) Fecal samples from postnatal months 50 to 60 shown on a
values, it is converted to a “1”; otherwise it is assigned a “0”. This binarized PCA space ordinated by the 15 taxa in (C). Heat maps illustrate the
covariance matrix is defined as Ci;j i;j
bin . Concatenating Cbin for all months fractional abundance of taxa responsible for the variance along each
creates a three-dimensional matrix, ðCi;j Þ
bin t. (B) Temporally conserved taxon- principal component. The blue box shown in the left portion of the
taxon covariance matrix. The binarized covariance values for each projection along PC1 highlights the subset of healthy children who have a
(i, j) pair of taxa in ðCi;j
bin Þt are averaged over all months to give a temporally high representation of P. copri relative to B. longum. (E) Graphical
weighted covariance value for each taxon-taxon pair (hCi;j bin it ). In the limit representation of the sparse covarying network of 15 taxa (green
that two taxa always covary with each other, hCi;j bin it = 1. If two taxa never nodes). See text for details.

A All taxa (118) Ecogroup Non-ecogroup the variance depicted by PC1, PC2, and PC3 as
taxa (15) taxa (103) compared to considering all taxa, and (ii) changes
Month 4
in their fractional abundances followed tempo-
ral patterns similar to those documented in the
(8%)
0.1 0
PC3
(9%)
(10%)
PC3
PC3
-0.1 0.1 -0.4 Bangladeshi cohort (fig. S10 and table S2F).
-0.05
0.08
0.2
Ecogroup configuration in acute
PC
PC
0.04 0.1
PC
0.1
2(
2(
malnutrition before and after treatment
2(
0 0.05 -0.08
11
0.08 0.15
12
-0.06 0
9%
0.06 -0.04 0.1
%)
0
%)
0.04 ) 0.05
-0.04 0.02 0 -0.02 1 (55% -0.1 Bangladeshi children with acute malnutrition
)
0 0%) -0.05 0 %)
1 (5 PC ( 1 0
PC PC1 have perturbed microbiota development; their
Month 10 gut communities appear younger than those of
0
chronologically age-matched individuals (14, 21).
(8%)
PC3
0.1
(10%)
-0.4
(9%)
PC3
PC3
0.1
-0.1 -0.05 We examined whether ecogroup taxa provide
0.2
0.08 a useful way to characterize the microbiota of
PC
0.1
PC
0.1
PC
children with moderate or severe acute mal-
2(
0.04
2(
2(
0.05 -0.08 0 0.15
9%
-0.06 0.1
12
11
0 -0.04 0 0.05 %)
nutrition (MAM and SAM, respectively) prior
)
0.06 0.08 0
%)
%
) -0.1
0 -0.02 (55% -0.05 (10
)
0.04
PC1 to and after food-based therapeutic interventions.
-0.04 0.02 ) PC1
0 (50% In the accompanying paper, Gehrig et al. describe
PC1
Month 20 63 children from Mirpur diagnosed with MAM,
(10%) 0
(8%)
PC3
(9%)
PC3
PC3
0.1 0.1 aged 12 to 18 months, who were enrolled in a
-0.4
-0.1 -0.05
0.2
double-blind, randomized, controlled feeding
0.08
trial of different microbiota-directed comple-
PC
PC
PC
0.1 0.1
0.04
2(
2(
2(
0.05 -0.08 0 0.15 mentary foods (MDCFs) (21). Fecal samples
9%
12
11
0 -0.06 0.1
0.05 were collected for 9 weeks at weekly intervals.
0.06 0.08
%)
-0.04
)
0 -0.1 0 )
10%
)
0.04 -0.02 ) -0.05

-0.04 0.02 ) 0 (55% PC 1 (
The first 2 weeks comprised a pretreatment ob-
0
PC1
(5 %
0 PC1
servation period. Over the next 4 weeks, chil-
dren received either one of three MDCFs, or a
B ready-to-use supplementary food (RUSF) rep-
0.08 resenting a form of conventional therapy that,
abundance, P. copri
Average fractional
unlike the MDCFs, was not designed to target

specific members of the gut microbiota and re-
1
0.04 pair community immaturity. The last 2 weeks
represented the post-treatment observation pe-
Average fractional abundance
riod. In total, we identified 945 97%ID OTUs

0
that had a fractional abundance of at least 0.001
1 2 3 4 5 10 20 40 60
Month (0.1%) in at least two fecal samples collected from
0.5 one or more participants prior to, during, and
after treatment (n = 531 samples). Gehrig et al.
(21) also describe another trial involving 54 hos-
pitalized Bangladeshi children with SAM, aged
6 to 36 months, where each participant was
1
0 2 treated with one of three standard therapeutic
m 3 foods and then followed over a 12-month period
gu ium cus 4
lon cter ti ini
s li 5 after discharge. In total, we identified 944 97%
B. a oly m co 0) 10
d ob all L .ru E. 94 les 9)
Month
ID OTUs that had a fractional abundance of at
Bi
f i
S.
g 14 dia 892 65
) 20
i i ( 5
s tri 8 18 ecta
le
4) 40 least 0.001 in at least two fecal samples collected
z o 5 a
n i t C l
r i (
( 8 5 .r 91 ell us 60 0 0.8
s o p i i E 40 ot hil lis from one or more participants in this trial (n =
ra
u . c i t z ( 8 ev op ca er
p P n r i P r
ae ist
F.
r a u s
. cop
e r m
E .f
D ial Average fractional 618 samples).
p P t h
F. S. abundance A matrix was created that included (i) all fecal
samples from the SAM trial, (ii) pretreatment
Fig. 2. Characterizing healthy gut microbiota development in the Bangladeshi birth cohort. samples from children with MAM enrolled in all
(A) PCA spaces were created. Each point in the spaces represents a fecal sample described by four arms of the MDCF trial, (iii) MAM samples
either all taxa present at a fractional abundance greater than 0.001 (0.1%) (118 taxa), ecogroup taxa obtained 2 weeks after treatment with one of
(15), or non-ecogroup taxa (103). The spatial distribution of fecal samples in each PCA space is the three MDCFs or the RUSF, and (iv) fecal
shown for the indicated postnatal months. (B) Bar graph illustrating average fractional abundance samples from age-matched healthy Bangladeshi
of ecogroup taxa as a function of postnatal month (see table S2E). Inset: Average fractional children (table S5). Each row of the matrix was a
abundance (±SD) of P. copri as a function of time. fecal sample, each column was an ecogroup taxon,
and each element in the matrix was the frac-
conserved covariance among taxa. However, to identified from the Bangladeshi birth cohort. PCA tional abundance of an ecogroup taxon within a
test how well the 15 ecogroup taxa identified in was performed on the rows of these matrices and particular fecal sample. PCA was performed
the Mirpur cohort could characterize the devel- the same analysis performed as described for on the rows of this matrix. Centroids for each
oping microbiota of children living in these the healthy Bangladeshi birth cohort. The re- cohort were computed and plotted on the PCA
countries, we created two matrices where each sults show that the ecogroup taxa identified in space (Fig. 3A). At the time of discharge, after
row was a fecal microbiota sample from the members of the healthy Bangladeshi cohort receiving standard therapeutic foods, the mi-
Indian or Peruvian cohorts and columns were also provide a concise description of commu- crobiota of children with SAM remained in an
either all taxa identified in the Peruvian and nity development in healthy members of these incompletely repaired state: Although there was
Indian samples or just the 15 ecogroup taxa other two birth cohorts; that is, (i) they capture some improvement at 1 month after discharge,

there was minimal additional improvement evi- whereas the microbiota of those treated with and mediators of metabolism, bone growth, cen-
dent at 6 or 12 months, at which times their MDCF-2 closely resembled that of healthy chil- tral nervous system development, and immune
microbiota resembled that of untreated children. Notably, MDCF-2 was also distinct among function [see (21) for details].
dren with MAM (Fig. 3A). The microbiota of the four treatment types in eliciting changes in PCA measures the effect of treatment on the
children with MAM that were treated with the plasma proteome indicative of improved gut microbiota by considering a constellation
MDCF-1, MDCF-3, and RUSF clustered together, health status, including changes in biomarkers of changes in fractional abundance of ecogroup
Fig. 3. Ecogroup taxa define the response of the microbiota of children with SAM and MAM to various nutritional interventions. (A) Centroids
of each indicated cohort are plotted on a PCA space. Arrows indicate the temporal progression of microbiota reconfiguration for children with SAM
treated with conventional therapy and children with MAM treated with a RUSF or a MDCF. (B) Matrix decomposition of the axes shown in (A) highlights
the taxa that are important for fecal sample variance observed along each principal component. (C and D) Average fractional abundance of ecogroup
taxa identified in (B) in the fecal microbiota of members of the SAM and MAM cohorts as a function of treatment (see table S2G).

taxa, with the premise that the fractional abun- as clearly separate MAM from healthy or (by [Mirpur-18 (21)] was added to food bowls con-
dances as well as the covariation of these taxa extension) the differential effects of MDCF taining Soweena. On postnatal day 7, piglets
are important for characterizing community con- treatment, although it does place MDCF-2– were gavaged with a second consortium con-
figuration. The left panel of Fig. 3B shows that treated microbiota closest to that of healthy sisting of 16 additional cultured sequenced eco-
the relationship between the fractional represen- children (fig. S11C). One explanation is that group taxa (fig. S13A) representing 13 of the 15
tations of B. longum (OTU 559527) and E. coli P. copri does not contribute as prominently to the species shown in Fig. 1C. During postnatal days
(OTU 1111294) determines microbiota position collective group of correlated taxa identified by 5 to 22, the amount of Mirpur-18 added to food
along PC1 in Fig. 3A. The center and right panels SparCC (fig. S11 and table S6, A and B). SPIEC- bowls was progressively increased while the
of Fig. 3B show that the relationship between the EASI identifies P. copri and other Prevotella amount of Soweena was decreased; once a fully
fractional representations of B. longum, E. coli, OTUs as key microbes (fig. S12, A and B, and table weaned state was achieved on day 22, animals
S. gallolyticus (OTU 349024), and P. copri (OTUs S6, C to E). However, SPIEC-EASI does not pro- were monotonously fed the Mirpur-18 diet un-
588929 and 840914) determines position along vide as informative a description of the temporal til they were euthanized on postnatal day 29.
PC2, whereas position along PC3 reflects pattern of healthy gut microbial development Piglets increased their weight by 185 ± 31%
the relationship between the abundances of as does the ecogroup taxa [note the relative lack (mean ± SD) between postnatal days 7 and 29.
S. gallolyticus and the two P. copri OTUs. B. longum, of movement over time of community configu- To define features in ecogroup strains that
S. gallolyticus, and E. coli are the predominant ration from right to left along PC1 in fig. S12C relate to their fitness during the series of dietary
ecogroup taxa represented in the microbiota of compared to Fig. 2A (ecogroup taxa) and fig. transitions that mimic those experienced by
children with untreated SAM (Fig. 3C and table S11B (SparCC)]. The 15 interacting taxa iden- children living in Mirpur, we performed short-
S2G). Treatment results in movement of their tified by SPIEC-EASI separate untreated and read shotgun sequencing of community DNA
microbiota along PC1 and PC3 in Fig. 3A; treated SAM and MAM microbiota from one prepared from rectal swabs obtained at 11 time
this movement is associated with a decrease in another and from healthy (fig. S12D). As with points spanning experimental days 5 to 29 (fig.
B. longum, S. gallolyticus, and E. coli (Fig. 3C and the two other approaches, although less clearly S13A) and along the length of the gut at the
table S2G). Differences between the microbiota than with the ecogroup taxa, SPIEC-EASI shows time of euthanasia. The results are presented in
of healthy children and those with SAM prior that MDCF-2 is most effective in changing the Fig. 4A and table S2H. After gavage of remain-
to and during the 12 months after treatment configuration of the MAM-associated micro- ing ecogroup members, the representation of
with standard therapeutic foods are manifest by biota toward a healthy state relative to MDCF-1, all B. longum strains diminished rapidly. From
differences in their respective positions along MDCF-3, and RUSF. Together, these findings pro- postnatal day 8 to day 22, as the animals were
PC1 and PC3 (Fig. 3A). These differences sig- vide support for considering temporally conserved being weaned, S. gallolyticus, E. coli, E. avium,
nify incomplete repair to a “healthy” state and taxon-taxon covariance when characterizing the L. salivarius, and P. copri exhibited distinct
highlight the need to achieve further decreases microbiota of children with undernutrition prior patterns of temporal change in their represen-
in the fractional abundance of B. longum (asso- to and after various therapeutic interventions. tation. After the animals were fully weaned, there
ciated with movement to the right of PC1) along was a pronounced increase in P. copri, which be-
with further decreases in the fractional abun- Ecogroup taxa in a gnotobiotic came the dominant member of the cecal, colonic,
dance of S. gallolyticus and increases in P. copri piglet model of postnatal Bangladeshi and fecal microbiota (Fig. 4A and fig. S13B). The
(associated with positive movement along dietary transitions relationship between the abundances of P. copri
PC3). The representation of B. longum, P. copri, Our observations raise questions about the and B. longum is comparable in these piglets to
S. gallolyticus, and E. coli in the microbiota of nature of the interactions among B. longum, that observed in the healthy Bangladeshi chil-
12- to 18-month-old children with untreated MAM P. copri, and other ecogroup taxa during post- dren who were used to evaluate the microbiota
accounts for their positive projection along PC1 natal development as a function of the dietary configurations of untreated and treated children
and PC3 relative to the microbiota of children transitions that occur when children progress with MAM and SAM (Fig. 3, C and D).
with untreated SAM (Fig. 3A). Among the tested from exclusive milk feeding to complementary The representations of 81 mcSEED metabolic
therapeutic foods, MDCF-2 was uniquely asso- feeding to a fully weaned state. To address this modules (see methods) in strain genomes were
ciated with a positive movement along PC1 (Fig. issue, we colonized germ-free piglets with used to make in silico predictions about their
3A); this corresponds to decreased fractional ecogroup taxa and tracked the dynamics of capacity to synthesize amino acids and B vita-
abundance of B. longum (Fig. 3D and table S2G) consortium members over time. We turned to mins, utilize a variety of carbohydrates, and
and more complete community repair. gnotobiotic piglets rather than mice because generate short-chain fatty acids. Predicted pheno-
Two other methods, SparCC and SPIEC-EASI, the former have physiologic and metabolic qual- types were scored as either a “1” or a “0” sig-
have been used to describe microbiota organi- ities more similar to that of humans (22). Piglets nifying auxotrophy or prototrophy in the case of
zation (9, 10). As these methods were designed were derived as germ-free at birth and were fed amino acid and B-vitamin biosynthesis, or the
for cross-sectional studies, we adapted them an irradiated sow’s-milk replacement (Soweena) ability or inability to utilize various carbohydrates
(see supplementary text) so we could compare for the first four postnatal days (fig. S13A). Piglets (table S8). PCA of a “binary phenotype matrix” of
their ability to identify (i) temporally conserved (n = 5) were then colonized, by oral gavage, all strains present at a fractional representation
aspects of community organization, and (ii) the with a consortium of seven cultured, sequenced of ≥0.001 in fecal samples collected from post-
degree to which SAM and MAM microbiota are B. longum strains recovered from the fecal mi- natal day 8 to day 18 identified 14 carbohydrate
repaired with different food-based interventions crobiota of children living in Mirpur, Bangladesh utilization pathways, plus the capacity to synthe-
with the approach we had used to identify the as well as three other countries (Peru, Malawi, size cysteine, folate, and pantothenate as genomic
ecogroup. SparCC identifies a subset of eco- and the United States) (fig. S13A). On the basis features that distinguish these strains from each
group taxa that describe healthy gut micro- of their genome sequences (table S7), six strains other (table S9). Hierarchical clustering by these
biota development in members of the 5-year were classified as B. longum subspecies infantis predicted metabolic phenotypes also grouped
healthy Bangladeshi cohort study (fig. S11, A and one as B. longum subspecies longum. The ga- these strains by their fitness (Fig. 4, B and C).
and B). SparCC clearly separates the microbiota vage mixture also contained two Bifidobacterium We performed microbial RNA-seq using cecal
of children with untreated SAM from healthy breve strains, which we used as comparators to contents to characterize the expression of genes
controls and shows that treatment with standard delineate factors that contribute to the fitness encoding components of mcSEED metabolic mod-
therapeutic foods fails to repair their microbiota of the B. longum strains, given the phylogenetic ules present within the ecogroup strains. [The frac-
to a healthy state, or even to a state seen in similarity of their genomes. Beginning on post- tional representations of these strains in the cecum
children with untreated MAM. Compared to the natal day 4, a diet representative of that con- and feces at the time of euthanasia were highly
approach described in Fig. 1A, SparCC does not sumed by 18-month-old children living in Mirpur correlated (r2 = 0.98; table S10).] Figure S14A

Fig. 4. Distinguishing genomic features related to the fitness coded according to whether they are predicted to endow the host
landscape of ecogroup strains in gnotobiotic piglets. (A) Average strain with prototrophy for amino acids and B vitamins or the capacity
fractional abundances of strains plotted over time (see table S10). to utilize the indicated carbohydrate. Strains are hierarchically
The summary of the experimental design shows when the various taxa clustered according to the representation of these metabolic pathways.
were first introduced by gavage and how the diet changed over time. See (C) Heat map depicting the fractional representation of the strains shown
fig. S13A for complete strain designations. (B) Genome features that in (B) at the indicated time points. Strains are hierarchically clustered
distinguish among strains whose average fractional abundances in the according to the mcSEED metabolic phenotypes in (B). Note that the
fecal microbiota of piglets was ≥0.001 between postnatal days 8 and 22. pattern of clustering defined by phenotypes also clusters strains by
These distinguishing features are mcSEED metabolic phenotypes color- their fitness.
illustrates the workflow used to generate a their mcSEED enrichment profiles correlated (ii) utilization of several carbohydrates (xylose
mcSEED “enrichment matrix” (ME) that signifies with their fractional representation (fitness) in and b-xylosides plus galacturonate/glucuronate/
the extent to which the aggregate transcript levels the cecal and fecal microbiota (Fig. 5A, inset). glucuronide); (iii) biosynthesis of queuosine; and
of components of a given mcSEED metabolic To identify which expressed components of (iv) uptake of cobalt related to cobalamin bio-
module in a given bacterial strain quantitatively mcSEED metabolic modules contribute to the synthesis (Fig. 5B and tables S2J and S11B).
differ from that of a reference strain. Because differences in the fractional representation, we Moreover, expression of four of these pathways
P. copri had the highest fractional representa- required a way to relate the principal compo- (cysteine and asparagine biosynthesis; xylose/
tion on postnatal day 29, it was used as the nents of the rows (metabolic modules) and col- b-xyloside and galacturonate/glucuronate/
reference (fig. S14B and table S2I). PCA was umns (strains) of the mcSEED enrichment matrix. glucuronide utilization) exclusively differentiate
performed on the mcSEED enrichment matrix To do so, we used singular value decomposition P. copri, B. luti, E. coli, and E. avium from all
(Fig. 5A and table S11A). The results revealed that (SVD; fig. S14, C and D). Relative to P. copri, the nine Bifidobacterium species and the other five
the transcriptomes of Bifidobacterium strains most distinguishing features of the Bifidobacterium strains whose transcripts were represented in
cluster together and are distinct from those of transcriptomes were markedly reduced or absent the community metatranscriptome (Fig. 5B).
P. copri, E. coli, B. luti, and E. avium. Moreover, expression of pathways involved in (i) biosynthesis The biological significance of expression of
the distribution of strains along PC1 based on of cysteine, tyrosine, tryptophan, and asparagine; these distinguishing mcSEED metabolic modules

their importance in maintaining their fitness

under these conditions. As such, the feature-
reduction approach used here provides a rationale
for testing nutritional interventions that target
these pathways in ecogroup members in chil-
dren at risk for, or who already have, perturbed
microbiota development.
Conclusions
We have developed a statistical approach to
identify a group of 15 covarying bacterial taxa
that we term an ecogroup. We found that the
ecogroup is a conserved structural feature of
the developing gut microbiota of healthy mem-
bers of several birth cohorts residing in dif-
ferent countries. Moreover, the ecogroup can
be used to distinguish the microbiota of chil-
dren with different degrees of undernutrition
(SAM, MAM) and to quantify the ability of their
gut communities to be reconfigured toward a
healthy state with a MDCF. Studies of gnoto-
biotic piglets subjected to a set of dietary tran-
sitions designed to model those experienced
by members of the Bangladeshi healthy birth
cohort demonstrate that temporal changes in
the fitness of ecogroup taxa can occur in the
absence of other gut community members. These
observations suggest that the approach used to
identify the ecogroup may be useful in charac-
terizing microbial community organization in
members of other longitudinally sampled (hu-
man) cohorts.
A critical feature of biological systems is that
they function reliably, yet adapt when faced with
environmental fluctuations (23, 24). An architec-
ture of sparse but tight coupling enables rapid
evolution to new functions in proteins (25, 26).
Studies of macro-ecosystems such as ant colonies
have argued that adaptive behaviors are depen-
dent on proper network organization (27). The
gut microbiota must satisfy the constraints of
survival: namely, withstanding insult and main-
Fig. 5. Distinguishing features of mcSEED metabolic module expression related to the fitness taining functionality (robustness) while still
of ecogroup strains in weaned gnotobiotic piglets. See fig. S13A for full strain designations. having the capacity for plasticity. “Embedding”
(A) The transcriptomes of cecal community members were classified on the basis of gene assignments a sparse network of covarying taxa in a larger
to 81 mcSEED metabolic modules (see count matrix in fig. S14B). Each strain is plotted on the first framework of independently varying organ-
two principal components of the enrichment matrix in fig. S14B. The inset shows that fractional isms could represent an elegant architectural
representation (fitness) of strains correlates with their expression profiles as judged by position solution developed by nature to maintain ro-
along PC1. (B) Singular value decomposition (SVD, fig. S14C) identifies which among the 81 bustness while enabling adaptation.
expressed metabolic modules most distinguish the indicated strains in the cecal community and
Mirpur-18 diet contexts (fig. S14D). (C) Expressed discriminatory metabolic modules identified by Methods
SVD in (B) are shown as complete or incompletely represented in the genomes of the indicated Human studies
strains by red pixels (predicted prototrophy for the amino acid or the ability to utilize the A previously completed NIH birth cohort study
carbohydrate shown) or by white pixels (auxotrophy or the inability to utilize the carbohydrate). (“Field Studies of Amebiasis in Bangladesh”;
Strains and metabolic modules are hierarchically clustered. ClinicalTrials.gov identifier NCT02734264) was
conducted at the International Centre for Diar-
demanded a further contextualization based have mcSEED binary phenotype profiles similar rhoeal Disease Research, Bangladesh (icddr,b).
on whether these systems were complete or to that of P. copri and contain complete meta- Anthropometric data and fecal samples were
incompletely represented in the strain genomes. bolic pathways for cysteine biosynthesis and collected monthly from enrollment through
Figure 5C shows that all of the Bifidobacterium xylose utilization (table S2J). These results in- postnatal month 60. Informed consent was ob-
strains contain complete metabolic pathways dicate that genomic features of the Bifidobac- tained from the mother or guardian of each
for tyrosine, asparagine, and tryptophan biosyn- terium strains examined limit their ability to child. The research protocol was approved by the
thesis but do not contain complete metabolic thrive in the context of the Mirpur-18 diet and institutional review boards of the icddr,b and the
pathways for cysteine biosynthesis; utilization a community that contains the other ecogroup University of Virginia, Charlottesville.
pathways for galactose, xylose, and glucuronides; strains. In contrast, the fact that P. copri and In the case of the MAL-ED birth cohort study
and B-vitamin synthetic pathways for queuosine other ecogroup strains contain and express (“Interactions of Enteric Infections and Mal-
and cobalamin. In contrast, E. coli and B. luti these metabolic pathways provides support for nutrition and the Consequences for Child Health

and Development”; ClinicalTrials.gov identifier tively (see supplementary text for a description of ian using protocols approved by the Washington
NCT02441426), anthropometric data and fecal the aggregate dataset). Model building for each University Animal Studies Committee.
samples were collected every month from enroll- birth cohort was initiated by regressing the re-
ment to 24 months of age. The study protocol lative abundance values of all identified 97%ID Diets
was approved by institutional review boards at OTUs in all fecal samples against the chronologic Piglets were initially bottle-fed with an irradiated
each of the study sites. age of each donor at the time each sample was sow’s milk replacement (Soweena Litter Life,
The accompanying paper by Gehrig et al. (21) procured (R package “randomForest,” ntree = Merrick; catalog number C30287N). Soweena
describes studies that enrolled (i) Bangladeshi 10,000). For each country site, OTUs were ranked powder (120-g aliquots in vacuum-sealed steri-
children with MAM in a double-blind, random- on the basis of their feature importance scores, lized packets) was gamma-irradiated (>20 Gy)
ized, four-group, parallel assignment inter- calculated from the observed increases in mean and reconstituted as a liquid solution in the gnoto-
ventional trial study of microbiota-directed square error (MSE) when values for that OTU biotic isolator (120 g per liter of autoclaved
complementary food (MDCF) prototypes con- were randomized. Feature importance scores were water). The procedure for producing Mirpur-18
ducted in Dhaka, Bangladesh (ClinicalTrials.gov determined over 100 iterations of the algorithm. is detailed in Gehrig et al. (21).
identifier NCT03084731); (ii) a reference cohort To determine how many OTUs were required to
of age-matched healthy children from the same create a RF-based model comparable in accuracy Husbandry
community; and (iii) a subcohort of 54 children to a model comprising all OTUs, we performed Feeding: The protocol used for generating germ-
with SAM who were treated with one of three dif- an internal 100-fold cross-validation where mod- free piglets was based on our previous publica-
ferent therapeutic foods and followed for 12 months els with sequentially fewer input OTUs were tion (29) with modifications (21). Piglets were
after discharge with serial anthropometry and compared to one another. Limiting the country- fed at 3-hour intervals for the first 3 postnatal
biospecimen collection (“Development and Field specific models to the top 30 ranked OTUs had days, at 4-hour intervals from postnatal days
Testing of Ready-to-Use Therapeutic Foods Made only minimal impact on accuracy (within 1% of 4 to 8, and at 6-hour intervals from postnatal
of Local Ingredients in Bangladesh for the Treat- the MSE obtained with all OTUs). In addition day 9 to the end of the experiment. Introduc-
ment of Children with SAM”; ClinicalTrials.gov to calculating the R2 of the chronological age tion of solid foods began on postnatal day 4
identifier NCT01889329). The research protocols versus predicted microbiota age for reciprocal and weaning was accomplished by day 22. Each
for these studies were approved by the Ethical cross-validation of the RF-derived models, we gnotobiotic isolator was equipped with four
Review Committee at the icddr,b. Informed con- also calculated the mean absolute error (MAE) stainless steel bowls and one 2-gallon waterer;
sent was obtained from the mother/guardian of and root mean square error (RMSE) for the ap- each 2-gallon waterer (Valley Vet, Maryville,
each child. Use of biospecimens and metadata plication of each model to each dataset to fur- KS; catalog number 17544) was equipped with
from each of the human studies for the analyses ther assess model quality (table S12). two 0.5-inch nipples (Valley Vet, catalog num-
described in this report was approved by the ber 17352). During the first 3 days after birth,
Washington University Human Research Protec- Comparing OTUs with DADA2 amplicon all four bowls were filled with Soweena. From
tion Office (HRPO). sequence variants (ASVs) (fig. S1) days 4 to 12, at each feeding, one bowl was filled
Each OTU in the ecogroup and each OTU in the with Mirpur-18 while the remaining three bowls
Collection and storage of fecal samples sparse RF-derived models that had 100% se- were filled with Soweena. On day 12, one bowl of
and clinical metadata quence identity to an ASV was identified; each milk was replaced with a bowl of water. From
Fecal samples were placed in a cold box with ice of these OTUs was defined as a “primary OTU day 15 to day 19, each daytime feeding consisted
packs within 1 hour of production by the donor sequence” and the ASV as the “correct ASV se- of placement of two bowls of water and two
and collected by field workers for transport back quence.” The primary OTU sequence was then bowls of Mirpur-18. In nighttime, one bowl of
to the lab (NIH Birth Cohort, MAL-ED study). mutated according to the maximum sequence water was replaced with Soweena (i.e., each iso-
For the “Development and Field Testing of Ready- variance accepted by QIIME for a ≥97%ID OTU lator at each feeding had two bowls of Mirpur-18,
to-Use Therapeutic Foods Made of Local In- (i.e., ≤3%) to create a library of 1000 derivative one bowl of water, and one bowl of Soweena).
gredients in Bangladesh for the Treatment of sequences. Each sequence in the library was then From postnatal days 20 and 21, only one bowl
Children with SAM” study, the healthy reference compared to a database of all ASVs produced was provided with Soweena, and the amount of
cohort, and the MDCF trial, samples were flash- from DADA2 analysis (28) of all 16S rDNA data- milk added was reduced by one half each day
frozen in liquid nitrogen–charged dry shippers sets generated from all birth cohorts described in during this period. On day 22, the last bowl of
(CX-100, Taylor-Wharton Cryogenics) shortly after this report and in Gehrig et al. (21). The ASV with milk was replaced with a bowl of water, thereby
their production by the infant or child. Biospeci- the maximum sequence identity to each mem- completing the weaning process. After weaning,
mens were subsequently transported to the local ber of each library of 1000 derivative sequences two bowls of fresh sterilized water and two bowls
laboratory and transferred to –80°C freezers was noted. If this ASV matched the correct ASV of fresh Mirpur-18 were introduced into each iso-
within 8 hours of collection. Samples were shipped sequence, the OTU derivative sequence in the lator every 6 hours to enable ad libitum feeding.
on dry ice to Washington University and archived library was assigned a “1”; otherwise it was as- The 2-gallon waterer was replenished with fresh
in a biospecimen repository at –80°C. signed a “0”. An average over all 1000 derivative sterilized water every 2 to 3 days. Mirpur-18 con-
sequences in a given library was then calculated. sumption was monitored by noting the amount
Sequencing bacterial V4-16S rDNA This process was iterated 10 separate times, of input food required to maintain a filled bowl
amplicons and assigning taxonomy creating 10 trials of 1000 derived sequences for during a 24-hour period. Piglets were weighed
Methods used for isolation of DNA from fro- each OTU. An average over all 10 trials was daily using a sling (catalog number 887600; Pre-
zen fecal samples, generation of V4-16S rDNA then calculated, thereby defining the prob- mier Inc., Charlotte, NC). Environmental enrich-
amplicons, sequencing of these amplicons, cluster- ability of an OTU being ascribed to the correct ment was provided within the isolators including
ing of sequencing reads into 97% ID OTUs, and as- ASV given the accepted sequence “entropy” of plastic balls for “rooting” activity and rubber hoses
signing taxonomy are described in Gehrig et al. (21). QIIME (15). The results demonstrated that V4- and stainless steel toys for chewing and manipu-
16S rDNA sequences comprising a 97%ID OTU lating. The behavior and health status of the pig-
Generation of RF-derived models of gut generated by QIIME map directly to the single lets were monitored every 3 to 4 hours throughout
microbiota development ASV sequence deduced by DADA2. the day and night during the first 13 postnatal days
We produced RF-derived models of gut micro- and then every 6 hours until the time of eutha-
biota development from the Peruvian, Indian, Studies of gnotobiotic piglets nasia on day 29.
and “aggregate” V4-16S rDNA datasets generated Experiments involving gnotobiotic piglets were Bacterial genome assembly, annotation,
from 22, 14, and 28 healthy participants, respec- performed under the supervision of a veterinar- in silico metabolic reconstructions, and phenotype

predictions: Barcoded, paired-end genomic libra- Euthanasia and assessment of community 11. V. Plerou et al., Random matrix approach to cross correlations
ries were prepared for each bacterial isolate, and composition along the length of the intestine: in financial data. Phys. Rev. E 65, 066126 (2002). doi: 10.1103/
PhysRevE.65.066126; pmid: 12188802
the libraries were sequenced (Illumina MiSeq Euthanasia was performed on experimental 12. S. W. Lockless, R. Ranganathan, Evolutionarily conserved
instrument; paired-end 150- or 250-nt reads). day 29 according to American Veterinary Med- pathways of energetic connectivity in protein families. Science
Reads were demultiplexed and assembled; con- ical Association (AVMA) guidelines. The small 286, 295–299 (1999). doi: 10.1126/science.286.5438.295;
tigs with greater than 10× coverage were initially intestine was divided into 20 sections of equal pmid: 10514373
13. N. Halabi, O. Rivoire, S. Leibler, R. Ranganathan, Protein
annotated using Prokka (30) followed by anno- length; the first 1 cm of the 1st, 5th, 10th, 15th, sectors: Evolutionary units of three-dimensional structure.
tation at various levels by mapping protein se- and 20th sections were opened with an incision, Cell 138, 774–786 (2009). doi: 10.1016/j.cell.2009.07.038;
quences to the Prokaryotic Peptide Sequence and luminal contents were harvested with sterile pmid: 19703402
database of the Kyoto Encyclopedia of Genes cell scraper (Falcon; catalog number 353085). 14. S. Subramanian et al., Persistent gut microbiota immaturity in
malnourished Bangladeshi children. Nature 510, 417–421
and Genomes (KEGG) as described in Gehrig et al. Luminal contents were also harvested from the (2014). doi: 10.1038/nature13421; pmid: 24896187
(21). Additional annotations were based on SEED, cecum, proximal colon (10 cm of the mid-spiral 15. J. G. Caporaso et al., QIIME allows analysis of high-throughput
a genomic integration platform that includes a region), and distal colon (10 cm from the anus). community sequencing data. Nat. Methods 7, 335–336 (2010).
growing collection of complete and nearly com- Methods for isolation of DNA from luminal and doi: 10.1038/nmeth.f.303; pmid: 20383131
16. A direct comparison of these OTUs and amplicon sequence
plete microbial genomes with draft annotations fecal samples, and short-read shotgun sequenc- variants (ASVs) identified using a bioinformatic pipeline
performed by the RAST server (31). SEED con- ing of community DNA samples (COPRO-seq), designed to reduce sequencing errors disclosed good agree-
tains a set of tools for comparative genomic are all detailed in Gehrig et al. (21). ment between the two methods (fig. S1 and methods).
analysis, annotation, curation, and in silico re- Microbial RNA-seq: Isolation of RNA from Therefore, we retained OTU designations for this study.
17. A. Hsiao et al., Members of the human gut microbiota involved
construction of microbial metabolism. Microbial cecal contents harvested from piglets at the in recovery from Vibrio cholerae infection. Nature 515,
Community SEED (mcSEED) is an application of time of euthanasia, depletion of ribosomal rRNA 423–426 (2014). doi: 10.1038/nature13738; pmid: 25231861
the SEED platform that we have used for manual (Ribo-Zero Kit, Illumina), and bacterial RNA pu- 18. T. Yatsunenko et al., Human gut microbiome viewed
curation of a large and growing set of bacterial rification were performed (21). Double-stranded across age and geography. Nature 486, 222–227 (2012).
doi: 10.1038/nature11053; pmid: 22699611
genomes representing members of the human complementary DNA and indexed Illumina li- 19. Each monthly covariance matrix was normalized against the
gut microbiota (currently ~2600). mcSEED sub- braries were prepared using the SMARTer Stranded highest covariance value for that month (see fig. S5, A to D,
systems (32) are user-curated lists/tables of RNA-seq kit (Takara Bio USA). Libraries were and table S2A for the example of month 60). Because some
specific functions (enzymes, transporters, tran- analyzed with a Bioanalyzer (Agilent) to determine taxon-taxon covariance values are zero as a result of the
absence of a taxon (e.g., fig. S5C), fitting a probability
scriptional regulators) that capture current (and fragment size distribution and then sequenced distribution over all of the covariance values becomes a
ever-expanding) knowledge of specific metabolic [Illumina NextSeq platform; 75-nt unidirectional practical constraint. Therefore, we retained the nonzero values
pathways, or groups of pathways, projected onto reads; 36.9 (±5.4) × 106 reads per sample (mean ± across months 20 to 60, yielding 80 of the original 118 taxa.
this set of ~2600 genomes. mcSEED pathways SD); n = 5 samples]. Fluorescence was not mea- Values in the normalized covariance matrix for each month
were then fit to a t-location scale probability distribution
are lists of genes comprising a particular meta- sured from the first four cycles of sequencing, as because the monthly normalized covariance histograms were
bolic pathway or module; they may be more gran- this library preparation strategy introduces three significantly heavy-tailed (e.g., fig. S5D). Given our desire to
ular than a subsystem, splitting it into certain nontemplated deoxyguanines. Transcripts were identify which taxon-taxon covariance values were consistently
aspects (e.g., uptake of a nutrient separately from quantified (34), normalized (transcripts per kilo- in the tails of these probability distributions over time, the
elements in each monthly covariance matrix were binarized to
its metabolism). mcSEED pathways are presented base per million reads; TPM), and then aggre- a “1” if they fell within the top or bottom 10% and a “0” if their
as lists of assigned genes and their annotations in gated according to their representation in mcSEED values were within the remaining 80% of the probability
table S7. As detailed in Gehrig et al. (21), predicted and KEGG subsystems/pathway modules (21). distribution; this isolated the most covarying taxon-taxon pairs
i;j
phenotypes are generated from the collection of [ðCbin Þt , where i and j are bacterial taxa and t designates the
RE FERENCES AND NOTES month]. Monthly binarized covariance matrices were then
mcSEED subsystems represented in a microbial averaged over time to create an 80 × 80 covariance matrix
genome and the results described in the form of 1. W. Z. Lidicker Jr., A clarification of interactions in
that signifies temporally conserved taxon-taxon covariation
ecological systems. Bioscience 29, 375–377 (1979).
a binary phenotype matrix (BPM; prototrophy or (hCi;jbin it , Fig. 1B).
doi: 10.2307/1307540
auxotrophy for an amino acid or B vitamin; the 20. MAL-ED Network Investigators, The MAL-ED study: A
2. K. Faust, J. Raes, Microbial interactions: From networks to
multinational and multidisciplinary approach to understand the
ability to utilize specific carbohydrates and/or models. Nat. Rev. Microbiol. 10, 538–550 (2012). doi: 10.1038/
relationship between enteric pathogens, malnutrition, gut
generate short-chain fatty acid products of fer- nrmicro2832; pmid: 22796884
physiology, physical growth, cognitive development, and
3. M. Layeghifard, D. M. Hwang, D. S. Guttman, Disentangling
mentation). Table S7 presents the supporting interactions in the microbiome: A network perspective.
immune responses in infants and children up to 2 years of age
evidence for assigning a given phenotype to an in resource-poor environments. Clin. Infect. Dis. 59,
Trends Microbiol. 25, 217–228 (2017). doi: 10.1016/
S193–S206 (2014). pmid: 25305287
organism. j.tim.2016.11.008; pmid: 27916383 21. J. L. Gehrig et al., Effects of microbiota-directed foods in
Colonization: Bacterial strains were cultured 4. A. R. Ives, B. Dennis, K. L. Cottingham, S. R. Carpenter, gnotobiotic animals and undernourished children. Science 365,
Estimating community stability and ecological interactions
under anaerobic conditions in pre-reduced from time-series data. Ecol. Monogr. 73, 301–330 (2003).
eaau4732 (2019).
Wilkins-Chalgren anaerobe broth (Oxoid Inc.) 22. E. Miller, D. Ullrey, The pig as a model for human nutrition.
doi: 10.1890/0012-9615(2003)073[0301:ECSAEI]2.0.CO;2 Annu. Rev. Nutr. 7, 361–382 (1987).
or MegaMedium (21, 33). Methods used for 5. D. R. Hekstra, S. Leibler, Contingency and statistical laws in 23. J. A. Draghi, T. L. Parsons, G. P. Wagner, J. B. Plotkin,
sequencing, assembling, and annotating bac- replicate microbial closed ecosystems. Cell 149, 1164–1173 Mutational robustness can facilitate adaptation. Nature 463,
(2012). doi: 10.1016/j.cell.2012.03.040; pmid: 22632978
terial genomes are described in Gehrig et al. 6. S. Weiss et al., Correlation detection strategies in microbial
353–355 (2010). doi: 10.1038/nature08694; pmid: 20090752
24. M. Kirschner, J. Gerhart, Evolvability. Proc. Natl. Acad.
(21). An equivalent mixture of each B. longum data sets vary widely in sensitivity and precision. ISME J. Sci. U.S.A. 95, 8420–8427 (1998). doi: 10.1073/
strain or additional ecogroup strain was prepared 10, 1669–1681 (2016). doi: 10.1038/ismej.2015.235; pnas.95.15.8420; pmid: 9671692
by adjusting the volumes of each culture based on pmid: 26905627 25. R. N. McLaughlin Jr., F. J. Poelwijk, A. Raman, W. S. Gosal,
7. K. Faust et al., Microbial co-occurrence relationships in the R. Ranganathan, The spatial architecture of protein function
optical density (OD600) readings. An equal volume human microbiome. PLOS Comput. Biol. 8, e1002606 (2012). and adaptation. Nature 491, 138–142 (2012). doi: 10.1038/
of pre-reduced PBS containing 30% glycerol was doi: 10.1371/journal.pcbi.1002606; pmid: 22807668 nature11500; pmid: 23041932
added to the mixture and aliquots were frozen 8. A. Zelezniak et al., Metabolic dependencies drive species 26. A. S. Raman, K. I. White, R. Ranganathan, Origins of allostery
and stored at –80°C until use. Each piglet re- co-occurrence in diverse microbial communities. Proc. Natl. and evolvability in proteins: A case study. Cell 166, 468–480
Acad. Sci. U.S.A. 112, 6449–6454 (2015). doi: 10.1073/ (2016). doi: 10.1016/j.cell.2016.05.047; pmid: 27321669
ceived an intragastric gavage (Kendall Kangaroo pnas.1421834112; pmid: 25941371 27. D. M. Gordon, The ecology of collective behavior. PLOS Biol.
2.7 mm diameter feeding tube; catalog number 9. J. Friedman, E. J. Alm, Inferring correlation networks from 12, e1001805 (2014). doi: 10.1371/journal.pbio.1001805;
8888260406, Covidien, Minneapolis, MN) of genomic survey data. PLOS Comput. Biol. 8, e1002687 (2012). pmid: 24618695
11 ml of a solution containing the bacterial con- doi: 10.1371/journal.pcbi.1002687; pmid: 23028285 28. B. J. Callahan et al., DADA2: High-resolution sample inference
10. Z. D. Kurtz et al., Sparse and compositionally robust inference from Illumina amplicon data. Nat. Methods 13, 581–583 (2016).
sortia listed in fig. S13A and Soweena (1:10 v/v). of microbial ecological networks. PLOS Comput. Biol. 11, doi: 10.1038/nmeth.3869; pmid: 27214047
The fecal microbiota was sampled using rectal e1004226 (2015). doi: 10.1371/journal.pcbi.1004226; 29. M. R. Charbonneau et al., Sialylated milk oligosaccharides
swabs on the days indicated in fig. S13A. pmid: 25950956 promote microbiota-dependent growth in models of infant

undernutrition. Cell 164, 859–871 (2016). doi: 10.1016/ collaboration and providing access to key clinical datasets; M. Meier, characterizing the role of diet-by-microbiota interactions in animal
j.cell.2016.01.024; pmid: 26898329 S. Deng, and J. Hoisington-López for superb technical assistance; health. W.A.P. serves as a consultant to TechLab Inc., a company that
30. T. Seemann, Prokka: Rapid prokaryotic genome annotation. D. O’Donnell, J. Serugo, and M. Talcott for their indispensable help makes diagnostic tests for enteric infections and has served as a
Bioinformatics 30, 2068–2069 (2014). doi: 10.1093/ with gnotobiotic piglet husbandry; and R. Olson for technical support consultant for Perrigo Nutritionals LLC, which produces infant
bioinformatics/btu153; pmid: 24642063 with the mcSEED-based genome analysis and subsystem curation. formula. Data and materials availability: Bacterial V4-16S rDNA
31. R. Overbeek et al., The SEED and the Rapid Annotation of Funding: Supported by the Bill & Melinda Gates Foundation as part of sequences in raw format (prior to postprocessing and data analysis),
microbial genomes using Subsystems Technology (RAST). the Breast Milk, Gut Microbiome, and Immunity (BMMI) Project. shotgun datasets generated from cultured bacterial strains, and
Nucleic Acids Res. 42, D206–D214 (2014). doi: 10.1093/nar/ The 5-year birth cohort study of Bangladeshi children was funded by COPRO-seq and microbial RNA-seq datasets obtained from
gkt1226; pmid: 24293654 NIH grant AI043596 (W.A.P.). A.S.R. is a postdoctoral fellow gnotobiotic piglets have been deposited at the European Nucleotide
32. R. Overbeek et al., The subsystems approach to genome supported by Washington University School of Medicine Physician Archive under study accession number PRJEB27068. Code has been
annotation and its use in the project to annotate 1000 genomes. Scientist Training Program and in part by NIH grant DK30292. D.A.R., archived at Zenodo (35). Fecal specimens from the MAL-ED birth
Nucleic Acids Res. 33, 5691–5702 (2005). doi: 10.1093/nar/ A.A.A., and S.A.L. were supported by Russian Science Foundation cohorts in Bangladesh (icddr,b, Dhaka), Brazil (Federal University of
gki866; pmid: 16214803 grant 19-14-00305. J.I.G. is the recipient of a Thought Leader award Ceará, Fortaleza), India (Christian Medical College, Vellore), Peru
33. A. L. Goodman et al., Extensive personal human gut from Agilent Technologies. Author contributions: R.H. and W.A.P. (JHSPH/AB PRISMA), South Africa (University of Venda), and from
microbiota culture collections characterized and designed and oversaw the 5-year birth cohort study; they, together the NIH birth cohort and SAM/MDCF studies at icddr,b, were provided
manipulated in gnotobiotic mice. Proc. Natl. Acad. with T.A., were responsible for coordinating various aspects of to Washington University under material transfer agreements.
Sci. U.S.A. 108, 6252–6257 (2011). doi: 10.1073/ biospecimen and metadata collection. S.H., M.M., R.H., W.A.P., and This work is licensed under a Creative Commons Attribution 4.0
pnas.1102938108; pmid: 21436049 T.A. (Bangladesh), M.N.K. (Peru), G.K. (India), P.O.B. (South Africa), and International (CC BY 4.0) license, which permits unrestricted use,
34. M. C. Hibberd et al., The effects of micronutrient deficiencies A.A.M.L. (Brazil) oversaw the MAL-ED studies. S.H., I.M., M.I., M.M., distribution, and reproduction in any medium, provided the original
on bacterial species from the human gut microbiota. and T.A. were responsible for studies involving the SAM and MAM work is properly cited. To view a copy of this license, visit http://
Sci. Transl. Med. 9, eaal4069 (2017). doi: 10.1126/ cohorts. J.L.G. and S.S. generated 16S rDNA datasets from human creativecommons.org/licenses/by/4.0/. This license does not apply
scitranslmed.aal4069; pmid: 28515336 fecal samples. M.J.B. managed the repository of biospecimens to figures/photos/artwork or other content included in the article
35. Github deposition of code; Zenodo doi: 10.5281/zenodo.3255003. and associated clinical metadata used for the studies described that is credited to a third party; obtain authorization from the rights
Also available for download at github.com/arjunsraman/ above. H.-W.C. performed the experiments with gnotobiotic piglets holder before using such material.
Raman_et_al_Science_2019. with the assistance of A.S.R. S.V., and M.C.H. D.A.R., A.A.A., S.A.L.,
and A.L.O. performed in silico metabolic reconstructions based on the
ACKN OW LEDG MEN TS SUPPLEMENTARY MATERIALS
genome sequences of bacterial strains introduced into gnotobiotic
We are indebted to the families of study subjects for their active piglets. A.S.R. conceived the mathematical approach and wrote all of science.sciencemag.org/content/365/6449/eaau4735/suppl/DC1
participation and assistance. We thank the staff and investigators at the computational workflow for identifying ecogroup taxa, performed Supplementary Text
icddr,b for their contributions to the recruitment and enrollment of the sensitivity analysis of the workflow, compared the SparCC and Figs. S1 to S16
participants in the 5-year Bangladeshi birth cohort study plus the SPIEC-EASI algorithms with the workflow, and undertook the analyses Tables S1 to S13
interventional studies of children with SAM and MAM, as well as the of gut microbial communities from subjects enrolled in the SAM, References (36–40)
collection of biospecimens and data. We also thank the study team MDCF, Peruvian, and Indian cohort studies as well as the gnotobiotic
members and health care workers involved in the MAL-ED birth piglet experiment, with J.L.G., S.V., M.J.B., and J.I.G. contributing in 13 June 2018; resubmitted 24 April 2019
cohort studies; M. Gottlieb, D. Lang, K. Tountas, and M. McGrath, who various supportive ways. A.S.R. and J.I.G. wrote the paper. Competing Accepted 7 June 2019
provided invaluable assistance in coordinating the MAL-ED interests: J.I.G. is a co-founder of Matatu Inc., a company 10.1126/science.aau4735

ARTIFICIAL INTELLIGENCE AUTONOMOUS VEHICLES BIOMATERIALS IN ROBOTICS HUMANOIDS LAND & UNDERSEA ROBOTS MEDICAL & SURGICAL ROBOTS
MICRO/NANO ROBOTS ROBOT ENGINEERING ROBOTS IN EDUCATION SPACE ROBOTS THEORETICAL ADVANCES WITH POSSIBLE APPLICATIONS
Transforming the Future

of Robotics
As a multidisciplinary online-only journal, Science Robotics publishes original, peer-reviewed,

research articles that advance the field of robotics. The journal provides a central forum for
communication of new ideas, general principles, and original developments in research and
applications of robotics for all environments.
Learn more at: ScienceRobotics.org
R ES E A RC H
◥ 51% of the cells encoding one of the three spa-

RESEARCH ARTICLE SUMMARY tial variables showing conjunctive tuning to at
least one other variable. The egocentric neuron
types identified in POR were largely absent in
NEUROSCIENCE
neighboring MEC and parasubiculum, areas
implicated in allocentric spatial processing.
A sense of space in postrhinal cortex Unlike neurons in these areas, POR neurons
rarely showed modulation of their spike trains
by theta rhythm (5 to 11 Hz). Spiking proper-
Patrick A. LaChance, Travis P. Todd, Jeffrey S. Taube* ties of POR cells were sufficient to decode an
animal’s allocentric position. All three spatial
INTRODUCTION: Navigation and spatial with brain areas thought to be involved in both ◥
cell types continued pro-
ON OUR WEBSITE cessing elements of space
learning depend upon a topographic repre- egocentric and allocentric spatial processing.
sentation of local space, which is defined with It has been suggested that cells encoding the Read the full article in the presence of objects
respect to the world (allocentric) and as op- egocentric bearing and distance of the geo- at http://dx.doi. and in the dark. Decreas-
posed to the observer (egocentric). Spatial metric center of the local environment, as org/10.1126/ ing the size of the arena
processing in the rodent brain has been prim- well as the animal’s allocentric head direction, science.aax4192 did not affect the tuning
..................................................
arily explored through recordings of neurons could provide a readout of allocentric self- slopes of center-distance
thought to encode elements of an allocentric position. Such a signal could drive naviga- cells. Rotation of the arena led to a corre-
spatial map, such as place cells in the hippo- tional behavior and support the high-level sponding shift in the preferred firing direc-
campus and grid cells in the medial entorhi- spatial maps exemplified by entorhinal grid tions of head direction cells without altering
nal cortex (MEC). Such a map, however, must cells and hippocampal place cells. the tuning of egocentric center-tuned cells.
be constructed from sensory information that
RESULTS: We recorded from 338 single neu- CONCLUSION: Our results reveal a popula-
is initially coded egocentrically. Theoretical
models have suggested mechanisms by which rons in POR (11 rats) as animals freely foraged tion code for allocentric space in POR that is
egocentric information can be transformed for sugar pellets in a 1.2-m square arena. more strongly tuned to the spatial layout of a
into an allocentric spatial reference frame, Thirty-nine percent of the cells encoded the local scene than to the contents of that scene.
though the precise neural mechanisms under- egocentric bearing of the center of the arena The cells supporting this code signal the in-
lying this integration are not well understood. (“center-bearing”). The tuning preferences of stantaneous egocentric bearing and distance
center-bearing cells were consistent across of the geometric center of the local environ-
RATIONALE: A potential locus for the inte- space and time. Seventeen percent of POR ment, as well as allocentric head direction.
gration of egocentric and allocentric spatial cells showed tuning to the animal’s distance Center-distance cells respond linearly to ab-
information in the rat brain is the postrhinal from the center of the arena (“center-distance”), solute distance from the environment center,
cortex (POR). The POR is the rodent homolog with both positive and negative linear responses whereas each center-bearing cell shows pre-
of the human parahippocampal cortex, which to center-distance observed. Thirty-eight per- ferential firing to a single egocentric bearing
is thought to be involved in topographic spa- cent of POR cells were tuned to the rat’s head of the apparatus center that remains constant
tial learning and processing of local spatial direction in an allocentric reference frame. across environmental manipulations. Responses
cues. The POR is also extensively connected Many POR cells were conjunctively tuned, with of these cells are tied to local cues and main-
tain their tuning properties in darkness. Ego-
centric encoding of the local environment’s
geometric center, as opposed to the encod-
ing of its boundaries or to discrete physical
cues within that environment, could allow
for the use of a common spatial reference frame
across environments with disparate geome-
tries. POR projects strongly to MEC, where
Center-bearing cells Head direction cells Center-distance cells grid cells are found. The egocentric and allo-
centric correlates identified in POR might help
to create or support that spatial metric. POR
Center bearing also projects to lateral entorhinal cortex, where
egocentric correlates and object signals have
been reported. The functional cell types inves-
tigated in POR may provide a foundation for
Head direction
spatial processing in both entorhinal subdivi-
sions. Representations from each area can then
Center distance be routed to the hippocampus and efficiently
integrated for high-level spatial processing and
2D position code
encoding of episodic memory.
▪
Spatial coding in postrhinal cortex. Neurons in the rat postrhinal cortex encode the egocentric
bearing and distance of the geometric center of the local environment during free foraging, Department of Psychological and Brain Sciences, Dartmouth
as well as the animal’s head direction in allocentric coordinates. Combining these firing correlates College, Hanover, NH, USA.
*Corresponding author. Email: jeffrey.s.taube@dartmouth.edu
reveals a code for representing two-dimensional allocentric space that could serve as a spatial Cite this article as P. A. LaChance et al., Science 365,
template for the neuronal activity in the downstream entorhinal cortex and hippocampus. eaax4192(2019). DOI: 10.1126/science.aax4192

R ES E A RC H
◥ regions. There was no difference in the propor-

RESEARCH ARTICLE tion of center-bearing cells between dorsal and
ventral subdivisions of POR [c2(1) = 1.54, P =
0.21]. The preferred bearings of these cells cov-
NEUROSCIENCE ered the full spectrum of angles but were ap-
proximately distributed bimodally, with peaks
A sense of space in postrhinal cortex at 30° (to the front left the animal) and 224° (to
the back right of the animal) (39) (Fig. 1D). The
tuning curves of center-bearing cells tended to be
Patrick A. LaChance, Travis P. Todd, Jeffrey S. Taube* broad (median MVL: 0.30, 95% confidence in-
terval (CI): [0.29, 0.33]) and appeared sinusoidal,
A topographic representation of local space is critical for navigation and spatial memory. In with all classified cells (132 out of 132) passing
humans, topographic spatial learning relies upon the parahippocampal cortex, damage to which a shuffle threshold for fit by a cosine function
renders patients unable to navigate their surroundings or develop new spatial representations. (median fit coefficient of determination, R2: 0.86;
Stable spatial signals have not yet been observed in its rat homolog, the postrhinal cortex. We fig. S4). Center-bearing cells showed no difference
recorded from single neurons in the rat postrhinal cortex whose firing reflects an animal’s in preferred bearing between the inner and outer
egocentric relationship to the geometric center of the local environment, as well as the animal’s segments of the arena (Rayleigh test, r = 0.86,
head direction in an allocentric reference frame. Combining these firing correlates revealed a P = 0; median shift: −1.18°; Wilcoxon signed rank
population code for a stable topographic map of local space.This may form the basis for higher- test, Z = 0.51, P = 0.70; fig. S5), eliminating wall-
order spatial maps such as those seen in the hippocampus and entorhinal cortex. hugging as a potential confounding variable.
Center-bearing tuning was largely absent in both
N
neighboring medial entorhinal cortex (MEC) (sup-
avigation and the encoding of spatial these measures with a sense of allocentric head erficial layers) and parasubiculum (PaS), with only
memory require an accurate sense of direction, determined by the principal axis of the 3% (9 out of 278) and 5% (7 out of 137) of cells
one’s location within the environment. local environment, would provide the necessary classified as encoding center bearing in these
Humans and animals form a topographic elements for constructing a map of allocentric regions, respectively (fig. S13 and table S2).
representation of space defined by local space as well as a set of tools for vector compu- To establish a vector representation of allo-
features and geometry (1–12). This representa- tations supporting navigation (13–15, 19, 20). This centric space, signals encoding the egocentric
tion is considered allocentric (world-centered), coding scheme would allow for representation of bearing of a reference point (i.e., environment
meaning it is independent of the observer’s local space using a common coordinate system center) should be accompanied by signals repre-
perspective. However, its formation and updating across environments with disparate geometries. senting egocentric distance of that reference
are accomplished via incoming sensory informa- point. Consistent with this framework, 59 of the
tion, which is egocentric (observer-centered). Ego- and allocentric spatial 338 POR cells (17%) showed tuning to center dis-
Thus, topographic spatial representation requires correlates in POR tance (median linear R2: 0.72, 95% CI: [0.70, 0.75])
the transformation of egocentric information We recorded the activity of 338 putative princi- (38) (Fig. 1E and figs. S3 and S9). The proportion
into an allocentric framework (13–20). pal neurons (N = 11 rats) in POR (Fig. 1A, fig. S1, of center-distance cells differed across layers
In humans, topographic spatial learning is and table S1) as rats foraged for randomly scat- [layer II (N = 38): 5%; layer III (N = 88): 7%;
thought to rely upon the parahippocampal cor- tered sucrose pellets in a 1.2-m square arena. We layer V (N = 120): 30%; layer VI (N = 92): 16%;
tex (PHC) (21), and damage to this area results performed an exhaustive search for potential ego- c2(3) = 24.02, P = 2.47e-05], such that deep layers
in topographical disorientation (22). Included centric bearing reference points throughout the had more distance-tuned cells than superficial
in this region is the parahippocampal place area local environment by plotting each cell’s firing layers [c2(1) = 15.99, P = 6.37e-05] and layer V
(PPA), which is activated by images of scenes rate against the instantaneous egocentric bear- had the most of all [c2(1) = 18.99, P = 1.31e-05; fig.
regardless of their contents (23) and is thought ings (sorted into 12° bins) of a 20 × 20 grid of S3]. There was no difference between dorsal and
to encode spatial layouts (24). The rodent homo- locations evenly spaced throughout the arena. ventral POR [c2(1) = 0.25, P = 0.61]. Both posi-
log of the PHC, postrhinal cortex (POR), is strongly Many POR cells showed strong tuning to the tive (37 out of 59 cells; median positive slope:
interconnected with brain regions thought to geometric center of the environment, indicated 0.038 Hz/cm, 95% CI: [0.32, 0.59]) and nega-
process elements of egocentric (15–18, 25) as well by high mean vector lengths (MVLs; fig. S2). For tive (22 out of 59 cells; median negative slope:
as allocentric space (4–7, 26–29). Reports of ego- example, a cell might fire whenever the left side −0.052 Hz/cm, 95% CI: [−0.082, −0.032]) tuning
centric encoding in areas connected with POR, of the rat’s head faced the apparatus center (Fig. slopes were observed, such that cells had their
including the lateral entorhinal cortex (LEC) (30), 1B). We subsequently classified POR cells (36, 37) peak firing rates at the corners or center of the
hippocampus (31), posterior parietal cortex (PPC) as encoding any of three navigational variables: environment, respectively. As with center-bearing
(32), postsubiculum (33), and dorsal striatum egocentric bearing of the environment center cells, and similar to POR layer II, center-distance
(34), support a potential role for POR as a hub of (“center bearing”), egocentric distance of the en- cells were largely absent in MEC and PaS, ac-
egocentric processing. POR is therefore well po- vironment center (“center distance”), and allo- counting for only 5% (14 out of 278) and 7% (9 out
sitioned to contribute to the integration of egocentric head direction (Fig. 1B). of 137) of cells, respectively (fig. S13 and table S2).
centric spatial information into an allocentric Of the 338 POR cells, 132 (39%) were classified In addition to representing the egocentric
spatial map (35). as encoding center bearing (38) (Fig. 1C and bearing and distance of a reference point, a vec-
This integration might be carried out by single figs. S3 and S8). Percentages differed by cell layer tor estimate of allocentric position requires a
neurons encoding egocentric bearing (i.e., the [layer II (N = 38): 5%; layer III (N = 88): 30%; sense of allocentric head direction (13–15, 19, 20).
angle of a reference point from an animal’s head- layer V (N = 120): 53%; layer VI (N = 92): 45%; Of the 338 recorded POR cells, 128 (38%) were
ing) and distance (i.e., the distance of a reference c2(3) = 31.86, P = 5.60e-07], such that center- classified as encoding allocentric head direction
point from an animal’s position) of the geometric bearing cells were more prevalent in deep than (12) (Fig. 1F and figs. S3 and S10). Here, there
center of the local environment (13). Combining in superficial layers [c2(1) = 22.80, P = 1.80e-06] were also significant differences across layers
and least prevalent in layer II [c2(1) = 18.97, P = [layer II (N = 38): 5%; layer III (N = 88): 30%;
Department of Psychological and Brain Sciences, Dartmouth
1.33e-05], where they were largely absent (fig. layer V (N = 120): 28%; layer VI (N = 92): 72%;
College, Hanover, NH, USA S3). Layers I and IV were not investigated here c2(3) = 69.256, P = 6.16e-15], with layer VI having
*Corresponding author. Email: jeffrey.s.taube@dartmouth.edu and below due to the lack of sampling from those the largest proportion of head direction (HD)
LaChance et al., Science 365, eaax4192 (2019) 12 July 2019 1 of 10

Fig. 1. Egocentric and allocentric spatial correlates in POR. (A) An animal’s path, dots show spike locations colored by head direction; color
example Nissl-stained sagittal section from one rat (PL61) showing the bar below) and center-bearing tuning curves for two example cells that
boundaries of POR. Black arrow indicates the border between dorsal and encode egocentric bearing of the environment center. (D) Histogram
ventral subdivisions of POR. (B) Schematic top-down view of a rat in of the preferred bearings of all center-bearing cells recorded in POR.
the recording arena illustrating three measures that together specify a Dashed red line shows a bimodal von mises mixture fit to the histogram,
single allocentric location in the environment. The dashed line extending with peaks at 30° and 224°. (E) Directional spike plots and center-
rightward from the rat indicates the reference axis for allocentric head distance curves for two example center-distance cells recorded in POR.
direction tuning; HD measurements increase with counterclockwise (F) Directional spike plots and head direction curves for two example head
rotation of the rat, such that the rat has an allocentric head direction of direction cells recorded in POR. (G) A directional spike plot and three
0° when it is facing “east” (in line with the reference axis) and 90° when it tuning curves (center bearing in blue, head direction in red, and center
is facing “north.” Center bearing is calculated as the eccentricity of the distance in green) for a POR cell tuned conjunctively to head direction and
environment center from the rat’s heading. The star indicates the location center bearing with a moderate linear response to center distance. Color
of the environment center. (C) Directional spike plots (gray trace shows bar indicates head direction for directional spike plots.
cells [c2(1) = 59.67, P = 1.12e-14] and layer II and 94% of POR HD cells (120 out of 128) pas- whereas 36% were conjunctively tuned to center
having the smallest [c2(1) = 17.82, P = 2.43e-05; sed a shuffle threshold for fit by a cosine func- distance. Center-bearing tuning was not an arti-
fig. S3]. There was no difference between dorsal tion (fig. S4). HD cell peak firing rates (mean: fact of conjunctive HD by two-dimensional (2D)
and ventral POR [c2(1) = 9.34e-4, P = 0.98]. MEC 7.01 Hz) were considerably lower than peak rates position tuning, as 128 of 132 originally classified
and PaS also contained sizable proportions of seen in other subcortical areas that contain HD center-bearing cells (97%) were still classified as
HD cells (12% and 35%, of cells, respectively; fig. cells [e.g., anterodorsal thalamus, mean: 41.1 Hz center-bearing cells after correcting for 2D po-
S13 and table S2). However, POR HD cells tended (40) and lateral mammillary nuclei, mean: 69.5 sition encoding (43). Center-distance cells were
to show broad tuning curves (median MVL: Hz (41)], but comparable to those seen in MEC especially conjunctive, with 81% conjunctively
0.33, 95% CI: [0.28, 0.37]; see table S3 for other (5.41 Hz) (42). tuned to center bearing and 44% tuned to head
directional properties) that were significantly POR cells tended to show conjunctive tuning direction. Of POR HD cells, 57% showed con-
more sinusoidal than those in MEC/PaS (POR to multiple behavioral variables (Fig. 1G and figs. junctive tuning to center bearing, whereas 20%
median fit R2: 0.75; MEC/PaS median fit R2: 0.70; S3, S9, S11, and S12). Of center-bearing cells, 55% were conjunctively tuned to center distance. Over-
Wilcoxon rank-sum test, Z = 2.49, P = 0.0063), showed conjunctive tuning to head direction, all, just over half (51%) of POR cells classified as

encoding at least one of these three variables and PaS cells showed worse performance ac- tween the sessions, indicating that the response
(99 out of 196) showed conjunctive tuning to at ross all measurements using both PV (MEC/ gain across center-distance cells did not change
least one other variable. Only 5 of 338 POR PaS median x fit R2: 0.56; Wilcoxon rank-sum (Wilcoxon signed rank test, Z = −1.32, P = 0.092;
cells (1%) showed modulation by theta rhythm test, Z = 12.22, P = 2.52e-34; MEC/PaS median Fig. 3, B and E).
(5 to 11 Hz), whereas 48% of MEC cells and 47% y fit R2: 0.55; Wilcoxon rank-sum test, Z = 12.22,
of PaS cells showed theta modulation. P = 2.52e-34; MEC/PaS median absolute error: Center-distance cells in POR encode
30 cm; Wilcoxon rank-sum test, Z = −343.24, absolute distances
POR cells code for allocentric location P = 0) and Bayesian methods (MEC/PaS median We next sought to determine the impact of the
Egocentric bearing and distance information x fit R2: 0.62; Wilcoxon rank-sum test, Z = 12.22, size of the local environment on the slopes of
might be combined with a sense of allocentric P = 2.52e-34; MEC/PaS median y fit R2: 0.56; distance response curves for center-distance
head direction to encode allocentric location Wilcoxon rank-sum test, Z = 12.22, P = 2.52e-34; cells. If movement to a smaller box caused an
(13–15, 19, 20). We employed two decoding strat- MEC/PaS median absolute error: 21 cm; Wilcoxon increase in the slope of distance tuning, it would
egies to determine if the firing properties of rank-sum test, Z = −344.90, P = 0). suggest that center-distance cells encode relative
POR cells could provide an accurate readout of distance proportional to the size of the local en-
an animal’s allocentric location: a simple popu- Center-bearing and center-distance cells vironment. If the slope remained the same, how-
lation vector (PV) decoding method (44, 45), in POR do not encode objects ever, it would suggest that cells encoded absolute
which is best suited to cells that encode angles If the cells recorded in POR were purely encod- distance. This latter finding would allow them
with sinusoidal tuning curves and distances with ing the spatial layout of the local environment, to support a universal distance mechanism, such
linear curves (44–46), and a two-step Bayesian they should encode elements of local space re- as that hypothesized to be encoded by down-
decoding algorithm (47, 48). Using spike counts gardless of its contents (23). We therefore re- stream entorhinal grid cells (49). We recorded
from a selection of separately recorded POR corded the activity of 35 center-bearing cells, from 22 center-distance cells as animals foraged
cells (30, 43, 45), PV and Bayesian methods were 12 center-distance cells, and 36 HD cells in POR in both a 1.2-m square enclosure and a visually
effective in decoding center bearing (PV median as animals foraged in a 1.2-m box with and similar 1-m square enclosure (Fig. 4A). Distance
R2: 0.94; Bayesian median R2: 0.95), center dis- without objects (Fig. 3A). Center-bearing cells tuning slopes were strongly correlated across the
tance (PV median R2: 0.79; Bayesian median R2: maintained their center-bearing preferences ac- two environments [Large1-Small, Pearson pro-
0.88), and head direction (PV median R2: 0.90; ross the two sessions, with mean angles show- duct moment correlation, t(20) = 5.55, P = 1.97e-
Bayesian median R2: 0.95). The POR decoders ing a nonuniform angular shift (Std1-Objects, 05, r = 0.78] and showed no significant difference
outperformed decoders based on pooled MEC Rayleigh test of uniformity, r = 0.97, P = 0) that in absolute slopes (Wilcoxon signed rank test,
and PaS cells across all measurements (fig. S7). was not different from zero (median shift: −2.72°; Z = 0.53, P = 0.70; Fig. 4, B and E). We also re-
The allocentric location correlates derived from Wilcoxon signed rank test, Z = −1.51, P = 0.065; corded from 49 center-bearing cells across these
these decoded values showed strong fits to the Fig. 3, C and F). HD cells also maintained their two sessions, which showed no difference in pre-
real locational data (Fig. 2A), with the PV meth- preferred directions across sessions (Std1-Objects, ferred center bearing (Large1-Small, Rayleigh test,
od explaining 83% of the variance for x location Rayleigh test, r = 0.96, P = 0; median shift: 1.61°; r = 0.93, P = 0; median shift: 0.70°; Wilcoxon
and 79% for y location, and the Bayesian algo- Wilcoxon signed rank test, Z = 0.21, P = 0.58; Fig. signed rank test, Z = 1.11, P = 0.87; Fig. 4, C and
rithm explaining 92% and 89% of the variance 3, D and G). The tuning slopes of center-distance F), as well as 40 HD cells, which also maintained
for x and y locations, respectively (Fig. 2B). The cells were highly correlated across standard and their preferred directions (Large1-Small, Rayleigh
median Euclidean decoding error was 18 cm for object sessions [Std1-Objects, Pearson product mo- test, r = 0.94, P = 0; median shift: 1.12°; Wilcoxon
the PV method and 11 cm for the Bayesian algo- ment correlation, t(10) = 6.99, P = 3.74e-05, r = signed rank test, Z = −0.53, P = 0.29; Fig. 4, D
rithm (Fig. 2C). Decoding based on pooled MEC 0.91] with no difference in absolute slopes be- and G).
Fig. 2. Coding for allocentric location by cells in POR. (A) Results from an example decoding iteration showing fit of two decoding algorithms
[population vector (PV) and two-step Bayesian] to the real x- and y-positions of an animal over a 3-min period. (B) Box plots showing variance in the
x- and y-positions of the animal explained by each decoder based on cells drawn from each region (POR and MEC/PaS). Center line indicates median, box
limits indicate upper and lower quartiles, and whiskers indicate 1.5× interquartile range. (C) Histograms showing counts of absolute error in decoded
location across all iterations of each decoder (left: population vector; right: Bayesian) for cells drawn from POR and MEC/PaS.

Fig. 3. POR spatial cell types in the presence

of objects. (A) Directional spike plots for
an example center-bearing by center-distance
cell showing tuning stability between standard
and object sessions. Color bar indicates head
direction. (B) Center-distance tuning curves
for an example distance-tuned cell showing
stability between standard and object sessions.
(C) Center-bearing tuning curves for an example
bearing-tuned cell showing stability across
sessions. (D) Head direction tuning curves for
an example HD cell showing stability across
sessions. (E) Scatter plot showing firing
rate slopes of center-distance tuning curves
between Standard 1 and Object sessions for all
recorded POR center distance cells. (F) Polar
plot showing shift in preferred center bearing
between Standard 1 and Object sessions for all
recorded POR center-bearing cells (each dot
represents one cell). (G) Polar plot showing shift
in preferred firing direction between Standard 1
and Object sessions for all recorded POR HD
cells (each dot represents one cell).
Spatial cell firing in POR is tied to visual cues. We recorded from 33 center-bearing The egocentric correlates found in POR could
local cues cells, 12 center-distance cells, and 31 HD cells as result from interactions with visuospatial areas
We next tested whether rotation of the local en- rats foraged in a 1-m square enclosure in both such as retrosplenial cortex (RSC) or PPC (25),
vironment caused a change in the spatial tuning light and darkness (Fig. 6A). Center-bearing cells where egocentric spatial correlates have been
of POR cells. Although we hypothesized that showed no difference in preferred direction be- reported (32, 50). Projections from both RSC
center-bearing and center-distance tuning would tween light and dark sessions (Std1-Dark, Rayleigh and PPC preferentially target deep layers of POR
not change, HD cells should exhibit a rotation test, r = 0.80, P = 0; median shift: 5.97°; Wilcoxon (51, 52), where the largest proportions of center-
of their preferred firing direction if their fir- signed rank test, Z = 0.86, P = 0.81; Fig. 6, C and bearing and center-distance cells were found
ing is truly tied to the layout or features of the F), nor did HD cells (Std1-Dark, Rayleigh test, r = (fig. S3). Neurons that resemble center-bearing
local environment. We recorded from 38 center- 0.83, P = 0; median shift: 1.82°; Wilcoxon signed cells have been reported in RSC (53) (“direction-
bearing cells, 15 center-distance cells, and 33 HD rank test, Z = −0.39, P = 0.35; Fig. 6, D and G). In dependent place cells”), so it is possible that POR
cells as animals foraged in a 1-m square box and addition, center-distance cells maintained their and RSC are functionally coupled in their pro-
the same box that had been rotated by 45° (Fig. distance tuning slopes [Std1-Dark, Pearson prod- cessing of local space. Return projections to PPC
5A). There was no change in the preferred bear- uct moment correlation, t(10) = 2.87, P = 0.017, originate primarily in deep layers of POR (52).
ings of center-bearing cells (Std1-Rotated, Rayleigh r = 0.67; Wilcoxon signed rank test on absolute Deep layers of POR also project strongly to the
test, r = 0.93, P = 0; median shift: −0.51°; Wilcoxon slopes, Z = 0.62, P = 0.73; Fig. 6, B and E]. dorsal striatum (54, 55), which has been impli-
signed rank test, Z = 0.11, P = 0.55; Fig. 5, C cated in egocentric spatial processing (34, 56).
and F) or the tuning slopes of center-distance Discussion Head direction signals in POR might arise from
cells [Std1-Rotated, Pearson product moment Single neurons in POR encode both the ego- afferents from several areas, including RSC
correlation, t(13) = 3.13, P = 0.0079, r = 0.66; centric bearing and distance of the geometric (25, 53); PPC (25, 32); postsubiculum (12, 27); or
Wilcoxon signed rank test on absolute slopes, center of the local environment, as well as al- the anterodorsal, lateral dorsal, and reuniens
Z = 0.58, P = 0.72; Fig. 5, B and E]. By contrast, locentric head direction. Moreover, they repre- thalamus (40, 57–59).
HD cells showed a significant shift in the direc- sent these spatial correlates with tuning curves A recent report of egocentric encoding in LEC
tion of environmental rotation (Std1-Rotated, ideally suited for a population code (44–46) from reported a subset of cells that were responsive to
median shift: 40°; Rayleigh test, r = 0.93, P = 0; which allocentric self-location information can external items within an egocentric reference
Wilcoxon signed rank test, Z = 6.23, P = 2.33e-10; be decoded (Fig. 2) (13–15, 19, 20). The tuning frame, including objects and salient locations
Fig. 5, D and G). properties of these cells were tied to local cues (30). The POR functional cell types investigated
and remained stable in response to environ- here encoded spatial elements of the local envi-
Center-bearing and center-distance cells mental manipulations, including darkness. They ronment with and without objects, with no prim-
in POR maintain their tuning properties continued to encode aspects of spatial layout ing of salient locations, suggesting that they are
in darkness even in the presence of objects, suggesting that preferentially involved in processing the spatial
We finally determined whether POR cells can they are specifically tuned to represent local layout of a scene. Another subset of LEC cells
maintain their tuning properties in darkness. As space. Thus, POR contains all the elements had egocentric spatial properties reminiscent
POR receives heavy visual input (25), its spatial for constructing a stable allocentric map of local of POR center-tuning (“bearingboundary” cells),
tuning properties might rely on the presence of space. including during object presentation. Although

Fig. 4. POR center-distance cells encode

absolute distances. (A) Directional spike plots
for an example center-bearing by center-
distance cell showing tuning stability across
the large box – small box – large box cycle. Color
bar indicates head direction. (B) Center-distance
tuning curves for an example distance-tuned
cell showing highly similar tuning slopes across
sessions. (C) Center-bearing tuning curves for
an example center-bearing cell showing stability
across sessions. (D) Head direction tuning
curves for an example HD cell showing stability
across sessions. The curves in (C) and (D)
are taken from one conjunctively tuned cell.
(E) Scatter plot showing firing rate slopes for
center-distance tuning curves between Large
1 and Small sessions for all recorded POR
center-distance cells. (F) Polar plot showing shift
in preferred center bearing between Large 1 and
Small sessions for all recorded POR center-
bearing cells (each dot represents one cell).
(G) Polar plot showing shift in preferred firing
direction between Large 1 and Small sessions
for all recorded POR HD cells (each dot
represents one cell).
POR primarily targets MEC, it also projects to MEC does not contain a large amount of center- Formation and updating of a spatial map re-
LEC (28) and could be the source of the ego- bearing or center-distance tuning (fig. S13 and quire alignment of one’s current perception of
centric signals found there. Egocentric spatial table S2) implies that inputs from POR are heav- the environment with a stored representation
representations originating in POR could be ily processed within MEC. (20). One possible mechanism for this alignment
used by LEC to place external objects and salient Spatial processing within POR has been sug- could be to match the perceived boundaries of
locations into a common spatial reference frame gested (35), though studies of single-neuron ac- the environment with a stored representation
that is subsequently routed to the hippocampus, tivity in POR have revealed cells that either of those boundaries, which is a fundamental
and could be a unifying factor in the “what” and change their locational correlates between record- process in some computational models of navi-
“where” pathways thought to be represented in ing sessions (61); encode conjunctions of objects, gation (17, 18). However, this process quickly be-
LEC and MEC processing streams, respectively locations, and rewards (62); or do not show loca- comes computationally cumbersome, as one
(35). Still, the combination of center-tuned cells tional correlates at all (63). Discrepancies between would need to establish a separate representa-
and colocalized (or conjunctive) HD cells in POR our study and the former two studies, as well as tion for every boundary configuration encount-
make it especially fit for population coding of the fact that the second study frequently ob- ered in every new environment. A more efficient
allocentric space. served theta modulation among POR cells (62) method consists of aligning the centroid and
POR is a major source of cortical input to MEC whereas we did not, could be due to our more principal axis of the represented and perceived
(28), and projections from POR synapse directly caudal and medial recording location immedi- environments (20). In the case of POR, the cen-
onto principal cells in superficial layers of MEC ately dorsal to caudal MEC, where grid cells are troid of the environment can be represented by
(29), where grid cells are most abundant (5, 26). primarily found (5, 26). Cells in POR project to center-bearing and center-distance cells, while
These projections originate primarily from cells cells in MEC that are almost directly ventral to the principal axis can be signaled by HD cells
in layers II/III and V of POR (28), with an em- them (29); thus, it is likely that some of the cells (13). Thus, POR may provide a computationally
phasis on superficial layers (29). Deep-layer ac- that we recorded project to the region of MEC efficient template for mapping space in dispar-
tivity is also conveyed to MEC through intrinsic that contains grid cells. The absence of theta ate environments. This template may serve as a
POR connections between deep and superficial modulation among our recorded POR cells could foundation for mapping local space in the ap-
layers (29). Return projections from MEC, though reflect a need for nonrhythmic signals to be propriate context, and it could provide a found-
less strong, preferentially target layer VI of POR compared with a downstream theta reference ation for the higher-level spatial maps observed
(28). Grid cells have been suggested to provide a signal for the encoding of self-location (13). in the hippocampus and entorhinal cortex.
distance metric for the spatial navigation system The activity observed in POR parallels the ac-
(49). Inputs from POR cells could provide sup- tivity of the homologous human PPA, which Materials and methods
port for vector computations that create or rein- shows preferential activation in response to For additional information on the classification,
force such a metric (13–15, 19, 20). That POR images of scenes, regardless of their contents artifact correction, decoding, and control proce-
center-distance cells appear to encode absolute (23, 24), as well as the human PHC in general. dures, please see the supplementary materials.
instead of relative distance supports this notion The cells investigated in this study could provide
(Fig. 4). It has been hypothesized that grid cells insights into the processing that takes place in Subjects
maintain a reference to the geometric center of these regions, as well as the severe spatial deficits Subjects were 11 female Long-Evans rats weigh-
the local environment (60). Our observation that that occur when the PHC is damaged (22). ing 220 to 470 g (median 277 g) at the start of

Fig. 5. POR spatial cell types during cue

rotations. (A) Directional spike plots for an
example POR HD cell showing a shift in preferred
firing direction across standard and rotated
sessions. Color bar indicates head direction.
(B) Center-distance tuning curves for an exam-
ple center-distance tuned cell showing similar
tuning slopes across sessions. (C) Center-
bearing tuning curves for an example center-
bearing tuned cell showing stability across
sessions. (D) Head direction tuning curves for an
example HD cell showing a shift in the direction
of local environmental rotation during the rota-
tion session. (E) Scatter plot showing firing
rate slopes for center-distance tuning curves
between Standard 1 and Rotated sessions for all
recorded POR center-distance cells. (F) Polar
plot showing shift in preferred center bearing
between Standard 1 and Rotated sessions for all
recorded POR center-bearing cells (each dot
represents one cell). (G) Polar plot showing shift
in preferred firing direction between Standard
1 and Rotated sessions for all recorded POR HD
cells (each dot represents one cell).
testing. Rats were individually housed in Plexi- surface. Electrodes targeting superficial layers of crossed a predefined amplitude threshold (30 to
glas cages and maintained on a 12 hours light/ the postrhinal cortex were positioned closer to 50 mV), they were time-stamped and digitized at
dark cycle. Prior to surgery, food and water were the transverse sinus than those targeting deep 32 kHz for 1 ms. The headstage was also equip-
provided ad libitum. All experimental procedures layers. The electrodes were also angled 10° for- ped with red and green light-emitting diodes
involving the rats were performed in compliance ward in the sagittal plane, such that the tetrodes (LEDs) spaced ~6 cm apart over the head and
with institutional standards as set forth by the were pointing anteriorly. All electrodes were sec- back of the animal, respectively. A color video
National Institutes of Health Guide for the Care ured to the skull using dental acrylic. camera positioned over the arena captured video
and Use of Laboratory Animals and approved frames with a sampling rate of 30 Hz, and an
by the Dartmouth Institutional Animal Care and Recovery and behavioral training automated video tracker extracted the x- and
Use Committee. Rats were allowed 7 days to recover from surgery, y-positions of the LEDs as well as their angle in
after which they were placed on food restriction an allocentric frame. The tracking frames were
Electrode construction such that their body weight reached 85-90% of timestamped so they could be matched up to
All animals were implanted with a movable mi- its pre-surgical level. During this time, the rats the neural data. If clearly isolated waveforms
crodrive consisting of a bundle of four tetrodes were also trained to forage for randomly scat- were visually apparent, a 20-min baseline record-
targeting the postrhinal cortex. The tetrodes were tered sucrose pellets within a gray square box ing session in the 1.2-m square box took place.
constructed by twisting together four strands (120 cm × 120 cm; 50 cm in height) surrounded Otherwise, tetrodes were advanced ~50 to 100 mm
of 17-mm nichrome wire. These twisted strands by a uniform black curtain. The box itself was and screened again at least 2 hours later or the
were subsequently threaded through a single featureless except for a white cardboard sheet next day. In a few cases, recordings were cut a
26-gauge stainless steel cannula, and the end of placed along one inside wall. The floor was com- few minutes short due to technical difficulties;
each wire was connected to a single pin of a posed of gray photographic backdrop paper. Re- these sessions were only included in analyses if
Mill-Max connector. The two center pins of the cording began when the animals’ walking paths the animal showed full sampling (>80%) of the
connector were attached to the cannula, which showed uniform coverage (>80%) of the entire environment.
acted as an animal ground. Three drive screws arena.
were secured around the connector using dental Spike sorting
acrylic, making the electrode driveable in the Recording of neural data Spike sorting was conducted offline. Spikes col-
dorsal-ventral plane. Over the course of weeks to months, tetrodes lected from a recording session were first auto-
were “screened” for cells as the animals foraged matically sorted into clusters using the automated
Electrode implantation for sucrose pellets in the open arena. Electrical clustering program Kilosort (64), after which
Animals were anesthetized with isoflurane. They signals were pre-amplified using unity-gain ope- manual cleanup was performed using the man-
were subsequently placed in a stereotaxic frame, rational amplifiers on an HS-18-MM headstage. ual clustering program SpikeSort3D (Neuralynx).
and an incision was made in the scalp to expose Signals from each tetrode wire were then dif- For the manual step, waveform features includ-
the skull. A single craniotomy was drilled and the ferentially referenced against a quiet channel ing peak, valley, height, width, and principal
electrode was implanted 0.25 to 0.45 mm ante- from a separate tetrode and bandpass filtered components were used to visualize the character-
rior to the transverse sinus, 4.2 to 4.6 mm lateral (600 Hz to 6 kHz) using a Cheetah 32 Data Ac- istics of individual spikes across multiple wires
to lambda, and 1 to 1.5 mm ventral to the cortical quisition System. If signals on a given tetrode of a tetrode simultaneously as a 3D scatter plot.

Fig. 6. POR spatial cell types in darkness.

(A) Directional spike plots for an example
center-bearing cell showing tuning stability
across light and dark sessions. Color bar
indicates head direction. (B) Center-distance
tuning curves for an example center-distance
tuned cell showing similar tuning slopes across
sessions. (C) Center-bearing tuning curves for
an example center-bearing tuned cell showing
stability across sessions. (D) Head direction
tuning curves for an example HD cell showing
stability across sessions. (E) Scatter plot
showing firing rate slopes for center-distance
tuning curves between Standard 1 and Dark
sessions for all recorded POR center-distance
cells. (F) Polar plot showing shift in preferred
center bearing between Standard 1 and Dark
sessions for all recorded POR center-bearing cells
(each dot represents one cell). (G) Polar plot
showing shift in preferred firing direction between
Standard 1 and Dark sessions for all recorded
POR HD cells (each dot represents one cell).
Cleanup of automatically sorted clusters, which attributed to cue instability (65). The rat was bility of the global environment and despite the
was not always required, was performed by then returned to the arena and allowed to forage rat’s own self-motion cues.
drawing a polygon around the visually apparent for sugar pellets in the presence of the objects for
boundaries of each cluster. Single-unit isolation a 20-min recording session. Dark sessions
was assessed using metrics such as L-ratio and Following the rotation session, the small box was
isolation distance, as well as assessment of tem- Arena size sessions rotated back to its standard orientation and the
poral autocorrelograms for the presence of a re- Arena size sessions typically followed the object rat was allowed to forage in the standard small
fractory period. Cross-correlograms were also session. Objects were removed from the arena box for a 10-min recording session (lights on). A
analyzed to make sure the same cells were not and the rat was then returned to the 1.2-m box 10-min dark session (lights off) followed this
recorded across different tetrodes. Despite signi- for a 20-min recording session; this recording session, with the rat brought into the recording
ficant advancement of the tetrodes between reconstituted the “large box” session. In cases where room in the dark. A final standard session in the
cording sessions, we sometimes found that the an object session was not run first, the baseline small box (lights on) then took place.
same cells were recorded multiple times on the session for the day was used as the “large box
same tetrodes across recording sessions (by session.” In either case, the 1.2-m box was then Histology
analyzing waveform shape and location in clus- replaced with a 1-m square box (height 50 cm). Once recordings were complete, animals were
ter space); in these cases we only used the first The smaller box was visually similar to the large deeply anesthetized with sodium pentobarbital
recording of the cell. For each well-isolated clus- box with gray walls and a white cue card along and small marking lesions were made at the
ter, we saved the timestamps for each spike the same inside wall. The rat was allowed to tetrode tips by passing a small anodal current
and then analyzed and matched them to the forage in the small box for a 10-min recording (15 mA, 15 to 20 s) through two active wires
tracking data. session (“small box” session). In some cases, this from separate tetrodes. Animals were then intra-
session was followed by a second “large box” ses- cardially perfused with saline followed by 10%
Behavioral testing sion, but usually it was followed by a rotation formalin solution, after which the brains were
Object session session. removed from the skull and postfixed in 10%
Following the initial baseline session in the 1.2-m formalin solution with 2% potassium ferrocya-
square box, the rat was returned to its home cage Rotation session nide for at least 24 hours. The brains were then
and the floor paper of the recording arena was Following the “small box” session, the small box transferred to 20% sucrose solution for at least
changed to reduce the presence of local cues was rotated by 45° and the rat was allowed to 24 hours, after which they were frozen and sliced
from the previous session. This procedure was forage in the rotated box for a 10-min recording sagitally (30 µm sections) using a cryostat. Sec-
repeated between all subsequent sessions. Three session. No attempt was made to disorient the tions were mounted on glass microscope slides
objects (glass jars with black lids) were then animal before recording. This procedure ensured and stained with thionin, after which electrode
placed in the arena. The locations of the objects that any shift in the preferred firing direction of tracks were examined using a light microscope.
were kept relatively constant from day to day head direction cells would be due to changes in Locations of recorded cells were determined by
so that any lack of object tuning could not be the local environment despite the perceived sta- measuring backward from the most ventral

location of the marking lesions or, if marking (20 × 20 for the 1.2-m box). The mean vector the path taken by the animal during foraging
lesions were not visible, the electrode tracks. De- lengths of the curves were extracted and visual- (gray trace) overlaid with dots indicating the
lineations of parahippocampal regions were drawn ized as a heatmap. Locations with strong ego- animal’s location when a single cell fired a spike.
mainly from (26), (66), and (67). centric tuning for a given cell could be visually The dots are colored (circular rainbow color pa-
discerned as a “hot spot” of high mean vector lette) according to the animal’s allocentric head
Computation of egocentric bearing length values, such that the cell fired preferen- direction when the spike was fired; red = 0°,
For a given egocentric reference point (i.e., envi- tially when that location occupied a certain angle green = 90°, blue = 180°, purple = 270°.
ronment center), we first computed the allocen- relative to the animal’s heading. The location
tric bearing of that location from the animal with the highest mean vector length for each Assessment of temporal and
(defined as the angle between the positive x axis cell was extracted as that cell’s preferred ego- spatial stability
with origin centered on the animal and a line centric reference location (fig. S2). Because many We employed several methods to establish that
drawn from the animal to the reference point) cells showed conjunctive tuning to allocentric center-bearing cells showed consistent tuning
for each time point in the recording session, using head direction that skewed the location of the over time and space. The GLM cross-validation
the following equation: preferred reference, we also corrected for head procedure (materials and methods) ensured that
direction tuning using a maximum likelihood tuning to each specific variable was consistent
Bearingallocentric = arctan2(yref – yanimal, xref – method (fig. S2). across the entire recording session. Additionally,
xanimal) we divided each 20-min baseline session in the
Egocentric distance tuning curves 1.2-m square box into four 5-min segments, and
The egocentric bearing of the reference point Egocentric distance is defined here as the in- the mean angles of the center-bearing tuning
relative to the animal was then computed by sub- stantaneous distance between an animal and a curves for each segment were compared (fig. S5).
tracting the animal’s allocentric head direction given reference location. Egocentric center dis- For stability across space, we divided the envi-
(HD) at each time point: tance was calculated as the distance between the ronment into inner and outer portions, the inner
animal and the geometric center of the environ- portion being a square area that had side lengths
Bearingegocentric = Bearingallocentric – ment. For tuning curve construction, 4-cm bins equal to one-half of a side length of the entire
HDallocentric were used. For each cell, the number of spikes arena. We compared mean angles between periods
fired over the course of the session per bin was when the animal occupied each region to assess
An egocentric bearing of 0° (“egocentric north”) divided by the amount of time the animal spent spatial stability. We also split the cells into two
(20) would indicate that the reference point was in each bin. A regression line was fit to each groups: those that showed preferred center bear-
in front of the animal (allocentric bearing equal curve and the R2 fit of the line was calculated as ings in front of or behind the animal (315° to 45°,
to allocentric heading), whereas an egocentric the linear tuning strength. Cells were only clas- 135° to 225°), and those that showed preferred
bearing of 180° indicates that the reference point sified as center-distance cells if they (i) passed bearings to the left and right of the animal (45° to
was behind the animal. Bearings of 90° and 270° the classification procedure for center-distance 135°, 225° to 315°). We then tested the stability of
indicate bearings to the left and right of the modulation (materials and methods), (ii) had each group across inner and outer zones (fig. S5).
animal, respectively. linear fit values > 95th percentile of a shuffle
distribution (discussed below), and (iii) had max- Theta modulation index
Egocentric bearing tuning curves imum firing rates >1 Hz in their center-distance Theta rhythmicity of each cell’s spike train was
Egocentric bearing tuning curves were created tuning curves. assessed using a theta index (69, 70). We first
using 12° bins. For each cell, the amount of time created a temporal autocorrelogram for each cell
that each bin was sampled and the number of Allocentric head direction tuning curves by tallying the number of spikes that occurred
spikes fired per bin over the course of a session Allocentric head direction tuning curves were within each 5-ms bin of a 1-s window centered on
were calculated, and the tuning curve was com- constructed using the same procedures as ego- each spike. A power spectrum was then com-
puted by dividing the number of spikes per bin centric bearing tuning curves but using instan- puted by performing a fast Fourier transform on
by the amount of sampling time per bin. The taneous allocentric head direction. Cells were the autocorrelogram. The strength of theta mod-
mean vector length and mean angle were then considered HD cells if they (i) passed the classi- ulation (theta index) was computed by first cal-
extracted to indicate tuning strength and pre- fication procedure for head direction modulation culating the mean power within 1 Hz on either
ferred firing direction, respectively. (materials and methods), (ii) had mean vector side of the frequency with the highest power
For center-bearing tuning (egocentric bearing lengths > 95th percentile of a shuffle distribution within a 5- to 11-Hz (theta) range and dividing
of the geometric center of the environment) a (discussed below), and (iii) had maximum firing this by the mean power between 1 and 125 Hz.
single period cosine curve was also fit to the rates >1 Hz in their head direction tuning curves. Only cells with a theta index >5 were considered
center-bearing tuning curve, and the explained HD cell properties shown in table S3 were calcu- theta-modulated.
variance (R2) value of the fit was calculated as lated based on previous methods and analyses
the cosine tuning strength. Units were only clas- [see (40, 68)]. Gridness score
sified as center-bearing cells if they (i) passed Gridness scores were computed as described
the classification procedure for center-bearing Allocentric position firing rate maps previously (42). Briefly, for each cell, a spatial
modulation (materials and methods), (ii) had The animal’s two-dimensional location through- autocorrelogram was computed for the smoothed
mean vector lengths > 95th percentile of a shuffle out the recording session was divided into 2 cm × allocentric position firing rate map which corre-
distribution (discussed below), and (iii) had max- 2 cm bins. For each cell, the number of spikes lated the map with itself (Pearson’s r) at all
imum firing rates >1 Hz in their center-bearing fired when the animal occupied each bin was possible spatial lags. If a cell was hexagonally
tuning curves. divided by the amount of time the animal spent periodic (like a grid cell) there would be a ring
in that bin to calculate a firing rate for each around the center of the autocorrelogram with
Egocentric bearing mean vector location in the environment. The resulting firing six evenly spaced peaks. The ring around the
length maps rate heatmaps were smoothed with a Gaussian center (not including the center) was then cor-
As any location within the environment could filter with a standard deviation of two bins. related with itself (Pearson’s r) at 3° offsets
constitute a potential egocentric bearing refer- from 0° to 180°. A hexagonally periodic signal
ence point, we created egocentric bearing tuning Directional spike plots would cause peaks at offsets of 60° and 120°
curves for a grid of possible reference locations To visualize the directional firing of cells across and troughs at offsets of 30°, 90°, and 150°. The
across the entire environment, spaced 6 cm apart space, we created directional spike plots that plot gridness score was calculated as the difference

between the smallest correlation value at 60° 6. T. Solstad, C. N. Boccara, E. Kropff, M. B. Moser, E. I. Moser, doi: 10.1002/(SICI)1096-9861(19980216)391:3<293::AID-
and 120° and the largest correlation value at Representation of geometric borders in the entorhinal CNE2>3.0.CO;2-X; pmid: 9492202
cortex. Science 322, 1865–1868 (2008). doi: 10.1126/ 29. N. Koganezawa, R. Gisetstad, E. Husby, T. P. Doan, M. P. Witter,
30°, 90°, and 150°. Cells with gridness scores > science.1166466; pmid: 19095945 Excitatory postrhinal projections to principal cells in the
0.4 were considered grid cells (30). 7. C. Lever, S. Burton, A. Jeewajee, J. O’Keefe, N. Burgess, medial entorhinal cortex. J. Neurosci. 35, 15860–15874 (2015).
Boundary vector cells in the subiculum of the hippocampal doi: 10.1523/JNEUROSCI.0653-15.2015; pmid: 26631468
Border score formation. J. Neurosci. 29, 9771–9777 (2009). doi: 10.1523/ 30. C. Wang et al., Egocentric coding of external items in the
Smoothed allocentric position firing rate maps JNEUROSCI.1319-09.2009; pmid: 19657030 lateral entorhinal cortex. Science 362, 945–949 (2018).
8. J. Krupic, M. Bauza, S. Burton, C. Barry, J. O’Keefe, Grid cell doi: 10.1126/science.aau4940; pmid: 30467169
were first thresholded to only include bins higher symmetry is shaped by environmental geometry. Nature 518, 31. M. Shahi, S. Dhingra, R. Sandler, R. Rios, C. Vuong, L. Acharya,
than 20% of each cell’s maximum firing rate. 232–235 (2015). doi: 10.1038/nature14153; pmid: 25673417 M. R. Mehta, A statistical model based approach to decipher
Firing fields were defined as above-threshold 9. D. Derdikman et al., Fragmentation of grid cell maps in a the mechanisms governing spatial and directional tuning of
contiguous groups of bins with size > 200 cm2. multicompartment environment. Nat. Neurosci. 12, 1325–1332 hippocampal neurons. Program No. 508.15.2018 Neuroscience
(2009). doi: 10.1038/nn.2396; pmid: 19749749 Meeting Planner. San Diego, CA: Society for Neuroscience
We next determined the firing field with the 10. R. U. Muller, J. L. Kubie, The effects of changes in the (2018).
most bins along one wall of the enclosure, and environment on the spatial firing of hippocampal complex- 32. A. A. Wilber, B. J. Clark, T. C. Forster, M. Tatsuno,
converted that number of bins into a distance spike cells. J. Neurosci. 7, 1951–1968 (1987). doi: 10.1523/ B. L. McNaughton, Interaction of egocentric and
along that wall, d. We then calculated the aver- JNEUROSCI.07-07-01951.1987; pmid: 3612226 world-centered reference frames in the rat posterior parietal
11. J. O’Keefe, N. Burgess, Geometric determinants of the place cortex. J. Neurosci. 34, 5431–5446 (2014). doi: 10.1523/
age distance of each of the bins in that firing field fields of hippocampal neurons. Nature 381, 425–428 (1996). JNEUROSCI.0511-14.2014; pmid: 24741034
from the associated wall, a. The border score doi: 10.1038/381425a0; pmid: 8632799 33. A. Peyrache, N. Schieferstein, G. Buzsáki, Transformation
was then computed according to the following 12. J. S. Taube, R. U. Muller, J. B. Ranck Jr., Head-direction cells of the head-direction signal into a spatial code. Nat. Commun.
equation (70): recorded from the postsubiculum in freely moving rats. 8, 1752 (2017). doi: 10.1038/s41467-017-01908-3;
I. Description and quantitative analysis. J. Neurosci. 10, pmid: 29170377
420–435 (1990). doi: 10.1523/JNEUROSCI.10-02-00420.1990; 34. J. R. Hinman, G. W. Chapman IV, M. E. Hasselmo, Neuronal
B = (d – a) / (d + a) pmid: 2303851 representation of environmental boundaries in egocentric
13. J. O’Keefe, An allocentric spatial model for the hippocampal coordinates. Nat. Commun. 10, 2772 (2019). doi: org/10.1038/
Cells that did not pass the gridness threshold cognitive map. Hippocampus 1, 230–235 (1991). doi: 10.1002/ s41467-019-10722-y
hipo.450010303; pmid: 1669295 35. J. J. Knierim, J. P. Neunuebel, S. S. Deshmukh, Functional
but had border scores > 0.5 were considered 14. D. S. Touretzky, A. D. Redish, H. S. Wan, Neural representation correlates of the lateral and medial entorhinal cortex: Objects,
border cells (30). of space using sinusoidal arrays. Neural Comput. 5, 869–884 path integration and local-global reference frames. Philos.
(1993). doi: 10.1162/neco.1993.5.6.869 Trans. R. Soc. Lond. B Biol. Sci. 369, 20130369 (2013).
Shuffling procedure 15. D. S. Touretzky, A. D. Redish, Theory of rodent navigation doi: 10.1098/rstb.2013.0369; pmid: 24366146
based on interacting representations of space. Hippocampus 6, 36. K. Hardcastle, N. Maheswaranathan, S. Ganguli, L. M. Giocomo,
Each cell’s spike train was randomly shifted be- 247–270 (1996). doi: 10.1002/(SICI)1098-1063(1996) A multiplexed, heterogenous, and adaptive code for navigation
tween 30 s and 30 s less than the length of the 6:3<247::AID-HIPO4>3.0.CO;2-K; pmid: 8841825 in medial entorhinal cortex. Neuron 94, 375–387.e7 (2017).
recording session, with entries beyond the end 16. B. L. McNaughton, B. Leonard, L. Chen, Cortical-hippocampal doi: 10.1016/j.neuron.2017.03.025; pmid: 28392071
wrapped to the beginning, to offset the spike interactions and cognitive mapping: A hypothesis based on 37. N. Burgess, F. Cacucci, C. Lever, J. O’keefe, Characterizing
reintegration of the parietal and inferotemporal pathways multiple independent behavioral correlates of cell firing in
data from the behavioral data without inter- for visual processing. Psychobiology 17, 230–235 (1989). freely moving animals. Hippocampus 15, 149–153 (2005).
rupting its temporal structure. Relevant tuning doi: 10.1007/BF03337774 doi: 10.1002/hipo.20058; pmid: 15558542
scores were then computed based on the shifted 17. P. Byrne, S. Becker, N. Burgess, Remembering the past and 38. Center-bearing and center-distance tuning among POR cells
spike train. This procedure was repeated 100 times imagining the future: A neural model of spatial memory and were tested against a number of alternative models, including
imagery. Psychol. Rev. 114, 340–375 (2007). doi: 10.1037/ egocentric encoding of nearby boundaries and egocentric
for each cell, and all of the shuffled measures 0033-295X.114.2.340; pmid: 17500630 bearing based on movement direction. Center-bearing and
were combined into a single shuffle distribution 18. A. Bicanski, N. Burgess, A neural-level model of spatial center-distance were the preferred firing correlates among
for each brain region. A 95th percentile cutoff memory and imagery. eLife 7, e33752 (2018). doi: 10.7554/ POR cells in all analyses (fig. S6).
was used to determine tuning significance for eLife.33752; pmid: 30176988 39. The slight offset from front (0°) and back (180°) could be due
19. B. L. McNaughton, J. J. Knierim, M. A. Wilson, Vector encoding to our electrode placement solely in the right hemisphere; it is
individual cells. possible that recordings from the left hemisphere (which
and the vestibular foundations of spatial cognition:
neurophysiological and computational mechanisms. In: primarily receives visual information from the opposite visual
Statistics Gazzaniga MS, editor. The Cognitive Neurosciences. Cambridge, field) could show an offset in the opposite direction.
Statistical analyses were carried out using cus- MA: MIT; 585–595 (1995). 40. J. S. Taube, Head direction cells recorded in the anterior
20. C. R. Gallistel, The Organization of Learning (Bradform Books/ thalamic nuclei of freely moving rats. J. Neurosci. 15,
tom Python and R code. As several distributions 70–86 (1995). doi: 10.1523/JNEUROSCI.15-01-00070.1995;
MIT, 1990).
appeared non-normal (e.g., center-distance slopes), 21. G. K. Aguirre, J. A. Detre, D. C. Alsop, M. D’Esposito, The pmid: 7823153
medians and bootstrapped 95% confidence in- parahippocampus subserves topographical learning in man. 41. R. W. Stackman, J. S. Taube, Firing properties of rat lateral
tervals were included in the text instead of means Cereb. Cortex 6, 823–829 (1996). doi: 10.1093/cercor/6.6.823; mammillary single units: Head direction, head pitch, and
pmid: 8922339 angular head velocity. J. Neurosci. 18, 9020–9037 (1998).
and standard errors. All tests were two-sided doi: 10.1523/JNEUROSCI.18-21-09020.1998; pmid: 9787007
22. G. K. Aguirre, M. D’Esposito, Topographical disorientation:
(except for GLM classifier cross-validation com- A synthesis and taxonomy. Brain 122, 1613–1628 (1999). 42. S. S. Winter, B. J. Clark, J. S. Taube, Spatial navigation.
parisons) and used an alpha level of 0.05. doi: 10.1093/brain/122.9.1613; pmid: 10468502 Disruption of the head direction cell network impairs the
23. R. Epstein, N. Kanwisher, A cortical representation of the parahippocampal grid cell signal. Science 347, 870–874
local visual environment. Nature 392, 598–601 (1998). (2015). doi: 10.1126/science.1259591; pmid: 25700518
doi: 10.1038/33402; pmid: 9560155 43. See materials and methods.
RE FE RENCES AND N OT ES 24. R. A. Epstein, W. E. Parker, A. M. Feiler, Where am I now? 44. A. P. Georgopoulos, A. B. Schwartz, R. E. Kettner,
1. K. Cheng, A purely geometric module in the rat’s spatial Distinct roles for parahippocampal and retrosplenial cortices Neuronal population coding of movement direction.
representation. Cognition 23, 149–178 (1986). doi: 10.1016/ in place recognition. J. Neurosci. 27, 6141–6149 (2007). Science 233, 1416–1419 (1986). doi: 10.1126/science.3749885;
0010-0277(86)90041-7; pmid: 3742991 doi: 10.1523/JNEUROSCI.0799-07.2007; pmid: 17553986 pmid: 3749885
2. J. W. Kelly, T. P. McNamara, B. Bodenheimer, T. H. Carr, 25. R. D. Burwell, The parahippocampal region: Corticocortical 45. A. P. Georgopoulos, R. E. Kettner, A. B. Schwartz,
J. J. Rieser, The shape of human navigation: How connectivity. Ann. N. Y. Acad. Sci. 911, 25–42 (2000). Primate motor cortex and free arm movements to visual
environmental geometry is used in maintenance of spatial doi: 10.1111/j.1749-6632.2000.tb06717.x; pmid: 10911865 targets in three-dimensional space. II. Coding of the
orientation. Cognition 109, 281–286 (2008). doi: 10.1016/ 26. C. N. Boccara et al., Grid cells in pre- and parasubiculum. direction of movement by a neuronal population. J. Neurosci.
j.cognition.2008.09.001; pmid: 18952206 Nat. Neurosci. 13, 987–994 (2010). doi: 10.1038/nn.2602; 8, 2928–2937 (1988). doi: 10.1523/JNEUROSCI.08-08-
3. W. Mou, T. P. McNamara, Intrinsic frames of reference in pmid: 20657591 02928.1988; pmid: 3411362
spatial memory. J. Exp. Psychol. Learn. Mem. Cogn. 28, 27. K. L. Agster, R. D. Burwell, Hippocampal and subicular 46. E. Salinas, L. F. Abbott, Vector reconstruction from firing rates.
162–170 (2002). pmid: 11827078 efferents and afferents of the perirhinal, postrhinal, J. Comput. Neurosci. 1, 89–107 (1994). doi: 10.1007/
4. J. O’Keefe, L. Nadel, The Hippocampus as a Cognitive Map and entorhinal cortices of the rat. Behav. Brain Res. 254, BF00962720; pmid: 8792227
(Oxford Univ.Press, 1978). 50–64 (2013). doi: 10.1016/j.bbr.2013.07.005; 47. E. N. Brown, L. M. Frank, D. Tang, M. C. Quirk, M. A. Wilson,
5. T. Hafting, M. Fyhn, S. Molden, M. B. Moser, E. I. Moser, pmid: 23872326 A statistical paradigm for neural spike train decoding applied
Microstructure of a spatial map in the entorhinal cortex. 28. R. D. Burwell, D. G. Amaral, Perirhinal and postrhinal cortices to position prediction from ensemble firing patterns of rat
Nature 436, 801–806 (2005). doi: 10.1038/nature03721; of the rat: Interconnectivity and connections with the hippocampal place cells. J. Neurosci. 18, 7411–7425 (1998).
pmid: 15965463 entorhinal cortex. J. Comp. Neurol. 391, 293–321 (1998). doi: 10.1523/JNEUROSCI.18-18-07411.1998; pmid: 9736661

48. K. Zhang, I. Ginzburg, B. L. McNaughton, T. J. Sejnowski, cortices of the rat. I. afferents. Hippocampus 26, 1189–1212 68. M. L. Mehlman, S. S. Winter, S. Valerio, J. S. Taube, Functional
Interpreting neuronal population activity by reconstruction: (2016). doi: 10.1002/hipo.22603; pmid: 27119220 and anatomical relationships between the medial precentral
Unified framework with application to hippocampal place cells. 58. S. J. Y. Mizumori, J. D. Williams, Directionally selective cortex, dorsal striatum, and head direction cell circuitry.
J. Neurophysiol. 79, 1017–1044 (1998). doi: 10.1152/ mnemonic properties of neurons in the lateral dorsal I. Recording studies. J. Neurophysiol. 121, 350–370 (2019).
jn.1998.79.2.1017; pmid: 9463459 nucleus of the thalamus of rats. J. Neurosci. 13, 4015–4028 doi: 10.1152/jn.00143.2018; pmid: 30427767
49. B. L. McNaughton, F. P. Battaglia, O. Jensen, E. I. Moser, (1993). doi: 10.1523/JNEUROSCI.13-09-04015.1993; 69. M. M. Yartsev, M. P. Witter, N. Ulanovsky, Grid cells
M. B. Moser, Path integration and the neural basis of pmid: 8366357 without theta oscillations in the entorhinal cortex of bats.
the ‘cognitive map’. Nat. Rev. Neurosci. 7, 663–678 (2006). 59. M. M. Jankowski et al., Nucleus reuniens of the thalamus Nature 479, 103–107 (2011). doi: 10.1038/nature10583;
doi: 10.1038/nrn1932; pmid: 16858394 contains head direction cells. eLife 3, e03075 (2014). pmid: 22051680
50. A. S. Alexander, D. A. Nitz, Retrosplenial cortex maps the doi: 10.7554/eLife.03075; pmid: 25024427 70. E. Kropff, J. E. Carmichael, M. B. Moser, E. I. Moser, Speed cells
conjunction of internal and external spaces. Nat. Neurosci. 18, 60. F. Savelli, J. D. Luck, J. J. Knierim, Framing of grid cells within in the medial entorhinal cortex. Nature 523, 419–424 (2015).
1143–1151 (2015). doi: 10.1038/nn.4058; pmid: 26147532 and beyond navigation boundaries. eLife 6, e21354 (2017). doi: 10.1038/nature14622; pmid: 26176924
51. B. F. Jones, M. P. Witter, Cingulate cortex projections to the doi: 10.7554/eLife.21354; pmid: 28084992 71. P. A. LaChance, T. P. Todd, J. S. Taube, GLM classification
parahippocampal region and hippocampal formation in the rat. 61. R. D. Burwell, D. M. Hafeman, Positional firing properties code. Zenodo, Version 1.0.1 (2019); doi: 10.5281/
Hippocampus 17, 957–976 (2007). doi: 10.1002/hipo.20330; of postrhinal cortex neurons. Neuroscience 119, zenodo.3172635
pmid: 17598159 577–588 (2003). doi: 10.1016/S0306-4522(03)00160-X;
52. G. M. Olsen, S. Ohara, T. Iijima, M. P. Witter, Parahippocampal pmid: 12770570 AC KNOWLED GME NTS
and retrosplenial connections of rat posterior parietal cortex. 62. S. C. Furtak, O. J. Ahmed, R. D. Burwell, Single neuron activity We thank M. L. Mehlman, S. S. Winter, and J. L. Marcroft for
Hippocampus 27, 335–358 (2017). doi: 10.1002/hipo.22701; and theta modulation in postrhinal cortex during visual technical assistance, and R. D. Burwell for assistance with
pmid: 28032674 object discrimination. Neuron 76, 976–988 (2012). histological analysis. Funding: This work was funded by NIH grants
53. J. Cho, P. E. Sharp, Head direction, place, and movement doi: 10.1016/j.neuron.2012.10.039; pmid: 23217745 NS053907 (J.S.T.) and MH116158 (T.P.T.). Author contributions:
correlates for cells in the rat retrosplenial cortex. Behav. 63. M. Fyhn, S. Molden, M. P. Witter, E. I. Moser, M. B. Moser, P.A.L. conceived of the project and designed experiments and
Neurosci. 115, 3–25 (2001). doi: 10.1037/0735-7044.115.1.3; Spatial representation in the entorhinal cortex. Science analyses. P.A.L. and T.P.T. performed surgeries and ran
pmid: 11256450 305, 1258–1264 (2004). doi: 10.1126/science.1099901 experiments. P.A.L. ran the analyses. P.A.L. and J.S.T. discussed
54. K. L. Agster, I. Tomás Pereira, M. P. Saddoris, R. D. Burwell, pmid: 15333832 the analyses and wrote the manuscript. Competing interests:
Subcortical connections of the perirhinal, postrhinal, and 64. M. Pachitaru, N. A. Steinmetz, S. Kadir, M. Carandini, None declared. Data and materials availability: Methods are
entorhinal cortices of the rat. II. efferents. Hippocampus 26, K. D. Harris, Fast and accurate spike sorting of high-channel as described in the text. Python code used for analyses is available
1213–1230 (2016). doi: 10.1002/hipo.22600; pmid: 27101786 count probes with Kilosort. Adv. Neural Inf. Process. Syst. on Github (71). Data are available as supplementary materials.
55. M. L. Mehlman, S. S. Winter, J. S. Taube, Functional and 29, 4448–4456 (2016).
anatomical relationships between the medial precentral cortex, 65. J. J. Knierim, H. S. Kudrimoti, B. L. McNaughton, Place SUPPLEMENTARY MATERIALS
dorsal striatum, and head direction cell circuitry. II. cells, head direction cells, and the learning of landmark
science.sciencemag.org/content/365/6449/eaax4192/suppl/DC1
Neuroanatomical studies. J. Neurophysiol. 121, 371–395 stability. J. Neurosci. 15, 1648–1659 (1995). doi: 10.1523/
(2019). doi: 10.1152/jn.00144.2018; pmid: 30427743 JNEUROSCI.15-03-01648.1995; pmid: 7891125
Figs. S1 to S13
56. M. G. Packard, J. L. McGaugh, Inactivation of hippocampus or 66. C. N. Boccara et al., A three-plane architectonic atlas of the rat
Tables S1 to S3
caudate nucleus with lidocaine differentially affects expression hippocampal region. Hippocampus 25, 838–857 (2015).
Reference (72)
of place and response learning. Neurobiol. Learn. Mem. 65, doi: 10.1002/hipo.22407; pmid: 25533645
Data Table S1
65–72 (1996). doi: 10.1006/nlme.1996.0007; pmid: 8673408 67. R. D. Burwell, Borders and cytoarchitecture of the perirhinal
57. I. Tomás Pereira, K. L. Agster, R. D. Burwell, Subcortical and postrhinal cortices in the rat. J. Comp. Neurol. 437, 17–41 21 March 2019; accepted 30 May 2019
connections of the perirhinal, postrhinal, and entorhinal (2001). doi: 10.1002/cne.1267; pmid: 11477594 10.1126/science.aax4192

R ES E A RC H
◥ multilayer NaCl area under frequency-shift (Df )

REPORTS feedback, in which constant Df AFM imaging
was used to locate molecules. To identify and
manipulate the charge state of molecules, we
SURFACE CHEMISTRY used Kelvin probe force spectroscopy, that is, Df
as a function of sample voltage V. Transitions in
Molecular structure elucidation the molecular charge state were observed as

steps between different D f(V ) parabolas (11–13),
in which each step upon ramping V in the pos-
with charge-state control itive (negative) direction corresponded to a de-
crease (increase) in charge state by attaching
one electron (hole).
Shadi Fatayer1*, Florian Albrecht1, Yunlong Zhang2, Darius Urbonas1, Diego Peña3,
We assign the charge states by comparison to
Nikolaj Moll1, Leo Gross1* the ion resonances and pristine charge states on
NaCl(2ML)/Cu(111) by using the scattering of
The charge state of a molecule governs its physicochemical properties, such as surface-state electrons and/or interface-state
conformation, reactivity, and aromaticity, with implications for on-surface synthesis, localization as indications (16). Moreover, the
catalysis, photoconversion, and applications in molecular electronics. On insulating, separation (gap) between transition voltages on
multilayer sodium chloride (NaCl) films, we controlled the charge state of organic the insulating film provided another indicator
molecules and resolved their structures in neutral, cationic, anionic, and dianionic (fig. S8). To avoid exchange of electrons with the
states by atomic force microscopy, obtaining atomic resolution and bond-order tip during imaging, we obtained AFM images at
discrimination using carbon monoxide (CO)–functionalized tips. We detected changes voltages between different charge transitions,
in conformation, adsorption geometry, and bond-order relations for azobenzene, in different charge states. The AFM images were
tetracyanoquinodimethane, and pentacene in multiple charge states. Moreover, for taken in constant-height mode, at a tip height z
porphine, we investigate the charge state–dependent change of aromaticity and near the minimum of Df(z) above the molecule,
conjugation pathway in the macrocycle. This work opens the way to studying chemical- which roughly corresponded to a distance of 3.9 Å
structural changes of individual molecules for a wide range of charge states. between the molecule imaged and the oxygen
T
atom of the tip (17). The atomic contrast of the
he charge-induced changes of a molecule sulating films, such as bilayer NaCl, in which AFM images confirmed that the tip remained
have fundamental implications for chem- tunneling to the substrate and STM operation is CO functionalized, even when biases of several
ical reactions, catalysis, electrochemistry, possible. For some molecules, two charge states volts were applied.
photoconversion, and charge transport. can be stabilized under special conditions of Azobenzene (A) (Fig. 1A) is an archetypical
Charged molecules can be studied in various energy-level alignment and relaxation energy, molecular electronics building block (18) and
environments (1), such as gas phase, solution, that is, when reorganization in the film/molecule photomechanical actuator (19) that can be
and solid. Experimentally, structural informa- leads to level shifting across the Fermi level. In switched between cis and trans isomers by light
tion is mostly obtained from x-ray diffraction such special cases, molecular conformational (20, 21), tunneling electrons (22), and electric
for molecular solids (2) and from vibrational and switching (6), tunnel barrier changes (7), and fields (23). The AFM image of neutral A0 (Fig. 1B)
optical spectroscopy for molecules in solution formation of a metal-molecule complex (8) have showed that it was adsorbed with both phenyl
and gas phase (3). Typically, counterions are been demonstrated. rings slightly tilted out of plane with respect
used to stabilize charged molecules and avoid On thicker insulating films, electrons cannot to the NaCl surface. The phenyl rings were ap-
fragmentation or charge leakage in solids and be exchanged with the substrate, and thus multiple proximately parallel (figs. S2 to S4), indicated by
solutions (4). Hence, the crystallization and charge states can be stabilized in general (9) and the same sides of the phenyl rings appearing
the counterions may affect the geometry of reorganization energies determined (10). In this bright and with comparable Df contrast in the
the molecules. study, we complement single-electron sensitivity AFM image.
The aforementioned experimental techniques (11–13) and charge-state control (9) with atomic An electron was attached to A0 when V was
predominantly probe a large number of molecules. resolution of AFM by using CO tips (14). We ramped above 2 V (Fig. 1A), creating the anion
The ability to detect single-molecule changes upon resolve molecules in multiple charge states (oxi- A−1. In A−1 (Fig. 1C), the phenyl rings were also
charging with atomic resolution would allow pre- dation states) with atomic resolution, including tilted out of plane but in opposite directions,
viously unknown phenomena to be observed bond-order analysis. The molecular charge states indicated by opposite sides of the two phenyl
(5) and the influence of the environment to are controlled by deliberately exchanging elec- rings appearing bright in the AFM image (fig. S5).
be quantified. Moreover, it would also affect trons between tip and molecule on an insulating Hence, the conformation changed when A0 was
applications such as devices based on single- substrate (9), accessing multiple charge states, reduced to A−1. A sequence of AFM images of A
electron transfer as well as thin-film devices including doubly charged ions. The changes in while alternating its charge state demonstrated
and provide fundamental insights into redox the molecular geometry of charged species can the reversibility and reproducibility of this charge-
reactions and charge-carrier injection. often be rationalized by using classical organic conformation switching (fig. S6).
Scanning tunneling microscopy (STM) and chemistry approaches such as the evaluation of The AFM measurements were in excellent
atomic force microscopy (AFM) imaging of ad- Lewis resonance structures or Clar’s sextet rule. agreement with AFM simulations (24) of the
sorbed molecules in different charge states have In this study, we investigated four model com- adsorbed molecule (Fig. 1, D and E) performed
been shown in specific cases on ultrathin in- pounds with different properties and applica- in the respective geometries obtained by den-
tions related to their charge-state transitions. sity functional theory (DFT) calculations (see
1
IBM Research–Zurich, Rueschlikon 8803, Switzerland.
We resolved minute changes in adsorption ge- supplemental material for details and tables S1
2
ExxonMobil Research and Engineering Company, Annandale, ometry, bond-order relations, and aromaticity. and S2) of A0 (Fig. 1F) and A−1 (Fig. 1H). For
NJ 08801, USA. 3Centro Singular de Investigación en Química All of the experiments were performed on an both oxidation states, A was in the trans con-
Biolóxica e Materiais Moleculares (CIQUS) and Departamento de NaCl film thicker than 20 monolayers (MLs) on formation, but there were small differences in
Química Orgánica, Universidade de Santiago de Compostela,
Santiago de Compostela 15782, Spain.
Cu(111) by using a qPlus-based (15) AFM under the molecular geometry. A0 was planar, and the
*Corresponding author. Email: sfa@zurich.ibm.com (S.F.); lgr@ ultrahigh vacuum at 5 K (materials and meth- molecular plane was tilted by 17° with respect to
zurich.ibm.com (L.G.) ods). The CO-functionalized tip approached the the surface plane. Conversely, A−1 was nonplanar

RE S EAR CH | R E P O R T S
Fig. 2. Measurements and calculations on pentacene. Constant-height AFM images of the

(A) cationic (P+1), (B) neutral (P0), (C) anionic (P−1), and (D) dianionic (P−2) molecule. Scale bars
represent 5 Å. (B) and (C) are imaged at a tip-sample distance that was 0.3 Å smaller than in (A) and
(D). V is (A) −3.3 V, (B) 0.5 V, (C) 2.5 V, and (D) 3.6 V. (E) DFT-calculated average C-C bond length of
each hydrocarbon ring for different charge states. (F) Possible resonance structures of P0 and P−2.
Fig. 1. Measurements and calculations on
azobenzene. (A) Df(V) spectrum recorded on tilting and thus also affect the apparent bond ture with three Clar aromatic sextets distributed
top of an azobenzene molecule. V was ramped length (17, 24). Thus, bonds should be compared in the first, third, and fifth rings. The measure-
from 1 to 3 V. The inset shows the chemical only if they are in similar local environments. ments of pentacene show that small charge-
structure of azobenzene. (B) Constant-height Before applying this method to resolve bond- induced effects in the bond-length relations
AFM image of A0 at V = 0.5 V. (C) Constant- order relations of molecules with oxidation can be resolved by AFM. Importantly for redox
height AFM image of A−1 at V = 2.5 V, tip-sample state–dependent functions, we characterized the reactions, we obtain the locations of sites with
distance reduced by 0.3 Å with respect to (B). well-studied model system of pentacene (P). The increased radical anion character upon charging.
(D and E) Simulated AFM images of on-surface D f(V) spectrum of P (fig. S8) revealed four dif- We applied our method to the electron ac-
A0 and A−1, respectively. All scale bars corre- ferent oxidation states, cation (P+1), neutral (P0), ceptor tetracyanoquinodimethane (T) (Fig. 3A),
spond to 5 Å. (F and H) Top view of the atomic anion (P−1), and dianion (P−2). The differences relevant for doping in organic electronics (26).
models of A0 and A−1, respectively. (G and I) observed in the AFM images (Fig. 2, A to D, and The Df(V) spectrum of T (fig. S14) revealed three
Chemical structures of A0 and A−1, respectively, fig. S9) revealed an apparent contraction (elon- different oxidation states, neutral (T 0), anion
with wedged bonds representing out-of-plane gation) of the molecule along its short (long) axis (T −1), and dianion (T −2). Consistent with its
conformations. with increased negative charge, which agrees large electron affinity, we observed the anion
with the calculations (fig. S10). However, when T −1 even at 0 V. The AFM image of T 0 (Fig. 3B)
with the phenyl rings tilted in opposite directions comparing apparent bond lengths obtained at showed four attractive lobes, two large ones and
by ~4° with respect to the surface plane. The different voltages, the electric field changes two small ones (fig. S15), which we assigned to
switch from a planar to a nonplanar confor- with different V and affects the spring constant an upstanding adsorption conformation (27) for
mation can be rationalized by the reduction of and tilt of the CO molecule at the tip apex (24). T 0 (fig. S16 and supplementary text SM2). AFM
the azo group (N=N) when switching from A0 In fact, we observed an overall compression measurements of T 0 at smaller tip-molecule dis-
to A−1, which alters the p conjugated system of the molecular image of P0 with increased tances than in Fig. 3B resulted in the molecule
and induces the distortion from planarity. The V (fig. S11 and supplementary text SM1). How- being picked up by the tip.
example of azobenzene showed that charge- ever, differences between individual bonds in In contrast, the AFM image of T −1 (Fig. 3C)
induced changes in the adsorption geometry one image, measured at identical voltages, could resolved the central carbon ring adsorbed par-
and molecular conformation—that is, tilts of be compared without such systematic error. For allel to the surface. The adsorption orientation
a few degrees of molecular moieties—could bond-order analysis, we compared only the con- switch between T 0 and T −1 was reversible and
be detected. trast within individual AFM images and bonds did not result in lateral movement of the mol-
Bond-order analysis can be a powerful tool for with a similar local environment. ecule (fig. S17). The contrast on the central ring
resolving structural and electronic properties In the images of P−1 and P−2, we observed a of T −1 indicates bond-length alternation (BLA).
of molecules in different charge states. In gen- D f modulation with the centers of the second Increased Df, indicative of shorter bond length,
eral, bond-order differences can be qualitatively and fourth ring appearing darker than the cen- was observed above the double bonds of the
resolved by AFM as different brightness (Df values) ters of the other rings, indicating that these ring as depicted in its neutral chemical struc-
and as different apparent bond length (17, 25). rings increased in diameter and/or exhibited a ture (Fig. 3A). The BLA in the ring suggests
Brighter Df contrast is caused by increased Pauli reduction in electron density. DFT calculations that T −1 does not have a perfect benzenoid char-
repulsion caused by increased electron density of the average C-C bond length in individual acter (28). However, in T −2, the Df contrast of the
and thus indicates smaller bond length. At small rings (Fig. 2E) showed that P−1 and P−2 fea- central ring became homogeneous (Fig. 3D),
tip-sample distances, the tilting of the CO at the tured second and fourth rings with an increased indicating no BLA and thus a change to a ben-
tip strongly affects the images and leads to a average bond length compared with the other zenoid character (Fig. 3E), in agreement with
magnifying effect, increasing bond-length dif- rings. For the dianion P−2, the effect could be previous calculations (29). In addition, compar-
ferences by about one order of magnitude (17). rationalized by considering the reduction of ing T −2 with T −1, the regions of the carbons
Moreover, background forces related to the lo- pentacene to form radical anions in the second connecting the cyano groups showed increased
cal potential landscape contribute to the CO and fourth rings (Fig. 2F), leading to a struc- Df for the dianion. Possible explanations include

R ES E A RC H | R E PO R TS
higher electron density at these carbons or

adsorption-height changes in the termination
of the molecule upon charge-state change. DFT
calculations (tables S9 and S10) indicated that
the cyano groups were more bent for T −1 than
for T −2. In the observed adsorption orientation
changes upon reduction of T 0, the prevalent
phenyl character changed from quinoid in T −1
Fig. 3. Tetracyanoquinodimethane model and measurements. (A) Chemical structure of T.
to benzenoid in T −2 and was accompanied by
Constant-height AFM images of the (B) neutral (T 0), (C) anionic (T −1), and (D) dianionic
small geometry changes in the dicyano moieties.
(T −2) molecule. Panels (C) and (D) are imaged with a tip-sample distance that is 1.9 Å smaller
We also applied our method to porphyrins,
than in (B). Scale bars represent 5 Å. V is indicated in each image. (E) Major resonance
which in multiple oxidation states fulfill essen-
structure proposed for T −2.
tial functions in medicine, biology, chemistry,
and physics (30). The aromaticity and conjuga-
tion pathway of porphyrins in different oxidation
states are debated (31). The parent compound
of porphyrins is porphine (F), a fully conjugated
substituent-free planar macrocycle (Fig. 4A). Ac-
cording to the [18]annulene model, the neutral
molecule has an aromatic conjugation pathway
involving 18p electrons (4n + 2), indicated by the
red-colored bonds of the resonance structures
shown in Fig. 4A, bypassing the NH in the pyrroles
and the outer CH=CH groups of the azafulvene
rings. Upon double reduction, the macrocyclic
conjugation pathway changes to a formally anti-
aromatic 20p-electron (4n) system, encompass-
ing the whole periphery of porphine (Fig. 4B).
The change in the macrocyclic p conjugation
can influence the global aromaticity of the
molecule, although local heterocyclic p circuits
(pyrrole subunits, 6p-electron, 4n + 2) would
contribute as well (31).
The Df (V ) spectrum of F (fig. S21) revealed
three different oxidation states, neutral (F 0),
anionic (F −1), and dianionic (F −2). The AFM
images and the respective Laplace-filtered images
are shown in Fig. 4, C to H, and in figs. S22
and S23. The location of the hydrogens inside
the cavity can be inferred from AFM images at
increased tip-sample distance (fig. S24) and is
indicated in Fig. 4C. We observed tautomeri-
zation switching (32) of F only at V > 3.7 V
(fig. S25). In the individual AFM images, we
observed apparent bond-length differences in
the pyrrole and azafulvene rings as well as in
the methine bridges, which connect the five-
membered rings. The outer C-C bonds of the
pyrrole and the azafulvene ring, labeled a and c
(Fig. 4I), respectively, can be compared within
an image (SM1) because of the similar environ-
ments (Fig. 4J).
For F 0, a appeared longer than c in con-
formity with the resonance structures shown
in Fig. 4A, in which c is a double bond and not
included in the conjugation pathway. For F −2,
a and c appeared with the same apparent length
within the measurement accuracy, in agreement
with the conjugation pathway highlighted in Fig.
4B including a and c. For F −1, bonds a and c had
Fig. 4. Analysis of porphine and its conjugation pathway. Chemical structure of (A) neutral (F0) qualitatively the same relation as in F 0—that is,
and (B) dianionic (F −2) porphine. The red path shows the expected annulene-type conjugation a larger than c—indicating that the conjugation
pathway for each charge state. Constant-height and corresponding Laplace-filtered AFM images of pathway of the neutral molecule was maintained
(C and D) F 0, (E and F) F −1, and (G and H) F −2. The constant-height AFM images in (E) and (G) in the anion, although a less stable 19p-electron
are taken at tip-sample distances larger by 0.5 Å and 0.4 Å, respectively, than the AFM image in (C). system was formed.
All scale bars correspond to 5 Å. V is indicated in each image. (I) Highlighted bonds in F. Measured Another possible comparison is in each bond
apparent bond lengths of (J) a and c and (K) l1 and l2 of F as a function of charge state. of the methine bridge, labeled l1 and l2 (Fig. 4K).

These bonds change appreciably for each charge 32. P. Liljeroth, J. Repp, G. Meyer, Science 317, 1203–1206 analyzed the data. S.F. and L.G. drafted the manuscript and
state, but a simple picture of resonance Kekulé (2007). finalized it with the input from F.A., Y.Z., D.U., D.P., and N.M.
33. L. L. Patera, F. Queck, P. Scheuerer, J. Repp, Nature 566, Competing interests: The authors declare no competing interests.
structures cannot rationalize their behavior. For Data and materials availability: All data are available in the
245–248 (2019).
both F 0 and F −2, l1 was shorter than l2, with the main text or the supplementary materials.
difference between l1 and l2 being larger in F −2 ACKN OWLED GMEN TS
than in F0. The latter indicates increased BLA We acknowledge G. Meyer, R. Allenspach, J. Repp, and S. J. Garden SUPPLEMENTARY MATERIALS
for F −2 and thus likely a reduced aromaticity for discussions. Funding: The project was supported by the science.sciencemag.org/content/365/6449/142/suppl/DC1
with respect to F 0. For the anion F −1, l1 was European Research Council Consolidator grant AMSEL (contract Materials and Methods
no. 682144). D.P. thanks Agencia Estatal de Investigación Supplementary Text
longer than l2; the asymmetry at the meso-carbon (MAT2016-78293-C6-3-R), Xunta de Galicia (Centro singular de Figs. S1 to S29
position of the methine bridge was inverted with investigación de Galicia, accreditation 2016–2019, ED431G/09), Tables S1 to S14
respect to both F 0 and F −2. Additional insight was and the European Regional Development Fund for financial References (34–41)
support. Author contributions: S.F. and L.G. designed the
gained by DFT calculations of F adsorbed on
experiments. S.F. carried out the experiments with support from 10 April 2019; accepted 12 June 2019
NaCl in different charge states (tables S11 to F.A. S.F. and N.M. carried out the DFT calculations. All authors 10.1126/science.aax5895
S13), in qualitative agreement with the exper-
iment (fig. S26). In porphine, we observed sub-
stantial changes of the bond-order relations
upon charging and used them to track the ARTIFICIAL MUSCLES
evolution of the macrocycle’s structure, aroma-
ticity, and conjugation pathway as a function
of its oxidation state. With the demonstrated
bond-order resolution and charge-state control,
Strain-programmable fiber-based
our method effectively complements redox po-
tential measurements (10) and mapping of artificial muscle
orbital densities (33) for the investigation of
charged molecules on surfaces. Mehmet Kanik1,2*, Sirma Orguc3*, Georgios Varnavides1,2,4, Jinwoo Kim2,
Thomas Benavides3, Dani Gonzalez5, Timothy Akintilo6, C. Cem Tasan2,
RE FE RENCES AND N OT ES
1. V. Balzani, Ed., Electron Transfer in Chemistry (Wiley-VCH
Anantha P. Chandrakasan3, Yoel Fink1,2, Polina Anikeeva1,2†
Verlag GmbH, Weinheim, Germany, 2001); http://doi.wiley.
com/10.1002/9783527618248. Artificial muscles may accelerate the development of robotics, haptics, and prosthetics.
2. J. S. Miller, D. A. Dixon, J. C. Calabrese, Science 240, Although advances in polymer-based actuators have delivered unprecedented strengths,
1185–1188 (1988).
producing these devices at scale with tunable dimensions remains a challenge. We applied
3. D. L. Jeanmaire, R. P. Van Duyne, J. Am. Chem. Soc. 98,
4029–4033 (1976). a high-throughput iterative fiber-drawing technique to create strain-programmable
4. M. K. Scheller, R. N. Compton, L. S. Cederbaum, Science 270, artificial muscles with dimensions spanning three orders of magnitude. These fiber-based
1160–1166 (1995). actuators are thermally and optically controllable, can lift more than 650 times their own
5. F.-R. F. Fan, J. Kwak, A. J. Bard, J. Am. Chem. Soc. 118,
weight, and withstand strains of >1000%. Integration of conductive nanowire meshes
9669–9675 (1996).
6. T. Leoni et al., Phys. Rev. Lett. 106, 216103 (2011). within these fiber-based muscles offers piezoresistive strain feedback and demonstrates
7. I. Swart, T. Sonnleitner, J. Repp, Nano Lett. 11, 1580–1584 long-term resilience across >105 deformation cycles. The scalable dimensions of these
(2011). fiber-based actuators and their strength and responsiveness may extend their impact from
8. F. Mohn et al., Phys. Rev. Lett. 105, 266102 (2010).
engineering fields to biomedical applications.
9. W. Steurer, S. Fatayer, L. Gross, G. Meyer, Nat. Commun. 6,
8353 (2015).
L
10. S. Fatayer et al., Nat. Nanotechnol. 13, 376–380 (2018).
11. L. Gross et al., Science 324, 1428–1431 (2009). inear actuators simultaneously offering ing within their compartments to achieve high
12. R. Stomp et al., Phys. Rev. Lett. 94, 056802 (2005). high temporal responsiveness, power-to- power-to-mass ratios and strains (17–20). Pro-
13. E. Bussmann, C. C. Williams, Appl. Phys. Lett. 88, 263108
(2006). mass ratio, and strain and capable of op- ducing such structures at scale with tunable
14. L. Gross, F. Mohn, N. Moll, P. Liljeroth, G. Meyer, Science 325, erating across micrometer-to-centimeter dimensions, however, presents a challenge. Fur-
1110–1114 (2009). spatial scales are poised to advance the thermore, low-latency linear actuation without
15. F. J. Giessibl, Appl. Phys. Lett. 73, 3956–3958 (1998). fields of robotics, prosthetic limbs, and trans- secondary transduction and integration of feed-
16. J. Repp, G. Meyer, S. Paavilainen, F. E. Olsson, M. Persson,
Phys. Rev. Lett. 95, 225503 (2005). portation (1). Although actuators based on shape- back mechanisms within these fiber-based arti-
17. L. Gross et al., Science 337, 1326–1329 (2012). memory alloys (2), stimuli-responsive polymers ficial muscles has not been demonstrated (21).
18. M. Del Valle, R. Gutiérrez, C. Tejedor, G. Cuniberti, (3–9), and carbon composites (10–14) offer light- We reasoned that differential thermal ex-
Nat. Nanotechnol. 2, 176–179 (2007). weight, compact, and cost-effective alternatives pansion within polymer bimorph structures
19. K. Kumar et al., Nat. Commun. 7, 11975 (2016).
20. C.-W. Chang, Y.-C. Lu, T.-T. Wang, E. W.-G. Diau, J. Am. Chem. to traditional hydraulic, pneumatic, and servo comprising an elastomer and a thermoplastic
Soc. 126, 10109–10118 (2004). designs, their temporal responsiveness remains polymer (22–24) amplified by the tendril-like
21. M. J. Comstock et al., Phys. Rev. Lett. 99, 038301 (2007). limited (15, 16). Recent research in polymer and spring geometry (fig. S1) would allow for linear
22. B.-Y. Choi et al., Phys. Rev. Lett. 96, 156106 (2006). composite actuators has drawn inspiration from actuation at low thermal stimuli. To produce
23. M. Alemani et al., J. Am. Chem. Soc. 128, 14446–14447
(2006). cucumber tendrils that rely on differential swell- bimorph structures with arbitrary lengths and
24. P. Hapala et al., Phys. Rev. B 90, 085421 (2014). lateral dimensions ranging from micrometers
25. A. Riss et al., Nano Lett. 14, 2251–2255 (2014). 1 to millimeters, we relied on a scalable fiber-
Research Laboratory of Electronics, Massachusetts Institute of
26. J. H. Lupinski, K. D. Kopple, Science 146, 1038–1039 Technology (MIT), Cambridge, MA 02139, USA. 2Department drawing process (25). Thermal drawing enables
(1964). of Materials Science and Engineering, MIT, Cambridge,
27. T. Esat, N. Friedrich, F. S. Tautz, R. Temirov, Nature 558,
lateral size reduction of preforms by factors of
MA 02139, USA. 3Department of Electrical Engineering and
573–576 (2018). Computer Science, MIT, Cambridge, MA 02139, USA.
101 to 105 through controlled application of heat
28. A. Hoekstra, T. Spoelder, A. Vos, Acta Crystallogr. B 28, 14–25 4
John A. Paulson School of Engineering and Applied Sciences, and tension while simultaneously delivering
(1972). Harvard University, Cambridge, MA, USA. 5Department of meters to kilometers of fiber (23, 26, 27). Unlike
29. B. Milián, R. Pou-Amérigo, R. Viruela, E. Ortí, J. Mol. Struct. Mechanical Engineering, MIT, Cambridge, MA 02139, USA. other fiber fabrication techniques such as electro-
THEOCHEM 709, 97–102 (2004). 6
Paul G. Allen School of Computer Science and Engineering,
30. J. H. Dawson, Science 240, 433–439 (1988). University of Washington, Seattle, WA 98195, USA.
and jet-spinning, thermal drawing is applicable
31. J. I. Wu, I. Fernández, P. V. R. Schleyer, J. Am. Chem. Soc. 135, *These authors contributed equally to this work. to multiple materials with a diversity of non-
315–321 (2013). †Corresponding author. Email: anikeeva@mit.edu cylindrical geometries (27). To be compatible

with thermal drawing, the layers of the bimorph cladding (35 mm by 26 mm outer cross-sectional mechanically stripped to release the bimorph
must, however, be composed of materials with dimensions) (Fig. 1A and fig. S3). Prior to draw- structures (Fig. 1C). The tension (70 to 100 mN)
similar viscosities at the drawing temperature. ing, the preforms were annealed under pressure experienced by the fibers during drawing led to
Simultaneously, maximizing the differences in of 50 bar and temperature of 125°C to promote the formation of a spring-like shape upon PMMA
thermal expansion coefficients (a) is necessary to adhesion between the PE and COCe layers. By cladding removal (coil diameter ~30 to 40 mm,
achieve robust actuation. Guided by these con- setting the drawing temperature (290° to 310°C) fig. S6).
straints, a high-density polyethylene (PE, melting and the relative feed (vf = 1 mm/min) and draw To produce actuated springs via strain pro-
temperature Tm = 120°C, a = 1.3 × 10−4 K–1) and (vd = 2 to 3 m/min) speeds, the cross-sectional gramming from the bimorph fibers, these struc-
cyclic olefin copolymer elastomer (COCe, melting area of the fiber was tuned between 50 mm by tures were cold drawn at strains of 50 to 1300%,
temperature Tm = 84°C, a = 2.6 × 10−5 K–1) were 35 mm and 5 mm by 3.5 mm over a 500-m length which induced plastic deformation in PE (Fig. 1D).
chosen as the constituents of the bimorph fibers (Fig. 1B). To further reduce the lateral dimen- Upon release, the elastomeric COCe component
(table S1). Finite element analysis was then ap- sions, the fibers produced by the first drawing attempted to contract to its original dimensions,
plied to select a cross-sectional geometry that step were stacked within another PMMA pre- and the resulting stress in the bimorph struc-
would optimize the thermal responsiveness of form and drawn at similar conditions to produce ture induced the formation of tendril-like springs
the bimorphs (fig. S2 and table S2). The final COCe-PE bimorph elements with cross-sectional (Fig. 1E and movies S1 and S2). The high pre-
design of the bimorph fibers comprised iden- areas as low as 13 mm by 8 mm (Fig. 1A and fig. strains between ~700 and 1300% were only tol-
tical PE and COCe layers with rectangular cross S4). The thermomechanical mismatch between erated by a subset of fibers, likely due to the
sections (25). COCe and PE set an upper limit of processing manufacturing defects. The diameters of springs
These fibers were then drawn from preforms 20 first-step fibers for every second-step PMMA scaled with the cross-sectional dimensions of the
composed of COCe and high-density PE blocks preform. Increasing the number of fibers re- precursor fibers, and a variety of springs could
(25 mm by 8 mm cross section, 200 mm long) sulted in contraction failure of the preform be obtained from the first and second iterative
joined within a poly(methyl methacrylate) (PMMA) (fig. S5). After drawing, the PMMA cladding was draws (Fig. 1, F and G) (25).
Fig. 1. Fabrication and morphological characterization (A) Bimorph (E) Cold drawing process for obtaining a spring actuator. Steps 1 to 4 show
fibers produced via two-step thermal drawing. HDPE, high density PE. the stretching process. Upon release, shown in steps 5 to 8, the fibers
(B) Photograph of ~60 m of PMMA-encapsulated bimorph fibers. formed springs. (F and G) Micrographs and photographs of the first-step and
(C) Illustration of the PMMA cladding removal. (D) Cross-sectional second-step artificial muscles [scale bars in (F): 200 mm]. (H) Fiber-based
scanning electron microscope (SEM) images of a fiber before and after muscle contracting in response to a temperature increase of DT = 14°C
cold drawing. Inset: COCe structure after stretching (scale bar: 20 mm). (spring index: k = 4.70, 6 turns/cm).

The fivefold mismatch in coefficients of the Darwin (28). The stochastic formation of these of PE and COCe alone exhibited stress-strain
thermal expansion between the COCe and PE bifurcations is predicted by applying the Kirchhoff characteristics of a tough plastic and an elasto-
enabled rapid thermal actuation. Upon tem- theory for thin rods to the fibers with elliptical mer, respectively, the bimorph fibers exhibited
perature increase, the PE underwent greater cross sections (figs. S7 to S9) (19, 25). Although intermediate behavior (Fig. 2A) (23, 29). Increas-
thermal expansion relative to COCe, which these structures do not substantially impede the ing the deformation rate from 10 to 50 mm/min
increased the tensile strain in COCe and induced thermal actuation performance, their formation during the cold drawing of the bimorph fibers
further tightening of the springs (fig. S7) (25). could, at least in part, be circumvented by simul- allowed for greater built-in strain within the
Figure 1H demonstrates a 50% linear contrac- taneously releasing the ends of the fibers after resulting artificial muscles. Further increasing the
tion in a fiber actuator with a 0.64 mm by 1 mm the application of prestrain (25). deformation rate, however, reduced the amount
cross-sectional area in response to a temperature To optimize the thermal actuation perform- of achievable strain (Fig. 2A). This was consistent
increase of DT = 14°C applied over 4 s. ance, we investigated the effects of cold drawing with an increase in the yield strength and the
We observed the spontaneous formation of parameters applied to COCe-PE bimorph fibers elastic modulus in thermoplastic materials with
unstable bifurcations, which are also found in on the mechanical properties of the resulting increasing deformation rate (29, 30). For a given
cucumber tendrils and termed “perversions” by fiber-based muscles. Whereas fibers composed fiber, the spring diameter and the spring index of
Fig. 2. Mechanical characterization of fiber-based artificial muscle. optical heat source was replaced with a micro-Peltier heater, and a force
(A) Stress-strain curves recorded at different extension rates for gauge had a higher resolution. (H) Temperature and force responses to
precursor fibers (without PMMA cladding) with a cross-sectional area of photothermal pulses collected for a fiber with 300 mm by 470 mm cross
300 mm 470 mm. (B) Change in the spring diameter with respect to the section (k = 5, 8 turns/cm). (I) Generated force versus the temperature
deformation rate. (C) Spring diameter and number of coils versus applied difference (number of cycles 300) for fibers with 200 mm by 312 mm
prestrain. Error bars and shaded areas represent average and standard cross section (k = 5, 10 turns/m). Navy blue, dark blue, cyan, and green
deviation, respectively. The number of samples n = 5. (D) Change in data point clusters represent different temperature ranges of 3.32 ±
the spring index with respect to spring diameter and actuation strain. 0.26°C, 5.28 ± 0.37°C, 9.74 ± 0.34°C, and 12.98 ± 0.74°C, respectively.
(E) Change in the actuation stress-strain (gray) and work capacity (green). (J) Temperature and force responses to thermal pulses for a fiber with an
Attributes of the fibers are k = 6, 6 turns/cm (continuous line); k = 5, 8 mm by 12.5 mm cross section (k = 4.6, 60 turns/cm). (K) Change in
10 turns/cm (dashed line); k = 5.5, 12 turns/cm (dotted line). (F) Change the efficiency with respect to actuation strain (blue line k = 6, 6 turns/cm;
in the residual stress with respect to the cross-sectional area. (G) The green line k = 5, 10 turns/cm; gray line k = 5.5, 12 turns/cm). (L) Force
setup used for the force measurement of the fibers with 300 mm by measured across 300 thermal actuation cycles applied over three con-
470 mm cross section (scale bar, 5 mm). For the fibers with an 8 mm by secutive days for a fiber with 200 mm by 312 mm cross section (k = 5,
12.5 mm cross section (scale bar, 200 mm), a similar setup was used. The 10 turns/cm). Inset: A single actuation cycle.

Fig. 3. Electrical feedback in fiber-based artificial muscle. (A) A schematic illustration of the setup for the resistance measurements for fibers with
300 mm by 470 mm cross-sectional area (k = 5, 8 turns/cm). Inset: A SEM image of silver nanowire mesh on the surface of a 300 mm by 470 mm2
fiber-based muscle. (B) Resistance waveforms collected at different numbers of extension and release cycles. (C) The change in the extended and
released resistance for ~12,000 cycles. (D) Hysteresis curve showing the resistance versus the applied strain for a single cycle of deformation. The lines
represent the average and error bars represent the standard deviation.
Fig. 4. Thermal actuation of fiber-based muscles. (A) Schematic dashed lines represent average and average ± 1 SD, respectively. (F) The
illustration of the vertical lift experiment, where t is time, m is mass, and maximum displacement for fiber bundles loaded with weights 1, 2, 3, 4, 5, 10,
Dx is displacement. (B) A photographic time series collected during a 50, 100, and 200 g. Error bars represent ± 1 SD. (G) A printed model of a
displacement experiment in (A). The heat was applied in 2-s pulses weight-lifting artificial limb, inspired by a human arm. (H) A photographic time
separated by 6-s rest epochs. The load mass is 1 g. (C) Waveforms for the series of the artificial limb lifting a 1-g load. The heat was applied with a heat
heat pulses (top), the corresponding change in the temperature at the fiber gun for 2 s and then followed by 5-s rest epochs. [(A) to (G)] Fiber cross-
surface (middle), and the displacement of the 1-g load (bottom). (D) The sectional area is 300 mm by 470 mm, spring index k = 5, and the number of
vertical displacement Dx of a 1-g load in response to temperature increase DT. turns per cm is 8. (I) The change in the fiber length, a piezoresistive strain
(E) Strain measurement across 100 cycles of thermal actuation. Solid and feedback signal, and the angle of the arm for the experiment in (H).

the actuators formed by cold drawing were con- marized in table S3. The devices (5 cm long, relaxed upon cooling to room temperature (Fig.
trolled by the deformation rate and the maxi- cross-sectional area 300 mm by 470 mm) were 4H and movie S5). The change in an angle be-
mum applied strain (Fig. 2, B to D). Consistent subjected to 300 cycles of thermal actuation tween the forearm and humerus during opera-
with the maximum allowable strain observed at over three consecutive days (DT = 13°C for 6 s, tion correlated linearly with the fiber muscle
50 mm/min for fibers with 300 mm by 470 mm followed by 30 s of rest, Fig. 2L), and no de- contraction and extension (Fig. 4I). A miniature
cross-sectional area, the spring diameters reached cline in performance was observed. arm was used to lift a 1-g dumbbell, and the
their minimum at this deformation rate (Fig. 2B). Fiber-based muscles were outfitted with con- platform was scalable to a larger limb with
At a fixed deformation rate, increasing the strain ductive meshes of silver nanowires (AgNW, greater weightlifting performance of 2 g afforded
applied to bimorph fibers during cold drawing diameter = 70 nm and length = 50 mm) to en- by two parallel fibers (movie S6).
produced actuators with smaller spring diame- able monitoring of their contraction and elon- This work presents a scalable strategy to
ters and a greater number of turns per centi- gation. These meshes were deposited directly produce fiber-based actuators with lateral di-
meter (Fig. 2C). Spring index (k) of the actuators onto the surfaces of the fiber muscles after the mensions ranging from millimeters to micro-
increased with the spring diameter and the cold drawing process of the bimorph fibers, meters and arbitrary lengths. Cold drawing
actuation strain (Fig. 2D), resulting in higher which was followed by the deposition of a pro- bimorph fibers composed of PE and COCe formed
work capacity (Fig. 2E). The latter was limited tective stretchable layer of polydimethylsiloxane springs with the spring index and residual stress
by the coil-to-coil contact at the highest actua- (PDMS) elastomer (23). Because percolation with- determined by the fiber cross-sectional dimen-
tion temperature. An increase in the prestrain in the AgNW mesh changes with the contraction sions, and the applied strain and deformation
decreased the spring diameter, and thus the spring or elongation of the underlying substrates, AgNW- rate. The mismatch in the thermomechanical
index. Decreasing cross-sectional dimensions coated fiber muscles act as piezoresistive sensors properties enabled reversible and repeatable
of the fibers yielded an increase in residual stress of deformation in response to stimuli. To eval- thermal actuation. Conceptual demonstrations
in the springs formed after the cold drawing pro- uate this sensing ability, one end of the fiber of load bearing and joint manipulation facili-
cess (Fig. 2F). muscles was connected to a direct current (DC) tated by the fiber-based artificial muscles indi-
The force generated by the cold-drawn (maxi- motor, and the resistance change was recorded cate the possibility of their future applications
mum strain of 700%, 50 mm/min deformation by a voltage divider (Fig. 3A). A fractional resist- in robotics and prosthetic limb technologies.
rate) fiber-based muscles with cross-sectional ance change of 0.47% was repeatable across
areas of 300 mm by 470 mm was characterized by ~12,000 cycles of 20% elongation (Fig. 3, B to REFERENCES AND NOTES
connecting the devices to a force gauge within a D). Although a slow increase in baseline resist- 1. M. Zupan, M. F. Ashby, N. A. Fleck, Adv. Eng. Mater. 4,
933–940 (2002).
custom-designed characterization setup (Fig. 2G ance was observed (Fig. 3, B and C), the relative 2. S. M. Mirvakili, I. W. Hunter, ACS Appl. Mater. Interfaces 9,
and fig. S10). A modulated broadband light sup- change between extension and release cycles 16321–16326 (2017).
plied thermal stimuli, and a thermistor was placed remained stable (Fig. 3D). 3. A. Maziz et al., Sci. Adv. 3, e1600327 (2017).
near the fiber muscles to monitor the change A single 5-cm-long fiber muscle with a cross- 4. A. Lendlein, R. Langer, Science 296, 1673–1676 (2002).
5. J. Park et al., Smart Mater. Struct. 26, 035048 (2017).
in the temperature concurrent with the gen- sectional area of 300 mm by 470 mm was able to 6. C. S. Haines et al., Proc. Natl. Acad. Sci. U.S.A. 113,
erated force measurements. Three-second illu- lift a 1-g weight by 5.12 ± 0.76 mm (12% strain) 11709–11716 (2016).
mination pulses separated by 10-s rest periods in response to a thermal stimulus of DT = 10°C 7. C. S. Haines et al., Science 343, 868–872 (2014).
induced temperature gradients of 3.45 ± 0.43°C/s (from room temperature) delivered by a heat 8. O. Kim, T. J. Shin, M. J. Park, Nat. Commun. 4, 2208
(2013).
(n = 6 cycles), causing a force of 36.23 ± 5.42 mN gun (Fig. 4, A and B, and movie S3). This be- 9. A. Lendlein, H. Jiang, O. Jünger, R. Langer, Nature 434,
in 5-cm-long fiber muscles (Fig. 2H). An actua- havior was reversible and repeatable across 879–882 (2005).
tion rate of 13.25 ± 1.66 N/s for a temperature multiple cycles of 2-s heat pulses separated by 10. K.-Y. Chun et al., Nat. Commun. 5, 3322 (2014).
increase rate of 1.11 ± 0.12°C/s and a power-to- 6-s rest periods, during which the fiber muscle 11. M. D. Lima et al., Science 338, 928–932 (2012).
12. J. Deng et al., Nat. Protoc. 12, 1349–1358 (2017).
mass ratio of 75 W kg−1 surpassed the average cooled down to room temperature (Fig. 4C). The 13. S. H. Kim et al., Sci. Rep. 6, 23016 (2016).
power-to-mass ratio of human muscle (50 W kg−1). vertical displacement was linearly correlated 14. P. Chen et al., Nat. Nanotechnol. 10, 1077–1083 (2015).
Increasing the exposure times led to a propor- with the thermal gradient (Fig. 4D). Perma- 15. T. Mirfakhrai, J. D. W. Madden, R. H. Baughman, Mater. Today
tionate increase in both heating and generated nent deformation was observed for temper- 10, 30–38 (2007).
16. G.-Z. Yang et al., Sci. Robot. 3, eaar7650 (2018).
force (Fig. 2I). No decline in performance was ature gradients DT ≥ 50°C, consistent with the 17. M. H. Godinho, J. P. Canejo, G. Feio, E. M. Terentjev,
observed across multiple actuation cycles at dif- thermomechanical properties of PE and COCe. Soft Matter 6, 5965–5970 (2010).
ferent temperatures (Fig. 2I). Artificial muscles Repeated application of 100 cycles of ther- 18. M. H. Godinho, J. P. Canejo, L. F. V. Pinto, J. P. Borges,
produced from the miniature bimorph fibers mal actuation (DT = 8.8 ± 0.6°C) delivered re- P. I. C. Teixeira, Soft Matter 5, 2772–2776 (2009).
19. S. J. Gerbode, J. R. Puzey, A. G. McCormick,
with cross-sectional areas of 8 mm by 12.5 mm producible actuation strains (10.1 ± 1.5%, Fig. L. Mahadevan; L. M. Sharon J. Gerbode, Science 337,
fabricated by the two-step thermal drawing 4E). Although an additive boost in strength 1087–1091 (2012).
were similarly evaluated for their actuator per- was afforded by bundling multiple fibers in 20. Y. Cheng et al., ACS Nano 12, 3898–3907 (2018).
formance (Fig. 2, G and J, and figs. S11 and S12). an oblique fashion, no observable change in 21. K. Jung, K. J. Kim, H. R. Choi, Sens. Actuators A Phys. 143,
343–351 (2008).
A micro-Peltier was used to apply a thermal stretchability was found (Fig. 4F, fig. S12, and 22. M. Shahinpoor, K. J. Kim, Smart Mater. Struct. 10, 819–833
stimulus, and a thermistor was placed near the movie S4). (2001).
fiber muscles to monitor the change in the tem- To further illustrate the potential application 23. C. Lu et al., Sci. Adv. 3, e1600955 (2017).
perature concurrent with the generated force of the fiber-based muscle as a model of biological 24. M. Amjadi, M. Sitti, ACS Nano 10, 10202–10210 (2016).
25. Materials, methods, and additional information are available as
measurement (fig. S13). A temperature gradient muscle, we designed and printed a weight-lifting supplementary materials.
of 11.09 ± 0.55°C/s (n = 10 cycles) produced a artificial limb inspired by a human arm (Fig. 4G). 26. M. Yaman et al., Nat. Mater. 10, 494–501 (2011).
force of 371 ± 40.7 mN in these 5-mm-long fiber One end of the fiber muscle (300 mm by 470 mm 27. S. Shabahang et al., Nature 534, 529–533 (2016).
muscles. The actuation rate of the miniature cross section) was then affixed to the radius and 28. C. Darwin, On the Movements and Habits of Climbing Plants
(London, 1865).
fiber muscle was 6.33 ± 0.72 N/s for a temper- ulna of the model forearm, while the other end 29. D. Kuhn, Howard Medlin, in ASM Handbook, vol. 8, Mechanical
ature increase rate of 130.3 ± 16.9°C/s, and the was affixed to the clavicle. As found in the hu- Testing and Evaluation (2000), pp. 26–48.
power-to-mass ratio was calculated as 90 W kg−1. man arm, the joint between the humerus and the 30. Y. Isogai et al., Polym. Eng. Sci. 58, E151–E157 (2018).
The power efficiency of the fiber-based muscles forearm was aligned with the clavicular muscle 31. E. P. Wismaier, H. Raff, K. T. Strang, Vander’s Human Physicology:
The Mechanisms of Body Function (McGraw-Hill, 2013).
was found to increase with actuation strain and connection slot to maximize the displacement
was higher for the fibers with larger cross- corresponding to a given bicep contraction (31). AC KNOWLED GME NTS
sectional areas (Fig. 2K). Actuator performance The artificial bicep was actuated by 2-s heat M. K. thanks D. Rhodes for the cucumber tendril image;
attributes of the fiber-based muscles are sum- pulses (DT = 10°C) delivered by a heat gun, and Z. Hazar Kanik for help in bioinspired artificial limb design; and

M. Tarkanian, D. Bono, S. Ayas, and S. Zearott for insightful conducted mechanical characterization. M.K. and G.V. designed quantities of physical samples of the described fibers are
discussions and comments on the manuscript. S.O. thanks Y. Terzioglu the theoretical model. M.K., S.O., and T.B. conducted silver available upon request.
for advice on electrical characterization. M.K. and P.A. thank Edgerton nanowire coating and resistance measurements. M.K., S.O.,
Center at MIT for assistance in photography. Funding: This work and J.K. characterized artificial muscle fibers. C.C.T. provided SUPPLEMENTARY MATERIALS
was supported in part by the National Institute of Neurological insight into the mechanical characterization of microscale fibers.
science.sciencemag.org/content/365/6449/145/suppl/DC1
Disorders and Stroke (5R01NS086804), National Science M.K. and S.O. designed all custom-made experimental apparatuses
Foundation (NSF) Center for Materials Science and Engineering used in the article. A.P.C. aided in the design of custom-designed
Supplementary Text
(DMR-1419807), and NSF Center for Neurotechnology (EEC- apparatuses for electrical characterization. M.K., S.O., and P.A.
Figs. S1 to S13
1028725). M.K. is a recipient of Simons Postdoctoral Fellowship. analyzed the data. All authors contributed to the writing of the
Tables S1 to S3
S.O. is supported in part by Delta Electronics, Inc. Author manuscript. Competing interests: M.K. and P.A. have applied for
contributions: M.K. and P.A. designed the study. M.K. and P.A. U.S. patent (Application no.16/427,540) related to the technology
Movies S1 to S6
designed the fibers. M.K. and T.A. designed the numerical models described in the manuscript. Data and materials availability:
for optimization. M.K., S.O., and D.G. fabricated preforms and All data needed to evaluate the conclusions in the paper are 2 December 2018; accepted 12 June 2019
fibers. Y.F. provided insight into fiber design. M.K. and D.G. present in the paper or the supplementary materials. Reasonable 10.1126/science.aaw2502
ARTIFICIAL MUSCLES yarn length and the nanotube direction). Be-

cause this angle decreases to zero on going from
Sheath-run artificial muscles

the yarn surface to the yarn center, the input en-
ergy delivered to the guest near the yarn center
is not effectively used. Second, muscle mechan-
ical power is limited by the chemical or thermal
Jiuke Mu1, Mônica Jung de Andrade1, Shaoli Fang1, Xuemin Wang2,3, Enlai Gao1,4, transport times to access yarn volume.
Na Li1,5, Shi Hyeong Kim1, Hongzhi Wang6, Chengyi Hou6, Qinghong Zhang6, Here, we describe a new muscle structure that
Meifang Zhu6, Dong Qian2, Hongbing Lu2, Dharshika Kongahage7, Sepehr Talebian7, addresses each of these problems. Rather than
Javad Foroughi7, Geoffrey Spinks7, Hyun Kim8, Taylor H. Ware8, Hyeon Jun Sim9, infiltrating a volume-changing yarn guest within
Dong Yeop Lee9, Yongwoo Jang9, Seon Jeong Kim9, Ray H. Baughman1* a yarn, such as for a HYAM, this guest is coated
as a yarn sheath. Because the dimensional and
Although guest-filled carbon nanotube yarns provide record performance as torsional and modulus changes of this sheath drive actuation,
tensile artificial muscles, they are expensive, and only part of the muscle effectively contributes we call the resulting actuators “sheath-run artifi-
to actuation. We describe a muscle type that provides higher performance, in which the cial muscles” (SRAMs).
guest that drives actuation is a sheath on a twisted or coiled core that can be an inexpensive CNTs were drawn as a sheet from a CNT forest
yarn. This change from guest-filled to sheath-run artificial muscles increases the maximum and twisted into the Archimedian yarn (fig. S1)
work capacity by factors of 1.70 to 2.15 for tensile muscles driven electrothermally or by (25, 26) used as muscle core. SRAMs were fabri-
vapor absorption. A sheath-run electrochemical muscle generates 1.98 watts per gram of cated (Fig. 1A) by drawing a vertically suspended,
average contractile power—40 times that for human muscle and 9.0 times that of the torsionally tethered twisted yarn through a large
highest power alternative electrochemical muscle.Theory predicts the observed performance droplet of polymer solution multiple times to
advantages of sheath-run muscles. achieve the targetted sheath thickness of dried
R
polymer. The solvent used was chosen to avoid
emarkable performance has been obtained CNT hybrid yarn artificial muscles (HYAMs) polymer infiltration into the twist-densified core
for tensile and torsional carbon nanotube can be made by inserting twist, or twist and yarn and provide a sharp interface between
(CNT) hybrid yarn muscles (1–5), whose coiling, into a guest-filled CNT yarn. Muscles sheath and core (Fig. 1C and figs. S2A, S3, and
actuation is driven by the volume change that are twisted (but not coiled), called twisted S7, C to F). Scanning electron microscope (SEM)
of a guest within a twisted or coiled CNT muscles, are mainly useful for torsional actua- measurements provided the sheath/core ratio
yarn. During thermally powered contraction, tion. High inserted twist results in coiled muscles (SCR; the ratio of sheath thickness to the interior
coiled hybrid muscles can deliver 29 times the that can deliver larger strokes than can human yarn diameter). To demonstrate that CNT yarns
work as the same weight human muscle (1). muscles (1). can be replaced with inexpensive yarns, we evalu-
Changing the structural relationship between Polymer fiber and yarn muscles are known ated commercial nylon 6, silk, and bamboo yarns
guest and host will provide major performance (6–11) that operate similarly to CNT HYAMs: as the muscle core as well as electrospun poly-
increases and the ability to replace expensive Muscle volume expansion drives muscle untwist, acrylonitrile (PAN) nanofibers.
CNT yarn with cheap commercialized yarns. which produces both torsional and tensile actu- The nomenclature used for a sheath X on a
ation. These thermally driven polymer muscles yarn core Y of a SRAM or an X guest inside a
are cheap because they can be made by inserting HYAM yarn Y is X@Y. Hence, PEO-SO3@CNT
1
Alan G. MacDiarmid NanoTech Institute, The University of extreme twist into fishing line or sewing thread. denotes a PEO-SO3 guest and a CNT yarn host,
Texas at Dallas, Richardson, TX 75080, USA. 2Department of
Mechanical Engineering, The University of Texas at Dallas,
Other twisted or coiled materials have been ex- where PEO-SO3 is a blend of poly(ethylene oxide)
Richardson, TX 75080, USA. 3Department of Mechanical ploited as fiber-like muscles, such as graphene and a copolymer of tetrafluoroethylene and sul-
Engineering, Georgia Southern University, Statesboro, GA oxide fiber (12), shape memory polymer fiber or fonyl fluoride vinyl ether (26). The yarn-bias-
30458, USA. 4Department of Engineering Mechanics, School metal alloy yarn (13, 14), cotton yarn composites angle dependence of the minimum SCR needed
of Civil Engineering, Wuhan University, Wuhan, Hubei
430072, China. 5Materials Science, MilliporeSigma,
(15), carbon fiber/polydimethylsiloxane yarn to prevent sheath cracking for a PEO-SO3@CNT
Milwaukee, WI 53209, USA. 6State Key Laboratory for (16), neat CNT yarns (1, 17–20), and spider-silk yarn is shown in fig. S7; this ratio approximately
Modification of Chemical Fibers and Polymer Materials, dragline (21). CNT HYAMs are especially useful maximizes torsional stroke for the high targeted
College of Material Science and Engineering, Donghua because guest choice results in muscles driven yarn bias angle. For comparative studies, the guest/
University, Shanghai 201620, China. 7Intelligent Polymer
Research Institute, Australian Institute for Innovative
thermally (1, 4), electrochemically (22, 23), or by host weight ratio was essentially the same for the
Materials, University of Wollongong, Wollongong, New South absorption (2, 3, 24). SRAM and HYAM, and the same mechanical load
Wales 2522, Australia. 8Department of Bioengineering, The The challenge is to develop a fundamentally was applied during twist insertion. HYAMs were
University of Texas at Dallas, Richardson, TX 75080, USA. new host-guest structure that eliminates the made by using the above droplet method by add-
9
Center for Self-Powered Actuation, Department of
Biomedical Engineering, Hanyang University, Seoul 04763,
liabilities of CNT HYAMs. First, the ability of ing polymer solution to a low-twist yarn, partially
South Korea. guest expansion to drive yarn untwist depends drying the solution to a gel-like state and then
*Corresponding author. Email: ray.baughman@utdallas.edu on the yarn’s bias angle (the angle between the adding additional twist to equal that of the SRAM.

If the guest/host weight ratio is too high for a softening, which combine to partially release elas- SO3 @CNT SRAM, the torsional strokes of a
HYAM (26), the guest will extrude from the host tically stored torsional energy in the core yarn. In PEO-SO3@PAN SRAM and a PEO-SO 3 @silk
yarn during twist insertion (fig. S12B). Fig. 2E, we compare the observed and predicted SRAM are predicted to be close to those for PEO-
“Self-coiled” yarn was fabricated by inserting dependence of torsional stroke on SCR for an SO3@CNT SRAMs that have the same core bias
further twist while the guest was in the gel state ethanol-driven PEO-SO3@CNT SRAM made from angle and diameter (26).
(Fig. 1B). To increase yarn stroke by increasing a 42°-bias-angle yarn. The maximum torsional All measurements show that a SRAM has im-
the spring index, twisted yarns or self-coiled yarns stroke (143°/mm) occurs for a SCR of 0.14, which portant performance advantages over the cor-
(Fig. 1, D and E) were coiled or supercoiled by agrees with the predicted 151°/mm stroke maxi- responding HYAM as a torsional actuator. The
wrapping around a mandrel. Afterward, the coiled mum for a SCR of 0.15. A much lower SCR cannot ratios of peak torsional speed of the SRAM to
yarn was thermally annealed (26). When describ- maintain the initially inserted twist before actua- that of the corresponding HYAM are nearly the
ing a muscle, the diameter is for the dry, twisted tion, and a much higher SCR ratio hinders twist same for PEO-SO3@CNT (1.75), PEO-SO3@silk
muscle before coiling. Unless otherwise described, release during actuation. As shown in fig. S6, the (1.74), and PEO-SO3@PAN (1.79) muscles pow-
gravimetric work and power densities are nor- torsional stroke is near maximum for yarn bias ered by ethanol vapor and close to that for water
malized to the weight of the dry muscle. The angles from 38° to 43° for a PEO-SO3@CNT SRAM vapor–powered nylon6@CNT muscles (1.86)
spring index is the ratio of the difference in outer that has a sheath/core weight ratio of 0.53. (Fig. 2, B and F, and figs. S2 and S9). However,
coil diameter and the fiber diameter to the fiber Torsional stroke is sensitive to vapor type (fig. greater variation arises in the ratios of peak stroke
diameter, where a fiber’s diameter is its width in S8) and the sheath’s ability to swell and soften for the SRAM to that of the HYAM (1.86, 1.67, 1.36,
its largest lateral dimension. by vapor absorption. Because ethanol produces and 1.63, respectively, for the above).
Torsional actuation of a one-end-tethered SRAM a much larger equilibrium volume expansion in By adding sufficient additional twist to the
is illustrated in Fig. 2A. Unless otherwise noted, PEO-SO3 (16.7%) than in polyvinyl alcohol (PVA) muscles used for torsional actuation, fully coiled
an equilibrium vapor pressure was delivered to (1.3%) or nylon 6 (0.5%) (fig. S5A), the torsional yarn muscles result that provide large strokes.
muscles in flowing dry air and then removed stroke of a CNT core SRAM was much larger By comparing the performance of coiled muscles
under vacuum, using the glass tube system of for a PEO-SO3 sheath (143°/mm) than for PVA made from yarns with nearly the same host and
Fig. 2B. For performance comparisons, a 60-mg- (22°/mm) or nylon 6 sheaths (11°/mm) (fig. S5B). guest weight per yarn length, we demonstrate
weight paddle at the yarn end, with 0.28 kg·mm2 High performance resulted for ethanol-powered the increases in stroke, stroke rate, contractile
moment of inertia, was used to characterize tor- torsional SRAMs in which the expensive CNT yarn work, and contractile power that results from
sional rotation angle and speed. Also, the SRAMs was replaced with a silk or electrospun PAN yarn transitioning from a HYAM to a SRAM.
and HYAMs were made from identical yarn, con- (Fig. 2F and fig. S2B). The bias angles of these The torsional rotor was replaced with a heavy,
tained the same inserted twist, and had nearly SRAMs (30° and 18°, respectively) are lower than nonrotating weight when changing from tor-
the same host/guest weight ratio. for the CNT yarn core SRAM (42°) because higher sional to tensile actuation. Allowing weight
In Fig. 2B, we compare the time dependence of twist broke the yarns. The lower bias angles and rotation decreases tensile contraction (fig. S10)
paddle rotation and speed for a PEO-SO3@CNT larger diameters of the PAN and silk core yarns because yarn untwist increases muscle length.
SRAM and HYAM and a pristine CNT muscle that provided smaller equilibrium torsional strokes When ethanol vapor–driven, a PEO-SO3@CNT
are undergoing one complete reversible cycle of (123 and 70°/mm, respectively) than for the PEO- SRAM delivered a higher stroke for all loads
ethanol vapor–powered actuation. The peak SO3@CNT SRAM (143°/mm). However, using and times than did a HYAM (Fig. 3, A and B, and
stroke and peak rotation speed for the SRAM the invariance of the product of torsional stroke fig. S11). One thousand fully reversible cycles
[143°/mm and 507 rotations per minute (rpm)] and yarn diameter when the yarn’s bias angle were demonstrated. Corresponding SRAM struc-
are about twice that for the HYAM (76°/mm and is constant (1) and results in fig. S6 for the bias- ture changes during 0.1 Hz actuation to provide
254 rpm) and much larger than for the pristine angle-dependence of torsional stroke for a PEO- 8.5% stroke are shown in movie S1. As shown in
yarn (4.7°/mm and 36 rpm). Steady-state mea-
surements of torsional stroke versus weight %
(wt %) ethanol in the muscles (Fig. 2C) show that
the ratio of torsional strokes for a PEO-SO3@CNT
SRAM to a PEO-SO3@CNT HYAM peaks at 6.7 for
4.1 wt % ethanol and then gradually decreases to
1.7 for 17.5 wt % ethanol. The small hysteresis in
torsional strokes for the SRAM and HYAM means
that both could reliably open and close valves in
response to absorbed vapor. However, the tor-
sional stroke of the SRAM is much more sen-
sitive to the amount of absorbed ethanol than
the HYAM.
As shown in Fig. 2D, PEO-SO3@CNT SRAMs
and HYAMs reversibly actuate over 3000 cycles
of ethanol absorption and desorption, despite
the absence of tethering. This reversibility results
because the guest acts as a torsional return spring.
By contrast, the torsional stroke of the pristine
yarn rapidly decreases from 27°/mm for the first
cycle to ~4.7°/mm on the 27th cycle, thereafter
stabilizing at this value for the next ~3000 cycles. Fig. 1. Muscle fabrication and structure for torsional and tensile actuation. (A) Schematic
Our theoretical model (26) predicts the depen- lateral and cross-sectional views of a twisted CNT yarn and a SRAM, made by coating a twisted CNT
dence of torsional stroke on the SCR by using the yarn with a polymer sheath. (B to E) SEM micrographs for PEO-SO3@CNT muscles. (B) A SRAM
torque balance between sheath and core, before made by self-coiling a sheath-coated twisted yarn. (C) The surface of a twisted SRAM, which was
and after actuation. This analysis captures the broken by untwisting in liquid N2, showing the distinct boundary between sheath polymer and CNT
two primary mechanistic contributions to SRAM core. (D) A mandrel-coiled twisted SRAM. (E) A SRAM that was self-coiled and then supercoiled
torsional actuation: sheath swelling and sheath around a mandrel. Scale bars, (B) to (E), 35, 15, 200, and 200 mm, respectively.

fig. S12A, the equilibrium isometric contractile SRAM, and this SRAM provided faster contrac- 1.51 W/g for the corresponding HYAM. The load-
stress generated by a PEO-SO3@CNT SRAM tion than that of cylindrical-mandrel SRAMs that optimized contractile work capacity and the max-
monotonically increases with increasing ethanol were coiled and supercoiled (fig. S13). imum average power density of coiled SRAMs
vapor concentration. By contrast, if the applied The SRAMs provide advantages in contractile are higher than for coiled HYAMs at all applied
load is low and the change in sheath thickness work capacity and maximum average contractile loads for sorption-driven, electrothermal, and
is large, the SRAM first contracts until intercoil power (Fig. 3B, figs. S14 to 16 and S19, and table electrochemical actuation (table S2). For loads
contact occurs and then expands as intercoil con- S2), which is the maximum ratio of contractile maximizing equilibrium contractile work capa-
tact drives actuation (fig. S12C). Mandrel coiling work to actuation time. The maximum average cities, the SRAM-to-HYAM work capacity ratio
greatly amplifies muscle stroke. A 70% tensile stroke contractile power output was 4.44 W/g for the was 1.84 for ethanol vapor–driven PEO-SO3@CNT
was obtained for a humidity-driven cone-mandrel ethanol vapor–driven PEO-SO3@CNT SRAM and muscles (Fig. 3B), 1.73 for electrothermally driven
PEO-SO3@CNT muscles (Fig. 3D), and 2.15 for
electrothermally driven PU@CNT muscles (Fig.
4D), where PU is an elastomeric polyurethane
(26). These SRAM-to-HYAM work capacity ratios
will approximately equal the ratio of energy con-
version efficiencies for sorption-powered muscles
in which the equilibrium gravimetric sorption
of guest in SRAM sheath and in HYAM core are
equal, and for thermal muscles in which the dif-
ferences in heat lost during high-rate contractile
work are negligible.
The SRAM-to-HYAM power density ratio
(table S2) is higher for ethanol vapor–driven
PEO-SO3@CNT muscles (2.94) than for electro-
thermally driven PEO-SO3@CNT muscles (1.69)
and PU@CNT muscles (2.06). This is likely be-
cause the power density ratio for the vapor-
driven muscle is enhanced by both the larger
equilibrium work capacity of the SRAM and
the more rapid vapor absorption, and the latter
diffusion-based enhancement term disappears
when actuation is from electrothermally heat-
ing the CNT yarn.
Because the rate of cooling is faster for the
SRAM than for the HYAM and the rate of cooling
has the greatest impact on full cycle perform-
ance, the high-frequency work capacity during
electrothermal actuation is much higher for a
SRAM than a HYAM. The PEO-SO3@CNT SRAMs
electrothermally operated in air and in room-
temperature water to produce 2.6 W/g (for 3.2%
stroke at 9 Hz) and 9.0 W/g (for 5.5% stroke at
12 Hz), respectively, of full-cycle contractile power
(fig. S15, C to F), which is much higher than the
typical contractile power of human natural mus-
cle (0.05 W/g, 5). When operated in air, this SRAM
muscle provided a stoke of 8.0% at 2 Hz, corre-
sponding to a power density of 1.2 W/g. Shown
in movie S2 is the electrothermal actuation of a
coiled PEO-SO3@CNT SRAM in water at 12 Hz
to generate a 5% stroke and a full-cycle con-
tractile power of 4.2 W/g.
Fig. 2. Torsional actuation of twisted PEO-SO3 SRAMs and HYAMs driven by ethanol vapor. We next predicted the stress dependence of
(A) Illustrations (left to right) of a PEO-SO3 SRAM before vapor exposure and during vapor sorption tensile stroke and contractile work capacity
and then desorption, which cause yarn untwist and uptwist, respectively. (B) Illustration of vapor for ethanol-powered actuation of coiled PEO-
delivery to a muscle and plots of the time dependence of torsional stroke and rotation speed for SO3@CNT SRAMs and HYAMs (fig. S21) (26).
one sorption-desorption cycle for a PEO-SO3@CNT SRAM and HYAM and for a pristine CNT yarn. A This analysis used the above theoretically derived
41-mm-diameter pristine yarn, with 72 turns/cm of twist, was used for fabricating the 45-mm-diameter torsional strokes of twisted, noncoiled muscles,
SRAM and 50-mm-diameter HYAM, which contained a 0.53 weight ratio of PEO-SO3 to CNT. the relationship between torsional stroke (DT )
(C) Equilibrium torsional stroke versus weight changes (black symbols) during ethanol absorption and tensile stroke for noncontacting coils if
and desorption for the muscles of (B), and the SRAM-to-HYAM stroke ratio during ethanol absorption muscle stiffnesses were constant, and the depen-
(blue pentagons). (D) Torsional stroke versus time for the muscles of (B). (E) The observed (black dence of PEO-SO3 modulus on ethanol absorption
squares) and predicted (blue line) dependence of torsional stroke on SCR for PEO-SO3@CNT (fig. S4A). Remarkable agreement was obtained
SRAMs (26). (F) Torsional stroke and rotation speed versus time for a sorption-desorption cycle between theory and experiment for the stress
of a PEO-SO3@silk SRAM and HYAM and a silk yarn. A 56-mm-diameter silk yarn (with 5.7 turns/cm of dependence of equilibrium stroke and contractile
twist) was used for fabricating the 90-mm-diameter SRAM and HYAM, which weighed 0.48 mg/cm work capacity without using a fitted parameter.
and contained a 0.27 weight ratio of PEO-SO3 to silk. The observed ratio of the maximum contractile

Fig. 3. Isobaric tensile actuation of self-coiled, sorption-powered, and versus applied stress for the sorption-actuated muscles of (A). (C) The
electrothermally powered SRAMs, HYAMs, and pristine CNT yarns. time dependence of tensile stroke for a PU@CNT SRAM and HYAM
(A) Tensile stroke versus time for a PEO-SO3@CNT SRAM and HYAM and a pristine CNT yarn when electrothermally actuated by using the
and a pristine yarn when actuated by ethanol absorption by using illustrated configuration, 42 MPa stress, and 0.25 W/cm power, which
the illustrated configuration and 33 MPa stress. Sorption was from provided temperatures of 85°, 93°, and 97°C, respectively. (Left) The
a near-equilibrium ethanol concentration in dry air and desorption was device structure. Before coiling, the diameters of the PU@CNT SRAM
by means of dynamic pumping. Before coiling, the diameters of the and HYAM and the pristine yarn were 65, 71, and 51 mm, respectively.
PEO-SO3@CNT SRAM and HYAM and the pristine yarn were 43, 47, and (D) Tensile stroke and contractile work capacity versus applied stress for
38 mm, respectively. (B) Tensile stroke and contractile work capacity the electrothermally actuated yarns in (C).
work capacity of the SRAM to that of the HYAM is volume changes produced by electrochemical more impressive. For an applied square-wave fre-
1.70, which is close to the predicted 1.52 (fig. S22). double-layer injection of anions and cations. For quency of ~0.3 Hz, this ratio is ~3.4 for all applied
Electrochemically powered artificial muscles the used electrolyte of 0.2 M tetrabutylammonium loads. At the highest measured frequency (5 Hz)
have key advantages over thermally powered hexafluorophosphate (TBA·PF6) in propylene car- and the highest applied load, this ratio is 14.6. The
muscles: (i) Their efficiency is not limited by the bonate, the calculated van der Waals volume (28) electrochemical actuation of a coiled CNT@nylon6
Carnot efficiency, and (ii) they have a natural of the TBA+ cation (~293 Å3) is much larger than SRAM to provide 14.3% stroke at 0.25 Hz, while
latching state, meaning that stroke can be main- for the PF6– anion (69 Å3). Potential scans (Fig. 4B) lifting a heavy load, is shown in movie S3.
tained without the input of substantial electrical for the SRAM and HYAM show that muscle The frequency dependences of work capacity
energy. A conventional electrochemical CNT yarn contractions increase on both sides of the poten- for a coiled CNT@nylon6 SRAM and a coiled
muscle is a HYAM, in which the yarn guest is the tial of zero charge and that the contraction is CNT HYAM are shown in Fig. 4D for square-
electrolyte. proportional to the volume of the injected ion. wave voltages between 0 and –3 V. For 1 Hz cycle
A CNT@nylon6 SRAM was made with the pro- These contractions are largest for the SRAM. frequency, the tensile stroke, work per cycle, and
cess shown in Fig. 4A, right. Similar to a process Because the electrical energy required for average contractile power density for the SRAM
used to make coiled CNT yarns for energy har- actuation increases with increasing amount were, respectively, 4.7%, 0.99 J/g, and 1.98 W/g,
vesting (27), a stack of CNT sheets was formed of electrochemically accessible CNTs, the con- as compared with 0.90%, 0.11 J/g, and 0.22 W/g
into a cylinder (Fig. 4A, left). A nylon yarn was tractile work per weight of CNT is an impor- for the HYAM. The high performance obtained
placed in the center of the cylinder. Initially, twist tant performance metric. For slow square-wave for the SRAM at relatively high frequencies ex-
is inserted only into the CNT cylinder. However, switching at 10 mHz between 0 and –3 V (Fig. 4C), pands the application possibilities for electro-
once the CNT cylinder collapses to form a sheath the load-maximized contractile work capacity chemical artificial muscles.
on the nylon 6 yarn, torque automatically trans- is slightly higher for the CNT@nylon6 SRAM The contractile energy conversion efficiencies
fers from this sheath to the yarn, enabling the (2.35 J/g) than for the CNT HYAM containing were obtained for optimized voltage scan rates
yarn to become fully coiled. the same CNT weight per yarn length (2.01 J/g). between 0 and –2.7 V. This peak efficiency in-
The electrolyte-filled CNT sheath of the SRAM However, for more practically applicable actua- creased from 2.96% at 80 mV/s scan rate for the
and the electrolyte-filled volume of the HYAM tion rates (fig. S17), the ratios of SRAM to HYAM CNT yarn muscle to 4.26% at 130 mV/s scan rate
provide electrochemical actuation because of work capacities for similar tensile loads are much for the SRAM (Fig. 4E). Using a higher potential

Fig. 4. Fabrication and electrochemical tensile actuation of coiled CNT@nylon6 SRAM and the 70-mm-diameter CNT HYAM were 0.88 and
CNT@nylon6 SRAM and coiled CNT HYAM yarns in 0.2 M TBA·PF6/PC 0.56, respectively. (D) The frequency dependence of work capacity for a
electrolyte. (A) Illustration of (left) cone spinning for fabricating CNT coiled CNT@nylon6 SRAM and a coiled CNT HYAM for square-wave voltages
yarns and (right) its modification for making SRAM yarns. SEM micrographs between 0 and –3 V. For 1 Hz cycle frequency, the tensile stroke, work-
of a coiled pristine yarn, a coiled CNT@nylon6 SRAM yarn, and a per-cycle, and average contractile power density for the SRAM at the highest
noncoiled nylon 6 yarn are shown. (B) Tensile stroke of the SRAM and loads were 4.7%, 0.99 J/g, and 1.98 W/g, compared with 0.90%, 0.11 J/g,
HYAM during a cyclic voltammetry scan at 20 mV/s, under 22 MPa and 0.22 W/g for the HYAM. (E) The scan rate dependence of work capacity
isobaric stress. (Inset) Actuator stroke at this load for this muscle versus and energy conversion efficiency for the SRAM and HYAM, using an applied
interelectrode voltage scan rate. (C) Tensile stroke and contractile stress of ∼30 MPa for the SRAM and HYAM. For (D) and (E), the spring
work capacity versus load when applying a 10-mHz square-wave voltage indices of the 87-mm-diameter CNT@nylon6 SRAM and the 79-mm-diameter
between 0 and –3 V. The spring indices of the 95-mm-diameter CNT HYAM were 0.97 and 0.67, respectively.
scan rate for both muscles (200 mV/s) (fig. S18), increased porosity when exposed to moisture, The 5.2-, 9.0-, and 9.0-fold advantages at 1 Hz
which increased stroke rates, provided a SRAM and flat-coil SRAMs were demonstrated (figs. of the SRAM over the corresponding HYAM in
efficiency (3.8%) that is 2.7 times the HYAM ef- S23 to S25). Analyte-powered sensors that in- electrochemical stroke, contractile work-per-cycle
ficiency (26). telligently respond in the body to open and density, and average contractile power density, re-
Because the SRAM technology enables replace- close valves that release drugs in response to spectively (Fig. 4D), are important for electrically
ment of expensive CNT yarns with inexpensive, antigens (30) or biochemicals such as glucose powered robotic devices in which stroke should
commercially available polymer yarns whose (31) are other possibilities. A CNT-free SRAM be maintained without consuming substantial
sheath responds to targeted ambient variables, that linearly contracts with increasing glucose electrical energy. Electrothermal PEO-SO3@CNT
they are attractive for intelligent structures (29). concentration was demonstrated (fig. S26), SRAMs operated in air and in room-temperature
Relevant for possible use in comfort-adjusting which could squeeze a pouch to release a drug water to produce 2.6 W/g (at 9 Hz) and 9.0 W/g (at
clothing, SRAMs were knitted into a textile that (fig. S20). 12 Hz) of full-cycle contractile power, respectively,

compared with the 0.05 W/g typical of human ARTIFICIAL MUSCLES

muscle (5). SRAM performance and realizable
low cost suggests their use for diverse applica-
tions, from fast, powerful muscles for humanoid
robots and exoskeletons to intelligent comfort-
Shape memory nanocomposite
adjusting clothing and drug-delivery systems.
RE FE RENCES AND N OT ES
fibers for untethered
1.
2.
3.
M. D. Lima et al., Science 338, 928–932 (2012).
X. Gu et al., Nanoscale 8, 17881–17886 (2016).
Y. Sun et al., Carbon 132, 394–400 (2018).
high-energy microengines
4. Y. Song et al., Nanoscale 10, 4077–4084 (2018).
5. S. M. Mirvakili, I. W. Hunter, Adv. Mater. 30, 1704407 (2018).
Jinkai Yuan1*, Wilfrid Neri1, Cécile Zakri1, Pascal Merzeau1, Karl Kratz2,
6. C. S. Haines et al., Science 343, 868–872 (2014). Andreas Lendlein2,3, Philippe Poulin1*
7. S. H. Kim et al., Energy Environ. Sci. 8, 3336–3344 (2015).
8. S. Aziz, S. Naficy, J. Foroughi, H. R. Brown, G. M. Spinks, Classic rotating engines are powerful and broadly used but are of complex design
Polym. Test. 46, 88–97 (2015).
9. P. Zhang, G. Li, Polymer (Guildf.) 64, 29–38 (2015).
and difficult to miniaturize. It has long remained challenging to make large-stroke,
10. M. Hiraoka et al., Sci. Rep. 6, 36358 (2016). high-speed, high-energy microengines that are simple and robust. We show that
11. A. M. Swartz, D. R. Higueras Ruiz, H. P. Feigenbaum, M. W. Shafer, torsionally stiffened shape memory nanocomposite fibers can be transformed upon
C. C. Browder, Smart Mater. Struct. 27, 114002 (2018). insertion of twist to store and provide fast and high-energy rotations. The twisted
12. H. Cheng et al., Adv. Mater. 26, 2909–2913 (2014).
13. J. Fan, G. Li, RSC Advances 7, 1127–1136 (2017).
shape memory nanocomposite fibers combine high torque with large angles of rotation,
14. S. M. Mirvakili, I. W. Hunter, ACS Appl. Mater. Interfaces 9, delivering a gravimetric work capacity that is 60 times higher than that of natural
16321–16326 (2017). skeletal muscles. The temperature that triggers fiber rotation can be tuned. This
15. J. Gong, H. Lin, J. W. C. Dunlop, J. Yuan, Adv. Mater. 29, temperature memory effect provides an additional advantage over conventional
1605103 (2017).
16. C. Lamuta, S. Messelot, S. Tawfick, Smart Mater. Struct. 27,
engines by allowing for the tunability of the operation temperature and a stepwise
055018 (2018). release of stored energy.
17. P. Chen et al., Nat. Nanotechnol. 10, 1077–1083 (2015).
M
18. J. Deng et al., Nat. Protoc. 12, 1349–1358 (2017).
19. W. Guo et al., Adv. Mater. 24, 5379–5384 (2012). iniature engines or motors are of prac- engines can be made by twisting polymer fibers.
20. F. Meng et al., Adv. Mater. 26, 2480–2485 (2014). tical interest for emerging applications For example, the twisted rubber band used in
21. I. Agnarsson, A. Dhinojwala, V. Sahni, T. A. Blackledge, J. Exp.
Biol. 212, 1990–1994 (2009).
ranging from microrobotics, lab-on-a- airplane toys untwists because of the entropic
22. J. Foroughi et al., Science 334, 494–497 (2011). chip technology, and smart textiles, to elasticity of polymer chains (11). However, this
23. J. A. Lee et al., Adv. Mater. 29, 1700870 (2017). microelectromechanical systems and type of engine suffers from a low energy density
24. S. H. Kim et al., Sci. Rep. 6, 23016 (2016). miniaturized medical devices (1). Making large- owing to the low Young’s and shear moduli of
25. M. D. Lima et al., Science 331, 51–55 (2011).
stroke, high-speed, high-energy rotating micro- elastomers. Twisted nylon-6,6 fibers contract and
26. Materials and methods are available as supplementary materials.
27. S. H. Kim et al., Science 357, 773–778 (2017). engines showing simplicity and robustness has untwist in response to thermal expansion with a
28. M. Ue, A. Murakami, S. Nakamura, J. Electrochem. Soc. 149, long remained challenging. Different mecha- high energy density but necessitate large tem-
A1385–A1388 (2002). nisms and materials have been sought to pro- perature changes (12). Buckled sheath-core rub-
29. M. A. McEvoy, N. Correll, Science 347, 1261689 (2015).
30. T. Miyata, N. Asami, T. Uragami, Nature 399, 766–769 (1999). vide torsional rotations, such as shape memory ber muscles that operate electrically have also
31. J. Lee et al., Small 12, 2085–2091 (2016). alloys (2), piezoelectric ceramics (3), and electro- been proposed as rotary motors (13). Here, we
active polymers (4). The most promising rotat- show that the twist insertion into shape memory
ACKN OW LEDG MEN TS
ing performances have been achieved using the nanocomposite fibers is efficient to create hook-
We thank Lintec of America, Nano-Science and Technology Center for
concept of twisted fibers (5), as the insertion of free and high-energy microengines. Because
providing CNT forests. Funding: Support in the United States was from
Air Force Office of Scientific Research grants FA9550-18-1-0510 and the twist amplifies the strokes and work capac- of the energy storage capability (14) and the
FA9550-17-1-0328; Office of Naval Research contract N68335-18-C- ities compared with those of nontwisted or triggering of the untwist by a small increase in
0368; National Science Foundation grants CMMI-1661246, CMMI- noncoiled fibers. temperature above the switching temperature
1636306, and CMMI-1726435; Robert A. Welch Foundation grant
Following the demonstration of electrochem- Tsw (which is in the range of an involved thermal
AT-0029; and the Louis Beecherl Jr. Endowed Chair. Australian support
was from the Australian Research Council for a Centre of Excellence ically driven motors based on twist-spun carbon transition) by environmental heating, these multi-
(CE140100012) and a DECRA Fellowship (DE12010517). Korean nanotube (CNT) yarns (6), highly coiled CNT functional micromachines can work untethered.
support was from the National Research Foundation of Korea for yarns have emerged. The helical topology enables We start with polyvinyl alcohol (PVA)–based
the Creative Research Initiative Center for Self-powered Actuation.
a conversion of the yarn volumetric expansion shape memory polymer (SMP) fibers with high
Chinese support was from the Science and Technology Commission of
Shanghai Municipality (16JC1400700). Authors contributions: J.M., into tensile contraction and torsional untwist, strength and toughness (15), in which crystallites
M.J.d.A., S.F., N.L., and R.H.B. conceived and initiated the project. delivering high gravimetric work capacities form the permanent netpoints. A 2-cm-long,
All authors contributed to experimental design, planning, and (7–10). The volume change is driven by the ex- 40-mm-diameter PVA fiber (Fig. 1A) was first
execution; data analysis; and manuscript writing. D.Q., G.S., H.L.,
pansion of infiltrated guest materials in response heated to the programming temperature Td ~
X.W., E.G., and R.H.B. built the model. Competing interests: J.M.,
M.J.d.A., S.F., N.L., and R.H.B. are listed as the inventors on a to heat (7), liquid adsorption (8), or by the sur- 100°C (which is above the glass transition tem-
provisional U.S. patent application (62/846,479) that describes face tension of liquids diffused through gaps of perature Tg ~ 80°C). Isobaric twist [under con-
sheath-run artificial muscles. Data and materials availability: All data hierarchically arranged helical CNT fibers (10). ditions of constant tensile force provided by a
needed to evaluate the conclusions in the paper are present in the
Nevertheless, the difficulty and high cost of fab- weight of 1.2 g attached to the fiber end (unless
paper or the supplementary materials.
ricating CNT muscles has restricted their ap- otherwise noted, the weight is fixed at 1.2 g)]
SUPPLEMENTARY MATERIALS plications. Alternatively, simple and low-cost was then inserted into this fiber at 7500 turns
science.sciencemag.org/content/365/6449/150/suppl/DC1 per meter of fiber length, with a rotation speed
Materials and Methods 1
Université de Bordeaux, CNRS, Centre de Recherche Paul
of 60 revolutions per minute (rpm). Afterwards,
Supplementary Text the twisted SMP fiber was quenched to room
Pascal, UMR5031, 33600 Pessac, France. 2Institute of
Figs. S1 to S26
Tables S1 and S2
Biomaterial Science and Berlin-Brandenburg Center for temperature Tr to fix the coiled structure (Fig. 1B).
Regenerative Therapies, Helmholtz-Zentrum Geesthacht, The coiled structure can be retained without
14513 Teltow, Germany. 3Institute of Chemistry, University of
Movies S1 to S3
Potsdam, 14476 Potsdam, Germany.
being hooked because of the glassy nonequili-
11 December 2018; accepted 11 June 2019 *Corresponding author. Email: jinkai.yuan@crpp.cnrs.fr (J.Y.); brium conformation of the helically configured
10.1126/science.aaw2403 philippe.poulin@crpp.cnrs.fr (P.P.) polymer chains. Upon reheating the twisted fiber

to a temperature above Tsw (14), the one end- torque needed to twist the fibers at Tr. The pure modulus (~1.3 GPa) of PVA fiber up to 2.3 and
tethered fiber rotated its free end to revert to its PVA fiber shows high toughness and a tensile 6.8 GPa, respectively. Relative to SWNTs aligned
original, equilibrium shape via untwist (Fig. 1C). Young’s modulus of 4.9 GPa at Tr (Fig. 1D). The along the fiber axis (17), GO has a more pro-
A practical example of the programming and incorporation of SWNTs or GO decreases the nounced effect on the improvement of torsional
untwist of the twisted fiber in response to heat strain to failure but increases the Young’s mod- properties because of its two-dimensional (2D)
(shape memory effect) is shown in movie S1. ulus up to 13.5 and 12.5 GPa, respectively. Both structure. The shear is applied perpendicular to
The energy that a rotating engine provides via nanofillers have nearly the same reinforcement the nanotube orientation direction but remains
shape recovery is a function of the energy ab- efficiency on the tensile properties. By contrast, within the plane of 2D GO platelets. The nano-
sorbed during the twist programming at Td. The a greater torque is needed to twist the PVA- sheets allow the fiber to sustain a high torque
polymer fibers can be stiffened by the inclusion GO fiber to a given angle compared with that under twist (fig. S2). We therefore chose the
of reinforcing nanofillers and can be made more needed to twist pure PVA and PVA-SWNT fibers strong PVA-GO fibers to investigate in more
efficient for high-energy microengines. We pre- (Fig. 1E). By considering the fiber to be a non- depth the torsional rotation of shape memory
pared single-walled carbon nanotube (SWNT)– coiled cylinder, the shear stress and shear strain microengines.
and graphene oxide (GO)–doped PVA fibers by can be calculated. The elastic shear moduli are We investigated the effect of programming
using a wet-spinning method upon injection of deduced from the ratio of shear stress and shear temperature Td on the torsional untwist for PVA-
a PVA-SWNT dispersion or PVA-GO solution in strain at small twist angles (shear strain < 0.02) GO fibers. A higher torque is needed to twist
an aqueous solution of Na2SO4 as a coagulating (fig. S1). Note that at higher strains, plastic de- the fibers at a lower Td, indicating that more
bath (16). The wet-spun composite fibers have formation takes place, followed by coiling at torsional mechanical energy is stored during
5 weight % (wt %) SWNT and 5 wt % GO nano- extreme twist levels. Nevertheless, the exact programming (Fig. 2A). When reheating the
particles (15). onset of coiling could not be directly measured fiber in free load conditions at a rate of 5°C/min
Quantitative characterizations were achieved from the shear stress–shear strain curves. The (unless otherwise noted, the heating rate is
by measuring the stress needed to stretch or the SWNT and GO nanoparticles boost the shear fixed at 5°C/min), higher full rotations are
Fig. 1. SMP fibers for

rotating microengines.
(A) Scanning electron
microscopy (SEM)
images of a 40-mm-
diameter PVA fiber. Its
cross-sectional morphol-
ogy is shown in fig. S11.
(B) SEM image of the
coiled PVA fiber. (C) Evolu-
tion of the morphology
of the twisted fiber as
it untwists in response to
heating. (D) Tensile stress
versus strain curves at
Tr for pure PVA fibers and
PVA nanocomposite fibers.
(E) The torque needed
to twist the pure PVA,
PVA-SWNT, and PVA-GO
fibers until their rupture at
Tr. The quantitative mea-
surements of the torque
and applied twist angle
were obtained using
homemade instruments
(fig. S6).

Fig. 2. Temperature memory of

twisted fibers. (A) 3-cm-long,
40-mm-diameter PVA-GO fibers
were programmed by inserting a
twist of 7500 turns/m with a
rotation rate of 60 rpm at
different temperatures Td ranging
from 80° to 140°C. The specific
torque versus twisted angle was
recorded at each temperature.
(B) Recovery rotation with free
load for the twisted fibers
that are programmed at different
Td . (C) Evolution of rotation
speed with temperature obtained
by the first derivative of the
curve of rotation versus time
shown in fig. S3. (D) Recovery-
specific torque as twisted fibers,
two ends of which are tethered to
prevent fiber untwisting, are heated.
Fig. 3. High energy density of shape

memory microengines. (A) Torque
needed to twist 3-cm-long, 23-mm-diameter
pure PVA, PVA-SWNT, and PVA-GO
fibers by 1885 rads (10,000 turns/m)
at a rate of 60 rpm at Td ~ 100°C.
After twisting, the fibers are quenched
in air to form coils with and without
being torque balanced for further
characterizations. (B) Recovery
torque generated by the coiled fibers
that have been quenched without
being hooked when they are reheated.
(C and D) Recovery angle upon
reheating in conditions of variable
applied torques tapplied for the
programmed fibers. (E and F) The
gravimetric work capacities as a
function of the ratio of applied
torque tapplied to the blocking
torque tmaximum. The coiled fibers
used for (C) and (E) are formed
by quenching in air without being
hooked, therefore losing some twist;
whereas for (D) and (F), the coiled
fibers are quenched by balancing
the twist inserted into the fibers.
The hook is removed before
the fiber is stimulated to rotate.

generated for fibers that have been initially pro- that needed for pure PVA and PVA-SWNT fibers composite fibers that can store mechanical energy
grammed at lower Td (Fig. 2B). A maximum because of the torsional reinforcement effect of GO and rotate in response to heat. This in-demand
value of 6123 turns/m was recorded by varying additives (Fig. 3A), which was also observed at shape recovery can be triggered by environmental
the temperature by ~160°C. Fibers that have Tr (Fig. 1E). heating at a temperature that depends on the
been twisted at greater Td recover their shape at Upon reheating the two ends-tethered twisted programming temperature (temperature mem-
higher temperatures but show decreased full ro- fibers from Tr to 210°C, the recovery torque ory). PVA fibers that are torsionally stiffened by
tations. The recovery rate Rr (the ratio of recov- reaches a maximum blocking torque tmaximum the inclusion of 2D GO lead to untethered micro-
ery number of rotations to the applied number of at 100°C (Fig. 3B). It reaches ~0.27 mN·m for engines with high gravimetric work capacity.
turns) decreases as the programming tempera- PVA-GO fiber, which is higher than 0.12 and Additionally, the distinctive temperature mem-
ture increases. The best Rr (82%) is obtained for a 0.11 mN·m, respectively, achieved for PVA-SWNT ory feature enables a large tunability of the stor-
programming temperature near 80°C, at which a and pure PVA fiber. We measured the recovery age and release of mechanical energy.
maximum mechanical energy was stored. angle against an applied torque tapplied below
On the basis of the first derivative of the curve tmaximum. As the temperature increases to an REFERENCES AND NOTES
of rotation versus time (fig. S3), the rotation rate operation window (DT, in the vicinity of Td ~ 1. G. Z. Yang et al., Sci. Robot. 3, eaar7650 (2018).
can be calculated and is plotted as a function of 100°C), in which the recovery torque remains sup- 2. K. J. Gabriel, W. S. N. Trimmer, J. A. Walker, Sens. Actuators 15,
temperature in Fig. 2C. The twisted fiber untwists erior to tapplied, the fiber starts to untwist and 95–102 (1988).
with a peak rate at a well-defined temperature, provide mechanical output. For PVA-GO fiber, as 3. J. Kim, B. Kang, Smart Mater. Struct. 10, 750–757
(2001).
which is equal to Td, showing a temperature mem- the applied torque increases from 20 to 80% of 4. Y. Fang, T. J. Pence, X. B. Tan, IEEE ASME Trans. Mechatron.
ory effect (TME) closely reminiscent of the TME the blocking torque, the recovery rotation de- 16, 656–664 (2011).
previously observed in tension or contraction creases and the needed operation window DT 5. C. S. Haines et al., Proc. Natl. Acad. Sci. U.S.A. 113,
11709–11716 (2016).
(18–21). The rotation rate largely depends on the narrows from 120° to 70°C (Fig. 3C). An optimum
6. J. Foroughi et al., Science 334, 494–497 (2011).
heating rate. Upon immediate heating from Tr work capacity is achieved at an intermediate 7. M. D. Lima et al., Science 338, 928–932 (2012).
to 100°C, the fiber rotates at its free end a 5000- load. The gravimetric work capacity is defined 8. M. D. Lima et al., Small 11, 3113–3118 (2015).
times-heavier syringe needle and a paper paddle here as the amount of work done by the fibers, 9. S. H. Kim et al., Sci. Rep. 6, 23016 (2016).
to a peak rotation rate of 600 rpm in 2 s. It main- divided by the mass of the fibers. Figure 3E shows 10. P. Chen et al., Nat. Nanotechnol. 10, 1077–1083
tains rotation for ~5 s and ~50 full turns (fig. S4 the gravimetric mechanical output as a function (2015).
11. J. Yuan, P. Poulin, Science 343, 845–846 (2014).
and movie S2). The cycling shape memory be- of the ratio of tapplied/tmaximum. In response to a
12. C. S. Haines et al., Science 343, 868–872 (2014).
havior of PVA-GO fibers is characterized (fig. S5) temperature variation of 100°C in the vicinity of 13. Z. F. Liu et al., Science 349, 400–404 (2015).
and stable cyclic performances are demonstrated Td, the twisted PVA-GO fiber can rotate against a 14. A. Lendlein, O. E. C. Gould, Nat. Rev. Mater. 4, 116–133
(15). Additionally, as thermally powered shape load of 50% of tmaximum to provide a maximum (2019).
memory nanocomposite microengines, the twisted gravimetric work capacity as high as 1800 J/kg 15. Materials and methods are available as supplementary
materials.
fibers can be principally triggered to untwist by (fig. S7). This energy density is far higher than
16. C. Mercader et al., J. Appl. Polym. Sci. 125 (S1), E191–E196
any kind of heating method. An example of heat- that of mammalian skeletal muscles (39 J/kg) (2012).
ing the fiber in viscous silicon oil is demonstra- (24) and higher than that of most of the pre- 17. B. Vigolo, P. Poulin, M. Lucas, P. Launois, P. Bernier,
ted in movie S3. viously reported artificial rotary engines (table S1). Appl. Phys. Lett. 81, 1210–1212 (2002).
We directly measure the generated torque upon The wet-spun PVA fibers do not 100% fix the 18. P. Miaudet et al., Science 318, 1294–1296 (2007).
19. T. Xie, Nature 464, 267–270 (2010).
reheating the programmed fiber at fixed defor- twisted deformation because of polymer chain
20. K. Kratz, S. A. Madbouly, W. Wagermaier, A. Lendlein, Adv.
mation from Tr to 220°C using homemade in- relaxation and thermal expansion. These effects Mater. 23, 4058–4062 (2011).
struments (15) (fig. S6). The fiber programmed result in the loss of some turns of twist after 21. M. Behl, K. Kratz, U. Noechel, T. Sauter, A. Lendlein, Proc. Natl.
at 100°C generated a maximum gravimetric torque being quenched in air. A torque-balanced struc- Acad. Sci. U.S.A. 110, 12555–12559 (2013).
22. H. C. Berg, Curr. Biol. 18, R689–R691 (2008).
of ~21 N·m/kg (Fig. 2D), which is lower than that ture can be added to prevent such untwist and
23. Y. Sowa, R. M. Berry, Q. Rev. Biophys. 41, 103–132
of bacterial flagella (200 N·m/kg) (22, 23) but further enhance the mechanical output. The (2008).
higher than that of previously developed artifi- hook is immediately removed before the fiber is 24. T. Mirfakhrai, J. D. W. Madden, R. H. Baughman, Mater. Today
cial torsional motors (table S1). Even by vary- stimulated to rotate. Figure 3D shows the re- 10, 30–38 (2007).
ing Td far from 100°C, this value decreases but covery rotations against different loads for the
AC KNOWLED GME NTS
still remains on the order of 10 N·m/kg. More- twisted PVA-GO fibers that have been quenched
We thank I. Ly for SEM images of the fibers. Funding: A.L.
over, the recovery torque peak occurs at the tem- and hooked. At the same load, the fibers rotate and K.K. were financially supported by the Helmholtz
perature at which the fibers were programmed. by a larger number of turns compared with Association through program-oriented funding. Author
This is a consequence of the TME. The distinctive twisted fibers quenched without being torque contributions: P.P. and J.Y. conceived and designed the
research project. W.N. prepared the wet-spun fibers. J.Y. and
feature of temperature memory with Tsw from balanced (Fig. 3C) and thus provide a greater
C.Z. performed the shape memory characterizations. P.M.
80° to 140°C, compared with other rotating en- gravimetric energy density of 2766 J/kg at an designed instruments for characterizations of torsional
gines, enables the shape memory nanocomposite optimum tapplied (50% of tmaximum) (fig. S8 and properties of fibers. J.Y., C.Z., K.K., A.L., and P.P. analyzed
microengines to be customized for releasing the Fig. 3F). the data. J.Y. and P.P. wrote the paper. All authors discussed
the results and commented on the manuscript. Competing
stroke or torque in the vicinity of a certain For comparison, PVA-CNT fibers and pure PVA
interests: A.L. and K.K are co-inventors on patents in the
temperature needed for a particular application. fibers were used to perform exactly the same pro- field of polymer-based shape memory materials and fibers.
The generated mechanical energy is expressed tocols to characterize their optimum energy den- Data and materials availability: All data are available in the
as U ¼ ∫Gdq when the fiber rotates by an angle q sity. Energy densities of 628 and 1115 J/kg were main text or the supplementary materials.
against a given torque G. Twist programming observed for twisted PVA-CNT fibers that had
was applied onto small-diameter pure PVA, PVA- been quenched without and with being torque SUPPLEMENTARY MATERIALS
SWNT, and PVA-GO fibers, which were obtained balanced, respectively (Fig. 3, E and F, and fig. science.sciencemag.org/content/365/6449/155/suppl/DC1
by thermally drawing the as-prepared fibers by S9); whereas pure PVA fibers showed similar Materials and Methods
200%. The reduction of the fiber diameter (from energy generation capabilities, 632 and 925 J/kg Supplementary Text
Figs. S1 to S19
40 to 23 mm) retards the onset of coiling, thus for the fibers quenched without and with being Table S1
largely increasing the amount of twist that can hooked, respectively (Fig. 3, E and F; fig. S10; and References (25–49)
be inserted before the direct contact of neighbor- table S1). Movies S1 to S4
ing coils (12). A greater torque is needed to twist With this work, we have created rotating mi- 13 December 2018; accepted 12 June 2019
PVA-GO fiber at Td ~ 100°C as compared with croengines based on multifunctional SMP nano- 10.1126/science.aaw3722

HOST-GUEST CHEMISTRY roform. Surprisingly, these affinities are achieved

using charge-neutral receptors, which circum-
vent the pH sensitivity typical of ionizable recep-
Chloride capture using a tors and avoid the introduction of additional
anions stemming from the ion exchange when
C–H hydrogen-bonding cage using charged receptors (9).

High size selectivity is needed to enable selec-
tive extraction of smaller anions such as ClÐ over
Yun Liu*, Wei Zhao, Chun-Hsing Chen, Amar H. Flood† larger anions such as iodide, IÐ. Large anions are
less hydrated and are easier to remove using
Tight binding and high selectivity are hallmarks of biomolecular recognition. Achieving liquid-liquid extractions according to the Hofmeister
these behaviors with synthetic receptors has usually been associated with OH and bias (14). One approach to size selectivity is to
NH hydrogen bonding. Contrary to this conventional wisdom, we designed a chloride- use cryptands (15). These receptors have a three-
selective receptor in the form of a cryptand-like cage using only CH hydrogen bonding. dimensional (3D) binding pocket and have
Crystallography showed chloride stabilized by six short 2.7-angstrom hydrogen bonds demonstrated improvements in size-selective
originating from the cage’s six 1,2,3-triazoles. Attomolar affinity (1017 M–1) was determined cation binding: When the 2D pocket of a crown
using liquid-liquid extractions of chloride from water into nonpolar dichloromethane ether is converted into a 3D pocket in a cryptand,
solvents. Controls verified the additional role of triazoles in rigidifying the three- the selectivity toward the size-matched K+ over the
dimensional structure to effect recognition affinity and selectivity: Cl– > Br– > NO3– > I–. larger Cs+ cation increases from a factor of 10 to a
This cage shows anti-Hofmeister salt extraction and corrosion inhibition. factor of 2500 (16). This design principle has
H
been used in NH hydrogen-bonding anion recep-
ydrogen bonding is an essential noncova- dichloroethane (DCE; e = 10.4), a typical sol- tors including LehnÕs cationic aza-cryptands
lent interaction used in the design of syn- vent used in liquid-liquid extraction (9), costs (17), SesslerÕs neutral pyrrolo cryptands (18), and
thetic receptors for selective molecular +52 kJ molÐ1, whereas the Na+ cation can be Bowman-JamesÕs polyamide cryptands, which are
recognition. Hydroxyl (OH) and amino transferred at a cost of just +25 kJ molÐ1 (10). selective for fluoride (19) and have been used by
(NH) groups are commonly used for this The high free energy of ClÐ transfer demands that Cummins and Nocera to stabilize peroxide (20).
purpose, and they are widely represented in a synthetic receptor has an affinity constant We have pioneered the use of nontraditional
biomolecular and chemical recognition. A recent exceeding 1 billion (DG of +52 kJ mol Ð1 ≡ K a CH donors originating from 1,2,3-triazoles in
reclassification of hydrogen bonding by the In- ~109 MÐ1). Furthermore, the larger size of ClÐ planar triazolophane macrocycles (Fig. 1C) (3).
ternational Union for Pure and Applied Chemis- (diameter = 3.8 Å) relative to Na+ (2.3 Å) neces- They display an alternating sequence of aryl and
try (1), however, assigns hydrogen bond donors sitates a larger binding cavity, which in turn triazole units to define a size-selective cavity for
more broadly by using differences in Pauling lowers charge density and therefore affinity. Only ClÐ and stabilization by CH⋅⋅⋅ClÐ hydrogen bonds.
electronegativities (Dc). OH and NH donors have a handful of receptors display affinities exceed- The triazolesÕ CH donors are polarized (3) by the
high differences (Dc = 1.24 and 0.84, respec- ing the free-energy penalty of transferring ClÐ cluster of three electronegative nitrogen atoms.
tively), whereas even a CH group with a differ- from water to typical nonpolar extraction sol- The ClÐ affinity of triazolophane macrocycles is
ence of 0.35 should support hydrogen bonding. vents: SindelarÕs fluorinated bambus[6]uril Ð38 kJ molÐ1 in dichloromethane (21). Although
Consistently, increasing examples of receptors (1010 MÐ1) measured in acetonitrile (7), MaedaÕs this affinity is relatively high, it is still insuffi-
relying on CH hydrogen bonding have emerged pyrrole macrocycle (1010 MÐ1) measured in dichlo- cient to overcome the water-DCE phase-transfer
(2Ð7) to expand the toolkit available for receptor romethane (CH2Cl2) (11), as well as DavisÕs steroid- free energy of chloride.
design. Nonetheless, CH hydrogen bonds are al tris-thiourea (1011 MÐ1) (12) and steroidal We designed a cryptand-like triazolo cage 1
usually relegated to the role of secondary con- squaramide (1014 MÐ1) (13) measured in wet chlo- (Fig. 1A) bearing only CH donors. The cavity
tacts during the conception of receptors designed
for high-affinity recognition. We challenge this
conventional wisdom by designing a CH hydrogen-
bonding cage (Fig. 1) that displays high affinity
and selective binding of chloride ions.
Interest in chloride has risen with intensive
and extensive water usage increasing the salinity
of terrestrial waters. Such lowered water quality
is anticipated to have negative impacts across a
range of industries including the food, energy,
and water sectors (8). Therefore, the creation of
high-affinity synthetic receptors for ClÐ has the
potential to inspire new strategies for effective
management of chloride salts, such as in the ex-
traction of ions from water (9).
Synthetic receptors capable of ion extraction
need affinities high enough to overcome the
associated hydration energies, which are sub-
stantially larger for anions than for cations.
The transfer of ClÐ from water (e = 80) to 1,2-
Department of Chemistry, Indiana University, Bloomington,

IN 47405, USA.
*Present address: Beckman Institute for Advanced Science and
Fig. 1. Synthesis and structure of the chloride-binding cage. (A) One-pot synthesis of
Technology, University of Illinois at Urbana-Champaign, Urbana, IL
61801, USA. triazolo cage 1. NaAsc, sodium ascorbate; TBTA, tris[(1-benzyl-1H-1,2,3-triazol-4-yl)methyl]amine.
†Corresponding author. Email: aflood@indiana.edu (B) Crystal structure of cage 1⋅NaCl. (C) Chemical structure of 2D triazolophane macrocycle 2.
SCIENCE sciencemag.org 12 JULY 2019 ¥ VOL 365 ISSUE 6449 159

size is identical to Bowman-James’s NH hydrogen- C3-symmetric building block tripropargylamine Affinities were initially determined in DMSO
bonding cryptand (19) but with triazoles in the and the readily prepared C2-symmetric bisazide (24). The high dielectric constant helps weaken
place of amides. This replacement increases (24). A crystal structure of the cage verified its binding (21) relative to less polar media, thereby
rigidity (fig. S3) by restricting rotation about geometry (Fig. 1B and fig. S27). The Cl– is stabi- allowing quantitative determination of affinity by
two bonds instead of just one bond in amides. lized by six short CH⋅⋅⋅Cl– hydrogen bonds from nuclear magnetic resonance (NMR) spectroscopy.
Overall, we hypothesized that the affinity and the triazoles (dH⋅⋅⋅Cl = 2.7 Å) and three from the Addition of Cl– as the tetra-n-butylammonium
selectivity shown by the rigidly preorganized phenylenes (dH⋅⋅⋅Cl = 2.9 Å), all below the 3.0 Å van (TBA+) salt to 1 in DMSO led to large down-
aryl-triazole macrocycles (3) would be enhanced der Waals contact distance. field shifts of triazole (Ha, 1.6 ppm) and phenyl-
further by the cryptand effect (22). Our results The 1⋅NaCl complex was consistently obtained ene (Hb, 0.8 ppm) resonances in the 1H NMR
agree with these ideas, but the performance sur- after column chromatography, providing an early spectra, indicating strong CH hydrogen bond-
passed expectations: An attomolar (1017 M–1) hint of the high Cl– affinity. The NaCl was not ing. Despite use of the polar solvent to weaken
affinity was found in dichloromethane (e = 8.9), intentionally added and was thought to have binding, the high affinity precludes direct mea-
whereas a nanomolar affinity (108 M–1) was been scavenged (11, 25) from the silica used in surement of the Cl– affinity.
found in the more polar dimethylsulfoxide (DMSO, the chromatographic purification of the cage. Most Competition titrations were used to sequen-
e = 46.8). This decreased affinity with an increased of the salt could be removed after six extractions tially measure anion affinities. The cage’s NO3–
dielectric constant is consistent with our previous with deionized water (fig. S4). By contrast, the affinity was measured (~104 M–1; fig. S6) by di-
finding on the 1/e solvent dependence of rigid 2D triazolophanes completely released chloride rect titration and then fitting the shifts of CH
triazolophanes (21). with just two extractions. Use of nitromethane donor proton peaks to a 1:1 binding isotherm
Triazolo cage 1 was designed such that each (e = 35.9) as a more polar solvent to weaken ion (fig. S7). The Br– affinity (~105 M–1; fig. S8) was
opening retained the same 24-membered chela- binding (21) offered no improvement (26). Pre- quantified using a competition titration of com-
tion site (Fig. 1B) as the 2D triazolophane macro- cipitation of AgCl by addition of AgNO3 was plex 1⋅NO3– with Br–. The equilibrium constant
cycle (Fig. 1C). A one-pot synthesis (15% yield) too rapid and led to irreversible loss of the for the conversion of 1⋅NO3– to 1⋅Br– was deter-
was achieved using copper(I)-catalyzed alkyne- cage (fig. S5). All subsequent experiments were mined from relative peak integrations (table S3).
azide cycloaddition (23) to combine a 2:3 stoi- performed on a 90 mol % salt-free batch of Finally, the nanomolar affinity of Cl– (~108 M–1;
chiometric mixture of the commercially available the cage. Fig. 2A and fig. S9) was determined using the
same method by monitoring conversion from
1⋅Br– to 1⋅Cl– (table S4). This chloride affinity
exceeds that of the 2D triazolophane (Fig. 2B)
and Bowman-James’s polyamide cryptand (19)
by four orders of magnitude. Slow anion ex-
change on the NMR time scale matches the be-
havior seen for cryptands binding to alkali cations
(22), which suggests that substantial structural
reorganization of the cage is needed to engage or
release anions.
The superior affinity suggested that the cage
would display elevated selectivities. Cl– was fa-
vored over the larger iodide, I–, by a factor of
1 million in DMSO (Fig. 2B). The 2D triazolo-
phane macrocycle showed only a factor of 500
difference between Cl– and I–. We synthesized a
second receptor as a control, triazolo tripod 3
(Fig. 2C). The cavity of 3 has the same spatial
arrangement of CH donors as the cage but with
more flexibility, and it only displayed a selectivity
factor of 10. Thus, both higher dimensionality
and rigidity contribute to the cage’s selectivity.
Weaker solvation of the cage’s cavity is also ex-
pected to lower the cost of its desolvation to en-
hance complexation (27).
We investigated the liquid-liquid extraction
of Cl– of various salts (24). On the basis of the
1/e solvent dependence of Cl– binding seen with
the 2D triazolophane (21), we estimated that the
affinity in dichloromethane would increase and
would far exceed the cost of aqueous extraction.
Extractions conducted at 1 mM on small scales
(2 ml) were quantified by 1H NMR spectroscopy,
and the results were verified using ionic con-
ductivity measurements on larger scales (80 ml).
Fig. 2. Quantifying and understanding chloride affinity and selectivity. (A) 1H NMR spectra of Extraction of Cl– was first tested using the
cage 1 upon titration with Br– and then by competitive exchange with Cl– ([1]0 = 0.5 mM, 298 K, tetraethylammonium (TEA+) salt. This cation con-
500 MHz, DMSO-d6). Peaks associated with free triazolo cage 1 are colored in red, 1⋅Br– in magenta, tributes a minor penalty to extraction (+8.4 kJ
and 1⋅Cl– in blue. (B) Anion affinities (K1) determined in DMSO-d6 for Cl–, Br–, NO3–, and I– as TBA+ mol–1) (28). Without the cage, extraction of the
salts. Colored lines are drawn to guide the eye. Errors were determined by root-mean-square deviations TEACl salt from aqueous solution into dichloro-
from the nonlinear fitting (peak shifts) or by averaging multiple data points (peak integral ratios). See methane was minimal (~2%), whereas with the
figs. S6 to S24 for more details. (C) Chemical structure of 3D flexible receptor 3. cage present it was quantitative (Fig. 3A and

6. M. Lisbjerg et al., J. Am. Chem. Soc. 137, 4948–4951

(2015).
7. H. Valkenier et al., Chem 5, 429–444 (2019).
8. M. A. Shannon et al., Nature 452, 301–310 (2008).
9. B. A. Moyer, P. V. Bonnesen, R. Custelcean, L. H. Delmau,
B. P. Hay, ChemInform 36, 10.1002/chin.200531276
(2005).
10. Y. Marcus, M. J. Kamlet, R. W. Taft, J. Phys. Chem. 92,
3613–3622 (1988).
11. Y. Haketa, H. Maeda, Chem. Eur. J. 17, 1485–1492
(2011).
12. A. J. Ayling, M. N. Pérez-Payán, A. P. Davis, J. Am. Chem. Soc.
123, 12716–12717 (2001).
13. S. J. Edwards, H. Valkenier, N. Busschaert, P. A. Gale,
A. P. Davis, Angew. Chem. Int. Ed. 54, 4592–4596
Fig. 3. Use of the cage for extraction of salts and as an anticorrosion agent. (A) Extraction
(2015).
efficiencies for salts as ammonium cations (black) and when assisted by triazolo cage 1 (red). [1]0 = 14. C. L. D. Gibb, B. C. Gibb, J. Am. Chem. Soc. 133, 7344–7347
0.2 mM, [X–]0 = 0.24 mM. See figs. S28 and S29. Extraction efficiency is defined as the amount of salts (2011).
extracted relative to the total salts (direct extractions) or total cage (limiting reagent in the assisted 15. S. O. Kang, J. M. Llinares, V. W. Day, K. Bowman-James, Chem.
Soc. Rev. 39, 3980–4003 (2010).
extractions). (B) Extraction efficiencies for salts as alkali cations (black) and when assisted by the cage 1
16. J. M. Lehn, J. P. Sauvage, J. Am. Chem. Soc. 97, 6700–6707
(red). [1]0 = 0.1 mM, [X–]0 = 0.5 mM. See fig. S30. (C) A thin film of the triazolo cage can protect mild (1975).
steel from corrosion in brine solution (5.6 M NaCl). 17. B. Dietrich, J.-M. Lehn, J. Guilhem, C. Pascard, Tetrahedron
Lett. 30, 4125–4128 (1989).
18. C. Bucher, R. S. Zimmerman, V. Lynch, J. L. Sessler, J. Am.
figs. S29 and S32). The cage overcomes the accu- forming 1⋅Cl– complexes (K1). Ion-pairing (Kipc = Chem. Soc. 123, 9716–9717 (2001).
mulated phase-transfer penalty of the TEACl salt 104.6 M–1) between the TEA+ counter cation and 19. S. O. Kang, J. M. Llinares, D. Powell, D. VanderVelde,
K. Bowman-James, J. Am. Chem. Soc. 125, 10152–10153
of +60 kJ mol–1. When we changed to the more 1⋅Cl– is negligible at 1 mM in dichloromethane. The (2003).
lipophilic TBA+ cation to favor extraction by –13 kJ following equation relates the observables to the 20. N. Lopez et al., Science 335, 450–453 (2012).
mol–1 (28), we saw complete removal and com- Cl– affinity (K1): 21. Y. Liu, A. Sengupta, K. Raghavachari, A. H. Flood, Chem 3,
plexation of various anions, including bromide, 411–427 (2017).
½Cage Cl o ½TEA þ o 22. J. M. Lehn, Acc. Chem. Res. 11, 49–57 (1978).
iodide, nitrite, and nitrate (Fig. 3A and fig. S29). Kw→o ¼ 23. M. Meldal, C. W. Tornøe, Chem. Rev. 108, 2952–3015
The high anion affinity of the cage also facili- ½Cageo ½Cl w ½TEA þ w
(2008).
tated the more difficult task of extracting alkali ¼ K1 Ktþ Kt ð1Þ 24. See supplementary materials.
metal salts (Fig. 3B) from water into dichloro- 25. Y. Hua, Y. Liu, C. H. Chen, A. H. Flood, J. Am. Chem. Soc. 135,
14401–14412 (2013).
methane. The difficulty stems from the transfer Using this equation, the cage’s Cl– affinity (K1) 26. We also refuted the hypothesis that the incomplete removal
free energy of Na+ (+25 kJ mol–1) being higher was estimated to be attomolar (1017 M–1) in wet of salt from the cage resulted from the exchange of
than that of TEA+ by a factor of 3 (27). To our CH2Cl2. This high affinity explains the chal- complexed Cl– with OH– ions from the aqueous washing
satisfaction, extraction efficiency using the cage lenges encountered when trying to remove Cl– layer (fig. S6).
27. D. J. Cram et al., J. Am. Chem. Soc. 107, 3645–3657
was highest for NaCl (18%), which was enhanced from the cage and the facile extraction of NaCl (1985).
over solvent alone by a factor of at least 108. Ex- from its surroundings. 28. M. H. Abraham, J. Liszi, J. Inorg. Nucl. Chem. 43, 143–151
traction efficiencies correlated with the cage’s bind- We next investigated whether the high-affinity (1981).
ing selectivity, where Br– extraction efficiency sequestration of Cl– could confer anticorrosion 29. G. Smolyakov et al., Desalination 432, 40–45 (2018).
30. J. Soltis, Corros. Sci. 90, 5–22 (2015).
was 8%, and there was no observable extraction properties. Cl– is classified as an aggressive ion 31. M. F. Montemor, Surf. Coat. Tech. 258, 17–37 (2014).
of NaI (Fig. 3B and fig. S31). The latter mani- that enhances the corrosion of iron by promot-
fested despite iodide’s substantially smaller phase- ing pitting (30). Consequently, we hypothesized ACKNOWLEDGMENTS
transfer penalty (+21 kJ mol–1). The extraction using that cage 1 would be able to inhibit corrosion by Y.L. thanks J. Fu for the inspirational discussion that led to
the triazolo cage thus displays anti-Hofmeister making Cl– inaccessible. This concept has been the conception of this project. Funding: Supported by the
Chemical Sciences, Geosciences and Biosciences Division,
selectivity, reversing the normal trends dictated explored previously using layered double hy- Office of Basic Energy Sciences, Office of Science, U.S.
by anion size (8, 14). droxide coatings (31). We deposited a small patch Department of Energy (DE-FG02-09ER16068). A.H.F. was
The Na+ salt showed higher extraction effi- (~1 cm2) of the cage (20 mg) on a test sample of supported by a Waterman Professorship. Author contributions:
ciency than other alkali cations (K+, Cs+). Pre- mild steel (0.1% carbon). The sample was sub- Y.L. conceived the project under the supervision of A.H.F.; Y.L.
sumably, Na+ forms ion pairs with complex 1⋅Cl–. merged in saturated brine (5.6 M) for 2 weeks.
conducted experiments with assistance from W.Z.; W.Z. duplicated
attomolar affinity determination and NaCl extraction experiments;
The 1H NMR signals of 1⋅NaCl (fig. S33) were The smooth film of the cage, which had formed C.-H.C. performed x-ray single crystal structural analysis; Y.L.
used to determine the diffusion coefficient before upon solvent-vapor annealing, remained adhered and A.H.F. analyzed the data and wrote the manuscript with
and after NaCl complexation and showed mini- input from all coauthors. Competing interests: Indiana University
and was found by visual inspection to inhibit the
has filed a patent application (PCT/US62/675572) on the
mal change (fig. S34), confirming that 1⋅NaCl is corrosion seen in the uncoated steel (Fig. 3C). work described for inventors A.H.F., Y.L., and W.Z. Data and
nonaggregated. The addition of 1 equivalent of The modularity of the cage lends itself to fur- materials availability: Crystallographic data are available free
18-crown-6 (18-C-6) helped overcome the last ther customization and incorporation into tech- of charge from the Cambridge Crystallographic Data Centre
portion of the penalty of coextraction of Na+ nologies for selective Cl– extraction as well as under reference CCDC No. 1533500. All other data supporting
the findings of this study are available in the manuscript or
(fig. S32) to effect quantitative extraction of salt protection of materials interfaced with water supplementary materials.
from water for NaCl at a concentration as low as that carries dissolved Cl–.
5 mM. The low concentration shifts ion partition- SUPPLEMENTARY MATERIALS
ing toward the aqueous phase (29), allowing us science.sciencemag.org/content/365/6449/159/suppl/DC1
to explore the limit of this dual-host strategy (9). Materials and Methods
1. E. Arunan et al., Pure Appl. Chem. 83, 1619–1636 (2011).
The measured equilibrium constant (KW →O = 2. W. B. Farnham, D. C. Roe, D. A. Dixon, J. C. Calabrese,
Supplementary Text
108.5 M–1) for extraction of TEACl from water to R. L. Harlow, J. Am. Chem. Soc. 112, 7707–7718 (1990).
Tables S1 to S8
Figs. S1 to S39
the organic phase (fig. S38) was used to evaluate 3. Y. Li, A. H. Flood, Angew. Chem. Int. Ed. 47, 2649–2652
the Cl– affinity (K1) in dichloromethane at 1 mM. (2008).
4. O. B. Berryman, A. C. Sather, B. P. Hay, J. S. Meisner, 1 January 2019; resubmitted 7 February 2019
Extraction can be described as overcoming the D. W. Johnson, J. Am. Chem. Soc. 130, 10895–10897 (2008). Accepted 6 May 2019
penalty of transferring the cation and anion (Kt+ ⋅ 5. S. Lee, C.-H. Chen, A. H. Flood, Nat. Chem. 5, 704–710 Published online 23 May 2019
Kt– = 10–8.5) from water to dichloromethane by (2013). 10.1126/science.aaw5145

IMMUNOTHERAPY main followed by CD28 and CD3z intracellular

domains; the cognate amph-ligand for this mu-
rine CAR is FITC-poly(ethylene glycol) (PEG)–
Enhanced CAR–T cell activity against 1,2-distearoyl-sn-glycero-3-phosphoethanolamine
(amph-FITC; Fig. 1B). When incubated with model
solid tumors by vaccine boosting APCs in vitro, amph-FITC was absorbed into the
plasma membrane in a dose-dependent manner,
and despite ongoing endocytosis, many mole-
through the chimeric receptor cules remained accessible to surface staining
with an anti-FITC antibody (Fig. 1, C and D).
Leyuan Ma1,2, Tanmay Dichwalkar1, Jason Y. H. Chang1, Benjamin Cossette1,
Amph-FITC–coated cells stimulated FITC–CAR-
Ts in a dose-dependent manner and were killed
Daniel Garafola1, Angela Q. Zhang1, Michael Fichter1, Chensu Wang1, Simon Liang1,
by FITC–CAR-Ts (Fig. 1, E and F).
Murillo Silva1, Sudha Kumari1, Naveen K. Mehta1,3, Wuhbet Abraham1,
On the basis of these findings, we next tested
Nikki Thai1, Na Li1, K. Dane Wittrup1,3,4, Darrell J. Irvine1,2,3,5,6*
whether amph-FITC molecules could decorate
APCs in LNs to prime FITC-CAR-Ts in vivo. Sub-
Chimeric antigen receptor–T cell (CAR-T) therapy has been effective in the treatment of
cutaneous (s.c.) immunization of mice with free
hematologic malignancies, but it has shown limited efficacy against solid tumors. Here
FITC did not result in accumulation in the drain-
we demonstrate an approach to enhancing CAR-T function in solid tumors by directly
ing LNs, whereas 10 nmol of amph-FITC was
vaccine-boosting donor cells through their chimeric receptor in vivo. We designed
detectable for 21 days (fig. S1A). Amph-FITC
amphiphile CAR-T ligands (amph-ligands) that, upon injection, trafficked to lymph nodes
primarily accumulated in draining LNs, with
and decorated the surfaces of antigen-presenting cells, thereby priming CAR-Ts in
low to negligible levels detectable in the liver,
the native lymph node microenvironment. Amph-ligand boosting triggered massive
spleen, and other organs (fig. S1B). Confocal
CAR-T expansion, increased donor cell polyfunctionality, and enhanced antitumor efficacy
imaging of LNs showed that amph-FITC ini-
in multiple immunocompetent mouse tumor models. We demonstrate two approaches
tially accumulated in interfollicular regions but
to generalizing this strategy to any chimeric antigen receptor, enabling this simple
partitioned onto CD11c+ dendritic cells (DCs) in
non–human leukocyte antigen–restricted approach to enhanced CAR-T functionality to
T cell areas over time (Fig. 2, A and B, and fig.
be applied to existing CAR-T designs.
S1C). Surface-displayed FITC could be detected
on sorted FITC+ CD11c+ cells stained with an anti-
C
himeric antigen receptor–T cell (CAR-T) the native TCR locus (thereby deleting the body against FITC (Fig. 2C and fig. S1D). In con-
immunotherapy targeting the CD19 anti- native TCR) have significantly enhanced acti- trast to the efficient amph-FITC insertion into
gen has produced some marked clinical vity (10). the membranes of many LN cell types in vitro,
responses in patients with leukemia and We recently developed a strategy to target vac- surface-accessible FITC was present primarily
lymphoma, including a high proportion of cines to lymph nodes by linking peptide antigens on macrophages and CD11c+ CD11b+ DCs in vivo
durable complete remissions (1, 2). However, to albumin-binding phospholipid polymers (11). (Fig. 2D and fig. S2, A and C). DCs line colla-
poor functional persistence of CAR-Ts in some Small peptides are normally rapidly dispersed gen conduits that carry lymph fluid into the
patients results in disease progression (3). De- into the blood after parenteral injection, but LN, and we hypothesize that the anatomic struc-
spite the success of CAR-T therapy in hemato- binding of amphiphile peptides to endogenous ture of LNs in part dictates preferential access
logical cancers, it has to date been much less albumin, which constitutively traffics from blood of these cells to amph-vax molecules entering
effective for solid tumors, and strategies to en- to lymph, retargets these molecules to lymph LNs (15). This is supported by the observation
hance efficacy in this setting remain an impor- nodes (LNs). In addition to exhibiting efficient that amph-FITC coinjected with a low-molecular-
tant goal (4, 5). Therapeutic vaccination is one lymph trafficking, these lipid-tailed molecules weight dextran [which is known to be trans-
well-established approach to enhance endoge- can also insert into cell membranes (12). We ported through the LN conduit system (16)]
nous T cell responses against cancer (6). Several therefore hypothesized that by attaching a small showed substantial colocalization in fiber-like
groups have demonstrated the concept of pre- molecule, peptide, or protein ligand for a CAR to structures extending from the sinuses (fig. S2D).
paring CAR-Ts from virus-specific endogenous the same polymer-lipid tail, CAR ligands could Immunization using amph-FITC together with the
lymphocytes or introducing a CAR together with be delivered by albumin to LNs and subsequently STING agonist adjuvant cyclic-di-GMP increased
a second antigen receptor specific for a target partition into membranes of resident antigen- the duration of amph-FITC display on multiple
peptide and then vaccinating recipients against presenting cells (APCs), thereby codisplaying the APCs and, as expected, led to up-regulation of
the viral or secondary antigen to boost CAR-T amphiphile ligand (amph-ligand) from the APC costimulatory molecules on amph-FITC+ DCs
therapy (7–9). However, these approaches suf- surface together with native cytokine-receptor (Fig. 2E and fig. S2E). Notably, however, surface-
fer from being human leukocyte antigen (HLA) costimulation (Fig. 1A). Here we show how the accessible FITC decayed quickly and persisted on
restricted, and the use of endogenous T cell dual properties of amph-ligands, lymph node only a small fraction of cells.
receptors (TCRs) may be superseded by recent targeting and membrane insertion, combine to To test the ability of amph-ligand immuniza-
advances where CARs genetically targeted to create a booster vaccine for CAR-Ts. This amph- tion to expand CAR-Ts in vivo, we transferred
ligand strategy safely expands CAR-Ts in vivo, CD45.1+ FITC-CAR-Ts into lymphocyte (lympho)-
1
David H. Koch Institute for Integrative Cancer Research,
while increasing their functionality and enhanc- depleted congenic CD45.2+ recipient mice and
Massachusetts Institute of Technology, Cambridge, MA ing antitumor activity in multiple models of solid subsequently vaccinated twice with amph-FITC
02139, USA. 2Howard Hughes Medical Institute, Chevy tumors. and adjuvant. The CAR-Ts expanded substan-
Chase, MD 20815, USA. 3Department of Biological To test the ability of amph-ligands to func- tially after amph-FITC vaccination, and expan-
Engineering, Massachusetts Institute of Technology,
Cambridge, MA 02139, USA. 4Department of Chemical
tionally decorate APCs in vivo, we first employed sion was increased by coadministering adjuvant
Engineering, Massachusetts Institute of Technology, a recently described “retargetable” CAR recogniz- (Fig. 2F). For example, transfer of 5 × 104 FITC-
Cambridge, MA 02139, USA. 5Department of Materials ing the small molecule fluorescein isothiocyanate CAR-T followed by amph-FITC vaccination with
Science and Engineering, Massachusetts Institute of (FITC), which is directed against tumors by co- adjuvant expanded these cells to a peak of
Technology, Cambridge, MA 02139, USA. 6Ragon Institute of
Massachusetts General Hospital, Massachusetts Institute of
administration of a FITC-conjugated antitumor ~70% of the total CD8+ T cell compartment,
Technology, Cambridge, MA 02139, USA. antibody (13). The anti-FITC scFv 4m5.3 peptide yielding a CAR-T population nearly double the
*Corresponding author. Email: djirvine@mit.edu (14) was fused to the CD8a transmembrane do- size achieved by administering a 200-fold-greater

number of CAR-Ts without vaccination (Fig. 2F). in this setting, two immunizations could expand ligand immunization, we depleted different cell
By 3 weeks after boost, the persisting CAR-Ts 106 transferred cells from undetectable levels to types in LNs. CAR-T expansion in response to
were a mixture of effector/effector memory and ~20% of the total CD8 compartment (Fig. 2H). amph-FITC immunization was not impaired
central memory cells (Fig. 2G). Amph-vax boost- To determine whether professional APCs played in Batf3−/− mice lacking cross-presenting DCs,
ing also expanded CAR-Ts in lympho-replete mice; an important role in CAR-T priming by amph- but depletion of total DCs in CD11c-diphtheria
A amph-ligand (i) peptide

Fig. 1. Design of an
O O amph-ligand vaccine to
O O
O O P O
N (OCH2CH2)45 N
R (ii) small molecule boost CAR-Ts. (A) Sche-
O H O- H H
NH4+ matic of the general
(iii) protein domain
O chemical structure of
albumin binding/membrane-inserting domain PEG2000 CAR ligand amph-ligands (top) and
the steps in amph-ligand
Albumin binding and trafficking to LN Transfer to APC cell surface CAR-T activation vaccine boosting in vivo
co-stimulatory co-stimulatory (bottom). Upon injection,
molecule 1 amph-ligand molecule 2
amph-ligands associate
cell surface
with albumin at the
APC
injection site and are
albumin DC subsequently trafficked to
the draining LNs. The
amphiphiles then transfer
to the membrane of lymph
node–resident cells,
including APCs. CAR-Ts
cell surface
that encounter decorated
CAR-T
CAR
APCs in the LNs are
cytokines CAR-T
signaling stimulated by the surface-
lymph node displayed amph-ligand
as well as costimulatory
OH
receptors and cytokines
B amph-ligand: O O E
O O produced by the APCs.
O P O
O
O H O-
N
H
(OCH2CH2)45N
H
O
4000 *** (B) Structures of amph-
O
NH4+ *** FITC and cognate FITC-
O- O
NH4+
*** **
cognate CAR: O
CAR and a representative
IFN secretion (pg/ml)

Untransduced
FITC-CAR 3000 * flow cytometry analysis
4m5.3 100
of Tcell surface expression
Normalized to mode
Myc tag scFv for FITC-CAR. (C and

80 2000
D) Flow cytometry analysis
60 at 24 hours (C) and
1000 confocal imaging after
40
CD28 30 min (D) of amph-FITC
20 insertion into DC2.4 cell
CD3ζ 0 membranes, by direct
0 0 25 100 500 FITC fluorescence or
CAR expression amph-FITC (nM) staining with an anti-FITC
C FBS amph-FITC antibody. (E and F) IFN-g
(nM)
F Untransduced T cells secretion (in picograms
500 FITC-CAR T cells per milliliter) (E) and killing
10% 100 *** (the percentage of target
25 60 *** cell death) (F) of amph-
***
500 FITC–coated DC2.4 cells
1% 100 50
after 6 hours coculture
25
with FITC–CAR-T or con-
Cell death %
500 40
trol untransduced T cells
0% 100
30 at a 10:1 effector:target
25
(E:T) ratio. Shown in (E)
0
20 and (F) are representative
FITC anti-FITC experiments with techni-
D 10
cal triplicates. P values
nucleus amph-FITC anti-FITC merged
100nM 0 were determined by
amph-FITC unpaired Student’s t test.
D .4
-F D 4
D .4
4
2.
2.
C 2
C 2
Error bars represent

IT DC
C
IT C
C
95% confidence intervals

-F
(CI). ***P < 0.0001;

ph
ph
10 µm **P < 0.01; *P < 0.05.

am
am

A CD3 B220 anti-FITC B CD3/anti-FITC CD11c/anti-FITC

Day 1 Day 7
500 µm 100 µm 25 µm
Day 7
C nucleus amph-FITC anti-FITC merged
10 µm
E amph-FITC amph-FITC + adjuvant

Day 1 Day 3
% of cells with surface FITC

% of cells with surface FITC
50 *** 8 ***
D Medullary MΦ CD11c+CD8+ DCs CD11c+CD11b+ DCs 40 ***
6
***
30
4
Day 3 20
2
Day 1 10 n.s n.s n.s n.s n.s
n.s
Control 0 0
lls
lls
lls
lls
s
s
C
C
M
M
ce
ce
ce
ce
D
D
b+
8+
ry
8+
b+
y
B
T
B
r
Surface FITC
la
la
11
D
11
ul
ul
C
c+
ed
D
ed
c+
C
F
c+
C
11
Day: -2 -1 0 7
c+
M
11
11
D
11
C
D
C
C
LD CAR-T Vax Vax 10x106 CAR-T
% of CD3+CD8+ T cells
80
5x104 CAR-T
in peripheral blood
5x104 CAR-T (amph-FITC) *** ***

60
5x104 CAR-T (amph-FITC + adjuvant) ***
1x104 CAR-T
40 ***
1x104 CAR-T (amph-FITC)
***
1x104 CAR-T (amph-FITC + adjuvant) ***
20
0
0 7 14 21 28
Days post vaccination
40
G ** Fig. 2. Amph-ligands accumulate on LN APCs and prime
% of CAR-T cells
amph-FITC
30 amph-FITC + adjuvant CAR-Ts in vivo. (A to E) C57BL/6 mice [n = 3 animals
Effector (CD62L- KLRG1+ IL7R-) per group for (A) to (C) or n = 5 animals per group for (D)
20
Effector Memory (CD62L- KLRG1- IL7R+) and (E)] were immunized s.c. with amph-FITC and cyclic
Central memory (CD62L+ KLRG1- IL7R+) di-GMP adjuvant [(A) to (C) and (E)] or amph-FITC alone
10
[(D) and (E)]. Shown are histological images of LNs
0 [(A) and (B)], confocal imaging of sorted amph-FITC–
r y y
cto or or coated CD11c+ cells isolated from LNs at 24 hours (C),
fe em em
Ef M m and flow cytometry analysis of the cellular biodistribution
ctor ral
fe nt
Ef Ce of amph-FITC 1 or 3 days after injection [(D) and (E)].
H Day: -1 0 7 mf, macrophages. (F to H) CD45.2+ C57BL/6 mice
(n = 7 animals per group) with [(F) and (G)] or without (H)
prior lympho-depletion (LD) were adoptively transferred
25 CAR-T Vax Vax 10x106 CAR-T

with CD45.1+ FITC–CAR-Ts and then vaccinated with
in peripheral blood
1x106 CAR-T
20 n.s
1x106 CAR-T (amph-FITC) *** amph-FITC. Shown are frequencies of peripheral blood
***
15 1x106 CAR-T (amph-FITC + adjuvant) *** CAR-Ts [(F) and (H)] and cellular phenotypes at day 30
5x104 CAR-T
n.s
(G). P values were determined by unpaired Student’s
10 5x104 CAR-T (amph-FITC)
p=0.12 * t test [(E) and (G)] and by an RM (repeated measures)
4
5x10 CAR-T (amph-FITC + adjuvant)
5 two-way analysis of variance (ANOVA) with Tukey’s
multiple-comparisons test [(F) and (H)]. Error bars
0
0 7 14 21 28 represent 95% CI. ***P < 0.001; **P < 0.01; *P < 0.05;
Days post vaccination
n.s., not significant.

toxin receptor (DTR) mice or macrophages using boosted CAR-Ts (Fig. 3E). In therapeutic studies, (TA99) that recognized the melanoma-associated
chlodronate liposomes led to significant reduc- animals receiving both CAR-T and repeated antigen TRP1 (Fig. 4, A and B). FITC/TRP1-CAR-
tions in CAR–T cell numbers (fig. S3, A to C). In amph-vax boosting had significantly delayed Ts were activated both by amph-FITC-coated
addition, the cytokine functionality of respond- tumor growth and prolonged survival (Fig. 3, F target cells and by TRP1-expressing B16F10 cells
ing CAR-Ts was reduced in all three settings (fig. and G). Treatment with 1 × 106 CAR-Ts alone led (fig. S8A), and killed TRP1+ target cells at levels
S3, A to C). In vivo blockade of a collection of co- to no long-term survivors, while this same CAR-T equivalent to those cells expressing mono-
stimulatory molecules expressed by APCs also dose boosted by amph-vaccination eliminated specific TRP1-CAR (Fig. 4C). In vivo, amph-
markedly suppressed both FITC–CAR-T expan- tumors in a majority of animals (Fig. 3, F and G). FITC vaccination stimulated FITC/TRP1-bispecific
sion and cytokine functionality in response to Administration of amph-pepvIII with adjuvant CAR-T proliferation (fig. S8B). Similar to obser-
amph-FITC immunization (fig. S3D). in the absence of CAR-Ts had no therapeutic vations in the EGFRvIII system, amph-vax boost-
A key concern with amph-ligand delivery is impact (fig. S6F). EGFRvIII–CAR-Ts from vacci- ing of FITC/TRP1–CAR-T in B16F10 tumor-bearing
the potential for toxicity from CAR-T–mediated nated animals persisted over time, and surviv- animals led to pronounced CAR-T expansion in
killing of decorated cells in LNs or other tissues. ing animals rejected tumor rechallenge at day the periphery and increased tumor infiltration
Consistent with the low fraction of any cell type 75 (fig. S6, G and H). Notably, animals that re- (fig. S8, C and D), with minimal serum cytokine
with detectable surface FITC ligand, no signifi- jected primary tumors after CAR-T plus amph- elevation and transient fluctuations in body
cant changes in viable LN cell populations were vax boosting therapy also rejected rechallenge weight after each vaccination (fig. S8, E and F).
detectable 1 day, 3 days, or 14 days after amph- with parental CT-2A tumor cells lacking the lig- Whereas adoptive therapy with FITC/TRP1–CAR-
FITC immunization (fig. S4, A to C). No changes and for the CAR-Ts , suggesting induction of an T alone had almost no effect on B16F10 tumor
in systemic liver enzymes, liver histopathology or endogenous T cell response against other tumor progression, repeated boosting after transfer with
CAR-T infiltration, or serum cytokine levels were antigens (fig. S6I). Motivated by this finding, we amph-FITC led to pronounced slowing in tumor
observed after amph-FITC boosting (fig. S4, D to evaluated the reactivity of splenocytes from CT- growth and extended survival (Fig. 4, D and E).
I). We further evaluated the functional integrity 2A-mEGFvIII tumor-bearing mice that received One resistance mechanism to CAR-T therapy is
of vaccinated LNs by administering an amph- CAR-Ts with or without two amph-pepvIII boosts. loss of surface antigen (20), but we did not ob-
FITC boost in the presence or absence of trans- Enzyme-linked immunosorbent spot (ELISPOT) serve apparent Trp1 loss upon tumor outgrowth
ferred FITC-CAR-Ts and then immunizing animals analysis of interferon-g (IFN-g) production by in this model (fig. S8, G and H). To assess poten-
with ovalbumin at the same site 5, 7, or 14 days splenocytes cultured with parental CT-2A cells tial autoimmune toxicity induced by amph-vax
later (fig. S4J). We observed decreased expan- revealed a strong endogenous T cell response boosting, we examined thymus and skin tis-
sion and functionality of endogenous SIINFEKL- against parental tumors (Fig. 3H). Similar to sues (which naturally express Trp1) from treated
specific T cells when animals were immunized amph-FITC–vaccinated mice, no antibody re- animals, but we found no changes in histopathology
5 days—but not 7 or 14 days—after amph-FITC sponse was elicited against pepvIII after three or CAR-T infiltration into the thymus with amph-
boost, suggesting that the combination of CAR-T rounds of weekly vaccination (fig. S6J). We also vax boosting (fig. S8, I to K). We also assessed
transfer and amph-FITC vaccination has a short- evaluated the therapeutic efficacy of CAR-T plus whether CAR-T therapy with vaccine boosting
term effect on priming of endogenous T cell amph vaccination in tumor-bearing mice with- would be more effective if mixed CD4/CD8 CAR-
responses [which recovers rapidly (fig. S4K)]. out lympho-depletion preconditioning. Tumor Ts were used. In vitro, both CD4+ and CD8+ CAR-Ts
Owing to the lack of T cell help, repeated amph- progression in animals receiving CAR-T alone were activated by culture with amph-ligand–
FITC immunization with adjuvant elicited no was indistinguishable from that in animals re- coated target cells (fig. S8L), and similar ther-
antibody response against the amph-ligand it- ceiving control untransduced T cells, whereas apeutic efficacy was observed when B16F10 tumors
self (fig. S5). CAR-T transfer combined with amph-pepvIII were treated with CD8 as with mixed CD4/CD8
We next evaluated if amph-ligands could be immunization delayed tumor growth and pro- FITC/Trp1–CAR-Ts boosted by amph-FITC vac-
used to prime a bona fide tumor antigen–specific longed animal survival (Fig. 3, I and J). In both cination (fig. S8, M and N).
CAR. The EGFRvIII-specific 139scFv CAR recog- the lympho-depleted and non–lympho-depleted To assess the broad applicability of this bi-
nizes a short linear epitope derived from EGFRvIII settings, amph-vax boosting was accompanied specific CAR platform irrespective of animal
(17). We prepared murine T cells expressing this by small transient alterations in animal body strain or haplotype and to evaluate treatment
CAR and synthesized an amph-vax molecule weight and minimal alterations in serum cyto- of metastatic disease, we prepared 4T1 tumor
composed of PEG-DSPE linked to the peptide kine levels (fig. S6, K and L). To assess the util- cells transduced to express mEGFRvIII and
ligand with or without an N-terminal FITC label ity of amph-vax boosting with a more potent luciferase, modeling EGFRvIII+ breast cancer
(amph-pepvIII; Fig. 3A). Similar to amph-FITC, “third-generation” CAR design, we generated an (21) on the BALB/c background. A cognate FITC/
amph-pepvIII inserted in cell membranes in vitro EGFRvIII-targeting CAR containing both CD28 EGFRvIII-bispecific CAR was generated, which
and the amph-pepvIII–coated cells stimulated and 41BB co-stimulatory domains. This CAR was was well-expressed in BALB/c T cells and was
EGFRvIII-CAR-Ts (fig. S6, A and B). Immunization well-expressed and functional in vitro (fig. S7, A functional in vitro and in vivo (fig. S9, A and
of mice with amph-pepvIII triggered EGFRvIII- and C). We then treated large (~50-mm2) estab- D). 4T1-mEGFRvIII tumor cells were injected
CAR-T proliferation in vivo (Fig. 3B). To test the lished CT-2A-mEGFRvIII tumors with EGFRvIII- intravenously (i.v.) into BALB/c mice to induce
therapeutic impact of vaccine boosting, we trans- 28BBzCAR-T cells, with or without amph-pepvIII lung metastases and then treated with FITC/
duced murine CT-2A glioma cells with EGFRvIII; boosting. In this high tumor burden setting, the EGFRvIII–CAR-T with or without amph-FITC
these cells were efficiently killed by EGFRvIII- CAR-Ts alone had a modest impact on tumor boosting. Tumor progression as assessed by
CAR-Ts in vitro (fig. S6, C and D). Transfer of progression, and amph-ligand boosting greatly bioluminescence imaging was significantly im-
EGFRvIII–CAR-T into lympho-depleted CT-2A- improved tumor control and enhanced overall pacted only when CAR-Ts were supplemented
mEGFRvIII tumor-bearing mice that were then survival (fig. S7, D and E). with amph-ligand boosting (fig. S9E), leading
immunized with amph-pepvIII expanded the Although use of a peptide ligand for CAR-Ts to prolonged survival and clearance of tumors
CAR-Ts substantially in the periphery (Fig. 3C). was effective, some CARs recognize three-dimensional in two of five animals (fig. S9F). In the CAR-T
Vaccination induced significant increases in structural epitopes (18). As an alternative strategy plus amph-vax–treated animals that relapsed,
the proportion of cells with an effector phenotype to amph-ligand boost with any CAR regardless EGFRvIII surface levels were markedly reduced,
(fig. S6E) and 5- to 10-fold increases in CAR– of the nature of its binding domain or specific- suggesting selection of low-antigen–expressing
T cell polyfunctionality (Fig. 3D). Amph-vax boost- ity, we devised a tandem scFv-based bispecific or null tumor cells during therapy (fig. S9G).
ing greatly increased CAR-T infiltration into tumors, CAR based on recently reported designs (19). Finally, to verify that this bispecific CAR ap-
and these tumor-infiltrating lymphocytes expressed The anti-FITC scFv was fused to the N-terminal proach could also be used to boost human CAR-T,
higher levels of granzyme B and Ki67 than un- extracellular domain of a tumor-targeting CAR we constructed a FITC/hCD19-bispecific human

A B C D
Day: 0 7 8 9 13 16 23 30 IFN +
amph-ligand: 9.1 12.1 100 IFN +TNF +
LEEKKGNYVVTDHC-PEG-DSPE 4x106 LD CAR-T Vax Peripheral blood analysis ***
tumor cells PBS
Untransduced CAR+ T cells Control T *** ***
80
(% of CAR-T cells)
Cytokine secretion
CAR- T cells 60 ***
EGFRvIII CAR 1x106 CAR-T
100
Normalized to mode
100 1x106 CAR-T + Vax ***
Normalized to mode
50
in peripheral blood
2.5x105 CAR-T 60
80 80 90.9 3.0
2.5x105 CAR-T + Vax *
IFNγ
40
60 60 6.1 69.5 40
30
40 40
Vax
20 20
20 20
10
0 0 0
0 38.9 0.8
CAR expression Cell trace dilution
2. 2 CA C trol
10 x1 T -T
AR A x
+ T
x
0 5 10 15 20 25
C 5C Va
Va
-T R-
5x .5 R- AR
6 n
6 0 o
5 0 +
Days post vaccination TNFα
10 1x1 C
E F
1x
Day: 0 4 5 6 13 Day: 0 7 8 9 16 23
6 6
10x106 LD 1x10 Vax TIL analysis 4x10 LD CAR-T Vax Vax Vax
tumor cells tumor cells
CAR-T
# of CAR-T cells per mg tumor
40000 ** 8000 12000 250

*** 10x106 Control T
** 2.5x105 CAR-T
Tumor area (mm2)

200 1x106 CAR-T ***
Granzyme B MFI
30000 6000 *** ***

10000 2.5x105 CAR-T + Vax p=0.088 ***
Ki67 MFI
150 1x106 CAR-T + Vax

20000 4000 10x106 CAR-T
100 10x106 CAR-T + Vax n.s
8000
10000 2000 50
0 0 6000 0
0 7 10 20 30
S
x
S
x
S
Va
PB
Va
PB
Va
PB
Days post tumor inoculation

III
III
III
pv
pv
pv
pe
pe
pe
G H Day: 0 4 5 6 13 20
6 6
100 10x10 LD 10x10 Vax Vax ELISPOT
10x106 Control T tumor cells CAR-T
Percent survival
10x106 CAR-T
p=0.16 ** ***
10 x106 CAR-T + Vax ** 80
1x106 CAR-T **
per 105 CD3 T cells

6
CAR-T + Vax
1x10 CAR-T + Vax * 60
50
# of Spots
2.5x105 CAR-T
2.5x105 CAR-T + Vax **
40 CAR-T
20
0 Control T
0 50 100 150 0
Days post tumor inoculation CT-2A No target
lT
-T R-T
x
Va
tro
AR CA
+
on
C
I Day: 0 8 9 16 23
J
6 C
4x10 CAR-T Vax Vax Vax
tumor cells
200
10x106 Control T 100 10x106 Control T
n.s n.s
(mm2)
Percent survival
10x106 CAR-T 10x106 CAR-T

150 *** ***
10x106 CAR-T + Vax *** 10x106 CAR-T + Vax ***
Tumor area
100 50
50
0 0
0 10 20 30 40 0 20 40 60
Days post tumor inoculation Days post tumor inoculation
Fig. 3. Amph-peptide ligands boost CAR-Ts in vivo for enhanced survival [(G) and (J)] of mEGFRvIII-CT-2A tumor-bearing mice treated
solid tumor immunotherapy in mice. (A) Structure of amph-pepvIII with EGFRvIII–CAR-T with or without amph-pepvIII vaccination for
and surface expression of EGFRvIII CAR. (B) Representative histogram animals that were lympho-depleted [(F) and (G) n = 5 animals per group;
showing EGFRvIII–CAR-T proliferation in LNs 48 hours after amph-pepvIII (H) n = 4 animals per group)], or lympho-replete [(I) and (J) n = 7
vaccination (n = 3 animals per group). (C and D) Expansion (C) and animals per group)] prior to adoptive transfer. The black arrow indicates
cytokine polyfunctionality at day 7 (D) of circulating EGFRvIII–CAR-Ts time of CT-2A-EGFRvIII tumor rechallenge. The red arrow indicates
following a single amph-pepvIII immunization (n = 5 animals per group). time of parental CT-2A tumor rechallenge. P values were determined
(E) Enumeration, granzyme B levels, and Ki67 levels of tumor-infiltrating by unpaired Student’s t test [(D), (E), and (H)], by an RM two-way
EGFRvIII–CAR-Ts (n = 4 animals per group) with or without amph-pepvIII ANOVA with Tukey’s multiple-comparisons test [(C), (F), and (I)], or
boost. (F to J) Tumor growth [(F) and (I)], ELISPOT of enriched CD3+ by log-rank test [(G) and (J)]. Error bars represent 95% CI. ***P < 0.001;
splenocytes cultured with irradiated parental CT-2A tumor cells (H), and **P < 0.01; *P < 0.05; n.s., not significant.

A APCs Tumor cells D Day: 0 4 5 6 13 20

6
5x105 LD 10x10 Vax Vax Vax
tumor cells CAR-T
250
FITC Control T
n.s
Tumor area (mm 2)

200 FITC/TRP1-CAR-T ***
FITC/TRP1-CAR-T + Vax **
4m5.3 scFv tumor antigen 150
specific scFV
100
CAR-T 50
0
0 5 10 15 20 25
CD28 Days post tumor inoculation
CD3ζ E
100
Percent survival
Control T
n.s
FITC/TRP1-CAR-T ***
B Untransduced C FITC/TRP1-CAR-T + Vax **
*** 50
FITC/TRP1-CAR 100 ***
100
80
Normalized to mode
80
0
Cell death %
60 0 10 20 30 40 50
60 Days post tumor inoculation
40 40
F Untransduced G Raji
20 20 FITC/hCD19-CAR
400 amph-FITC K562
100
***
IFN secretion(pg/ml)
0 0
Normalized to mode
***
80 300
-T
-C -T
/T 1-C l T
CAR expression
AR
P1 AR
TC P tro
FI TR on
60
C
200 ***
R
40 ***
100
20
Fig. 4. Amph-FITC ligands boost the antitumor activity of
0 0
bispecific CAR-Ts. (A) Schematic of bispecific CAR design:
lls
-T
-T
CAR expression
ce
AR
AR
the FITC-binding scFv 4m5.3 is fused through a short linker to a
-C
-C
d
19
tumor antigen–specific CAR, enabling the T cell to be triggered
19
ce
D
D
du
C
hC
by binding to either FITC-decorated cells or tumor cells.
ns
/h
TC
ra
(B) Representative T cell surface expression of FITC/TRP1-
nt
FI
U
CAR. (C) Killing of TRP1-expressing B16F10 cells in vitro after
6-hour coculture with FITC/TRP1–CAR-T, monospecific TRP1–CAR-T, expression of FITC/hCD19-bispecific CAR on human T cells. (G) FITC/TRP1-
or control untransduced T cells at an E:T of 10:1. (D and E) Tumor growth bispecific CAR-Ts responding to either hCD19+ Raji cells or amph-FITC–
(D) and survival (E) of B16F10 tumor-bearing mice (n = 7 animals per coated K562 cells as monitored by IFN-g secretion. Shown in (C) and
group) treated with 10 × 106 CAR–Ts alone or CAR-Ts plus amph-FITC (G) are representative experiments with technical triplicates. P values were
vaccination. P values were determined by an RM two-way ANOVA with determined by an unpaired Student’s t test [(C) and (G)]. Error bars
Tukey’s multiple-comparisons test (D) or by log-rank test (E). (F) Surface represent 95% CI. ***P < 0.0001; **P < 0.01; *P < 0.05; n.s., not significant.
CAR using the established FMC63 antibody RE FERENCES AND NOTES 14. E. T. Boder, K. S. Midelfort, K. D. Wittrup, Proc. Natl. Acad. Sci.
against CD19 (22) and expressed this CAR in 1. A. D. Fesnak, C. H. June, B. L. Levine, Nat. Rev. Cancer 16, U.S.A. 97, 10701–10705 (2000).
566–581 (2016). 15. M. Sixt et al., Immunity 22, 19–29 (2005).
human T cells (Fig. 4F). Human FITC/hCD19– 16. J. E. Gretz, C. C. Norbury, A. O. Anderson, A. E. Proudfoot,
2. D. N. Khalil, E. L. Smith, R. J. Brentjens, J. D. Wolchok,
CAR-Ts were stimulated by both CD19+ Raji cells Nat. Rev. Clin. Oncol. 13, 273–290 (2016). S. Shaw, J. Exp. Med. 192, 1425–1440 (2000).
as well as amph-FITC–coated target cells (Fig. 4G). 3. S. Guedan et al., JCI Insight 3, e96976 (2018). 17. J. H. Sampson et al., Clin. Cancer Res. 20, 972–984 (2014).
Altogether, we present here a new vaccine ap- 4. K. Newick, S. O’Brien, E. Moon, S. M. Albelda, Annu. Rev. Med. 18. S. N. De Oliveira et al., J. Transl. Med. 11, 23 (2013).
68, 139–152 (2017). 19. E. Zah, M. Y. Lin, A. Silva-Benedict, M. C. Jensen, Y. Y. Chen,
proach to boosting CAR-T numbers and func- Cancer Immunol. Res. 4, 498–508 (2016).
5. C. H. June, R. S. O’Connor, O. U. Kawalekar, S. Ghassemi,
tionality in vivo with low toxicity, enabling M. C. Milone, Science 359, 1361–1365 (2018). 20. E. Sotillo et al., Cancer Discov. 5, 1282–1295 (2015).
enhanced efficacy in syngeneic solid tumor 6. S. H. van der Burg, R. Arens, F. Ossendorp, T. van Hall, 21. C. A. Del Vecchio et al., Cancer Res. 72, 2657–2671 (2012).
models. Although not directly evaluated here, C. J. Melief, Nat. Rev. Cancer 16, 219–233 (2016). 22. J. N. Kochenderfer et al., Blood 116, 4099–4102 (2010).
7. M. Tanaka et al., Clin. Cancer Res. 23, 3499–3509 (2017).
this approach might be further enhanced by
8. X. Wang et al., Clin. Cancer Res. 21, 2993–3002 (2015). AC KNOWLED GME NTS
nascent strategies to improve CAR function, 9. C. Y. Slaney et al., Clin. Cancer Res. 23, 2478–2490 (2017). We thank the Koch Institute Swanson Biotechnology Center for
such as insertion of the CAR into the TRAC 10. J. Eyquem et al., Nature 543, 113–117 (2017). technical support, specifically, the whole-animal imaging core
locus (10). The bispecific vaccinable CAR design 11. H. Liu et al., Nature 507, 519–522 (2014). facility, histology core facility, and flow cytometry core facility. We
12. H. Liu, B. Kwong, D. J. Irvine, Angew. Chem. Int. Ed. Engl. 50, thank T. Seyfried for providing the CT-2A cell line. Funding: This
with amph-FITC vaccine offers a simple and 7052–7055 (2011). work was supported by the NIH (award EB022433), the Marble
universal solution to boosting CAR-Ts with any 13. J. S. Ma et al., Proc. Natl. Acad. Sci. U.S.A. 113, E450–E458 Center for Nanomedicine, and Johnson & Johnson. D.J.I. is an
antigen specificity. (2016). investigator of the Howard Hughes Medical Institute. The project

was also supported by award no. T32GM007753 from the National C.W., S.L., M.S., M.F., N.K.M., W.A., N.T., and N.L. performed SUPPLEMENTARY MATERIALS
Institute of General Medical Sciences. M.F. was supported by experiments. Competing interests: D.J.I. and L.M. are inventors science.sciencemag.org/content/365/6449/162/suppl/DC1
Deutsche Forschungsgemeinschaft grant FI 2249/1-1:1. The on international patent application PCT/US2018/051764 Materials and Methods
content is solely the responsibility of the authors and does not submitted by Massachusetts Institute of Technology, which covers Figs. S1 to S9
necessarily represent the official views of the National Institute the use of amphiphile-vaccine technology as a vaccine for CAR-Ts. References (23–28)
of General Medical Sciences or the National Institutes of Health. D.J.I. is a consultant for Elicio Therapeutics that has licensed
Author contributions: L.M., D.J.I., and K.D.W. designed the IP related to this technology. Data and materials availability:
studies. L.M. and D.J.I. analyzed and interpreted the data and Materials are available under a material transfer agreement 28 October 2018; accepted 10 June 2019
wrote the manuscript. L.M, T.D, D.G., A.Q.Z., J.Y.H.C., S.K., B.C, (contact person D.J.I.). 10.1126/science.aav8692
INORGANIC CHEMISTRY ions have been exclusively the subject of theo-

retical and gas-phase studies (24–26).
Recently, our group disclosed a protocol for
Characterization of the facile generation of donor-stabilized trialkyl-
substituted as well as previously unavailable silyl-
hydrogen-substituted silylium substituted silylium ions based on the heterolytic

cleavage of activated and even inert Si–C(spn)
bonds (n = 1 to 3) by protonation with Reed’s
ions in the condensed phase carborane acid [C6H6·H]+[CHB11H5Br6]– (Fig. 1A,
iii) (27). Encouraged by this work, we envisioned
Qian Wu, Elisabeth Irran, Robert Müller, Martin Kaupp,
the synthesis of hydrogen-substituted silylium
ions by protolysis of a suitable precursor silane
Hendrik F. T. Klare*, Martin Oestreich*
(28, 29). Here, we report the successful imple-
mentation of this strategy for the preparation
Hydrogen-substituted silylium ions are long-sought reactive species. We report a protolysis
and structural characterization of counteranion-
strategy that chemoselectively cleaves either an Si–C(sp2) or an Si–H bond using a carborane
stabilized secondary and primary silylium ions,
acid to access the full series of [CHB11H5Br6]–-stabilized R2SiH+, RSiH2+, and SiH3+ cations,
as well as the simplest silylium ion H3Si+, the
where bulky tert-butyl groups at the silicon atom (R = tBu) were crucial to avoid substituent
heavier analog of the methylium cation H3C+
redistribution. The crystallographically characterized molecular structures of [CHB11H5Br6]–-
(Fig. 1D).
stabilized tBu2HSi+ and tBuH2Si+ feature pyramidalization at the silicon atom, in accordance
Previous efforts by Müller and co-workers to
with that of tBu3Si+[CHB11H5Br6]–. Conversely, the silicon atom in the H3Si+ cation adopts a
generate a secondary silylium ion by classical hy-
trigonal-planar structure and is stabilized by two counteranions. This solid-state structure
dride transfer from dihydrosilanes were unsuc-
resembles that of the corresponding Brønsted acid.
cessful, leading either to triarylsilylium ions by
R
the reaction with the aromatic solvent, as in the
eactive intermediates play a key role in the (Fig. 1A, i, left) (14–19). Although steric shielding case of diaryldihydrosilane 3 (3 → 4; Fig. 1C) (9),
understanding of chemical reactions (1). of the empty 3p orbital at the silicon atom using or to intramolecular rearrangements, as in the
Accordingly, knowledge of the nature and bulky 2,6-disubstituted aryl substituents (i.e., case of cyclic disilyl-substituted dihydrosilane 5
reactivity of silylium ions (R3Si+) (2, 3), the Mes or Dur; duryl = 2,3,5,6-tetramethylphenyl) (5 → 6; Fig. 1C) (10). These studies indicate that
heavier analogs of classical carbenium ions allows for the formation of strictly tricoordinate the synthesis of hydrogen-substituted silylium
(R3C+) (4), is of fundamental interest. Relative to (“free”) silylium ions (6), the high steric demand ions is hampered by their exceptionally high elec-
their carbon homologs, however, the chemistry requires a more elaborate synthetic route, involv- trophilicity and tendency to undergo substituent
of silylium ions is much less developed (2, 3). ing remote attack of a strong electrophile (El+) exchange reactions (21–23). Although the use of
For instance, it took almost one century after on an allyltriarylsilane, the so-called allyl leaving sterically hindered aryl substituents and stabi-
the discovery of the first stable carbenium ion, group approach (Fig. 1A, ii, left) (20). A more lizing silyl substituents turned out to be ineffec-
the triphenylmethylium (trityl) cation (Ph3C+), convenient route to triarylsilylium ions was in- tive, we probed the hydride abstraction from a
until evidence of a corresponding silicon conge- troduced by the group of Müller, making use of a dialkyl-substituted dihydrosilane and chose di-
ner, the trimesitylsilylium ion (Mes3Si+; mesityl = substituent redistribution reaction in the hydride tert-butyldihydrosilane (7) as a test substrate,
2,4,6-trimethylphenyl), was provided in the con- abstraction of heteroleptic methyl(diaryl)silanes as we knew from our previous study that the
densed phase by Reed, Lambert, and co-workers (Fig. 1A, ii, right) (21, 22). This process can also bulky tert-butyl group prevents any substituent
(5, 6). In contrast, secondary and primary silylium be tuned to deliver donor-stabilized trialkyl- redistribution (23). To investigate a possible
ions with hydrogen substituents at the silicon silylium ions, as recently reported by our labo- influence of the counteranion, we used Reed’s
atom remain unknown (7, 8), presumably be- ratory (Fig. 1A, i, right) (23). However, all these carborane-based trityl salt [Ph3C]+[CHB11H5Br6]–
cause of the lack of a general method to access approaches invariably lead to triorganosubstituted as a hydride abstracting agent. Indeed, the for-
these reactive species (9, 10). The standard strat- tertiary silylium ions. mation of the secondary counteranion-stabilized
egy to generate silylium ions is the Bartlett- Attempts to generate hydrogen-substituted silylium ion tBu2HSi+[CHB11H5Br6]– (8) was ob-
Condon-Schneider protocol for hydride transfer silylium ions by standard hydride abstraction served in this case (7 → 8; Fig. 1C). However, the
from a hydrosilane to the trityl cation paired with have failed so far (Fig. 1C) (9, 10), although Jutzi reaction was extremely slow, not reaching full
a weakly coordinating anion ([WCA]–) (11, 12), and Bunte accessed a secondary silylium ion by conversion even after 2 weeks. Overall, these
also known as the Corey reaction (13). How- protonation of decamethylsilicocene (1 → 2; Fig. attempts demonstrate the challenges associ-
ever, this protocol is mainly limited to trialkyl- 1B) (7). However, the extreme sensitivity of the ated with the synthesis of hydrogen-substituted
substituted hydrosilanes, eventually providing silylium ion 2 impeded its crystallographic char- silylium ions.
donor-stabilized tetracoordinate silylium ions acterization, and quantum chemical calculations The efficiency of our recently introduced
revealed a nonclassical bonding situation, where method to generate donor-stabilized tertiary sily-
the positive charge is intramolecularly stabilized by lium ions by protonation (Fig. 1A, iii) prompted
Institut für Chemie, Technische Universität Berlin, 10623
Berlin, Germany.
the p-bonded pentamethylcyclopentadienyl lig- us to target the synthesis of a secondary silylium
*Corresponding author. Email: hendrik.klare@tu-berlin.de ands (8). Aside from this isolated and unique ion by means of this strategy. Initially, we de-
(H.F.T.K.); martin.oestreich@tu-berlin.de (M.O.) contribution, secondary and primary silylium cided to use phenyl (Ph) as the leaving group

because dephenylation (protodesilylation) proved drosilane substrate tBu2PhSiH (d = 13.2 ppm), a drogenation was determined to be higher than
to be a facile and fast process in our previous clear indication for the development of silylium 95:5. The competing protolysis of the Si–H bond
study (27). Treatment of a pale yellow suspension ion character. Unambiguous evidence for the was not completely surprising, as we had already
of benzenium ion [C6H6·H]+[CHB11H5Br6]– in structure of tBu 2 HSi+ [CHB11H 5Br 6]– (8) was applied this process to the electrophilic C–H sily-
benzene with equimolar di-tert-butyl(phenyl) eventually provided by crystallographic char- lation of electron-rich (hetero)arenes (30). Hence,
silane (9) at room temperature immediately re- acterization (Fig. 2A, bottom). Single crystals we considered the protolysis of a dihydrosilane
sulted in a colorless solution from which an off- suitable for x-ray diffraction were obtained as a complementary pathway for the generation
white solid precipitated. After 15 min, the solid from a solution of silylium carborane salt 8 in of a secondary silylium ion. Indeed, dihydrogen
was collected by filtration, washed with addi- ortho-dichlorobenzene by vapor diffusion with gas immediately evolved from the reaction of di-
tional benzene, and dissolved in deuterated ortho- n-hexane at room temperature. The key geomet- tert-butylsilane (7) with equimolar amounts of
dichlorobenzene for nuclear magnetic resonance ric parameters are summarized in Table 1 and are the carborane acid, cleanly affording tBu2HSi+
(NMR) spectroscopic analysis. Clean formation of discussed in comparison to the other obtained [CHB11H5Br6]– (8) as the sole product (7 → 8;
the secondary silylium ion tBu2HSi+[CHB11H5Br6]– silylium ion species. Fig. 2A, top right).
(8) was observed (9 → 8; Fig. 2A, top left). The 1H Careful NMR spectroscopic analysis of the To further probe the potential of our protolysis
NMR spectrum showed a diagnostic signal at d = protolysis of di-tert-butyl(phenyl)silane (9) re- method, we next focused on the generation of a
5.1 ppm with a 1J(Si,H) coupling constant of 233 Hz vealed that traces of tBu2PhSi+[CHB11H5Br6]– primary silylium ion. Both the dephenylation
determined from the 29Si satellites, revealing an also formed as a result of competitive Si–H bond protocol (10 → 12; Fig. 2B, top left) and the de-
intact Si–H bond. The associated doublet at d = cleavage. This tertiary dialkylarylsilylium ion ex- hydrogenation protocol (11 → 12; Fig. 2B, top
73.0 ppm in the 1H/29Si heteronuclear single hibits higher solubility and can thus be separated right) proceeded smoothly, providing access to
quantum coherence (HSQC) spectrum without from hydrogen-substituted silylium ion 8 by sim- the primary counteranion-stabilized silylium ion
decoupling in F1 dimension (fig. S6) was shifted ple washing with small amounts of benzene. The tBuH2Si+[CHB11H5Br6]– (12) from either tert-
downfield (Dd = 59.8 ppm) relative to the hy- chemoselectivity of dephenylation over dehy- butyl(phenyl)silane (10) or tert-butylsilane (11).
Fig. 1. Synthetic routes to silylium ions. (A) Reported methods to generate tertiary silylium ions ([WCA]– omitted for clarity) (13–23). (B) Jutzi and
Bunte’s synthesis of an unconventional secondary silylium ion (7, 8). (C) Attempts to generate secondary silylium ions by hydride abstraction (9, 10).
(D) Our planned protolysis strategy to access hydrogen-substituted, counteranion-stabilized silylium ions. Do, donor; El, electrophile; LG, leaving
group; Ph, phenyl; rt, room temperature; WCA, weakly coordinating anion.

However, as a result of the volatility of the latter

starting material [boiling point (b.p.) 34.4°C],
the dephenylation process proved more practi-
cal. The measured 1H NMR chemical shift of d =
5.0 ppm in deuterated ortho-dichlorobenzene was
similar to that of tBu2HSi+[CHB11H5Br6]– (9), but
a larger 1J(Si,H) coupling constant of 254 Hz was
observed. The 29Si NMR resonance at d = 27.0 ppm
in the 1H/29Si HSQC spectrum was shifted down-
field relative to the starting material tBuPhSiH2
(d = –14.5 ppm) and tBuSiH3 (d = –39.7 ppm), but
was shifted upfield (Dd = 46 ppm) relative to
secondary silylium ion 8. Slow diffusion of n-
hexane into a solution of tBuH2Si+[CHB11H5Br6]–
(12) in ortho-dichlorobenzene provided color-
less crystals suitable for x-ray diffraction (Fig.
2B, bottom). The key parameters are discussed
below.
Encouraged by the successful generation of
both a secondary and a primary silylium ion,
we then targeted the synthesis of the smallest
member of hydrogen-substituted silylium ions,
namely the elusive H3Si+ cation. Although there
are several theoretical and gas-phase studies on
this heavier methylium congener (24–26), its
synthesis in noncovalent form in the condensed
phase has not yet been accomplished (29, 31). In
this case, both the hydride abstraction and the
dehydrogenative protolysis approach (Fig. 2C,
top right) are impeded by the pyrophoric and
explosive nature of monosilane (SiH4) gas. Grati-
fyingly, the reaction of easy-to-handle phenyl-
silane (13; b.p. 120°C) with equimolar benzenium
ion [C6H6·H]+[CHB11H5Br6]– led again to se-
lective Si–C(sp2) heterolysis and formation of
H3Si+[CHB11H5Br6]– (14) under release of ben-
zene (Fig. 2C, top left). The poor solubility of this
silylium ion in benzene facilitated its isolation,
and 14 was obtained as an off-white solid in pure
form after simple filtration and washing with
small amounts of C6D6. The presence of three
hydrogen substituents at the silicon atom was
initially supported by the 1H/29Si HSQC spec-
trum in deuterated ortho-dichlorobenzene with
coupling in F1 dimension, showing a quartet for
the silicon atom at d = –27.6 ppm with a 1J(Si,H)
coupling constant of 287 Hz (fig. S27). The corres-
ponding singlet at d = 4.8 ppm in the 1H NMR
spectrum was again observed in a similar range
as for the secondary and primary silylium ions 8
and 12. The exclusively hydrogen-substituted silyl
cation H3Si+ was found to have limited stability in
solution, gradually transforming into PhH2Si+
by electrophilic C–H silylation of the benzene
solvent and subsequent dihydrogen elimina-
tion (fig. S1). This reaction pathway had already
been described by Müller and co-workers (see
3 → 4, Fig. 1C) (9, 10). Despite its high reac-
tivity in aromatic solvents, we eventually suc-
ceeded in the crystallographic characterization
of H3Si+[CHB11H5Br6]– (14). Colorless crystals
suitable for x-ray diffraction were grown from
a benzene solution by vapor diffusion with
n-hexane at room temperature (Fig. 2C, bottom).
Fig. 2. Synthesis and solid-state molecular structures of hydrogen-substituted, counteranion- In the solid state, silylium carborane salt 14
stabilized silylium ions. (A) Generation of a secondary silylium ion. (B) Generation of a primary is stable under inert atmosphere for several
silylium ion. (C) Generation of H3Si+. months at –30°C.

Table 1. Key geometric parameters for the molecular structures of silylium carboranes tBu(3–n)HnSi+[CHB11H5Br6]– (n = 0 to 3). For complete
crystallographic data and details of structure refinement, see supplementary materials. Data for tBu3Si+ are from (15).
tBu3Si+ tBu2HSi+ (8) tBuH2Si+ (12) H3Si+ (14)
Si–Br (•) 2.465 ± 0.005 2.4110 ± 0.0014 2.3777 ± 0.0016 2.477 ± 0.004 [Br1]
2.770 ± 0.004 [Br3]
............................................................................................................................................................................................................................................................................................................................................
B–Brcoord (•) 2.04 ± 0.02 2.007 ± 0.005 2.017 ± 0.006 1.988 ± 0.015 [Br1]
1.988 ± 0.016 [Br3]
............................................................................................................................................................................................................................................................................................................................................
B–Br uncoord (•) 1.92–1.93 ± 0.02 1.937–1.958 ± 0.006 1.941–1.961 ± 0.006 1.923–1.959 ± 0.016
............................................................................................................................................................................................................................................................................................................................................
Si–Br–B (¡) 125.0 ± 0.5 110.53 ± 0.16 104.42 ± 0.16 105.5 ± 0.4 [Br1, B6]
104.7 ± 0.4 [Br3, B8]
............................................................................................................................................................................................................................................................................................................................................
6 10 2 6 2 1C
R–Si–R (¡) 117.7 ± 0.7 [C , C ] 121.7 ± 0.2 [C , C ] 117 ± 3 [C , H ] 119.1 ± 1.7 [H1C, H1D]
115.9 ± 0.6 [C2, C6] 116 ± 2 [C2, H1B] 114 ± 3 [C2, H1B] 119.0 ± 1.7 [H1B, H1D]
2 10 6 1B 1B 1C
115.1 ± 0.7 [C , C ] 110 ± 2 [C , H ] 111 ± 4 [H , H ] 118.8 ± 1.7 [H1B, H1C]
P
............................................................................................................................................................................................................................................................................................................................................
R–Si–R (¡) 348.7 347.7 342 356.9
............................................................................................................................................................................................................................................................................................................................................
Mean R–Si–R (¡) 116.2 115.9 114 119.0
............................................................................................................................................................................................................................................................................................................................................
Br–Si–Br (¡) 179.0 ± 0.2
............................................................................................................................................................................................................................................................................................................................................
carborane upon dissolution. A comparison of ex-

Table 2. Experimental and calculated 29Si NMR chemical shifts of silylium carboranes perimental and computed 29Si NMR chemical
R3Si+[CHB11H5Br6]–. Experimental values were determined by 1H/29Si heteronuclear multiple shifts (table S1) suggests that the counteranion-
quantum coherence (HMQC) NMR spectroscopy in 1,2-Cl2C6D4 at 293 K. Fully relativistic stabilized silylium ions exist as ion pairs in solu-
4c-mDKS/PBE0/IGLO-III/Dyall(TZ) calculations were performed at PBE0-D3(BJ)/def2-TZVPP/ tion. However, we cannot fully exclude dynamic
COSMO(1,2-Cl2C6H4) structures. and, as such, partial exchange of the counter-
anion and the solvent because the chemical shifts
slightly vary in ortho-dichlorobenzene and ben-
Me3Si+ tBu2HSi+ tBuH2Si+ H3Si+ zene. Deuterium incorporation into the carbo-
experimental( Si)
29
93.5 73.0 27.0 –27.6
d..................................................................................................................................................................................................................... rane anion is in fact evidence of the formation
29
–21.5 of these deuterated arene adducts (37); these
calculated( Si) 85.9 74.5 24.5
d.....................................................................................................................................................................................................................
are at the same time strong Brønsted acids
and a source of D+, which in turn engages in
electrophilic aromatic substitution of the car-
For higher sensitivity, 29Si NMR chemical the long C–Si bonds and less effective orbital borane anion. H/D exchange at the silicon atom
shifts were measured by 1H/29Si HSQC spec- overlap. Instead, more electronic charge is with- was not observed.
troscopy (Table 2). Using this technique with drawn from the silicon atom as a consequence of The 1H NMR chemical shifts of the hydrogen
coupling in F1 dimension also allows for the the higher electronegativity (c) of the tert-butyl substituents at the silicon atom are observed in
determination of the number of hydrogen sub- group relative to the hydrogen atom [c(tBu), a narrow range of d = 4.8 to 5.1 ppm, shifted
stituents at the silicon atom. The spectra were 2.501; c(H), 2.176] (32). Alkyl substitution thus downfield relative to the neutral tetracoordinate
recorded in deuterated ortho-dichlorobenzene leads to deshielding of the silicon nucleus in the silane precursors. The increase of the 1J(Si,H)
because the silylium carborane salts are not very 29
Si NMR spectrum. This general trend is also coupling constant from tBu2HSi+ (233 Hz) and
soluble in benzene. The experimental 29Si NMR consistent with calculations in the gas phase tBuH2Si+[CHB11H5Br6]– (254 Hz) to H3Si+ (287 Hz)
chemical shifts were further supported by relati- by Olsson and Cremer (33). However, a 29Si NMR is in agreement with the decreasing HOMO
vistic density functional theory (DFT) calcu- chemical shift of d = 270.2 ppm is expected (highest occupied molecular orbital)–LUMO (low-
lations. The agreement between measured and for an ideal trigonal-planar, three-coordinate est unoccupied molecular orbital) gap (table S3).
computed 29Si NMR chemical shifts is excellent H3Si+ cation without any nucleophile interac- In accordance with the reported molecular
(fig. S35), confirming the high quality of the tion (33). The experimentally observed upfield structures of tertiary counteranion-stabilized
chosen computational protocol. shift is ascribed to counteranion as well as sol- trialkylsilylium ions R3Si+[CHB11H5Br6]– [R =
The 29Si NMR chemical shifts of reported vent coordination in the condensed phase and methyl (Me), ethyl (Et), isopropyl (iPr), tBu,
counteranion-stabilized tertiary trialkylsilylium is in line with our and previous quantum chem- tBu2Me] (14–16, 23), the positively charged sili-
ions of type R3Si+[CHB11H5Br6]– (R = alkyl) are ical computations (34–36). For instance, even the con atom in secondary tBu2HSi+[CHB11H5Br6]–
observed to be shifted downfield in the range of approach of weak nucleophiles such as methane (8) and primary tBuH2Si+[CHB11H5Br6]– (12) is
93 to 113 ppm (Table 2 and table S2) (14–16, 23). or benzene to the H3Si+ cation is predicted to coordinated by one bromine atom from the
Whereas the experimental chemical shift of lower the 29Si NMR chemical shift by 206 ppm pentagonal belt of the icosahedral carborane
the tri-tert-butyl derivative is unavailable (15), (36) and 294 ppm (33), respectively. anion, leading to pyramidalization at the silicon
the calculated value is 112.1 ppm (tables S1 and S2). We also analyzed H3Si+[CHB11H5Br6]– by solid- atom (Fig. 2, A and B, bottom). The Si–Br bond
Successive replacement of the tert-butyl substi- state NMR spectroscopy using cross-polarization lengths decrease from tBu3Si+[CHB11H5Br6]–
tuents by hydrogen atoms leads to pronounced magic angle spinning (CPMAS) on the material (2.465 Å) to 8 (2.411 Å) and 12 (2.377 Å). The
upfield shifts of 39 ppm (from tBu3Si+ to tBu2HSi+), identical to that characterized by x-ray crystal- reduced steric demand is likely to account for
46 ppm (from tBu2HSi+ to tBuH2Si+), and 55 ppm lography (fig. S28). The recorded resonance at the stronger attractive interactions (23). How-
(from tBuH2Si+ to H3Si+). Although the tert- d = –64.7 ppm corresponds to an upfield shift of ever, the Si–Br bond distance of 12 is still elon-
butyl substituent might be expected to stabilize Dd = 37.1 ppm relative to the sample in ortho- gated by more than 0.14 Å relative to the Si–Br
the positive charge at the silicon atom by hyper- dichlorobenzene, indicating breaking of the lin- bond in neutral bromosilane Me3SiBr (2.235 Å)
conjugation, this effect is negligible because of ear chain structure and solvation of the silylium (38), reflecting the weakly coordinating nature
SCIENCE sciencemag.org 12 JULY 2019 • VOL 365 ISSUE 6449 17 1

20. J. B. Lambert, Y. Zhao, H. Wu, W. C. Tse, B. Kuhlmann, J. Am.

Chem. Soc. 121, 5001Ð5008 (1999).
21. A. Schäfer et al., Angew. Chem. Int. Ed. 50, 12636Ð12638
(2011).
22. A. Schäfer et al., Organometallics 32, 4713Ð4722 (2013).
23. L. Omann et al., Chem. Sci. 9, 5600Ð5607 (2018).
24. C. Maerker, P. R. Schleyer, in The Chemistry of Organic Silicon
Compounds, vol. 2, Z. Rappoport, Y. Apeloig, Eds. (Wiley,
1998), pp. 513Ð555.
25. N. Goldberg, H. Schwarz, in The Chemistry of Organic Silicon
Compounds, vol. 2, Z. Rappoport, Y. Apeloig, Eds.
(Wiley, 1998), pp. 1105Ð1142.
26. S. Fornarini, in The Chemistry of Organic Silicon Compounds,
vol. 3, Z. Rappoport, Y. Apeloig, Eds. (Wiley, 2001),
pp. 1027Ð1057.
27. Q. Wu et al., Angew. Chem. Int. Ed. 57, 9176Ð9179 (2018).
28. O. Allemann, S. Duttwyler, P. Romanato, K. K. Baldridge,
J. S. Siegel, Science 332, 574Ð577 (2011).
29. A. R. Bassindale, T. Stout, J. Organomet. Chem. 271, C1ÐC3
(1984).
30. Q.-A. Chen, H. F. T. Klare, M. Oestreich, J. Am. Chem. Soc. 138,
7868Ð7871 (2016).
Fig. 3. Comparison of carborane-stabilized silylium ions and reported carbenium ions. 31. Y. Xiong, S. Yao, M. Driess, Z. Naturforsch. B 68, 445Ð452 (2013).
32. N. Inamoto, S. Masuda, Chem. Lett. 11, 1003Ð1006 (1982).
33. L. Olsson, D. Cremer, Chem. Phys. Lett. 215, 433Ð443
(1993).
of the anion and the ionic character of silylium (R = Me, Et, iPr, tBu, tBu2Me) (14–16, 23). The 34. L. Olsson, C.-H. Ottosson, D. Cremer, J. Am. Chem. Soc. 117,
carborane salts 8 and 12. Accordingly, the re- secondary and primary analogs are similar in 7460Ð7479 (1995).
duced steric bulk around the silicon atom and both the carbenium and silylium ion series, fea- 35. M. Arshadi, D. Johnels, U. Edlund, C.-H. Ottosson, D. Cremer,
J. Am. Chem. Soc. 118, 5120Ð5131 (1996).
the formation of tighter contact ion pairs lead turing a tetracoordinate pyramidalized structure.
36. C.-H. Ottosson, D. Cremer, Organometallics 15, 5309Ð5320
to more pronounced pyramidalization at the sili- However, the carbenium carboranes show more (1996).
con atom, as indicated by the decreasing sum covalent bonding and halonium ion character 37. L. Omann et al., Angew. Chem. Int. Ed. 57, 8301Ð8305
of all C–Si–C angles from tBu3Si+[CHB11H5Br6]– that increases in the order iPr < Et < Me (42). (2018).
(348.7°) to 8 (347.7°) and 12 (342°). Notably, the smallest members of these reactive 38. M. D. Harmony, M. R. Strand, J. Mol. Spectrosc. 81, 308Ð315
(1980).
Intriguingly, the molecular structure of H3Si+ intermediates are distinctly different again: 39. E. S. Stoyanov, S. P. Hoffmann, M. Juhasz, C. A. Reed,
[CHB11H5Br6]– (14) differs markedly from these Whereas the methylium cation is covalent (43), J. Am. Chem. Soc. 128, 3160Ð3161 (2006).
alkyl-substituted tetracoordinate silylium carbo- the heavier H3Si+ analog is ionic and adopts a 40. I. Fleming, Chem. Soc. Rev. 10, 83Ð111 (1981).
rane salts (Fig. 2C, bottom). The H3Si+ cation is trigonal-planar (bipyramidal) geometry. Stabi- 41. T. Kato, C. A. Reed, Angew. Chem. Int. Ed. 43, 2908Ð2911
(2004).
coordinated by two counteranions via the bro- lization by two carborane counteranions through 42. T. Kato, E. Stoyanov, J. Geier, H. Grützmacher, C. A. Reed,
mine atoms from the pentagonal belt of the car- the typical coordination of the halide leads to a J. Am. Chem. Soc. 126, 12451Ð12457 (2004).
borane cluster, thereby forming linear polymeric H3Si+-bridged linear chain structure. This bond- 43. C. Bolli et al., Angew. Chem. Int. Ed. 49, 3536Ð3538
chains with a Br–Si–Br angle of 179.0°. Hence, the ing situation resembles exactly that of the cor- (2010).
44. S. Shaik, D. Danovich, W. Wu, P. C. Hiberty, Nat. Chem. 1,
pentacoordinate silicon atom is not pyramidal- responding Brønsted acid H+[CHB11H5Br6]– (39). 443Ð449 (2009).
ized but adopts a trigonal-planar (bipyramidal) Hence, the H3Si+ cation resembles a proton more
geometry with an average of 119.0° and a sum of than it resembles a methylium ion. AC KNOWLED GME NTS
356.9° for the H–Si–H bond angles. The two dif- We thank D. Stalke and A. N. Paesch for hosting the stay of
ferent Si–Br bond lengths of 2.477 Å and 2.770 Å RE FERENCES AND NOTES Q.W. to obtain an improved set of crystallographic data for
H3Si+[CHB11H5Br6]Ð (Georg-August-Universität Göttingen);
reveal an unsymmetrical H3Si+ bonding. How- 1. R. A. Moss, M. S. Platz, M. Jones Jr., Eds., Reactive
Q.-A. Chen for initial contributions to the dehydrogenation
ever, both distances are longer than those ob- Intermediate Chemistry (Wiley, 2004).
approach for silylium ion generation; L. Omann for sharing his
2. V. Ya. Lee, A. Sekiguchi, in Organosilicon Compounds,
served in the alkyl-substituted tetracoordinate V. Ya. Lee, Ed. (Academic Press, 2017), vol. 1, pp. 197Ð230.
expertise with Q.W.; and S. Kemper and J. D. Epping for expert
silylium carboranes, reflecting the weak and advice with the NMR spectroscopic measurements (all Technische
3. T. Müller, in Functional Molecular Silicon Compounds I,
Universität Berlin). Funding: Supported by an Alexander von
noncovalent nature of these Si–Br interactions. D. Scheschkewitz, Ed. (Springer, 2014), pp. 107Ð162.
Humboldt Foundation postdoctoral fellowship (2017Ð2019) (Q.W.)
Nonetheless, H3Si+[CHB11H5Br6]– is preferably 4. G. A. Olah, J. Org. Chem. 66, 5943Ð5957 (2001).
and by an Einstein Foundation Berlin endowed professorship
5. J. B. Lambert, Y. Zhao, Angew. Chem. Int. Ed. Engl. 36,
termed as a silanium ion, referring to a penta- 400Ð401 (1997).
(M.O.). The computational work was supported by the Deutsche
coordinate silicon cation (31). The solid-state Forschungsgemeinschaft (SFB 1109, project A3). Author
6. K.-C. Kim et al., Science 297, 825Ð827 (2002).
structure of H3Si+[CHB11H5Br6]– is reminiscent of 7. P. Jutzi, E.-A. Bunte, Angew. Chem. Int. Ed. Engl. 31, 1605Ð1607
contributions: All experiments were conducted by Q.W. X-ray
diffraction analysis was performed by E.I. Computations were run
Reed’s carborane acid H+[CHB11H5Br6]– (39), (1992).
by R.M. and led by M.K. H.F.T.K. and M.O. directed the project
8. T. Müller, P. Jutzi, T. Kühler, Organometallics 20, 5619Ð5628 (2001).
which suggests that H3Si+ can be viewed as a fat 9. C. Gerdes, W. Saak, D. Haase, T. Müller, J. Am. Chem. Soc. 135,
and co-wrote the manuscript. Competing interests: The authors
proton. That relation between silyl groups and declare no competing financial interests. Data and materials
10353Ð10361 (2013).
availability: X-ray crystallographic data are available free of
protons had already been established by Fleming 10. L. Albers, J. Baumgartner, C. Marschner, T. Müller, Chem. Eur. J.
charge from the Cambridge Crystallographic Data Centre:
nearly 40 years ago (40). 22, 7970Ð7977 (2016).
CCDC 1913125 for tBu2HSi+[CHB11H5Br6]Ð (8), CCDC 1913126
11. S. P. Fisher et al., Chem. Rev. acs.chemrev.8b00551 (2019).
A comparison of these silylium carboranes for tBuH2Si+[CHB11H5Br6]Ð (12), and CCDC 1913127 for
12. I. M. Riddlestone, A. Kraft, J. Schaefer, I. Krossing, Angew.
(14–16, 23) with related carbenium salts (41–43) H3Si+[CHB11H5Br6]Ð (14), respectively. All other data are
Chem. Int. Ed. 57, 13982Ð14024 (2018).
available in the main text or the supplementary materials.
once again reveals the difference between silicon 13. J. Y. Corey, J. Am. Chem. Soc. 97, 3237Ð3238 (1975).
and carbon (44) despite their proximity in the 14. C. A. Reed, Z. Xie, R. Bau, A. Benesi, Science 262, 402Ð404 (1993).
15. Z. Xie, R. Bau, A. Benesi, C. A. Reed, Organometallics 14, SUPPLEMENTARY MATERIALS
periodic table (Fig. 3). Whereas the tertiary 3933Ð3941 (1995).
tert-butyl cation exists as a free carbenium ion 16. Z. Xie et al., J. Am. Chem. Soc. 118, 2922Ð2928 (1996).
as a result of hyperconjugation and polariza- 17. J. B. Lambert, S. Zhang, C. L. Stern, J. C. Huffman, Science
Figs. S1 to S35
tion (41), these effects are impeded by the larger 260, 1917Ð1918 (1993).
Tables S1 to S5
18. J. B. Lambert, S. Zhang, S. M. Ciro, Organometallics 13,
size and higher electrophilicity of the silicon 2430Ð2443 (1994).
atom, leading to stabilization by the counter- 19. S. P. Hoffmann, T. Kato, F. S. Tham, C. A. Reed, Chem. 4 May 2019; accepted 17 June 2019
anion and pyramidalization in the case of R3Si+ Commun. 2006, 767Ð769 (2006). 10.1126/science.aax9184

CATTLE DOMESTICATION date from Mesolithic to early Islamic periods,

and despite poor preservation, which is typical
of the region, we obtained an average genome
Ancient cattle genomics, origins, and coverage of 0.9× (table S1).
The pattern of genetic variation in extant cat-
rapid turnover in the Fertile Crescent tle is well established. European B. taurus,
West African B. taurus, and B. indicus of South
Asian origin represent three distinct apices in
Marta Pereira Verdugo1*, Victoria E. Mullin1,2*, Amelie Scheu1,3*, Valeria Mattiangeli1, plotted principal components (PCs) (Fig. 1A).
Kevin G. Daly1, Pierpaolo Maisano Delser1,4, Andrew J. Hare1, Joachim Burger3, Geographically intermediate populations, such
Matthew J. Collins5,6, Ron Kehati7 , Paula Hesse8, Deirdre Fulton9, as Near Eastern and East African animals, fall
Eberhard W. Sauer10, Fatemeh A. Mohaseb11,12, Hossein Davoudi12,13,14, in genetically intermediate positions (7, 8, 10).
Roya Khazaeli12, Johanna Lhuillier15, Claude Rapin16, Saeed Ebrahimi17, Projecting ancient cattle genomes (provenance
Mutalib Khasanov18, S. M. Farhad Vahidi19, David E. MacHugh20,21, Okan Ertuğrul22, shown in Fig. 1B) against this genetic land-
Chaido Koukouli-Chrysanthaki23, Adamantios Sampson24, George Kazantzis25, scape (Fig. 1A), we observe that to the left of
Ioannis Kontopoulos5, Jelena Bulatovic26, Ivana Stojanović27, Abdesalam Mikdad28, PC1, earlier (Neolithic and Bronze Age) genomes
Norbert Benecke29, Jörg Linstädter30, Mikhail Sablin31, Robin Bendrey10,32,
fall in three geographical clusters (a, Balkans; b,
Anatolia/Iran; and c, southern Levant) along
Lionel Gourichon33, Benjamin S. Arbuckle34, Marjan Mashkour11,12,13, David Orton5,
with modern European and African B. taurus,
Liora Kolska Horwitz35, Matthew D. Teasdale1,5, Daniel G. Bradley1à
whereas B. indicus breeds are separated and
represented on the far right of the PC plot (Fig.
Genome-wide analysis of 67 ancient Near Eastern cattle, Bos taurus, remains reveals
1A). This suggests that cattle origins included
regional variation that has since been obscured by admixture in modern populations.
two divergent aurochs populations that formed
Comparisons of genomes of early domestic cattle to their aurochs progenitors identify
the basis of the B. indicus–B. taurus divide.
diverse origins with separate introgressions of wild stock. A later region-wide Bronze
Six ancient aurochs genomes, including four
Age shift indicates rapid and widespread introgression of zebu, Bos indicus, from
from the greater Near East, provide additional
the Indus Valley. This process was likely stimulated at the onset of the current geological
context: two ~9000-year-old samples from the
age, ~4.2 thousand years ago, by a widespread multicentury drought. In contrast to
Levantine Aceramic village of Abu Ghosh (Abu1
genome-wide admixture, mitochondrial DNA stasis supports that this introgression was
and Abu2), a 7500-year-old sample from the early
male-driven, suggesting that selection of arid-adapted zebu bulls enhanced herd
Anatolian settlement Çatalhöyük (Ch22), and a
survival. This human-mediated migration of zebu-derived genetics has continued through
7000-year-old Armenian aurochs (Gyu2) (11).
millennia, altering tropical herding on each continent.
These four genomes fall close to the Anatolia
T
and Iran ancient domestic cattle cluster (Fig. 1A,
he extinct Eurasian aurochs (Bos primigenius) from the Indus Valley region (6, 7). Zebu ge- cluster b) and reveal this as the oldest ancestral
was domesticated circa 10,500 years be- nomic influence is pervasive in modern Near stratum of B. taurus. The genomic signature of
fore present (yr B.P.) within the restricted Eastern herds (8). Two theories account for this earliest population has been obscured in mod-
locality of the Upper Euphrates and Tigris this: one suggests an origin from genomically ern Near Eastern cattle by later admixture. From
drainages of the Fertile Crescent (1, 2). How- intermediate Near Eastern aurochs, whereas a this group, we sequenced a well-preserved 8000-
ever, the true extent and nature of interactions second hypothesizes that these Near Eastern herds year-old Anatolian genome (Sub1) (11) to 13.5×
between humans and aurochs resulting in mod- resulted from an introgression of domestic zebu coverage and use this in D statistics testing for zebu
ern day domestic cattle are obscure. genomes into the region from the east, either in introgression in other ancient individuals (Fig. 2).
Mitochondrial DNA (mtDNA) diversity in mod- a discrete active process—perhaps responding B. indicus cattle are adapted to, and predo-
ern Bos taurus cattle suggests a highly restricted to climate fluctuation—or a passive diffusion minate in, modern arid and tropical regions of
initial domestic pool of ~80 females (3–5). How- over many millennia (9). the world (11). Zebu cattle originated circa
ever, a more complex relationship with wild To analyze now-obscured early cattle genome 8000 yr B.P. (12). However, despite archaeo-
populations is evidenced by introgression from strata from the region of B. taurus domestica- logical evidence for contact between civilizations
local aurochs into British cattle and the ge- tion, we retrieved genome-wide data from 67 of the Fertile Crescent region and the Indus
nomic divergence of B. indicus (zebu) cattle ancient bovines (including six aurochs). These Valley (9), the influence of the zebu genome is
1
Smurfit Institute of Genetics, Trinity College Dublin, Dublin D02 PN40, Ireland. 2Department of Earth Sciences, Natural History Museum, London SW7 5BD, UK. 3Palaeogenetics Group, Institute
of Organismic and Molecular Evolution (iOME), Johannes Gutenberg-University Mainz, 55099 Mainz, Germany. 4Department of Zoology, University of Cambridge, Cambridge CB2 3EJ, UK.
5
BioArCh, University of York, York YO10 5DD, UK. 6Museum of Natural History, University of Copenhagen, Copenhagen, Denmark. 7448 Shvil Hachalav Street, Nir Banim 7952500, Israel. 8Jewish
Studies Program, Department of Classics and Ancient Mediterranean Studies, The Pennsylvania State University, University Park, PA 16802, USA. 9Department of Religion, Baylor University,
Waco, TX 76798, USA. 10School of History, Classics and Archaeology, University of Edinburgh, Edinburgh EH8 9AG, UK. 11Archéozoologie et Archéobotanique (UMR 7209), CNRS, MNHN, UPMC,
Sorbonne Universités, Paris, France. 12Bioarchaeology Laboratory, Central Laboratory, University of Tehran, 1417634934 Tehran, Iran 13Osteology Department, National Museum of Iran,
1136918111 Tehran, Iran. 14Department of Archaeology, Faculty of Humanities, Tarbiat Modares University, 111-14115 Tehran, Iran. 15Archéorient Université Lyon 2, CNRS UMR 5133, Maison de
l’Orient et de la Méditerranée, 69365 Lyon, France. 16Archéologie d’Orient et d’Occident (AOROC, UMR 8546, CNRS ENS), Centre d'archéologie, 75005 Paris, France. 17Faculty of Literature and
Humanities, Islamic Azad University, 1711734353 Tehran, Iran. 18Uzbekistan Institute of Archaeology of the Academy of Sciences of the Republic of Uzbekistan, 703051 Samarkand, Uzbekistan.
19
Agricultural Biotechnology Research Institute of Iran-North branch (ABRII), Agricultural Research, Education and Extension Organization, 4188958883 Rasht, Iran. 20Animal Genomics
Laboratory, UCD School of Agriculture and Food Science, University College Dublin, Dublin D04 V1W8, Ireland. 21UCD Conway Institute of Biomolecular and Biomedical Research, University
College Dublin, Dublin D04 V1W8, Ireland. 22Veterinary Faculty, Ankara University, 06110 Ankara, Turkey. 23Hellenic Ministry of Culture and Sports, Department of Prehistoric and Classical
Antiquities, and Museums, Serres 62 122, Greece. 24Department of Mediterranean Studies, University of the Aegean, 85132 Rhodes, Greece. 25Archaeological Museum of Aeani, 500 04, Kozani,
Western Macedonia, Greece. 26Laboratory for Bioarchaeology, Department of Archaeology, University of Belgrade, 11000 Belgrade, Serbia. 27Institute of Archaeology, 11000 Belgrade, Serbia.
28
Institut National des Sciences de l‘Archéologie et du Patrimoine de Maroc (INSAP) Hay Riad, Madinat al Ifrane, Rabat Instituts, 10000 Rabat, Morocco. 29Department of Natural Sciences,
German Archaeological Institute, 14195 Berlin, Germany. 30Deutsches Archäologisches Institut, Kommission für Archäologie Außereuropäischer Kulturen (KAAK), 53173 Bonn, Germany.
31
Zoological Institute of the Russian Academy of Sciences, 199034 St Petersburg, Russia. 32Department of Archaeology, University of Reading, Reading RG6 6AB, UK 33Université Côte d'Azur,
CNRS, CEPAM (UMR 7264), 06357 Nice, France. 34Department of Anthropology, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA. 35National Natural History Collections,
Faculty of Life Sciences, The Hebrew University, 9190401 Jerusalem, Israel.
*These authors contributed equally to this work. †Independent researcher
‡Corresponding author. Email: dbradley@tcd.ie

detectable in ancient Southwest Asian cattle change event (9, 15–17). This multicentury drought tectable B. indicus influence. Third, it was plausi-
only 4000 years later (Fig. 2). However, after coincided with empire collapse in both Mesopo- bly driven by bull choice, as we observe up to
~4000 yr B.P., hybrid animals (median 35% in- tamia and Egypt as well as a decline in the Indus ~70% autosomal genome change but a retained
dicine ancestry) are found across the Near East, civilization and has been accepted as the bound- substratum of B. taurus mtDNA haplotypes (Fig. 2
from Central Asia and Iran to the Caucasus and ary defining the onset of our current geological and table S3). Hybrid B. taurus–B. indicus herds
Mediterranean shores of the southern Levant age, the Meghalayan (18). may have enabled the survival of communities
(table S2 and fig. S1). During this period, de- Three features of this zebu influx after ~4000 yr under stress and perhaps facilitated expansion of
pictions and osteological evidence for B. indicus B.P. attest that the influx was likely driven by herding into more-peripheral regions. Restocking
also appear in the region (9, 13). In contrast to adaptation and/or human agency rather than after herd decline may have also been a factor.
autosomal data, but similar to earlier work (14), passive diffusion. First, the extent of indicine Westward human migration has been documented
we find persistence of B. taurus mitochondria, introgression does not follow a simple east-to- around this time (19, 20) along with archaeological
suggesting introgression may have been medi- west gradient; for example, it is pronounced in evidence for the appearance of other South Asian
ated by bulls (Fig. 2). Levantine genomes from the western edge of the taxa such as water buffalo and Asian elephants in
This sharp influx may have been stimulated Near East. Second, the introgression was wide- the Near East (21), suggesting the movement of
by the onset of a period of increased aridity spread and took place in a relatively restricted large animals by people.
known as the 4.2-thousand-year abrupt climate time interval after four millennia of barely de- Before zebu admixture, ancient southern
Levantine animals occupy a distinctive space
within the PC plot (Fig. 1A, cluster c), toward
A modern African cattle and adjacent to a 9000-
yr-B.P. Epipalaeolithic Moroccan aurochs (Th7).
A 7000-yr-B.P. Mesolithic British aurochs ge-
European Bos taurus Bos indicus nome (CPC98) (6) also plots away from the core
Anatolia/Iran ancestral Near Eastern cluster and
close to Neolithic Balkan (cluster a) and modern
European cattle. These genetic affinities in ancient
a
cattle suggest an early secondary recruitment
from diverse wild populations.
Concordantly, D statistic tests of allele sharing
b d by cattle population pairs with three divergent
Near Eastern aurochs confirm that early cattle exhibit asym-
metric relationships with different wild pop-
African Bos taurus x indicus ulations (Fig. 3). The most extreme deviations
are found in comparisons featuring the B. taurus
Levantine population (Fig. 1, cluster c); these
c
share the least affinity with the British and
Bos primigenius Anatolia Armenian aurochs (z-score > 5.67; P < 10−5) but
Domesticated Bos Balkans more with the Moroccan Epipalaeolithic sam-
Neolithic Georgia ple. We infer that a distinct strain of aurochs,
Bronze Age Central Asia probably from the Levant and similar to those
Iron Age Iran ranging across North Africa, had considerable
Post Iron Age Iraq input into early cattle in the southern Levant. The
African Bos taurus Modern Southern Levant Mesolithic British aurochs also shows asymmet-
ric affinity with the Neolithic Balkans samples,
implying that the hybridization of European
aurochs (6) was initiated more than 7000 years
B ago, close to the onset of human herding of cattle
in Europe. These findings are supported by a
qpgraph analysis (figs. S2 and S3) and cannot be
explained by cattle-into-aurochs admixture, as
both the British and Moroccan aurochs have
securely predomestic dates. Although each of
these three aurochs have divergent mtDNA
haplotypes falling outside normal B. taurus
variation, ancient domesticates display typical
modern domestic haplotypes (fig. S4). This
points toward common matrilineal origins for
domestic taurine cattle and away from an ar-
chaeologically less parsimonious interpreta-
tion that our observed ancient genetic structure
may have arisen from separate domestications;
Fig. 1. Procrustes projection principal components analysis of ancient cattle. (A) Ancient it also suggests that introgression may have
animals are projected on modern 770K Bovine single-nucleotide polymorphism (SNP) genotypes, been via mating with wild males. Sexually ma-
shown as background gray asterisks (figs. S1 and S2). Four clusters are highlighted: Neolithic ture bulls, because of size and aggression, were
Balkans, which plot with modern Europeans (a); a group of mainly Anatolian and Iranian cattle close likely the most dangerous stock in Neolithic
to four aurochs from the Near East (b); and Levantine cattle that fall into two groups, a cluster of villages, and thus unsupervised field insemi-
earlier samples (c) and a cluster of later samples (d) close to contemporary Near Eastern cattle with nation by aurochs bulls may have played a role
B. indicus admixture. (B) Approximate geographical distribution of ancient sample sites. in early herd management (22).

Fig. 2. Zebu introgression through time in

the greater Near East. Whole genome D
0.3 Anatolia _
Z>|3|
Balkans _
Z<|3|
statistic calculations with a gaur (B. gaurus) as Bos primigenius
an outgroup and the Neolithic Anatolian Central Asia
domestic genome Sub1 as a representative 0.2 Georgia
Iran Sub1 ancient B.indicus B.gaurus
nonadmixed individual (see inset). Lower Iraq
coverage samples with <200 informative sites Levant
were excluded. A step change in zebu intro- 0.1
gression is apparent circa 4000 yr B.P. MtDNA
D
counts of taurine (gray) and the single
zebu (black) mtDNA in ancient domestic 0
cattle are graphed (bottom, with a shared
time axis) in 250-year intervals. -0.05
10
B.taurus mtDNA
5
B.indicus mtDNA
5
9000 8000 7000 6000 5000 4000 3000 2000 1000 0
Years BP
Fig. 3. Clade integrity of ancient popula-

A B C
tion pairs with respect to aurochs intro-
gression. (A) The test D(gaur, aurochs;
ancient group1, ancient group2) reveals
asymmetric affinities of aurochs genomes -
with pre-4000-yr-B.P. cattle populations. +
Levantine cattle show reduced allele sharing Ancient Ancient Aurochs Gaur
relative to other populations with the domestic domestic
Armenian (Gyu2) and British (CPC98)
Gyu2 CPC98 Th7
aurochs but more with the Moroccan aurochs
Levant Anatolia
(Th7). Balkan cattle show asymmetric
affinities with the British aurochs. Bars denote
two standard errors. (B) Geographical Levant Iran
location of aurochs tested. (C) Distribution
of ancient domestic cattle groups tested; post- Levant Balkans
4000-yr-B.P. Near Eastern samples were
excluded because of their zebu admixture. Anatolia Balkans
Iran Balkans
Anatolia Iran
-0.2 -0.1 0.0 0.1 0.2 -0.2 -0.1 0.0 0.1 0.2 -0.2 -0.1 0.0 0.1 0.2
D D D
Armenian aurochs British aurochs Moroccan aurochs
Distinct genotypes and phenotypes in B. taurus region gained heterogeneous inputs from di- 3. C. S. Troy et al., Nature 410, 1088–1091 (2001).
cattle native to Africa, such as tolerance of trop- verse aurochs strains, including contributions 4. R. Bollongino et al., Mol. Biol. Evol. 29, 2101–2104 (2012).
5. A. Scheu et al., BMC Genet. 16, 54 (2015).
ical infections, have been attributed to either specific to European and African cattle ances- 6. S. D. E. Park et al., Genome Biol. 16, 234 (2015).
local domestication or introgression from Afri- tors. After ~4200 yr B.P., gross genome turn- 7. Bovine HapMap Consortium, Science 324, 528–532
can aurochs (10, 23). However, ancient Levan- over reflecting the spread of B. indicus, and (2009).
8. R. T. Loftus et al., Mol. Ecol. 8, 2015–2022 (1999).
tine genome affinity with North African aurochs likely associated with climate change, was ef- 9. R. Matthews, Antiquity 76, 438–446 (2002).
hints that this distinctiveness may have origins fected by cattle herders throughout southwest 10. D. E. MacHugh, M. D. Shriver, R. T. Loftus, P. Cunningham,
in the southern Fertile Crescent. Supporting this, and central Asia, representing the start of a global D. G. Bradley, Genetics 146, 1071–1086 (1997).
the B. taurus mtDNA haplogroup (T1), which is B. indicus genome diaspora (25) that continues 11. See the supplementary materials.
12. A. K. Patel, R. H. Meadow, in The Oxford Handbook of
almost fixed in African cattle populations (24), is today. Zooarchaeology, U. Albarella, M. Rizzetto, H. Russ, K. Vickers,
the most frequent in the southern Levant, includ- S. Viner-Daniels, Eds. (Oxford Univ. Press, 2017), pp. 280–303.
ing earliest samples, but was not found among RE FERENCES AND NOTES 13. C. Grigson, Levant 47, 5–29 (2015).
other ancient domesticates (table S3). 1. D. Helmer, L. Gourichon, H. Monchot, J. Peters, M. S. Segui, 14. M. Meiri et al., Sci. Rep. 7, 701 (2017).
in The First Steps of Animal Domestication, J.-D. Vigne, 15. M. H. Wiener, Radiocarbon 56, S1–S16 (2014).
B. taurus were initially derived from a re- J. Peters, D. Helmer, Eds. (Oxbow Books, 2005), pp. 86–95. 16. A. Sharifi et al., Quat. Sci. Rev. 123, 215–230 (2015).
stricted northern Fertile Crescent genetic back- 2. H. Hongo, J. Pearson, B. Öksüz, G. Ilgezdi, Anthropozoologica 17. M. Staubwasser, H. Weiss, Quat. Res. 66, 372–387 (2006).
ground, but early domestic cattle outside this 44, 63–78 (2009). 18. M. J. C. Walker et al., J. Quat. Sci. 27, 649–659 (2012).

19. M. Haber et al., . Am. J. Hum. Genet. 101, 274–282 (2017). G. Grabež, J. Kuzmanović-Cvetković, A. Bulatović, M. Spasić, research, with input from J.B. and M.J.C.; M.P.V., V.E.M., A.S., V.M.,
20. I. Lazaridis et al., Nature 536, 419–424 (2016). E. Rosenstock, K. Kaniuth, and B. Eivers. We thank the Israel K.G.D., M.D.T., and A.J.H. performed ancient DNA laboratory work;
21. C. Çakırlar, S. Ikram, Levant 48, 167–183 (2016). Antiquities Authority for permitting sampling of the Israeli sites M.D.T. and V.M. performed modern DNA laboratory work; M.P.V.
22. J. Peters, B. S. Arbuckle, N. Pöllath, in The Neolithic in Turkey, (under permit). We thank the Çatalhöyük Research Project and and V.E.M. performed the computational analyses with input from
M. Özdogan, N. Baflgelen, P. Kuniholm, Eds. (Archaeology and Ministry of Culture and Tourism of the Republic of Turkey. We D.G.B., M.D.T., P.M.D., and K.G.D.; other coauthors provided data
Art Publications, 2012), vol. 6, pp. 1–65. thank the Iranian Cultural Heritage Handicraft and Tourism and samples. D.G.B., M.P.V., V.E.M., A.S., and M.D.T. wrote the
23. J. E. Decker et al., PLOS Genet. 10, e1004254 (2014). organization and the National Museum of Iran (NMI) (J. Nokandeh, paper and supplementary information with input from all
24. D. G. Bradley, D. E. MacHugh, P. Cunningham, R. T. Loftus, director, and F. Biglari, cultural deputy). We are grateful to H. Laleh coauthors. Competing interests: The authors declare that they
Proc. Natl. Acad. Sci. U.S.A. 93, 5131–5135 (1996). and A. Aliyari, directors of the Bioarchaeology Laboratory, have no competing interests. Data and materials availability:
25. N. Chen et al., Nat. Commun. 9, 2337 (2018). Central Laboratory, University of Tehran. The ATM Project of MNHN Raw reads have been deposited at the European Nucleotide
26. dblabTCD, dblabTCD/NE_Cattle_2019: NE_Cattle_2019_v1.0, Version supported sampling of several sites as well as the LIA HAOMA Archive (ENA) with project number PRJE31621. The code used in
v1.0, Zenodo (2019); http://doi.org/10.5281/zenodo.3206663. CNRS project. The participation of ZIN RAS (state assignment this study for manipulating sequence files can be found at
AAAA-A17-117022810195-3) to this research is acknowledged. Zenodo (26). Mitochondrial phylogenies and modern SNP data
ACKN OW LEDG MEN TS Funding: Supported by ERC Investigator grant 295729-CodeX. are available at https://osf.io/vne7t/.
We thank L. Frantz, G. Larson, and M. Djamali for valuable Additional support from Science Foundation Ireland Award 12/
suggestions. We thank L. Cassidy for helpful discussions and ERC/B2227, Trinseq and the SFI/HEA Irish Centre for High-End SUPPLEMENTARY MATERIALS
scripts, and R. Verdugo for contributions in figure design. We thank Computing (ICHEC). M.D.T. was partially supported by the Marie
excavators, archaeozoologists, and museums who permitted Skłodowska-Curie Individual Fellowship H2020-MSCA-IF-2016
sampling of bones from their excavations and collections without 747424 (SCRIBE). V.E.M. was partially supported by the NERC
Figs. S1 to S14
which this project would not have been possible, including E. Galili, Grant NE/P012574/1. M.J.C. acknowledges support from DNRF128.
Tables S1 to S13
T. Levy, C. Grigson, R. Gophna, A. Maeir, S. Gitin, A. Ben-Tor, P.M.D. was supported by the HERA Joint Research Programme
D. Master, E. van den Brink, S. Cohen, E. Rosenstock, P. Biehl, “Uses of the Past” (CitiGen) and the European Union’s Horizon
I. Hodder, J.J. Roodenberg, S. Alpaslan Roodenberg, D. Helmer, 2020 research and innovation program (649307). Author 28 August 2018; accepted 14 June 2019
H. Greenfield, J. Vuković, M. Radivojević, B. Roberts, M. Marić, contributions: D.G.B. conceived of the project and designed 10.1126/science.aav1002
REPRODUCTION
times as pairs (syncytin-1 and -2 in humans and
syncytin-A and -B in mice) (3, 8, 9), occurred inde-
IFITM proteins inhibit placental pendently from different ERVs in diverse mamma-
lian lineages 10 to 85 million years ago. Knocking
syncytiotrophoblast formation out the syncytin-A gene or both syncytin-A and

syncytin-B genes leads to fetal growth restriction
and mid-gestational lethality (8, 9).
and promote fetal demise The immune-related IFN-induced transmem-
brane proteins IFITM1, IFITM2, and IFITM3 are
restriction factors that protect uninfected cells
Julian Buchrieser1,2*†, Séverine A. Degrelle3,4,5*, Thérèse Couderc6,7*, Quentin Nevers1,2*,
from viral infection. They block the entry into
Olivier Disson6,7, Caroline Manet8, Daniel A. Donahue1,2, Françoise Porrot1,2,
host cells of many enveloped viruses by inhibit-
Kenzo-Hugo Hillion9, Emeline Perthame9, Marlene V. Arroyo1,2,10, Sylvie Souquere11, ing virus–cell fusion at the hemifusion stage
Katinka Ruigrok12, Anne Dupressoir13,14, Thierry Heidmann13,14, Xavier Montagutelli8, (10–12) and act by altering the biophysical prop-
Thierry Fournier3,4‡, Marc Lecuit6,7,15‡, Olivier Schwartz1,2,16†‡ erties or cholesterol content of the cellular mem-
branes in which they are found (11–13). IFITMs
Elevated levels of type I interferon (IFN) during pregnancy are associated with intrauterine modify the rigidity and/or curvature of the mem-
growth retardation, preterm birth, and fetal demise through mechanisms that are not branes without evidence of a direct interaction
well understood. A critical step of placental development is the fusion of trophoblast cells with the fusogenic viral envelopes (11, 12). Be-
into a multinucleated syncytiotrophoblast (ST) layer. Fusion is mediated by syncytins, cause of different sorting motifs, IFITM1 mostly
proteins deriving from ancestral endogenous retroviral envelopes. Using cultures of human
trophoblasts or mouse cells, we show that IFN-induced transmembrane proteins (IFITMs), 1
a family of restriction factors blocking the entry step of many viruses, impair ST formation Virus and Immunity Unit, Department of Virology, Institut
Pasteur, Paris, France. 2CNRS-UMR3569, Paris, France.
and inhibit syncytin-mediated fusion. Moreover, the IFN inducer polyinosinic:polycytidylic 3
INSERM, UMR-S1139, Faculté de Pharmacie de Paris,
acid promotes fetal resorption and placental abnormalities in wild-type but not in Paris, France. 4Université Paris Descartes, Sorbonne Paris
Ifitm-deleted mice. Thus, excessive levels of IFITMs may mediate the pregnancy complications Cité, Paris, France. 5Inovarion, Paris, France. 6Institut
observed during congenital infections and other IFN-induced pathologies. Pasteur, Biology of Infection Unit, Paris, France. 7Institut
National de la Santé et de la Recherche Médicale U1117,
T
Paris, France. 8Mouse Genetics Laboratory, Department of
he placenta is the primary maternal–fetal centa, the mouse placenta has a labyrinthic struc- Genomes and Genetics, Institut Pasteur, Paris, France.
9
Institut Pasteur, Bioinformatics and Biostatistics Hub,
barrier, achieving metabolic exchanges, ture and exhibits two ST layers that separate fetal C3BI, USR 3756 IP CNRS, Paris, France. 10Department of
hormone production, and protection from capillaries from maternal blood (4). In gestating Biochemistry and Molecular Biophysics and Department of
pathogens and the maternal immune sys- mice, interferon-beta (IFN-b) triggered by Zika or Microbiology and Immunology, Howard Hughes Medical
tem. The placental tissue stems from em- other viruses or by administration of polyinosinic: Institute, Columbia University, New York, NY 10032, USA.
11
Plateforme de Microscopie Electronique Cellulaire, UMS
bryonic cytotrophoblasts. The structural and polycytidylic acid (Poly-IC), promotes IUGR and AMMICA, Gustave Roussy, Villejuif, France. 12Institut Pasteur,
functional unit of human placenta is the chori- fetal demise (2, 5–7). IFN-b signaling leads to ab- Structural Virology Unit and CNRS UMR3569, Paris, France.
13
onic villous, which includes proliferative mono- normal labyrinth and ST structures, with the pres- Unité Physiologie et Pathologie Moléculaires des Rétrovirus
nuclear villous cytotrophoblasts (VCTs) at the ence of many unfused cells and reduced fetal Endogènes et Infectieux, CNRS UMR 9196, Gustave Roussy,
Villejuif, France. 14UMR 9196, Université Paris-Sud, Orsay,
basement and the syncytiotrophoblast (ST) layer blood vessels (7). France. 15Paris Descartes University, Department of
at the surface. The multinuclear ST arises from Cytotrophoblast fusion is mediated by en- Infectious Diseases and Tropical Medicine, Necker-Enfants
the differentiation and fusion of VCTs. In hu- velope glycoprotein (env)–derived genes of en- Malades University Hospital, APHP, Institut Imagine, Paris,
mans, an abnormal ST is observed in pregnancy dogenous retroviruses (ERVs) that have been France. 16Vaccine Research Institute, Créteil, France.
*These authors contributed equally to this work.
pathologies including intrauterine growth retar- captured by mammals (3). These genes were †Corresponding author. Email: julian.buchrieser@pasteur.fr
dation (IUGR), preeclampsia, lupus, and Down termed “syncytins” on the basis of their fusogenic (J.B.); olivier.schwartz@pasteur.fr (O.S.)
syndrome (1–3). In contrast to the human pla- capacity. The capture of syncytin genes, some- ‡These authors contributed equally to this work.

Fig. 1. IFITMs inhibit A 293T 100 ns

+Syncytin-1
% Fusion (Area of GFP)

syncytin-mediated cell
GFP1-10 GFP11
fusion. (A) Left: 293T- Control IFITM2 80
GFP1-10 and -GFP11 were
60
cocultured at a 1:1 ratio
and cotransfected with Transfection
40
syncytin-1 and IFITM or
Syncytin-1 ± IFITMs
control plasmids. Cell 20
fusion was quantified by
measuring the GFP+ area 0
M lm
IF M3 1
IF 1
M Ub
l
IF l
3Δ 3
3
tro
tro
IT 1-2
M
M
IT TM
2+
a
at 18 hours. Middle: Rep-
IT
IT Δ
IF Δ p
on
on
IT
1+
IF IFI
GFP+
3
resentative images of
IT
IF
GFP-Split 293T cells cotrans-
+ Syncytin-1
fected with syncytin-1 and
control or IFITM2 expression B C
plasmids. GFP areas are out- BeWo
MEF
(Normalised CPRG Signal)

ns
Gal-α Gal-ω 1.0 40
lined in white. Scale bar, ± IFITMs ± IFITMs WT or IfitmDel
Syncytia per well

200 mm. Right: Quantifica-
Fusion Index
30
tion of GFP areas. Results
are shown as mean ± SD 20
0.5
from four independent Forskolin
experiments. (B) Left: ± mIfitm3 10
BeWo b-Gal-a and b-Gal-w GFP Syncytin-A
0
were transduced with 0.0
el
el
T
T
W
W
D
D
IFITM or control vectors.
lin
l
tm
tm
ro
M
β-Gal+
M
ko
t
IT
on
IT
IT
Ifi
Ifi
Cells were cocultured at a
rs
IF
IF
IF
C
Fo
1:1 ratio and fusion was Control +mIfitm3
o-
N
induced by forskolin for
48 hours. Right: Fusion index (b-Gal activity) measured with a colorimetric (chlorophenol red-b-D-galactopyranoside, CPRG) assay. Results are shown as
mean ± SD from four independent experiments. (C) Left: WT or IfitmDel MEFs were cotransfected with GFP and syncytin-A plasmids and syncytia were
quantified after 24 hours. Right: Quantification of syncytia (cells with more than three nuclei) per well. Results are shown as mean ± SD from three to six
independent experiments. Statistical analysis: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, one-way analysis of variance (ANOVA); ns, not significant.
resides at the plasma membrane, whereas IFITM2 ubiquitination motifs, respectively, and accumu- their ability to produce syncytia-independent
and IFITM3 accumulate in the endolysosomal late in the cell (15, 16), were more active at in- b-Gal (fig. S3 A to D). Similar results were ob-
compartment after transiting at the cell surface. hibiting fusion than was wild-type (WT) IFITM3. served with the GFP-based system. Videomicro-
IFITM proteins are expressed at various basal By contrast, an IFITM3Dpalm mutant, whose scopy analysis showed delayed and reduced
levels in different cell types and are up-regulated levels are reduced (15, 16), was inactive. The fusion with each IFITM (fig. S2, C to E, and
by IFNs and other cytokines (11). In addition to coexpression of three IFITMs strongly inhibited movie S1).
their antiviral activity, the physiological cellular fusion (Fig. 1A). IFITMs similarly inhibited fusion We next tested the capacity of endogenous
functions of IFITM proteins remain poorly char- mediated by syncytin-2 (fig. S1A). IFITMs did not IFITMs to block fusion in mouse embryonic
acterized. Transgenic mice in which the cluster decrease the levels of syncytin-1 in transfected fibroblasts (MEFs) derived from WT or IfitmDel
of Ifitm genes is knocked out (hereafter referred cells (fig. S1B). To distinguish the effect of IFITMs mice (14). To this aim, MEFs were cotransfected
to as IfitmDel mice) are more sensitive to various in syncytin-1+ (donor) cells from that in target with syncytin-A and GFP plasmids (Fig. 1C and
viral infections (12) but exhibit no overt abnor- cells (expressing SLC1A5, the syncytin-1 receptor), fig. S4A). Syncytin-A induced numerous and large
malities apart from a slight overweight (14). we used an HIV Tat and long terminal repeat GFP+ syncytia (with up to 20 nuclei) in IfitmDel
Little is known about the impact of IFITMs on luciferase transactivation coculture system, in MEFs, and fewer and smaller syncytia in WT
fusion events mediated by cellular proteins or by which cell fusion generates luciferase activity MEFs. Transduction of IfitmDel MEFs with a
ERV-derived env proteins. Here, we investigated (13). IFITMs inhibited fusion when present in murine Ifitm3 vector (fig. S4B) restored resist-
whether IFITMs impair syncytin-mediated fusion either donor or target cells (fig. S1C). ance to fusion (Fig. 1C and fig. S4A). WT MEFs
and thus affect fetal development. We next examined the effect of IFITMs on were poorly sensitive to fusion, likely due to high
We first investigated whether IFITMs inhibit endogenous syncytins. In trophoblast-like BeWo basal IFITM3 levels (fig. S4B).
cell fusion mediated by exogenously expressed human choriocarcinoma cells, addition of the We then evaluated the effect of type I IFN and
human syncytins. We generated 293T cells car- adenylate cyclase activator forskolin triggers IFITMs in primary human trophoblasts. We first
rying a green fluorescent protein (GFP)–Split syncytin-1 production and syncytium formation asked whether IFITMs were up-regulated by
complementation system, in which two cells (17). To quantify fusion, we generated BeWo cells type I IFN in human placental explants. It has
separately produce half of the reporter protein, carrying two complementation systems based on been reported that in human midterm placental
generating a signal only upon fusion (Fig. 1A). either b-galactosidase-a/w (b-Gal-a/w) (Fig. 1B) chorionic villi explants, addition of IFN-b, but
The extent of fusion was quantified by measuring or GFP-Split (fig. S2C). These cells were then not IFN-l, leads to defects including cellular
the GFP+ area (Fig. 1A). Transfection of syncytin-1 transduced with lentiviral vectors to express a damage, decreased production of human chori-
induced the appearance of multinucleated GFP+ control protein or FLAG-tagged IFITM1-3 (fig. onic gonadotropin (hCG), and the appearance
cells. Fusion was significantly decreased when S2, A and D). As expected, forskolin triggered of syncytial knots (7). In such explants, IFN-b
syncytin-1 was cotransfected with FLAG-tagged fusion and production of b-Gal, which was de- up-regulated hundreds of IFN-stimulated genes
IFITM1, 2 or 3, but not with a control plasmid tected in fixed cells and quantified in cell lysates (ISGs), including IFITM1 (7). We took advantage
(Fig. 1A). The extent of inhibition of fusion varied (Fig. 1 B and fig. S2B). The fusion efficacy was of this large set of published RNA data to in-
between individual IFITMs. The IFITM3D1-21 and significantly reduced by IFITMs. The presence vestigate the expression of IFITMs, syncytins,
IFITM3Dub mutants, which lack endocytic and of IFITMs did not modify BeWo growth nor alter and their receptors. IFITMs were up-regulated by

Fig. 2. IFN-b inhibits A Control GW9662 B

fusion of primary ST
(% of GATA3-negative nuclei)
Mononucleated
human VCTs. 100
VCTs
(A) Left: Quantification
(GATA3+)
Fusion index (%)

of fusion. Mononucleated 80
VCTs express the nuclear ST 60
GATA3 protein, which
is down-regulated after 40
ST
fusion. Right: After 20
IFN-β or 24h
48 hours, nuclei were GW9662 ST
ST
visualized by 4′,6- 0
72h
0U
U
l
62
tro
diamidino-2-phenylindole 50 µm 50 µm
00
GATA3 E-Cad DAPI
96
10
on
48h
10
W
C
(DAPI) and plasma
G
100U IFN-β 1000U IFN-β IFN-β
membranes with Spontaneous fusion C
E-cadherin. Repre- 1.5
sentative images are
Normalized hCG
shown. Syncytia,
secretion
1.0
defined as large cells ST
containing multiple
GATA3– nuclei, are ST
0.5
outlined in white. ST ST
(B) The fusion index Syncytia
(GATA3-) 0.0
was quantified as the
0U
U
W l
62
tro
00
96
10
ST
on
percentage of GATA3-
10
50 µm 50 µm
G
negative nuclei. (C) hCG IFN-β
levels in culture super-
natants at 72 hours. Results are shown as mean ± SD from eight independent experiments. Statistical analysis: ****P < 0.0001, one-way ANOVA.
IFN-b compared with mock- and IFN-l–treated expressing IFITM3 or GFP) precluded precise We next performed histological analysis of
explants, whereas syncytins and their receptors quantification of syncytia. However, a visual the placentas to examine the consequences of
were not modulated (fig. S5). examination indicated that IFITM3+ cells re- Poly-IC treatment. Placentas from WT and
These observations were made in whole pla- mained mononucleated, even in the vicinity of IfitmDel fetuses were stained for E-cadherin
cental explants, in which nontrophoblastic cells large syncytia, whereas GFP-transfected cells to label trophoblasts (Fig. 3D). As expected, ST
are also present. We thus studied the impact of were able to fuse (fig. S7). These results dem- formation was detectable in untreated placenta
IFN-b on primary VCTs isolated from eight term onstrate that IFN-b induces IFITMs in VCTs from both WT and IfitmDel fetoplacental units
human placentas. VCTs can be cultured for up to and that both IFN-b and IFITMs inhibit ST (arrows). By contrast, with Poly-IC, less ST was
3 days and spontaneously differentiate in multi- formation. detected in WT placentas, whereas it was still
nucleated ST (18). Treating VCTs with IFN-b (100 We then evaluated the role of IFITMs in preg- present in IfitmDel placentas. We next triple
or 1000 IU/mL for 48 hours) strongly up-regulated nant mice. Administration of Poly-IC has been stained placenta 14 hours after Poly-IC in-
IFITM1 and IFITM2-3, as shown by immuno- extensively used as a model of type I IFN induc- jection for IFITM3, E-cadherin, and CD31 (a
fluorescence and Western blot (fig. S6, A and tion, triggering fetal growth retardation and marker of endothelial cells) (fig. S9D). As ex-
C). IFN-b slightly up-regulated RNA levels of resorption (2, 6, 7). We thus tested the effect of pected, IFITM3 was up-regulated in placentas
syncytin-2 but did not affect those of syncytin-1 Poly-IC in gestating WT, Ifnar−/− or IfitmDel of Poly-IC-treated WT mice, but not in those
and syncytin receptors (fig. S6D). We next quan- dams that had been mated to males of the same of IfitmDel mice. IFITM3 overexpression upon
tified VCT fusion, using a method based on the genotypes (Fig. 3A). Poly-IC was injected at em- Poly-IC injection was observed in CD31+ E-
detection of the transcription factor GATA3 by bryonic day 7.5 (E7.5), a time point that shortly cadherin– cells (endothelium) and CD31– E-
immunofluorescence and its down-regulation precedes ST formation (~E8.5) (4). A dose of cadherin+ cells (trophoblasts), demonstrating
upon ST differentiation (18). Cells were also 200 mg, which results in resorption of almost all that IFITM3 is induced in trophoblasts in re-
stained for E-cadherin to visualize plasma mem- fetuses within 48 hours (7), was injected. With sponse to Poly-IC, consistent with in vitro results.
branes. IFN-b inhibited VCT fusion to the same Poly-IC, WT fetuses, but neither Ifnar−/− nor Altogether, these results strongly suggest that
extent as GW9662, a peroxisome proliferator– IfitmDel fetuses, were resorbed at E9.5 (Fig. 3B). IFITM up-regulation in trophoblasts inhibits
activated receptor gamma (PPARg) antagonist The fetus size was similar in Ifnar−/− and IfitmDel ST formation in vivo, which likely contributes to
known to block this process (18) (Fig. 2, A and B). mice (Fig. 3C and fig. S8) and slightly reduced type I IFN-associated placental dysfunction and
Because it acts independently of IFN signaling, compared with untreated animals (Fig. 3C and fetal demise in mice.
GW9662 did not induce IFITMs (fig. S6, A and fig. S8). We confirmed that IfitmDel and WT In this study, we have uncovered a previously
C). The effect of IFN-b on ST differentiation was mice similarly responded to Poly-IC, as shown unknown function for the IFITM family of anti-
supported by a decreased release of hCG, which by the induction of various ISGs (Irf 7, Oas1a, viral restriction factors. We report that IFITMs
is secreted by ST but not by VCTs (Fig. 2C). The Stat1) measured in the liver 14 hours after injec- impair the fusogenic activity of syncytins required
reduction of fusion was not due to a cytotoxic tion, indicative of a systemic inflammation (fig. for ST formation and maintenance. Our results
effect of IFN-b, because levels of an apoptosis S9B). A local response was also observed in pla- provide a possible molecular explanation for pla-
marker (cCK18) were not augmented (fig. S6, B cental extracts of WT mice, which up-regulated cental dysfunctions associated with IFN-mediated
and C). Ifitm1, Ifitm3, and Irf7 RNAs. As expected, IfitmDel disorders such as IUGR and “TORCH” infections
To further assess the effect of IFITMs in this placentas did not express Ifitm genes but up- (toxoplasmosis, other, rubella, cytomegalovirus,
system, VCTs were transfected with a FLAG- regulated Irf 7 (fig. S9 C). Levels of syncytin-A and herpes) (19). In addition to infection, genetic
tagged IFITM3 or a control GFP plasmid. The and -B RNAs were not significantly affected by and autoimmune interferonopathies such as
poor efficiency of transfection (10% of the cells Poly-IC (fig. S9C). Aicardi–Goutières syndrome and systemic lupus

A B ns ns C PBS Poly-IC
****
2/33 49/50 6/22 9/27 4/33 13/61
100
WT
E7.5
Viable
% Resorption
Resorbed
± Poly-IC 48h
Ifnar-/-
50
E9.5
IfitmDel
0
Viable
Embryos Poly-IC - + - + - +
Resorptions
WT Ifnar-/- IfitmDel
D PBS Poly -IC

WT IfitmDel WT IfitmDel
E-Cad
Hoechst
60 µm
60 µm
Fig. 3. IFITMs are key mediators of IFN-induced fetal demise in Numbers in brackets indicate the number of resorptions/total number
mice. (A) Gestating dams were injected intraperitoneally with Poly-IC of fetoplacental units. Statistical analysis: ****P < 0.0001, Mann-Whitney
or phosphate-buffered saline (PBS) at E7.5. The number, size, and test; ns, not significant. (C) Representative images of E9.5 embryos.
resorption of the embryos were assessed at E9.5. Numbers of litters Scale bars, 500 mm. (D) WT and IfitmDel placentas (E9.5) stained for
were as follows: WT PBS, n = 3; WT Poly-IC, n = 6; Ifnar−/− PBS, E-cadherin (red) and Hoechst (blue). White arrows indicate ST. Bottom
n = 3; Ifnar−/− Poly-IC, n = 3; IfitmDel PBS, n = 3; and IfitmDel panels are magnified areas indicated by white boxes in the top panels.
Poly-IC, n = 7. (B) Percentage of resorption for the indicated conditions. Representative images from three independent experiments are shown.
erythematosus (SLE) are associated with preg- pathologies of unknown etiology, such as certain 4. S. E. Ander, M. S. Diamond, C. B. Coyne, Sci. Immunol. 4,
nancy complications (2). High serum IFN level preeclampsia or early spontaneous abortions, eaat6114 (2019).
5. J. J. Miner et al., Cell 165, 1081–1091 (2016).
is a marker of poor pregnancy outcome in SLE implicate IFITM proteins and variants. It is also 6. J. E. Thaxton et al., J. Immunol. 190, 3639–3647 (2013).
(20). Down syndrome in trisomy 21 (T21) patients tempting to suggest that blockade of IFITMs 7. L. J. Yockey et al., Sci. Immunol. 3, eaao1680 (2018).
is also associated with serious birth defects (1). may represent a possible intervention strategy 8. A. Dupressoir et al., Proc. Natl. Acad. Sci. U.S.A. 106,
In vitro, VCTs from T21 patients poorly fuse into to prevent pregnancy complications linked to 12127–12132 (2009).
9. A. Dupressoir et al., Proc. Natl. Acad. Sci. U.S.A. 108,
ST (1). This may be due to the location of the IFN interferonopathies. E1164–E1173 (2011).
receptor on chromosome 21, rendering T21 cells 10. A. L. Brass et al., Cell 139, 1243–1254 (2009).
hyperresponsive to IFN (21) and thus potentially 11. G. Shi, O. Schwartz, A. A. Compton, Retrovirology 14, 53
expressing high amounts of IFITMs. (2017).
1. G. Pidoux et al., Placenta 33 (suppl.), S81–S86 (2012). 12. A. Zani, J. S. Yount, Curr. Clin. Microbiol. Rep. 5, 229–237 (2018).
As other restriction factors, Ifitm genes are 2. L. J. Yockey, A. Iwasaki, Immunity 49, 397–412 (2018). 13. A. A. Compton et al., Cell Host Microbe 16, 736–747 (2014).
polymorphic in humans and primates (11, 12). It 3. P. A. Bolze, M. Mommert, F. Mallet, Prog. Mol. Biol. Transl. Sci. 14. U. C. Lange et al., Mol. Cell. Biol. 28, 4688–4696 (2008).
will be worth determining whether placental 145, 111–162 (2017). 15. A. A. Compton et al., EMBO Rep. 17, 1657–1671 (2016).

16. J. S. Yount, R. A. Karssemeijer, H. C. Hang, J. Biol. Chem. 287, revivification, M. Cohen-Tannoudji for hosting mouse experiments. We performed mouse experiments. E.P. and K.-H.H. performed the
19631–19641 (2012). thank the Cellular and Molecular Imaging facility of the Faculty of bioinformatic analysis. A.D. and T.H. provided vital materials and
17. K. Orendi, M. Gauster, G. Moser, H. Meiri, B. Huppertz, Pharmacy of Paris (UMS3612 CNRS/US25 INSERM), and the UtechS expert advice. J.B. and O.S. wrote the manuscript and all authors
Reproduction 140, 759–766 (2010). Photonic BioImaging (Imagopole). Funding: Work in the laboratory agreed on the final version. Competing interests: The authors declare
18. S. A. Degrelle, T. Fournier, Methods Mol. Biol. 1710, 165–171 of O.S. is funded by Institut Pasteur, ANRS, Sidaction, the Vaccine no competing interests. Data and materials availability: All data are
(2018). Research Institute (ANR-10-LABX-77), Labex IBEID (ANR-10-LABX-62- available in the manuscript or the supplementary materials.
19. C. B. Coyne, H. M. Lazear, Nat. Rev. Microbiol. 14, 707–715 IBEID), “TIMTAMDEN” ANR-14-CE14-0029, “CHIKV-Viro-Immuno”
(2016). ANR-14-CE14-0015-01, CNRS and the Gilead HIV cure program. Work in
20. D. Andrade et al., Arthritis Rheumatol. 67, 977–987 (2015). SUPPLEMENTARY MATERIALS
the laboratory of M.L. is funded by Institut Pasteur, INSERM, ERC,
21. K. D. Sullivan et al., eLife 5, e16220 (2016). Labex IBEID (ANR-10-LABX-62-IBEID), Institut Convergence, and science.sciencemag.org/content/365/6449/176/suppl/DC1
Université Paris Descartes. C.M. was supported by a fellowship from Materials and Methods
ACKN OW LEDG MEN TS grant no. ANR-10-LABX-62-IBEID. M.V.A. was supported by NSF GRFP Figs. S1 to S9
We thank members of the Virus and Immunity Unit for discussions grant no. 1644869. Author contributions: J.B., D.A.D., T.H., X.M., References (22–37)
and help, M. Maaran Rajah for critical reading of the manuscript. T.F., M.L., and O.S. designed the experimental strategy. J.B., Q.N., Movie S1
We thank patients who participated in the study. We also thank D.A.D., F.P., M.V.A., and K.R. designed and performed experiments with
members of Embryology and Core Breeding teams (DTPS-C2RA- cell lines. S.A.D. and T.F. designed and performed experiments with 25 January 2019; accepted 23 May 2019
Central Animal Facility platform) for technical support with strain primary trophoblasts. J.B., T.C., C.M., Q.N., O.D., and X.M. designed and 10.1126/science.aaw7733
NEUROSCIENCE red and green targets, respectively. Because the

locations of the red and green targets were
randomly interleaved across trials, there was
Posterior parietal cortex plays a a flexible mapping between motion category
or direction and the saccade direction. In the
causal role in perceptual and MDC task, 10 motion directions were shown
at three angular distances from the category
categorical decisions boundary (Fig. 1B). In the MDD task, two di-
rections of motion were shown, each at three
levels of coherence (i.e., signal to noise). To
Yang Zhou* and David J. Freedman* assess LIP’s contribution to stimulus evaluation
and motor planning, each task was tested in
Posterior parietal cortex (PPC) activity correlates with monkeys’ decisions during blocks of trials using three spatial configura-
visual discrimination and categorization tasks. However, recent work has questioned tions of motion stimuli and saccade targets
whether decision-correlated PPC activity plays a causal role in such decisions. with respect to the location of the inactivated
That study focused on PPC’s contribution to motor aspects of decisions (deciding visual field (IVF) contralateral to muscimol in-
where to move), but not sensory evaluation aspects (deciding what you are looking at). jection (Fig. 1C). In the “Stimulus IN” (SIN) con-
We employed reversible inactivation to compare PPC’s contributions to motor and dition, the motion stimulus was shown within
sensory aspects of decisions. Inactivation affected both aspects of behavior, but the IVF, with both saccade targets outside the
preferentially impaired decisions when visual stimuli, rather than motor response targets, IVF. In the “Target IN” (TIN) condition, the mo-
were in the inactivated visual field. This demonstrates a causal role for PPC in tion stimulus was outside the IVF and one of
decision-making, with preferential involvement in evaluating attended task-relevant the colored saccade targets was shown in the
sensory stimuli compared with motor planning. IVF. In the “Both OUT” (BOUT) condition, neither
the motion stimuli nor the saccade targets
D
were in the IVF.
ecision-making requires evaluating task- and selective attention), as opposed to select- We examined the impact of LIP inactivation
relevant sensory stimuli to select appropri- ing motor responses, during decision-making. on evaluating visual motion in the S IN con-
ate motor responses. The primate posterior Furthermore, neuronal recordings have dem- dition (Fig. 2). LIP inactivation produced a
parietal cortex (PPC) is well suited to onstrated correlates of more flexible categorical significant behavioral impairment in both the
mediate decision-making because of its or rule-based tasks in PPC (13), raising the pos- MDC and MDD tasks, shown by a decrease in
anatomical position at a midpoint in the sensory– sibility that the primate PPC may be more en- overall accuracy (Fig. 2, A, E, G, and K; MDC:
cognitive–motor cortical hierarchy (1–3). Indeed, gaged in decisions requiring greater cognitive P(accuracy) = 2.2e-08, t(37) = –7.1; MDD: P(accuracy) =
there is a correlation between PPC activity and flexibility or abstraction than in the classic per- 5.8e-11, t(37) = –9.1, inactivation versus control,
monkeys’ decisions during visual discrimination ceptual decision tasks. unpaired t test) and increase in reaction time
and categorization tasks (4–12). Much of this We directly compared LIP’s role in sensory (RT) (Fig. 2, B, F, H, and L) (MDC: P(RT) =
previous work has focused on the lateral intra- stimulus evaluation and motor-planning func- 0.0062, t(37) = 2.9; MDD: P(RT) = 0.0011, t(37) =
parietal area (LIP) (13–15). Yet, PPC’s role in tions during flexible visual perceptual and cat- 3.5, unpaired t test). In both tasks, a greater
sensory stimulus evaluation and motor planning egorical decisions. We reversibly inactivated LIP behavioral impairment was observed for one
aspects of decisions has been debated. Recent in one hemisphere with intracortical injections of the motion categories or directions (fig. S1;
work showed that reversible inactivation of LIP of muscimol during several variants of motion- MDC: P = 6.8e-04, t(16) = 4.2; MDD: P = 0.0048,
does not affect motor aspects of decisions during direction categorization (MDC) and motion- t(16) = 3.3, paired t test). Psychometric curve
a visual motion discrimination task (16). This direction discrimination (MDD) tasks. Within fitting showed a significant change in bias and
leaves open the possibility that LIP is instead each task, monkeys reported their decisions an increase in psychophysical threshold during
more engaged in processes related to evaluat- about the category membership (MDC task) both tasks (Fig. 2, M to P, and fig. S2; MDC:
ing sensory stimuli (e.g., stimulus feature pro- or motion direction (MDD task) of a sample R2(inactivation) = 0.90 ± 0.034, R2(control) = 0.94 ±
cessing, perceptual or abstract categorization, stimulus with a saccade to a target of the as- 0.044, P(bias) = 2.7e-04, t(37) = 4.0, P(threshold) =
sociated color (Fig. 1A). The mappings between 1.9e-08, t(37) = 7.1; MDD: R2(inactivation) = 0.91 ±
motion categories or directions and target color 0.069, R2(control) = 0.98 ± 0.025, P(bias) = 9.5e-4,
Department of Neurobiology, The University of Chicago,
Chicago, IL 60637, USA.
were arbitrarily assigned at the start of training t(37) = 3.6, P(threshold) = 1.2e-7, t(37) = 6.5, un-
*Corresponding author. Email: yangz1@uchicago.edu (Y.Z.); (and fixed across the study), with the two cate- paired t test). Behavioral impairments in the
dfreedman@uchicago.edu (D.J.F.) gories (MDC) or directions (MDD) assigned to SIN condition of both tasks were significantly

greater than those in the BOUT condition (Fig. 2, gaze position, microsaccades, or peak saccade each monkey) and was confirmed using a re-
E, F, K, and L; MDC: P(accuracy) =7.5e-06, t(16) = velocity (figs. S8 to S10). ceiver operating characteristic (ROC) analysis
–6.5, P(RT) = 0.0011, t(16) = 4.0; MDD: P(accuracy) = We hypothesized that the inactivation-related (Fig. 4D).
1.1e-09, t(16) = –12.6, P(RT) = 1.9e-04, t(16) = 4.8, SIN behavioral deficits in the SIN condition of the To further test whether DS in LIP was de-
versus BOUT, paired t test), in which we did not MDD task were due to disrupting neuronal ac- cision related, we compared activity on correct
observe a global impact on accuracy (Fig. 2, C, tivity related to evaluating stimulus motion versus error low-coherence trials (see materials
E, I, and K; MDC: P(accuracy) = 0.17, t(37) = –1.4; within the IVF. Thus, we recorded from 194 and methods). LIP population activity was more
MDD: P(accuracy) = 0.12, t(37) = –1.6, inactivation LIP neurons (monkey M: n = 78; monkey B: n = similar in trials in which monkeys decided that
versus control, unpaired t test) or RT (Fig. 2, D, 116) during the SIN condition, targeting the same different motion directions were the same ver-
F, J, and L; MDC: P(RT) = 0.60, t(37) = 0.53; cortical locations within the same hemispheres sus trials in which the same motion directions
MDD: P(RT) = 0.71, t(37) = 0.37, unpaired t test). as in the inactivation experiments. More than were treated as different (Fig. 4E; P = 1.9e-04,
LIP’s role in motor aspects of decisions was half of LIP neurons (monkey M: 50/78, monkey t(100) = 3.9, paired t test). Accordingly, LIP ac-
assessed in the TIN condition (Fig. 3), which B: 54/116) showed significant motion direction tivity covaried more closely with the monkeys’
is the spatial configuration nearly always tested selectivity (DS) in the MDD task (one-way trial-by-trial decisions about motion direction
in past studies of LIP with the MDD task ANOVA, P < 0.01). DS emerged before the mon- than it did with the physical motion direction
(7, 8, 12, 14, 15, 17) [but see (18)]. Although the keys’ mean RTs (fig. S11A), and the magnitude (Fig. 4F; P = 6.8e-04, t(100) = –3.4, paired t test,
motion stimulus was outside the IVF, note that of DS was positively correlated with motion comparing ROC values on chosen versus physical
the TIN condition did require evaluating the coherence at the single-neuron and population direction of motion). Furthermore, LIP activity
color of the in-IVF saccade target. After LIP levels, but did not converge to a fixed threshold covaried with the monkeys’ RTs, with most neu-
inactivation, the monkeys’ saccadic choices were near the decision time (Fig. 4, A to D, and fig. rons showing a greater response to their pre-
significantly biased away from the IVF in both S11B). Neuronal activity on zero-coherence mo- ferred motion direction on shorter versus longer
tasks (Fig. 3, A, C, G, and I; MDC: P = 1.3e-07, tion trials (in which monkeys could only guess RT trials for all motion-coherence levels (and a
t(16) =8.9, MDD: P = 3.0e-04, t(16) = –5.4, paired about motion direction) reflected the monkeys’ weaker response to their nonpreferred direc-
t test), and a greater RT increase was observed trial-by-trial decisions, consistent with a role in tion; figs. S14 and S15). Even in low-coherence
on trials in which the saccade was directed the decision process. Decision-correlated activ- (Fig. 4G) and zero-coherence (Fig. 4H) trials, DS
toward versus away from the IVF (Fig. 3, B, D, ity across all motion-coherence levels was evi- evolved more rapidly on shorter versus longer
H, and J; MDC: P = 7.0e-06, t(16) = –6.5, MDD: dent in single-neuron activity (Fig. 4, A and B, RT trials (comparing slope: P(low) < 0.01, P(zero) <
P = 6.0e-05, t(16) = 4.6, paired t test). Although and fig. S12) and in the LIP population (Fig. 4C; 0.01, bootstrap). At the population level, the slope
LIP inactivation produced an ipsilateral saccade see fig. S13 for results reported separately for of DS negatively correlated with monkeys’ RT
bias, there was no significant difference in mean
accuracy between inactivation and control ses-
sions for either task (Fig. 3, A, E, G, and K; MDC:
P = 0.13, t(37) = –1.5; MDD: P = 0.56, t(37) = –0.59,
unpaired t test), and accuracy on both tasks was
less influenced by inactivation in TIN than by
that in SIN (Fig. 3, E and K; MDC: P = 1.6e-08,
t(16) = –10.4; MDD: P =1.2e-06, t(16) = –7.5, paired
t test). The magnitudes of RT effects between
TIN and SIN were statistically indistinguishable
(Fig. 3, F and L; MDC: P = 0.40, t(16) = 0.86;
MDD: P = 0.093, t(16) = 1.8, paired t test). Psy-
chometric curve fitting showed a significant bias
in saccades away from the IVF in both tasks,
but no increase in TIN threshold in either task
(Fig. 3, M to P, and fig. S3; MDC: R2(inactivation) =
0.92 ± 0.070, R2(control) = 0.94 ± 0.047, P(bias) =
4.0e-07, t(37) = 6.1, P(threshold) = 0.15, t(37) = 1.5;
MDD: R2(inactivation) = 0.98 ± 0.010, R2(control) =
0.98 ± 0.010, P(bias) = 0.0013, t(37) = 3.5, P(threshold) =
0.33, t(37) = 0.99, unpaired t test). These ef-
fects were consistent across monkeys (figs. S4
and S5).
We examined how distinct components of the
decision process were affected by inactivation.
We fitted our results with fixed-boundary drift Fig. 1. Behavioral task. (A) Monkeys reported their categorical (MDC task) or directional
diffusion models, which captured monkeys’ de- (MDD task) decisions about visual motion stimuli by choosing either the green or the red
cision behavior across different task conditions saccade target. The positions of red and green targets were randomly chosen on each trial.
in both the MDD and MDC tasks, including Using an RT design, monkeys could initiate their saccade as soon as they had made their
both accuracy and the RT distributions (figs. decision. The red and green spots indicate the positions of the saccade targets and the
S6 and S7). This revealed that LIP inactivation yellow spot indicates the position of fixation. (B) The MDC task required grouping 10 motion
substantially slowed the evidence accumula- directions (indicated by the direction of the arrows) into two categories (indicated by the
tion process (drift rate) only in SIN, while ap- color of the arrows) defined by a learned category boundary (black dashed line). In the MDD task,
preciably changing the saccade choice bias two motion directions (the two category center directions from the MDC) were shown at
(starting point of the diffusion process) and three coherence levels. (C) Three spatial stimulus configurations tested LIP’s role in sensory
decision boundary only in TIN across monkeys evaluation (SIN), saccade planning (TIN), and a control condition assessing nonspatial aspects
and tasks (tables S1 to S4). Inactivation also of the tasks (BOUT). The dark shaded areas and the dashed circle represent the IVF and the
did not produce substantial impairments of position of the motion stimulus, respectively.

Fig. 2. Causal evidence for decision-related sensory evaluation in LIP. [(G) and (H)] and BOUT [(I) and (J)] conditions are shown in the same
(A) Psychometric curves for the SIN condition of the MDC task. Task format as for the MDC task [(A) to (D)]. Monkeys showed a significantly
accuracy pooled across both monkeys is plotted as the proportion of greater deficit in the SIN than in the BOUT condition in the MDD task [(K)
choosing the primary category, defined as the category for which each and (L)]. (M to P) Paired comparisons between inactivation and control
monkey showed a greater decrease in accuracy (on average across all sessions for choice bias and threshold in the SIN conditions of the MDC task
sessions) after LIP inactivation (figs. S4 and S5 show data for each (M) and (N)] and the MDD task [(O) and (P)]. The open and filled symbols
monkey separately). (B) Chronometric curves are shown for the SIN denote the inactivation sessions in which the majority of the recorded
condition of the MDC task. (C and D) The psychometric and chronometric neurons at the targeted cortical locations preferred the primary (open) and
curves in the BOUT condition of the MDC task [same format as (A) and nonprimary (filled) category or direction, respectively (see the supplemen-
(B)]. (E and F) Comparisons of the averaged behavioral deficits after LIP tary materials). Asterisks indicate statistical significance: *P < 0.05,
inactivation in the SIN and BOUT conditions of the MDC task. (G to L) **P < 0.01, ***P < 0.001, unpaired t test, multiple tests in [(A) to (D)] and
Monkeys’ behavioral performance in the MDD task is plotted for inactivation [(G) to (J)] are Bonferroni corrected. The error bars denote ± SEM. P, primary;
and control sessions. Psychometric and chronometric curves for the SIN NP, nonprimary; cat, category; dir, direction; H, high; M, middle; L, low.

(fig. S15; r = –0.9455, P = 0.0044). Elevated DS be interesting to determine whether LIP plays results when we analyzed inactivation behavioral
was even observed before stimulus onset on low- a similar role during tasks with fixed sensory- data with respect to neurons’ stimulus prefer-
coherence trials (Fig. 4H), suggesting that the motor mappings. ences at the targeted LIP sites (figs. S17 and
state of LIP activity before stimulus onset was It is well established that LIP helps to me- S18). This may relate to biased representations
predictive of the monkeys’ upcoming decision diate covert spatial attention (21–23) and sac- of categories and directions observed in LIP
in that trial. Furthermore, a partial correlation cades (1, 24, 25). However, the current study during the MDD and MDC tasks (27), or it
analysis examined choice- and stimulus-related highlights LIP’s role in evaluating the abstract could reflect anatomical clustering of such
components of DS (i.e., r(choice) and r(stimulus); behavioral significance of visual stimuli during representations in LIP. Third, LIP inactivation
Fig. 4I and fig. S16) (19). This revealed that decision-making. Although attention contributes did not produce an obvious impairment of eye
the choice and stimulus components of LIP DS toward the selection and representation of task- movements or profound spatial neglect, although
were positively correlated (Fig. 4J, r = 0.6276, relevant stimuli, and disruption of attention by we did observe a mild ipsilateral bias during the
P < 0.0001). LIP inactivation may contribute to the current free choice saccade task (fig. S19), consistent with
Taken together, our results suggest that LIP results, our inactivation results in the SIN con- previous studies (16, 23, 28). Nevertheless, future
plays an important role in visually based percep- dition are unlikely to arise primarily from dis- work using more precise causal approaches such
tual and categorical decisions, with preferential rupting attention- or saccade-related functions as stimulating or silencing pools of neurons with
involvement in stimulus evaluation compared for several reasons. First, the LIP neurons tar- specific direction or category preferences can
with motor-planning aspects of such decisions. geted for inactivation showed activity that cor- more precisely dissect LIP’s contributions toward
This idea is supported by behavioral impairments related with MDD task decisions, indicating decision-making tasks.
on discrimination and categorization tasks that LIP’s role extends beyond the comparatively A recent study came to a different conclusion
caused by reversibly inactivating LIP, neuro- fixed representation of motion direction in the regarding LIP’s role in perceptual decisions,
physiological recordings revealing decision- middle temporal (MT) area (4, 6, 11, 14, 16). The finding no apparent impact of inactivation
correlated LIP activity during the MDD task, partial correlation analysis suggests that LIP DS on MDD task performance (16). However, that
as well as work from multiple groups showing may be more closely choice correlated compared study only considered LIP’s contributions to
sensory-decision–related LIP activity in sev- with other parietal areas (19, 26), although a di- motor-planning (TIN) aspects of the MDD task,
eral tasks (11, 13, 20), including categorization rect comparison across multiple areas during the not sensory stimulus evaluation (SIN). Another
(6, 9). Both tasks tested in this study had a same tasks will be needed. Second, inactivation study found that LIP activity was decision cor-
flexible mapping between the decisions about produced greater impairments for specific di- related in both SIN and TIN conditions of an MDD
motion stimuli and the direction of the sac- rections and categories, rather than uniformly task but concluded that LIP activity correlated
cade used to report those decisions. Thus, it will affecting performance, and we found similar more closely with motor aspects (however, that
Fig. 3. Causal evidence for decision-

related saccade planning in LIP. (A) Psy-
chometric curves for TIN condition of MDC
task. The choice accuracy is plotted as the
proportion of contralateral saccades rela-
tive to the inactivated hemisphere. Data
from both monkeys were pooled based on
target location. (B) Chronometric curves
for the TIN condition of the MDC task.
(C and D) Comparisons of behavioral
impairments after LIP inactivation between
ipsilateral and contralateral target trials in
the TIN condition. Monkeys’ saccade
choices were biased toward the ipsilateral
target after LIP inactivation, as shown
by both accuracy (C) and RT (D).
(E and F) Comparisons of overall behavior
deficits after LIP inactivation between
the SIN and TIN conditions of the MDC task.
Monkeys showed significantly greater
behavioral impairment in the SIN condition
than in the TIN condition in accuracy (E) but
not RT (F). (G to L) Monkeys’ behavioral
performance in the TIN condition of the
MDD task. Psychometric and chronometric
curves [(G) and (H)] are in the same format
as those in the MDC task [(A) and (B)].
Monkeys showed consistent saccadic
choice biases after LIP inactivation in
both the MDD and MDC tasks [(I) and (J)].
In the MDD task, a greater deficit was
observed after LIP inactivation in the SIN
than in the TIN condition in accuracy (K)
but not RT (L). (M to P) Paired comparisons between inactivation and control sessions for choice bias and threshold in the TIN condition of the MDC
task [(M) and (N)] and the MDD task [(O) and (P)]. Asterisks indicate statistical significance in the same way as in Fig. 2.

Fig. 4. LIP activity reflects decision-related sensory evaluation. trials is compared for the monkeys’ decisions about motion
(A and B) Activity is shown for each motion-coherence level for direction compared with the physical direction of stimulus motion.
two example direction-selective LIP neurons in the MDD task. The Asterisks indicate time periods in which there was a significant
motion stimulus but not the targets appeared in neurons’ RF. The difference (P < 0.01, paired t test). (G and H) DS on low-coherence
zero-coherence trials were grouped based on the monkeys’ choices. trials (G) and choice selectivity (CS) on zero-coherence trials
The two vertical dashed lines mark the time of target and motion (H) is compared between shorter RT and longer RT trials. Shaded
stimulus onset, respectively. (C) Average normalized population areas denote ± SEM. (I and J) Partial correlation analysis. (I) The
activity across all direction-selective neurons is shown for each value of r(stimulus) (the partial correlation between neuronal activity
motion-coherence level, aligned to stimulus onset (left panel) or and stimulus direction, given the monkeys’ choices) and r(choice)
saccade onset (right panel). Activity shown in the left panel is (the partial correlation between neuronal activity and monkeys’
truncated at the monkeys’ mean RT for each coherence level. choice, given the stimulus direction) are plotted across time.
(D) The DS (determined by ROC) for the different coherence levels (J) Correlation between r(stimulus) and r(choice). Note that most LIP
is shown as in (C). (E) Average activity on low-coherence trials is neurons showed a congruent sign between r(stimulus) and r(choice)
shown for neurons’ preferred and nonpreferred directions, separately values. P, preferred; NP, nonpreferred; H, high; M, middle; L, low;
for correct and error trials. (F) Neuronal selectivity on low-coherence Z, zero; C, correct; E, error.
study did not directly test LIP’s causal contribu- decision-related motion encoding might be trans- 9. S. K. Swaminathan, D. J. Freedman, Nat. Neurosci. 15, 315–320
tions) (18). Consistent with the earlier inacti- formed into saccadic signals via coordination (2012).
10. S. K. Swaminathan, N. Y. Masse, D. J. Freedman, J. Neurosci.
vation study, we found that LIP inactivation did between sensory and motor pools of LIP neurons,
33, 13157–13170 (2013).
not impair average MDD-task accuracy in the TIN in contrast to LIP reading out sensory inputs 11. Z. M. Williams, J. C. Elfar, E. N. Eskandar, L. J. Toth, J. A. Assad,
condition. However, we did observe a saccade exclusively from upstream visual areas such as Nat. Neurosci. 6, 616–623 (2003).
bias away from the IVF in the TIN condition, MT (30). 12. M. N. Shadlen, W. T. Newsome, J. Neurophysiol. 86, 1916–1936
suggesting a deficit in mapping decisions to Perceptual and categorical decisions rely on (2001).
13. D. J. Freedman, J. A. Assad, Annu. Rev. Neurosci. 39, 129–147
actions. Several factors may account for this contributions from and coordination among a (2016).
discrepancy. First, we used an RT rather than a network of cortical and subcortical areas span- 14. J. I. Gold, M. N. Shadlen, Annu. Rev. Neurosci. 30, 535–574
fixed-interval version of the MDD task, perhaps ning the sensory, cognitive, and motor hierarchy (2007).
prompting different strategies or speed–accuracy (13–15, 31, 32). Although the current study es- 15. A. C. Huk, L. N. Katz, J. L. Yates, Annu. Rev. Neurosci. 40,
tradeoffs. Second, our tasks employed a flexible tablishes primate PPC as an important node in 349–372 (2017).
16. L. N. Katz, J. L. Yates, J. W. Pillow, A. C. Huk, Nature 535,
mapping (based on saccade target color) be- that network, along with recent work in rodents 285–288 (2016).
tween motion stimuli and saccades. Thus, LIP (33–35), a more complete understanding neces- 17. T. D. Hanks, J. Ditterich, M. N. Shadlen, Nat. Neurosci. 9,
may have been more engaged by decisions re- sitates the characterization of local and long- 682–689 (2006).
quiring greater cognitive flexibility or abstrac- range circuits—within different animal models 18. S. Shushruth, M. Mazurek, M. N. Shadlen, J. Neurosci. 38,
6350–6365 (2018).
tion. Furthermore, the TIN condition in our tasks and humans—that flexibly transform sensory
19. A. Zaidel, G. C. DeAngelis, D. E. Angelaki, Nat. Commun. 8, 715
required discriminating the color of the in-IVF encoding into decisions and actions and the (2017).
saccade target to plan the correct saccade, so process by which task-related neuronal encod- 20. G. Ibos, D. J. Freedman, eLife 6, e23743 (2017).
that condition had greater sensory evaluation ing emerges during learning. 21. J. W. Bisley, M. E. Goldberg, Annu. Rev. Neurosci. 33, 1–21
demands than the traditional MDD task (13). (2010).
RE FERENCES AND NOTES 22. C. L. Colby, M. E. Goldberg, Annu. Rev. Neurosci. 22, 319–349
Compared with the previous study, our animals
1. R. A. Andersen, Annu. Rev. Neurosci. 12, 377–403 (1989). (1999).
learned both the MDD and MDC tasks, and the 23. C. Wardak, E. Olivier, J. R. Duhamel, Neuron 42, 501–508
2. G. J. Blatt, R. A. Andersen, G. R. Stoner, J. Comp. Neurol. 299,
strategy used to solve the MDD task could have 421–445 (1990). (2004).
been affected by MDC training. The greater be- 3. J. W. Lewis, D. C. Van Essen, J. Comp. Neurol. 428, 112–137 24. R. A. Andersen, C. A. Buneo, Annu. Rev. Neurosci. 25, 189–220
havioral deficit observed in the MDD task (which (2000). (2002).
4. E. P. Cook, J. H. Maunsell, Nat. Neurosci. 5, 985–994 25. Y. Liu, E. A. Yttri, L. H. Snyder, Nat. Neurosci. 13, 495–500
employed low-coherence motion stimuli) com- (2010).
(2002).
pared with the MDC task (which used only 100% 5. V. de Lafuente, M. Jazayeri, M. N. Shadlen, J. Neurosci. 35, 26. X. Yu, Y. Gu, Neuron 100, 715–727.e5 (2018).
coherent motion) is consistent with LIP integrat- 4306–4318 (2015). 27. J. K. Fitzgerald et al., Neuron 77, 180–191 (2013).
ing noisy sensory evidence (12, 14, 17), and the 6. D. J. Freedman, J. A. Assad, Nature 443, 85–88 (2006). 28. P. F. Balan, J. Gottlieb, J. Neurosci. 29, 8166–8176
7. J. D. Roitman, M. N. Shadlen, J. Neurosci. 22, 9475–9489 (2009).
saccadic choice bias in the TIN condition sup-
(2002). 29. M. N. Shadlen, R. Kiani, T. D. Hanks, A. K. Churchland,
ports LIP’s involvement in mediating the sac- 8. M. N. Shadlen, W. T. Newsome, Proc. Natl. Acad. Sci. U.S.A. 93, “Neurobiology of decision making: An intentional framework”
cades used for reporting decisions (14, 29). Thus, 628–633 (1996). in Better than Conscious? Decision Making, the Human Mind,

and Implications for Institutions, C. Engel, W. Singer, Eds. decisions (2019); https://dx.doi.org/10.6084/m9.figshare. financial interests. Data and materials availability: All data
(MIT Press, 2008), pp. 71–101. 8283251 and analyses necessary to understand and assess the conclusions
30. J. L. Yates, I. M. Park, L. N. Katz, J. W. Pillow, A. C. Huk, of the manuscript are presented in the main text and in the
Nat. Neurosci. 20, 1285–1292 (2017). supplementary materials. Data and analysis code from this study
ACKN OWLED GMEN TS
31. T. B. Crapse, H. Lau, M. A. Basso, Neuron 97, 181–194.e6 are available at Figshare (36).
We thank J. Assad, N. Masse, K. Latimer, K. Mohan, and
(2018).
W. Johnston for their comments on an earlier version of this SUPPLEMENTARY MATERIALS
32. C. A. Seger, E. K. Miller, Annu. Rev. Neurosci. 33, 203–219
manuscript. We also thank the veterinary staff at The University of
(2010). science.sciencemag.org/content/365/6449/180/suppl/DC1
Chicago Animal Resources Center for expert assistance.
33. A. M. Licata et al., J. Neurosci. 37, 4954–4966 Materials and Methods
Funding: This study was supported by National Institutes of
(2017). Figs. S1 to S19
Health grant no. NIH R01EY019041. Author contributions: Y.Z.
34. T. D. Hanks et al., Nature 520, 220–223 (2015). Tables S1 to S4
and D.J.F. designed the experiments and wrote the manuscript.
35. L. Zhong et al., Nat. Neurosci. 22, 963–973 (2019). References (37–46)
Y.Z. trained monkeys, collected the behavioral and neuronal data,
36. Y. Zhou, D. J. Freedman, Data and analysis code for: Posterior analyzed the data, and made the figures. D.J.F. supervised the 29 January 2019; accepted 18 June 2019
parietal cortex plays a causal role in perceptual and categorical project. Competing interests: The authors declare no competing 10.1126/science.aaw8347
PROTEIN NETWORKS We hypothesized that the most strongly co-

evolving protein pairs would likely physically
interact. We assessed this by constructing a “gold-
Protein interaction networks revealed standard” PPI set from E. coli protein complexes
(20) in the Protein Data Bank (PDB) (table S1)
by proteome coevolution and a negative control set consisting of pro-
tein pairs drawn from two different complexes
(table S2), with no experimental data (21–23)
Qian Cong1,2, Ivan Anishchenko1,2, Sergey Ovchinnikov 3, David Baker1,2,4* indicating interactions between them (we can-
not be sure that all such pairs do not interact,
Residue-residue coevolution has been observed across a number of protein-protein but the fraction is likely to be small). Selection
interfaces, but the extent of residue coevolution between protein families on the of coevolving pairs with high MI increases the
whole-proteome scale has not been systematically studied. We investigate coevolution frequency of physically interacting pairs com-
between 5.4 million pairs of proteins in Escherichia coli and between 3.9 millions pairs in pared with negative control pairs (Fig. 1C).
Mycobacterium tuberculosis. We find strong coevolution for binary complexes involved in Considerably better discrimination (Fig. 1C,
metabolism and weaker coevolution for larger complexes playing roles in genetic green versus red curves) of the positive from the
information processing. We take advantage of this coevolution, in combination with negative control could be achieved by down-
structure modeling, to predict protein-protein interactions (PPIs) with an accuracy that weighting proteins that appear to coevolve with
benchmark studies suggest is considerably higher than that of proteome-wide two-hybrid many others through an average product correc-
and mass spectrometry screens. We identify hundreds of previously uncharacterized PPIs in tion (APC) (19).
E. coli and M. tuberculosis that both add components to known protein complexes and We selected the top 961,929 pairs (to the left of
networks and establish the existence of new ones. the black vertical line in Fig. 1C) for further anal-
C
ysis using the global methods direct coupling
oevolution-based prediction of interact- pairs, have prevented coevolution-based ap- analysis (DCA) (6) and generative regularized
ing residues from aligned protein sequen- proaches from being used to systematically iden- models of proteins (GREMLIN) (16). These
ces has enabled considerable progress in tify PPIs on the whole-proteome scale. methods improved discrimination of the gold-
predicting the structures of monomeric pro- To systematically investigate coevolution in the standard set with DCA followed by GREMLIN
teins (1–3) and complexes of proteins known E. coli proteome (Fig. 1B), we began by using the performing better than DCA alone (Fig. 1D,
to interact (4–7). However, using coevolution “reciprocal best hit” criterion (12, 13) to identify, purple versus red curves); still better discrim-
to identify previously uncharacterized protein- when possible, putative orthologs for each of the ination was achieved by again penalizing res-
protein interactions (PPIs) over the whole pro- 4262 E. coli proteins in each of the 40,607 repre- idues and proteins that coevolve with many
teome, which requires fishing out the 0.1% of sentative bacterial proteomes. We aligned these others (Fig. 1D, green versus purple curves).
pairs (8) that interact among the vast majority orthologs (14, 15) and constructed paired align- Choosing a threshold balancing sensitivity and
of noninteracting pairs, remains a formidable task ments for all 4262 × (4262 − 1) ÷ 2 = 9,080,191 specificity, we selected the top 21,818 pairs (to
(9, 10). Determining the extent of coevolution protein pairs. The alignments for 5,433,039 pairs the left of the black vertical line in Fig. 1D). As
between residues in two different protein fam- contain sufficient sequence information (Nf 90 >¼ a final screen, three-dimensional models for
ilies requires pairing each protein in one family 16, see Fig. 1 legend) to assess coevolution (Fig. 1A). proteins in each pair were docked together
with its partner in the other (7, 11). This pairing is Coevolution detection methods that eliminate (24), guided by distance constraints between
not straightforward if either family includes transitivity using global statistical models, which coevolving residue pairs, and we selected 804
multiple paralogues with different functions consider all residue pairs simultaneously (16–18), protein pairs that exhibited the strongest co-
and partners in a species; previous studies have are too slow for datasets of this size. Instead, evolution across the docked interface (Fig. 1E
thus been largely limited to proteins encoded on we used the residue-residue mutual information
and materials and methods M7.2). With each
P
the same operon (4–7). These difficulties, and the Pðaa1; aa2Þ increasingly stringent screen step, the num-
MI; Pðaa1; aa2Þ log Pðaa1ÞPðaa2Þ as an
complexity of working with millions of protein aa1;aa2 ber of negative control pairs is reduced greatly,
initial screen (19); this is a local statistical model whereas the recovery of the gold-standard pairs
1
Department of Biochemistry, University of Washington,
because each residue pair is considered indepen- decreases only slightly (table S3). We inves-
Seattle, WA 98105, USA. 2Institute for Protein Design, dently. We used the maximum value of the MI tigated whether deep learning methods (25)
University of Washington, Seattle, WA 98105, USA. 3John over all residue pairs as a metric for protein- trained on contacts in monomeric proteins could
Harvard Distinguished Science Fellowship Program, Harvard protein coevolution (rather than an average or provide better discrimination but found they
University, Cambridge, MA 02138, USA. 4Howard Hughes
Medical Institute, University of Washington, Seattle, WA
sum over the most strongly coevolving residue did not improve performance (fig. S1; contacts
98105, USA. pairs) to reduce the impact of lack of indepen- are overpredicted as the prior probability of
*Corresponding author. Email: dabaker@uw.edu dence due to transitivity. residue-residue interactions between proteins

is far lower than within proteins); such meth- two hybrid (Y2H) (8) and affinity purification- derived from x-ray and cryo–electron microscopy
ods likely will require direct training on PPIs to be mass spectrometry (APMS) (26, 27) over several complexes of E. coli proteins in the PDB and from
useful for this purpose. benchmark sets. One benchmark is from the gold-standard complexes in Ecocyc (20) (tables S4
We compared the accuracy of the coevolution- Y2H study, one is from one of the two APMS (26) to S7). We evaluate the performance of each meth-
based interaction predictions with those of inter- studies (the other APMS study did not contain a od on each benchmark using precision (TP/P,
actions inferred from high-throughput yeast benchmark), and two additional benchmarks were where TP is the number of true interactions that
Fig. 1. PPI identification by using

coevolution. (A) Distribution of E. coli
pffiffiffi
protein family sizes. Nf90 ¼ N90 = L, where
L is the number of aligned positions in the
alignment, and N90 is the number of
sequences in the alignment, filtered at
90% sequence identity. The black box indi-
cates selected protein pairs. (B) Screening
pipeline. (C) Protein pairs were ranked
by MI, and lines sweep MI threshold
values from high (left) to low (right). The
number (P) of pairs above a MI threshold,
the number (T) of gold-standard pairs,
and their overlap (TP) are used to calculate
Precision (TP/P, y-axis) and Recall (TP/T,
x-axis). Baseline (blue) represents random
ranking of pairs. Improved performance
(green) is achieved by using an average
product correction (APC). (D) Enhanced
recovery of gold-standard pairs by using global
statistical methods (DCA and GREMLIN).
Green curve includes APC-like procedures to
penalize false-positive hubs. (E) Further
increase in precision through protein-protein
docking calculations. Pairs were ranked
by the sum of the probability of contacts
made in the best-fitting docked complex.
(F) Performance of experimental and
coevolution screens on diverse benchmarks.
The size of each benchmark is shown in parentheses. Cells are colored by performance: green for the best and red for the worst. Coevolution+,
increased coverage by supplementing input to GREMLIN and docking screens with pairs missed in initial stages but identified in previous
experimental studies (materials and methods M6.1); F-score, harmonic mean of precision and recall; Pre, precision; Rec, recall; TP, true positives.
Fig. 2. Coevolution in known protein

complexes. The extent of coevolution is higher
in complexes with fewer subunits (A) and
varies with the function of the complexes
(B). (C to E) Obligate and transient interactions
revealed by coevolution provide insights into
function. Bars connecting coevolving residues are
in green if an experimental structure containing
the interface has been determined
and in red if not. Black arrows indicate inferred
movements of proteins. (C) LPS transporter
consisting of periplasmic LPS-binding protein
LptA (orange), LptE (yellow), and outer-
membrane b-barrel LptD (light blue). (D) Acetyl-
CoA carboxylase complex consisting of biotin
carboxylase (AccC, yellow), biotin carboxyl carrier
(AccB, pink), and carboxyltransferase subunits
(AccA and AccD, light blue and orange). (E) Self-
inhibitory mechanism of DNA polymerase V
(umuD2C). The magenta b-strand in umuD
(yellow, top) is cleaved upon activation by RecA.
The remaining umuD′ dimerizes (bottom) and
causes the green b strand blocking the active site
(black spheres below the magenta strand) in
umuC (light blue) to move away and release
inhibition of polymerase activity.

are correctly predicted, and P is the number of was such a wide range of coevolution across are incompatible with a single static complex,
predicted interactions) and recall (TP/T, where T known protein-protein interfaces, we compared suggesting dynamic interactions important for
is the number of true interactions in a bench- the properties of strongly coevolving protein function. We observed coevolution between the
mark). Our coevolution screen outperforms the complexes with those showing little coevolu- periplasmic lipopolysaccharide (LPS)–binding
experimental methods (Fig. 1F) in both precision tion (table S9). Overall, coevolution across in- subunits LptA and LptE and between the N-
and recall except for a worse recall on the Y2H terfaces in binary complexes (two components) terminal tail of LptE and the outer-membrane
benchmark—this includes more transient inter- was stronger than that in larger complexes (Fig. b-barrel LptD in the LPS transporter. LptE sits
actions that are not well conserved among many 2A); in large assemblies with interfaces between inside LptD in the cocrystal structure (PDB ID:
species and hence are harder to detect by using several protein pairs, any single interface may 4RHB), where it cannot bind LptA (Fig. 2C). The
coevolution methods. The interacting partners be less critical for complex formation, reducing coevolution data suggest a handoff mechanism:
predicted by coevolution, like those in structur- the extent of coevolution. Coevolution was also LptE dips into the periplasmic space to accept
ally confirmed complexes, have more closely rel- lower in complexes that contain nucleic acids LPS from LptA, maintaining interaction with
ated functions (fig. S2) than those identified in (Fig. 2B): the protein–nucleic acid interactions LptD through the N terminus, and then delivers
the large-scale experiments. The fast coevolution may contribute to the stability of the complex LPS to the extracellular space by transitioning
detection methods used in the early, higher- along with the PPIs. Less coevolution was also to the conformation seen in the crystal structure.
throughput steps in our protocol may miss in- observed for small and low-affinity interfaces In addition to the experimentally determined
teractions that can be recognized by the slower (fig. S3) and for interfaces exhibiting more var- interfaces between the biotin carboxylase sub-
but more sensitive methods in later steps. There- iation between different species (fig. S4). On the unit AccC and the biotin carboxyl carrier AccB
fore, we input protein pairs reported to interact basis of these observations, we expect that the (PDB ID: 4HR7) and between the carboxyltrans-
in experimental studies or on the same operon large set of strongly coevolving protein pairs ferase subunits AccA and AccD (PDB ID: 2F9Y)
directly into the GREMLIN and docking screens, we have identified includes most of the higher- in the acetyl–coenzyme A (CoA) carboxylase com-
resulting in 814 additional pairs (1618 in total, affinity binary complexes in the core prokaryotic plex, we observed coevolution between AccC
table S8) that pass our coevolution and docking proteome (we are not able to observe coevolu- and AccD and between AccB and AccA (Fig. 2D).
thresholds (coevolution+ protocol). tion between proteins present in small num- These interactions suggest a dynamic model in
We observed strong coevolution (predicted in- bers of species) but likely is quite incomplete which AccB shuttles biotin carboxyl from AccC
teracting probability > 0.7) across the interfaces in the coverage for larger protein and protein– (which produces it) to AccA-AccD complex,
of 40% of the gold-standard PPI set but little co- nucleic acid assemblies. which then transfers the carboxyl group from
evolution (interacting probability < 0.2) for 20% It is instructive to consider three examples biotin to the substrate acetyl. For polymerase V
of interfaces in this set. To understand why there in which the interfaces revealed by coevolution (UmuD2C), coevolution data predict a model
Fig. 3. Examples of new components of

known complexes and newly identified
complexes. (A) Fractions of coevolving
complexes that are consistent with
previous structural and experimental data.
(B) Predicted interactions between
nonribosomal proteins and core ribosomal
proteins are indicated by bars color-coded as
in (A) (full names are in table S15).
(C and D) Previously unknown interfaces
extending those in crystal structures.
(E to H) Interactions supported by large-scale
experiments. (I to T) Previously unidentified
interactions. (C) Coevolution suggests that
both the C- (shown) and N-terminal (not
shown, in cocrystal) domains of antitoxin
MqsA interact with toxin MqsR, possibly
forming a higher-order complex. (D) DNA
mismatch repair proteins MutS and MutL
(C terminus). (E) Sec translocon accessory
protein YajC and membrane protein insertase
YidC. (F) Cell division protein FtsX and
murein hydrolase activator EnvC. (G) DNA
polymerase III subunit delta and ferredoxin
YfhL. (H) Protein YciI and riboflavin bio-
synthesis protein RibD. (I) Thioesterase TesA
and protein YbbP. (J) tRNA methyltransferase
TrmD and tRNA sulfurtransferase ThiI.
(K) 1,2-phenylacetyl-CoA epoxidase, subunits
C and D. (L) RNA polymerase sigma factor
FliA (green), flagellar biosynthetic protein FlhB (orange), and and ribosome hibernation promoting factor Hpf (pink). (P) Transcrip-
secretion chaperone FliS (yellow). (M) D-ribose pyranase RbsD tional factor BolA and ribosome modulation factor Rmf. (Q) LPS
and sigma D regulator Rsd. (N) Cell division topological specificity exporter ATPase LptB and protein YbbN. (R) Membrane protein
factor MinE and tRNA-modifying protein YgfZ. (O) Phosphate quality-control factor QmcA and protein YbbJ. (S) Nucleoside
transporter ATPase PstB (green), phosphate transporter accessory triphosphatase RdgB and DNA utilization protein HofN. (T) Macro-
protein PhoU (blue), phosphate regulon sensor protein PhoR (yellow), domain Ter protein MatP and protein YjjV.

of the complex between UmuD and UmuC that and table S15). Some of these PPIs have been ary phase; these may link the metabolic status
can explain the self-inhibition mechanism (Fig. structurally characterized (green bars) or in- of the cell with the transition to stationary
2E). In the predicted UmuD-UmuC complex, ferred in high-throughput experiments (blue phase.
the active site of UmuC is obstructed by a seg- bars), whereas others to our knowledge have Beyond the binary interactions considered
ment of UmuD that strongly coevolves with not been reported previously (magenta bars). above, we observe extended networks of mutual-
residues around the active site. After cleavage Our coevolution-guided docking models (Fig. ly coevolving proteins (Fig. 4). Some of these in-
of the 24 N-terminal amino acids of UmuD to 3, C to T) reveal interfaces that may provide new volve proteins with similar biochemical functions
generate UmuD′, which then dimerizes (28), the insights into biological processes: for example, —for example, networks of sequentially acting
active site–blocking segment changes conforma- how the type II toxin MqsR and antitoxin MqsA enzymes (fig. S6) and of adenosine triphosphate
tion and relieves the inhibition. These examples intermesh, enabling antibiotic-resistant cells (ATP)–binding cassette (ABC) transporters (fig.
show that coevolution can suggest transient inter- to dominate in a bacteria community (Fig. 3C); S7). Given the number of components and range
actions difficult to capture by using conventional how MutS cooperates with MutL to repair DNA of functions, many of the networks are unlikely
structural biology methods. mismatches (Fig. 3D); and how the expression, fold- to form single complexes; rather, they may form
As shown in Fig. 3A, 936 (332 + 604) of the ing, and secretion of flagellar proteins is controlled multiple reconfiguring complexes involved in more
strongly coevolving pairs have been reported by the dedicated sigma factor FliA and regu- complex functions (fig. S8). Some of the larger
previously. Because the false-positive rate is non- latory proteins such as FlhB (Fig. 3L). Other networks involve proteins mediating the same
negligible (10 to 20% overall, see materials and predicted interactions suggest functional roles biological processes, such as transcriptional regu-
methods M8), we searched for additional sup- for poorly characterized proteins. For example, lation (Fig. 4A), outer-membrane integrity main-
porting data for each of the remaining 682 we predict that uncharacterized proteins YfhL tenance (Fig. 4B), and flagella biosynthesis and
(1618 − 936) new predictions. We found homol- (Fig. 3G), YbbJ (Fig. 3R), and YjjV (Fig. 3T) are assembly (Fig. 4F). Other networks connect dif-
ogous PDB templates consistent with coevolu- involved in regulation of DNA replication, mem- ferent processes and perhaps enable bacteria to
tion data for 126 pairs, nearby genomic locations brane protein quality control, and chromosome modulate DNA replication (Fig. 4D), or tran-
for an additional 143 pairs, and an additional segregation based on their coevolution with scription and translation (Fig. 4C) on the basis of
231 pairs with related functions and/or particu- DNA polymerase III subunit delta, membrane the internal and external environment—for ex-
larly strong coevolution across the docked in- protein quality-control factor QmcA, and mac- ample, under stress conditions (Fig. 4E).
terface (fig. S5 and table S10). The predicted PPIs rodomain Ter protein MatP, respectively. Partic- We investigated the applicability of coevolution-
include both previously unknown complexes and ularly interesting cases involve cross-talk between based interaction prediction to Mycobacterium
previously uncharacterized components of different pathways (Fig. 3, F and M to P). For tuberculosis, a human pathogen evolutionarily
known complexes (table S11). The ribosome pro- example, some metabolic enzymes (Fig. 3M) and distant from E. coli: only 41% of M. tuberculosis
vides a notable illustration of the latter. We find transporters (Fig. 3O) coevolve with transcrip- proteins have clear E. coli homologs (BLAST
considerable coevolution between core ribo- tional and translational regulators (ribosome e-value < 10−5). Using the protocol developed for
somal proteins and other components, extending hibernation factor and sigma D regulator) in- E. coli, we inferred 911 PPIs in M. tuberculosis
over the full surface of the assembly (Fig. 3B volved in the transition from growth to station- with an expected precision of 83%. Of the
predicted PPIs, 662 do not have E. coli orthologs,
and 593 do not have E. coli homologs (BLAST
A B e-value < 10−5). The majority (95%, fig. S9) of
these predicted PPIs have not been previously
described, because of the limited experimental
characterization of M. tuberculosis proteins
(for E. coli, 42% have not been characterized).
Forty percent of the predicted interacting part-
C D
ners are functionally related according to the
STRING (functional protein networks) database
(23) (Fig. 5); in comparison, the agreement
between pairs identified in the bacterial two-
hybrid screen (29) and functional networks in
STRING is almost as low as that of randomly
selected pairs and much lower than Y2H or
E APMS studies on E. coli (fig. S2), reflecting the
challenge of carrying out experimental screens
on nonmodel organisms (poor protein expres-
sion, etc.). In contrast, our coevolution screen is
likely as accurate for M. tuberculosis as it is for
E. coli. We provide a full list of the predicted PPIs
F in table S12. Among these predictions, 293 link
poorly characterized proteins to partners with
well-annotated function (table S13), and 70 in-
volve proteins that were suggested to contribute
to the virulence of M. tuberculosis (table S14)
(30). We expect this list to be useful for thera-
Fig. 4. Examples of coevolving protein networks. Blue lines connect coevolving protein pairs, and peutic target identification and deciphering the
green lines connect proteins interacting in experimentally determined structures. (A) Network of hard-to-study biology of M. tuberculosis.
transcription elongation factors. (B) Outer-membrane integrity maintenance network. (C) Linkage The large number of previously unpredicted
between phosphate transport and regulation of transcription and ribosome activity. (D) Chaperones binary interactions, networks, and protein com-
and tRNA modification enzymes are coupled to DNA replication initiation, perhaps decreasing it plex structures described here is notable because
under stress conditions. (E) Stress response network. (F) Network connecting flagella components no new experiments were required; these results
and regulators of their synthesis. instead leverage ongoing genome sequencing

17. D. S. Marks, T. A. Hopf, C. Sander, Nat. Biotechnol. 30,

Fig. 5. Functional relatedness of predicted
1072–1080 (2012).
interacting partners in M. tuberculosis. 18. F. Morcos et al., Proc. Natl. Acad. Sci. U.S.A. 108, E1293–E1301
Functional relatedness was assessed by using (2011).
the M. tuberculosis functional network in the 19. S. D. Dunn, L. M. Wahl, G. B. Gloor, Bioinformatics 24, 333–340
STRING database (high: STRING combined (2008).
20. I. M. Keseler et al., Nucleic Acids Res. 39 (Database),
score ≥ 0.4, missing: not in the STRING database).
D583–D590 (2011).
A considerable fraction (40%) of coevolution- 21. S. Orchard et al., Nucleic Acids Res. 42 (D1), D358–D363
based predictions (green) involve partners (2014).
that are predicted to be functionally related, 22. I. Xenarios et al., Nucleic Acids Res. 28, 289–291 (2000).
whereas a much lower fraction (orange, 2.0%) 23. D. Szklarczyk et al., Nucleic Acids Res. 43 (D1), D447–D452
of PPIs identified in a previous experimental (2015).
24. I. A. Vakser, Protein Eng. 8, 371–378 (1995).
screen involves functionally related partners,
25. D. T. Jones, S. M. Kandathil, Bioinformatics 34, 3308–3315
almost as low (gray, 1.3%) as randomly selected (2018).
pairs. The 384 predicted PPIs involving partners 26. M. Babu et al., Nat. Biotechnol. 36, 103–112 (2018).
lacking homologs in E. coli and without STRING 27. P. Hu et al., PLOS Biol. 7, e1000096 (2009).
annotations (blue bar on left) are likely to be of most interest to M. tuberculosis researchers. 28. Q. Jiang, K. Karata, R. Woodgate, M. M. Cox, M. F. Goodman,
Mtb, M. tuberculosis. Nature 460, 359–363 (2009).
29. Y. Wang et al., J. Proteome Res. 9, 6665–6677 (2010).
30. B. Liu, D. Zheng, Q. Jin, L. Chen, J. Yang, Nucleic Acids Res. 47
efforts. Despite the inevitable errors in such have the potential to more accurately predict (D1), D687–D692 (2019).
large-scale studies, our benchmarking suggests contacts across interfaces with fewer sequen-
AC KNOWLED GME NTS
a considerably higher accuracy than previous ces and further facilitate the identification of
We thank N. V. Grishin, H. S. Malik, L. Stewart for inspiring discussions
large-scale experimental screens and, hence, a PPIs in eukaryotes.
about this project, I. Vakser for sharing the latest version of GRAMM
sound starting point for detailed experimental software, D. E. Kim for help in setting up the pipeline for protein
testing by biochemistry and mutagenesis. It will structure prediction, and L. Goldschmidt for maintaining the computers
RE FERENCES AND NOTES used in this study. We also thank Rosetta@home and Charity Engine
be particularly interesting to follow up on inter-
1. D. S. Marks et al., PLOS ONE 6, e28766 (2011). participants for donating their computer time. Funding: This project
actions that shed light on the function of pre-
2. S. Ovchinnikov et al., Science 355, 294–298 (2017). has been funded in part with Washington Research Foundation,
viously uncharacterized proteins and to investigate 3. S. Wang, S. Sun, Z. Li, R. Zhang, J. Xu, PLOS Comput. Biol. 13, National Institute of General Medical Sciences (grant no. R01-
the suggested couplings between different bio- e1005324 (2017). GM092802-07), National Institute of Allergy and Infectious Diseases
logical processes such as metabolism and trans- 4. T. A. Hopf et al., eLife 3, e03430 (2014). (contract no. HHSN272201700059C), and Office of the Director of
5. S. Ovchinnikov, H. Kamisetty, D. Baker, eLife 3, e02030 the National Institutes of Health (grant no. DP5OD026389). This
lation regulation. The coevolution screen can be (2014). research used resources of the National Energy Research Scientific
carried out on organisms like M. tuberculosis, for 6. M. Weigt, R. A. White, H. Szurmant, J. A. Hoch, T. Hwa, Computing Center (contract no. DE-AC02-05CH11231). Author
which experimental PPI screens are difficult or Proc. Natl. Acad. Sci. U.S.A. 106, 67–72 (2009). contributions: Q.C. and D.B. formulated the research goals and
intractable. By carrying out the analysis on many 7. A. F. Bitbol, R. S. Dwyer, L. J. Colwell, N. S. Wingreen, drafted the manuscript. Q.C. designed the methodology, implemented
Proc. Natl. Acad. Sci. U.S.A. 113, 12180–12185 (2016). most of the code, collected most of the data, and performed the study.
organisms, it should be possible to follow the 8. S. V. Rajagopala et al., Nat. Biotechnol. 32, 285–290 I.A. and S.O. participated in the data collection, methodology design,
evolutionary dynamics of PPI networks. The use (2014). and code implementation. All authors participated in the final draft.
of coevolution to unravel interaction networks in 9. S. Hashemifar, B. Neyshabur, A. A. Khan, J. Xu, Bioinformatics Competing interests: The authors declare no competing interests.
the core eukaryotic proteome will likely require 34, i802–i810 (2018). Data and material availability: All data are available in the manuscript
improved decoupling of coevolutionary and phy- 10. A. F. Bitbol, PLOS Comput. Biol. 14, e1006401 (2018). or the supplementary materials.
11. H. Zeng et al., Nucleic Acids Res. 46 (W1), W432–W437
logenetic contributions to residue-residue covar- (2018).
iation and more genome sequence data on less SUPPLEMENTARY MATERIALS
12. D. P. Wall, H. B. Fraser, A. E. Hirsh, Bioinformatics 19,
complex eukaryotes spanning a wide evolution- 1710–1711 (2003). science.sciencemag.org/content/365/6449/185/suppl/DC1
13. S. F. Altschul, W. Gish, W. Miller, E. W. Myers, D. J. Lipman, Materials and Methods
ary distance. Deep learning methods, which have Figs. S1 to S24
J. Mol. Biol. 215, 403–410 (1990).
greatly improved the contact prediction in indi- Tables S1 to S16
14. S. R. Eddy, PLOS Comput. Biol. 7, e1002195 (2011).
vidual proteins by considering the full spectrum 15. F. Sievers et al., Mol. Syst. Biol. 7, 539 (2011).
of amino acid substitutions (3, 11, 25) rather than 16. H. Kamisetty, S. Ovchinnikov, D. Baker, Proc. Natl. Acad. Sci. U.S.A. 15 January 2019; accepted 7 June 2019
just the overall magnitude of the covariation, 110, 15674–15679 (2013). 10.1126/science.aaw6718

LIFE SCIENCE TECHNOLOGIES Produced by the Science/AAAS Custom Publishing Office
new pro d uc ts
Microbiome Kits
Feeder-Free Culture Medium Loop Genomics provides researchers with ultrahigh-resolution micro-
AMS Biotechnology introduces StemFit

!
Basic04, a next-generation feeder-free -
!! /"/'
medium for the maintenance of induced full-length 16S kit with picogram input amounts and zero human DNA
pluripotent stem (iPS) and embryonic "&, -," -,#

stem (ES) cells during the reprogram-

#!$ -
uous region of the genome (this kit also has a low-biomass version), and
phases of stem cell culture. With its fully "%,

.+

/"/' /"/'
StemFit Basic04 is suited to both research studies and translation to !(

*
cell therapy. The medium enables market-leading colony expansion )

(scalable 100-fold expansion/week), allowing single-cell cloning and insights with a plug and play kit that utilizes low-cost core sequencing
consumables when compared to other long-read platforms.

- Loop Genomics
ules, with lower-than-standard media volume consumption, plus fewer +
%#!! ##!$
passages and media changes. Provided in a single bottle, StemFit www.loopgenomics.com
# !"

cDNA Expression Vectors
protocol. 1 ,

AMS Biotechnology transfecting into mammalian cells, and delivering genome-wide coverage
For info: 617-945-5033 of human (100,000), mouse (300,000), and rat cells (150,000) as well as
www.amsbio.com/feeder-free-stem-cell-culture.aspx (
)0')0'

%5 (

)0'

&5 13+

13+

CRISPR Cas12a Enzyme /))- ,+2 & 5 13+,

&
'(%,
',

&*+)* !.
13+
13+
/))- ,+2 &

%
*& "# &"-
, "6 13+ 13+
/))- *

& $
driven by a cytomegalovirus promoter. Authentic cDNA transcripts are
activity across a wider temperature range than the wild type enzyme,
2(3 5

making it useful for genome editing in additional organisms, including

)0'

13+

,

2
4 (

- available for pathway studies or high-throughput screening.

ing to medical gene therapy to agricultural biotechnology. OriGene
Integrated DNA Technologies / +*** $() &&%(
-('!!$#'#&&" www.origene.com
www.idtdna.com
Total Residual Oxidant (TRO) Analyzer
Signaling Pathway Reporters 0 5!3!.
1 , -
3
* .-# !

4 / 3 0 )")"
pathway reporters. These popular lentivectors are now upgraded with 420,

a clever design that enables reliable generation of stable cell lines. They measurements down to 1 ppb with 1 ppb resolution. The system will give

allowing the possibility of performing a dual-spectral luciferase assay into natural water sources do not exceed the safety threshold or pose a
and delivers greater sensitivity for in vivo applications than conventional
!/

.-

#'!-

" ! # !
conventional DPD (N,N-diethyl-p-phenylenediamine) colorimetric meth-

42,

+05
.- 4#).-1 color interferences, which can have a negative impact on result accuracy

+05
!/ )")"

has negligible activity, resulting in little-to-no luciferase activity or green long periods of time with minimal instrument drift (<5% for >180 days
.-1 / 42,

!
42,
+05 addition, the systemís self-cleaning capability prevents chemical and bio-
promoter drive expression of both luciferase and GFP in a dose- logical fouling of the measurement cell and sensor, minimizing downtime
dependent fashion. The result is the ability to quantitatively measure and facilitating continuous testing.
pathway activation using luciferase activity or while imaging using GFP.
System Biosciences For info: 866-984-3766

-('''#&&%!&& !

www.systembio.com
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.
190 12 JULY 2019 • VOL 365 ISSUE 6449 sciencemag.org/custom-publishing SCIENCE
0712Product.indd 190 7/9/19 11:12 AM

online @sciencecareers.org
Faculty Position in Stem Cell Research
The Princess Margaret Cancer Centre at the University Health Network (UHN) is seeking an outstanding Scientist with a research program focused on
the biology and translational applications of stem cells in cancer. Expertise in genetics, epigenetics, or metabolism would be of asset.
We are seeking to recruit a PhD or MD/PhD scientist within the first years of completing their post-doctoral fellowship. The successful candidate will have
an outstanding research track record during their PhD and postdoctoral training and be expected to develop an independent world-class research program.
The Princess Margaret Cancer Centre, one of the top cancer research centres worldwide, has a longstanding history of landmark discoveries in stem
cell research dating back to the original identification of stem cells by Drs. Till and McCulloch. Understanding stem cell biology in cancer and normal
hematopoiesis remains a core strength and priority of the centre.
The Princess Margaret Cancer Centre is a member of the University Health Network (UHN) which encompasses Toronto General Hospital, Toronto Western
Hospital, Toronto Rehabilitation Institute, Michener Institute of Education and Princess Margaret Cancer Centre. Research across UHN’s research institutes
spans the full spectrum of diseases and disciplines, including cancer, cardiovascular sciences, transplantation, neural and sensory sciences, musculoskeletal
health, rehabilitation sciences, and community and population health. The breadth of research, the complexity of the cases treated, and the magnitude of
its educational enterprise has made UHN a national and international resource for patient care, research and education. With a long tradition of ground-
breaking firsts and a purpose of “Transforming lives and communities through excellence in care, discovery and learning”, UHN, Canada’s largest research
teaching hospital, bringing together over 16,000 employees, more than 1,200 physicians, 8,000+ students, and many volunteers. UHN is a caring, creative
place where amazing people are amazing the world.
The successful candidate will be eligible for appointment at the Assistant Professor level in the Department of Medical Biophysics or related departments
at the University of Toronto. Compensation for this position will be commensurate with experience and be consistent with UHN compensation policy
(range of $90,000 - $180,000 plus benefit package). The full-time permanent position is available immediately, but the search will remain open until the
position is filled. The position will be located in Toronto.
Interested candidates should apply below and also send their CV, as well as a description of their research interests and program to Dr. Aaron Schimmer,
Research Director, Senior Scientist, Staff Physician, Princess Margaret Cancer Centre, 610 University Avenue, Rm 7-504, Toronto, Ontario,
Canada M5G 2M9; E-mail: Aaron.Schimmer@uhn.ca
UHN is a respectful, caring, and inclusive workplace. We are committed to championing accessibility,
diversity and equal opportunity. All qualified candidates are encouraged to apply.
BIOCONTAINMENT RESEARCH ASSOCIATE

II or III
MICROBIOLOGY AND IMMUNOLOGY
A Biocontainment Research Associate Position is available immediately in the laboratory of Dr. Thomas
Geisbert, to support studies evaluating vaccines and treatments against BSL-4 viruses and to support work
characterizing the pathogenesis of these agents in animal models. Applicants must have a Bachelor’s degree
in basic science. The applicant must have at least two years of experience working with laboratory animals
Confused about your at ABSL-2. The applicant must meet all State and Federal requirements necessary for working with Select
Agents.
next career move? Preferred Qualifcations: LAT or LATG certifcation is preferred, but not required. More than two years of
experience working with viruses in BSL-3 or BSL-4 is preferred but not required. The ideal candidate would
Download Free Career have a DOJ clearance, LAT or LATG certifcation, and more than two years of BSL-3 or BSL-4 experience
Advice Booklets! working with laboratory animals infected with viruses or bacteria.
Duties: • Performs, coordinates and conducts standard research experiments following technical instructions
ScienceCareers.org/booklets in BSL-2, BSL-3, and BSL-4 work environments • Performs functions of research projects utilizing
hazardous viruses that require BSL-4 containment • Performs technical services including but not limited
to performing a variety of procedures on laboratory animals including exposure to viruses, blood collection,
tissue collection, vaccination, treatment, euthanasia, and necropsy. Other technical services include but are
not limited to performing clinical pathology assays (hematology, clinical pathology, blood coagulation)
and processing blood and tissue samples from infected animals and maintaining mammalian cell cultures •
Contributes to scientifc analysis of projects • Writes and reviews methods sections of manuscripts • Orders
and maintains lab inventory of supplies, animals, and equipment • Maintains records of all tests performed
and data collected, compiles and tabulates data, and provides analysis of results • Maintains knowledge
of BSL-2, BSL-3, and BSL-4 operations • Adheres to internal controls established for the department •
Performs related duties as required
Salary range is between $37,920 and $84,125 annually, commensurate with experience. Depending on the
experience and education, the candidate might be considered for a Biocontainment Research Associate III.
UTMB has several highly interactive research centers, biomedical institutes, and a national biocontainment
laboratory with excellent infrastructure to conduct research at BSL2, -3 and -4 on diverse animal models of
infectious diseases. The Department, with 31 full-time faculty members, is ranked among the top of its peer
departments in NIH funding. Interested candidates should apply to Job id #63819, Biocontainment Research
Associate II via the UTMB careers website at https://www.utmb.edu/careers/
UTMB Health strives to provide equal opportunity employment without regard to race, color, national
origin, sex, age, religion, disability, sexual orientation, gender identity or expression, genetic information
or veteran status. As a Federal Contractor, UTMB Health takes affrmative action to hire and advance
women, minorities, protected veterans and individuals with disabilities.
0712Recruitment_SC.indd 191 7/9/19 11:10 AM

CAREER FEATURE:
Postdoctoral
Careers
Issue date: August 30
Reserve space by August 15
Ads accepted until August 23
if space allows
Deliver your message to a

global audience of targeted,
qualifed scientists.
DIY and Science Hacking: What’s the best way to evolve

129,564 from garage scientist to full-time paid researcher? This
subscribers in print feature examines the DIY movement and its growth over
every week the last 10 years. This article will be read by scientists
looking for postdoc opportunities so be sure to include
473,652 your openings among the pages of this article.
monthly unique browsers
on ScienceCareers.org
What makes Science the best choice for recruiting?
55%  Read and respected by 400,000 readers around the globe

 Your ad dollars support AAAS and its programs, which
of our weekly readers
are Ph.D.s strengthens the global scientifc community.
Why choose this feature for your advertisement?
To book your ad, contact:

 Relevant ads lead oO the career section with a special
advertise@sciencecareers.org postdoctoral banner.
The Americas Expand your exposure by posting your print ad online:

202 326 6577
Europe  Link on the job board homepage to a landing page for
+44 (0) 1223 326527
postdoctoral jobs
Japan
+81 3 6459 4174  Additional marketing driving relevant job seekers to
China/Korea/Singapore/ the job board.
Taiwan
+86 131 4114 0012
Produced by the Science/AAAS Custom Publishing Ofce.
SCIENCECAREERS.ORG
FOR RECRUITMENT IN SCIENCE, THERE’S ONLY ONE SCIENCE.

myIDP:
A career plan customized
for you, by you.
For your career in science, there’s only one
Features in myIDP include:

 Exercises to help you examine your skills,
interests, and values.
 A list of 20 scientifc career paths with a
prediction of which ones best ft your skills
and interests.
 A tool for setting strategic goals for the
coming year, with optional reminders to
keep you on track.
 Articles and resources to guide you through
the process.
 Options to save materials online and print
them for further review and discussion.
 Ability to select which portion of your IDP
you wish to share with advisors, mentors,
or others.
 A certifcate of completion for users that
fnish myIDP.
Visit the website and start

planning today!
myIDP.sciencecareers.org
In partnership with:

WORKING LIFE
By Kazumasa Z. Tanaka
A father’s odyssey
T
he human resources representatives were amazed when I requested parental leave. I was one
of the first male postdocs to ask in the 100-year history of the institution, they told me. The
response didn’t surprise me. Here in Japan, where roughly 85% of academic scientists are men
and most have stay-at-home spouses, it’s rare for fathers to take leave from work after a baby is
born. But the real surprise came later, as I learned how hard it was to secure the leave—in spite
of careful planning and my supervisor’s support.
I wanted to buck Japan’s cultural I recently returned to work, along

norms and take paternity leave be- with my wife, and our daughters are
cause I felt strongly that I needed now well taken care of at day care.
to support my wife, who is also an I’m glad I persevered in my request
early-career scientist. She works for leave. I don’t think my gender ex-
as a postdoc at a different institu- plains the challenges I experienced,
tion and had arranged 3 months of but being a pioneer didn’t help. If
unpaid maternity leave. Caring for you’re in a place or situation where
a newborn is a lot of work, and I parental leave is a novelty, here are
didn’t want her to shoulder all the some tips to help smooth the way.
physical and mental exhaustion,
on top of recovering from the de- START PAPERWORK EARLY. Adminis-
livery. I also wanted to be home trative staff may have to chart new
with her and the baby during those waters, enacting a new policy or
first weeks. otherwise navigating the ins and
Our first child was born when we outs of processing such a request. It
were in the United States, where all takes time.
it’s more common for both parents “I wanted to buck Japan’s
to take leave. We worked in the cultural norms ASK FOR HELP. Talk to parents who
same lab—I was a Ph.D. student and have taken leave. When I was try-
my wife was a postdoc—and our and take paternity leave.” ing to figure out how the process
adviser was supportive of us taking worked, a female colleague who had
3 months of leave. I greatly appreciated having that time at children heard that I was trying to take leave and reached
home. When our second child was on the way, it was clear out to offer advice. Her help was invaluable.
to me that I needed to make the same thing happen.
My postdoc supervisor was on board with my plan to STAND FIRM. You only have one chance at that time with your
take 3 months off. He congratulated me on my wife’s preg- kids. Unfortunately, some employers will not grant parental
nancy and told me it was my right to take leave—support leave. Others will retaliate after requests are made, giving
for which I was grateful. I made my initial inquiry 6 months employees poor evaluations or even terminating them. Get
before our baby was due, thinking that would give my insti- to know your legal rights by talking to a lawyer or labor
tution plenty of time to figure out how to grant my request. union. In Japan, and many other countries, retaliation is
But then the hassle began. Under Japanese law, an em- illegal and parents are entitled to a period of parental leave.
ployee isn’t eligible for paid parental leave if their contract
is set to end before the child reaches 18 months of age. My RAISE YOUR VOICE. If more parents demand better parental
3-year postdoc contract only extended to 3 months after leave policies, the road will become easier for those who
ILLUSTRATION: ROBERT NEUBECKER

our child would be born. The same law prevented my wife follow. Better family-friendly policies will help all parents
from taking paid leave, and we didn’t want to go months succeed in science—fathers and mothers alike. j
without a paycheck. So, my supervisor and I came up with
a plan to draw up a new extended contract for me—but Kazumasa Z. Tanaka is a research scientist at RIKEN in
various legal and administrative hurdles meant that it Wako, Japan. Do you have an interesting career story?
took months to be finalized. Send it to SciCareerEditor@aaas.org.
0712WorkingLife.indd 194 7/3/19 3:43 PM

B I O LO GY ° E A R T H ° E N V I R O N M E N T ° L I F E ° I N T E R D I S C I P L I N A RY ° P H YS I CA L ° M AT E R I A L ° S O C I A L S C I E N C E S
O P E N AC C E S S , D I G I TA L , A N D F R E E T O A L L R E A D E R S
Pushing the Boundaries of Knowledge

As AAAS’s frst multidisciplinary, open access journal, Science Advances publishes
research that refects the selectivity of high impact, innovative research you expect
from the Science family of journals, published in an open access format to serve a
vast and growing global audience. Check out the latest fndings or learn how to
submit your research: ScienceAdvances.org
0712Product.indd 195 7/2/19 9:00 AM

Share Your Robotics
Research with the World.
Transforming the Future of Robotics in Research !

As a multidisciplinary online-only journal, Science Robotics publishes original, peer-reviewed,
research articles that advance the feld of robotics. The journal provides a central forum for
communication of new ideas, general principles, and original developments in research and
applications of robotics for all environments.
Submit your research today. Learn more at: ScienceRobotics.org
0712Product.indd 196 7/2/19 9:00 AM

Science20190712 DL Artificial Muscles

Uploaded by

Copyright:

Available Formats

Science20190712 DL Artificial Muscles

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Science20190712 DL Artificial Muscles

Uploaded by

Copyright:

Available Formats

Can biodiversity survive on Boosting CAR–T cells for Promoting reproducibility

0712Product.indd 98 7/2/19 9:00 AM

NEWS 111 WAS OUR SPECIES IN EUROPE 210,000

104 News at a glance 122 STROMAL DIRECTIVES CAN

SCIENCE sciencemag.org 12 JULY 2019 • VOL 365 ISSUE 6449 99

0712TOC.indd 99 7/9/19 5:07 PM

RESEARCH ARTICLES 176 REPRODUCTION

141 NEUROSCIENCE DEPARTMENTS

RESEARCH Enhanced CAR–T cell activity

100 12 JULY 2019 • VOL 365 ISSUE 6449 sciencemag.org SCIENCE

0712TOC.indd 100 7/9/19 5:07 PM

Balancing science and security

SCIENCE sciencemag.org 12 JULY 2019 • VOL 365 ISSUE 6449 101

0712Editorial.indd 101 7/9/19 4:34 PM

Best of both worlds:

0712Product.indd 102 7/2/19 9:00 AM

Decoding the immune system fresh-frozen and FFPE samples. The

subpopulations of cancer cells,” according to Jorge Reis-Filho and References

0712Product.indd 103 7/2/19 9:00 AM

Hungary’s academy loses control

#MeToo activist out at Vanderbilt

PHOTO: NASA/ESA/J. HESTER AND A. LOLL/ARIZONA STATE UNIVERSITY

104 12 JULY 2019 • VOL 365 ISSUE 6449 sciencemag.org SCIENCE

0712NewsInBrief.indd 104 7/9/19 5:28 PM

A 7.1-magnitude earthquake on 5 July

SCIENCE sciencemag.org 12 JULY 2019 • VOL 365 ISSUE 6449 105

0712NewsInBrief.indd 105 7/9/19 5:28 PM

Polio eradication campaign loses ground

106 12 JULY 2019 • VOL 365 ISSUE 6449 sciencemag.org SCIENCE

0712NewsInDepth.indd 106 7/9/19 5:29 PM

administration on the use of federal fund-

SCIENCE sciencemag.org 12 JULY 2019 • VOL 365 ISSUE 6449 107

0712NewsInDepth.indd 107 7/9/19 5:29 PM

dent Barack Obama’s administration later

108 12 JULY 2019 • VOL 365 ISSUE 6449 sciencemag.org SCIENCE

0712NewsInDepth.indd 108 7/9/19 5:29 PM

Gut microbes may help malnourished children

SCIENCE sciencemag.org 12 JULY 2019 • VOL 365 ISSUE 6449 109

0712NewsInDepth.indd 109 7/9/19 5:29 PM

Psychologist aims to study

PHOTO: DANIEL HAUN/MAX PLANCK INSTITUTE FOR EVOLUTIONARY ANTHROPOLOGY

110 12 JULY 2019 • VOL 365 ISSUE 6449 sciencemag.org SCIENCE

0712NewsInDepth.indd 110 7/9/19 5:29 PM

Was our species in Europe 210,000 years ago?

SCIENCE sciencemag.org 12 JULY 2019 • VOL 365 ISSUE 6449 111

0712NewsInDepth.indd 111 7/9/19 5:29 PM

MAKING PEACE WITH OIL PALM

By Dyna Rochmyaningsih, in Libo on Sumatra, Indonesia

112 12 JULY 2019 • VOL 365 ISSUE 6449 sciencemag.org SCIENCE

SCIENCE sciencemag.org 12 JULY 2019 • VOL 365 ISSUE 6449 113

5 Nest boxes that abandoning herbicides improved the

114 12 JULY 2019 • VOL 365 ISSUE 6449 sciencemag.org SCIENCE

SCIENCE sciencemag.org 12 JULY 2019 • VOL 365 ISSUE 6449 115

Seabird clues to ecosystem health

116 12 JULY 2019 • VOL 365 ISSUE 6449 sciencemag.org SCIENCE

0712Perspectives.indd 116 7/3/19 2:21 PM

Ecosystem-based management, an alterna- such as mangrove cover, fish abundance, 10.1126/science.aaw9999

SCIENCE sciencemag.org 12 JULY 2019 • VOL 365 ISSUE 6449 117

0712Perspectives.indd 117 7/3/19 2:21 PM