Science20200103 DL

What’s in store in 2020?
A sci-f legend’s legacy Earliest cooked root

Science looks ahead p. 6 lives on p. 20 vegetables p. 87
$15
3 JANUARY 2020
sciencemag.org
TINY
CARNIVOROUS
PLANTS
Convoluted shapes
from simple rules pp. 24 & 91
TRILLIONS
OF MICROBES
ONE ESSAY
The NOSTER Science Microbiome Prize is an international prize
that rewards innovative research by investigators, under the age
of 35, who are working on the functional attributes of the micro-
biota. The research can include any organism that has potential to
contribute to our understanding of human or veterinary health and
disease, or to guide therapeutic interventions. The winner and
/nalists will be chosen by a committee of independent scientists,
chaired by a senior editor at Science. The top prize includes a
complimentary membership to AAAS, an online subscription to
Science, and $25,000 1USD4. Submit your research essay today.
Apply by 1/24/20 at www.sciencemag.org/noster

Sponsored by NOSTER Bio-Institute, Nitto Pharmaceutical Industries, Ltd.
0103Product.indd 2 12/20/19 8:48 AM

71
CONTENTS
3 JA N UA RY 2 0 2 0 • VO LU M E 3 6 7 • I S S U E 6 47 3
What makes
knots tight?
9
Afterglow of a 1962 nuclear test in space
CREDITS (FROM TOP): (IMAGE) V. P. PATIL ET AL.; (PHOTO) SCIENCE HISTORY IMAGES/AL AMY STOCK PHOTO; (PHOTO) TONY HEALD/MINDEN PICTURES
NEWS
13 Congress again rejects Trump cuts,
smiles on science agencies
2020 spending bill increases research
INSIGHTS
budgets, but next year will be tougher as
IN BRIEF lawmakers will have less to spend BOOKS ET AL.
By J. Merivs
6 What’s coming up in 2020 20 Asimov at 100
PODCAST 14 Global polio eradication falters From epic space operas to rules for
in the final stretch robots, the prolific author’s literary
IN DEPTH Vaccine-derived outbreaks may force a legacy endures By J. Gunn
9 U.S. military tests radiation belt change in “endgame” strategy By L. Roberts
PERSPECTIVES
cleanup in space
Radio waves could sweep belts clean FEATURES 22 Uncovering the ART of antimalarial
of satellite-killing particles after nuclear 16 Fighting words resistance
sneak attack By R. Stone Virologist Roberto Burioni has become a A key mechanism of resistance to the
celebrity in Italy by sparring with vaccine antimalarial drug artemisinin is identified
10 Past megadroughts hit North skeptics By D. Starr By D. Marapana and A. F. Cowman
and South America in tandem RESEARCH ARTICLE p. 51
Strong La Niña conditions drove
deep medieval droughts By P. Voosen 23 Majorana fermions go for a ride
11 Study pushes emergence of measles

6 Evidence for propagating Majorana
quasiparticles is found
back to antiquity in a topological superconductor
The virus may have entered the human By S. Tewari and T. D. Stanescu
population when cities grew large enough to REPORT p. 104
sustain outbreaks By K. Kupferschmidt
24 Building a carnivorous trap
12 Computer scientist in line to Experiments and computations reveal
become next NSF director developmental origins of cup-shaped leaves
Sethuraman Panchanathan would By D. E. Moulton and A. Goriely
succeed France Córdova By J. Mervis REPORT p. 91
SCIENCE sciencemag.org 3 JANUARY 2020 • VOL 367 ISSUE 6473 3
0103TOC.indd 3 12/27/19 3:48 PM

CONTENTS
83 Neuroscience
Dendritic action potentials and computation
in human layer 2/3 cortical neurons
A. Gidon et al.
87 Archaeology
Cooked starchy rhizomes in Africa
170 thousand years ago L. Wadley et al.
91 Plant science
Evolution of carnivorous traps from planar
leaves through simple shifts in gene
expression C. D. Whitewoods et al.
PERSPECTIVE p. 24; PODCAST
30 96 Evolutionary biology
Protein-coding changes preceded
cis-regulatory gains in a newly evolved
26 Electrostatics affect the glow RESEARCH ARTICLES transcription circuit C. S. Britton et al.
Chromophore twisting is probed 40 Nanoparticles
with unnatural amino acids By C. Hu et al. Oriented attachment induces fivefold twins 100 Tubulin
REPORT p. 76 by forming and decomposing high-energy TTC5 mediates autoregulation of tubulin
grain boundaries M. Song et al. via mRNA degradation Z. Lin et al.
27 Cancer in sub-Saharan Africa PERSPECTIVE p. 29
Knowledge of cancer in Africa brings needed 45 Multiplex genomics
diversity to improve health worldwide Massively multiplex chemical transcriptomics 104 Topological matter
By T. R. Rebbeck at single-cell resolution S. R. Srivatsan et al. Evidence for dispersing 1D Majorana
channels in an iron-based
29 Gene expression regulated by 51 Malaria superconductor Z. Wang et al.
RNA stability A Kelch13-defined endocytosis pathway PERSPECTIVE p. 23
The factor responsible for autoregulation mediates artemisinin resistance in malaria
of tubulin RNA stability is identified parasites J. Birnbaum et al.
By O. Shoshani and D. W. Cleveland PERSPECTIVE p. 22
DEPARTMENTS
REPORT p. 100
REPORTS 5 Editorial
POLICY FORUM 59 Topological optics Clarity in 2020 By H. Holden Thorp
30 Sustainable minerals and metals A single photonic cavity with two
independent physical synthetic dimensions 114 Working Life
for a low-carbon future Strange dreams By Desiree Dickerson
Policy coordination is needed for global A. Dutt et al.
supply chains By B. K. Sovacool et al.
64 Topological matter
Absence of evidence for chiral Majorana ON THE COVER
LETTERS
modes in quantum anomalous Hall- Immature trap of the
34 NextGen Voices: superconductor devices M. Kayyalha et al. carnivorous plant
Making science accessible
Utricularia gibba (~200
68 Superconductivity µm in width), illustrat-
Atomic manipulation of the gap in ing its two-layered
RESEARCH Bi2Sr2CaCu2O8+x F. Massee et al. cellular structure. Such
cup-shaped leaves have
PHOTO: MICHAEL ROBINSON CHAVEZ/THE WASHINGTON POST VIA GETTY IMAGES

71 Applied physics evolved multiple times
Topological mechanics of knots and as a mechanism for trap-
IN BRIEF tangles V. P. Patil et al. ping animals. Analyzing U. gibba, researchers
36 From Science and other journals used a combination of molecular genetics
76 Fluorescent proteins and computer modeling to reveal how shifts
REVIEW Electrostatic control of photoisomerization in gene expression domains can account for
pathways in proteins M. G. Romei et al. repeated evolution of cup shapes from spe-
39 Neuroscience cies with planar leaves. See pages 24 and 91.
PERSPECTIVE p. 26
Memory engrams: Recalling the past Image: Karen Lee and Claire Bushell
and imagining the future
S. A. Josselyn and S. Tonegawa 79 Optics
REVIEW SUMMARY; FOR FULL TEXT: On-chip integrated laser-driven particle New Products ..............................................110
DX.DOI.ORG/10.1126/SCIENCE.AAW4325 accelerator N. V. Sapra et al. Science Careers .......................................... 111
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 © 2020 by the American Association for the Advancement of Science. The title SCIENCE is a registered trademark of the AAAS. Domestic
individual membership, including subscription (12 months): $165 ($74 allocated to subscription). Domestic institutional subscription (51 issues): $2148; Foreign postage extra: Air assist delivery: $98. First class, airmail, student, and
emeritus rates on request. Canadian rates with GST available upon request, GST #125488122. Publications Mail Agreement Number 1069624. Printed in the U.S.A.
Change of address: Allow 4 weeks, giving old and new addresses and 8-digit account number. Postmaster: Send change of address to AAAS, P.O. Box 96178, Washington, DC 20090–6178. Single-copy sales: $15 each plus shipping and
handling available from backissues.sciencemag.org; bulk rate on request. Authorization to reproduce material for internal or personal use under circumstances not falling within the fair use provisions of the Copyright Act can be obtained
through the Copyright Clearance Center (CCC), www.copyright.com. The identification code for Science is 0036-8075. Science is indexed in the Reader’s Guide to Periodical Literature and in several specialized indexes.
4 3 JANUARY 2020 • VOL 367 ISSUE 6473 sciencemag.org SCIENCE
0103TOC.indd 4 12/27/19 3:48 PM

EDITORIAL
Clarity in 2020
W
e begin 2020 (or should I say 20/20) awash One of the questions I am often asked is how the Sci-
in references to ophthalmology. In that vein, ence journals correct or retract papers. With the rapid
we hope to make science clearer and bright- analysis that can happen on social media, these questions
er in the months ahead—which brings me are being raised faster than ever before. We don’t seek
to the issue of transparency. to hide from these efforts: Image sleuth Elisabeth Bik
The Science family of journals looks for- (@MicrobiomDigest) recently stated on Twitter that
ward to a year in which the editors will strive she needed a list of journal editors’ emails and I im-
for greater transparency and reproducibility in the sci- mediately responded with mine (which is not hard to H. Holden Thorp
ence that we publish. Policies on conflicts of interest find at right).
Editor-in-Chief,
and professional behavior for authors are now uniform Science journals prefer to publish a retraction signed
Science journals.
across our journals and strengthen our standards for dis- by all authors. Deciding to retract in cases where not
hthorp@aaas.org;
closure. For the first time, we will require general disclo- all authors agree can require a lengthy investigation
sures from Science’s Board of Reviewing Editors; prior by universities. In such cases, we publish an Editorial @hholdenthorp
to this year, they were only asked to disclose conflicts Expression of Concern (EEoC) in the meantime, to
that arose for papers on which they were consulted. quickly alert readers that concerns have been raised
As readers of Science’s news section about the reported work. We hope
would know, I have been active for de- that making this information avail-
cades in the biotechnology business
and disclosed competing interests
“…editors will able promptly helps mitigate the
waste of time, effort, and funds by
to the American Association for the
Advancement of Science (AAAS, pub-
strive for greater researchers who might otherwise
base future work on papers that
lisher of the Science journals) upon
joining the editorial team. These inter-
transparency are later retracted. During my short
time as editor-in-chief, when we
ests are stated on my web bio (www.
sciencemag.org/about/leadership-
and reproducibility…” have contacted universities to be-
gin this process, I have been giving
and-management#HoldenThorp), a deadline by which we will decide
and I commit to keeping them updated for full public if we are going to proceed with an EEoC. After run-
disclosure. The Science journals believe that the com- ning academic units for the past 14 years, I know that
mercial application of scientific findings is critical to it sometimes takes a nudge to get things to the top of
gaining public support and to ensuring that scientists an inbox. We also publish an EEoC when authors alert
are involved in the realization of their ideas. At the us to problems with their published paper and need
same time, vigorous disclosure of these competing in- time to determine whether the findings hold. Authors
terests is a must. taking the initiative to correct the record strengthens the
We will also continue to strive for greater reproduc- integrity of the scientific enterprise.
ibility in the science that we publish. Last year, Science The year 2020 brings a presidential election to the
helped to develop a framework setting out minimal United States and a time of transition across the At-
expectations for materials, design, analysis, and re- lantic. There will be great opportunities for everyone
porting (MDAR) and also participated with other to stand up for science and for their beliefs. Break-
publishers in piloting a checklist that operational- throughs in all fields will continue with new ideas, in-
ized this framework (https://cos.io/blog/journals-test- sights, and applications, hopefully for the good of all
materials-design-analysis-reporting-mdar-checklist/). members of society and for the Earth that we cherish.
Authors were admirably enthusiastic about this pro- And it will be exciting for AAAS, as our new chief ex-
cess, and we are pleased to endorse the framework ecutive officer, Sudip Parikh, begins next week. I have
and implement the checklist this year for life science known Sudip for 25 years and look forward to his gen-
papers in Science. This should allow information to be erous and thoughtful leadership.
used by others seeking to reproduce findings and will
hopefully pave the way for more such standardization. –H. Holden Thorp
PHOTO: CAMERON DAVIDSON
10.1126/science.aba6293
0103Editorial.indd 5 12/20/19 5:55 PM

NEWS
IN BRIEF
AREAS TO WATCH
New goals for saving biodiversity
What’s coming up in 2020 | This year will see an
C O N S E R VAT I O N
attempt to revitalize the ambitious Aichi
I
ncessant political turmoil in the United Kingdom, United Biodiversity Targets, named for the city in
States, and other nations will likely last well into the new year, Japan where they were negotiated. Since
they were approved 10 years ago, there has
complicating many researchers’ work. The U.K. election last been little to no progress in meeting most
month made the country’s departure from the European Union of those 20 goals, such as preventing the
a near-certainty, and its scientists now face losing EU science decline of endangered species. That alarm-
grants and scientific collaborators. In the United States, a presi- ing situation was highlighted last year in
PHOTO: TONY HEALD/MINDEN PICTURES

a major scientific assessment by another
dential election in November will determine the role of scientists organization, the Intergovernmental
in future policy deliberations; many experts on climate change Science-Policy Platform on Biodiversity
and other environmental issues assert that the Trump administra- and Ecosystem Services. But in October,
tion has ignored scientific evidence. In this section, Science’s news nations will have a chance to try to set a
more effective course when they meet in
staff forecasts other areas of policy and research likely to make
Kunming, China, to review and revise the
news this year amid the chaos, from dark matter detectors to new Convention on Biological Diversity,
efforts to rein in loss of species. the world’s flagship conservation pact.
0103NewsInBrief.indd 6 12/27/19 11:32 AM

Despite long-standing Without stepped-up efforts, there is little can apply them to understand more about
biodiversity conservation hope the world can keep future warming older fossils lacking DNA, including the sex
plans, both African elephants below 2°C, the level scientists forecast and age of remains of enigmatic ancient
and African wild dogs will produce catastrophic damage to hominins. Most hominins are known by
are listed as endangered. human communities and ecosystems. bones and teeth alone, and proteins could
provide a new tool for sorting them in
family trees and to identify fragments too
Counting on the census small to classify. Although tooth enamel
DEMOGRAPHICS | The U.S. government has offers the best source of ancient proteins,
conducted a decennial census since 1790. researchers are also extracting them from
But the 2020 census that kicks off on 1 April bones and hair. In addition, proteins can
faces unprecedented political challenges. reveal new information about artifacts
Although civil rights groups won a fight to made of plant and animal materials, and
block a question about citizenship that the researchers hope this year to analyze
Trump administration wanted to add, it parchment manuscripts and the beeswax
has ordered the Census Bureau to generate once used to seal documents. Scientists
the equivalent data using existing govern- are also analyzing residues on pots for
ment records so that states can use the more clues to whether early pastoralists in
information when they redraw boundaries the steppelands of Mongolia, for example,
for federal and state elections. Researchers drank camel or goat milk first—and what
fear that assignment may not be doable, people living on the edge of the Roman
and that the political debate has further Empire in England ate.
alienated those already hardest to count.
Demographers also worry that the census’
use of a new way to protect respondents’
privacy could distort analyses of demo-
graphic trends. At stake are not only how
more than $1.5 trillion in federal funds are
distributed each year, but also the integrity
of the nation’s largest statistical agency.
CRISPR’s big clinical tests

BIOMEDICINE | The CRISPR gene-editing
tool faces key tests this year of its promise
to treat cancer and genetic diseases. A small
U.S. clinical trial is using CRISPR to disable Proteins in these
three genes in T cells that are then returned 400-year-old bone
to a cancer patient’s body, an approach that fragments, found in
could help these immune system soldiers Iroquois settlements
stop malignant cells from growing and in Canada, revealed whether
extend patients’ lives. More results may also they were animal or human.
come from separate CRISPR cancer trials
in China. Other researchers are working to
treat people with sickle cell disorder and ‘Foreign influence’ worries grow
Crunch time for climate policy thalassemia by using the DNA editor to turn | The political debate
N AT I O N A L S E C U R I T Y
POLICY | The politics of climate change on the gene for a fetal version of hemoglo- over how to respond to China’s emer-
faces crucial moments this year. The bin to compensate for a defective adult form gence as a scientific superpower is likely
Trump administration’s opposition to of the oxygen-carrying protein; last fall, sci- to intensify this year. In the United States,
regulations reducing fossil fuel emissions entists reported success in two patients and some federal agencies have banned their
PHOTO: K. MCGRATH ET AL., SCIENTIFIC REPORTS 9, 11027 (2019)
has emerged as a primary talking point in 2020 will present longer-term results for employees from participating in foreign
for the president’s Democratic challengers. a larger group. Another clinical trial in the talent recruitment programs—an approach
One day after the U.S. presidential election United States could show whether CRISPR that China has used to connect with thou-
on 3 November, the country, the second improves vision in people with an inherited sands of scientists—to prevent disclosure
largest emitter of greenhouse gases, is set disorder that causes progressive blindness. of information that could damage national
to leave the Paris climate accord, although security and U.S. economic competitiveness.
a Democratic president could quickly rejoin Two new bodies created by Congress will
after taking office in 2021. Less than 1 week Proteins tell ancient tales work to harmonize practices across federal
later, the United Nations will convene | Ancient proteins will
A R C H A E O L O GY agencies and chew over how best to balance
in Glasgow, U.K., for its most important shed new light this year on the identity openness and security. U.S. academic lead-
climate summit since 2015, where nations and behavior of humans and other animals ers are hoping to convince policymakers
are expected to increase their pledges to that lived more than 1 million years ago. not to fence off certain types of research,
cut greenhouse gas emissions—even though Proteins are more stable than DNA, and as which they say would throttle U.S. innova-
they are behind on meeting existing ones. analytical methods improve, researchers tion. A new report to the National Science

NEWS | I N B R I E F
Foundation says teaching students and only through the feeble weak nuclear force. person. Transplants from these edited pigs
faculty members about acceptable and This year, the XENON-NT detector, which to monkeys, a key test of safety and efficacy
unacceptable behavior is a better approach. contains 8 tons of frigid liquid xenon, will before human trials, have demonstrated
turn on in the subterranean Gran Sasso long-term viability in their new hosts.
National Laboratory in Italy. At the Sanford
Japan boosts neutrino efforts Underground Research Facility in South
PA R T I C L E P H YS I C S | Japan is expanding Dakota, the LUX-ZEPLIN (LZ) detector, Exascale computer to debut
neutrino research to better understand which contains 10 tons of liquid xenon, COMPUTER SCIENCE | This year, China is
properties of the phantom particles and the will also power up. If XENON-NT and expected to win the race to build the world’s
cosmic processes that produce them. This the LZ see nothing in the next few years, first exascale computer, capable of carrying
spring, scientists will increase the sensitiv- dark matter hunters could push for bigger out 1 billion billion (1018) calculations per
ity of the 22-year-old Super-Kamiokande WIMP detectors or set their sights on other second, also known as an exaflop. Just which
neutrino observatory by doping water in hypothesized forms of dark matter. The supercomputer will be the first remains
its observation chamber with the rare- Italian lab’s future also remains uncertain, uncertain, as China has set up a competi-
earth metal gadolinium. The detector will as former lab officials face prosecution for tion between three institutions: the National
then watch for signals generated when allegedly allowing contamination of local Supercomputing Center of Tianjin, the
neutrinos from supernovae hit the water, drinking water. National Supercomputing Center in Jinan,
providing clues about the dynamics within and Dawning Information Industry Co., a
those exploding stars. Japan’s legislature manufacturer also known as Sugon. The
is expected to fund an even bigger step: Making xenotransplants survive new Chinese supercomputers, and others to
construction of the 72 billion Japanese yen BIOMEDICINE | The genome editor CRISPR follow in the European Union, Japan, and
($660 million) Hyper-Kamiokande. Ten is reinvigorating the beleaguered field of the United States, will be used to analyze
times larger than its predecessor, it will xenotransplantation, which aims to surgi- vast data sets from astronomical and genetic
capture that much more data about neutri- cally replace human organs or tissues with surveys, and will support the continued rise
nos emanating from the Sun, distant stars, ones harvested from animals such as pigs. of artificial intelligence. Some computer
and supernovae. Novel clinical trials of the strategy could scientists expected the exascale milestone
launch this year. Xenotransplantation to have come sooner; delays resulted in part
has long promised to alleviate a chronic from the need to develop energy efficient
Dueling dark matter detectors shortage of human livers, hearts, and other computer chips.
| The race to detect
A S T R O P H YS I C S organs. It could also provide corneas to cure
hypothetical particles of dark matter—the blindness and insulin-producing islet cells
invisible stuff that binds together the galax- to replace those destroyed by diabetes. But ALSO IN 2020
ies with its gravity—enters a new phase this time and time again in earlier tests, human
year with the startup of two powerful new immune systems have quickly destroyed ALZHEIMER’S DRUG The U.S. Food and
underground detectors. Since the 1980s, the foreign transplants. Recent CRISPR Drug Administration will decide whether
physicists have used ever bigger and more experiments have modified genes in pigs to approve aducanumab, an antibody
sensitive ones to search for so-called weakly to prevent or dampen human immune drug designed to bust the brain-clogging
interacting massive particles (WIMPs), responses to their tissue and have removed amyloid plaques of Alzheimer’s disease.
theorized to weigh 100 times as much as DNA from the porcine genome that could The experimental treatment has shown
protons and to interact with other matter spawn potentially dangerous viruses in a mixed success in clinical trials.
OCEAN CONSERVATION The United

Nations intends to finish plans for a
Decade of Ocean Science to begin in
2021. The goal is to coordinate work
by scientists around the world to help
improve ocean health. One expected
emphasis is mapping more of the world’s
vulnerable marine ecosystems and bio-
PHOTO: MATTHEW KAPUST/SANFORD UNDERGROUND RESEARCH FACILITY

diversity hot spots and more of the
ocean’s bottom, only about 4% of which
has been charted in high resolution.
STEM CELL FUNDING California voters will

decide in November whether to allocate
$5.5 billion from bond sales to keep alive
the California Institute for Regenerative
Medicine. The funding agency was cre-
ated through a $3 billion ballot initiative
in 2004 to translate stem cell research
into new therapies.
SCIENCEMAG.ORG/NEWS
The LUX-ZEPLIN dark matter detector is readied to record data at an underground lab in South Dakota. Read more news from Science online.

IN DEP TH
Aurorae were seen widely

after Starfish Prime, a
1962 nuclear test in space.
SPACE PHYSICS
U.S. military tests radiation belt cleanup in space

Radio waves could sweep belts clean of satellite-killing particles after nuclear sneak attack
By Richard Stone experiments—one now in orbit and two be- 10 years ago, we just know so much more
ing readied for launch in 2021—aim to gather about how these wave-particle interactions
T
he U.S. military thought it had cleared data on how to drain high-energy electrons work,” says Geoff Reeves, a space physicist
the decks when, on 9 July 1962, it trapped by Earth’s magnetic field in radia- at Los Alamos National Laboratory.
heaved a 1.4-megaton nuclear bomb tion belts encircling the planet. The process, Now, researchers are ready to try artifi-
some 400 kilometers into space: Or- called radiation belt remediation (RBR), al- cial remediation, by beaming radio waves
biting satellites were safely out of ready happens naturally, when radio waves into the belts. Physicists have tested using
range of the blast. But in the months from deep space or from Earth—our own the U.S. Navy’s very low frequency (VLF) an-
that followed the test, called Starfish Prime, radio chatter, for example, or emissions tenna towers, powerful facilities used to com-
satellites began to wink out one by one, in- from lightning—knock electrons trapped in municate with submarines, says Dan Baker,
cluding the world’s first communications Earth’s Van Allen radiation belts into the up- director of the Laboratory for Atmospheric
satellite, Telstar. There was an unexpected per atmosphere, where they quickly shed en- and Space Physics at the University of Colo-
aftereffect: High-energy electrons, shed by ergy, often triggering aurorae. rado, Boulder, and a lead investigator on the
radioactive debris and trapped by Earth’s “Natural precipitation happens all the Van Allen Probes. The antennae of the High-
magnetic field, were fritzing out the satel- time,” says Craig Rodger, a space physicist frequency Active Auroral Research Program
lites’ electronics and solar panels. at the University of Otago. But it would in Alaska and the giant dish of the Arecibo
Starfish Prime and similar Soviet tests not nearly be fast enough to drain nuclear- Observatory in Puerto Rico might also be en-
might be dismissed as Cold War misadven- charged radiation belts, where electron listed to generate cleansing radio beams.
PHOTO: SCIENCE HISTORY IMAGES/ALAMY STOCK PHOTO
tures, never to be repeated. After all, what fluxes can be millions of times higher than An orbiting RBR platform, closer to the
nuclear power would want to pollute space in Earth’s Van Allen belts. target, could be more effective. In June
with particles that could take out its own Scientists got a glimpse of a potential 2019, the U.S. Air Force launched what it
satellites, critical for communication, naviga- solution from NASA’s Van Allen Probes, bills as the largest uncrewed structure ever
tion, and surveillance? But military planners which launched in 2012 and ducked in and flown in space: the DSX dipole antenna.
fear North Korea might be an exception: It out of Earth’s radiation belts until the mis- Nearly as long as a U.S. football field, DSX’s
has nuclear weapons but not a single func- sion ended last summer. It offered a deep primary mission is to transmit VLF waves
tioning satellite among the thousands now in dive into natural remediation processes, into the Van Allen belts and measure pre-
orbit. They quietly refer to a surprise orbital showing how radio waves resonate with cipitating particles with onboard detectors.
blast as a potential “Pearl Harbor of space.” high-energy electrons, scattering them “It’s a new way to prod the belts and explore
And so, without fanfare, defense scien- down the magnetic field lines and sweep- basic questions in space physics,” says DSX’s
tists are trying to devise a cure. Three space ing them out of the belts. “Compared to principal investigator, James McCollough at
0103NewsInDepth.indd 9 12/27/19 3:50 PM

NEWS | I N D E P T H
the Air Force Research Laboratory.

A team of scientists at Los Alamos and
NASA’s Goddard Space Flight Center is
spearheading a second experiment in VLF
precipitation. In April 2021, the team plans
to launch a sounding rocket carrying the
Beam Plasma Interactions Experiment, a
miniature accelerator that would create
a beam of electrons, which in turn would
generate VLF waves capable of sweeping
up particles. Reeves, who leads the ex-
periment, believes the compact electron
accelerator could ultimately be a better
broom than a gigantic VLF antenna. “If we
validate it with this experiment, we have a
lot more confidence we can scale it up to
higher power,” he says.
A third experiment would coax the at- CLIMATE
mosphere itself to kick up turbulent waves
that would draw down electrons. In the
summer of 2021, the Naval Research Labo-
ratory plans to launch a mission called the
Past megadroughts hit North
Space Measurements of a Rocket-Released
Turbulence. A sounding rocket will fly
into the ionosphere—an atmospheric layer
and South America in tandem
hundreds of kilometers up that’s awash in Strong La Niña conditions drove deep medieval droughts
ions and electrons—and eject 1.5 kilograms
of barium atoms. Ionized by sunlight, the By Paul Voosen But pinning down the exact timing wasn’t
barium would create a ring of moving possible in the 1990s. Ancient drought is
F
plasma that emits radio waves: essentially or 10 years, central Chile has been typically detected in variations in the width
a space version of a magnetron, the gadget gripped by unrelenting drought. With of tree rings, but tree ring records then were
used in microwave ovens. 30% less rainfall than normal, verdant spotty. Since, however, tree ring scientists
The missions should help show which landscapes have withered, reservoirs have put together “drought atlases” that
RBR system is most feasible, although an are low, and more than 100,000 farm provide consistent records for much of the
operational system may be years off. What- animals have died. The dry spell has world. “We’ll have most hemispheric land ar-
ever the technology, it could bring risks. lasted so long that researchers are calling it a eas covered by the end of the year,” says Ed
A full-scale space cleanup might dump as “megadrought,” rivaling dry stretches centu- Cook, a tree ring scientist at Columbia.
much energy into the upper atmosphere ries ago. It’s not so different from the decade- Steiger combined these records with
as the geomagnetic storms caused by the long drought that California, some 8000 kilo- thousands of other proxies for dryness
Sun’s occasional eruptions. Like them, meters away, endured until last year. and temperature from trees, corals, ocean
it could disrupt airplane navigation and By analyzing tree ring records, scientists sediments, and ice cores, and fed them
communication. And it would spawn heaps have now found evidence that such tandem into a global climate model. Aligning it-
of nitrogen oxides and hydrogen oxides, droughts are more than a coincidence: They self to the records, it generated a global
which could eat away at the stratospheric are surprisingly common over the past 1200 view of the changing climate, even in
ozone layer. “We don’t know how great years, and they may often share a common places with sparse proxies. The model
the effect would be,” says Allison Jaynes, cause—an abnormally cool state of the east- confirmed that, from 800 to 1600 C.E.,
a space physicist at the University of Iowa. ern Pacific Ocean known as La Niña. “We multiple megadroughts occurred simulta-
Besides safeguarding against a nuclear did not expect there to be as much coher- neously across the hemispheres. “It’s there,”
burst, RBR technology could have a cience as we see,” says Nathan Steiger, a paleo- Cook says. “Without question, it’s there.”
vilian dividend, Jaynes notes. NASA and climatologist at Columbia University who Besides correlating the varied climate re-
other space agencies have long wrestled presented the work last month at a meeting cords, the model also identified the key fac-
with shielding astronauts from the Van of the American Geophysical Union. “They tors driving the climate variations. Steiger
Allen belts and other sources of radia- just happen together.” The results suggest and his co-authors, including Cook, first
tion on their way to and from deep space. that, in the future, extreme aridity could used the new tool to look at megadroughts
PHOTO: RODRIGO GARRIDO/REUTERS/NEWSCOM

VLF transmitters might be used to clear strike all along the Americas’ western coast. in the U.S. Southwest. Their study, published
out high-energy electrons just before a Evidence for synchronous, hemisphere- last year in Science Advances, is “amazing,”
spacecraft enters a danger zone. “When wide droughts first emerged in a 1994 study in says David Stahle, a tree ring scientist at the
we become more active space travelers,” Nature, which documented dead tree stumps University of Arkansas in Fayetteville. “It’s a
she says, “it could provide a safe passage in the middle of lakes and rivers in both Pa- bit like … they took those black-and-white
through the radiation belts.” j tagonia and California’s Sierra Nevada. For films and colorized them.” They found that
trees to grow in stream- and lakebeds, the megadroughts in the Southwest were in-
Richard Stone is senior science editor at the droughts must have lasted for decades, and fluenced by three factors: an anomalously
Howard Hughes Medical Institute’s Tangled Bank at least one of these megadroughts seemed to warm North Atlantic Ocean, small global
Studios in Chevy Chase, Maryland. have hit both continents simultaneously. temperature rises driven by factors such as

Ongoing drought INFECTIOUS DISEASES
has dried up
a lagoon near
Santiago, Chile. Study pushes emergence of
measles back to antiquity
The virus may have entered the human population when
cities grew large enough to sustain outbreaks
By Kai Kupferschmidt, in Berlin century B.C.E., rather than in medieval
times, as previous research had suggested.
O
n 3 June 1912, a 2-year-old girl at the The work is technically brilliant, says
Charité University Hospital here died evolutionary biologist Mike Worobey of the
of pneumonia following a measles University of Arizona: “Just being able to
a brightening Sun, and, especially, La Niña. infection. The next day, doctors took get the measles virus out of these old, wet
The cold cousin of El Niño, La Niña can per- out her lungs, fixed them in forma- specimens. That sets the stage for all sorts
sist for years, deflecting rainstorms away lin, and added them to a collection of exciting work.” Monica Green, a historian
from their usual tracks. of anatomical specimens started by Rudolf of infectious diseases at Arizona State Uni-
In the new work, the team finds that La Virchow, the “father of pathology.” There versity, Tempe, calls the sequencing “very
Niña is almost the sole driver of the South they languished for more than 100 years— impressive” as well but says the study lacks
American megadroughts. And because La until Sebastien Calvignac-Spencer, an enough data points to “provide decisive
Niña affects conditions on both sides of the evolutionary biologist at the Robert Koch answers” about measles’ emergence. The
equator, it could plausibly trigger simultane- Institute, came across them in the basement authors agree. They hope sequences from
ous droughts in both hemispheres. of Berlin’s Museum of Medical History. antiquity, preserved in naturally mummi-
Jessica Tierney, a paleoclimatologist at Calvignac-Spencer and his team took a fied or frozen bodies, may one day do so.
the University of Arizona, says this mar- sample from the lungs, isolated RNA from Measles, which killed an estimated
riage of proxy records and models is a it, and subsequently pieced together what 142,000 people in 2017, is one of the most in-
powerful tool for understanding past cli- is the oldest known genome of the measles fectious human diseases. But when, where,
mates. “Nathan has really been leading the virus. Its sequence helped them shed light and how it became a human pathogen is
way on that,” she says. But she cautions on a much earlier period in measles’ history. still debated. The closest relative of the
that models don’t perfectly simulate the La In a study slated for submission to the pre- measles virus is one that causes rinderpest,
Niña cycle. It’s also uncertain whether the print submitted bioRxiv this week, the team a disease that affected cattle, deer, buffalo,
link between La Niña and distant droughts concludes that the virus may have entered and other even-toed ungulate species before
is a stable dynamic that lasts centuries or the human population as early as the fourth it was eradicated in 2011. Most researchers
might change over time. And the random- believe both viruses had a common ances-
ness of weather is always a factor: The cur- tor that infected cattle. “The challenge is
rent South American drought, for example, that … measles has left so few clear traces in
has endured through both El Niño and La historical disease descriptions,” Green says.
Niña conditions. Because measles spreads so fast and in-
Nor is it clear how the drought patterns fection confers lifelong immunity, scientists
will change as climate warms. A warming estimate it needs populations of 250,000 to
atmosphere alone seems certain to make a half-million people to avoid burning itself
megadroughts more frequent, especially in out. Historians believe that the largest cit-
the Southwest. But scientists remain divided ies reached that size around the fourth cen-
on how climate change will affect the El tury B.C.E. But when researchers in Japan
Niño-La Niña cycle. Models suggest El Niño used available genomes of the measles and
will dominate, but in the past few decades, La rinderpest viruses to build a phylogenetic
Niña has seemed to be more frequent. “There tree, enabling them to date the branches,
are still camps,” says Daniel Griffin, a paleo- they concluded in 2010 that measles didn’t
climatologist at the University of Minnesota. emerge until the 11th or 12th century C.E.
“I see people trying to hold their noses for The uncertainty stems in part from a
consensus statements.” surprising lack of historic sequences. Only
As bad as the drought in Chile is today, it three genomes from measles viruses occur-
PHOTO: KAI KUPFERSCHMIDT/SCIENCE
barely qualifies as a megadrought when com- ring before 1990 are known; the oldest is
pared with the medieval ones, which were one isolated in 1954 that was turned into the
longer and more severe. Clearly, there was first measles vaccine. So Calvignac-Spencer
something about that period that switched turned to the Berlin museum, whose shelves
off in recent centuries, Cook says. If that pat- are filled with thousands of tissues and or-
tern somehow came back, with greenhouse gans floating in formalin-filled glass cases,
warming amplifying it, Cook says, “then Scientists assembled a measles virus genome from like aquariums of human anatomy.
things could get quite catastrophic.” j the preserved lungs of a girl who died in 1912. Formalin fixes tissue by cross-linking

proteins and other large molecules, includ-

ing RNA, which the measles genome is
made of. To extract RNA from such samples,
scientists use techniques pioneered about
10 years ago by cancer researchers inter-
ested in formalin-fixed biopsies. “We put
them at 98° for 15 minutes and that breaks
the cross-links,” Calvignac-Spencer says.
This also breaks up RNA, but modern meth-
ods allow scientists to sequence the frag-
ments and piece them back together.
Calvignac-Spencer’s team drew up a new
phylogenetic tree using the 1912 genome
as well as a new one from 1960, pieced to-
gether from a sample in another collection,
and other available genomes. The result-
ing tree suggests the disease could have
jumped to humans as early as 345 B.C.E.—
right around the time human populations
reached the critical size.
The earlier date for measles’ emergence TRUMP ADMINISTRATION
also reflects the models that the team used
to analyze the viral sequences. When draw-
ing up a family tree using differences in ge-
nomes, researchers must estimate the speed
Computer scientist in line
at which viral genomes diverge. In the past,
their estimates were often too high, because
some deleterious mutations tend to disap-
to become next NSF director
pear over time. The new model accounts Sethuraman Panchanathan would succeed France Córdova
for this effect, called purifying selection. It
pushes back the divergence of measles and By Jeffrey Mervis One of those jobs is being a member of
rinderpest even without including the 1912 the National Science Board (NSB), NSF’s
T
genome. But the genome strengthens the he computer scientist whom Presi- presidentially appointed oversight body.
new timeline, Calvignac-Spencer says. dent Donald Trump picked last His 5-year stint has given him an insider’s
The researchers can’t rule out that the month as the next director of the look at the $8 billion agency, and a chance
measles virus first circulated in humans National Science Foundation (NSF) to work with Trump’s science adviser,
and then jumped to cattle, but that seems has followed the path taken by an Kelvin Droegemeier, who served on NSB to-
unlikely, says Albert Osterhaus of the Uni- untold number of foreign-born re- gether with Panchanathan for 2 years.
versity of Veterinary Medicine in Hanover, searchers by seeking greater opportunities Droegemeier, who hailed the president’s
Germany. For one, ungulate herds probably in the United States. If the Senate confirms 19 December 2019 announcement of the
reached the critical population size long be- him, as seems likely, 58-year-old, India-born pending nomination as “a win for science
fore humans did. And the closest relative of Sethuraman Panchanathan will become in the Trump administration,” has strength-
the two viruses, which is even older, causes not only the second NSF director of Asian ened ties between NSB and the White House
peste des petits ruminants, a sheep and American descent, but a living embodiment Office of Science and Technology Policy since
goat disease that probably crossed to cattle of how the international flow of talent has becoming OSTP director in January 2019.
more easily than to humans. helped fuel U.S. leadership in global science. Some observers credit Droegemeier with not
Similar studies have suggested that HIV Panchanathan, who goes by “Panch,” is only championing Panchanathan for the NSF
and other pathogens also took off in the executive vice president for research and post, but also persuading the White House
wake of major changes in the human popu- chief innovation officer at Arizona State to queue up a new director before France
lation structure, Worobey says. “It seems University (ASU) where he has worked Córdova finishes her 6-year term in March.
like changes in human ecology really did since 1997. A former chair of its computer Panchanathan’s interests dovetail with
coincide with the successful emergence of science and engineering department, he NSF priorities, says Kambhampati, who at-
these viruses.” founded the university’s school of comput- tended the same prestigious school, the In-
The ability to fish viral RNA out of very ing and informatics and created and leads dian Institute of Technology (IIT), Madras,
old samples has renewed interest in the its center for cognitive ubiquitous comput- that launched Panchanathan’s career. “He’s
Virchow collection, says Thomas Schnalke, ing, a technology that uses artificial intel- very strong on interdisciplinary and use-
PHOTO: ARIZONA STATE UNIVERSITY

head of the museum. “It came as a kind of ligence and machine learning to help those inspired basic research, as well as entrepre-
revolution for us that researchers are com- with disabilities. neurship and partnerships with industry.”
ing and saying: ‘Your samples are interest- “His star has kept rising,” says Subbarao Curiously, Panchanathan will be the sec-
ing for us again.’” Calvignac-Spencer has Kambhampati, an ASU computer scientist ond IIT Madras alum to lead NSF; fellow
already labeled additional specimens he’d and a longtime colleague and friend. “I alum Subra Suresh, a materials scientist,
like to study with orange stickers. “It’s a think he’s got three or four jobs. But he has directed the agency for 2.5 years during
treasure trove,” he says. “A window to the a crazy amount of energy, and he seems to the first term of former President Barack
past that we can open now.” j like going full speed all the time.” Obama. Suresh, who is 5 years older than

Sethuraman Panchanathan’s research has focused on U.S. BUDGET
building technologies to help those with disabilities.
Panchanathan, attended IIT Madras in the

mid-1970s, and he says its graduates—who
Congress again rejects Trump
are drawn from the most scientifically pre-
cocious of the country’s vast pool of young
people—“have done very well” in life.
cuts, smiles on science agencies
Open borders fostered his own career, 2020 spending bill increases research budgets, but next
says Suresh, who left NSF in 2013 to become
president of Carnegie Mellon University,
year will be tougher as lawmakers will have less to spend
and is now president of Nanyang Techno-
logical University. Leading NSF could put By Jeffrey Mervis Survey that the Trump administration had
Panchanathan on a similar leadership tra- targeted. And the bill includes the first fed-
L
jectory. In 2017, he was runner-up to become ast summer, U.S. congressional lead- eral funding for gun violence research in
president of the University of Arizona, and ers and the White House agreed to decades and language urging reductions in
Kambhampati says he has been considered raise federal spending over the next the use of animals in research.
for other top academic positions. “Panch 2 years by $125 billion above pro- However, this year’s totals for research
enjoys playing a leadership role within the jected levels. The deal effectively may be a high-water mark for Trump’s
scientific community,” Kambhampati says. removed the threat of large cuts to 4-year term. Unless the White House and
“He tends to have a big vision on issues.” research agencies that President Donald Democrats take the unlikely step of reopen-
Panchanathan declined to comment on his Trump had proposed in his 2020 budget ing the budget deal, Congress will have just
nomination. But in October 2019, he got the request. But those agencies didn’t learn $5 billion to add to agency budgets in 2021,
chance to appear on a national policy stage, precisely how much money they would get far less than the $44 billion extra it was
testifying at a Senate hearing on “ensuring until late last month, when Trump signed able to hand out this year. (The July 2019
America’s economic and strategic leader- a $1.37 trillion spending bill for the 2020 agreement erased draconian cuts sched-
ship” through research and innovation. fiscal year, which ends on 30 September. uled to go into effect in 2021.) So any ma-
Senators spared Panchanathan when In the short term, the news is good. Once jor spending increases next year will have
they pressed a panel of top scientists on again, the National Institutes of Health to come at the expense of cuts elsewhere.
the sensitive issue of combating threats to (NIH) did the best, with Congress overrid- The bad news came early for the National
U.S. research from foreign influences. But ing Trump’s proposed $5 billion cut and Science Foundation (NSF), which received
he weighed in when they asked about ways handing the agency a fifth straight annual a second straight year of modest growth.
to eliminate gender discrimination and increase of at least $2 billion. Science pro- The agency was already under siege from
remove barriers to entry for those groups grams at the Department of Energy and Trump, who wanted to cut its $8 billion
underrepresented in science. NASA made a similar rebound from a pro- budget by 12%. Legislators had fought back:
He cited a program at ASU that has at- posed double-digit cut to a healthy increase, A spending bill passed by the House of Rep-
tracted more undergraduate women into and DOE’s Advanced Research Projects resentatives in June 2019 would have grown
the mostly male discipline of engineering Agency-Energy escaped a proposed termi- NSF’s budget by $565 million, and in Oc-
CREDITS: (GRAPHIC) D. MALAKOFF/SCIENCE; (DATA) U.S. CONGRESS/WHITE HOUSE OFFICE OF MANAGEMENT AND BUDGET
by showing how it relates to the social sci- nation and instead will grow by a robust tober the Senate voted for a bump of $242
ences, where women are often a majority. 16%. Congress also strengthened climate million. But rather than splitting the differ-
The social sciences also play an important and environmental research programs at ence, Congress ended up giving NSF only a
role in his research. But he refused to of- the National Oceanic and Atmospheric $205 million increase, less than both marks
fer legislators a silver bullet for creating a Administration and the U.S. Geological and only 2.5% above last year.
more diverse scientific workforce. Under-
represented groups “need more role mod-
els,” he said. “And culture change is hard.” Reversal of fortune
Even those who don’t know him well Congress not only rejected deep budget cuts to major science agencies proposed by President Donald Trump
expect him to be a good fit for NSF. “He’s for 2020, but also provided sometimes hefty increases in a final spending deal.
got all the right qualifications,” says Arden
Bement, an emeritus professor at Purdue Trump request versus 2019 2020 1nal appropriations ($ increase)
University who was nominated by former 10
President George W. Bush and preceded ($2.6 billion) ($410 million) ($230 million) ($205 million)
Suresh as NSF director. “And I think he’ll be 5
a good partner with Kelvin. ... There’s going
Percentage change
to be a lot of uncertainty in the next year or 0

so,” Bement adds, drily, “and having those
two hands at the helm should help science –5
to stay on course.”
Córdova, nominated by Obama to head –10
NSF, says she’s “thrilled with Panch’s nom-
ination.” She deflected a question about –15
her plans after leaving the agency, saying,
“The most important thing for me is to –20
NIH DOE NASA NSF
finish strong.” j science science

Competition with other programs about $480 million per year on research PUBLIC HEALTH
funded within the same pot of money, to- comparing the benefits of medical treat-
gether with a cut to that pot that came late
in the process, is a major reason. Fund-
ing for the Executive Branch is spread
ments. Congress created PCORI under the
2010 Affordable Care Act, and patient ad-
vocacy and research organizations success-
Global polio
across 12 appropriations subcommittees.
The Commerce, Justice, and Science (CJS)
panel funds NSF, as well as NASA and the
fully pushed for PCORI’s renewal (Science,
6 December 2019, p. 1179). Congress also
gave the institute more leeway to consider
eradication
Commerce and Justice departments.
Senator Richard Shelby (R–AL), chair-
man of both the Senate’s CJS panel and the
costs in assessing competing treatments.
The bill also marks a turning point in
the long-running debate over federal fund-
falters in the
full committee, had already committed to
giving the Commerce Department’s Cen-
ing for research on gun violence. NIH and
the Centers for Disease Control and Pre-
final stretch
sus Bureau the additional $3.5 billion it vention will split $25 million, ending a 24-
needed to conduct the 2020 census. Shelby year drought for awards dedicated to such Vaccine-derived outbreaks
is also a big supporter of NASA, which research. Congress also ordered the direc- may force a change in
spends billions of dollars in his home state tors of NIH and CDC to report to Congress
of Alabama. That huge economic impact within 30 days on how it plans to support “endgame” strategy
translates into a vocal constituency. “ideologically and politically unbiased re-
In contrast, although NSF enjoys broad search projects.” Scientists “spent so many By Leslie Roberts
support among legislators, the agency “is years wondering whether we could even do
T
everyone’s second choice” this research. This is a clear he “endgame” in the decadeslong
when it comes to spending signal that not only can we campaign to eradicate polio suffered
priorities, says Joel Widder, “An increase for do it, but they want us to do major setbacks in 2019. While the ef-
whose Federal Science Part-
ners lobbies for universities
NSF that is it,” says Charles Branas, a
firearm violence epidemio-
fort lost ground in Afghanistan and
Pakistan, which recorded 116 cases of
and research institutions.
So, after a 6-month nego-
above inflation is logist at Columbia University.
Other provisions go be-
wild polio—four times the number in
2018—an especially alarming situation de-
tiation between House and a good thing and yond spending. At the urging veloped in Africa. In 12 countries, 196 chil-
Senate negotiators left the of animal rights advocates, dren were paralyzed not by the wild virus,
CJS panels with less money we should be Congress ordered the Depart- but by a strain derived from a live vaccine
than either had planned for,
NSF lost out.
thankful for it …” ment of Veterans Affairs to
devise a plan to reduce or end
that has regained its virulence and ability to
spread. Fighting these flare-ups will mean
Research advocates say Joel Parriot, dog, cat, and primate testing difficult decisions in the coming year.
it’s not such a bad deal. American Astronomical Society by 2025. Similarly, it directed The culprit in Africa is vaccine-derived
“An increase for NSF that the Food and Drug Adminis- polio virus type 2, and the fear is that it
is above inflation is a good thing and we tration to develop a strategy and timeline for will jump continents and reseed outbreaks
should be thankful for it given all the phasing out tests that involve primates and across the globe. A brand new vaccine is
other CJS pressures,” says Joel Parriott of retiring its research monkeys, and NIH to tell now being rushed through development to
the American Astronomical Society. “And Congress how it is moving to reduce the use quash type 2 outbreaks. Mass production
compared to the president’s request, this of primates in research. has already begun, even though the vaccine
is crazy good.” The budget also appears to settle, for is still in clinical trials; it could be rolled
One big winner within NSF’s new bud- now, a debate over whether the United out for emergency use as early as mid-2020.
get are institutions planning new or up- States should remain fully engaged in At the same time, the Global Polio Eradica-
graded “midscale research infrastructure,” ITER, the giant international experimen- tion Initiative (GPEI) is debating whether to
such as small telescopes or communica- tal fusion reactor under construction near combat the resurgent virus by re-enlisting a
tions networks, that cost from $6 million Cadarache in France. Some legislators have triple-whammy vaccine pulled from global
to $70 million. Scientists say NSF funding wanted DOE to withdraw from the project use in 2016. That would be a controversial
has failed to keep up with the demand for because of cost overruns and delays, and move, setting back the initiative several
these types of tools. Congress had scaled back U.S. contribu- years, as well as a potential public relations
For the first time in decades, NSF this year tions. But lawmakers approved an 83% in- disaster—an admission that the carefully
requested $30 million for midscale projects crease, to $242 million in 2020, including crafted endgame strategy has failed.
in its research account, as well as $45 million $85 million in cash and $157 million for “All options are on the table,” says viro-
in an account normally used to build more manufacturing parts. That funding essen- logist Mark Pallansch of the U.S. Centers
costly facilities. Congress liked the idea so tially puts the U.S. contribution back on for Disease Control and Prevention, one
much that it upped the amount in the large the track DOE envisioned in 2017. of the five partner organizations in GPEI.
facilities account to $65 million, bringing the The new year means a new budget cycle. “We are clearly in the most serious situation
total for midscale projects to $95 million. Trump will deliver his State of the Union we have been in with the program,” adds
Lawmakers used the appropriations address to Congress on 4 February and Roland Sutter, who recently stepped down
process to settle some policy matters. For shortly after deliver his 2021 spending plan as the director of polio research at the
instance, the final bill renews, for 10 years, to Congress, which will then have its say. j World Health Organization (WHO).
a government-created nonprofit organiza- The heart of the problem is the live oral
tion called the Patient-Centered Outcomes With reporting by Adrian Cho, David Grimm, polio vaccine (OPV), the workhorse of the
Research Institute (PCORI), which spends Jocelyn Kaiser, and Meredith Wadman. eradication program—the only polio vac-

still be a big benefit,” Wenger says, but the
program is hoping for more.
Sutter worries GPEI is “putting all of its
eggs into the nOPV basket.” The novel vac-
cine could quickly lose its genetic stability
if it exchanges key chunks of DNA with re-
lated viruses, he says. But how often these
critical “recombination events” occur won’t
be known until the vaccine is used in larger
populations. GPEI’s Independent Monitor-
ing Board noted recently that the program is
“rather starry-eyed” about nOPV2’s prospects.
If novel OPV2 doesn’t work or vaccine-
derived outbreaks spiral out of control
before it is ready, the program might have
little choice but to resurrect trivalent live
vaccine, which would reintroduce immu-
nity against type 2 in young children while
maintaining protection against serotypes 1
and 3. The vaccine might be used in cam-
paigns across Africa, reintroduced into rou-
tine immunization, or both.
The program is now struggling to de-
fine the “triggers” that would warrant this
Polio vaccinators in northwestern Nigeria, where the live-virus vaccine itself is causing new infections. move. Is it reestablishment of type 2 across
Africa? In Asia? The failure of nOPV2? The
cine powerful enough to stop viral circu- drop if supplies run critically low, despite depletion of the mOPV2 emergency stock-
lation. Given as two drops into a child’s what it calls “a relatively weak level of evi- pile? “It is actually a hard question. … It’s a
mouth, OPV for decades contained a mix dence” that the smaller dose is as effective.) public health judgment call,” Wenger says.
of three weakened polio viruses, one for Meanwhile, the risk of explosive outbreaks “People have different ideas on timing and
each of the three wild serotypes that have around the globe is ratcheting up, because triggers,” Zaffran adds. But officials need
long plagued humanity. All three serotypes millions of children born since the switch to decide soon whether to ramp up pro-
in the vaccine have the potential to revert have little or no immunity to type 2 virus. duction of OPV3 again, which could take
to more dangerous versions; that’s why the WHO’s Michel Zaffran, who leads GPEI, several years.
endgame strategy calls for deploying OPV says there’s room to make better use of Some experts fervently hope to avoid
in massive campaigns to eradicate the wild mOPV2 by detecting outbreaks sooner, get- reintroduction of the trivalent vaccine.
virus, then ending its use entirely. ting money and vaccines to countries ear- “It would be an enormous blow to the po-
Wild serotype 2 was last sighted in 1999, lier, and reaching more children. “There are lio program and to international public
so in 2016, as a first step in the endgame, things we can do even without a new tool,” health,” says Nicholas Grassly, a modeler
all 155 countries using OPV replaced the agrees Jay Wenger of the Bill & Melinda and epidemiologist at Imperial College
trivalent version with a bivalent one, lack- Gates Foundation, a partner in GPEI. London. Sutter, on the other hand, favors
ing the type 2 component. Announced with But hopes are pinned on a novel OPV reintroduction sooner rather than later.
great fanfare, “the switch” was billed as the (called nOPV2) that doesn’t revert so eas- Trivalent OPV “is the only thing we know
biggest vaccine rollout ever. Some type 2 ily. A Gates-funded research consortium has eradicated type 2 in the past and prob-
outbreaks would inevitably occur for sev- is developing two candidates, each with ably could eradicate it again,” he says. But
eral years, GPEI realized, but those would changes at multiple nucleotides to increase he agrees it would be a hard decision to
be fought, somewhat paradoxically, by rush- genetic stability. Small phase I clinical tri- communicate, given the huge global ef-
ing in essentially the same vaccine that als suggested both trigger an immune re- fort that went into persuading countries to
gave rise to them in the first place: a live, sponse and are safe and unlikely to regain switch to the bivalent vaccine in the first
monovalent vaccine targeted against type virulence. Phase II studies are underway in place. “How do we explain to the world
2 (mOPV2). If used in well-run campaigns, Belgium and Panama, but GPEI has already that we have to go backward, not forward?”
and only in outbreak regions, mOPV2 could started to manufacture one candidate and Sutter asks.
stop outbreaks without seeding new ones, hopes to have at least 100 million doses There’s a bigger issue, too. No vaccine
models suggested. available this summer. GPEI is also push- can stop polio if it doesn’t get into children’s
PHOTO: PIUS UTOMI EKPEI/AFP/GETTY IMAGES
It often has not turned out that way. In- ing for an Emergency Use Listing, a never- mouths, program leaders and their advisers
stead of fading away, the number of type 2 before-used WHO mechanism that would caution—and that has been a long-standing
outbreaks in Africa almost tripled from enable the program to deploy the vaccine problem anywhere the virus, vaccine-derived
2018 to 2019. Most of today’s outbreaks stem while it collects more data. or wild, still circulates. The polio eradica-
from mOPV2 responses to previous ones, It’s a risky strategy. The vaccine could tion program has been struggling with
and GPEI is burning through its emergency fail or be delayed, and it won’t solve all the complacency, fatigue, resistance, and poor
stockpile of mOPV2 faster than it can be re- problems. It won’t be better at stopping planning—all human issues that technology
plenished. (Based on a small study in Mo- outbreaks, just less likely to seed new ones. can’t fix. j
zambique, a WHO advisory panel recently How much less likely remains to be seen.
recommended halving the dose to one “Even if it is just 100 times safer, that will Leslie Roberts is a journalist in Washington, D.C.

NEWS
NEWS
FEATURES
FIGHTING
WORDS
Virologist Roberto Burioni
has become a celebrity
in Italy by sparring with
vaccine skeptics
By Douglas Starr
I
n May 2016, Roberto Burioni, a viro- sage simple: “The Earth is round, gasoline is journalist Alessandro Milan called Burioni’s
logist at Vita-Salute San Raffaele flammable, and vaccines are safe and effec- rebuttal to Red Ronnie “the 13 most beauti-
University in Milan, Italy, was asked tive,” he said. “All the rest are dangerous lies.” ful words heard on TV in the last year.”
to appear on a popular TV talk show “It went off like a bomb,” Burioni recalls. So began the unlikely media career of
to face off against two opponents of Emails from viewers poured into the show, Roberto Burioni. In just a few years, he
vaccines—a former DJ, Red Ronnie, with some questioning how the publicly has gone from being a respected but little-
PHOTO: TANIA/CONTRASTO/REDUX
and an actress and TV personality, funded TV network hosting it could allow known professor to a major media person-
Eleonora Brigliadori. The host gave such ill-informed personalities to speak ality and an internet savvy advocate for
most of the air time to the Italian about medicine. Burioni took up the theme science. In a country where the government
celebrities. Then, with just a few minutes on his Facebook page, asking how one has sometimes promoted dubious medi-
left, he turned to Burioni. branch of government could broadcast lies cine, such as unproven stem cell therapies,
Burioni realized he didn’t have time to about vaccines while its health agency pro- Burioni has become an outspoken advocate
make the usual arguments about statistics moted immunization. More than 5 million for scientific evidence on vaccines and other
and scientific uncertainty, so he kept his mes- people responded to his comments. Radio medical topics, and a harsh critic of pseudo-

NE WS
Roberto Burioni had never appeared on TV before more from disease cures than from vaccines,” than $200,000 from the government, on the
a 2016 appearance catapulted him to fame. he declared. “So if you don’t vaccinate your grounds that the measles vaccine distrib-
children, the pharmaceutical multinationals uted by the national health service caused
science. Nearly 480,000 people now follow will be sincerely grateful to you.” his condition. News of the decision spread
him on Facebook—an impressive number in That post and others hit a nerve. Within on the internet like a contagion, igniting
a country of 60 million. A web page he and weeks his social media followers grew from false beliefs and conspiracy theories. “The
colleagues established to provide general about 100 to six digits. A voice like Burioni’s— year 2012 was identified as the breaking
health information gets more than 100,000 that of an expert who sees no reason to point in the public’s confidence in vaccina-
visitors per month. mince words or suffer fools—was evidently tion in Italy,” researchers at the University
Burioni, with his shock of graying hair, what many Italians were looking for. of Pisa wrote in a study of the web’s effect
peaked eyebrows, and ironic smile, appears As in many Western nations, concern on vaccinations. An appeals court reversed
often on TV and at public events. His four about vaccines had surged in Italy in the late that judge’s decision in 2015, but the dam-
recent science books for popular audiences age was done.
have become best sellers. The Italian edi- Antivaccination sentiment infected not
tion of Forbes magazine named him one of “He’s the one scientist only the courts, but also entertainment and
Italy’s top five internet game changers, and
a former health minister nominated him for
who stood up and politics. In the late 1990s, for example, a
comedian named Beppe Grillo had become
Italy’s gold medal in public health.
Internet prominence brings trolls, and
said, ‘This is bullshit.’” famous in Italy for denouncing vaccination.
Ten years ago, he co-founded the Five Star
Burioni has been forced to worry about Guido Silvestri, Emory University Movement, a libertarian political party that
security. Some respected health research- became a dominant member of Italy’s ruling
ers and journalists have also been critical, 1990s after U.K. doctor Andrew Wakefield coalition and embraced antiscience positions.
saying his blunt, even abrasive manner published his now notorious study in The Vaccination rates, which had been climb-
inflames an already polarized conflict. But Lancet linking autism to the measles, ing since the 1990s, started to slide. Uptake
many public health experts credit him with mumps, and rubella (MMR) vaccine. The of the MMR vaccine declined from a peak
changing Italy’s debate about vaccination study was later shown to be fraudulent, and of nearly 94% in 2010 to just over 85% in
and elevating the profile of science there. Wakefield lost his medical license. But that 2015—one of the lowest rates in Europe,
“I think he’s had a major impact on the didn’t stop him from continuing to prosely- and well below the 95% needed for herd
public’s understanding on the topic of vac- tize against vaccines or prevent vaccine op- immunity. Almost in lockstep, the nation’s
cinations and science in general,” says Pier ponents from embracing his study. measles rate climbed to the second highest
Luigi Lopalco, who studies epidemiology The Italian government insisted vaccines in Europe, after Romania’s.
and public health at the University of Pisa. were safe, but a series of medical scandals
“He’s re-established the right of scientists to had damaged its credibility. The most re- THAT WAS THE SITUATION Burioni waded
speak directly to the people without having cent involved an experimental treatment into when he made his appearance on Ital-
a DJ or actor intervene.” called Stamina therapy, developed by an ian TV. He’d never been on TV before, and
entrepreneur who claimed to be able to re- when the network called he assumed he’d
BURIONI MIGHT seem an unlikely media per- generate nerves from stem cells and cure be speaking to other medical experts. He
sonality. He followed a rigorous academic conditions such as Parkinson’s disease and was shocked to find that he’d be sharing the
track: a medical degree in Rome; a Ph.D. muscular dystrophy. The inventor, Davide broadcast with two people who knew noth-
in microbiology in Geneva, and several Vannoni, received huge fees from desper- ing about vaccines. Hence his curt reply,
years of postdoctoral research at the U.S. ate patients, despite having produced no and the public’s enthusiastic response.
Centers for Disease Control and Preven- clinical studies or peer-reviewed papers. As he saw it, a door had opened and he
tion, the University of Pennsylvania, the Scientist denounced the treatment, and the had to walk through it. The public seemed
University of California, San Diego, and Italian Medicines Agency ruled it unsafe. hungry for straight talk from an expert,
the Scripps Research Institute. As a profes- But the Italian Senate, bowing to public and he obliged. “I realized that the lan-
sor in Milan, he develops monoclonal anti- opinion, permitted doctors and hospitals to guage of social media needs to be differ-
body therapies for herpes, hepatitis C, and administer it and funded a nearly $4 mil- ent than the language used in conferences,
other viral diseases—work that has led to lion clinical trial that was never completed. with colleagues or even with patients,” he
30 international patents. Stamina therapy was finally outlawed says, “so I tried to use not a single diffi-
He made his first foray into social media in 2014, and Vannoni received a 22-month cult word.” He turned instead to concrete,
in 2015, when a friend who had created a suspended sentence for fraud. (He died in nonmedical metaphors in his Facebook
Facebook group for mothers asked him to December 2019.) Later, it was revealed that and Twitter postings. “Does an aircraft en-
write an explainer about vaccines. Burioni, Italy’s current prime minister, attorney gineer take a vote among the passengers
already irked by Italy’s growing antivaccine Giuseppe Conte, won a court case in 2013 as to how many wheels to put on an air-
movement, agreed. “I felt it was my duty to that allowed Stamina therapy to be given plane?” he asked. “No—the engineer is the
do something as a doctor and a professor to a girl suffering from an incurable neuro- expert, he’s trained for this job, and it’s his
and as a father of an 8-year-old daughter.” He logical condition. The child later died. job to decide.”
worried that the push to resist vaccinations Courts in Italy have compounded the When one of the largest publishers in It-
could put her and her classmates at risk. problem. Local judges often lack scientific aly asked him to write a book on vaccines,
He posted a five-point rebuttal of popular expertise and are allowed pick their own Burioni cranked it out in 4 months. A night
vaccine misconceptions and conspiracy the- consultants on technical matters. In one owl, he wrote after his wife and daughter
ories—including the notion that drug compa- egregious case in 2012, a judge in Rimini went to bed. “I never realized that I was good
nies promote vaccination in order to increase relied on Wakefield’s discredited research at writing,” he says. “Here I was at 54 years
profits. “Pharmaceutical houses earn much to award the family of an autistic boy more old; how would you say it—a late bloomer?”

The result, Vaccines are not an Opinion:
Vaccinations Explained to Those Who Really
Don’t Want to Understand, was a novelty in
a country where scientists rarely communi-
cate in colloquial language. In it, Burioni ex-
plains how vaccines work, traces the history
of vaccination and of vaccine denialism,
and dismantles the denialists’ arguments.
He brings up tragic case histories of people
who died young for want of a vaccine and
skewers prominent vaccine opponents,
such as Jenny McCarthy, who declared on
Oprah that she learned her biology “from
the university of Google.” He hammers on
the theme that you can’t substitute opinion
for facts. The book’s popularity made him a
fixture on radio and TV. “The speed of light
can’t be decided by a show of hands,” he’d
say in talk show appearances.
In his later bestsellers, Burioni widened
his case against pseudoscience in Italy. “We
are a country forever wobbling between sci-
ence and superstition,” he says. He railed
against judges who ruled that unproven
treatments such as Stamina therapy had to
be given to patients who requested them,
often in publicly funded hospitals. He also
criticized the Italian public health system
for reimbursing patients for homeopathy—
pseudoscientific medical treatments that
use an extremely dilute concentration of a
substance that causes symptoms similar to
those of the disease.
Burioni says he wants to promote a respect
for expertise. “I know something about vac-
cines, viruses, and bacteria because I have
been studying them for a lifetime,” he writes
in Conspiracy of Dunces: Why Science Can’t
PHOTOS: (TOP TO BOTTOM) AUGUSTO CASASOLI/A3/CONTRASTO/REDUX; STEFANO MONTESI/CORBIS VIA GETTY IMAGES
be Democratic, his second book. “But I have
no idea how to bake a cake or wire a lamp, so
I go to a bakery or call an electrician.”
Trouble begins, he says, when electri-
cians, bakers, and other nonscientists feel
qualified to weigh in on vaccination. The
internet abets the problem, he says: Un-
filtered by editors, it levels the playing field
between experts and “dunces.” A mass me-
dia committed to presenting both sides of
every issue makes things worse. “I can’t sup-
port a world in which lies are given the same
dignity as the truth,” he heatedly remarked
on an interview show. “Enough already!
Enough!” The audience burst into applause.
Burioni’s books, postings, and media
presence have made him a celebrity sci-
entist in Italy akin to, say, Neil deGrasse
Tyson in the United States. Hardly a week
goes by without him appearing on TV
or in a glamour shot in a newspaper or
magazine. At a recent conference in Milan
about future technologies, the audience—
A 2017 law requiring childhood vaccinations triggered protests and threats in Italy. In the Senate, Roberto including many influential business leaders—
Burioni’s allies displayed a doctored photo that circulated online, showing his face grafted onto an image of swarmed him after his talk, asking for advice
Aldo Moro, an Italian prime minister murdered in 1978. In Rome, opponents of the law flooded the streets. and autographs.

NE WS | F E AT U R E S
“I feel a bit embarrassed by it all,” Burioni their children. That’s why scientists have bring up a list of antivaccination groups
says, seeming not at all embarrassed but examined vaccines with such care, she tells among top results. Now, the first hits in-
very much amused. (In fact he seems to rel- viewers. “People are much more educated clude Burioni’s websites and the vaccine
ish the publicity, having recently appeared than they used to be,” she says. “They want information sites created by the World
on an Italian comedy show in a version of to understand what we are doing to their Health Organization and Italy’s health au-
Name That Tune.) He’s proud that despite children. So you cannot approach them in a thorities. Similarly, radio and TV shows
all the media appearances he hasn’t missed paternalistic way.” have become more likely to book scientists
giving a lecture and continues to work full- Others appreciate Burioni’s no-holds- instead of actors and DJs to discuss vac-
speed in his lab. barred style. “I know he likes to make jokes cination, according to Lopalco and others
Yet it hasn’t all been fun. His time for fam- that sometimes can be seen as abrasive,” who monitor the media.
ily and leisure has suffered. More disturb- says Guido Silvestri, a longtime friend and a Recently, Burioni expanded his cam-
ing, at one point police had to stand watch pathologist at Emory University. “But what paign for science. In late 2018, he and half
over his house after someone threatened his kind of debate can you have with someone a dozen colleagues created Medical Facts,
daughter online—one of many death threats. who says that vaccines are a conspiracy to a web portal that posts news, advice, and
In 2018, Burioni and his family were vaca- kill children? He’s the one scientist who comments on a variety of health issues.
tioning at the beach town of Rimini when a stood up and said, ‘This is bullshit.’” In January 2019, he and Silvestri posted
vaccine opponent caught sight of him and a manifesto called “The Pact for Science,”
posted on the web. Within minutes an on- THINGS ARE IMPROVING. In the summer of supporting research and education and
line posse had weighed in with suggestions 2017, after the nadir in vaccination rates calling for the use of objective science in
on how to harass him; on the advice of police and the spike in preventable disease, the government decisions. It has attracted
he and his family went elsewhere.
He’s also been the subject of
more substantive criticism for Setback and recovery
his “Burioni blasts”—devastating After rising for years, measles vaccination rates in Italy fell until 2015 because of unfounded safety concerns, abetted by some
CREDITS: (GRAPHIC) N. DESAI/SCIENCE; (DATA) EPICENTRO; F. D’ANCONA ET AL., EURO SURVEILL. 24(26), (2019); INTEGRATED SURVEILLANCE OF MEASLES AND RUBELLA IN ITALY, 55, (2019)
replies to even the mildest op- government and court actions; measles cases spiked. Public education and a 2017 law have boosted rates since then.
position. When a Facebook fol-
lower said he was trying to figure
out the vaccine issue in his own 100 %
head, Burioni replied, “When
Vaccination rate
you go on social media you can 90

be reasonably sure that the inte-
rior of that head is as empty as a
tire tube.” He’s also been known 80
to ban even mild critics from
his Facebook page and discount
70
the work of science journalists
on the vaccine issue. “This is not
0
what you would call public out-
reach,” says Sergio Pistoi, a science 10000
Number of cases
writer and molecular biologist

in Tuscany.
Burioni’s broadsides polarize 5000
and oversimplify a complicated
discussion, adds Fabio Turone, 0
a science writer and director of 2000 2005 2010 2015 2018
the Center for Ethics in Science
and Journalism. Fewer than 1% of Italians government passed a law that set up pub- thousands of signatures from people as
are hard-core vaccine deniers, he says; but lic education programs and requires all diverse as the head of the left-leaning
about 15% are “vaccine hesitant” because schoolchildren to get 10 essential vaccina- democratic party and Grillo himself, one
they have concerns about vaccines or find tions before kindergarten. Even though the of Italy’s original vaccine deniers. In June
them hard to get. (In Italy, doctors at the law is loosely enforced, it’s working. Vacci- 2019, Burioni and colleagues created a
public health service—not the child’s pe- nation rates have rebounded, according to group related to the pact that will advocate
diatrician—administer vaccinations, and Italy’s National Institute of Health. After for science and help local judges gather
some health service centers closed after bottoming out at 85.3% in 2015, measles better information when adjudicating
the budget crash of 2008.) Such people are vaccination has risen to 94.1%, within strik- cases involving science and health.
best reached by persuasion, not mockery, ing distance of the 95% needed to prevent “The world has changed,” Burioni says,
Turone says. outbreaks when single cases pop up. The acknowledging both the good and “cata-
Roberta Villa, a journalist and doctor, law has proved so effective that France and strophic” effects of social media. “Science
calls for a more empathetic approach. Her Germany passed similar laws in 2019. needs to find a new voice—not the lan-
YouTube videos show her sitting at her Burioni and others don’t claim he is re- guage of scientific congresses, but a lan-
kitchen table, a cup of coffee in hand and sponsible for the turnaround, but public guage that’s understandable, passionate,
a drawing of Snoopy in the background, ex- health researchers say the “Burioni effect” and convincing.” j
plaining that as a mother of six she under- has altered public discourse. Four years
stands that all parents want to protect ago, googling “vaccines” in Italy would Douglas Starr is a journalist in Boston.

B O OKS et al .
INSIGHTS SCIENCE FICTION
Asimov at 100
From epic space operas
to rules for robots, the
prolific author’s literary
legacy endures
By James Gunn
G
ertrude Blugerman once asked of
her husband, Isaac Asimov, “What
will you say at the end of your life if
you have written one hundred books
but have missed living?” “Only one
hundred?” he replied. Asimov wrote
of this incident in Opus 100, his hundredth
book, published in 1969. He would write or
edit more than 500 books in his lifetime.
Asimov was brought to the United States
at the age of 3 from Petrovichi, a small town
in the still relatively new Soviet Union. His
birthday—2 January 1920—was one settled
on by his parents. (Records in Petrovichi
were unreliable, and he may have been
born as early as November 1919.) The fam-
ily settled in Brooklyn, New York, where his
father opened a candy store that also sold
newspapers and magazines. It was an en-
terprise in which all members of the family
participated. Perhaps not surprisingly, Isaac
was drawn to the magazines, particularly the
colorful science fiction publications.
Asimov was a gifted student with a capa-
cious memory who moved easily through
grade school and high school. In the first of
his autobiographies (he published two com-
prehensive volumes and, later, two supple-
mental tomes), he recounted how he would
acquire his textbooks for the semester, read
them on his walk home, and never open
them again. When I interviewed him for
the book I wrote about his life and work (1),
I asked if he ever forgot anything, and he said
that once he had been reciting the second
PHOTO: COURTESY OF THE RALPH MCQUARRIE ARCHIVES

verse of the U.S. national anthem and, for a
moment, could not think of how it started.
Asimov attended Seth Low Junior College,
a branch of Columbia University in Brooklyn,
graduating in 1939 from Columbia when
Seth Low closed in 1938. He majored in
chemistry after discovering in his freshman
year that to major in zoology, one would be
required to dissect cats.
This pensive android appeared on the cover

of Asimov’s 1990 collection Robot Visions.
0103Books.indd 20 12/20/19 5:32 PM

The year 1939 was also when Asimov’s mentioned Opus 100—and was, it turned should appear one night in a thousand years,
first science fiction story was published in out, only just getting started. Soon his books, how would men believe and adore; and
the magazine Astounding Science Fiction. written on topics ranging from the Bible to preserve for many generations the remem-
The sale of his stories paid for his college the human body, began appearing as often brance of the city of God which had been
expenses, including the master’s degree he as monthly, leading Harvard paleontologist shown! But every night come out these en-
would later earn in chemistry. (He was re- George C. Simpson to call him “one of our voys of beauty, and light the universe with
jected twice for medical school but would go natural wonders and national resources” (1). their admonishing smile.” And in an article
on to earn his doctorate, again in chemistry.) Asimov was a popular public speaker in Isaac Asimov’s Science Fiction Magazine
In 1942, Asimov began research as a chem- and a regular participant in science fiction (3), Asimov credited my words with helping
ist at the Philadelphia Navy Yard. Meanwhile, conventions, where, despite having earned him overcome his fears about writing a se-
he continued to sell stories but considered a formidable reputation as one of the “big quel to the Foundation trilogy. The resulting
himself a third-rate writer until his nov- three” science fiction authors of the era, he series (4) brought the Foundation trilogy and
elette “Nightfall” received a cover story in remained approachable. He participated his robot novels together into a single future
Astounding Science Fiction in 1941. The fol- in a variety of social organizations, includ- history and became his first bestseller.
lowing year, the first story of his Foundation ing Mensa; the Humanist Society (of which A case can be made that, like H. G. Wells,
series was published. During this period, he he was named the honorary president); the Asimov came along at the right time. (Wells
began the first of his robot stories, which Baker Street Irregulars, a society dedicated once commented that he made his writing
were published together in 1950 as I, Robot. to the appreciation of Sherlock Holmes; and debut in the 1890s, when the public was
In September 1945, Asimov was drafted the Trap-Door Spiders, a luncheon group looking for new writers.) But Asimov also
into the U.S. Army and served for 6 months that served as the inspiration for the fic- had a restless and productive mind. His
before being honorably discharged. He re- tional Black Widowers club in a series of early experience of reading, and then writ-
turned to Columbia, where he earned his mystery stories and novels he wrote. ing, science fiction gave his popular science
doctoral degree in 1948 before accept- writing a rare narrative model, while
ing his first (and only) academic ap- his fiction similarly benefited from his
pointment at Boston University. scientific training.
Science fiction received a boost from Some of Asimov’s critics complained
World War II. The number of magazines that his writing lacked style. He re-
publishing it increased in the aftermath sponded by asserting that he had a
of the war, and book publishers soon fol- style: clarity. But it also was true that
lowed. The subsequent Space Age and he was able to adopt new methods, par-
concerns evoked by Sputnik also led to ticularly in his later works.
publishing opportunities in the realm Asimov’s fiction was based on the
of science popularization. Asimov’s en- presumption that humanity would
gagement with the latter genre began solve its problems by thinking coolly
with a 1953 biology text coauthored with and logically. In his nonfiction writ-
two other faculty members. This would ing, he often grappled with the messier
be the beginning of a series for which realities of human nature. There are
the earnings—together with those of his no records of how many minds he in-
science fiction and other writing—soon fluenced with the latter, but his abil-
exceeded his university salary. Asimov brought drama and narrative to his nonfiction science ity to communicate difficult scientific
When Asimov was reprimanded for writing and often grounded his fiction in real scientific principles. ideas in simple language has not been
failing to conduct any research, he re- equaled since.
plied that he considered his writing his re- Kurt Vonnegut is reported to have once Asimov once told a friend that if he had a
search. Shortly afterward, he was terminated. asked Asimov how it felt to be the man who hang-up, it was his desire to write, saying that
He had previously been promoted to the rank knows everything, to which Asimov is said he wanted to die with his nose stuck between
of associate professor with tenure, and after a to have replied that he only knew how it felt two typewriter keys. It did not happen that
2-year battle it was determined that he would to have the reputation of omniscience. On way. His second wife and widow, Janet, con-
be allowed to stay on. He chose instead to another occasion, an editor reportedly en- firmed after his death that Asimov had con-
leave to focus on his writing. couraged Asimov to write an autobiography. tracted AIDS from a blood transfusion during
By the time I filmed him for my literature “But I’ve never done anything,” he protested. open-heart surgery a decade before. He had
PHOTO: MONDADORI PORTFOLIO/CONTRIBUTOR/GETTY IMAGES
of science fiction film series in 1972, Asimov He returned a year later with a thick manu- been persuaded by his doctors to keep this
was living in a high-rise off of Central Park script. When the editor failed to protest its information confidential, because of concerns
and was in the middle of divorcing his first length, Asimov left and returned with an- that it would deter people from undergoing
wife. “Science fiction writers and readers other manuscript, just as thick. “What would necessary surgery. He died from complica-
didn’t put a man on the moon,” he reflected you have written if you had done anything?” tions of the disease in 1992 at the age of 72.
during our interview, “but they created a cli- the editor is said to have replied. But his legacy, and his books, remain. j
mate of opinion in which the goal of putting Although he claimed no false modesty—
REF ERENCES AND NOTES
a man on the moon became acceptable.” “nor true modesty either” (2)—he insisted on
1. J. E. Gunn, Isaac Asimov: The Foundations of Science
Asimov had, by this time, already pub- acknowledging the role of others in his suc- Fiction (Oxford Univ. Press, 1982).
lished his hundredth book—the previously cess. He credited editor John W. Campbell 2. J. E. Gunn, Alternate Worlds: The Illustrated History of
with the invention of the three laws of robot- Science Fiction (A & W Visual Library, 1975).
3. I. Asimov, “The story behind the ‘Foundation’,” Isaac
ics and with introducing him to the Ralph Asimov’s Science Fiction Magazine, December 1982.
The reviewer is the founder of the Gunn Center for the Study
of Science Fiction, University of Kansas, Lawrence, KS 66045, Waldo Emerson quotation that inspired 4. I. Asimov, Foundation’s Edge (Doubleday, 1982).
USA. Email: jgunn@ku.edu Asimov to write “Nightfall”: “If the stars 10.1126/science.aba0303
0103Books.indd 21 12/20/19 5:32 PM

INSIGHTS
PERSPECTIVES
INFECTIOUS DISEASE
Uncovering the ART of antimalarial resistance

A key mechanism of resistance to the antimalarial drug artemisinin is identified
By Danushka Marapana and Alan F. Cowman However, the initial identification of ART this carboxyl-terminal Kelch-repeat region.
resistance in western Cambodia in 2008 Directed mutagenesis studies of both ART-
T
he identification of artemisinin (ART) followed by rapid spread throughout the naïve strains and clinical isolates have
in 1971 allowed treatment of malaria Greater Mekong Subregion of Southeast confirmed the causal role of PfKelch13 in
resistant to chloroquine, the prevail- Asia forewarned of a major issue for the mediating ART resistance (7, 8). Multiple
ing drug at the time, and provided malaria elimination agenda in that re- hypotheses implicate PfKelch13 as a re-
hope for a malaria-free world (1). To- gion (3). Additionally, the potential spread sponder to downstream effects of ART
day, malaria control efforts have been of ART resistance to sub-Saharan Africa activation, especially in up-regulation of
very successful, with 32% fewer deaths over would make elimination programs even pathways involved in the cellular stress
the past 8 years (2). However, the emergence more challenging. response and reduced protein translation
of resistance to ART and other antimalari- ART resistance manifests as enhanced in the presence of ART-induced stress (9–
als threatens to become a major problem in survival and delayed clearance of young 11). However, the possibility of PfKelch13
the continuing program to eliminate and ring-stage parasites after a concen- acting upstream as a conduit for ART ac-
eventually eradicate malaria (3). ART com- trated exposure to ART or its derivatives. tivation was not investigated prior to the
bination therapies (ACTs) are the Birnbaum et al. study.
current gold standard for the treat- Birnbaum et al. found that
ment and control of malaria, and Gateway for activation PfKelch13 and its interacting pro-
how parasites that cause malaria Malaria parasites ingest hemoglobin to produce heme from the teins are localized in vesicles close
in humans mediate this resistance host red blood cell using cytostomes. The interaction of heme in the to cytostomes, which are erythro-
is of intense interest for preventing parasite food vacuole with artemisinin (ART) causes activation of cyte-cytosol containing structures
the spread of drug resistance. On the drug and parasite killing. PfKelch13-containing vesicles regulate produced by the parasite. They
page 51 of this issue, Birnbaum et the uptake of hemoglobin by Plasmodium falciparum and thereby discovered that proteins in this
al. (4) answer this critical question affect the amount of ART activation through reduced heme availability. PfKelch13 compartment are re-
by identifying the molecular mecha- quired for the endocytic uptake of
nism of ART resistance in the most hemoglobin; however, PfKelch13 it-
lethal human malaria parasite, Plas- Cytostome self is essential for this process only
modium falciparum. in the young ring-stage parasites.
As malaria parasites infect and PfKelch13 By investigating the link between
grow within human erythrocytes vesicles PfKelch13, hemoglobin uptake, and
(red blood cells), they actively en- ART
ART activation, Birnbaum et al.
gulf and digest host hemoglobin Heme found that PfKelch13-inactivated
in a dedicated parasite-derived parasites and those carrying ART
food vacuole (see the figure). Hemoglobin resistance–conferring Pfkelch13
Hemoglobin is broken down into mutations display depleted con-
amino acids, which are used for the Nucleus centrations of the protein. These
Red blood Food
synthesis of parasite proteins, and cell vacuole PfKelch13-depleted parasites ex-
iron-bound heme, which is gradu- Plasmodium hibit decreased hemoglobin uptake
ally detoxified within hemozoin falciparum Endoplasmic as well as enhanced ring-stage sur-
crystals. Heme binding results in parasite reticulum vival during high-dose ART treat-
cleavage of the endoperoxide bond ment. Thus, PfKelch13 is a facilita-
of ART, enabling drug activation. Active Molecular genotyping of ART-resistant par- tor of ART activation, and Birnbaum et al.
ART produces free radicals and reactive asites uncovered single nonsynonymous propose an elegant model to merge a large
oxygen species that attack protein and mutations in a P. falciparum–specific gene body of prior research into a comprehen-
lipid molecules of the developing para- called Pfkelch13 (6). The PfKelch13 protein sive pathway of ART resistance.
site. ART treatment of intraerythrocytic contains three main functional regions: a In this model, mutations in the Kelch
GRAPHIC: KELLIE HOLOSKI/SCIENCE

Plasmodium parasites results in rapid kill- parasite-specific localization sequence, a propeller domains of PfKelch13 impair the
ing of all life cycle stages, including the BTB/POZ domain that typically facilitates ability of parasites to endocytose hemoglo-
young “ring” stage of infection, which is ubiquitin-mediated degradation, and a bin during the young ring stage of infection.
resistant to most antimalarial drugs (5). carboxyl-terminal Kelch propeller repeat Consequently, hemoglobin degradation is
region that is predicted to function as a reduced, and less heme becomes available
scaffold for protein-protein interactions for ART activation. Hemoglobin catabolism
Department of Medical Biology, Faculty of Medicine,
Dentistry and Health Sciences, University of Melbourne, (6). Nearly all clinically relevant ART- also provides the parasite with a source of
Parkville, VIC 3052, Australia. Email: cowman@wehi.edu.au resistance mutations are localized within amino acids for protein synthesis. Therefore,
0103Perspectives.indd 22 12/20/19 5:58 PM

it seems that resistance to ART carries a bio- PHYSICS
logical trade-off for the parasite: It grows at
a slower rate owing to the decreased avail-
ability of amino acids for protein translation.
Therefore, the unfolded protein response,
Majorana fermions go for a ride
which alleviates cellular stress caused by Evidence for propagating Majorana quasiparticles
accumulation of unfolded and misfolded
proteins, is up-regulated in parasites in an
is found in a topological superconductor
attempt to withstand the toxic effects of re-
active ART metabolites (10, 11). By Sumanta Tewari1 and Tudor D. Stanescu2 scribe the “regular” particles (for example,
How does PfKelch13 function in he- electrons and neutrons), whereas the nega-
E
moglobin endocytosis? A likely scenario nrico Fermi described Ettore Ma- tive-energy solutions—initially regarded by
is that the cysteine-containing Kelch do- jorana as having the mind of a ge- even Dirac as unphysical—are now known
mains, which allow protein-protein inter- nius. The Majorana fermion, born to describe their so-called “antiparticles”
actions, function as a molecular scaffold as a testimony to the truthfulness of (for example, positrons for electrons and
to engage PfKelch13 partner proteins to mathematical aesthetics, has recently antineutrons for neutrons). For every quan-
facilitate hemoglobin transport. Inhibitors returned to the center stage of mod- tum mechanical particle that obeys the
against eukaryotic Kelch proteins disrupt ern physics. These are particles that are Dirac equation, there are subatomic anti-
protein-protein interactions by chemi- also their own antiparticles. For decades, particles with physical properties that are
cally modifying the Kelch cysteine residues Majorana’s theory was considered a math- identical in some respects to those of the
(12). This could be an avenue to disrupt ematical curiosity that has little to do with corresponding particle (for example, the
PfKelch13-mediated hemoglobin digestion reality. However, they have now become same mass) and exactly opposite in some
and nutrient uptake by the parasite. A rela- the key concept associated with certain others (for example, baryon numbers and
tively unexplored facet of PfKelch13 is the types of quasiparticles in condensed-mat- opposite electric charges). Thus, the nega-
BTB/POZ domain (6). In other eukaryotes, ter systems (1). In the condensed-matter tively charged electron has an antielec-
this domain forms a Cullin-3–E3 ubiquitin context, they are not fundamental particles tron called a positron that is the positively
ligase complex and promotes the protea- like electrons or neutrinos but emerging charged counterpart with equal mass. The
somal degradation of multiple protein sub- excitations that we term quasiparticles. On neutron and the antineutron have the same
strates (13). However, P. falciparum does page 104 of this issue, Wang et al. (2) pro- mass but opposite baryon numbers.
not appear to encode a canonical homolog vide strong evidence for the observation of In the late 1930s, Ettore Majorana showed
of Cullin-3, and because PfKelch13 itself Majorana quasiparticles in an iron-based that the Dirac equation accepts a class of so-
is localized within a distinct subcompart- superconductor, FeSexTe1–x. lutions that describes particles that are iden-
ment in the parasite cytoplasm (14), it will The concept of Majorana fermions has its tical in every respect to their antiparticles.
be interesting to identify how PfKelch13 roots in a celebrated equation discovered in The original candidates for Majorana fermi-
ubiquitinates proteins. the late 1920s by physicist Paul Dirac. The ons—neutrons and neutrinos—appeared to
Pfkelch13 mutation is not the only mech- Dirac equation seamlessly brings together have distinct antiparticles. For neutrinos,
anism of ART resistance, because parasites quantum mechanics and the special theory the jury is still out, but the concept of Ma-
that display an increased ring-stage sur- of relativity and provides the quantum me- jorana fermions has become central to de-
vival phenotype but lack Pfkelch13 muta- chanical description of spin-half fermions, velopments in supersymmetry, dark matter,
tions have been reported (15). It would be such as electrons, protons, and neutrons. and, most recently, certain types of emerg-
intriguing to investigate whether these The equation has the property that if a solu- ing quasiparticles in condensed-matter sys-
parasites also show defects in hemoglobin tion exists with an energy +E, –E is also a tems. In condensed-matter systems, they
uptake. In addition to ART, ACTs include a solution. The positive-energy solutions de- can emerge in a special class of supercon-
partner drug that typically targets the he-
moglobin digestion mechanism of the par-
asite. Future formulations of ACTs might Inferring Majorana modes
be more effective by using partner drugs Indirect observations support the existence of propagating Majorana quasiparticles (QPs) in the presence
targeting other biochemical pathways. j of surface topological superconducting states (blue) emerging from the bulk states (red).
REFERENCES AND NOTES
1. Y. Tu, Nat. Med. 17, 1217 (2011). Surface Domain wall
2. “World malaria report 2018” (World Health
Organization, 2018); www.who.int/malaria/ G
publications/world-malaria-report-2018/en/. X
3. A. M. Dondorp et al., N. Engl. J. Med. 361, 455 (2009). Z
4. J. Birnbaum et al., Science 367, 51 (2020).
5. C. L. Hartwig et al., Biochem. Pharmacol. 77, 322 (2009). G
6. F. Ariey et al., Nature 505, 50 (2014). X
7. J. Straimer et al., Science 347, 428 (2015).
Bulk
8. M. Ghorbal et al., Nat. Biotechnol. 32, 819 (2014).
X G X X G X
GRAPHIC: C. BICKEL/SCIENCE
9. C. Dogovski et al., PLOS Biol. 13, e1002132 (2015).

10. S. Mok et al., Science 347, 431 (2015). Setting the stage Defning the directions Topological surface Majorana QPs
11. P. J. Shaw et al., BMC Genomics 16, 830 (2015). The iron-based The reciprocal lattice states Along the domain wall, a
12. A. Cuadrado et al., Nat. Rev. Drug Discov. 18, 295 (2019). topological superconduc- points of the bulk In the bulk, valence and linear set of allowed low-
13. L. Pintard, A. Willems, M. Peter, EMBO J. 23, 1681 (2004). tor has a domain wall (X, G, Z) and surface conducting bands (red) do energy states (orange)
14. J. Birnbaum et al., Nat. Methods 14, 450 (2017).
(black line) separating states (X, G) are shown. not touch. Surface valence would connect the gapped
15. A. Mukherjee et al., Malar. J. 16, 195 (2017).
two regions shifted by and conduction bands bands of the topological
10.1126/science.aba0445 one-half of a unit cell. (blue) touch at one point. superconductor.

INSIGHTS | P E R S P E C T I V E S
ductors called topological superconductors tified topological superconducting surface DEVELOPMENTAL BIOLOGY
as either quasiparticles localized at certain states with high-resolution angle-resolved
types of defects, such as vortices, or as de-
localized quasiparticles propagating along
the boundaries or the walls between differ-
photoemission spectroscopy (9) and ob-
served sharp zero-bias peaks inside vortex
cores (10). The authors’ scanning tunnel-
Building a
ent domains of superconductivity (see the
figure). In condensed-matter physics, the
Majorana quasiparticles are predicted to
ing microscopy studies show a flat (bias-
independent) differential conductance
along the domain wall. This signature is
carnivorous trap
obey fundamentally new particle statistics the hallmark of the linearly dispersing Ma- Experiments and
that generalize from the Fermi-Dirac statis-
tics governing identical particles with half-
jorana quasiparticles propagating along a
one-dimensional defect. The spatial distri-
computations reveal
integer spin. Experimental observations of bution of domain wall states as a function developmental origins of
sharp tunneling conductance peaks that
are consistent with the presence of Majo-
of energy is also consistent with the evo-
lution expected from the Majorana quasi-
cup-shaped leaves
rana quasiparticles in semiconductor-su- particle being localized along the domain
perconductor heterostructures (3–5) and wall at zero energy to become delocalized By Derek E. Moulton and Alain Goriely
ferromagnetic atomic chains deposited on as its energy approaches the gap edge.
V
superconductors (6) have been reported The authors also show a robust zero bias ariation, according to evolutionary bi-
in the past few years, although definitive anomaly observed in vortex cores below ologist Stephen Jay Gould, is “nature’s
evidence is still lacking (7). These observa- the superconducting transition tempera- only irreducible essence” (1). The vari-
tions have only been on static quasipar- ture Tc , which is a signature of companion ation and diversity of shapes in nature
ticles, otherwise called localized or bound Majorana quasiparticles localized at vortex is a central focus of both evolutionary
states. By contrast, evidence for Majorana cores consistent with previous studies (10). and developmental biologists. Uni-
quasiparticles that propa- The experiments from fied under the unlikely roof of “evolutionary
gate along the boundaries
or walls between different
“Experimental Wang et al. provide a com-
pelling case for topological
developmental biology,” the ultimate goal of
these scientists is to understand how varia-
domains in superconductors observations… superconductivity and propa- tion arises both through natural selection (on
has not been observed. gating Majorana quasipar- geological time scales) and during develop-
The challenge associated are consistent ticles in a class of iron-based ment (on embryological time scales). On page
with the experimental dem-
onstration of Majorana qua-
with the presence superconductors. This system
is attractive because the ob-
91 of this issue, Whitewoods et al. (2) present
a fascinating example of evolutionary devel-
siparticles stems, essentially,
from the complexity of the
of Majorana servations and theory sug-
gest Majorana behavior occur
opmental biology in a carnivorous plant.
The herbarium of our early school years
proposed host systems. These quasiparticles…” in a single material, without taught us that leaves come in many different
systems are multicomponent a heterostructure. The find- sizes, shapes, and textures that have evolved
heterostructures (3–6) predicted to behave ings open a new chapter in the field of by subtle gene rearrangements to solve vari-
as topological superconductors under con- iron-based superconductors and represent ous packing and arrangement problems (3).
trolled external conditions. By contrast, a large step in the quest for Majorana fer- Whereas many leaves are nearly flat, so as
Wang et al. exploit a key advantage of the mions in condensed-matter systems. For to present their best face to the Sun, others
iron-based superconductor FeSexTe1–x in practical applications, Majorana systems have been sculpted by evolution to function
that it consists of a single material that sup- in condensed-matter physics are attractive as mechanical devices; ropes, springs, spikes,
ports all the key ingredients necessary for for quantum computing because their fun- claws, spears, hooks, catapults, and traps are
Majorana physics. This requires a nontrivial damentally new particle statistics may be the medieval weapons that plants use in their
electronic band structure, superconductiv- useful for the development of fault-tolerant daily struggle with the environment. Perhaps
ity, and special types of domain walls that topological quantum computation (11, 12). most intriguing are carnivorous traps, the re-
can support propagating Majorana qua- This makes these iron-based systems im- venge of the plant kingdom and one of the
siparticles. The surface of this system is a portant both for fundamental science and lesser-known interests of Darwin (4).
topological superconductor with a special quantum technology. j The humped bladderwort (Utricularia
band structure, based on an argument put gibba), is an inconspicuous, easy-to-grow
forward by Fu and Kane (8). A domain wall aquatic plant found on all inhabited con-
1. N. Read, D. Green, Phys. Rev. B 61, 10267 (2000).
separating regions of the crystalline lattice 2. Z. Wang et al., Science 367, 104 (2020).
tinents. Yet, it has developed a nearly
shifted by half a unit-cell should support a 3. J. D. Sau, R. M. Lutchyn, S. Tewari, S. Das Sarma, Phys. spherical cage and a sophisticated release
pair of counterpropagating Majorana quasi- Rev. Lett. 104, 040502 (2010). mechanism that can swallow an unsuspect-
particles identifiable with a linear energy- 4. V. Mourik et al., Science 336, 1003 (2012). ing crustacean in a few milliseconds (5).
5. H. Zhang et al., Nature 556, 74 (2018).
momentum dispersion relation (8). How can a leaf develop into such an elegant
6. S. Nadj-Perge et al., Science 346, 602 (2014).
Wang et al. used a combination of pre- 7. C. Moore, C. Zeng, T. D. Stanescu, S. Tewari, Phys. Rev. B and complicated structure?
vious studies and experimental and theo- 98, 155314 (2018). With its small genome, the bladderwort
retical evidence to show that this scenario 8. L. Fu, C. L. Kane, Phys. Rev. Lett. 100, 096407 (2008). turns out to be an excellent model system. In
exists for FeSexTe1–x. Previous studies iden- 9. P. Zhang et al., Science 360, 182 (2018). these plants, the same branch supports both
10. D. Wang et al., Science 362, 333 (2018).
11. A. Kitaev, Phys. Uspekhi 44, 131 (2001).
needle-like leaves and bladder-shaped traps
1
12. C. Nayak, S. H. Simon, A. Stern, M. Freedman, S. Das
Department of Physics and Astronomy, Clemson Sarma, Rev. Mod. Phys. 80, 1083 (2008).
University, Clemson, SC 29634, USA. 2Department Oxford Centre for Industrial and Applied Mathematics
of Physics and Astronomy, West Virginia University, (OCIAM), Mathematical Institute, University of Oxford,
Morgantown, WV 26506, USA. Email: stewari@clemson.edu 10.1126/science.aaz6961 Oxford OX2 6GG, UK. Email: goriely@maths.ox.ac.uk

(see the figure). Both structures develop from by inducing expression of one of the genes in tion of cell division? The new model demon-
the same initial dome-shaped organ primor- an abnormal position on a leaflet. strates a logical sequence of differential gene
dia. The selection of one shape or the other Having established relevant gene expres- expression preceding growth polarity, which
occurs through the physical translation of a sion profiles, Whitewoods et al. turned to precedes cell-division orientation.
delicate variation in differential gene expres- a computational model for leaf morpho- By showing that spherical traps, conical
sion and morphogen production, which can genesis. Most computational work in plant needles, and flat leaves all can be generated
be probed experimentally and theoretically biology tends to model morphology by track- from the same initial tissue shape through
through computational modeling. ing cell growth and division (7). One of the small shifts in gene expression and growth
During its development, the leaf must singular features of the authors’ research differentials inspired by morphogen distribu-
solve a problem in geometry. There is a fun- was their study of growth deformations at tion, the new study opens exciting lines of re-
damental difference between a flat sheet the tissue level. In their model, each point search. For example, can these polarity fields
and a sphere, a fact that can be appreciated in the budding organ—treated as a three- be explicitly identified by measurements of
by trying to flatten an orange peel. The two dimensional continuum—was given differ- gene expression at the cellular level? How
states are geometrically incompatible, so ent growth rates in each of three selected are the polarity fields influenced by chemi-
transforming one into the other involves directions. These directions are linked to a cal and mechanical stimuli (12)? And how
either stretching or cutting. Because the in- polarity field obtained by the diffusion of a exactly does the developmental process form
a functioning trap? The trap mechanism in-
volves the slow build-up and rapid release of
Bladderwort beginnings mechanical energy (13), which is intimately
During development of the humped bladderwort, the same initial primordia (arrow) on a single branch linked to morphological changes during de-
can be transformed into either needle-like leaves (left, top circle) or carnivorous traps (left, bottom circle). velopment (14), but the connection has not
The key to shaping an organ is differential gene expression, which creates differential growth that is either yet been explored.
heterogeneous or anisotropic. More broadly, Whitewoods et al. offer key
insights into the competing pressures that
Primordia ultimately shape every living thing. Develop-
ment is inherently a physical process and is
Varied rates
thus the end result of physical forces subtly
of growth in manipulated by genetic clues. To understand
diferent such a process requires analysis across mul-
parts of the tiple scales as well as the integrated tools of
organ
mechanics, mathematics, and biology (15).
Heterogeneous growth
However, this multidisciplinary approach
tells only half the story of evolutionary de-
velopmental biology. On the species scale,
evolution is driven by forces that enable one
Entrance organism to successfully reproduce while
Varied rates
Trap of growth in another dies out. To properly connect evolu-
door diferent tionary and embryonic forces across vastly
directions different scales is to understand the very
Anisotropic growth nature of variation: “the hard reality” (1). j
1. S. J. Gould, Discover 6, 40 (1985).
tegrity of the leaf is preserved through de- morphogen. Their model is an adaptation of 2. C. D. Whitewoods et al., 367, 91 (2020).
velopment, the incompatible morphological the theory of morphoelasticity (8), which al- 3. D. Kierzkowski et al., Cell 177, 1405 (2019).
change from dome to needle or trap can oc- lows for continuous changes resulting from 4. C. Darwin, Q. J. Microsc. Sci. 1874, 185 (1874).
5. O. Berg, M. D. Brown, M. J. Schwaner, M. R. Hall, U.
cur only through differential growth that can both mechanics and growth. This theory has K. Müller, J. Exp. Zool. A Ecol. Integr. Physiol. 333, 29
be either anisotropic (varied rates of growth been used successfully in animal morpho- (2020).
in different directions) or heterogeneous genesis to describe the formation of a wide 6. E. E. Kuchen et al., Annu. Rev. Plant Biol. 62, 365 (2012).
7. P. Prusinkiewicz, A. Runions, New Phytol. 193, 549
(varied rates of growth in different parts of range of structures, from folds in the brain (2012).
the organ). For example, a small sphere of to seashell architecture (9, 10). 8. A. Goriely, “The Mathematics and Mechanics of
material may be deformed into an ellipsoid Notably, by simply varying the growth Biological Growth” (Springer, 2017).
9. D. Ambrosi et al., J. R. Soc. Interface 16, 20190233
or with a bulge on one side (see the figure). rates, the computational model showed how (2019).
The challenge, then, is to decipher the ge- the same dome can develop into either a 10. D. E. Moulton, A. Goriely, R. Chirat, Proc. Natl. Acad. Sci.
netic underpinnings of the necessary growth needle-like cone or a planar leaf shape. By U.S.A. 10.1073/pnas.1916520116 (2019).
11. K. Fukushima et al., Nat. Commun. 6, 6450 (2015).
differentials. Much is known about the effects making these growth rates nonuniform in
12. O. Hamant et al., Science 322, 1650 (2008).
of particular genes on the shape of leaves (6). space, Whitewoods et al. further demon- 13. O. Berg, K. Singh, M. R. Hall, M. J. Schwaner, U. K. Müller,
In developing leaves, key genes are expressed strated a simple mechanism for generating Integr. Comp. Biol. 2019, icz144 (2019).
differentially in zones on the adaxial (upper) cup-shaped traps and other features, such as 14. H. Hofhuis et al., Cell 166, 222 (2016).
GRAPHIC: N. DESAI/SCIENCE
15. V. Mirabet, P. Das, A. Boudaoud, O. Hamant, Annu. Rev.

versus abaxial (lower) surfaces. Whitewoods the ridges found in a related cousin, Sarrace- Plant Biol. 62, 365 (2011).
et al. revealed that these same genes are ex- nia purpurea (11). The model also provided
pressed differently in leaflets that form traps some insight into a chicken-and-egg prob- ACKNOWL EDGMENTS
versus ones that form needle-like leaves. This lem in development: Does the orientation of A.G. is supported by the Engineering and Physical Sciences
Research Council, grant EP/R020205/1.
crucial observation was confirmed by show- cell division generate growth anisotropy, or
ing that trap development can be inhibited does growth anisotropy generate the orienta- 10.1126/science.aba3797

FLUORESCENT PROTEINS in the transition state, the P-ring and I-ring

are nearly perpendicular (9). This geometry
Electrostatics affect the glow reduces the amount of p-p orbital overlap
between the P-ring and I-ring, which re-
sults in negative charge flowing back from
Chromophore twisting is probed with unnatural amino acids the I-ring to the P-ring. Alternatively, in
the S1 state, the P-ring can rotate to reach
another fluorescence off-state 2, and this
By Cheng Hu1, Xiaohong Liu2, Jiangyun Wang2 light produces the nonprotonated nonfluo- process causes more negative charge flows
rescent trans form. This switching enables from the P-ring to the I-ring. Electron-
T
he chromophores of fluorescent pro- applications in super-resolution imaging donating group substitution on Tyr63 can
teins (FPs) form through self-cata- (4) and optogenetics (5). In this protein, the lower the excited-state barrier for P-bond
lyzed posttranslational modifications chromophore is formed from Cys62, Tyr63, rotation and decreases FQY.
(1). In the original green FP (GFP) and Gly64. Investigation of the contribution Romei et al. show that the electric field
isolated from the jellyfish Aequorea of the electrostatics to FQY requires fine- exerted by the chromophore’s environment
victoria, Ser65, Tyr66, and Gly67 resi- tuning electrostatic parameters in the com- can either promote or hinder charge trans-
dues form the 4-(p-hydroxybenzylidene)- plex protein environment. Genetic code ex- fer, and thereby could effectively control
5-imidazolinone (HBI) chromophore that pansion (6) allows unnatural amino acids the choice of bond rotation and isomeri-
contains a phenolate ring (P-ring), an im- (UAAs) to be introduced into FPs. Yu et al. zation pathway after photoexcitation. For
idazoline ring (I-ring), and a monomethine (7) and others established methods for the optogenetic applications, photoinduced
bridge (1). The protein cage excludes wa- genetic incorporation of large numbers of rotation around a specific bond of the FP
ter that can quench fluorescence, but also Tyr analogs to modify the chromophore. chromophore or a retinal chromophore
enhances the fluorescence quantum yield Upon photon absorption, the chromo- triggers a distinct conformational change
(FQY) by restricting bond-twisting photo- phore enters the S1 state, in which negative of the protein, which results in specific
kinase activation (5) or ion conductance
(10). Engineering the electrostatic and ste-
Excited-state outcomes ric environment of the chromophore by
Romei et al. used genetic code expansion of a Tyr residue to change the H atom (denoted X) of a protein introducing charged, hydrogen-bonding,
chromophore to groups that withdraw electrons (such as Cl) or donate electrons (such as OCH3) to explore the or unnatural amino acids in the protein
effects of electrostatics on fluorescence (R1 and R2 are contacts to the protein). scaffold could lead to more precise control
of chromophore twisting and downstream
Carbon Hydrogen Nitrogen Oxygen R group X group signaling pathways, or create photoswitch-
able FPs with higher total photon number
Electron fow e– Electron fow e– e– Electron fow and faster on-off state switching for super-
Rotation Rotation
resolution imaging (4).
X Recent advances in time-resolved serial
P-bond I-bond
rotation rotation femtosecond crystallography with x-ray
R2 free electron lasers (XFELs) have visual-
U I-ring ized bond-rotation events in proteins (9).
R1
P-ring I-ring P-ring I-ring Transient absorption spectroscopy could
P-ring
probe how genetically encoded UAAs influ-
To of-state 2 On state (cis) To of-state 1 (trans) ence the I-bond rotation and P-bond rota-
Rotation around the P-ring Photoexcitation creates the cis Crossing the bond-rotation tion pathways. These insights could in turn
leads to an of-state and moves fuorescent S1 state. Charge fows barrier creates the trans of-state.
more charge to the I-ring. from the P-ring to the I-ring. Charge fows back to the P-ring.
inform new mutagenesis efforts to develop
improved FPs. j
isomerization of the HBI chromophore. charge flows from the P-ring to the I-ring
1. A. Acharya et al., Chem. Rev. 117, 758 (2017).
However, the protein could also improve (see the figure). Romei et al. found that 2. M. G. Romei, C.-Y. Lin, I. I. Mathews, S. G. Boxer, Science
FQY through electrostatic effects. As re- replacing Tyr63 with analogs bearing an 367, 76 (2020).
ported on page 76 of this issue, Romei et al. electron-withdrawing group required more 3. A. C. Stiel et al., Biochem. J. 402, 35 (2007).
4. Z. Fu et al., Nat. Methods 10.1038/s41592-019-0613-6
(2) studied the effect of introducing groups energy to transfer the electron, which blue- (2019).
that donate or withdraw chromophore shifted the absorption maximum, and sub- 5. X. X. Zhou, L. Z. Fan, P. Li, K. Shen, M. Z. Lin, Science 355,
electrons on the FQY of the photoswitch- stitution with an electron-donating group 836 (2017).
6. J. W. Chin, Nature 550, 53 (2017).
able FP Dronpa2 (3). red-shifted the absorption maximum. Both 7. Y. Yu, X. Liu, J. Wang, Acc. Chem. Res. 52, 557 (2019).
Mutations of residues near the chromo- substitutions decreased FQY and likely low- 8. C. Y. Lin, J. Both, K. Do, S. G. Boxer, Proc. Natl. Acad. Sci.
phore can fine-tune the bond-rotation en- ered the energy barrier for bond rotation. U.S.A. 114, E2146 (2017).
9. N. Coquelle et al., Nat. Chem. 10, 31 (2018).
ergy barrier to create photoswitchable vari- The next challenge was to decipher the
GRAPHIC: A. KITTERMAN/SCIENCE
10. K. K. Yang, Z. Wu, F. H. Arnold, Nat. Methods 16, 687
ants. In Dronpa2, green light produces the mechanism of how bond twisting is modu- (2019).
protonated fluorescent cis form, and blue lated by electrostatics (8). In the transfor-
ACKNOWL EDGMENTS
mation from the S1 state to the fluorescence
1
Institute of Synthetic Biology, Shenzhen Institutes of Supported by the National Science Foundation of China
off-state 1, a cis-trans isomerization of the (21750003, 21837005, 21890743, and U1632133) and the
Advanced Technology, Chinese Academy of Sciences, double bond connecting the P-ring and Chinese Academy of Sciences (QYZDB-SSW-SMC032).
Shenzhen, China. 2Institute of Biophysics, Chinese
Academy of Sciences, Chaoyang District, Beijing, China. I-ring is required. Previous studies of the
Email: jwang@ibp.ac.cn photoswitchable FP rsEGFP2 showed that 10.1126/science.aba0571

Research is needed to understand the
projected increase in cancer incidence
and mortality in sub-Saharan Africa.
in the tumor suppressor gene TP53, muta-

tions in GATA binding protein 3 (GATA3),
and greater genomic mutational burden (in-
dicating aggressive biology), compared with
breast tumors from African Americans or
Caucasians (4). This suggests different tumor
etiologies by race or geography, perhaps re-
flecting particular environmental exposures
to carcinogens, and may provide knowledge
about the spectrum of breast cancer molec-
ular phenotypes that may be composed of
distinct molecular subtypes and represent
GLOBAL HEALTH different frequencies of known subtypes in
non-Africans. Knowledge of these differences
Cancer in sub-Saharan Africa can lead to optimized monitoring and treat-

ment across all populations.
Early age at diagnosis is a hallmark of
Knowledge of cancer in Africa brings needed cancer in SSA, which has been proposed to
reflect a higher rate of hereditary cancer in
diversity to improve health worldwide SSA. This is supported by the observation
that although 29% of SSA women and 33% of
By Timothy R. Rebbeck in SSA launched the research that identified Caucasian women are between ages 25 and
Epstein-Barr virus as the causal agent of 49, 58% of SSA women are diagnosed with
C
ancer is an increasing global pub- Burkitt's lymphoma (2), thus providing some breast cancer before they are 50 years old,
lic health burden. This is especially of the earliest knowledge about the infectious compared with 21% of Caucasian women (5).
the case in sub-Saharan Africa (SSA); and molecular etiology of cancer. Similarly, Breast tumors in SSA Black cases are twice as
high rates of cancer—particularly of the high prevalence of HIV infection and the likely as SSA Caucasian cases to be the triple-
the prostate, breast, and cervix—char- attendant large number of HIV-associated negative subtype, meaning that the estrogen,
acterize cancer in most countries in cancers in SSA have enabled research that progesterone, and HER2 receptors are not
SSA. The number of these cancers in SSA is has led to understanding of the causes and expressed by the breast cancer cells, making
predicted to more than double in the next treatment of cancers exacerbated by HIV them highly aggressive and difficult to treat
20 years (1). Both the explanations for these infection, including cervical cancer, Kaposi's (6). These molecular features are hallmarks
increasing rates and the solutions to address sarcoma, and non-Hodgkin's lymphoma (3). of hereditary cancers; genetic testing for
this cancer epidemic require SSA-specific Patients with cancer in SSA are often di- pathogenic sequence variants in the tumor
data and approaches. The histopathologic agnosed when their disease is in advanced suppressor genes breast cancer 1 (BRCA1)
and demographic features of these tumors stages. This is in part a consequence of inad- and BRCA2 in a series of Nigerian breast can-
differ from those in high-income countries equate resources for cancer prevention and cer cases suggested that the rate of cancers
(HICs). Basic knowledge of the epidemiology, early detection. Delayed diagnosis coupled in women with BRCA1 or BRCA2 germline
clinical features, and molecular characteris- with inadequate treatment options is a ma- pathogenic sequence variants is higher than
tics of cancers in SSA is needed to build pre- jor reason for the continent’s cancer mortal- in Caucasian populations (7). Although large
vention and treatment tools that will address ity rates, which are 1.5- to 4-fold higher than population studies of hereditary cancer in
the future cancer burden. The distinct dis- in HICs for leading cancers (1). Studies that SSA do not yet exist, BRCA1 and BRCA2 mu-
tribution and determinants of cancer in SSA focus on features of late-stage and aggressive tations of SSA origin are also found in Afri-
provide an opportunity to generate knowl- disease will be required to better understand can Americans, and the type of mutations in
edge about cancer risk factors, genomics, and how to manage such disease states. This re- BRCA1 and BRCA2 differs between SSA and
opportunities for prevention and treatment search is required to develop and implement non-SSA populations (8).
globally, not only in Africa. early detection and treatment modalities that Generation of knowledge about cancer in
The most frequent cancers in African coun- can be implemented in low-resource settings SSA will lead to both improved cancer pre-
tries include prostate, lung, liver, leukemia, such as SSA, where availability of laboratory vention and treatment. Protocols and net-
non-Hodgkin’s lymphoma, and Kaposi's sar- or imaging technologies is limited. works are being formed to study cancer in
coma in men and breast and cervical cancer Although molecular and other biologi- SSA patients and translate this knowledge to
PHOTO: STRINGER/REUTERS/NEWSCOM
in women (see the figure). Distinct risk fac- cal data that address cancer etiology and health care in SSA (9). These protocols may
tors in SSA contribute to the understanding progression in SSA are limited, emerging begin with those used in HICs but must be op-
of cancer etiology in ways that may not be as evidence suggests that distinct tumor histo- timized to low-resource settings, which have
easily studied in HICs. For example, the ob- pathology, tumor subtypes, and molecular limited access to equipment, trained person-
servation of high rates of Burkitt's lymphoma signatures exist in SSA. For example, Nige- nel, and therapies. Although the availability
rian breast cancer cases were defined by in- of basic research infrastructure is more lim-
creased mutational signature associated with ited in SSA than in HICs, there are numerous
Dana-Farber Cancer Institute and Harvard T. H. Chan
School of Public Health, Boston, MA, USA. deficiency of the homologous recombination basic science and translational research insti-
Email: timothy_rebbeck@dfci.harvard.edu DNA repair pathway, pervasive mutations tutes in SSA, including those that can address

cancer genomics and molecular biology. tions been included in the original study co- (12). Selection against a prostate cancer sus-
Pan-African initiatives, such as the Human hort. Similarly, it has been observed that use ceptibility locus on chromosome 2q37 has led
Heredity and Health in Africa (H3Africa) ini- of artificial intelligence and other “big data” to lower frequencies of this risk allele in Cau-
tiative, have developed common protocols, approaches in diagnostic and therapeutic casians and higher frequencies in Africans
data systems, and collaborative networks applications will be suboptimal if these ap- and African Americans (12). KhoeSan (an in-
that are developing the infrastructure and proaches are trained and validated in nondi- digenous southern African people) ethnicity
resources required to address cancer needs verse populations. has also been associated with prostate cancer
in SSA. Africa-based organizations that foster A limited scope of population data may risk (13). These data from African popula-
cancer research, translation, infrastructure, perpetuate or exacerbate existing biases in tions may help elucidate the causes of cancer
and training also exist, including the African the information needed to generate effec- disparities, particularly in African Americans.
Academy of Sciences and the African Organi- tive clinical and public health interventions Diagnostic, monitoring, and treatment
zation for Research and Training in Cancer. (11). Thus, it is not only the underrepresented technologies developed out of necessity in
The World Economic Forum has initiated a population that will benefit from increased low-resource settings provide an opportunity
for cost-efficient and accessible
technologies that can be imple-
Cancer in Africa compared with high-income countries mented in HICs. In particular,
The occurrence of the common types of cancer in sub-Saharan Africa (SSA) differs from that in high-income countries (HICs). The development of point-of-care
newly diagnosed cancer cases in 2018, as a percentage of the total population, are shown for SSA versus HICs. It is estimated that technologies that provide rapid
incidence of the most common cancers will approximately double between 2018 and 2040 in SSA. Data were obtained from (15). and accurate cancer diagnosis
or treatment in SSA could be ap-
Kaposi’s plied in community settings to
Bladder Breast Cervix Uterus Esophagus sarcoma Liver address disparities in access to
cancer services in HICs. When
1.2%
Female SSA 34.2% 25.8% 23.3% 2.1% 2.3% 2.4% 2.8%
technologies, including data-
Female HICs 2.1% 24.5% 2.1% 5.2% 0.6% <0.1% 1.5% derived models of risk or prog-
nosis, are not developed so that
Male SSA 2.9% 4.2% 6.2% 7.2% they can be applied in diverse
settings and populations, health
Male HICs 6.0% 1.8% 0.1% 2.9%
disparities can be created or
increased (14). Development of
Non-Hodgkin's technologies that can be used
Lung lymphoma Ovary Pancreas Prostate Stomach Thyroid
in SSA could provide knowledge
Female SSA 1.6% 3.1% 3.4% 1.1% 2.4% 1.9% about optimal implementation
of these technologies world-
Female HICs 9.5% 3.1% 2.7% 3.2% 2.9% 4.7%
wide. Advances in science and
technology and the concomitant
Male SSA 4.6% 5.6% 1.8% 21.5% 4.1% 0.9%
increase in capacity for health
Male HICs 12.0% 3.2% 2.8% 18.9% 4.5% 1.2% improvements will increase di-
versity as a means to remove
bias from cancer genomics stud-
“Leapfrogging with Precision Medicine” ini- research representation but all populations. ies and improve the quality of cancer pre-
tiative that will advance the use of genetics There is growing evidence that genetic risk vention and treatment globally. j
and genomics in cancer prevention and treat- prediction models developed in Cauca-
ment. These and many other institutions and sian populations may not be appropriate in REF ERENCES AND NOTES
activities have the potential to develop the African-descent populations. These results 1. D. Parkin et al., Eds., Cancer in Sub-Saharan Africa (IARC
Scientific Publication no. 167, IARC, 2018).
knowledge and sustainable resources needed suggest that ancestry-specific models may 2. D. Burkitt, Br. J. Surg. 46, 218 (1958).
to address the cancer burden in SSA. be required to optimally predict cancer 3. L. Chinula et al., Curr. Opin. HIV AIDS 12, 89 (2017).
In addition to benefiting the treatment of risk. Such models may use genetic variants 4. J. J. Pitt et al., Nat. Commun. 9, 4181 (2018).
5. Population Division, United Nations Department of
cancer patients in SSA, the knowledge gained defined from the discovery of relevant race- Economic and Social Affairs, 2019 Revision of World
from research on the continent will be inform- specific disease-associated variants and/or Population Prospects; https://population.un.org/wpp.
ative for cancer globally. Evidence suggests use ancestry-specific genomic markers to de- 6. V. A. McCormack et al., Breast Cancer Res. 15, R84
(2013).
that diversity in study populations will im- fine ancestry rather than self-identified race 7. Y. Zheng et al., J. Clin. Oncol. 36, 2820 (2018).
prove the ability to generate and generalize or ethnicity. 8. T. R. Rebbeck et al., Hum. Mutat. 39, 593 (2018).
knowledge that can be applied to cancer and Because of the underlying genetic and ge- 9. C. Andrews et al., J. Glob. Oncol. 4, 1 (2018).
10. A. K. Manrai et al., N. Engl. J. Med. 375, 655 (2016).
other diseases. For example, rare mutations nomic relationships between Africans and
11. G. Kuhnen, A. Rebhan (2019); www.advisory.com/
judged to be pathogenic on the basis of Cau- members of the African diaspora (primarily daily-briefing/2019/02/04/ai.
casian mutation frequency data were later in North America and Europe), knowledge 12. J. Lachance et al., Cancer Res. 78, 2432 (2018).
found to occur commonly in African Ameri- gained from research in SSA can be used to 13. D. C. Petersen et al., BMC Med. Genomics 12, 82 (2019).
14. D. Weiss et al., PLOS ONE 13, e0195447 (2018).
can populations (10). Knowledge of mutation address health disparities that are prevalent
GRAPHIC: X. LIU/SCIENCE
15. The Global Cancer Observatory; https://gco.iarc.fr.

frequency subsequently led to the opposite in members of the African diaspora. West
conclusion that these variants were non- African genomic ancestry (the ancestral ori- ACKNOWL EDGMENTS
pathogenic. This misclassification of patho- gin of most African Americans) has been re- The author is supported by National Institutes of Health grant
U01-CA184734.
genicity led to molecular misdiagnoses that ported to confer the highest genomic risk for
could have been avoided had diverse popula- prostate cancer of any population worldwide 10.1126/science.aay4743

CELL BIOLOGY
Gene expression regulated by RNA stability

The factor responsible for autoregulation of tubulin RNA stability is identified
By Ofer Shoshani and Don W. Cleveland mosomes. By inactivating both maternal and tion that the tubulin concentration directly
paternal copies of the gene encoding TTC5, participates in the ribosomal complex with
O
ne of the first discoveries of gene Lin et al. demonstrate that tubulin autoregu- TTC5 and the MREI peptide. Instead, the
expression mediated by controlling lation is essential for maintaining faithful findings of Lin et al. support a counterintui-
messenger RNA (mRNA) stability is chromosome segregation, with a modest in- tive model in which cells ordinarily contain a
autoregulation of tubulin synthesis. crease in chromosome segregation errors in cytosolic factor that prevents TTC5 binding
In this regulatory process, the con- the absence of TTC5. Errors in chromosome to MREI-ribosome complexes and that this
centration of tubulin subunits modu- inheritance are key drivers of tumorigenesis, inhibitor is inactivated when the tubulin di-
lates the stability of the mRNAs from which so maintenance of the genome is important mer concentration increases (see the figure).
they are translated (1, 2). In the 1980s it was (10). However, Lin et al. have determined The study of Lin et al. is a major step in
found that only translated tubulin mRNAs that cells with inactivated tubulin autoregu- deciphering a regulatory pathway for con-
are autoregulated (3) and that translation lation are viable and can continue to survive trolling expression of an important cellular
had to continue through at least 41 amino and duplicate. This is unexpected because product (tubulin) through cotranslationally
acids (4). This is enough for the nascent tu- disruption of autoregulation would be pre- mediated mRNA instability. It should be
bulin polypeptide to emerge from the ribo- dicted to yield runaway tubulin synthesis, noted, however, that important steps in the
some (5). Later work established that when which in turn should have severely disrupted autoregulatory pathway remain unidentified,
the tubulin subunit pool is high, the first microtubule assembly dynamics and func- including (i) the factor that inhibits TTC5
four amino acids (Met-Arg-Glu-Ile, MREI) tion. Perhaps additional factors are involved, binding activity when tubulin levels are nor-
emerging during nascent tubulin translation the activities of which might compensate, to mal, (ii) the newly proposed autoregulatory
serve as a regulatory tag. Recognition of this some extent, for the lack of TTC5 activity. signal generated by an increased pool of tu-
tag promoted the degradation of the trans- The new work also casts doubt on the no- bulin subunits, (iii) the nuclease(s) that medi-
lating tubulin mRNA (4, 6–8). More than 30 ate tubulin mRNA degradation, and (iv) the
years later, on page 100 of this issue, Lin et activation of those nucleases once TTC5 rec-
al. (9) identify tetratricopeptide repeat pro- Autoregulation of tubulin RNA ognizes the nascent tubulin peptide.
tein 5 (TTC5) as the regulator that binds to Tubulin messenger RNA (mRNA) instability is mediated This may only be the tip of the iceberg for
nascent tubulin polypeptides. by cotranslational binding of tetratricopeptide repeat cotranslational control of gene expression.
With the exception of a high-resolution protein 5 (TTC5) to the tubulin amino-terminal Work in yeast suggests that mRNA decay
confirmation that an elevated pool of tu- tetrapeptide MREI and activation of one or more through cotranslational regulation is wide-
bulin subunits selectively represses tubulin ribosome-associated nucleases. spread (11) and that it involves a 59-to-39
synthesis (8), no progress toward under- RNA exoribonuclease 1 (xrn1) (12). The TTC5
standing the autoregulation of tubulin ex- a/b-tubulin Unknown structure bound to a ribosome, as beautifully
I E R M T
TTC5
pression was made since 1988. The most concentration
ncent determined by Lin et al., provides clues that
attractive model for how the pool size of inhibitor
inh
inhibit might allow the identification of additional
tubulin subunits could trigger rapid mRNA regulators (perhaps containing the same tet-
NORMAL
degradation to suppress new tubulin syn- ratricopeptide repeats found in TTC5) that
thesis was that it was the tubulin a/b dimer, TTC
TTC5
T TC55 can bind the ribosome exit tunnel and simul-
the unit that assembles into microtubules, Transfer
Tra
a taneously recognize targets encoded by dif-
RNA
that cotranslationally bound to the MREI Ribosome ferent mRNAs. j
tetrapeptide. Lin et al. disprove this model. AAA
They use mass spectrometry and in vitro 5' Tubulin mRNA
1. T. J. Yen, D. A. Gay, J. S. Pachter, D. W. Cleveland, Mol. Cell.
translation of an mRNA encoding the first Biol. 8, 1224 (1988).
94 amino acids of b-tubulin to identify 2. D. W. Cleveland, M. A. Lopata, P. Sherline, M. W.
Kirschner, Cell 25, 537 (1981).
TTC5 as the protein that recognizes the na- Tubulin autoregulation activated 3. D. W. Cleveland, M. F. Pittenger, J. R. Feramisco, Nature
scent b-tubulin MREI tetrapeptide in com- 305, 738 (1983).
plex with the large ribosomal subunit. 4. D. A. Gay, T. J. Yen, J. T. Lau, D. W. Cleveland, Cell 50, 671
(1987).
a- and b-tubulin form a heterodimer that 5. J. S. Pachter, T. J. Yen, D. W. Cleveland, Cell 51, 283 (1987).
Unknown
Unk
Un 6. T. J. Yen, P. S. Machlin, D. W. Cleveland, Nature 334, 580
serves as the building block for microtubule TTC inhibitor
TTC5 (1988).
polymers, the tracks along which cargoes HIGH is inactivated
ina 7. A. Yonath, K. R. Leonard, H. G. Wittmann, Science 236,
are moved by dynein and kinesin family mo- 813 (1987).
GRAPHIC: V. ALTOUNIAN/SCIENCE
tors. During cell duplication, microtubule- 8. I. Gasic, S. A. Boswell, T. J. Mitchison, PLOS Biol. 17,
e3000225 (2019).
directed trafficking is essential for delivery to 9. Z. Lin et al., Science 367, 100 (2020).
each daughter cell of a complete set of chro- Nuclease 10. U. Ben-David, A. Amon, Nat. Rev. Genet. 21, 44 (2020).
11. V. Pelechano, W. Wei, L. M. Steinmetz, Cell 161, 1400
5' (2015).
12. W. Hu, T. J. Sweet, S. Chamnongpol, K. E. Baker, J. Coller,
Ludwig Institute for Cancer Research and Department of Nature 461, 225 (2009).
Cellular and Molecular Medicine, University of California, San Tubulin
Diego, La Jolla, CA 92093, USA. Email: dcleveland@ucsd.edu mRNA degraded 10.1126/science.aba0713

INSIGHTS
cama Desert, destroying fragile ecosystems,

P OLICY FORUM and converting meadows and lagoons into
salt flats. The extraction, crushing, refining,
ENERGY and processing of cadmium, a by-product of
zinc mining, into compounds for recharge-
Sustainable minerals and able nickel cadmium batteries and thin-film

photovoltaic modules that use cadmium tel-
luride (CdTe) or cadmium sulfide semicon-
metals for a low-carbon future ductors can pose risks such as groundwater
or food contamination or worker exposure to
hazardous chemicals, especially in the supply
Policy coordination is needed for global supply chains chains where elemental cadmium exposures
are greatest. REEs, such as neodymium and
the less common dysprosium, are needed for
By Benjamin K. Sovacool1, Saleem H. Ali2,3,4, more than 7100 GW (3). The materials and magnets in electric generators in wind tur-
Morgan Bazilian5, Ben Radley6, Benoit metals demanded by a low-carbon economy bines and motors in EVs, control rods for nu-
Nemery7, Julia Okatz8, Dustin Mulvaney9 will be immense (4). One recent assessment clear reactors, and the fluid catalysts for shale
concluded that expected demand for 14 gas fracking. But REE extraction in China has
C
limate change mitigation will cre- metals—such as copper, cobalt, nickel, and resulted in chemical pollution from ammo-
ate new natural resource and supply lithium—central to the manufacturing of nium sulfate and ammonium chloride and
chain opportunities and dilemmas, renewable energy, EV, fuel cell, and storage tailings pollution that now threaten rural
because substantial amounts of raw technologies will grow substantially in the groundwater aquifers as well as rivers and
materials will be required to build next few decades (5). Another study projected streams. Several metals for green technolo-
new low-carbon energy devices and increases in demand for materials between gies are found as “companions” to other ores
infrastructure (1). However, despite attempts 2015 and 2060 of 87,000% for EV batteries, with differential value and unsustainable
at improved governance and better corpo- 1000% for wind power, and 3000% for solar supply chains (9).
rate management, procurement of many cells and photovoltaics (6). Although they are
mineral and metal resources occurs in areas only projections and subject to uncertainty, POLICY RECOMMENDATIONS
generally acknowledged for mismanage- the World Bank put it concisely that “the With these sobering social and environmental
ment, remains environmentally capricious, clean energy transition will be significantly aspects of current mineral extraction in mind,
and, in some cases, is a source of conflict mineral intensive” (7) (see the figure). we suggest four policy recommendations.
at the sites of resource extraction (2). These Many of the minerals and metals needed
extractive and smelting industries have thus for low-carbon technologies are considered Diversify mining enterprises for local
left a legacy in many parts of the world of “critical raw materials” or “technologically ownership and livelihood dividends
environmental degradation, adverse impacts critical elements,” terms meant to capture Although large-scale mining is often eco-
to public health, marginalized communities the fact that they are not only of strategic or nomically efficient, it has limited employ-
and workers, and biodiversity damage. We economic importance but also at higher risk ment potential, only set to worsen with the
identify key sustainability challenges with of supply shortage or price volatility (8). But recent arrival of fully automated mines. Min-
practices used in industries that will supply their mining can produce grave social risks. ing can concentrate occupational hazards as
the metals and minerals—including cobalt, A majority of the world’s cobalt, used in the well as environmental risk, as demonstrated
copper, lithium, cadmium, and rare earth most common battery chemistries for EVs most severely by tailings pond disasters and
elements (REEs)—needed for technologies and stationary electricity storage, is mined mining wastewater contamination. Even
such as solar photovoltaics, batteries, elec- in the Democratic Republic of Congo (DRC) where there is relative political stability and
tric vehicle (EV) motors, wind turbines, fuel (see the map), a country struggling to recover stricter regulatory regimes in place, there
cells, and nuclear reactors. We then propose from years of armed conflict. There, women can still be serious environmental failures, as
four holistic recommendations to make min- and sometimes children often work in or exemplified by the recent global rise in dam
ing and metal processing more sustainable around mines for less pay or status than their failures at settling ponds for mine tailings.
and just and to make the mining and extrac- male and adult counterparts, without basic The level of distrust of extractive industries
tive industries more efficient and resilient. safety equipment (see the photo). Owing to a has even led to countrywide moratoria on all
Between 2015 and 2050, the global EV lack of preventative strategies and measures new mining projects, such as in El Salvador
stock needs to jump from 1.2 million light- such as drilling with water and proper ex- and the Philippines.
duty passenger cars to 965 million passenger haust ventilation, many cobalt miners have Traditional labor-intensive mechanisms of
cars, battery storage capacity needs to climb extremely high levels of toxic metals in their mining that are possible to undertake with
from 0.5 gigawatt-hour (GWh) to 12,380 body and are at risk of developing respiratory less mechanization and without major capital
GWh, and the amount of installed solar pho- illness, heart disease, or cancer. investments are called artisanal and small-
tovoltaic capacity must rise from 223 GW to In addition, mining frequently results in scale mining (ASM). Although ASM is not
severe environmental impacts and commu- immune from poor governance or environ-
1
University of Sussex, Brighton, UK. 2University of nity dislocation. Moreover, metal produc- mental harm, it provides livelihood potential
Delaware, Newark, DE, USA. 3University of Queensland, tion itself is energy intensive and difficult to for at least 40 million people worldwide, with
Brisbane, Queensland, Australia. 4United Nations
International Resource Panel, United Nations Environment decarbonize. Mining for copper, needed for an additional three to five times more people
Programme, Nairobi, Kenya. 5Colorado School of Mines, electric wires and circuits and thin-film solar indirectly supported by the sector (10). It is
Golden, CO, USA. 6London School of Economics, London, cells, and mining for lithium, used in batter- also usually more strongly embedded in local
UK. 7Katholieke Universiteit Leuven, Leuven, Belgium.
8
SYSTEMIQ Ltd., London, UK. 9San José State University, ies, has been criticized in Chile for depleting and national economies than foreign-owned,
San José, CA, USA. Email: b.sovacool@sussex.ac.uk local groundwater resources across the Ata- large-scale mining, with a greater level of
0103PolicyForum.indd 30 12/20/19 6:02 PM

value retained and distributed within the ability to increase productivity and mecha- committed to “responsible sourcing” of raw
country. Diversifying mineral supply chains nize production, even in hostile regulatory materials for batteries.
to allow for greater coexistence of small- and and governance environments. More space Traceability schemes, however, may be im-
large-scale operations is needed. Yet, efforts for and support to ASM to pursue this trajec- possible to fully enforce in practice and could,
to incorporate artisanal miners into the for- tory would enhance its capacity to meet the in the extreme, merely become an exercise in
mal economy have often resulted in a scar- increased demand for minerals required in public relations rather than improved gover-
city of permits awarded, exorbitant costs for the move toward a low-carbon future. One nance and outcomes for miners. In the east-
miners to legalize their operations, and ex- place to begin is with the redistribution of ern DRC, for example, cassiterite, the mineral
tremely lengthy and bureaucratic processes dormant mining concessions previously that tin is extracted from, is exported through
for registration. granted to (but unused by) mining compa- a traceability system yet can nonetheless have
Development donors need to focus on nies so that local ASM operators can legally contributed to conflict financing or labor and
bottom-up formalization efforts rather than work in these locations, as has been taking human rights abuses while simultaneously
merely facilitating government efforts to bet- place recently in Tanzania. introducing heavy financial costs onto local
workers for the right to participate in the sys-
tem (11). Nonetheless, traceability is not with-
out promise, and examples from Blockchain
technology show how the use of artificial in-
telligence algorithms for data processing has
the potential for greater assurance but ulti-
mately relies on the accuracy of data being
fed into the supply chain.
Transparency of supply chains is a means
to an end and will only be effective if con-
sumers or regulators start to differentiate
between products being provided. There are
effective lessons on traceability and transpar-
ency arising from the Kimberley Process for
conflict diamonds; the Extractive Industries
Transparency Initiative for oil, gas, and min-
eral resources; and the Fairmined Standard
for gold that could be applied to the mineral
supply chains needed for decarbonization.
Paramount among these is an acknowledg-
ment that traceability schemes offer a largely
technical solution to profoundly political
problems and that these political issues can-
not be circumvented or ignored if meaningful
solutions for workers are to be found. Trace-
ability schemes ultimately will have value if
the market and consumers trust their au-
thenticity and there are few potential oppor-
tunities for leakage in the system.
A creuseur, or digger, descends into a Congolese copper and cobalt mine in Kawama. Wages are low, and working
conditions are dangerous, often with no safety equipment or structural support for the tunnels. Explore new resource streams
Although primary emphasis must be placed
ter regulate the sector for increased tax rev- Acknowledge the limits of traceability on resource efficiency (higher output or us-
enues. There needs to be a focus on policies A great deal of attention has focused on fos- age of product per unit of resource input)
PHOTO: MICHAEL ROBINSON CHAVEZ/THE WASHINGTON POST VIA GETTY IMAGES
that recognize its livelihood potential in areas tering transparency and accountability of and recycling, there will likely be a need
of extreme poverty. Moreover, formalization mineral mining by means of voluntary trace- for primary resource extraction as well
of the sector should focus on creating stron- ability or even “ethical minerals” schemes. owing to clean-energy infrastructure de-
ger, more accountable arrangements to drive International groups, including Amnesty mand. New resource streams—including
greater value of resource revenues down the International, the United Nations, and the metal availability in seawater (desalina-
supply chain to ASM miners to ensure better Organisation for Economic Co-operation tion) and groundwater (geothermal brines),
environmental and safety mechanisms and and Development, have all called on mining material substitution or material intensity
expand their access to markets. The recent companies to ensure that supply chains are reductions, and materials recovery and
decision of the London Metals Exchange to not sourced from mines that involve illegal recycling—also hold promise for diversify-
have a policy of “nondiscrimination” toward labor and/or child labor. In concert, Eur- ing supply chains, as long as they maintain
ASM is a positive sign in this regard. Certain asian Resources Group (ERG) launched their environmental sustainability and protect
industry actors have demonstrated a com- Clean Cobalt Framework in 2018, First Cobalt worker safety.
mitment to, and the benefits of, this type has their Responsible Cobalt Initiative, RCS Although mining in terrestrial areas is
of approach, such as Fairphone’s sourcing Global has its Better Cobalt program, Am- likely to continue to meet the demands of
of the mineral columbite-tantalite (coltan) nesty International is working on an Ethical low-carbon technologies in the nearer term,
used in mobile phones. At the level of gov- Battery framework, and the World Economic we need to carefully consider mineral sources
ernment policy, ASM has demonstrated its Forum launched a Global Battery Alliance beneath the oceans in the longer term. The

INSIGHTS | P O L I C Y F O RU M
International Seabed Authority, set up under initially but can become more challenging the end of their useful life and encourages
the United Nations (UN) Convention on the to procure over time. Moreover, substitution durability, extended product lifetimes, and
Law of the Sea, is in the process of issuing may be limited to particular innovations or designs that are easy to reuse, repair, or re-
regulations related to oceanic mineral extrac- niches. Alternatives to lithium-ion batteries, cover materials from. A successful EPR pro-
tion. This process is a rare opportunity to be such as sodium-ion batteries, are becoming gram known as PV Cycle has been in place in
proactive in setting forth science-based envi- more practical and feasible. But finding sub- Europe for photovoltaics for about a decade
ronmental safeguards for mineral extraction. stitutes for metals like platinum group met- and has helped drive a new market in used
For metals such as cobalt and nickel, ocean als in key technologies such as fuel cells has photovoltaics that has seen 30,000 metric
minerals hold important prospects on the become increasingly difficult, and reserves tons of material recycled. To date, EPR has
continental shelf within states’ exclusive eco- are dwindling. mainly shaped collection, recycling, and
nomic zones as well as the outer continental Recycling and better resource efficiency waste management to ensure safe and re-
shelf regions. Within international waters, can play a part at extending and enhancing sponsible disposal of specific classes of prod-
metallic nodules found in the vast Clarion- the lifetimes of products and also stretch- ucts like e-waste, paint, and pharmaceuticals,
Clipperton Zone of the Pacific as well as in ing out mineral reserves. Closed-loop sup- but, in concept, it is also meant to help drive
cobalt and tellurium crusts, which are found ply chains based on circular economy ideas more sustainable design as well as options
in seamounts worldwide, provide some of the in addition to advancements in metallurgy, for reuse and repair. There is evidence of
richest deposits of metals for green technolo- reverse logistics, waste separation, materi- EPR’s influence on green design in the global
gies. Difficult extraction and declining re- als science, waste processing, and advanced solar industry. For example, thin-film manu-
facturer First Solar screens new materials to
ensure that they will not negatively influence
Growth in mineral needs for low-carbon energy technology their recycling process, through which they
currently recover 90% of their CdTe semicon-
Production Demand ductor material and 90% of their glass. To
Mineral (kilo–metric tons) more easily recycle the plastics and copper
2017 2050 from photovoltaics, some manufacturers are
Lithium 965% 43 415 seeking out halogen-free components.
110 644 Space mining, although potentially use-
Cobalt 585%
ful for developing lunar and planetary bases
Graphite 383% 1200 4590 farther into the future, has less potential for
Indium 241% 0.72 1.73 meeting the demand for minerals for imme-
80 138 diate decarbonization on Earth. A possible
Vanadium 173%
exception to this may be platinum group
Nickel 108% 2100 2268 metals from asteroids, but here, too, the time
Silver 60% 25 15 frame and quantity of production would pre-
Neodymium 23 8𰀏𰀕 clude its use in meeting immediate technol-
37%
ogy needs for climate mitigation.
Molybdenum 11% 290 33
Aluminum 9% 60,000 5583 Incorporate minerals into climate
19,700 1378 and energy planning
Copper 7%
Given the centrality of minerals and metals
Manganese 4% Percentage = ( 2017
2050 demand
production
) 16,000 694 to the future diffusion of low-carbon technol-
0 100% ogies, materials security should be actively
incorporated into formal climate planning.
All production and demand data refect annual values. 2017 data re.ect annual production for all uses. 2050 data re.ect estimated demand for only This could be connected to ongoing planning
low-carbon energy technology uses. Data from (7).
as part of the nationally determined contri-
butions (NDCs) under the Paris Accord, the
serves of some terrestrial minerals, as well as recycling can all enhance the longevity and European Commission’s National Energy
social resistance against terrestrial mining, continual reuse of minerals and metals. Re- and Climate Plans (NECPs), or even energy
may lead to oceanic mineral reserves becom- searchers at the U.S. National Renewable En- policy-making at the national scale. Climate
ing more plausible sources. Minerals near ergy Laboratory estimate that 65% of the U.S. planners could begin by mapping out their
hydrothermal vents are in more pristine and domestic cobalt demand in 2040 could be NDC contributions alongside a list of “criti-
distinctive ecosystems and should likely re- supplied by end-of-life lithium-ion batteries, cal minerals” for energy security (see supple-
main off-limits for mineral extraction for the provided a robust take-back and recycling in- mentary materials).
foreseeable future. frastructure is in place. Although care must be taken to ensure
Technological substitution can play an im- Extended producer responsibility (EPR) that the NDC process does not become too
portant role as well. Copper offers an illustra- is a framework that stipulates that produc- broad or research intensive, we believe the
tive example. Higher copper prices in recent ers are responsible for the entire lifespan of NDCs are the most tangible international
years have incentivized replacement in new a product, including at the end of its useful- policy consensus mechanism on this mat-
applications in the automotive industry, such ness. EPR would, in particular, shift respon- ter. The NDCs can incorporate some of the
GRAPHIC: X. LU/SCIENCE
as wire harnesses and replacing copper with sibility for collecting the valuable resource mineral sourcing challenges through ef-
aluminum winding in motors. However, sub- streams and materials inside used electron- forts at resource efficiency. The Group of
stitution to other primary metals or even syn- ics from users or waste managers to the com- Seven (G7) has taken on this linkage, and
thetics could merely shift resource demand to panies that produce the devices. EPR holds policies to motivate resource efficiency can
another material that may be more abundant producers responsible for their products at be a means of keeping track of material

and mineral supply chains. For example, a Countries accounting for the largest share of critical raw materials
materials assessment for particular infra-
structure options for climate change miti- Russia
46% Palladium
gation or adaptation could be included in
cost-benefit analyses. Recent work has sug-
China
gested that the social acceptability of ty- 87% Antimony
ing resource-efficient products to climate France
USA 44% Baryte
43% Hafnium
change mitigation efforts is strong (12). 90% Beryllium 82% Bismuth
73% Helium 64% Fluorspar
Having each country create a list of criti- 73% Gallium
cal minerals within its NDC process and 67% Germanium
show possible trade-offs and shortfalls Thailand 57% Indium
DRC
could lead to several benefits. More efforts Brazil 64% Cobalt 32% Natural rubber 87% Magnesium
90% Niobium 69% Natural graphite
on national critical material analysis could Rwanda 44% Phosphate rock
result in improved mapping of mineral South Africa 31% Tantalum 38% Phosphorus
85% Iridium 66% Scandium
supply chains, for which there is already 70% Platinum 61% Silicon metal
a notable gap across many developing 83% Rhodium 84% Tungsten
countries and regions. The analytical ef- 93% Ruthenium 53% Vanadium
forts would enhance our understanding of 95% LREEs
95% HREEs
supply constraints and demand patterns,
DRC, Democratic Republic of Congo; LREEs, light rare earth elements; HREEs, heavy rare earth elements. Figure modifed from European
which in turn could lead to a better un- Commission, "Third list of critical raw materials for the EU of 2017" (European Commission, 2017); https://ec.europa.eu/growth/sectors/
derstanding of future prices and drivers, raw-materials/specifc-interest/critical_en.
especially those beyond the control of gov-

ernments and policy as agents of change. Consideration should also be given to REF ERENCES AND NOTES
The process of mapping mineral demands where mining is most likely to have a 1. Intergovernmental Panel on Climate Change (IPCC),
“Global warming of 1.5°C,” V. Masson-Delmotte et al.,
for NDCs, NECPs, and national energy positive development footprint while also Eds. (IPCC Special Report, IPCC, 2018); www.ipcc.ch/
policies could lead to new linkages and having more manageable environmental sr15.
networks and a raising of awareness, con- impacts (14). Utilizing tools such as the Re- 2. D. M. Franks et al., Proc. Natl. Acad. Sci. U.S.A. 111, 7576
(2014).
necting the traditional minerals and met- sponsible Mining Index and platforms such 3. International Renewable Energy Agency (IRENA),
als community to other research and social as the Responsible Minerals Initiative or “Global energy transformation: A roadmap to 2050”
communities, especially in climate policy the Intergovernmental Forum on Mining, (IRENA, Abu Dhabi, 2018).
4. International Resource Panel, “Global resources outlook
and energy studies. In this way, climate Minerals, Metals and Sustainable Develop- 2019: Natural resources for the future we want” (United
mitigation could be twinned with minerals ment may be a way forward. Although there Nations Environment Programme, Nairobi, 2019); www.
security and industrial strategy as a way to may be treaty fatigue among policy-makers, resourcepanel.org/reports/global-resources-outlook.
5. E. Dominish, N. Florin, S. Teske, “Responsible miner-
meet broad sets of goals (environmental, an intertreaty protocol on mineral supply als sourcing for renewable energy” (Institute for
political, and economic) in one stroke. chains to ensure that the goals of existing Sustainable Futures, University of Technology, Sydney,
treaties are met could enhance effective gov- 2019).
6. A. Månberger, B. Stenqvist, Energy Policy 119, 226
AN ETHICAL CONUNDRUM ernance. Conversations in this vein should (2018).
Mineral and metal supplies are geologi- be attempted among the parties to the UN 7. World Bank, “Climate-smart mining: Minerals for
cally determined, yet socially mediated. Framework Convention on Climate Change, climate action” (World Bank, 2018); www.world-
bank.org/en/topic/extractiveindustries/brief/
Even if supplies are enhanced through co- through the UN Environment Assembly, as climate-smart-mining-minerals-for-climate-action.
products of other industries, new resource well as more focused mechanisms such as 8. European Commission, “Communication from the
Commission to the European Parliament, the Council,
streams, and considerable expansion of re- the U.S. government’s recently launched the European Economic and Social Committee, and the
cycling and increased recovery rates, there Energy Resource Governance Initiative, the Committee of the Regions on the 2017 list of critical
are likely to be bottlenecks across metal World Bank’s Climate-Smart Mining Facil- raw materials for the EU” (COM/2017/490, European
Commission, Brussels, 2017).
supply chains (13). This is exacerbated by ity, or the European Institute of Innovation 9. N. T. Nassar, T. E. Graedel, E. M. Harper, Sci. Adv. 1,
poorly functioning markets, as least for and Technology for Raw Materials. e1400180 (2015).
the minor metals. Hence, trade policy will Having just marked the 150th anniver- 10. World Bank, “2019 state of the artisanal and small-scale
mining sector” (World Bank, Washington, DC, 2019).
need to become more deftly aligned with sary of the formulation of the periodic 11. B. Radley, C. Vogel, Extr. Ind. Soc. 2, 406 (2015).
mineral supply in ways which are both table, it is high time we realize that the ele- 12. C. E. Cherry, K. Scott, J. Barrett, N. F. Pidgeon, Nat. Clim.
GRAPHIC: ADAPTED FROM EUROPEAN UNION/CC-BY BY X. LU/SCIENCE
economically and ecologically more effi- ments, and the minerals in which they are Chang. 8, 1007 (2018).
13. S. H. Ali et al., Nature 543, 367 (2017).
cient. Furthermore, more robust reporting embedded, are essential to our attainment 14. L. J. Sonter, S. H. Ali, J. E. M. Watson, Proc. R. Soc. London
and emissions data will be required across of low-carbon goals. There is an ethical co- Ser. B 285, 20181926 (2018).
the supply chain. For example, although nundrum to addressing climate change only
ACKNOWL EDGMENTS
the U.S. government strategy for mineral by aggravating other social and ecological
This project received funding from the European Union’s
supply security released in June 2019 high- problems related to unsustainable mineral Horizon 2020 research and innovation program under
lights the importance of trade with allies and metal supply chains. But done sustain- grant agreement no. 730403 “Innovation pathways,
and partners, it does not consider where ably, an impending mining boom could help strategies and policies for the low-carbon transition in
Europe (INNOPATHS).” The content of this deliverable
it is most ecologically efficient to source lift communities out of poverty, accelerate does not reflect the official opinion of the European Union.
minerals. Pursuing decarbonization si- technical innovation for decarbonization, Responsibility for the information and views expressed herein
multaneously with principles of a circular and further the realization of energy and lies entirely with the author(s).
economy, coupled with increased market climate targets. Which direction it takes SUPPL EMENTARY MATE RIALS
transparency mechanisms and full life- will depend considerably on how metal and science.sciencemag.org/content/367/6473/30/suppl/DC1
cycle reporting, could yield important so- mineral supply chains are governed over
cial and environmental benefits. the next few critical years. j 10.1126/science.aaz6003

Although this serves as important experi-
ence, students with disabilities may find it
prohibitively challenging to work so many
consecutive hours. A reduction in required
on-call hours would allow students with
and without disabilities to thrive in their
training. Students pursuing specialties
with extended-call commitments could
elect to gain this exposure.
Cody Lo
Faculty of Medicine, University of British Columbia,
Vancouver, BC V6T 1Z2, Canada. Twitter: @cody_lo
Despite being the most popular event at

scientific conferences, poster sessions are
probably the least accessible. As someone
who struggles with low vision, I find poster
sessions frustrating. Most posters have
unreadable fonts, unclear images, and
LET TERS poor color contrast. Conference organiz-
ers should encourage poster presenters to
make a 2-minute video summary of their
NEXTGEN VOICES work. The video and the presenter’s con-
tact information could then be accessed by
Making science accessible scanning a QR code. These QR codes would

help build camaraderie among research-
ers; unlike business cards, conference
To mark the 30th anniversary of the Americans with Disabilities Act, goers would have details of the presenters
we asked young scientists this question: What one thing would you together with their presentations.
change about the training or careers in your field to improve Edmond Sanganyado
Provincial Key Laboratory of Marine Biotechnology,
accessibility for people with visible and/or invisible disabilities? Shantou University, Shantou, Guangdong, 515063,
China. Twitter: @ESanganyado
A selection of their responses is below. Follow NextGen Voices on
Twitter with hashtag #NextGenSci. Read previous NextGen Voices survey
In physical anthropology, the majority of
results at https://science.sciencemag.org/collection/nextgen-voices. assessments of ancestry, sex, and age of
a skeleton are based on morphological
features. However, only a limited number
Provide logistical support discussions. Alternatively, written sum- of models are provided to demonstrate the
Because I have dyslexia, I use read-out- maries prepared beforehand should differences. The inclusion of 3D models
loud software, but it is not well suited to accompany visual and oral presentations. with differing variations of features would
journal articles with citations. Listening This would allow learners to read the be helpful to every student, not just the
to citations makes keeping track of the material as explained by the professor, ones who are differently abled. I believe it
paper difficult and can add hours of rather than a textbook, multiple times for would bridge the divide between recently
reading time. Some expensive software stronger comprehension. graduated students and experienced practi-
allows you to select what to read, but it Juliet Tegan Johnston tioners, allowing for greater mentorship of
can take days to mark the citations and is Department of Civil, Environmental, and Geo- other technical aspects of the work.
Engineering, University of Minnesota, Minneapolis,
impractical when reading many journal Kristy A. Winter
MN 55455, USA. Twitter: @queermsfrizzle
articles. A tool that allows the listener to College of Humanities, Arts, and Social
Sciences, Flinders University, Bedford Park, SA
skip the citations would help.
5042, Australia. Twitter: @Winterhiccup
Kristen Tuosto Wet labs are designed for people with no
Center for the Advanced Study of Human physical challenges. Many of us who could
Paleobiology, The George Washington University,
Washington, DC 20052, USA. contribute to science have difficulties not Having a disability can be very expensive.
Email: ktuosto@gwu.edu with the techniques or knowledge but In addition to costs such as out-of-pocket
rather with the need to spend hours sitting fees for therapy, people with mobility
on a lab stool. Adjustable benches and impairments may pay more for rent, as
ILLUSTRATION: ROBERT NEUBECKER

Movies have subtitles for a reason. People equipment would improve accessibility and accessible apartments tend to be more
might be hearing impaired or easily dis- the wet lab experience for all scientists. expensive, or more for food because of
tracted, have trouble deciphering accents, Carol Connolly specific dietary needs. These higher costs
or understand written information better. Dublin 18, Ireland. contribute to a higher risk of poverty
Email: Hateyes@yahoo.com
Speech-to-text technology, which enables for people with disabilities. Therefore,
instructors to instantly provide written increasing pay for Ph.D. students would
versions of their lecture materials, should Many fields of medicine require physicians make it more likely that students with
be used in real time during classroom to work on call for more than 24 hours. a disability—as well as students from
0103Letters.indd 34 12/23/19 9:33 AM

INSI GHTS
disadvantaged backgrounds—could pursue for a job when you have a disability. Will allow all graduate students the opportunity
a scientific career. department leaders be willing to make to develop a better work-life balance and
Tanja Roembke accommodations, and will they worry that become more productive scientists.
Institute of Psychology, RWTH Aachen University, hiring me will affect productivity? Academia Ashley B. Heim
Aachen, NRW 52062, Germany. needs to avoid direct comparison between School of Biological Sciences, University of
Email: tanja.roembke@psych.rwth-aachen.de Northern Colorado, Greeley, CO 80639, USA.
researchers with and without disabilities.
Email: ashley.heim@unco.edu
The academic paths of researchers with
Much of the scientific work in the field of disabilities may differ, for example, because
chemistry depends on colors. For those who of prolonged stays in hospitals or numerous The shame and stigma of disabilities
are color blind, the use of techniques such as surgeries and rehabilitations. However, these caused by depression and post-traumatic
acid-base titrations or spectroscopy presents “hardships” make us better researchers: stress disorder prevent the people who have
a challenge. Technology could help. If an app They push us to develop excellent organi- them from asking for help. Breaking this
could analyze a smartphone camera photo zational skills, creativity, stress resilience, barrier would be easier if colleagues and
and correct different types of color blindness emotional intelligence, and grit. Students in especially principal investigators and group
by converting the colors into decipherable STEM fields need role models in the form of leaders were trained to identify the signs and
patterns, it could be used not only in the lab lecturers or professors with disabilities. They reach out to ensure their support. Building
but also to convert figures in textbooks and need to see that it can be done and that the basic bridge of trust and safety is the
papers into more accessible images. universities and departments are inclusive most important first step in accessing people
Wagner Eduardo Richter in reality and not only on paper. with hidden disabilities. Supervisors should
Department of Chemical Engineering, Aleksandra Kosanic be trained to create and communicate a
Technological Federal University–Paraná, Paraná, Ecology Team, Department of Biology, safe, kind environment and to access further
Brazil. Email: richter@utfpr.edu.br University of Konstanz, D 78457 Konstanz, resources for help.
Germany. Twitter: @SashaKosanic
Pragya Srivastava
Hearing impairment affects a sizeable Department of Bioengineering, University
of Pennsylvania, Philadelphia, PA 19104, USA.
minority of world inhabitants. To better
Twitter: @Prags0211
welcome all scientists, we must incorporate
sign-language translation and closed-
captioning services in scientific meetings Facilitating accessible learning spaces for
and conferences. Computer technologies students with invisible disabilities poses
are advancing, and machine translation unique challenges, since neither student nor
services are now available at a fraction of instructor is initially aware of the other’s
their former cost. This simple modification condition. Moreover, some invisible diagno-
could have a lasting impact on the progres- ses, such as autism spectrum disorder (ASD),
sion of science through inclusion. carry a degree of social stigma. If they feel
Kyle J. Isaacson comfortable doing so, I believe it would be
Department of Biomedical Engineering, University valuable for course instructors and graduate
of Utah, Salt Lake City, UT 84112, USA. mentors affected by ASD to acknowledge
Twitter: @kjisaacson
this fact to their students. This would make
it easier for students to open up about their
In my field of astronaut training and own invisible challenges and reduce both
Foster inclusive community operations, it is clear that humans some- perceived and actual stigmatization.
When onboarding a new member, the times need equipment to thrive in their P. William Hughes
research group leader should affirm the environment. Just as an astronaut needs Department of Ecology, Environment, and Plant
Sciences, Stockholm University, Stockholm, 10691,
entire group’s commitment to accommo- specialized technology to survive in space Sweden. Twitter: @pwilliamhughes
dating anyone with a disability. Affirming and a diver needs scuba gear to survive
this commitment to every new member in water, people with disabilities benefit
ensures people with invisible disabilities from access to the right tools. Technology, We need to change the narrative to value
are not unintentionally ignored, empow- whether a brain-controlled exoskeleton, the experience of overcoming adversity
ers people with a disability to disclose artificial limbs, or a computer-generated rather than viewing those who have over-
information and request accommodations, voice, allows us to overcome environmental come challenges as inferior or damaged. I’d
and sets clear expectations for those with- challenges and transform into super- love to see a seminar series where estab-
out disabilities to make accommodation humans. If more people understood this lished principal investigators could openly
requests a priority. Having this conversa- concept, more individuals with disabilities discuss the challenges they faced and
tion during onboarding also helps form would be involved in scientific fields. how they overcame those challenges. This
a community that is supportive of people Loredana Bessone would increase representation and a sense
with disabilities and in turn benefits from European Space Agency, Cologne, NRW 51147, of belonging among people with invisible
their unique perspectives. Germany. Twitter: @esa_caves disabilities who might not otherwise know
Michael Raitor who else is out there. It would also reduce
Department of Mechanical Engineering, In graduate school, mental health is often stereotyping from able scientists.
Stanford University, Stanford, CA 94305, USA. put on the back burner. Encouraging
Email: mraitor@stanford.edu Caitlin M. Aamodt
more discussions about and providing Neuroscience Interdepartmental PhD Program,
more institutional support for mental University of California, Los Angeles, Los Angeles,
CA 90025-4816, USA. Twitter: @CaitlinAamodt
As a researcher with cerebral palsy, I know health would foster a more welcoming
first-hand the fear that comes with applying and accepting environment. It would also 10.1126/science.aba6129
0103Letters.indd 35 12/23/19 9:33 AM

RESEARCH
IN S CIENCE JOURNAL S
Edited by Michael Funk
Light micrograph
showing cortical
neurons with
branching dendrites
NEUROSCIENCE
Human dendrites are special
A
special developmental program in the human brain drives somatodendritic recordings revealed previously unknown classes
the disproportionate thickening of cortical layer 2/3. This of action potentials in the dendrites of these neurons, which
suggests that the expansion of layer 2/3, along with its make their activity far more complex than has been previously
numerous neurons and their large dendrites, may contribute thought. These action potentials allow single neurons to solve two
to what makes us human. Gidon et al. thus investigated the long-standing computational problems in neuroscience that were
dendritic physiology of layer 2/3 pyramidal neurons in slices taken considered to require multilayer neural networks. —PRS
from surgically resected brain tissue in epilepsy patients. Dual Science, this issue p. 83
MULTIPLEX GENOMICS capture gene expression profiles provides a key insight into the the knot as the two strands are
from thousands of experimental twin-forming process, which pulled apart. The authors show
Single-cell chemical conditions in a single experiment. might be exploited for develop- why some common knots slip
transcriptomics —LMZ ing nanoparticles suitable for a easily and untie, whereas others
Single-cell transcriptomic Science, this issue p. 45 variety of applications. —BG hold tight. —MSL
technologies have emerged as Science, this issue p. 40 Science, this issue p. 71
powerful tools to explore cellular
NANOPARTICLES
heterogeneity at the resolution
APPLIED PHYSICS
of individual cells. Srivatsan Finding the fivefold path OPTICS
et al. now add another layer of Crystals formed in nature or It’s knot what you know Miniaturizing particle
information and complexity by synthesized in a lab can have Why is it that some knots seem
combining single-cell transcrip- domains with different orienta- to hold tight while others readily accelerators
tomics with oligo hashing and tions that are called twins. Having slip apart? Patil et al. develop a Particle accelerators are usually
small molecule screening in a five of these domains is relatively theoretical analysis of the stabil- associated with large national
method called sci-Plex. Because common, but how these fivefold ity of knots and find links between facilities. Because photons
screening many chemical com- twins form is still a mystery. Song topological parameters (twist are able to impart momentum
pounds requires the ability to et al. combined in situ trans- charge, crossing numbers, hand- to electrons, there are also PHOTO: CNRI/SCIENCE SOURCE
profile many cells, and because mission electron microscope edness) and mechanical stability. efforts to develop laser-based
screens perturb cells in many observations with simulations The theory is confirmed using particle accelerators. Sapra
different ways, the authors to find two different pathways simulations and experiments et al. developed an integrated
demonstrate the effects of 188 for forming fivefold twins in on color-changing fibers that particle accelerator using pho-
compounds in three cancer gold, platinum, and palladium optically show localized stress tonic inverse design methods to
lines. The sci-Plex method can nanoparticles. This combination differences in different parts of optimize the interaction between

the light and the electrons. They or changes in cis-regulatory
IN OTHER JOURNALS Edited by Caroline Ash
show that an additional kick of sequences? Britton et al. exam-
and Jesse Smith
around 0.9 kilo–electron volts ine the Mata2 protein in a
(keV) can be given to a bunch Saccharomycotina clade of
of 80-keV electrons along just fungi. They show that a newly
30 micrometers of a specially evolved transcription circuit Artist’s conception
designed channel. Such miniatur- involving repression of the of a Snowball Earth
ized dielectric laser accelerators a-specific genes by the ancient almost entirely
could open up particle physics to homeodomain protein Mata2 covered by ice
a number of scientific disciplines. occurred in two stages sepa-
—ISO rated by millions of years. In the
Science, this issue p. 79 first stage, Mata2 acquired
several coding changes followed
by changes in cis-regulatory
ARCHAEOLOGY
sequences. This clade-specific
Middle Stone Age cooking requirement explains how the
Early evidence of cooked starchy coding changes of Mata2 were
plant food is sparse, yet the con- in place long before the new
sumption of starchy roots is likely a-specific gene repression circuit
to have been a key innovation arose. —BAP
in the human diet. Wadley et al. Science, this issue p. 96
report the identification of whole,
charred rhizomes of plants of the
genus Hypoxis from Border Cave, PARASITIC INFECTIONS
South Africa, dated up to 170,000
years ago. These archaeobotani-
Parasite perturbation
cal remains represent the earliest of immunity
direct evidence for the cooking of Helminths—a broad grouping
underground storage organs. The of parasitic worms—infect bil-
edible Hypoxis rhizomes appear lions of people and are known
to have been cooked and con- to modulate host immune
sumed in the cave by the Middle responses to promote their own
Stone Age humans at the site. survival. De Ruiter et al. used SNOWBALL EARTH
Hypoxis has a wide geographi- mass cytometry to determine
cal distribution, suggesting that which immune cells are affected Beating the freeze
D
the rhizomes could have been a by helminth infection. They espite multiple episodes during the Cryogenian period
ready and reliable carbohydrate analyzed samples from rural in which the planet was mostly covered in ice for millions
source for Homo sapiens in Indonesians before and after of years—essentially shutting down air–sea exchange
Africa, perhaps facilitating the deworming treatment. These and causing widespread ocean anoxia—marine aerobic
mobility of human populations. were compared to samples eukaryotes somehow survived those Snowball Earth
conditions. How did they do it? Lechte et al. show that oxygen-
PHOTO (FROM LEFT): SURADECH KONGKIATPAIBOON/ALAMY STOCK PHOTO; CHRIS BUTLER/SCIENCE SOURCE
—AMS from Europeans and urban

Science, this issue p. 87 Indonesians, neither of which ated glacial meltwater was supplied to the oceans near ice
had been exposed to helminths. sheet grounding lines, offering a lifeline to microbial communi-
Helminths expanded specific ties there. This process may have created glaciomarine oxygen
EVOLUTIONARY BIOLOGY type 2 and regulatory immune oases that were critical to the survival and evolution of those
cells, and many immune altera- organisms during episodes of extreme glaciation. —HJS
Generating a new tions reverted upon deworming. Proc. Natl. Acad. Sci. U.S.A. 116, 25478 (2019).
transcriptional network These details on host–pathogen
Organismal novelties result interaction could inform future
from changes in transcriptional targeted therapies. —LP
circuits. But what comes first, Sci. Transl. Med. 12, T CELL SIGNALING the targeting of either alone, but
changes in regulatory protein eaaw3703 (2020). how they functionally cross-talk
Checkpoint cross-talk is not well defined. Zhao et al.
Immunotherapy using antibod-
report that PD-L1 (the main
ies that block the programmed
ligand of PD-1) and CD80 (shared
cell death (ligand) 1 [PD-(L)1] ligand for CTLA-4 and the central
or cytotoxic T lymphocyte– costimulatory receptor CD28),
associated protein 4 (CTLA-4) heterodimerize in cis. Using their
immune checkpoint pathways model system, the authors found
has resulted in impressive the PD-L1:CD80 cis complexes
responses for some cancer to be defective in binding either
Modern example of storage organs patients. Combined inhibition PD-1 or CTLA-4, but the abil-
from a species of the Hypoxis genus, of both pathways has generally ity of CD80 to activate CD28
which includes the African potato delivered better responses than appeared to be fully preserved.

RESE ARCH | I N O T H E R J O U R NA L S
METALS IN BIOLOGY
Copper on the half shell
O
ysters, like many marine mollusks, have
a complex life cycle that requires changes
in their body and biochemistry. Weng et al.
examined how the abundance and distribu-
tion of copper, a trace element essential for
metabolism, changes in larval and juvenile oysters
as they develop. Using a combination of imag-
ing techniques, they see copper accumulation in
certain organs during developmental transitions,
especially in the cells’ mitochondria. Both cell
proliferation and programmed cell death involve
biochemical processes linked to mitochondrial
copper concentrations. Understanding metal
ion distributions may offer insights valuable for
conservation and cultivation. —MAF
Environ. Sci. Technol. 53, 14724 (2019).
Oysters have a particularly critical need for the trace element

copper during their late pelagic larval stage (pictured).
The interaction of checkpoint the lines and edges converge on and inactivate perforin. —SMH that selective transplantation of
pathways might allow opportuni- her mental model of a scene, thus Nat. Commun. 10, 5396 (2019). extinct taxa could aid a healthy
ties to improve therapy. —PNK providing a window into her visual microbiota to recover after anti-
Immunity 51, 1059 (2019). cortex. —PRS biotics exposure. —CA
MICROBIOTA
J. Neurosci. 39, 9410 (2019). Cell Host Microbe 26, 650 (2019).
NEUROSCIENCE
Postantibiotic recovery
Antibiotic consumption by
CELL BIOLOGY
Filling a blank canvas humans is ubiquitous. Although PHYSICS
Natural scenes can be decoded in Kill, but don’t be killed they are lifesaving and valuable
Tailoring electronic
PHOTO: STEPHANIE ALEXANDER, UNIVERSITY OF MARYLAND CENTER FOR ENVIRONMENTAL SCIENCE HORN POINT OYSTER HATCHERY
the brain in the early visual cortex. Cytotoxic T lymphocytes (CTLs) medicines, antibiotics can harm
Decoding is a product of planned recognize and destroy infected resident mutualistic micro- properties
output (feedforward) corrected or malignant cells. CTLs secrete organisms in unpredictable ways The layered material a-RuCl3
by experience (feedback). Morgan the pore-forming protein per- leading to physiological damage. has piqued the interest of physi-
et al. occluded one quadrant of forin into the immune synapse, Ng et al. systematically explored cists as a potential platform for
the visual field of human partici- where perforin disrupts the the long-term effects of antibi- the so-called Kitaev model—a
pants and then presented them target cell plasma membrane otic treatment on transplanted type of quantum spin liquid
with images of landscape scenes. and initiates cell death pathways. human microbiota in mice. The that hosts exotic excitations.
Responses in the visual cortex Paradoxically, the secreted alpha-diversity of Bacteroides However, in addition to the
(areas V1 and V2) were measured perforin only damages the spp. suffered mass extinction magnetic interactions of the
for each person by functional target cell, leaving the producer by streptomycin within hours of Kitaev model, a-RuCl3 also hosts
magnetic resonance imaging. lymphocytes unscathed and treatment, whereas Firmicutes non-Kitaev interactions. Biswas
Even in the occluded region, able to attack another target cell. were more resilient. Such taxa- et al. investigated theoretically
information about the scene was What protects the lymphocyte specific selection depended on whether a heterostructure
being decoded. To access what is unclear. Using a variety of the antibiotic used, influenced consisting of a a-RuCl3 mono-
the brain had modeled for the approaches, Rudd-Schmidt et the recovered community com- layer on top of graphene would
occluded region, participants al. found that two distinct but position, and affected resilience be a better approximation of
were asked to complete the image coordinated mechanisms were to subsequent treatments. the Kitaev model. Using ab initio
by line drawings on an electronic deployed by CTLs to protect Diet was an important variable: calculations, the researchers
drawing tablet. The drawings were themselves. Both mechanisms Without microbiota-accessible found that the a-RuCl3 layer
shown to all participants, who depended on the dynamic regu- fiber in the diet, mucus graz- would experience strain and
scored how well the “averaged” lation of plasma membrane lipid ers predominated, which may become doped by graphene
drawing for each scene matched composition and topology. High facilitate the activity of patho- electrons. The strain would
the complete image. Mental membrane lipid order made gens, such as Acinetobacter make the Kitaev interactions
models at the first visual cortical the CTL presynaptic membrane muciniphila. Recolonization from more dominant, and the doping
stage were predicted by the line refractory to perforin binding. untreated cage mates was also may provide a route towards
drawings. So, faced with a blank Furthermore, phosphatidyl- vital for healthy reestablishment p-wave superconductivity in this
canvas, an artist may be using serine, exposed on the killer cell of the microbiota, especially system. —JS
cortical feedback processing until membrane, appeared to bind of Bacteroidetes. It is possible Phys. Rev. Lett. 123, 237201 (2019).

RESE ARCH
ALSO IN SCIENCE JOURNALS Edited by Michael Funk
NEUROSCIENCE MALARIA associated with condensed mat- PLANT SCIENCE

ter systems. —ISO
The neural substrate An artemisinin resistance Science, this issue p. 59
Subtle origin for
of memory mechanism complex shapes
The ability to form memory is Species of the malaria parasite The cup-shaped leaves of
an essential trait that allows Plasmodium live in red blood TOPOLOGICAL MATTER carnivorous plants have evolved
learning and the accumula- cells and possess a highly multiple times from ancestors
tion of knowledge. But what is conserved gene called kelch13.
Looking for chiral with flat leaves. Studying devel-
a memory? There has been a Single point mutations in this Majoranas opment of the carnivorous trap
long history of searching for the gene are associated with resis- Chiral Majorana modes have been in the humped bladderwort,
neuronal substrate that forms tance to the frontline artemisinin predicted to exist in heterostruc- Utricularia gibba, Whitewoods
memory in the brain, and the drugs. Birnbaum et al. found that tures consisting of a quantum et al. identified genes similar to
emerging view is that ensembles Kelch13 and associated proteins anomalous Hall insulator and a those expressed in surfaces of
of engram cells explain how comprise an endocytic compart- superconductor. Kayyalha et al. flat leaves (see the Perspective
memories are formed and ment associated with feeding fabricated more than 30 such by Moulton and Goriely).
retrieved. In a Review, Josselyn on host erythrocytes (see the samples and used transport Ectopic expression and compu-
and Tonegawa discuss the Perspective by Marapana and measurements to look for signa- tational modeling reveals how
evidence for engram cells as a Cowman). Hot targets for arte- tures of the Majorana modes. The slight shifts in gene expression
substrate of memory, particu- misinin research also occur in data indicated that the transport domains make the difference
larly in rodents; what we have this compartment, including the signatures previously thought between a flat leaf and a convo-
learned so far about the features proteins UBP1, AP-2m, and the to be associated with Majorana luted trap structure. Flexibility
of memory, including memory parasite homolog of the endocy- physics could, in their samples, be in growth rates in orthogonal
formation, retrieval over time, tosis protein Eps15. Inactivation explained using a more mundane polarity fields allows for diver-
and loss; and future directions of Kelch13 compartment mechanism. —JS sity in shapes formed through
to understand how memory proteins revealed that these Science, this issue p. 64 development. —PJH
becomes knowledge. —GKA are required for endocytosis of Science, this issue p. 91;
Science, this issue p. 39 host hemoglobin. Artemisinins see also p. 24
are activated by hemoglobin TOPOLOGICAL MATTER
degradation products, so these
GLOBAL HEALTH mutations render the parasite
A possible propagating TUBULIN
Cancer in diverse resistant to these drugs to differ- Majorana Mechanism of tubulin
ent extents. —CA Majorana states in solid-state
populations Science, this issue p. 51; systems may one day form a autoregulation
Much of what is known about see also p. 22 basis for topological quantum Cells tightly control the abun-
the signature genetic altera- computing. Most of the candi- dance of key housekeeping
tions and clinical features of dates identified so far have been factors, such as ribosomes and
cancer comes from studying TOPOLOGICAL OPTICS Majorana bound states, but chaperones, to maintain them
Caucasian patients in high- theorists have predicted that at optimal levels needed for
income countries. However, this
Optically contorting into propagating Majorana states may homeostasis. Most abundance
does not reflect the diversity of new dimensions exist as well. Wang et al. looked control mechanisms involve
heritage in most societies and Creating synthetic dimensions for such a state on the surface feedback regulation of mRNA
therefore does not allow patients has generated interest in many of the material FeSe0.45Te0.55 transcription, but others, such as
to be diagnosed and treated branches of science, ranging (see the Perspective by Tewari tubulins, are regulated by highly
effectively. In a Perspective, from ultracold atomic physics and Stanescu). Using scanning specific mRNA degradation. Lin
Rebbeck discusses the emerging to photonics. The ability to do tunneling spectroscopy, the et al. found that tetratricopep-
evidence that cancers in patients so provides a versatile platform researchers measured a flat, tide protein 5 (TTC5) binds to
in sub-Saharan Africa are clini- for realizing effective gauge bias-independent density of nascent alpha and beta tubulins
cally and molecularly distinct. potentials and novel topological states along a particular type of on translating ribosomes to
Understanding these differences physics that might be difficult domain wall, which was consis- trigger degradation of their
should improve treatment of or impossible to realize in real tent with a theoretical prediction associated mRNAs when excess
patients in sub-Saharan Africa as systems. Dutt et al. show that a for a propagating Majorana tubulin is present (see the
well as the diaspora. Assessment structured optical ring cavity can state in this material. Although Perspective by Shoshani and
of the diversity of genetic sustain more than one synthetic topologically trivial origins of this Cleveland). In the absence of
alterations can also improve how dimension. Under modulation, finding are difficult to completely TTC5-mediated tubulin autoreg-
cancer is diagnosed and classi- coupling the different degrees of rule out, the work is likely to ulation, cells display error-prone
fied in all populations. —GKA freedom within the resonator is stimulate interest in iron-based chromosome segregation, a
Science, this issue p. 27 used synthesize two additional superconductors as hosts of process critically dependent on
dimensions. The authors are Majorana states. —JS tubulin concentration. —SMH
then able to emulate many com- Science, this issue p. 104; Science, this issue p. 100;
plex physical phenomena usually see also p. 23 see also p. 29
38-B 3 JANUARY 2020 • VOL 367 ISSUE 6473 sciencemag.org SCIENCE

RE S E ARC H
FLUORESCENT PROTEINS stemming from the inhomoge-

neity of the chemical doping
Electrostatics guide process. In particular, the size of
chromophore twist the spectroscopic gap can vary
Photoisomerization—the widely across a single sample.
twisting of bonds in a molecule Massee et al. used the tip of a
in response to absorption of scanning tunneling microscope
light—is exploited in biology to to manipulate atoms on the
sense light and can influence surface of a member of the
the photophysical properties Bi2Sr2CaCu2O8+x cuprate family.
of fluorescent proteins used in Moving bismuth atoms up or
imaging applications. Romei et down caused adjacent atoms to
al. studied this behavior by intro- shift laterally, leading to revers-
ducing unnatural amino acids ible local changes in the size of
into the photoswitchable green the gap. It is expected that the
fluorescent protein Dronpa2, technique can be used to probe
thus systematically altering the the influence of the local lattice
electronic properties of the chro- on the electronic states of other
mophore (see the Perspective correlated materials. —JS
by Hu et al.). Crystal structures Science, this issue p. 68
and spectroscopic analyses of a
series of these variants support a
model in which the electrostatic
interactions between the chro-
mophore and its environment
influence the barrier heights for
twisting around different bonds
during photoisomerization. These
insights may guide future design
of photoswitchable proteins with
desired properties. —MAF
Science, this issue p. 76;
see also p. 26
CANCER
A movement orchestrated
by DNA damage
Doxorubicin is a commonly used
therapeutic that kills cancer
cells by inducing genotoxic
stress. Harvey et al. found
that treatment of cancer cells
with clinically relevant doses
of doxorubicin promoted the
phosphorylation of the transla-
tion factor eIF2a, which, rather
than inhibiting protein syn-
thesis as might be expected,
led to increased migration.
Inhibiting eIF2a phosphoryla-
tion with a U.S. Food and Drug
Administration–approved drug
restricted doxorubicin-induced
cell migration. —WW
Sci. Signal. 12, eaaw6763 (2019).
SUPERCONDUCTIVITY
Manipulating the gap
Cuprate superconductors
typically have a large amount
of spatial inhomogeneity, partly
SCIENCE sciencemag.org 3 JANUARY 2020 • VOL 367 ISSUE 6473 38-C

RESEAR CH
◥ memory consolidation and retrieval processes.

REVIEW SUMMARY Third, it is now possible to artificially manip-
ulate memory encoding and retrieval processes
NEUROSCIENCE to generate false memories, or even create a
memory in mice without any natural sensory
Memory engrams: Recalling the past and ON OUR WEBSITE
◥
experience (implantation
of a memory for an expe-
imagining the future Read the full article rience that did not occur).
at http://dx.doi. Fourth, “silent” engrams
Sheena A. Josselyn* and Susumu Tonegawa* org/10.1126/ were discovered in amne-
science.aaw4325 sic mice; artificial reac-
..................................................
tivation of silent engrams
BACKGROUND: The idea that memory is stored There are several recent advances in engram induces memory retrieval, whereas natural
as enduring changes in the brain dates back research. First, eligible neurons within a given cues cannot. Endogenous engram silencing
at least to the time of Plato and Aristotle (circa brain region were shown to compete for al- may contribute to the change in memory over
350 BCE), but its scientific articulation emerged location to an engram, and relative neuronal time (e.g., systems memory consolidation) or
in the 20th century when Richard Semon excitability determines the outcome of this in different circumstances (e.g., fear memory
introduced the term “engram” to describe the competition. Excitability-based competition extinction). These findings suggest that once
neural substrate for storing and recalling also guides the organization of multiple en- formed, an engram may exist in different
memories. Essentially, Semon pro- states (from silent to active) on the
posed that an experience activates a basis of their retrievability. Although
population of neurons that undergo initial engram studies focused on sin-
persistent chemical and/or physical gle brain regions, an emerging concept
changes to become an engram. Sub- is that a given memory is supported
sequent reactivation of the engram by an engram complex, composed of
by cues available at the time of the functionally connected engram cell
experience induces memory retrieval. ensembles dispersed across multiple
After Karl Lashley failed to find the brain regions, with each ensemble
engram in a rat brain, studies attempt- supporting a component of the overall
ing to localize an engram were largely memory.
abandoned. Spurred by Donald O. Hebb’s
theory that augmented synaptic strength OUTLOOK: The ability to identify and
and neuronal connectivity are critical manipulate engram cells and brain-
for memory formation, many research- wide engram complexes has intro-
ers showed that enhanced synaptic duced an exciting new era of memory
strength was correlated with memory. research. The findings from many labs
Nonetheless, the causal relationship are beginning to define an engram as
between these enduring changes in the basic unit of memory. However,
synaptic connectivity with a specific, many questions remain. In the short
behaviorally identifiable memory at term, it is critical to characterize how
the level of the cell ensemble (an en- information is stored in an engram,
gram) awaited further advances in including how engram architecture
experimental technologies. affects memory quality, strength, and
An engram cell alongside a nonengram cell. Within the hippocam-
precision; how multiple engrams in-
pus, dentate gyrus cells were filled with biocytin (white) to examine
ADVANCES: The resurgence in research teract; how engrams change over time;
morphology. Engram cells active during context fear conditioning were
examining engrams may be linked and the role of engram silencing in
engineered to express the red fluorescent protein mCherry, which
to two complementary studies that these processes. The long-term goal
appears pink owing to overlap with biocytin signals. Axons of the
applied intervention strategies to tar- of engram research is to leverage the
perforant path (green) express the excitatory opsin channelrhodopsin 2
get individual neurons in an engram fundamental findings from rodent en-
and a fluorescent marker (enhanced yellow fluorescent protein). The
CREDIT: ADAPTED FROM T. J. RYAN ET AL., SCIENCE 348, 1007 (2015).
supporting a specific memory in mice. gram studies to understand how infor-

upper blade of the dentate gyrus granule cell layer is revealed by the
One study showed that ablating the mation is acquired, stored, and used in
nuclear stain 4′,6-diamidino-2-phenylindole (DAPI, blue).
subset of lateral amygdala neurons humans and facilitate the treatment of
allocated to a putative engram dis- human memory, or other information-
rupted subsequent memory retrieval (loss grams in the brain and determines how these processing, disorders. The development of
of function). The second study showed that engrams interact. Second, research examin- low- to noninvasive technology may enable
artificially reactivating a subset of hippocam- ing the nature of the off-line, enduring changes new human therapies based on the growing
pal dentate gyrus neurons that were active
during a fearful experience (and, therefore,
in engram cells (neurons that are critical com-
ponents of an engram) found increased synap-
knowledge of engrams in rodents.
▪
part of a putative engram) induced memory tic strength and spine density in these neurons
retrieval in the absence of external retrieval as well as preferential connectivity to other
cues (gain of function). Subsequent findings downstream engram cells. Therefore, both The list of author affiliations is available in the full article online.
from many labs used similar strategies to increased intrinsic excitability and synaptic *Corresponding author. Email: sheena.josselyn@sickkids.ca
(S.A.J.); tonegawa@mit.edu (S.T.)
identify engrams in other brain regions sup- plasticity work hand in hand to form engrams, Cite this article as S. A. Josselyn and S. Tonegawa, Science
porting different types of memory. and these mechanisms are also implicated in 367, eaaw4325 (2020). DOI: 10.1126/science.aaw4325

RES EARCH
◥ veloped a cell assembly theory (similar to

REVIEW Semon’s engram complex) (15). Hebb hypothe-
sized that a cell assembly is formed between
NEUROSCIENCE reciprocally interconnected cells that are sim-
ultaneously active during an experience. Suffi-
Memory engrams: Recalling the past and cient activity within the cell assembly induces
growth and/or metabolic changes that strength-
imagining the future en the connections between these cells [a con-
cept distilled in the phrase “neurons that fire
Sheena A. Josselyn1,2,3,4,5* and Susumu Tonegawa6,7* together, wire together” (16)]. These synap-
tic and metabolic changes (perhaps including
In 1904, Richard Semon introduced the term “engram” to describe the neural substrate for storing memories. changes in intrinsic neuronal excitability) have
An experience, Semon proposed, activates a subset of cells that undergo off-line, persistent chemical implications for the function of a cell assem-
and/or physical changes to become an engram. Subsequent reactivation of this engram induces memory bly. For instance, reactivation of only a frac-
retrieval. Although Semon’s contributions were largely ignored in his lifetime, new technologies that tion of assembly cells was hypothesized to
allow researchers to image and manipulate the brain at the level of individual neurons has reinvigorated produce reactivation of the entire assembly
engram research. We review recent progress in studying engrams, including an evaluation of evidence for (15) [a process similar to pattern completion
the existence of engrams, the importance of intrinsic excitability and synaptic plasticity in engrams, and the (17–19)]. By contrast, destruction of a fraction
lifetime of an engram. Together, these findings are beginning to define an engram as the basic unit of memory. of assembly cells would not necessarily produce
catastrophic failure of the entire represent-
M
ation (but rather gracefully degrade the rep-
emory is the ability to use the past in change in some aspect of brain state but was resentation). Interestingly, Semon also proposed
service of the present or future (1, 2). suitably cautious when asked to speculate on similar types of properties for an engram (5).
Memory is central to our everyday lives the precise neural mechanisms underlying an Together, these (and other) scientists helped
and defines who we are. Without it, engram, “To follow this into the molecular define and describe an engram. However, there
we are condemned to an eternal pre- field seems to me…a hopeless undertaking at was a paucity of studies examining the biol-
sent. That memory persists after an experience the present stage of our knowledge and for my ogical basis of engrams. More than 100 years
suggests that an internal representation of part, I renounce the task” (7, p. 154). ago, Semon wrote that to examine the neuro-
this experience is stored in the brain and that A few years later, though, Karl Lashley, a biological basis of an engram represented a
later this representation can be reconstructed geneticist turned psychologist, took up this “hopeless undertaking.” This may no longer be
and used. In 1904, Richard Semon, an evolu- challenge by systematically attempting to lo- true. Recent excitement surrounding engram
tionary zoologist turned memory theorist, in- calize an engram in a mammalian brain (12–14). research may stem directly from the develop-
troduced the term “engram” to describe such In a typical study, Lashley trained rats over ment of new tools allowing cell ensembles to
memory representations (3, 4). Semon defined many days to solve a maze by running a dis- be imaged and manipulated at the level of the
an engram as “…the enduring though primar- tinct route to collect a reward. Hypothesizing individual cell. We begin by briefly reviewing
ily latent modifications in the irritable sub- that some critical component of the engram the neurobiological evidence supporting the
stance produced by a stimulus…” (5, p. 12; 6). supporting this maze-route memory is local- existence of engrams in the rodent brain and
He postulated a fundamental “law of engra- ized in the cortex, Lashley removed cortical our collective ability to not only find but also
phy” in which “all simultaneous excitations… tissue of varying sizes from varying locations manipulate engrams to better understand
form a connected simultaneous complex of and then tested the rats’ memory for the maze memory. Then, we discuss the current state of
excitations which, as such, act engraphically, route. Although the amount of cortical tissue engram research by examining the results of
that is to say leaves behind it a connected, removed correlated with overall memory im- explicit engram studies and previous memory
and to that extent, unified engram-complex” pairment, the location of the lesion did not. and plasticity findings from an engram point of
(7, p. 159–160). An engram, therefore, is rough- After more than 30 years of searching, Lashley view. Guided by Semon, we define an engram
ly equivalent to a “memory trace.” failed to find an engram, declaring it “elusive.” as an enduring off-line representation of a past
Semon’s innovative ideas were largely over- The next leap in engram-related research experience (Box 1). It is important to note that
looked or dismissed during his lifetime. How- came when Donald O. Hebb, a psychologist, an engram is not yet a memory but rather pro-
ever, his theories foreshadowed many prominent memory theorist, and student of Lashley, de- vides the necessary physical conditions for a
contemporary memory concepts (8–11). Semon
defined an engram as an off-line, physical
1
Box 1. Engram definitions.
Program in Neurosciences & Mental Health, Hospital for
Sick Children, Toronto, Ontario M5G 1X8, Canada.
2
Department of Psychology, University of Toronto, Toronto, An “engram” refers to the enduring offline physical and/or chemical changes that were elicited by
Ontario M5S 3G3, Canada. 3Department of Physiology, learning and underlie the newly formed memory associations.
University of Toronto, Toronto, Ontario M5G 1X8, Canada.
4
“Engram cells” are populations of cells that constitute critical cellular components of a given
Institute of Medical Sciences, University of Toronto,
Toronto, Ontario M5S 1A8, Canada. 5Brain, Mind &
engram. These cells may (or may not) also be critical components of engrams supporting other
Consciousness Program, Canadian Institute for Advanced memories. Engram cells are (i) activated by a learning experience, (ii) physically or chemically modified
Research (CIFAR), Toronto, Ontario M5G 1M1, Canada. by the learning experience, and (iii) reactivated by subsequent presentation of the stimuli present at
6
RIKEN-MIT Laboratory for Neural Circuit Genetics at the
the learning experience (or some portion thereof), resulting in memory retrieval.
Picower Institute for Learning and Memory, Department of
Biology and Department of Brain and Cognitive Sciences, An “engram cell ensemble” refers to the collection of engram cells localized within a brain region.
Massachusetts Institute of Technology, Cambridge, MA 02139, Engram cell ensembles in each brain region are connected, forming an “engram complex,” which is the
USA. 7Howard Hughes Medical Institute, Massachusetts
entire brainwide engram supporting a memory that is stored in sets of engram cell ensembles in
Institute of Technology, Cambridge, MA 02139, USA.
*Corresponding author. Email: sheena.josselyn@sickkids.ca different brain regions connected via an engram cell pathway.
(S.A.J.); tonegawa@mit.edu (S.T.)
Josselyn and Tonegawa, Science 367, eaaw4325 (2020) 3 January 2020 1 of 14

RES EARCH | R E V I E W
memory to emerge (20). Memories are re- semble, Mayford and colleagues (41) tagged neurons after training (58). Ablating CREB-
trieved when appropriate retrieval cues success- neurons active during auditory fear condition- overexpressing neurons disrupted freezing to
fully reactivate an engram in a process Semon ing. In this commonly used memory task, an subsequent tone presentation, as if the mem-
dubbed “ecphory.” initially innocuous tone (a conditioned stimu- ory was erased (Fig. 1). Importantly, mice were
lus) is paired with an aversive footshock (an capable of learning a new fear conditioning
Experimental strategies to evaluate engrams unconditioned stimulus) in a conditioning task (showing overall LA function was not
To evaluate the existence of engrams, we adapt context. When subsequently reexposed to the compromised), and ablating a similar number
the criteria and experimental strategies dis- tone or conditioning context, rodents freeze of non–CREB-overexpressing cells (nonengram
cussed by Morris and colleagues (21, 22) in (the active, learned conditioned response), cells) did not disrupt memory (showing speci-
their landmark papers evaluating the impor- showing memory of the training experience ficity of the memory disruption at the cellular
tance of synaptic plasticity in memory. Spe- (44). In this experiment, mice were replaced in level).
cifically, we discuss evidence from four types the conditioning context 3 days after training, Subsequent studies using diverse methods to
of studies. First, observational studies sup- and active neurons were marked with zif268 permanently or reversibly inactivate allocated
porting the existence of engrams in the rodent immunohistochemistry. Consistent with the or tagged neurons across several brain areas
brain should show that the same (or overlap- existence of an engram supporting this con- hypothesized to be part of an engram, in many
ping) cell populations are activated both by an ditioned fear memory, the overlap of neurons memory tasks, produced comparable results
experience and by retrieval of that experience active during training (tagged) and testing (40, 48, 53, 63, 66, 67). Together, these findings
and that, furthermore, learning should induce (zif268+) in the basal amygdala nucleus ex- suggest that neurons active during an expe-
long-lasting cellular and/or synaptic modifica- ceeded chance (~11% total cells) (41). rience become engram cells that are indispens-
tions in these cells. Second, loss-of-function Similar results, using different tagging meth- able (or somehow necessary) for successful
studies should show that impairing engram ods, across multiple brain regions [including subsequent memory expression.
cell function after an experience impairs sub- dorsal hippocampus (40, 45–55), amygdala Why were these loss-of-function studies per-
sequent memory retrieval. Third, gain-of- (41, 45, 49, 51, 55, 56), and cortex (42, 45, 55, 57)] haps successful in “finding an engram” when
function studies should show that artificially were reported for a variety of different mem- Lashley was not? First, Lashley may have used
activating engram cells induces memory ret- ory tasks (including contextual fear condition- an inappropriate behavioral test to probe an
rieval, in the absence of any natural sensory ing, auditory fear conditioning, and novel engram. The well-learned maze task Lashley
retrieval cues. Fourth, mimicry studies should object exploration). Control studies revealed typically used could be solved using different
artificially introduce an engram of an experi- that tagged cells were only reactivated by the strategies and, therefore, may have been in-
ence that never happened into the brain and corresponding conditioned stimulus and not sensitive to damaging a distinct brain region.
show that rodents use the information of an by stimuli unrelated to the training experience Second, Lashley may have targeted the wrong
artificial engram to guide behavior. (45). Although most observational studies did brain region for this type of spatial memory
Memory traces, or at least physiological cor- not address directly the enduring, learning- task (68).
relates of memory, have been examined in in- induced changes hypothesized by Semon, over-
vertebrate species, such as flies (23–27), octopus all, these results (and their notable consistency Gain-of-function studies
(28, 29), Aplysia sea slugs (30, 31), honey bee across methods, tasks, and labs) provide broad Gain-of-function studies attempt to induce
(32), and Hermissenda sea slugs (33). More- support for the existence of engrams. How- memory retrieval in the absence of natural
over, pioneering studies in mammals (34–36) ever, causal studies are necessary to show that retrieval cues by artificially reactivating en-
greatly informed our current understanding of these reactivated putative engram cells indeed gram cells. Tonegawa and colleagues (69)
the neural basis of memory but did not ex- function as part of the internal representation provided the first gain-of-function evidence
amine memory at the cell ensemble level. The of an experience. for the existence of an engram. Hippocampal
discussion here is limited primarily to rodent dentate gyrus (DG) neurons active during con-
experiments examining memory of an explicit Loss-of-function studies textual fear conditioning (in which a context
experience that probe memory at the level of Loss-of-function studies attempt to “capture” was paired with a footshock) were tagged (41)
an engram. engram cells and specifically disrupt their func- to express the excitatory opsin channelrhodop-
tion before a memory test. Josselyn and col- sin 2 (ChR2) (70). When tested in a nontraining
Observational studies leagues (58) performed the first loss-of-function context, mice did not freeze. However, photo-
Typically, observational studies take advantage memory study at the level of a cell ensemble. stimulation of tagged engram cells was suffi-
of immediate early genes (IEGs) such as c-Fos, An allocation strategy was used to capture cient to induce freezing, the learning-specific
Arc (activity-regulated cytoskeleton-associated putative engram cells in the amygdala lateral conditioned response (44), even though mice
protein), or Zif268 (zinc finger protein 225) nucleus (LA) supporting an auditory fear con- had never been shocked in this nontraining
(37–39) to visualize active neurons. Cells active ditioned memory in mice. That is, a small, context (Fig. 2). Importantly, light-induced
during a memory test are marked using IEG random population of LA neurons was biased freezing was not due to activation of pre-
immunohistochemistry, whereas cells active for inclusion (or allocation) into a putative en- wired learning-independent neural circuits or
during a training experience are “tagged” gram using a neurotropic virus expressing a simple reflex response, because similar photo-
through the use of temporally inducible IEG CREB (Ca++/cyclic AMP–responsive element- stimulation of tagged DG neurons failed to
promoters that drive the expression of more binding protein). CREB is a transcription fac- induce freezing if downstream CA1 neurons
enduring fluorescent (or other) reporter pro- tor that increases both neuronal excitability were silenced during training (thereby prevent-
teins (40–43). Above-chance overlap between (59–64) and dendritic spine density (60, 65). ing learning) (71).
these two cell populations (“active during train- Therefore, neurons infected with this CREB Artificial optogenetic or chemogenetic (72, 73)
ing” and “active during test”) within a brain vector were hypothesized to be biased for in- reactivation of tagged or allocated engram cells
region (or throughout the brain) is suggestive clusion into an engram. A virus expressing across several brain regions similarly induced
of an engram. both CREB (to allocate neurons) and an indu- memory expression without external sensory
In an initial observational study designed to cible construct that produces cell-autonomous retrieval cues in a variety of tasks (42, 53, 74–81).
examine a memory at the level of a cell en- ablation was used to specifically kill allocated Therefore, artificial engram cell reactivation

Fig. 1. Engram loss-of-function studies disrupt subsequent memory during the fear memory test (green filled circles), suggesting that allocated
retrieval. (A) Ablating allocated neurons. Lateral amygdala principal (excitatory) neurons are preferentially recruited to an engram supporting this conditioned
neurons were experimentally allocated to an engram (blue circles) by means of fear memory. Specifically ablating experimentally allocated neurons (red circles)
overexpression of the transcription factor CREB (122). Mice received auditory before a second memory test disrupts memory retrieval. (B) Ablating a similar
fear conditioning during which a tone (conditioned stimulus) was paired with a number of random, nonallocated neurons does not disrupt memory retrieval.
footshock (unconditioned stimulus). The majority of allocated neurons are active [Images: Adapted from (122)].
serves as a sufficient retrieval cue to “reawaken” been shocked in this context. That is, mice brain as mice explored a new context (context A).
a dormant engram to induce memory expres- retrieved an artificial memory. Mice also froze Chemogenetically reactivating these neurons
sion, similar to Semon’s original definition of in context B (showing natural memory retriev- while mice were fear conditioned in context
ecphory [“the influences which awaken the al), but not in a third distinct context (context B produced a “hybrid or synthetic” context
mnemic trace or engram out of its latent state C), indicating freezing was a context-specific, representation that was not retrievable by
into one of manifested activity” (5, p. 12)]. and not a generalized, response (46). Both either context alone [unlike (46), above]. How-
memories produced by “natural” and “arti- ever, mice froze in a test session that more
Mimicry experiments ficial” means could only be retrieved by their closely matched the training conditions (place-
During natural memory retrieval, the sensory respective conditioned stimuli, indicating both ment in context B while chemogenetically ac-
conditioned stimulus (e.g., the training con- memories retained their identities. Similar tivating context A engram cells), suggesting
text) is thought to reactivate engram cells to to a compound conditioned stimulus in which that this hybrid memory incorporated both
induce memory retrieval. The first gain-of- both a tone and light predict footshock, the natural and artificial cues. Differences in the
function study (69) was designed to mimic strength of the natural and artificial mem- spatial and temporal properties of artificial
this retrieval process by directly reactivating ories were roughly 50% of a single “normally engram reactivation (more acute optogenetic
engram cells by means of optogenetic stimu- induced” memory, suggesting cue competi- activation of localized tagged DG neurons
lation, thus circumventing the need for the tion between the natural and artificial condi- versus longer-term chemogenetic activation of
conditioned stimulus. That is, artificial stim- tioned stimuli [as originally described by (82)]. nonlocalized tagged neurons across the brain)
ulation replaced the natural conditioned stim- Therefore, when a biologically important event may account for the discrepant outcomes of
ulus to induce memory retrieval. Optogenetic (e.g., footshock) occurs while an animal is these two artificial conditioned stimuli studies.
stimulation of engram cells has also been used retrieving a previously formed but perhaps Neurons active during presentation of an
to artificially retrieve a previously experienced unrelated memory, the two stimuli can be as- unconditioned stimulus have also been tagged
sensory stimulus during the formation of a sociated to form a new but false episodic mem- and artificially reactivated (85). Neuronal en-
new memory. For instance, DG neurons ac- ory. An analogous mechanism may underlie sembles active during context exploration (the
tive during exploration of a new context (con- human false memories, except that in humans, conditioned stimulus) and footshock (the un-
text A) were photostimulated when mice later the previously acquired memory would be re- conditioned stimulus) were tagged separately
received footshocks in a different context (con- trieved by natural processes (83). in the CA1 subfield of the hippocampus and
text B). During a memory test, mice replaced Mayford and colleagues (84) used a similar the basolateral complex of the amygdala, re-
in context A froze, even though they had never approach but tagged active neurons across the spectively. Synchronous optogenetic activation

of these ensembles while mice were in the

homecage was sufficient to induce a false mem-
ory; mice froze in the tagged (but nonshocked)
A B context, as if the conditioned stimulus and
unconditioned stimulus had been paired.
Finally, a recent study investigated whether
47
3n
a memory could be implanted through artifi-
m
cial means in the total absence of natural
stimuli (either conditioned stimulus or uncon-
ditioned stimulus). To be a true memory im-
plantation, such an experiment should satisfy
several criteria (86). First, the “learning expe-
rience” should occur entirely within the brain
through, for example, direct stimulation of pu-
tative conditioned-stimulus and unconditioned-
C D stimulus neural pathways. Second, the presence
of the implanted memory should be probed
40
through presentation of a “real” external re-
trieval cue (not just the internal neural cue).
Freezing (%)
30
Finally, behavioral manifestation of this mem-
ory should reflect the predicted memory con-
20
tent and be retrieved only by the “trained”
conditioned stimulus (not to similar cues).
10
In this study, optogenetic stimulation of a
genetically specific olfactory glomerulus (the
0
Off On Off On conditioned stimulus) was paired with opto-
genetic stimulation of either appetitive or
E aversive neural pathways (the unconditioned
stimuli) (86). After this entirely intracranial
conditioning, mice showed either an attrac-
tion or aversion, respectively, to the real odor
that activated this olfactory glomerulus. In
short, a memory was made in the absence of
experience. These results satisfy the mimicry
criterion of experimental evidence outlined
by Martin and colleagues (21, 22) and, as such,
provide another line of persuasive evidence for
the existence of engrams.
Understanding memory through engrams

The “enduring changes” of an engram
The ability to label in vivo engram cells sup-
porting a specific memory provided an op-
portunity to investigate the nature of the
Fig. 2. Gain-of-function method for engram identification and distributed engram ensembles. “enduring changes” proposed by Semon. Guided
(A) A c-fosÐtTA transgenic mouse is injected with AAV9-TRE-ChR2-mCherry (allowing active neurons by Hebb’s influential theory on the critical
in the absence of doxycycline to express the excitatory opsin ChR2) and implanted with an optical fiber importance of synaptic plasticity (the increase
to target blue light to activate ChR2-expressing neurons in the DG. (B) Basic experimental scheme. Mice in synaptic strength between neurons) in mem-
are habituated to context A with light stimulation while on doxycycline for 5 days and are then taken off ory [e.g., (21, 22)], Tonegawa and colleagues
doxycycline for 2 days (to open the tagging window) and exposed to contextual fear conditioning (CFC) in showed that learning augmented synaptic
context B. Mice are put back on doxycycline (to close the tagging window) and tested for 5 days in context A strength, specifically in engram cells. First, 1 day
with light stimulation. (C) Representative image showing the expression of ChR2-mCherryÐpositive (red) after training, hippocampal DG granule en-
engram cells in a mouse that was taken off doxycycline for 2 days and underwent CFC training. [Image credit: gram cells tagged during contextual fear con-
X. Liu and S. Ramirez (Tonegawa lab)] (D) Mice expressing ChR2 in engram cells from CFC in context B (red) ditioning showed greater synaptic strength
show greater freezing during test light-on epochs in context A than a control group expressing mCherry only. [higher AMPA/NMDA ratio, which is a means
Error bars indicate standard error of the mean. [Graph: Adapted from Liu et al. (69)] (E) A part of the engram of assessing basal strength of excitatory synap-
cell ensemble complex for contextual fear memory. It is generally thought that the engram for a specific ses by examining the relative expression of
memory is distributed in more than one brain region. For instance, for contextual fear memory, the engram amino-3-hydroxy-5-methyl-4-isoxazole pro-
cell ensemble in the entorhinal cortex layer II (EC-II) as well as hippocampal subfields [DG, CA3, CA2, CA1, pionic acid receptor (AMPAR)–mediated sy-
and subiculum (Sub)] may represent context, whereas amygdala engram cell ensembles represent fear naptic currents to N-methyl-D-aspartate receptor
information. These engram cell ensembles are functionally connected to form an engram cell ensemble (NMDAR)–mediated synaptic currents of a
complex. Thus, a concept has emerged that a specific pattern of cellular connectivity within an engram cell population of stimulated synapses (87)] and
ensemble complex serves as the substrate for a specific memory. US, unconditioned stimulus; LA, lateral increased spine density at entorhinal cortex
nucleus of the amygdala; BLA, basolateral nucleus of the amygdala; CS, conditioned stimulus. junctions than nonengram DG cells (71).

Second, compared to nonengram CA3 cells, to examine overlap between neurons active at wave ripples), tending to co-occur with rhythmic
downstream CA3 engram cells were more contextual fear training and testing, Luo and firing of cortical neurons (termed spindles)
functionally connected with upstream DG en- colleagues (42) showed that retrieval of a re- (114). Disrupting either sharp-wave ripple–
gram cells (71). Moreover, Kaang and colleagues mote (14 day) contextual fear memory engaged spindle coupling (115, 116) or sharp-wave ripple–
showed that the number and sizes of spines more neurons in prelimbic cortex than retrieval associated replay of hippocampal place cells
on CA1 engram cells tagged during contex- of a recent (1 day) memory, suggesting that (104, 105, 117, 118) impairs memory recall. The
tual fear conditioning receiving input from an engram changes over time [consistent with precise role of these rhythmic oscillations with
CA3 engram cells was greater than on non- the findings of (100)]. Finally, a preliminary respect to engram cells is unclear. Sharp-wave
engram CA1 cells. This enhanced interregional study (99) mapped candidate engram ensem- ripples promote synaptic depression of CA1
connectivity between CA3 and CA1 engram bles representing a contextual fear condition- hippocampal neurons (119, 120). A recent study
cells correlated with memory strength and ing memory in 409 brain regions in mice. suggests that CA1 engram cells tagged during
occluded long-term potentiation (LTP), sug- Roy and colleagues tagged cells active at train- context exploration are more likely than non-
gesting a previous LTP-like phenomenon endo- ing and those active at recall throughout the engram neurons to participate in sharp-wave
genously occurred (88). Similarly, LA engram brain in the same mouse using a CLARITY- ripple events, perhaps allowing these engram
cells tagged during auditory fear conditioning like tissue-clearing technique (101) dubbed cells to escape this SWS-induced synaptic de-
showed enhanced synaptic connectivity with SHIELD (stabilization under harsh conditions pression (120). In this way, postencoding re-
presynaptic neurons (56, 89). Finally, shrink- via intramolecular epoxide linkages to prevent activation of engram cells during oscillatory
ing potentiated synapses in primary motor degradation) (102), thereby permitting the en- rhythms may help refine an engram by de-
cortex (M1) engram cells supporting a motor tire intact brain to be imaged at once. From creasing irrelevant “noise” of nonengram neu-
memory disrupted subsequent performance this activation data, these researchers devel- ronal activity during memory consolidation.
of this, and not a similar, motor memory (90). oped an “engram index” (defined as the degree
Together, these studies are beginning to inte- to which cells in a given brain region were Lifetime of an engram
grate previous research on synaptic plasticity active at memory encoding and retrieval) Birth of an engram
with engrams and suggest preferential en- that allowed the rank ordering of different Josselyn, Silva, and colleagues discovered that
gram cell–to–engram cell connectivity is a crit- brain regions. Using optogenetic and chemo- during engram formation, eligible neurons
ical part of the enduring changes to an engram genetic methods to interrogate the effects of in a given brain region compete against each
generated by learning. Overall these findings artificially activating regions with a high en- other for allocation (or recruitment) to an en-
suggest an update of Hebb’s axiom: Engram gram index, this study showed many of these gram. Neurons with relatively increased intrin-
cells that fire together, wire together. engram ensembles are functionally connected sic excitability win this allocation competition
and activated simultaneously by an experience. to become engram cells (58, 63, 66, 76, 77, 121–126)
Distributed engram ensembles These findings suggest that an experience is (Fig. 3). Competitive excitability-based alloca-
Although one specific brain region is often ex- represented in specifically connected multiple tion to an engram occurs in other brain re-
amined in engram studies, it is generally ap- engram ensembles distributed across multiple gions and supports different types of memories
preciated that an engram supporting a specific brain regions and provide experimental sup- [e.g., dorsal CA1 region of hippocampus (91–93)
experience may be widely distributed through- port for Semon’s “unified engram complex” and prefrontal cortex (126) (for a contextual
out the brain. Engram cell ensembles in dif- hypothesis. fear memory), insular cortex (127) (conditioned
ferent brain regions may support distinct taste-aversion memory), and retrosplenial cor-
aspects of an experience. For instance, in Engrams, place cells, and sleep tex (128) (spatial memory)].
contextual fear memory, hippocampal (DG, Location-specific firing of CA1 place cells is In addition to aversive memories, LA neu-
CA3, and CA1) engram cell ensembles may well established (103). Stable place cells may rons experimentally made more excitable during
represent the context (40, 48, 91–93), whereas be important in engrams supporting spatial training were also preferentially allocated to
amygdala engram cell ensembles may repre- or contextual memories (104–106). Recently, an engram supporting a cocaine-cue reward-
sent valence information (69, 71, 75), and cor- McHugh and colleagues (107) contrasted the ing memory (66). Similarly, increasing the ex-
tical engram cell ensembles may represent roles of CA1 place cells and engram cells in citability of a small, random portion of piriform
distinct sensory information (79, 94–96). memory. While mice explored a new context, cortex principal neurons resulted in their allo-
Several studies have examined potential engram cells were tagged and place cells iden- cation to an engram supporting either a re-
engram cell ensembles supporting contextual tified using tetrode recordings. Most tagged warding or an aversive olfactory memory,
fear memories across the brain (42, 97–99). engram cells were also place cells, but the depending on the nature of the training ex-
For instance, Frankland and colleagues com- majority of place cells were not tagged. Non- perience (129). Excitability-based neuronal allo-
pared the brainwide (84 brain regions) dis- tagged place cells behaved like traditional cation is predicted by computational modeling
tribution of active cells after retrieval of recent place cells (stable in the same context but re- (130–132), occurs endogenously (56, 89), and is
(1 day after training) versus remote (36 days mapping in a new context). By contrast, tagged consistent with previous research implicating
after training) contextual fear memory. On place cells fired in a context-specific manner, intrinsic excitability in the formation of inver-
the basis of coactivation, graph theory was albeit with imprecise spatial information, and tebrate memory traces (33, 133–135). Together,
used to construct functional connectome “mem- were not active (did not remap) in a new con- these findings suggest that in some brain re-
ory maps” (97) and identify hub-like regions text. Therefore, engram cells may provide gen- gions, at any given time, a small portion of
hypothesized to play privileged roles in mem- eral contextual information, with nontagged eligible neurons are “primed” to become part
ory retrieval. Subsequent chemogenetic in- place cells providing precise spatial information. of an engram (should an experience occur),
hibition confirmed that these identified hub Postencoding reactivation or replay of hip- regardless of experience valence.
regions were necessary for subsequent mem- pocampal place cell firing, especially during Although stable place cells and engram cells
ory retrieval (98). Using a combination of slow-wave sleep (SWS) (108, 109), is thought in dorsal CA1 of the hippocampus differ (107),
engram tagging technology [targeted recom- to be important for memory consolidation some mechanisms underlying their formation
bination in active populations 2 (TRAP2) trans- (110–113). During SWS, hippocampal neurons may be shared. In a given environment, only
genic mice] and IEG immunohistochemistry fire in an oscillatory rhythm (termed sharp- a small subset of CA1 neurons are place cells,

size (89, 122) [for review, see (146)]. Rather, a

stronger memory engages a greater number
of synapses between engram cells (88).
Several lines of evidence suggest that one
mechanism constraining engram size involves
inhibitory neurons. Thus, inhibiting parvalbumin-
containing interneurons in the basolateral
amygdala complex increased the size of an
engram in the LA supporting an auditory
fear memory through a process involving di-
synaptic inhibition (145), in which an excitatory
neuron inhibits another excitatory neuron via
an intervening inhibitory neuron. Moreover,
inhibiting somatostatin-containing interneu-
rons increased the size of a DG contextual fear
memory engram through a lateral-inhibition
like process (147). The importance of inhibitory
neurons in engrams has also been highlighted
in human studies. For instance, evidence sug-
gests that in the cortex, associative memories are
represented in excitatory engrams and matched
(equal and opposite) inhibitory engrams. Mem-
ories are expressed upon disinhibition of the
excitatory engram (148–150). Further explora-
tion of excitatory-inhibitory balance in engram
formation, storage, and retrieval is necessary
to understand how these opposing forces in-
teract to support memory function.
Silent engrams in memory loss

Engrams may become damaged, such that a
memory becomes forever unavailable. How-
ever, engrams may also be temporarily inac-
cessible, such that the engram still exists but
cannot be retrieved by natural means. Silent
Fig. 3. Neuronal allocation to an engram. Eligible neurons compete for allocation to an engram engrams, engrams that cannot be retrieved
supporting a memory, and neurons with increased relative excitability at the time of training “win” by natural retrieval cues but can be retrieved
this competition for allocation. (A) Neurons that were endogenously more excitable than their with direct optogenetic stimulation, were first
neighbors at the time of training or were experimentally manipulated to become relatively more excitable revealed in an experiment in which the pro-
(blue circles) are preferentially allocated to an engram (green filled circles). Subsequent disruption tein synthesis inhibitor, anisomycin, was ad-
of these allocated or engram neurons disrupts memory retrieval (top right), whereas artificial ministered immediately after contextual fear
reactivation of these neurons elicits memory retrieval in the absence of normal sensory retrieval conditioning in mice (71). Inhibiting protein
cues (bottom right). (B) Neurons with relatively decreased excitability at the time of training (either synthesis before or immediately after an ex-
endogenously or through experimental manipulation) (purple circles) are preferentially excluded perience is known to induce amnesia (151, 152)
from the engram (green filled circles). Subsequent disruption of nonallocated or nonengram neurons and block cellular consolidation (153–155). Cel-
does not impact memory retrieval. lular consolidation refers to the relatively fast
process of memory stabilization thought to
involve the expression of genes necessary to
because the majority of CA1 neurons are silent persist. Both allocation and selective stabiliza- strengthen synapses. By contrast, systems con-
(136). Those neurons with relatively higher ex- tion resonate with the conceptual framework solidation (discussed below) refers to the slower,
citability immediately before placement in a of Darwinian competition. time-dependent reorganization of memories
novel environment are more likely to become Observational and tagging experimental stud- over distributed brain circuits (156–159).
place cells in that environment (137–139), and ies agree with computational theories [e.g., In this study, mice administered anisomy-
experimentally increasing the excitability of (144)] that an engram is sparsely encoded. cin immediately after training showed little
an initially silent cell biased this cell toward That is, not all neurons within a given brain freezing when replaced in the training con-
becoming a place cell (140, 141). region become an engram cell supporting a text 1 day later (71). Therefore, as expected,
It is interesting to note the similarities be- particular memory. The size of an engram disrupting protein synthesis induced retro-
tween data from current allocation studies and within a given brain region (that is, the num- grade amnesia by blocking cellular consolid-
the long-standing idea of selective stabilization ber of engram cells) is stable and invariant to ation. However, optogenetic reactivation of DG
(142, 143). Selective stabilization proposes memory strength. For instance, the size of an engram neurons tagged during contextual fear
that multiple prerepresentations are endog- LA engram (number of LA engram cells) is training was sufficient for memory recovery,
enously generated in the brain and only one or similar for an auditory fear conditioned mem- even 8 days after training (Fig. 4). These re-
a few that fit the situation are selected at any ory and a cocaine-cue memory (66, 122, 145), sults indicate that the engram was formed and
given point of time to control behavior and/or and memory strength does not affect engram persisted for several days but that this engram

variant of an auditory fear conditioning task

in which the tone conditioned stimulus was
replaced by optogenetic activation of LA axon
Non engram cell Silent engram cell Reactivated silent engram cell Active engram cell Reactivated active engram cell
terminals from neurons originating in the
A Active Engram Silent Engram medial geniculate nucleus and auditory cor-
In amnesia and early Alzheimer’s disease

Natural recall Natural recall Artificial recall tex. Immediately after conditioning, long-term
(ChR2 + blue laser) depression (LTD)–like optogenetic stimula-
Context A Context A Context B tion was administered. LTD is thought to
weaken synaptic efficacy and decrease spine
density (162–165). Consistent with the inter-
pretation that LTD-like stimulation silenced
Spine density Spine density Spine density the engram, this opto-LTD stimulation im-
HIGH LOW LOW paired subsequent memory recall. However,
LTP-like optogenetic stimulation allowed the
memory to be retrieved (consistent with the
interpretation that the engram was “unsi-
lenced”). Again, subsequent LTD-like opto-
Freezing No Freezing Freezing genetic stimulation silenced this memory,
whereas LTP-like optogenetic stimulation al-
B Basal level Recent natural recall Remote natural recall
lowed recovery of this memory.
These findings raise the question of whether
engrams (and the memories they support) in
other amnesic conditions are truly “lost” or are
Hippocampus
CFC
Dox(-) Dematuration simply inaccessible such that they cannot be
During Systems Consolidation
retrieved under natural conditions. Silent en-

grams were reactivated by artificially stim-
ulating engram cells in amnestic mice used to
Spine density: HIGH Spine density: LOW study the early stages of Alzheimer’s disease
(AD) (166, 167). These transgenic mice [APP/
PS1 mice containing human transgenes with
the familial AD mutation in both amyloid pre-
CFC cursor protein (APP) and presenilin 1 (PSEN1)]
Dox(-) Maturation
mPFC
showed contextual fear memory deficits (166).

However, optogenetic reactivation of ChR2-
labeled DG engram cells induced robust freez-
ing comparable to control mice (166). Consistent
Spine density: LOW Spine density: HIGH with other examples of silent engram cells,
No Freezing Freezing Freezing DG engram cells in these mice used to study
AD showed decreased spine density. However,
Fig. 4. Active and silent engram cells in amnesia and during memory systems consolidation.
LTP-like optogenetic stimulation at entorhinal
(A) Active engram cells have higher spine density and are activated in the conditioned context A to
cortex engram cell inputs onto DG engram
produce the conditioned response, freezing. Silent engram cells generated in amnesia and in a mouse designed
cells restored not only spine density in DG en-
to model early Alzheimer’s disease show lower spine density and cannot be activated in the conditioned
gram cells but also the ability of natural re-
context A to produce a conditioned response but can be activated by blue light in an unconditioned context B if
trieval cues to elicit memory retrieval (thereby
they were tagged with ChR2 during encoding. (B) During memory systems consolidation, active engram
unsilencing the engram) (166). These findings
cells with high spine density are formed in the hippocampus during contextual fear conditioning and for several
in mice are consistent with reports that mem-
days, the conditioned context can evoke a conditioned response. However, by two weeks (remote recall),
ory retrieval in people with early-stage AD
these hippocampal engram cells demature to become silent, with reduced spine density. In the mPFC, engram
may be enhanced by particular retrieval cues
cells are formed during CFC but are silent with low spine density. During the following 2 weeks, these mPFC
(168, 169). Therefore, under certain conditions,
silent engram cells acquire higher spine density and become active engram cells.
a previously inaccessible memory may be re-
trieved in human AD, consistent with the in-
terpretation that some engrams in early-AD
could not be retrieved by natural means. Silent structural plasticity within engram cells. Con- brains may be silent rather than lost.
DG engram cells showed weaker physiological sistent with this, genetic restoration of spine Apart from clinical implications, the finding
(increased synaptic strength) and structural density [targeted overexpression of p-21 acti- of silent engrams is relevant to discussions
(increased dendritic spine density) alterations vated kinase (PAK 1)] also allowed a silent en- on the role of protein synthesis–dependent
than normal engram cells (in control mice), gram to be reactivated and memory expressed cellular consolidation in terms of memory stor-
suggesting that a silent engram may be the by natural retrieval cues (160). age versus retrieval. There has been persistent
result of disrupting the synaptic strengthening The idea that engrams may be silenced by debate on this issue (170–173). The majority
normally induced by training. That optoge- disrupting synaptic efficacy and spine den- of neuroscientists examining cellular memory
netic activation of DG engram cells was able sity and reawakened by enhancing synaptic consolidation may favor the view that disrupt-
to induce memory retrieval suggests that di- plasticity is consistent with findings from a ing protein synthesis disrupts memory storage.
rect optogenetic activation was able to cir- nonengram study examining auditory fear However, in many amnesia experiments, mem-
cumvent this requirement for synaptic and conditioning (161). Rats were trained in a ory storage is conflated with memory retrieval.

The finding that optogenetically stimulating a LA neurons, that is, the reversal of synaptic DG neurons active during training remains an
silent engram in an otherwise amnestic mouse, potentiation induced by fear conditioning open question (40, 51).
even 1 week after training, induces memory re- (179, 180). Moreover, after fear conditioning, Finally, a recent study examined fear ex-
trieval challenges the view that protein synthesis– LTD-like electrical stimulation of external cap- tinction engrams in the amygdala and found
dependent cellular consolidation is important sule inputs to the LA induces synaptic depot- that extinction engram cells were formed
for memory storage. Instead, these findings entiation and decreases fear behavior (181), in a genetically distinct and “reward-responsive”
suggest that the role of cellular consolidation resembling both extinction and engram silenc- subpopulation of basal amygdala neurons.
is to enhance subsequent retrievability of an ing. Finally, shortly after extinction training, These fear extinction engram cells suppressed
engram, consistent with the idea of engram the chemogenetic artificial activation of cells the fear engram neurons that were also present
“retrieval handles” that are established after tagged brainwide during context fear training in basal amygdala and, furthermore, induced
memory formation and may be remodeled (the putative fear engram) was reported to in- appetitive behavior when optogenetically stim-
after memory retrieval (1). Importantly, silent crease freezing levels (182), suggesting that the ulated (195). These findings in mice are con-
engrams are consistent with the pioneering original fear engram was silenced during ex- sistent with the results of a recent study in
cognitive psychologist Endel Tulving’s (174) tinction. The similarities between engram fruit flies (26) and highlight the similarities
conceptual distinction between memory avail- silencing and extinction are consistent with between fear extinction and reward processes
ability and accessibility, in which memory fail- theoretical views that during extinction, the across species. Moreover, these results are
ure may reflect the absence of the information conditioned stimulus–unconditioned stimulus consistent with the general idea of competi-
or difficulties accessing the information [see contingency is “unlearned” (183, 184). tion between memory traces in the control
(175) for review]. However, other accounts stress that extinc- of behavior.
tion does not reflect unlearning the original
Silent engrams in normal memory association (perhaps by silencing the original Silent engrams and time
Memory may change with time and circum- engram) but rather reflects learning a new The representation of a memory in the brain
stance. Might these changes in memory be “conditioned stimulus–no unconditioned stim- may change with time. For instance, dorsal
mediated by endogenous engram silencing? ulus” association (185, 186) with a correspond- hippocampal lesions in rodents disrupt ex-
This was explored in a social discrimination ing new extinction engram. That the original pression of contextual fear memories in the
task in which mice interact more with a new, memory is not “erased” by extinction is sug- days, but not weeks after training (196–198).
rather than a familiar, mouse. This social dis- gested by findings that after extinction train- At more remote times, cortical areas, includ-
crimination memory lasts roughly an hour ing, the conditioned response may return if ing anterior cingulate cortex or medial pre-
after exposure to a familiar mouse (the train- the conditioned stimulus is presented (i) in a frontal cortex (mPFC), become preferentially
ing experience) and is absent 24 hours after new nonextinction context (renewal), (ii) after engaged (100). The time-dependent reorgan-
training (176). The dorsal CA2 to ventral CA1 a stressor (reinstatement), or (iii) after the pas- ization of memory reflects systems consoli-
(vCA1) hippocampal circuit plays a pivotal role sage of time (spontaneous recovery) (187–192). dation, a process that typically refers to initially
in social discrimination (177), with a vCA1 en- A recent study concluded that contextual fear hippocampal-based episodic-like memories
gram representing the familiar mouse (178). extinction may be supported by a novel fear (158, 159). Systems consolidation was recently
Consistent with the time course of social dis- extinction engram in the DG that is distinct examined at the level of the engram in the
crimination memory, the familiar mouse en- from and suppresses the contextual fear DG hippocampus and mPFC, where findings in-
gram in vCA1 becomes silent an hour after engram with a time course that corresponds dicate time-dependent silencing of active en-
training. However, artificially reactivating this to the emergence of spontaneous recovery grams and conversions of silent engrams to
engram 24 hours after training (when the (53). In this experiment, spontaneous recovery active engrams (51, 199). During contextual
social discrimination memory normally has was observed remotely (29 days), but not re- fear conditioning, active mPFC neurons were
dissipated) reinstates social discrimination cently (6 days), after extinction training. More- labeled to express ChR2. When placed in the
memory, as if the trained-but-forgotten famil- over, the original fear engram was reactivated conditioning context, mice showed robust
iar mouse is being remembered. Besides arti- at the remote, but not recent, memory test freezing when tested either 2 days or 13 days
ficial engram reactivation, the accessibility after extinction training. The opposite pat- after training. However, the engram ensemble
of vCA1 engram (and social discrimination tern of results was observed for active cells components supporting memory retrieval
memory) is prolonged by interventions such tagged after extinction training (the presumed differed with test time. Tagged mPFC neurons
as group housing. These findings provide a fear extinction engram). Interestingly, artifi- were reactivated 13 days, but not 2 days, after
hint that engram silencing may be one way in cial reactivation of the fear extinction engram training, suggesting that the mPFC engram
which the brain normally regulates mnemonic prevented spontaneous recovery of the origi- was silent shortly after training but active after
processes. nal fear memory, even at remote times. These longer delays. DG engram cells showed an
Additional evidence comes from memory ex- results suggest that the original fear engram opposite pattern; DG engram cells were reac-
tinction studies. After conditioning, repeated and the extinction engram compete for con- tivated shortly after training but silenced more
presentation of the conditioned stimuli alone trol over behavior; the extinction engram first remotely. Similar to other instances of silent
(in the absence of the unconditioned stimulus) suppressed or silenced the original fear engram, engrams discussed above, the mPFC engram
produces a gradual decrease of the condi- but, with time, the fear extinction engram was cells shortly after training and the DG engram
tioned response (82)—a phenomenon called itself silenced. Conversely, activation of a re- cells at longer delays after training showed re-
extinction. Therefore, after extinction training, mote DG contextual fear engram (labeled duced spine density, and, furthermore, opto-
the ability of the conditioned stimulus to in- 25 days after contextual fear conditioning) genetic activation of these silent engrams was
duce memory retrieval is diminished, an out- itself may also be important for subsequent sufficient to induce memory retrieval. Inter-
come that is similar phenomenologically to fear memory extinction (52), perhaps sim- estingly, posttraining tetanus toxin–induced
engram silencing. Might engram silencing ac- ilar to a process referred to as reconsolidation- inhibition of the input from DG engram cells
count for extinction? Consistent with this updating (193, 194). However, the extent to to mPFC engram cells blocked the maturation
general idea, some auditory fear extinction which DG neurons that were activated 25 days of the silent mPFC engram cells to an accessible
protocols induce synaptic depotentiation of after contextual fear conditioning overlap with state, suggesting coordinated network function

between different engram ensemble components autophagy made the engram unavailable rather ilar memories (two fear memories or two
is important in systems consolidation. than simply inaccessible. contextual memories), two aversive, but oth-
Memories may also become less precise and erwise dissimilar memories (a conditioned
more generalized with time (200–202). Ac- From engrams to knowledge fear and a conditioned taste aversion mem-
cording to memory transformation theory, Thus far, we have discussed engrams support- ory), were integrated by repeated coretrieval
changes in the nature and quality of memories ing a single memory. Of course, animals (in- of these memories (220). Overall, these data
correspond to changes in neural representa- cluding humans) learn and remember many from rodent experiments agree with results
tions, with hippocampal-dependent context- things. Some of these experiences may be best from human memory experiments showing
specific detailed memories transforming into remembered as distinct episodes, rich with epi- that the representations of memories for events
gist-like schematic memories represented in sodic details (207–209). However, in other cir- experienced close in time or with related con-
cortical structures over time (201, 203, 204). cumstances, it may be advantageous to link tent overlap may be integrated or linked, thus
The neural processes governing remote mem- related experiences, thereby creating a gen- enabling generalization and flexible use of this
ory generalization at the engram level suggest eral concept or principle (210–214). This raises shared information [e.g., (212, 221–224)].
that the availability of the DG engram is crit- the question of how engrams representing Memory retrieval also transiently reactivates
ical for memory specificity (205). In this experi- different experiences interact. The mecha- engram cells (89, 215, 219). This increase in
ment, shortly after contextual fear conditioning nisms governing neuronal allocation to an excitability both enhances the precision and
(1 day), mice froze in the training context only, engram supporting a single experience also efficiency of memory retrieval (219) and opens
whereas at more remote time points (16 days serve to either coallocate neurons to overlap- a new “coallocation window” (215), perhaps
after training), mice also froze in a nonshocked ping engrams (thereby linking experiences) explaining how new information is integrated
context. This finding is consistent with previous or disallocate neurons to nonoverlapping en- into preexisting knowledge.
reports of contextual fear memory generalizing grams (thereby disambiguating experiences)
over time (51, 201). At the recent, but not re- (121, 215–217) (Fig. 5). In this way, relative Conclusions and perspectives
mote, time, DG engram cells showed greater neuronal excitability is critical not only for Overall, these studies provide persuasive evi-
connectivity to parvalbumin-expressing CA3 initial engram formation but also in organiz- dence for the existence of engrams in rodent
basket cells (thereby inhibiting CA3 pyramidal ing different memory representations across brains. We agree with Endel Tulving who
neurons through feedforward inhibition) than the brain. stated “As a scientist I am compelled to the
nonengram DG cells, suggesting that greater Neurons that are relatively more excitable conclusion—not postulation, not assumption,
feedforward inhibition in DG-CA3 circuits helps than their neighbors at the time of an exper- but conclusion—that there must exist certain
maintain memory precision. Interestingly, opto- ience are more likely to be allocated to the physical-chemical changes in the nervous tissue
genetic activation of DG engram cells 10 days engram supporting the memory of that ex- that correspond to the storage of informa-
after training did not induce memory retrieval perience (121). Increased excitability in engram tion, or to the engram, changes that constitute
(suggesting that this engram had become un- cells is also maintained for several hours after the necessary conditions of remembering. (The
available), except if feedforward inhibition of an experience (215, 218, 219). Therefore, if a alternative stance, that it may be possible for
CA3 pyramidal neurons was genetically en- related experience occurs in this time win- any behavior or any thought to occur indepen-
hanced. Moreover, mice with genetically endow, these same (or overlapping) engram dently of physical changes in the nervous sys-
hanced feedforward inhibition also showed cells are more excitable than their neighbors tem, as all your good readers know, is sheer
precise memory, even when tested at more and thus coallocated to the engram support- mysticism)” (225). The findings from many labs
remote times. Together, these data suggest that ing the memory of the second experience. using different methods to examine many types
enhanced feedforward inhibition onto CA3 Because the memories of the two experiences of memory converge to support the idea that
neurons maintains DG engram cell availability are coallocated to overlapping engram cells, complex information may not be represented
and delays the loss of context specificity asso- these two memories become linked (or inte- in single cells [e.g., a “grandmother cell”
ciated with remote memories. grated); thinking of one experience automat- (226, 227)]; instead, these findings suggest
These findings suggest that engram silenc- ically makes one think of the second. For that the basic unit of computation in the brain
ing may represent a continuum of a natural example, LA neurons allocated to one fear is an engram (228, 229).
state of an engram. That is, an engram may be memory were coallocated to a second fear To understand a complex, multilayered sys-
(i) unavailable (neither natural conditioned memory if the second event occurred minutes tem such as the brain, it is crucial to causally
stimuli nor artificial reactivation induces to hours (30 min to 6 hours), but not 24 hours, link a process or phenomenon occurring at a
memory expression), (ii) silenced (only arti- after the first (215). This linking occurred even lower level of complexity to those at higher
ficial reactivation is sufficient to induce mem- if the conditioned stimuli used in the two levels. Traditionally, such studies have been
ory expression), (iii) dormant or latent, as training sessions were of different modalities carried out using interventions such as tissue
initially named by Semon (natural condi- (e.g., a light and a tone or a context and a lesion or pharmacological disruption. Many
tioned stimuli may induce memory retriev- tone). Similarly, coallocation of CA1 engram of the studies discussed in this review took
al), or (iv) active (currently being retrieved). cells supporting memories of two distinct advantage of state-of-the-art intervention
Different processes may mediate these distinct contexts was observed if exposure to the con- techniques and their combinations, includ-
engram states. For example, similar to silenc- texts was separated by a short time interval ing temporally inducible targeted transgenics
ing a DG engram, posttraining anisomycin (216). Behaviorally extinguishing one memory and optogenetics, that may generally per-
administration silenced an LA engram sup- produced extinction for the second mem- mit the identification of more precise cause-
porting an auditory fear memory (79). How- ory, even though the second memory was consequence relationships. Nevertheless, even
ever, if in addition to anisomycin, a peptide not behaviorally extinguished, indicating that advanced interventions inevitably artificially
to induce autophagy (a mechanism of pro- the two memories were functionally linked manipulate the brain and therefore provide in-
tein degradation) was administered after (215). Coallocated memories may maintain formation as to what an engram can do, but not
training, then optogenetic reactivation of in- their distinct identity by engaging specific necessarily what it does do (physiologically).
puts to the LA was no longer sufficient to in- synapses within shared engram cells (79). This point has been articulated in several other
duce memory retrieval (206), suggesting that Moreover, in addition to integrating two sim- reviews on memory research [e.g., (230)].

signed to study the memory deficits of AD

hint at the extraordinary translational po-
tential of this line of research.
Some additional general themes emerge
from the results of engram studies. The first
theme is that findings from engram studies
are reminiscent of reconsolidation studies.
Upon retrieval, a memory may enter a labile
and modifiable state that lasts for several
hours. The process of restabilizing this mem-
ory is referred to as reconsolidation. Although
reconsolidation has a longer history (234), the
modern reawakening of this phenomenon
stems from a finding by Nader, LeDoux, and
Schafe (235). At the time that this ground-
breaking study was conducted, the general
thinking was that memories become stabi-
lized in a process of cellular consolidation that
occurs once, shortly after a learning experi-
ence. However, Nader, LeDoux, and Schafe
challenged this view by showing that memory
Fig. 5. Neuronal allocation and memory linking. Neurons with increased excitability at the time of event
retrieval opens a several-hour “reconsolidation
1 (blue) are allocated to the engram supporting this memory (blue filled circles outlined in orange).
window” during which different interventions
These allocated engram neurons remain more excitable than their neighbors for several hours after event 1. If
may weaken or strengthen the original mem-
a similar event 2 (green) occurs during this time, neurons allocated to the engram supporting event 1 are
ory. For instance, disrupting protein synthesis
more excitable and, therefore, also allocated to the engram supporting event 2 (blue and green filled circles
during the reconsolidation window of a condi-
outlined in orange). In this way, neurons are coallocated to events 1 and 2. By virtue of coallocation, these
tioned fear memory produced apparent am-
two memories become linked. After some time, neurons allocated to the engram supporting event 1 become
nesia for this memory. This result was replicated
less excitable than their neighbors (“refractory”), and if event 2 occurs in this time window, a new population
and generalized to several types of memory
of more excitable neurons wins the competition for allocation to the engram supporting event 2. This
(156, 236, 237). There are many similarities
disallocation allows the two memories to be remembered separately. Circles with red dashed outlines represent
between this reconsolidation blockade and
less excitable neurons.
engram silencing. For instance, reconsolidation
blockade is only observed when a memory is
being actively retrieved, because administer-
However, the results of these intervention memory, memory systems consolidation, and ing anisomycin (or another similar interven-
studies provide direction as to which processes perhaps fear extinction training) silences en- tion) in the absence of memory reactivation
we should focus our efforts to understand how grams such that they can no longer be ac- does not impair its subsequent retrieval. Viewed
the brain actually forms and retrieves mem- cessed by normal sensory cues. However, these from an “engram conceptual framework,” re-
ory. Furthermore, the high specificity of the studies show that silent engrams still exist in trieval of a specific memory would activate the
state-of-the-art intervention methods, spanning the brain and that the information they rep- underlying engram, and disrupting protein syn-
from the molecular level up to the behavioral resent may not be forever lost. The pioneering thesis shortly after this activation might silence
level, have already revealed mechanisms that psychologist and behaviorist Edward Tolman this engram. The function of reconsolidation
would have been difficult to study using other (231) advanced the concept of latent learning may be to update a memory (1, 211, 238–240).
techniques. For instance, these artificial inter- and latent memory: learning that occurs with- That the reconsolidation window is not unlike
vention studies allowed the field to identify out reinforcement, the memory of which is not the coallocation window suggests that these
the silent state of an engram and the mech- revealed or expressed until the need or mo- two processes might be similar ways of ex-
anism underlying memory allocation. tivation for the acquired knowledge arises plaining the same (or similar) phenomenon
More than 100 years ago, Semon put forth a (232, 233). It would be interesting to deter- at different levels of analysis.
law of engraphy. Combining these theoretical mine whether at least some latent memories A second emerging theme is that of com-
ideas with the new tools that allow researchers are based on silent engrams and, if so, use the petition. Allocation to an engram involves
to image and manipulate engrams at the level conversion of silent engram to active engram as competition between eligible neurons within
of cell ensembles facilitated many important a means of identifying and characterizing the a given brain region at the time of memory
insights into memory function. For instance, brain circuits mediating the relevant motivation. encoding. Competition represents a funda-
evidence indicates that both increased intrin- A continuum of engram accessibility states mental property of many biological systems
sic excitability and synaptic plasticity work may exist. Engrams may be entirely unavail- and has been previously shown to be im-
hand in hand to form engrams and that these able and not retrievable, even through artifi- portant in other mnemonic phenomena. For
processes may also be important in memory cial means (the memory would be forgotten). instance, memory traces may compete for con-
linking, memory retrieval, and memory con- Or, engrams may be silenced such that mem- trol of behavior at the time of retrieval (241).
solidation. Interestingly, disrupting synaptic ories may be retrieved by artificially reactivat- In addition, human studies reveal that mem-
plasticity in engram cells either by disease ing engram cells. The processes that silence or ories may compete if they are linked to a
processes (as in mice used to study AD) or erase an engram, as well as strategies for un- common retrieval cue. Retrieval of a target
amnestic drugs (such as protein synthesis in- silencing engrams, are a subject for further memory may lead to retrieval-induced for-
hibitors) or during some natural behaviors investigation. That it was possible to arti- getting of currently irrelevant competing mem-
(housing condition in social discrimination ficially reactivate silent engrams in mice de- ories (242).

Although recent engram studies have of- ries and organize information in the brain 24. D. Owald et al., Activity of defined mushroom body
fered important insights into memory, several may motivate the development of new algo- output neurons underlies learned olfactory behavior in
Drosophila. Neuron 86, 417–427 (2015). doi: 10.1016/
key questions remain. First, although the ma- rithms and AI architectures to better allow j.neuron.2015.03.025; pmid: 25864636
jority of observational studies reveal that the these agents to form generalizations and 25. E. Perisse et al., Aversive learning and appetitive motivation
overlap between populations of neurons active schema. In addition, machine learning and toggle feed-forward inhibition in the Drosophila mushroom
body. Neuron 90, 1086–1099 (2016). doi: 10.1016/
during training and testing exceed chance deep neural networks may inspire or generate j.neuron.2016.04.034; pmid: 27210550
levels, the overall correspondence between these testable theories at the level of the engram for 26. J. Felsenberg et al., Integration of parallel opposing memories
two populations is relatively low (roughly 10 neuroscientists to investigate. In this way, unit- underlies memory extinction. Cell 175, 709–722.e15 (2018).
to 40%, depending on the study). That this ing the foundational theories of AI pioneer doi: 10.1016/j.cell.2018.08.021; pmid: 30245010
27. T. Miyashita, E. Kikuchi, J. Horiuchi, M. Saitoe, Long-term
overlap does not approach 100% suggests a Alan Turing with those of Endel Tulving could memory engram cells are established by c-Fos/CREB
number of possibilities. First, the methods to benefit both AI and memory research. transcriptional cycling. Cell Rep. 25, 2716–2728.e3 (2018).
label active neurons using IEG promoters may doi: 10.1016/j.celrep.2018.11.022; pmid: 30517860
28. J. Z. Young, Computation in the learning system of
be imprecise (either “overtagging” or “under- RE FERENCES AND NOTES
cephalopods. Biol. Bull. 180, 200–208 (1991). doi: 10.2307/
tagging” the “real” engram at training and/or 1. Y. Dudai, The neurobiology of consolidations, or, how stable is
1542389; pmid: 29304688
the engram? Annu. Rev. Psychol. 55, 51–86 (2004).
testing). Alternatively, engrams may be dynam- 29. J. Z. Young, in Cephalopod Neurobiology: Neuroscience
doi: 10.1146/annurev.psych.55.090902.142050; pmid: 14744210
Studies in Squid, Octopus and Cuttlefish, N. J. Abbott,
ic, even over relatively short (days) periods of 2. D. L. Schacter, Constructive memory: Past and future.
R. Williamson, L. Maddock, Eds. (Oxford Univ. Press, 1995),
time, with cells “dropping into” or “dropping Dialogues Clin. Neurosci. 14, 7–18 (2012). pmid: 22577300
pp. 431–443.
3. D. L. Schacter, Stranger Behind the Engram: Theories of Memory
out of” the engram as it is refined or consolid- and the Psychology of Science (Erlbaum Associates, 1982).
30. A. Bédécarrats, S. Chen, K. Pearce, D. Cai, D. L. Glanzman,
ated (243, 244). It will be interesting to deter- 4. D. L. Schacter, J. E. Eich, E. Tulving, Richard Semon’s
RNA from trained Aplysia can induce an epigenetic engram
for long-term sensitization in untrained Aplysia. eNeuro 5,
mine how the mechanisms of engram silencing theory of memory. J. Verbal Learn. Verbal Behav. 17, 721–743
ENEURO.0038-0018.2018 (2018). doi: 10.1523/
(1978). doi: 10.1016/S0022-5371(78)90443-7
contribute to and/or interact with this refine- 5. R. Semon, The Mneme (G. Allen & Unwin, 1921).
ENEURO.0038-18.2018; pmid: 29789810
ment process and the implications this may 6. R. Semon, Die Mneme als erhaltendes Prinzip im Wechsel des 31. T. J. Carew, R. D. Hawkins, T. W. Abrams, E. R. Kandel, A test
organischen Geschehens, W. Engelmann, Ed. (Leipzig, 1904). of Hebb’s postulate at identified synapses which mediate
have on memory quality, precision, or strength. classical conditioning in Aplysia. J. Neurosci. 4, 1217–1224
7. R. W. Semon, Mnemic Psychology (G. Allen & Unwin, 1923).
Moreover, it will be important to determine 8. S. A. Josselyn, S. Köhler, P. W. Frankland, Finding the (1984). doi: 10.1523/JNEUROSCI.04-05-01217.1984;
how engrams change over more prolonged engram. Nat. Rev. Neurosci. 16, 521–534 (2015). pmid: 6726327
32. R. Menzel, Searching for the memory trace in a mini-brain,
periods of time. For example, do all engrams doi: 10.1038/nrn4000; pmid: 26289572
the honeybee. Learn. Mem. 8, 53–62 (2001). doi: 10.1101/
9. S. A. Josselyn, S. Köhler, P. W. Frankland, Heroes of
(engrams representing different types of mem- lm.38801; pmid: 11274250
the engram. J. Neurosci. 37, 4647–4657 (2017).
ories such as episodic, semantic, or even pro- doi: 10.1523/JNEUROSCI.0056-17.2017; pmid: 28469009 33. D. L. Alkon, Calcium-mediated reduction of ionic currents:
A biophysical memory trace. Science 226, 1037–1045 (1984).
cedural or motor memories, with different 10. S. Tonegawa, X. Liu, S. Ramirez, R. Redondo, Memory
engram cells have come of age. Neuron 87, 918–931 (2015). doi: 10.1126/science.6093258; pmid: 6093258
valence) change over time, gradually engaging 34. D. A. Mccormick et al., The engram found? Role of the
doi: 10.1016/j.neuron.2015.08.002; pmid: 26335640
more cortical regions? Is there a role for top- 11. D. L. Schacter, Forgotten Ideas, Neglected Pioneers: Richard cerebellum in classical conditioning of nictitating membrane
down (mPFC to hippocampal) processing in Semon and the Story of Memory (Psychology Press, 2001). and eyelid responses. Bull. Psychon. Soc. 18, 103–105
12. K. S. Lashley, in Society of Experimental Biology Symposium, (1981). doi: 10.3758/BF03333573
the dematuration of hippocampal engrams and 35. J. M. Fuster, J. P. Jervey, Inferotemporal neurons distinguish
No. 4: Psychological Mechanisms in Animal Behavior,
a possible role of silent hippocampal engrams J. F. Danielli, R. Brown, Eds. (Academic Press, 1950), and retain behaviorally relevant features of visual stimuli.
in remote memory recall? pp. 454–482. Science 212, 952–955 (1981). doi: 10.1126/science.7233192;
13. K. S. Lashley, Brain Mechanisms and Intelligence: pmid: 7233192
Second, how can we leverage our knowledge 36. Y. Miyashita, Neuronal correlate of visual associative long-
A Quantitative Study of Injuries to the Brain (Dover
of engrams in rodents to better understand Books on Psychology, vol. T1038, Dover Publications, 1963). term memory in the primate temporal cortex. Nature 335,
human memory? There is good evidence for 14. K. S. Lashley, Mass action in cerebral function. Science 73, 817–820 (1988). doi: 10.1038/335817a0; pmid: 3185711
general engram-like memory representations 245–254 (1931). doi: 10.1126/science.73.1888.245; 37. M. E. Greenberg, E. B. Ziff, Stimulation of 3T3 cells induces
pmid: 17755301 transcription of the c-fos proto-oncogene. Nature 311,
in humans [e.g., (245)], but, to date, there are 15. D. O. Hebb, The Organization of Behavior: A Neuropsychological 433–438 (1984). doi: 10.1038/311433a0; pmid: 6090941
no compelling findings at the cellular ensem- Theory (Wiley, 1949). 38. T. Curran, J. I. Morgan, Superinduction of c-fos by nerve
ble level. To extend the findings from rodent 16. C. J. Shatz, The developing brain. Sci. Am. 267, growth factor in the presence of peripherally active
60–67 (1992). doi: 10.1038/scientificamerican0992-60; benzodiazepines. Science 229, 1265–1268 (1985).
engram studies to humans, it may be neces- doi: 10.1126/science.4035354; pmid: 4035354
pmid: 1502524
sary to develop non- to low-invasive methods 17. D. Marr, Simple memory: A theory for archicortex. Philos. 39. J. F. Guzowski, B. L. McNaughton, C. A. Barnes,
to image and manipulate engrams at the single- Trans. R. Soc. Lond. Ser. B 262, 23–81 (1971). doi: 10.1098/ P. F. Worley, Imaging neural activity with temporal and
rstb.1971.0078; pmid: 4399412 cellular resolution using FISH. Curr. Opin. Neurobiol. 11,
cell or specific ensemble level in humans. 579–584 (2001). doi: 10.1016/S0959-4388(00)00252-X;
18. M. R. Hunsaker, R. P. Kesner, The operation of pattern
Progress in this general area of human “arti- separation and pattern completion processes associated pmid: 11595491
ficial memory manipulation” has been made with different attributes or domains of memory. 40. C. A. Denny et al., Hippocampal memory traces are
by harnessing the power of reconsolidation Neurosci. Biobehav. Rev. 37, 36–58 (2013). doi: 10.1016/ differentially modulated by experience, time, and adult
j.neubiorev.2012.09.014; pmid: 23043857 neurogenesis. Neuron 83, 189–201 (2014). doi: 10.1016/
(194, 235, 237, 246, 247) in which engram cells 19. J. J. Knierim, J. P. Neunuebel, Tracking the flow of j.neuron.2014.05.018; pmid: 24991962
are thought to be specifically reactivated by hippocampal computation: Pattern separation, pattern 41. L. G. Reijmers, B. L. Perkins, N. Matsuo, M. Mayford,
memory retrieval. Pharmacological blockade completion, and attractor dynamics. Neurobiol. Learn. Mem. Localization of a stable neural correlate of associative
129, 38–49 (2016). doi: 10.1016/j.nlm.2015.10.008; memory. Science 317, 1230–1233 (2007). doi: 10.1126/
of reconsolidation and noninvasive techniques pmid: 26514299 science.1143839; pmid: 17761885
that “update” memory during reconsolidation 20. M. Moscovitch, in Science of Memory: Concepts, 42. L. A. DeNardo et al., Temporal evolution of cortical ensembles
have shown some success in manipulating H. L. I. Roediger, Y. Dudai, S. M. Fitzpatrick, Eds. (Oxford promoting remote memory retrieval. Nat. Neurosci. 22,
Univ. Press, 2007), pp. 17–29. 460–469 (2019). doi: 10.1038/s41593-018-0318-7;
human memories (248, 249). 21. S. J. Martin, P. D. Grimwood, R. G. Morris, Synaptic pmid: 30692687
Finally, it is important that the links be- plasticity and memory: An evaluation of the hypothesis. 43. A. T. Sørensen et al., A robust activity marking system for
tween neuroscience and artificial intelligence Annu. Rev. Neurosci. 23, 649–711 (2000). doi: 10.1146/ exploring active neuronal ensembles. eLife 5, e13918 (2016).
annurev.neuro.23.1.649; pmid: 10845078 doi: 10.7554/eLife.13918; pmid: 27661450
(AI) are leveraged to inform both fields. Un- 44. M. S. Fanselow, L. S. Lester, in Evolution and Learning,
22. S. J. Martin, R. G. Morris, New life in an old idea: The synaptic
derstanding how the brain encodes, stores, plasticity and memory hypothesis revisited. Hippocampus 12, R. C. Bolles, M. D. Beecher, Eds. (Lawrence Erlbaum
and uses information, especially at the level of 609–636 (2002). doi: 10.1002/hipo.10107; pmid: 12440577 Associates, Inc., 1988), pp. 185–212.
23. D. Yu, Y. Tan, M. Chakraborty, S. Tomchik, R. L. Davis, 45. K. K. Tayler, K. Z. Tanaka, L. G. Reijmers, B. J. Wiltgen,
the engram, can help inspire the development
Elongator complex is required for long-term olfactory Reactivation of neural ensembles during the retrieval of
of more intelligent machines. For instance, en- memory formation in Drosophila. Learn. Mem. 25, 183–196 recent and remote memory. Curr. Biol. 23, 99–106 (2013).
grams and how engrams serve to link memo- (2018). doi: 10.1101/lm.046557.117; pmid: 29545390 doi: 10.1016/j.cub.2012.11.019; pmid: 23246402

46. S. Ramirez et al., Creating a false memory in the 67. E. Koya et al., Targeted disruption of cocaine-activated 91. S. Park et al., Neuronal allocation to a hippocampal
hippocampus. Science 341, 387–391 (2013). doi: 10.1126/ nucleus accumbens neurons prevents context-specific engram. Neuropsychopharmacology 41, 2987–2993 (2016).
science.1239073; pmid: 23888038 sensitization. Nat. Neurosci. 12, 1069–1073 (2009). doi: 10.1038/npp.2016.73; pmid: 27187069
47. W. Deng, M. Mayford, F. H. Gage, Selection of distinct doi: 10.1038/nn.2364; pmid: 19620976 92. M. J. Sekeres et al., Increasing CRTC1 function in the dentate
populations of dentate granule cells in response to inputs as 68. P. M. Milner, Cell assemblies: Whose idea? Psycholoquy 10, gyrus during memory formation or reactivation increases
a mechanism for pattern separation in mice. eLife 2, e00312 1 (1999). memory strength without compromising memory quality.
(2013). doi: 10.7554/eLife.00312; pmid: 23538967 69. X. Liu et al., Optogenetic stimulation of a hippocampal J. Neurosci. 32, 17857–17868 (2012). doi: 10.1523/
48. K. Z. Tanaka et al., Cortical representations are reinstated engram activates fear memory recall. Nature 484, 381–385 JNEUROSCI.1419-12.2012; pmid: 23223304
by the hippocampus during memory retrieval. Neuron 84, (2012). doi: 10.1038/nature11028; pmid: 22441246 93. M. J. Sekeres, R. L. Neve, P. W. Frankland, S. A. Josselyn,
347–354 (2014). doi: 10.1016/j.neuron.2014.09.037 70. E. S. Boyden, F. Zhang, E. Bamberg, G. Nagel, K. Deisseroth, Dorsal hippocampal CREB is both necessary and sufficient
pmid: 25308331 Millisecond-timescale, genetically targeted optical control for spatial memory. Learn. Mem. 17, 280–283 (2010).
49. M. Zelikowsky, S. Hersman, M. K. Chawla, C. A. Barnes, of neural activity. Nat. Neurosci. 8, 1263–1268 (2005). doi: 10.1101/lm.1785510; pmid: 20495061
M. S. Fanselow, Neuronal ensembles in amygdala, hippocampus, doi: 10.1038/nn1525; pmid: 16116447 94. T. Sacco, B. Sacchetti, Role of secondary sensory cortices
and prefrontal cortex track differential components of contextual 71. T. J. Ryan, D. S. Roy, M. Pignatelli, A. Arons, S. Tonegawa, in emotional memory storage and retrieval in rats. Science
fear. J. Neurosci. 34, 8462–8466 (2014). doi: 10.1523/ Engram cells retain memory under retrograde amnesia. 329, 649–656 (2010). doi: 10.1126/science.1183165;
JNEUROSCI.3624-13.2014; pmid: 24948801 Science 348, 1007–1013 (2015). doi: 10.1126/science. pmid: 20689011
50. Y. Nakazawa, A. Pevzner, K. Z. Tanaka, B. J. Wiltgen, aaa5542; pmid: 26023136 95. W. B. Kim, J. H. Cho, Synaptic targeting of double-projecting
Memory retrieval along the proximodistal axis of CA1. 72. B. N. Armbruster, X. Li, M. H. Pausch, S. Herlitze, B. L. Roth, ventral CA1 hippocampal neurons to the medial prefrontal
Hippocampus 26, 1140–1148 (2016). doi: 10.1002/ Evolving the lock to fit the key to create a family of G cortex and basal amygdala. J. Neurosci. 37, 4868–4882
hipo.22596; pmid: 27068122 protein-coupled receptors potently activated by an inert (2017). doi: 10.1523/JNEUROSCI.3579-16.2017;
51. T. Kitamura et al., Engrams and circuits crucial for systems ligand. Proc. Natl. Acad. Sci. U.S.A. 104, 5163–5168 (2007). pmid: 28385873
consolidation of a memory. Science 356, 73–78 (2017). doi: 10.1073/pnas.0700293104; pmid: 17360345 96. Y. Yang et al., Selective synaptic remodeling of
doi: 10.1126/science.aam6808; pmid: 28386011 73. C. D. Nichols, B. L. Roth, Engineered G-protein coupled amygdalocortical connections associated with fear memory.
52. O. Khalaf et al., Reactivation of recall-induced neurons receptors are powerful tools to investigate biological Nat. Neurosci. 19, 1348–1355 (2016). doi: 10.1038/nn.4370;
contributes to remote fear memory attenuation. Science processes and behaviors. Front. Mol. Neurosci. 2, 16 (2009). pmid: 27595384
360, 1239–1242 (2018). doi: 10.1126/science.aas9875 doi: 10.3389/neuro.02.016.2009; pmid: 19893765 97. A. L. Wheeler et al., Identification of a functional connectome
pmid: 29903974 74. K. K. Cowansage et al., Direct reactivation of a for long-term fear memory in mice. PLOS Comput. Biol.
53. A. F. Lacagnina et al., Distinct hippocampal engrams control coherent neocortical memory of context. Neuron 84, 9, e1002853 (2013). doi: 10.1371/journal.pcbi.1002853;
extinction and relapse of fear memory. Nat. Neurosci. 432–441 (2014). doi: 10.1016/j.neuron.2014.09.022; pmid: 23300432
22, 753–761 (2019). doi: 10.1038/s41593-019-0361-z; pmid: 25308330 98. G. Vetere et al., Chemogenetic interrogation of a brain-wide
pmid: 30936555 75. R. L. Redondo et al., Bidirectional switch of the valence fear memory network in mice. Neuron 94, 363–374.e4
54. H. Nomura, C. Teshirogi, D. Nakayama, M. Minami, Y. Ikegaya, associated with a hippocampal contextual memory engram. (2017). doi: 10.1016/j.neuron.2017.03.037; pmid: 28426969
Prior observation of fear learning enhances subsequent Nature 513, 426–430 (2014). doi: 10.1038/nature13725; 99. D. S. Roy et al., Brain-wide mapping of contextual fear
self-experienced fear learning with an overlapping neuronal pmid: 25162525 memory engram ensembles supports the dispersed
ensemble in the dorsal hippocampus. Mol. Brain 12, 21 (2019). 76. A. P. Yiu et al., Neurons are recruited to a memory trace engram complex hypothesis. bioRxiv 668483 [Preprint].
doi: 10.1186/s13041-019-0443-6; pmid: 30871580 based on relative neuronal excitability immediately before 12 June 2019. .doi: 10.1101/668483
55. S. Trouche, J. M. Sasaki, T. Tu, L. G. Reijmers, Fear extinction training. Neuron 83, 722–735 (2014). doi: 10.1016/ 100. P. W. Frankland, B. Bontempi, L. E. Talton, L. Kaczmarek,
causes target-specific remodeling of perisomatic inhibitory j.neuron.2014.07.017; pmid: 25102562 A. J. Silva, The involvement of the anterior cingulate cortex in
synapses. Neuron 80, 1054–1065 (2013). doi: 10.1016/ 77. T. Rogerson et al., Synaptic tagging during memory remote contextual fear memory. Science 304, 881–883
j.neuron.2013.07.047; pmid: 24183705 allocation. Nat. Rev. Neurosci. 15, 157–169 (2014). (2004). doi: 10.1126/science.1094804; pmid: 15131309
56. A. Nonaka et al., Synaptic plasticity associated with a doi: 10.1038/nrn3667; pmid: 24496410 101. K. Chung et al., Structural and molecular interrogation of
memory engram in the basolateral amygdala. J. Neurosci. 34, 78. J. Kim, J. T. Kwon, H. S. Kim, S. A. Josselyn, J. H. Han, intact biological systems. Nature 497, 332–337 (2013).
9305–9309 (2014). doi: 10.1523/JNEUROSCI.4233-13.2014; Memory recall and modifications by activating neurons doi: 10.1038/nature12107; pmid: 23575631
pmid: 25009263 with elevated CREB. Nat. Neurosci. 17, 65–72 (2014). 102. Y. G. Park et al., Protection of tissue physicochemical
57. H. Xie et al., In vivo imaging of immediate early gene expression doi: 10.1038/nn.3592; pmid: 24212670 properties using polyfunctional crosslinkers. Nat. Biotechnol.
reveals layer-specific memory traces in the mammalian brain. 79. K. Abdou et al., Synapse-specific representation of the (2018). pmid: 30556815
Proc. Natl. Acad. Sci. U.S.A. 111, 2788–2793 (2014). doi: 10.1073/ identity of overlapping memory engrams. Science 103. J. O’Keefe, J. Dostrovsky, The hippocampus as a spatial map.
pnas.1316808111 pmid: 24550309 360, 1227–1231 (2018). doi: 10.1126/science.aat3810 Preliminary evidence from unit activity in the freely-moving
58. J. H. Han et al., Selective erasure of a fear memory. Science pmid: 29903972 rat. Brain Res. 34, 171–175 (1971). doi: 10.1016/0006-8993
323, 1492–1496 (2009). doi: 10.1126/science.1164139; 80. A. Guskjolen et al., Recovery of “lost” infant memories in (71)90358-1; pmid: 5124915
pmid: 19286560 mice. Curr. Biol. 28, 2283–2290.e3 (2018). doi: 10.1016/ 104. V. Ego-Stengel, M. A. Wilson, Disruption of ripple-
59. Y. Dong et al., CREB modulates excitability of nucleus j.cub.2018.05.059; pmid: 29983316 associated hippocampal activity during rest impairs
accumbens neurons. Nat. Neurosci. 9, 475–477 (2006). 81. K. Ghandour et al., Orchestrated ensemble activities constitute a spatial learning in the rat. Hippocampus 20, 1–10 (2010).
doi: 10.1038/nn1661; pmid: 16520736 hippocampal memory engram. Nat. Commun. 10, 2637 (2019). pmid: 19816984
60. H. Marie, W. Morishita, X. Yu, N. Calakos, R. C. Malenka, doi: 10.1038/s41467-019-10683-2; pmid: 31201332 105. G. Girardeau, K. Benchenane, S. I. Wiener, G. Buzsáki,
Generation of silent synapses by acute in vivo expression of 82. I. Pavlov, Conditioned Reflexes (Oxford Univ. Press, 1927). M. B. Zugaro, Selective suppression of hippocampal ripples
CaMKIV and CREB. Neuron 45, 741–752 (2005). 83. D. L. Schacter, E. F. Loftus, Memory and law: What can impairs spatial memory. Nat. Neurosci. 12, 1222–1223
doi: 10.1016/j.neuron.2005.01.039; pmid: 15748849 cognitive neuroscience contribute? Nat. Neurosci. 16, (2009). doi: 10.1038/nn.2384; pmid: 19749750
61. M. H. Han et al., Role of cAMP response element-binding 119–123 (2013). doi: 10.1038/nn.3294; pmid: 23354384 106. T. J. McHugh et al., Dentate gyrus NMDA receptors mediate
protein in the rat locus ceruleus: Regulation of neuronal 84. A. R. Garner et al., Generation of a synthetic memory trace. rapid pattern separation in the hippocampal network.
activity and opiate withdrawal behaviors. J. Neurosci. 26, Science 335, 1513–1516 (2012). doi: 10.1126/ Science 317, 94–99 (2007). doi: 10.1126/science.1140263;
4624–4629 (2006). doi: 10.1523/JNEUROSCI.4701-05.2006; science.1214985; pmid: 22442487 pmid: 17556551
pmid: 16641242 85. N. Ohkawa et al., Artificial association of pre-stored 107. K. Z. Tanaka et al., The hippocampal engram maps
62. M. Lopez de Armentia et al., cAMP response element-binding information to generate a qualitatively new memory. Cell Rep. experience but not place. Science 361, 392–397 (2018).
protein-mediated gene expression increases the intrinsic 11, 261–269 (2015). doi: 10.1016/j.celrep.2015.03.017; doi: 10.1126/science.aat5397; pmid: 30049878
excitability of CA1 pyramidal neurons. J. Neurosci. 27, pmid: 25843716 108. M. A. Wilson, B. L. McNaughton, Reactivation of hippocampal
13909–13918 (2016). doi: 10.1523/JNEUROSCI.3850- 86. G. Vetere et al., Memory formation in the absence of ensemble memories during sleep. Science 265, 676–679
07.2007; pmid: 18077703 experience. Nat. Neurosci. 22, 933–940 (2019). doi: 10.1038/ (1994). doi: 10.1126/science.8036517; pmid: 8036517
63. Y. Zhou et al., CREB regulates excitability and the allocation s41593-019-0389-0; pmid: 31036944 109. S. Diekelmann, J. Born, The memory function of sleep.
of memory to subsets of neurons in the amygdala. 87. J. A. Kauer, R. C. Malenka, Synaptic plasticity and addiction. Nat. Rev. Neurosci. 11, 114–126 (2010). doi: 10.1038/nrn2762;
Nat. Neurosci. 12, 1438–1443 (2009). doi: 10.1038/nn.2405; Nat. Rev. Neurosci. 8, 844–858 (2007). doi: 10.1038/ pmid: 20046194
pmid: 19783993 nrn2234; pmid: 17948030 110. H. R. Joo, L. M. Frank, The hippocampal sharp wave-ripple in
64. E. Benito, A. Barco, CREB’s control of intrinsic and synaptic 88. J. H. Choi et al., Interregional synaptic maps among memory retrieval for immediate use and consolidation.
plasticity: Implications for CREB-dependent memory engram cells underlie memory formation. Science Nat. Rev. Neurosci. 19, 744–757 (2018). doi: 10.1038/
models. Trends Neurosci. 33, 230–240 (2010). doi: 10.1016/ 360, 430–435 (2018). doi: 10.1126/science.aas9204; s41583-018-0077-1; pmid: 30356103
j.tins.2010.02.001; pmid: 20223527 pmid: 29700265 111. G. Buzsáki, Hippocampal sharp wave-ripple: A cognitive
65. D. Sargin et al., CREB regulates spine density of lateral 89. L. A. Gouty-Colomer et al., Arc expression identifies the biomarker for episodic memory and planning.
amygdala neurons: Implications for memory allocation. lateral amygdala fear memory trace. Mol. Psychiatry 21, Hippocampus 25, 1073–1188 (2015). doi: 10.1002/
Front. Behav. Neurosci. 7, 209 (2013). doi: 10.3389/ 364–375 (2016). doi: 10.1038/mp.2015.18; pmid: 25802982 hipo.22488; pmid: 26135716
fnbeh.2013.00209; pmid: 24391565 90. A. Hayashi-Takagi et al., Labelling and optical erasure of 112. M. F. Carr, S. P. Jadhav, L. M. Frank, Hippocampal replay in
66. H. L. Hsiang et al., Manipulating a “cocaine engram” in mice. synaptic memory traces in the motor cortex. Nature the awake state: A potential substrate for memory
J. Neurosci. 34, 14115–14127 (2014). doi: 10.1523/ 525, 333–338 (2015). doi: 10.1038/nature15257; consolidation and retrieval. Nat. Neurosci. 14, 147–153 (2011).
JNEUROSCI.3327-14.2014; pmid: 25319707 pmid: 26352471 doi: 10.1038/nn.2732; pmid: 21270783

113. H. S. Kudrimoti, C. A. Barnes, B. L. McNaughton, Reactivation 136. M. A. Wilson, B. L. McNaughton, Dynamics of the Annu. Rev. Psychol. 61, 49–79, C1–C4 (2010). doi: 10.1146/
of hippocampal cell assemblies: Effects of behavioral hippocampal ensemble code for space. Science 261, annurev.psych.093008.100523; pmid: 19575620
state, experience, and EEG dynamics. J. Neurosci. 19, 1055–1058 (1993). doi: 10.1126/science.8351520 158. P. W. Frankland, B. Bontempi, The organization of recent and
4090–4101 (1999). doi: 10.1523/JNEUROSCI.19-10-04090.1999; pmid: 8351520 remote memories. Nat. Rev. Neurosci. 6, 119–130 (2005).
pmid: 10234037 137. J. D. Cohen, M. Bolstad, A. K. Lee, Experience-dependent doi: 10.1038/nrn1607; pmid: 15685217
114. A. G. Siapas, M. A. Wilson, Coordinated interactions shaping of hippocampal CA1 intracellular activity in novel 159. B. J. Wiltgen, R. A. M. Brown, L. E. Talton, A. J. Silva, New
between hippocampal ripples and cortical spindles during and familiar environments. eLife 6, e23040 (2017). circuits for old memories: The role of the neocortex in
slow-wave sleep. Neuron 21, 1123–1128 (1998). doi: 10.1016/ doi: 10.7554/eLife.23040; pmid: 28742496 consolidation. Neuron 44, 101–108 (2004). doi: 10.1016/
S0896-6273(00)80629-7; pmid: 9856467 138. J. Epsztein, M. Brecht, A. K. Lee, Intracellular determinants of j.neuron.2004.09.015; pmid: 15450163
115. F. Xia et al., Parvalbumin-positive interneurons mediate hippocampal CA1 place and silent cell activity in a novel 160. D. S. Roy, S. Muralidhar, L. M. Smith, S. Tonegawa, Silent
neocortical-hippocampal interactions that are necessary for environment. Neuron 70, 109–120 (2011). doi: 10.1016/ memory engrams as the basis for retrograde amnesia. Proc. Natl.
memory consolidation. eLife 6, e27868 (2017). doi: 10.7554/ j.neuron.2011.03.006; pmid: 21482360 Acad. Sci. U.S.A. 114, E9972–E9979 (2017). doi: 10.1073/
eLife.27868; pmid: 28960176 139. P. D. Rich, H. P. Liaw, A. K. Lee, Large environments reveal pnas.1714248114; pmid: 29078397
116. A. I. Abbas et al., Somatostatin interneurons facilitate the statistical structure governing hippocampal 161. S. Nabavi et al., Engineering a memory with LTD and LTP. Nature
hippocampal-prefrontal synchrony and prefrontal spatial representations. Science 345, 814–817 (2014). doi: 10.1126/ 511, 348–352 (2014). doi: 10.1038/nature13294; pmid: 24896183
encoding. Neuron 100, 926–939.e3 (2018). doi: 10.1016/ science.1255635; pmid: 25124440 162. M. H. Monfils, G. C. Teskey, Induction of long-term depression is
j.neuron.2018.09.029; pmid: 30318409 140. D. Lee, B. J. Lin, A. K. Lee, Hippocampal place fields emerge associated with decreased dendritic length and spine density
117. D. Dupret, J. O’Neill, B. Pleydell-Bouverie, J. Csicsvari, The upon single-cell manipulation of excitability during behavior. in layers III and V of sensorimotor neocortex. Synapse 53, 114–121
reorganization and reactivation of hippocampal maps predict Science 337, 849–853 (2012). doi: 10.1126/science.1221489; (2004). doi: 10.1002/syn.20039; pmid: 15170823
spatial memory performance. Nat. Neurosci. 13, 995–1002 pmid: 22904011 163. Q. Zhou, K. J. Homma, M. M. Poo, Shrinkage of dendritic
(2010). doi: 10.1038/nn.2599; pmid: 20639874 141. J. P. Rickgauer, K. Deisseroth, D. W. Tank, Simultaneous spines associated with long-term depression of hippocampal
118. S. P. Jadhav, C. Kemere, P. W. German, L. M. Frank, Awake cellular-resolution optical perturbation and imaging of place synapses. Neuron 44, 749–757 (2004). doi: 10.1016/
hippocampal sharp-wave ripples support spatial memory. cell firing fields. Nat. Neurosci. 17, 1816–1824 (2014). j.neuron.2004.11.011; pmid: 15572107
Science 336, 1454–1458 (2012). doi: 10.1126/ doi: 10.1038/nn.3866; pmid: 25402854 164. J. N. Bourne, K. M. Harris, Balancing structure and function
science.1217230; pmid: 22555434 142. J.-P. Changeux, A. Danchin, Selective stabilisation of at hippocampal dendritic spines. Annu. Rev. Neurosci. 31,
119. S. Lewis, Sleep: Ever-decreasing ripples. Nat. Rev. Neurosci. developing synapses as a mechanism for the specification of 47–67 (2008). doi: 10.1146/annurev.
19, 184 (2018). pmid: 29515191 neuronal networks. Nature 264, 705–712 (1976). neuro.31.060407.125646; pmid: 18284372
120. H. Norimoto et al., Hippocampal ripples down-regulate doi: 10.1038/264705a0; pmid: 189195 165. M. F. Bear, R. C. Malenka, Synaptic plasticity: LTP and LTD.
synapses. Science 359, 1524–1527 (2018). doi: 10.1126/ 143. J. Z. Young, Learning as a process of selection and Curr. Opin. Neurobiol. 4, 389–399 (1994). doi: 10.1016/0959-
science.aao0702; pmid: 29439023 amplification. J. R. Soc. Med. 72, 801–814 (1979). 4388(94)90101-5; pmid: 7919934
121. S. A. Josselyn, P. W. Frankland, Memory allocation: Mechanisms doi: 10.1177/014107687907201103; pmid: 552442 166. D. S. Roy et al., Memory retrieval by activating engram cells
and function. Annu. Rev. Neurosci. 41, 389–413 (2018). 144. P. Kanerva, Sparse Distributed Memory (MIT Press, 1988). in mouse models of early Alzheimer’s disease. Nature 531,
doi: 10.1146/annurev-neuro-080317-061956; pmid: 29709212 145. D. J. Morrison et al., Parvalbumin interneurons constrain the 508–512 (2016). doi: 10.1038/nature17172; pmid: 26982728
122. J. H. Han et al., Neuronal competition and selection during size of the lateral amygdala engram. Neurobiol. Learn. Mem. 167. J. N. Perusini et al., Optogenetic stimulation of dentate gyrus
memory formation. Science 316, 457–460 (2007). 135, 91–99 (2016). doi: 10.1016/j.nlm.2016.07.007; engrams restores memory in Alzheimer’s disease mice.
doi: 10.1126/science.1139438; pmid: 17446403 pmid: 27422019 Hippocampus 27, 1110–1122 (2017). doi: 10.1002/hipo.22756;
123. J. H. Han et al., Increasing CREB in the auditory thalamus 146. P. Rao-Ruiz, J. Yu, S. A. Kushner, S. A. Josselyn, Neuronal pmid: 28667669
enhances memory and generalization of auditory conditioned competition: Microcircuit mechanisms define the sparsity of 168. M. El Haj, V. Postal, P. Allain, Music enhances
fear. Learn. Mem. 15, 443–453 (2008). doi: 10.1101/ the engram. Curr. Opin. Neurobiol. 54, 163–170 (2019). autobiographical memory in mild Alzheimer’s disease.
lm.993608; pmid: 18519545 doi: 10.1016/j.conb.2018.10.013; pmid: 30423499 Educ. Gerontol. 38, 30–41 (2012). doi: 10.1080/
124. S. A. Josselyn, Continuing the search for the engram: Examining 147. T. Stefanelli, C. Bertollini, C. Lüscher, D. Muller, P. Mendez, 03601277.2010.515897
the mechanism of fear memories. J. Psychiatry Neurosci. 35, Hippocampal somatostatin interneurons control the size of 169. A. Herlitz, R. Adolfsson, L. Bäckman, L. G. Nilsson, Cue
221–228 (2010). doi: 10.1503/jpn.100015; pmid: 20569648 neuronal memory ensembles. Neuron 89, 1074–1085 (2016). utilization following different forms of encoding in mildly,
125. A. J. Silva, Y. Zhou, T. Rogerson, J. Shobe, J. Balaji, Molecular and doi: 10.1016/j.neuron.2016.01.024; pmid: 26875623 moderately, and severely demented patients with Alzheimer’s
cellular approaches to memory allocation in neural circuits. 148. H. C. Barron, T. P. Vogels, T. E. Behrens, M. Ramaswami, disease. Brain Cogn. 15, 119–130 (1991). doi: 10.1016/0278-
Science 326, 391–395 (2009). doi: 10.1126/science.1174519; Inhibitory engrams in perception and memory. Proc. Natl. 2626(91)90020-9; pmid: 2009170
pmid: 19833959 Acad. Sci. U.S.A. 114, 6666–6674 (2017). pmid: 28611219 170. E. K. Warrington, L. Weiskrantz, Amnesic syndrome:
126. M. R. Matos et al., Memory strength gates the involvement of 149. H. C. Barron et al., Unmasking latent inhibitory connections in Consolidation or retrieval? Nature 228, 628–630 (1970).
a CREB-dependent cortical fear engram in remote memory. human cortex to reveal dormant cortical memories. Neuron doi: 10.1038/228628a0; pmid: 4990853
Nat. Commun. 10, 2315 (2019). doi: 10.1038/s41467-019- 90, 191–203 (2016). doi: 10.1016/j.neuron.2016.02.031; 171. R. R. Miller, L. D. Matzel, Retrieval failure versus memory loss
10266-1; pmid: 31127098 pmid: 26996082 in experimental amnesia: Definitions and processes.
127. Y. Sano et al., CREB regulates memory allocation in the insular
150. G. Hennequin, E. J. Agnes, T. P. Vogels, Inhibitory plasticity: Learn. Mem. 13, 491–497 (2006). doi: 10.1101/lm.241006;
cortex. Curr. Biol. 24, 2833–2837 (2014). doi: 10.1016/
Balance, control, and codependence. Annu. Rev. Neurosci. pmid: 17015845
j.cub.2014.10.018; pmid: 25454591
40, 557–579 (2017). doi: 10.1146/annurev-neuro-072116- 172. O. Hardt, S. H. Wang, K. Nader, Storage or retrieval deficit:
128. R. Czajkowski et al., Encoding and storage of spatial
031005; pmid: 28598717 The yin and yang of amnesia. Learn. Mem. 16, 224–230
information in the retrosplenial cortex. Proc. Natl. Acad. Sci.
151. S. Maren, C. R. Ferrario, K. A. Corcoran, T. J. Desmond, (2009). doi: 10.1101/lm.1267409; pmid: 19304892
U.S.A. 111, 8661–8666 (2014). doi: 10.1073/pnas.1313222111;
K. A. Frey, Protein synthesis in the amygdala, but not the auditory 173. L. R. Squire, Lost forever or temporarily misplaced? The
pmid: 24912150
129. G. B. Choi et al., Driving opposing behaviors with ensembles thalamus, is required for consolidation of Pavlovian fear long debate about the nature of memory impairment.
conditioning in rats. Eur. J. Neurosci. 18, 3080–3088 (2003). Learn. Mem. 13, 522–529 (2006). doi: 10.1101/lm.310306;
of piriform neurons. Cell 146, 1004–1015 (2011).
doi: 10.1016/j.cell.2011.07.041; pmid: 21925321 doi: 10.1111/j.1460-9568.2003.03063.x; pmid: 14656303 pmid: 17015849
130. D. Kim, D. Paré, S. S. Nair, Assignment of model amygdala 152. G. E. Schafe, J. E. LeDoux, Memory consolidation of auditory 174. E. Tulving, Z. Pearlstone, Availability versus accessibility of
neurons to the fear memory trace depends on competitive Pavlovian fear conditioning requires protein synthesis and information in memory for words. J. Verbal Learn. Verbal Behav.
synaptic interactions. J. Neurosci. 33, 14354–14358 (2013). protein kinase A in the amygdala. J. Neurosci. 20, RC96 5, 381–391 (1966). doi: 10.1016/S0022-5371(66)80048-8
doi: 10.1523/JNEUROSCI.2430-13.2013; pmid: 24005288 (2000). doi: 10.1523/JNEUROSCI.20-18-j0003.2000; 175. P. W. Frankland, S. A. Josselyn, S. Köhler, The neurobiological
131. D. Kim, D. Paré, S. S. Nair, Mechanisms contributing to the pmid: 10974093 foundation of memory retrieval. Nat. Neurosci. 22, 1576–1585
induction and storage of Pavlovian fear memories in the 153. P. J. Hernandez, T. Abel, The role of protein synthesis in (2019). doi: 10.1038/s41593-019-0493-1; pmid: 31551594
lateral amygdala. Learn. Mem. 20, 421–430 (2013). memory consolidation: Progress amid decades of 176. J. H. Kogan, P. W. Franklandand, A. J. Silva, Long-term
doi: 10.1101/lm.030262.113; pmid: 23864645 debate. Neurobiol. Learn. Mem. 89, 293–311 (2008). memory underlying hippocampus-dependent social
132. D. Kim, P. Samarth, F. Feng, D. Pare, S. S. Nair, Synaptic doi: 10.1016/j.nlm.2007.09.010; pmid: 18053752 recognition in mice. Hippocampus 10, 47–56 (2000).
competition in the lateral amygdala and the stimulus 154. G. E. Schafe, N. V. Nadel, G. M. Sullivan, A. Harris, J. E. LeDoux, doi: 10.1002/(SICI)1098-1063(2000)10:1<47::AID-
specificity of conditioned fear: A biophysical modeling study. Memory consolidation for contextual and auditory fear HIPO5>3.0.CO;2-6; pmid: 10706216
Brain Struct. Funct. 221, 2163–2182 (2016). doi: 10.1007/ conditioning is dependent on protein synthesis, PKA, and MAP 177. F. L. Hitti, S. A. Siegelbaum, The hippocampal CA2 region is
s00429-015-1037-4; pmid: 25859631 kinase. Learn. Mem. 6, 97–110 (1999). pmid: 10327235 essential for social memory. Nature 508, 88–92 (2014).
133. D. L. Alkon, Changes of membrane currents during learning. 155. S. A. Josselyn, S. Kida, A. J. Silva, Inducible repression of doi: 10.1038/nature13028; pmid: 24572357
J. Exp. Biol. 112, 95–112 (1984). pmid: 6150967 CREB function disrupts amygdala-dependent memory. 178. T. Okuyama, T. Kitamura, D. S. Roy, S. Itohara, S. Tonegawa,
134. D. L. Alkon, I. Lederhendler, J. J. Shoukimas, Primary changes Neurobiol. Learn. Mem. 82, 159–163 (2004). doi: 10.1016/ Ventral CA1 neurons store social memory. Science
of membrane currents during retention of associative learning. j.nlm.2004.05.008; pmid: 15341801 353, 1536–1541 (2016). doi: 10.1126/science.aaf7003;
Science 215, 693–695 (1982). doi: 10.1126/science.7058334; 156. Y. Dudai, M. Eisenberg, Rites of passage of the engram: pmid: 27708103
pmid: 7058334 Reconsolidation and the lingering consolidation hypothesis. 179. J. Kim et al., Amygdala depotentiation and fear extinction.
135. K. P. Scholz, J. H. Byrne, Long-term sensitization in Aplysia: Neuron 44, 93–100 (2004). doi: 10.1016/ Proc. Natl. Acad. Sci. U.S.A. 104, 20955–20960 (2007).
Biophysical correlates in tail sensory neurons. Science j.neuron.2004.09.003; pmid: 15450162 doi: 10.1073/pnas.0710548105; pmid: 18165656
235, 685–687 (1987). doi: 10.1126/science.2433766; 157. S. H. Wang, R. G. Morris, Hippocampal-neocortical interactions 180. I. Hong et al., Extinction of cued fear memory involves a
pmid: 2433766 in memory formation, consolidation, and reconsolidation. distinct form of depotentiation at cortical input synapses

onto the lateral amygdala. Eur. J. Neurosci. 30, 2089–2099 204. M. Moscovitch, R. Cabeza, G. Winocur, L. Nadel, Episodic assembly as the unit of cognition? Neurosci. Lett. 680, 88–93
(2009). doi: 10.1111/j.1460-9568.2009.07004.x pmid: 20128847 memory and beyond: The hippocampus and neocortex in (2018). doi: 10.1016/j.neulet.2017.04.006; pmid: 28389238
181. C.-H. Lin, C.-C. Lee, P.-W. Gean, Involvement of a calcineurin transformation. Annu. Rev. Psychol. 67, 105–134 (2016). 230. C. A. Denny, E. Lebois, S. Ramirez, From engrams to
cascade in amygdala depotentiation and quenching of fear doi: 10.1146/annurev-psych-113011-143733; pmid: 26726963 pathologies of the brain. Front. Neural Circuits 11, 23 (2017).
memory. Mol. Pharmacol. 63, 44–52 (2003). doi: 10.1124/ 205. N. Guo et al., Dentate granule cell recruitment of feedforward doi: 10.3389/fncir.2017.00023; pmid: 28439228
mol.63.1.44; pmid: 12488535 inhibition governs engram maintenance and remote memory 231. E. C. Tolman, C. H. Honzik, Introduction and removal of
182. T. Yoshii, H. Hosokawa, N. Matsuo, Pharmacogenetic generalization. Nat. Med. 24, 438–449 (2018). doi: 10.1038/nm.4491; reward, and maze performance in rats. Univ. Calif. Publ.
reactivation of the original engram evokes an extinguished pmid: 29529016 Psychol. 4, 257–275 (1930).
fear memory. Neuropharmacology 113, 1–9 (2017). 206. M. Shehata et al., Autophagy enhances memory erasure 232. G. T. Philips, E. I. Tzvetkova, S. Marinesco, T. J. Carew, Latent
doi: 10.1016/j.neuropharm.2016.09.012; pmid: 27639988 through synaptic destabilization. J. Neurosci. 38, 3809–3822 memory for sensitization in Aplysia. Learn. Mem. 13, 224–229
183. R. L. Clem, D. Schiller, New learning and unlearning: (2018). doi: 10.1523/JNEUROSCI.3505-17.2018; pmid: 29555855 (2006). doi: 10.1101/lm.111506; pmid: 16585798
Strangers or accomplices in threat memory attenuation? 207. C. B. Kirwan, C. E. Stark, Overcoming interference: An fMRI 233. R. E. Lubow, A. U. Moore, Latent inhibition: The effect of
Trends Neurosci. 39, 340–351 (2016). doi: 10.1016/ investigation of pattern separation in the medial temporal nonreinforced pre-exposure to the conditional stimulus.
j.tins.2016.03.003; pmid: 27079843 lobe. Learn. Mem. 14, 625–633 (2007). doi: 10.1101/lm.663507; J. Comp. Physiol. Psychol. 52, 415–419 (1959). doi: 10.1037/
184. N. V. Luchkina, V. Y. Bolshakov, Mechanisms of fear learning pmid: 17848502 h0046700; pmid: 14418647
and extinction: Synaptic plasticity-fear memory connection. 208. J. K. Leutgeb, S. Leutgeb, M. B. Moser, E. I. Moser, Pattern 234. D. J. Lewis, Psychobiology of active and inactive memory.
Psychopharmacology (Berl.) 236, 163–182 (2019). separation in the dentate gyrus and CA3 of the hippocampus. Psychol. Bull. 86, 1054–1083 (1979). doi: 10.1037/0033-
doi: 10.1007/s00213-018-5104-4; pmid: 30415278 Science 315, 961–966 (2007). doi: 10.1126/science.1135801; 2909.86.5.1054; pmid: 386401
185. C. A. Orsini, S. Maren, Neural and cellular mechanisms of fear pmid: 17303747 235. K. Nader, G. E. Schafe, J. E. Le Doux, Fear memories require
and extinction memory formation. Neurosci. Biobehav. Rev. 209. K. A. Norman, R. C. O’Reilly, Modeling hippocampal and protein synthesis in the amygdala for reconsolidation after
36, 1773–1802 (2012). doi: 10.1016/j.neubiorev.2011.12.014; neocortical contributions to recognition memory: A complementary- retrieval. Nature 406, 722–726 (2000). doi: 10.1038/
pmid: 22230704 learning-systems approach. Psychol. Rev. 110, 611–646 (2003). 35021052; pmid: 10963596
186. K. M. Myers, M. Davis, Mechanisms of fear extinction. Mol. doi: 10.1037/0033-295X.110.4.611; pmid: 14599236 236. J. Przybyslawski, S. J. Sara, Reconsolidation of memory after
Psychiatry 12, 120–150 (2007). doi: 10.1038/sj.mp.4001939; 210. A. Gilboa, H. Marlatte, Neurobiology of schemas and schema- its reactivation. Behav. Brain Res. 84, 241–246 (1997).
pmid: 17160066 mediated memory. Trends Cogn. Sci. 21, 618–631 (2017). doi: 10.1016/S0166-4328(96)00153-2; pmid: 9079788
187. M. E. Bouton, R. F. Westbrook, K. A. Corcoran, S. Maren, doi: 10.1016/j.tics.2017.04.013; pmid: 28551107 237. S. Kida et al., CREB required for the stability of new and
Contextual and temporal modulation of extinction: Behavioral and 211. S. McKenzie, H. Eichenbaum, Consolidation and reactivated fear memories. Nat. Neurosci. 5, 348–355
biological mechanisms. Biol. Psychiatry 60, 352–360 (2006). reconsolidation: Two lives of memories? Neuron 71, 224–233 (2002). doi: 10.1038/nn819; pmid: 11889468
doi: 10.1016/j.biopsych.2005.12.015; pmid: 16616731 (2011). doi: 10.1016/j.neuron.2011.06.037; pmid: 21791282 238. S. J. Sara, Retrieval and reconsolidation: Toward a
188. J. Ji, S. Maren, Hippocampal involvement in contextual 212. M. L. Schlichting, A. R. Preston, Memory integration: Neural neurobiology of remembering. Learn. Mem. 7, 73–84 (2000).
modulation of fear extinction. Hippocampus 17, 749–758 mechanisms and implications for behavior. Curr. Opin. Behav. Sci. 1, doi: 10.1101/lm.7.2.73; pmid: 10753974
(2007). doi: 10.1002/hipo.20331; pmid: 17604353 1–8 (2015). doi: 10.1016/j.cobeha.2014.07.005; pmid: 25750931 239. R. G. Morris et al., Memory reconsolidation: Sensitivity of spatial
189. M. E. Bouton, Context and behavioral processes in extinction. 213. D. Tse et al., Schemas and memory consolidation. Science 316, memory to inhibition of protein synthesis in dorsal hippocampus
Learn. Mem. 11, 485–494 (2004). doi: 10.1101/lm.78804; 76–82 (2007). doi: 10.1126/science.1135935; pmid: 17412951 during encoding and retrieval. Neuron 50, 479–489 (2006).
pmid: 15466298 214. D. Zeithamova, A. R. Preston, Temporal proximity promotes doi: 10.1016/j.neuron.2006.04.012; pmid: 16675401
190. G. J. Quirk, D. Mueller, Neural mechanisms of extinction integration of overlapping events. J. Cogn. Neurosci. 29, 240. J. L. Lee, Memory reconsolidation mediates the updating of
learning and retrieval. Neuropsychopharmacology 33, 56–72 1311–1323 (2017). doi: 10.1162/jocn_a_01116; pmid: 28253077 hippocampal memory content. Front. Behav. Neurosci. 4, 168
(2008). doi: 10.1038/sj.npp.1301555; pmid: 17882236 215. A. J. Rashid et al., Competition between engrams influences (2010). doi: 10.3389/fnbeh.2010.00168; pmid: 21120142
191. M. E. Bouton, J. B. Nelson, Context-specificity of target fear memory formation and recall. Science 353, 383–387 241. M. Eisenberg, T. Kobilo, D. E. Berman, Y. Dudai, Stability of
versus feature inhibition in a feature-negative discrimination. (2016). doi: 10.1126/science.aaf0594; pmid: 27463673 retrieved memory: Inverse correlation with trace dominance.
J. Exp. Psychol. Anim. Behav. Process. 20, 51–65 (1994). Science 301, 1102–1104 (2003). doi: 10.1126/
216. D. J. Cai et al., A shared neural ensemble links distinct
doi: 10.1037/0097-7403.20.1.51; pmid: 8308493 science.1086881; pmid: 12934010
contextual memories encoded close in time. Nature 534,
192. R. A. Rescorla, C. D. Heth, Reinstatement of fear to an extinguished 242. M. C. Anderson, R. A. Bjork, E. L. Bjork, Remembering can
115–118 (2016). doi: 10.1038/nature17955; pmid: 27251287
conditioned stimulus. J. Exp. Psychol. Anim. Behav. Process. 1, cause forgetting: Retrieval dynamics in long-term memory.
217. M. Sehgal et al., Memory allocation mechanisms underlie
88–96 (1975). doi: 10.1037/0097-7403.1.1.88; pmid: 1151290 J. Exp. Psychol. Learn. Mem. Cogn. 20, 1063–1087 (1994).
memory linking across time. Neurobiol. Learn. Mem. 153,
193. M. H. Monfils, K. K. Cowansage, E. Klann, J. E. LeDoux, doi: 10.1037/0278-7393.20.5.1063; pmid: 7931095
21–25 (2018). doi: 10.1016/j.nlm.2018.02.021; pmid: 29496645
Extinction-reconsolidation boundaries: Key to persistent 243. B. A. Richards, P. W. Frankland, The conjunctive trace.
218. P. Rao-Ruiz et al., Engram-specific transcriptome profiling of
attenuation of fear memories. Science 324, 951–955 (2009). Hippocampus 23, 207–212 (2013). doi: 10.1002/hipo.22089;
contextual memory consolidation. Nat. Commun. 10, 2232
doi: 10.1126/science.1167975; pmid: 19342552 pmid: 23389924
(2019). doi: 10.1038/s41467-019-09960-x; pmid: 31110186
194. D. Schiller et al., Preventing the return of fear in humans 244. A. Rubin, N. Geva, L. Sheintuch, Y. Ziv, Hippocampal ensemble
219. M. Pignatelli et al., Engram cell excitability state determines
using reconsolidation update mechanisms. Nature 463, dynamics timestamp events in long-term memory. eLife 4,
the efficacy of memory retrieval. Neuron 101, 274–284.e5
49–53 (2010). doi: 10.1038/nature08637; pmid: 20010606 e12247 (2015). doi: 10.7554/eLife.12247; pmid: 26682652
(2019). doi: 10.1016/j.neuron.2018.11.029; pmid: 30551997
195. X. Zhang, J. Kim, S. Tonegawa, Amygdala reward neurons 245. S. Brodt et al., Fast track to the neocortex: A memory engram
220. J. Yokose et al., Overlapping memory trace indispensable for
form and store fear extinction memory. Neuron 10.1016/ in the posterior parietal cortex. Science 362, 1045–1048 (2018).
linking, but not recalling, individual memories. Science 355,
j.neuron.2019.12.025 (2020). doi: 10.1126/science.aau2528; pmid: 30498125
398–403 (2017). doi: 10.1126/science.aal2690; pmid: 28126819
196. J. J. Kim, M. S. Fanselow, Modality-specific retrograde 246. K. Nader, Memory traces unbound. Trends Neurosci. 26, 65–72
221. B. J. Levy, A. D. Wagner, Measuring memory reactivation with
amnesia of fear. Science 256, 675–677 (1992). doi: 10.1126/ (2003). doi: 10.1016/S0166-2236(02)00042-5; pmid: 12536129
functional MRI: Implications for psychological theory.
science.1585183; pmid: 1585183 247. Y. Dudai, The restless engram: Consolidations never end.
197. S. Maren, G. Aharonov, M. S. Fanselow, Neurotoxic lesions of Perspect. Psychol. Sci. 8, 72–78 (2013). doi: 10.1177/
1745691612469031; pmid: 25484909 Annu. Rev. Neurosci. 35, 227–247 (2012). doi: 10.1146/
the dorsal hippocampus and Pavlovian fear conditioning in annurev-neuro-062111-150500; pmid: 22443508
rats. Behav. Brain Res. 88, 261–274 (1997). doi: 10.1016/ 222. M. L. Mack, B. C. Love, A. R. Preston, Building concepts one
episode at a time: The hippocampus and concept formation. 248. D. Schiller, E. A. Phelps, Does reconsolidation occur in
S0166-4328(97)00088-0; pmid: 9404635 humans? Front. Behav. Neurosci. 5, 24 (2011). doi: 10.3389/
198. K. K. Tayler, K. Z. Tanaka, L. G. Reijmers, B. J. Wiltgen, Neurosci. Lett. 680, 31–38 (2018). doi: 10.1016/j.neulet.2017.07.061;
pmid: 28801273 fnbeh.2011.00024; pmid: 21629821
Reactivation of neural ensembles during the retrieval of
223. M. L. Schlichting, A. R. Preston, Memory reactivation during 249. M. Kindt, M. Soeter, B. Vervliet, Beyond extinction: Erasing human
recent and remote memory. Curr. Biol. 23, 99–106 (2013).
rest supports upcoming learning of related content. Proc. fear responses and preventing the return of fear. Nat. Neurosci.
doi: 10.1016/j.cub.2012.11.019
Natl. Acad. Sci. U.S.A. 111, 15845–15850 (2014). doi: 10.1073/ 12, 256–258 (2009). doi: 10.1038/nn.2271; pmid: 19219038
199. S. Tonegawa, M. D. Morrissey, T. Kitamura, The role of
engram cells in the systems consolidation of memory. pnas.1404396111; pmid: 25331890
AC KNOWLED GME NTS
Nat. Rev. Neurosci. 19, 485–498 (2018). doi: 10.1038/ 224. D. Zeithamova, M. L. Schlichting, A. R. Preston, The
s41583-018-0031-2; pmid: 29970909 hippocampus and inferential reasoning: Building memories to We thank our many colleagues for interesting conversations that
200. B. J. Wiltgen, A. J. Silva, Memory for context becomes less navigate future decisions. Front. Hum. Neurosci. 6, 70 (2012). shaped this review. In particular, we would like to acknowledge the
specific with time. Learn. Mem. 14, 313–317 (2007). doi: 10.3389/fnhum.2012.00070; pmid: 22470333 contributions of Y. Dudai, P. Frankland, S. Köhler, M. Pignatelli, and
doi: 10.1101/lm.430907; pmid: 17522020 225. M. S. Gazzaniga, Conversations in the Cognitive S. Waddell, as well as J. Lau (for figure preparation) and D. Roy and
201. B. J. Wiltgen et al., The hippocampus plays a selective role in the Neurosciences (MIT Press, 1997). J. Yu (for a sorted publication list); and the members of the Josselyn,
retrieval of detailed contextual memories. Curr. Biol. 20, 226. H. B. Barlow, in The Cognitive Neurosciences, M. S. Gazzaniga, Tonegawa, and Frankland labs for helpful discussions. Funding:
1336–1344 (2010). doi: 10.1016/j.cub.2010.06.068 pmid: 20637623 Ed. (MIT Press, 1995), pp. 415–435. Supported by the Canadian Institute of Health Research (CIHR, FDN-
202. S. H. Wang, C. M. Teixeira, A. L. Wheeler, P. W. Frankland, The 227. C. G. Gross, Genealogy of the “grandmother cell”. Neuroscientist 388455), the Natural Science and Engineering Research Council
precision of remote context memories does not require the 8, 512–518 (2002). doi: 10.1177/107385802237175; (NSERC) Discovery Grant, the Canadian Institute for Advanced Studies
hippocampus. Nat. Neurosci. 12, 253–255 (2009). pmid: 12374433 (CiFAR) Grant, and the NIH (NIMH, 1 R01 MH119421-01) (to S.A.J);
doi: 10.1038/nn.2263; pmid: 19182794 228. R. Yuste, From the neuron doctrine to neural networks. and by RIKEN’s Center for Brain Science, Howard Hughes Medical
203. G. Winocur, M. Moscovitch, Memory transformation and Nat. Rev. Neurosci. 16, 487–497 (2015). doi: 10.1038/ Institute (HHMI), and JPB Foundation (to S.T.). Competing interests:
systems consolidation. J. Int. Neuropsychol. Soc. 17, 766–780 nrn3962; pmid: 26152865 The authors declare no competing interests.
(2011). doi: 10.1017/S1355617711000683; 229. H. Eichenbaum, Barlow versus Hebb: When is it time to
pmid: 21729403 abandon the notion of feature detectors and adopt the cell 10.1126/science.aaw4325

RES EARCH
◥ lations (17), (ii) the successive twining and

RESEARCH ARTICLES growth of tetrahedral units (4, 18, 19), and (iii)
partial dislocation slipping (20). Experimental
NANOPARTICLES studies were mainly based on morphology evo-
lution, in contrast to the decahedral symmetry
Oriented attachment induces fivefold twins by of 5-FTs (4, 18, 19). Furthermore, in these ex-
perimental studies, the particles had already
forming and decomposing high-energy undergone twinning by the time they were
observed (18, 20). Oriented attachment (OA)
grain boundaries of particles, which is a candidate pathway for
5-FT formation, was observed directly during
Miao Song1*, Gang Zhou2*, Ning Lu3*, Jaewon Lee1,4, Elias Nakouzi1, Hao Wang2, Dongsheng Li1† the formation of twin interfaces (21). We pri-
marily focused on face-centered cubic Au nano-
Natural and synthetic nanoparticles composed of fivefold twinned crystal domains have distinct particles (NPs) as a model system with which to
properties. The formation mechanism of these fivefold twinned nanoparticles is poorly understood. investigate the twinning process at the atomic
We used in situ high-resolution transmission electron microscopy combined with molecular dynamics scale. In order to observe the onset of twinning,
simulations to demonstrate that fivefold twinning occurs through repeated oriented attachment we chose small (~3 nm) Au NPs because of the
of ~3-nanometer gold, platinum, and palladium nanoparticles. We discovered two different mechanisms size effect on 5-FTs, which are stable with the
for forming fivefold twinned nanoparticles that are driven by the accumulation and elimination of strain. size of 3 to 14 nm owing to thermodynamics
This was accompanied by decomposition of grain boundaries and the formation of a special class of (22–24). Twinning exists in Au NPs of ≥3 nm
twins with a net strain of zero. These observations allowed us to develop a quantitative picture of the in our experiment (fig. S1, B and E) (25).
twinning process. The mechanisms provide guidance for controlling twin structures and morphologies
across a wide range of materials. Multiply twinned structures formed
through OA
T
We drop-cast spherical Au NPs (~3 nm) (fig.
winning in materials occurs when two mechanisms are unclear and subject to debate S1) embedded in an organic matrix of 1-
crystals that share the same crystal because of the challenges of making direct dodecanethiol (fig. S2) on a transmission elec-
lattice plane intergrow through certain observations of the formation process at the tron microscope (TEM) grid. The organics
symmetry operations. Crystallographic atomic scale. Proposed formation mechanisms decomposed under electron-beam (e-beam)
twins are widespread, being found in a include (i) direct nucleation through atom-by- irradiation and the surface of the Au NPs
wide range of materials that include minerals atom addition on the basis of theoretical simu- self-adjusted to decrease the surface energy
(such as rutile and wurtzite) (1, 2), metals [such
as copper (Cu), silver (Ag), and gold (Au)] (3, 4),
and ceramics [barium titanate (BaTiO3)] (5).
Twinning leads to a variety of structures and
morphologies that affect physical and chemi-
cal properties. For example, the stress of five-
fold twins (5-FTs) substantially increases the
Young’s modulus of nanowires (6), whereas
multitwinned Cu nanowires exhibit excellent
methane selectivity during reduction of car-
bon dioxide (7). Multiply twinned structures
have attracted substantial attention for ap-
plications in crystal growth (8), mechanical
engineering (9), biomedical diagnosis (10),
optics (11), and catalysis (12).
Understanding the properties of these
twinned materials required determining the
formation mechanisms that enables control
of the growth and the final shape. Fivefold
twins are a common multiply twinned struc-
ture that was discovered nearly 200 years ago
(13, 14), and the formation mechanisms and
potential applications have been widely studied
(15, 16). However, the underlying formation
1
Physical and Computational Sciences Directorate, Pacific
Northwest National Laboratory, Richland, WA 99352, USA.
2
Division of Titanium Alloys, Institute of Metal Research,
Fig. 1. Examples of resulting crystal structures after OA processes of Au NPs. (A) Formation of a
Chinese Academy of Sciences, Shenyang 110016, China.
3
Department of Materials Science and Engineering, University single crystal. (B) Formation of a twin structure with a twin boundary. (C to F) Formation of crystals with
of Michigan, Ann Arbor, MI 48109, USA. 4Department of multiple twin interfaces, including (C) parallel twin interfaces, (D) crossed twin interfaces with an angle
Biomedical, Biological and Chemical Engineering, University of ~109° between S3s and a concave surface angle of ~89°, (E) crossed twin interfaces with an angle of ~70°
of Missouri, Columbia, MO 65211, USA.
*These authors contributed equally to this work. between S3s and a concave surface angle of ~98°, and (F) crossed twin interfaces with an angle of ~71°
†Corresponding author. Email: dongsheng.li2@pnnl.gov between S3s and a concave surface angle of ~149°.

RESE ARCH | R E S E A R C H A R T I C L E S
Fig. 2. 5-FT formation mechanism 1 through OA,

atomic surface diffusion, and after nucleation
and growth of ZST. (A and B) Formation of
S32 and ~90° concave surface after the OA process.
(C) Migration of atoms to the concave surface
region (blue arrow) with the formation of the S9 GB
and S32 migration of two atomic layers toward
region III from II. (D to G) Oscillation between
S27-(200)I and S27-(111)IV and corresponding a.
(H) Nucleation of ZST on S27 near the twin pole.
(I) Formation of 5-FT. The red arrowhead in
(G) denotes one layer of twin interface through
partial dislocation slipping.
Fig. 3. S27 oscillatory evolution process

before the 5-FT formation. (A and B) Enlarged
high-resolution TEM images of the NP in
Fig. 2 showing a oscillation. (C) Theoretical
angle misfit between (200)I and (111)IV.
(D to F) Variation of a, bI, bII, and bIII with
time. Approximately 20 to 80 measurements
were taken for each data point in Fig. 3E.
SE, standard error. (G and H) Atomic structure
and energy profile of the whole NP during the
oscillatory evolution process of S27 based
on MD simulation, respectively.

RES EARCH | R E S E A R C H A R T I C L E S
(26), presenting a hexagonal two-dimensional twinned crystal (with the first S31) underwent At the 89.9° concave surface (Fig. 2B and
(2D) projection composed of four {111} and OA with another single crystal, resulting in movie S1, the same OA process as shown in
two {200} planes (Fig. 1A). During this process, two types of concave surfaces with angles of Fig. 1E), adjacent atoms migrated to this high-
the Au NPs moved and aggregated. The decom- ~94° ({111}+{200} surfaces) (Figs. 1, D and E, energy surface because of the small curvature
position of the surfactant led us to make the and 2, A and B) and ~150° (two {111} surfaces) radius (Fig. 2C and movie S1) (29), forming a
assumption that the type does not affect the (Fig. 1F). The angle between two S3s was new high-energy S9, identified by the 39° angle
mechanisms for this system (26). To understand either ~109° (Fig. 1D) or ~70° (Fig. 1, E and F). between {111} planes in regions I and IV (Fig.
the relation between the two grains, we used In addition, we often observed twinning and 2C and fig. S4, A to C). The GB energy (EGB)
the degree of fit (S) to define the grain bound- detwinning processes through the slipping of of S9 is ~542 mJ/m2 (30), which is high com-
ary (GB) based on coincident site lattice theory, atoms in {111} planes by 61 ½112—that is, partial pared with the ~17.5 mJ/m2 EGB of S3 (31).
where S is the ratio of the total number of sites dislocation slipping with a Burgers vector (b) of Because of this high EGB, S9 subsequently de-
6 ½112 (fig. S3). composed into a stable S3 (the third S33) and
1
to coincidence sites (27). During aggregation,
NPs specifically underwent OA along the {111} a new high-energy S27 (EGB ~ 560 mJ/m2) (30),
planes, forming either interface-free single Mechanism 1: Formation of 5-FT by identified by the 32° angle between the {111}
crystals (Fig. 1A) or S3 twin structures (Fig. 1B), forming and decomposing high-energy planes in regions I and IV (Fig. 2D and fig. S4,
similar to Au and iron oxide particle OA pro- grain boundaries D to F). The S9 decomposition follows the
cesses previously reported in liquid phases The angle of ~70° between two S3s (Fig. 1, E dissociation theory of the coincidence site lat-
(21, 28). In addition, multiple twin interfaces and F) provided the possibility of 5-FT forma- tice [S(A/B) ↔ SA + SB or S(A × B) ↔ SA +
often formed. For example, repeated OA pro- tion, in which the theoretical angle between SB] (32). As atoms further filled the concave
cesses induced both parallel (Fig. 1C) and three S3s is 70.53°. The two types of concave surfaces surface through either particle aggregation
types of crossed-twin interfaces (Fig. 1, D to F). with angles of ~98° (Fig. 1E) and ~150° (Fig. 1F) (Fig. 2, C and D) or surface atom diffusion (fig.
A second twin boundary S32 formed when a lead to two mechanisms for the 5-FT formation. S5), stacking faults formed in region IV, re-
sulting in a S27 between stacking faults and
{200}I planes in region I [S27-(200)I] (Fig. 2E).
This unstable S27-(200)I evolved into a dif-
ferent S27 between (111)IV and the region I
[S27-(111)IV] with the disappearance of stack-
ing faults (Fig. 2F and movie S1), resulting in
an oscillation between them (Figs. 2, E to G,
and 3). After three consecutive 64-s oscillation
cycles (Fig. 2, D to H; fig. S6; and movie S1),
twin lamellae—previously termed a zero-strain
twin (ZST)—that exhibit localized strain, but
possess strain fields that sum to zero in the
surrounding crystal lattice, nucleated at the
twin pole along S27 (Figs. 2H and 4) (33, 34).
The S27 immediately decomposed into the
fourth and fifth twin boundaries (S34 and S35)
within 0.2 s and formed a 5-FT (Fig. 2I), cor-
responding to S27→S34 + S35 + 7.35° (S81 =
S35 + 7.35°). The misfit angle of 7.35° between
the twin planes after 5-FT formation is ac-
commodated by lattice strain in twin units
(30), and the lattice strain can also be relaxed
through surface adjustment, such as forming
a truncated (Marks) decahedron (fig. S7) (35).
The evolution between S27-(200)I and S27-
(111)IV was accompanied by an oscillation of the
angle a (between S27 and S32) (Figs. 2, E to G,
and 3, A to D). The theoretical a is 125.3° and
148.4° for S27-(200)I and S27(111)IV, respective-
ly, which leads to a large angle misfit of 23.1°
between (200)I and (111)IV (Fig. 3C). During this
process, the angle misfit was mainly accom-
modated by either stacking faults in region IV
or lattice deformation in region I. In a perfect
Au crystal, the angle (b) between {111} planes is
70.53°. With stacking faults (a ~ 130°) (Fig. 3A),
the lattice deformed slightly in region I, with bI
Fig. 4. Decomposition of S27 through ZST forming S34 and S35. (A and B) Enlarged high-resolution of 72.5°. When the stacking faults disappeared
TEM images of twin pole regions of the NP at 76.8 and 80.6 s, respectively, showing the formation of a ZST (a ~ 150°), the lattice in region I deformed fur-
near the twin pole. (C) Schematics of the ZST. (D) MD simulation (model 1) with {111} surfaces. (E to H) ZST ther, with bI of ~74° (Fig. 3E), to compensate the
nucleation on S27 and growth in region I at 800 K. (I) Enlarged black-box area in (G). (J) Energy profile lattice mismatch. Region III was also affected
of the NP during decomposition of S27. during the oscillatory evolution of S27, with

bIII ranging from 71° to 74° (~72° on average) than the 94° concave surface, and the atoms Similarly, a symmetrical 5-FT can evolve into
(Fig. 3F). By contrast, region II was barely seldom migrated to the concave surface area. an asymmetrical structure through particle
affected, and bII remained at ~70° (Fig. 3F). Under e-beam irradiation, one layer of twin aggregation (fig. S18, F to J). On the basis of
However, bII increased instantly to ~72° as the structures formed at the concave area through 15 observations of 5-FT formation, the prob-
5-FT formed at 80.6 s (Fig. 3F) and remained GB decomposition and partial dislocation slip- ability of these two twinning mechanisms oc-
at ~72°, sharing the angle misfit (7.35°) with the ping (Fig. 5, C, D, and G; and fig. S14). Twin curring is almost equal (7:8) because of the
other twin units. Molecular dynamics (MD) interfaces subsequently migrated from the random selectivity of the attached {111} sur-
simulations confirmed a similar oscillation periphery of the particle toward the center faces during OA. We collected a number of
(Fig. 3G), showing comparable energies be- and formed S33, S34, and S35 (Fig. 5, D and E). other examples (figs. S15 and S17) of these two
tween them (Fig. 3H). MD results also con- On the basis of our eight observations of this types of twinning processes (fig. S19).
sistently demonstrated the relatively minor process, twin boundaries only migrate one to
shear strain in region I with stacking faults five atomic layers, resulting in the unstable Formation of 5-FT through OA in other
(Fig. 3G, a ~ 132.4°) and an increase of shear asymmetric 5-FT (Fig. 5, D and E, and fig. S15). metal systems
strain near the S27 boundary without stacking The asymmetric 5-FT underwent detwinning We observed the two fivefold twinning mech-
faults (Fig. 3G, a ~ 146.4°). through partial dislocation slipping, as dem- anisms not only in Au NPs with a low twin-
As described above, region I was highly onstrated by the varying number of {111} layers fault energy (~26 mJ/m2) (37) but also in other
strained without stacking faults, with bI of in each twin unit (Fig. 5, E and F), to decrease metal NP systems with high twin-fault ener-
~74°; the shear strain was concentrated near the twin interface and volume strain energies, palladium (Pd) NPs (~106 mJ/m2) (Fig. 6, A
the twin pole along S27-(111)IV (Fig. 4, A and E), gies (36). During detwinning processes, twin to H) (37) and platinum (Pt) NPs (~179 mJ/m2)
where at 80.6 s, three layers of {111} planes boundary migration and pole splitting were (Fig. 6, I to P) (37). For example, a twinned
deformed into {200} planes (Fig. 4B and fig. observed (more examples are provided in figs. NP (Fig. 6A) underwent OA with another NP
S8). This deformation did not affect the neigh- S16 and S17), similar to what we reported pre- (Fig. 6B), inducing the second S32 and a ~85°
boring lattice structures, indicating nucleation viously, the detwinning of Ag 5-FT particles concave surface; after atomic surface dif-
of a ZST. A ZST is typically formed by atoms in solution (36). fusion, a high-energy S9 GB formed (Fig. 6,
deforming from {111} into {200} planes through In addition, the asymmetric 5-FT can also A and B). Subsequently, S9 decomposed into
partial dislocation slipping of three consecutive grow into a stable structure through particle another high-energy GB S27 and the third
{111} planes (Fig. 4C) with Burgers vectors (b) aggregation (OA or non-OA) (fig. S18, A to E). S33 (Fig. 6, B and C). Last, S27 evolved into
of b1 ¼ 61 ½211, b2 ¼ 61 ½112, and b3 ¼ 61 ½1
21,
resulting in a total b of zero (Fig. 4C) (33, 34).
We found with MD simulation that the shear
strain concentration on the S27-{111}IV (Fig.
4E) induces nucleation of ZST (Fig. 4, D to G
and I; figs. S9 and S10; and movies S2 and S3).
The ZST grows laterally along the three layers
of the {111}I planes in region I (Fig. 4, F and G,
yellow arrows), resulting in strain relaxation
on the unstable S27 (Fig. 4G) and formation
of S35 in region I (Fig. 4G). The continuous
nucleation and growth of ZST (movies S2 and
S3) lead to the complete decomposition of S27-
(111)IV into S34 and S35, forming the 5-FT (Fig.
4H) with a large energy decrease (Fig. 4J).
Residual layers of {111}I planes (less than three
layers) near the NP surface can be twinned
through partial dislocation slipping (Fig. 2G).
Fully decomposing S27 through {111} partial
dislocation slipping nucleated on the surface
of the NP is unfavorable because of the angle
mismatch between {111} planes of regions I
and IV (fig. S11), although it has been observed
to form a few layer twin units. There was a
substantial configuration adjustment of the
NP during the S27 oscillatory evolution pro-
cess (Fig. 2, D to H, and fig. S12), indicating
surface adjustment of the NP. During this pe-
riod, a nucleation attempt of the ZST at 71.0 s
occurred, but it disappeared in ~0.3 s (fig. S13),
indicating a failure to overcome the energy
barrier to nucleate and grow.
Fig. 5. Formation mechanism 2 through OA and partial dislocation slipping. (A and B) Formation of
Mechanism 2: Formation of 5-FT through S32 and a concave surface after the OA process. (C) Formation of S9 after diffusion of atoms onto
partial dislocation slipping the concave surface. (D) Formation of 5-FT induced through GB decomposition and partial dislocation slipping.
At the resulting ~150° concave surface (Fig. 5, (E and F) Migration of twin boundaries. (G) Schematic illustration of the decomposition of S9 into three S3
A to C), the surface is relatively more stable GBs via partial dislocation slipping.

to a ~94° concave surface with a small radius

of curvature, OA provides a path to create
behavior that is far from equilibrium. A large
strain quickly accumulates while the surface
energy is minimized through atomic surface
diffusion to the curved surface, resulting in
lattice deformation and high-energy grain
boundaries (Fig. 3). To release this high strain
energy, a small twin forms through slipping
on three consecutive {111} planes, which does
not create extra strain in the nearby lattice or
require high energy—there is zero net strain.
Continuous nucleation and growth of this
twin structure releases the strain, eventually
forming a 5-FT.
In mechanism 2, after surface diffusion at
~150° concave surfaces, OA only creates a
small S9 (Fig. 5C) because of the large radius
of curvature, and the crystal lattice is less de-
formed than in mechanism 1. Consequently,
asymmetrical 5-FTs form through partial dis-
location slipping of only a few layers of {111}
planes because further slipping creates extra
strain energy in the lattice because of large
interfaces. By contrast, in mechanism 1 a highly
strained lattice is already formed through sur-
face diffusion after the OA event. The forma-
tion of 5-FTs within the bulk of the particle
(with large interfaces) results in a reduction of
strain. Similar processes of mechanisms 1 and
2 are also found in high twin-fault energy sys-
tems, such as Pd and Pt, but result in smaller
twin units.
Fig. 6. Examples of formation of 5-FTs in Pd and Pt NPs by means of mechanisms 1 and 2. (A to Multiply twinned structures, including
D) Formation process of mechanism 1 in Pd NPs. (E to H) Formation process of mechanism 2 in Pd NPs. 5-FTs, have attracted increased interest owing
(I to L) Formation process of mechanism 1in Pt NPs. (M to P) Formation process of mechanism 2 in Pt NPs. to their distinct properties (6, 7) and have been
widely used in mechanical engineering (9),
optics (11), and catalysis (12). OA events have
S34 and S35, forming 5-FT (Fig. 6, C and D) two layers of {111} planes by means of mech- been recognized as a common pathway for
by means of mechanism 1. At a ~149° concave anism 2 (Fig. 6, I to L and M to P). On the other crystal growth and occur widely in diverse
surface induced through OA (Fig. 6, E and hand, because of the higher twin-fault energy, systems, including oxides, metals, semicon-
F), 5-FT formed through partial dislocation more energy is required to form twin inter- ductors, organics, and biomineral phases (1–5).
slipping—that is, by means of mechanism 2 faces for Pt and Pd than Au, resulting in small Therefore, what we learned from our observa-
(Fig. 6, F and G). twin units. For the same reason, few 5-FTs in tion may generalize to a wide range of mate-
On the basis of most observations of Pd and Pt, which has the highest twin-fault energy, rials and enable the formation of multiply
Pt 5-FTs, the sizes of twin units are one to were reported previously (38). Many other fac- twinned structures besides 5-FTs, such as the
three layers of {111} planes (Fig. 6 and fig. S20), tors may affect the stability of 5-FTs, such as formation of multiply twinned iron oxide in
which is smaller than that of Au 5-FTs (4 to particle morphologies, external environments, solution (21), depending on specific crystal
10 layers) (Fig. 2I and figs. S15 and S17). During and surface absorbents (38). Large 5-FTs with structure and the nature of twin boundaries
the mechanism 1 process, we did not observe a high twin-fault energies may form under dif- in each system. Our findings place disparate
ZST (composed of three layers of {111} planes). ferent conditions. systems into the context of well-developed
A ZST may or may not form depending on the In total, we observed 57 (Au), 33 (Pd), and 57 theories for multiple-twin-formation mecha-
size of twin units. The average particle size of (Pt) OA processes and ~26% (Au), ~15% (Pd), nisms, provide a guide for interpreting and
the initial Pt and Pd NPs was ~2.7 nm (figs. S21 and ~9% (Pt) of them led to 5-FTs by means of controlling twinned crystal structures and
and S22), which is only slightly smaller than mechanism 1 or 2, indicating that the two 5-FT morphologies, and hopefully will result in
that of the Au NPs (~3 nm) (fig. S1). Therefore, formation mechanisms are applicable for var- advances in materials design and synthesis
the small twin units are unlikely to be attrib- ious metal systems. for diverse applications.
utable to the initial Pd or Pt particle size. For
Discussion and outlook REFERENCES AND NOTES
example, by means of mechanism 1, after the
1. N. Daneu, H. Schmid, A. Rečnik, W. Mader, Am. Mineral. 92,
OA of two ~3-nm Pt NPs (Fig. 6, I and J), the Our findings reveal that OA processes lead to 1789–1799 (2007).
twin pole of the resulting 5-FT is near the sur- multiply twinned structures, among which 2. V. Šrot, A. Rečnik, C. Scheu, S. Šturm, B. Mirtič, Am. Mineral.
face, with small twin units of two to three layers two types of concave surfaces lead to 5-FTs 88, 1809–1816 (2003).
3. L. Lu, X. Chen, X. Huang, K. Lu, Science 323, 607–610 (2009).
of {111} planes (Fig. 6, K and L). Another Pt through two well-defined and reproducible 4. M. R. Langille, J. Zhang, M. L. Personick, S. Li, C. A. Mirkin,
example shows small twin units with one to mechanisms. In mechanism 1, when OA leads Science 337, 954–957 (2012).

5. C. L. Jia, K. Urban, Science 303, 2001–2004 (2004). MULTIPLEX GENOMICS

6. J. Y. Wu, S. Nagao, J. Y. He, Z. L. Zhang, Nano Lett. 11,
5264–5273 (2011).
7. Y. Li et al., Nano Lett. 17, 1312–1317 (2017). Massively multiplex chemical transcriptomics at
single-cell resolution
8. Y. Xia, Y. Xiong, B. Lim, S. E. Skrabalak, Angew. Chem. Int. Ed.
48, 60–103 (2009).
9. S. Narayanan, G. Cheng, Z. Zeng, Y. Zhu, T. Zhu, Nano Lett. 15,
4037–4044 (2015).
10. H. Li, H. Hu, D. Xu, Anal. Chem. 87, 3826–3833 (2015). Sanjay R. Srivatsan1,2*, José L. McFaline-Figueroa1*, Vijay Ramani1,3*, Lauren Saunders1, Junyue Cao1,
11. B. Pietrobon, V. Kitaev, Chem. Mater. 20, 5186–5190 (2008). Jonathan Packer1, Hannah A. Pliner1, Dana L. Jackson1, Riza M. Daza1, Lena Christiansen4,
12. X. Wang et al., J. Am. Chem. Soc. 137, 15036–15042 (2015).
Fan Zhang4, Frank Steemers4, Jay Shendure1,5,6,7†‡, Cole Trapnell1,5,7†‡
13. G. Rose, Ann. Phys. 99, 196–204 (1831).
14. H. Hofmeister, Nanotechnology 3, 431–452 (2004).
15. L. D. Marks, L. Peng, J. Phys. Condens. Matter 28, 053001 (2016). High-throughput chemical screens typically use coarse assays such as cell survival, limiting what can
16. H. Hofmeister, Cryst. Res. Technol. 33, 3–25 (1998). be learned about mechanisms of action, off-target effects, and heterogeneous responses. Here, we
17. F. Baletto, A. Rapallo, G. Rossi, R. Ferrando, Phys. Rev. B 69,
235421 (2004).
introduce “sci-Plex,” which uses “nuclear hashing” to quantify global transcriptional responses to
18. K. Yagi, K. Takayanagi, K. Kobayashi, G. Honjo, J. Cryst. Growth thousands of independent perturbations at single-cell resolution. As a proof of concept, we applied sci-Plex
28, 117–124 (1975). to screen three cancer cell lines exposed to 188 compounds. In total, we profiled ~650,000 single-cell
19. M. R. Langille, J. Zhang, C. A. Mirkin, Angew. Chem. Int. Ed. 50, transcriptomes across ~5000 independent samples in one experiment. Our results reveal substantial
3543–3547 (2011).
20. Y. T. Zhu, X. Z. Liao, R. Z. Valiev, Appl. Phys. Lett. 86, 103112
intercellular heterogeneity in response to specific compounds, commonalities in response to families
(2005). of compounds, and insight into differential properties within families. In particular, our results with
21. D. Li et al., Science 336, 1014–1018 (2012). histone deacetylase inhibitors support the view that chromatin acts as an important reservoir of
22. A. S. Barnard, N. P. Young, A. I. Kirkland, M. A. van Huis, H. Xu,
ACS Nano 3, 1431–1436 (2009).
acetate in cancer cells.
23. N. P. Young, M. A. van Huis, H. W. Zandbergen, H. Xu,
H
A. I. Kirkland, Ultramicroscopy 110, 506–516 (2010).
24. K. Koga, T. Ikeshoji, K. Sugawara, Phys. Rev. Lett. 92, 115507 igh-throughput screens (HTSs) are a However, the per-sample and per-cell costs of
(2004).
25. Materials and methods are available as supplementary materials.
cornerstone of the pharmaceutical drug- most scRNA-seq technologies remain high,
26. L. Vitos, A. V. Ruban, H. L. Skriver, J. Kollár, Surf. Sci. 411, discovery pipeline (1, 2). However, con- precluding even modestly sized screens. Re-
186–202 (1998). ventional HTSs have at least two major cently, several groups have developed “cellular
27. H. Grimmer, W. Bollmann, D. H. Warrington, Acta Crystallogr. A
30, 197–207 (1974).
limitations. First, the readout of most hashing” methods, in which cells from differ-
28. C. Zhu et al., Nat. Commun. 9, 421 (2018). are restricted to gross cellular phenotypes, ent samples are molecularly labeled and mixed
29. Z. A. Peng, X. Peng, J. Am. Chem. Soc. 124, 3343–3353 (2002). e.g., proliferation (3, 4), morphology (5, 6), before scRNA-seq. However, current hashing
30. D. Wolf, Acta Metall. Mater. 38, 781–790 (1990).
or a highly specific molecular readout (7, 8). approaches require relatively expensive re-
31. A. Howie, L. D. Marks, Philos. Mag. A Phys. Condens. Matter
Defects Mech. Prop. 49, 95–109 (1984). Subtle changes in cell state or gene expres- agents [e.g., antibodies (17) or chemically
32. V. Randle, Acta Mater. 47, 4187–4196 (1999). sion that might otherwise provide mecha- modified DNA oligos (18, 19)], use cell-type–
33. L. Liu, J. Wang, S. K. Gong, S. X. Mao, Phys. Rev. Lett. 106, nistic insights or reveal off-target effects are dependent protocols (20), and/or use scRNA-
175504 (2011).
34. X. L. Ma, H. Zhou, J. Narayan, Y. T. Zhu, Scr. Mater. 109, 89–93 routinely missed. seq platforms with a high per-cell cost.
(2015). Second, even when HTSs are performed in To enable cost-effective HTSs with scRNA-
35. S. Patala, L. D. Marks, M. Olvera de la Cruz, J. Phys. Chem. C conjunction with more comprehensive mo- seq–based phenotyping, we describe a new
117, 1485–1494 (2013).
36. M. Song, Z. Wu, N. Lu, D. Li, Chem. Mater. 31, 842–850 (2019).
lecular phenotyping such as transcriptional sample labeling (hashing) strategy that relies
37. N. M. Rosengaard, H. L. Skriver, Phys. Rev. B 47, 12865–12873 profiling (9–12), a limitation of bulk assays is on labeling nuclei with unmodified single-
(1993). that even cells ostensibly of the same “type” stranded DNA oligos. Recent improvements
38. S. Zhou, M. Zhao, T. H. Yang, Y. Xia, Mater. Today 22, 108–131
(2019).
can exhibit heterogeneous responses (13, 14). in single-cell combinatorial indexing (sci-
Such cellular heterogeneity can be highly rel- RNA-seq3) have lowered the cost of scRNA-
ACKN OW LEDG MEN TS evant in vivo. For example, it remains largely seq library preparation to <$0.01 per cell,
We thank P. Sushko and J. De Yoreo for their comments unknown whether the rare subpopulations with millions of cells profiled per experi-
regarding revisions to our manuscript. Funding: This research
was supported by the U.S. Department of Energy (DOE), Office of
of cells that survive chemotherapeutics are ment (21). Here, we combine nuclear hash-
Science, Office of Basic Energy Sciences, through Early Career doing so on the basis of their genetic back- ing and sci-RNA-seq into a single workflow
Research Program Award KC0203020:67037. The work was ground, epigenetic state, or some other as- for multiplex transcriptomics in a process
conducted in the William R. Wiley Environmental Molecular
Sciences Laboratory, which is a DOE Office of Science User
pect (15, 16). called “sci-Plex.” As a proof of concept, we
Facility and located at Pacific Northwest National Laboratory In principle, single-cell transcriptome se- use sci-Plex to perform HTS on three cancer
(PNNL). PNNL is a multiprogram national laboratory operated for quencing (scRNA-seq) represents a form of cell lines, profiling thousands of indepen-
DOE by Battelle under contract DE-AC05-76RLO1830. Author
high-content molecular phenotyping that could dent perturbations in a single experiment.
contributions: M.S. conducted experiments, data analysis, and
results discussion. G.Z. conducted MD simulations. N.L. assisted enable HTSs to overcome both limitations. We further explore how chemical transcrip-
with data analysis. J.L. synthesized Au NPs. E.N. assisted with tomics at single-cell resolution can shed light
image processing. H.W. assisted with MD simulations. D.L. on mechanisms of action. Most notably, we
supervised the study, designed the experiments, and carried out
the results discussion. Competing interests: The authors declare
1
Department of Genome Sciences, University of Washington, find that gene-regulatory changes consequent
no competing interests. Data and materials availability: All Seattle, WA, USA. 2Medical Scientist Training Program, to treatment with histone deacetylase (HDAC)
data are available in the main text or the supplementary materials. University of Washington, Seattle, WA, USA. 3Department
of Biochemistry and Biophysics, University of California,
inhibitors are consistent with the model that
SUPPLEMENTARY MATERIALS San Francisco, San Francisco, CA, USA. 4Illumina Inc., they interfere with proliferation by restrict-
San Diego, CA, USA. 5Allen Discovery Center for Cell Lineage ing a cell’s ability to draw acetate from chro-
science.sciencemag.org/content/367/6473/40/suppl/DC1
Tracing, Seattle, WA, USA. 6Howard Hughes Medical
Materials and Methods matin (22, 23).
Institute, University of Washington, Seattle, WA, USA.
Figs. S1 to S25 7
Brotman Baty Institute for Precision Medicine, Seattle,
References (39–70) Results
WA, USA.
Movies S1 to S3
*These authors contributed equally to this work. †These authors Nuclear hashing enables multisample sci-RNA-seq
17 April 2019; resubmitted 7 October 2019 contributed equally to this work.
Accepted 15 November 2019 ‡Corresponding author. Email: shendure@uw.edu (J.S.); Single-cell combinatorial indexing (sci-) methods
10.1126/science.aax6511 coletrap@uw.edu (C.T.) use split-pool barcoding to specifically label the

molecular contents of large numbers of single lar identifiers (UMIs) per cell] and hash oligos followed by single-cell transcriptional profil-
cells or nuclei (24). Samples can be barcoded (median 270 UMIs per cell). ing as a high-content phenotypic assay. We
by these same indices, e.g., by placing each We devised a statistical framework to iden- exposed A549, a human lung adenocarcinoma
sample in its own well during reverse tran- tify the hash oligos associated with each cell at cell line, to one of four compounds: dexametha-
scription in sci-RNA-seq (21, 25), but such a frequency exceeding background (table S1). sone (a corticosteroid agonist), nutlin-3a (a p53-
enzymatic labeling at the scale of thousands We observed 99.1% concordance between spe- Mdm2 antagonist), BMS-345541 (an inhibitor
of samples is operationally infeasible and cost cies assignments on the basis of hash oligos of nuclear factor kB–dependent transcrip-
prohibitive. To enable single-cell molecular versus endogenous cellular transcriptomes tion), or vorinostat [suberoylanilide hydrox-
profiling of a large number of independent (Fig. 1C and fig. S1, C to F). Additionally, the amic acid (SAHA), an HDAC inhibitor], for
samples within a single sci- experiment, we set association of hash oligos and nuclei was sta- 24 hours across seven doses in triplicate for
out to develop a low-cost labeling procedure. ble to a freeze–thaw cycle, highlighting the a total of 84 drug–dose–replicate combina-
We noticed that single-stranded DNA (ssDNA) opportunity to label and store samples (Fig. 1D tions and additional vehicle controls (Fig. 2A
specifically stained the nuclei of permeabilized and fig. S1, G and H). These results demon- and fig. S2A). We labeled nuclei from each well
cells but not intact cells (Fig. 1A and fig. S1A). strate that hash oligos stably label nuclei in a and subjected them to sci-RNA-seq2 (fig. S2, B
We therefore postulated that a polyadenylated manner that is compatible with sci-RNA-seq. to D, and table S1).
ssDNA oligonucleotide could be used to label In sci- experiments, “collisions” are instances We used Monocle 3 (21) to visualize these
populations of nuclei in a manner compatible in which two or more cells are labeled with the data using Uniform Manifold Approxima-
with sci-RNA-seq (Fig. 1B and fig. S1B). To test same combination of barcodes by chance (24). tion and Projection (26) (UMAP) and Louvain
this concept, we performed a “barnyard” experi- To evaluate hashing as a means of detecting community detection to identify compound-
ment. We separately seeded human (HEK293T) doublets resulting from collisions, we varied specific clusters of cells, which were distributed
and mouse (NIH3T3) cells to 48 wells of a the number of nuclei loaded per polymerase in a dose-dependent manner (Fig. 2, B and C,
96-well culture plate. We then performed nu- chain reaction well, resulting in a range of and fig. S2, E and F). To quantify the “pop-
clear lysis in the presence of 96 well-specific predicted collision rates (7 to 23%) that was ulation average” transcriptional response of
polyadenylated ssDNA oligos (“hash oligos”) well matched by observation (fig. S1I). Hash A549 cells to each of the four drugs, we mod-
and fixed the resulting nuclear suspensions oligos facilitated the identification of the vast eled each gene’s expression as a function of
with paraformaldehyde. Having labeled or majority of interspecies doublets (95.5%) and dose through generalized linear regression.
“hashed” the nuclei with a molecular barcode, otherwise undetectable within-species doublets A total of 7561 genes were sensitive to at least
we pooled nuclei and performed a two-level (Fig. 1E and fig. S1, J and K). one drug, and 3189 genes were differentially
sci-RNA-seq experiment. Because the hash expressed in response to multiple drugs (fig.
oligos were polyadenylated, they had the po- sci-Plex enables multiplex chemical S3A and table S2). These included canon-
tential to be combinatorially indexed identi- transcriptomics at single-cell resolution ical targets of dexamethasone (Fig. 2D) and
cally to endogenous mRNAs. As intended, we We next evaluated whether nuclear hashing nutlin-3a (Fig. 2E). Gene ontology analysis of
recovered reads corresponding to both endog- could enable chemical screens by labeling cells differentially expressed genes revealed the
enous mRNAs [median 4740 unique molecu- that had undergone a specific perturbation, involvement of drug-specific pathways (e.g.,
Fig. 1. sci-Plex uses poly-

adenylated single-stranded
oligonucleotides to label
nuclei, enabling cell hashing
and doublet detection.
(A) Fluorescent images of
permeabilized nuclei after
incubation with DAPI (top) and
an Alexa Fluor-647–conjugated
single-stranded oligonucleotide
(bottom). (B) Overview of
sci-Plex. Cells corresponding
to different perturbations are
lysed in-well, their nuclei labeled
with well-specific “hash” oligos,
followed by fixation, pooling, and
sci-RNA-seq. (C) Scatter plot
depicting the number of UMIs
from single-cell transcriptomes
derived from a mixture of hashed
human HEK293T cells and
murine NIH3T3 cells. Points are
colored on the basis of hash
oligo assignment. (D) Boxplot
depicting the number of mRNA UMIs recovered per cell for fresh versus frozen human and mouse cell lines. (E) Scatter plot of overloading experiment;
axes are as in (C). Identified hash oligo collisions (red) identify cellular collisions with high sensitivity.

hormone signaling for dexamethasone; p53

A D ANGPTL4 GDF15
Cells (percent)
signaling for nutlin-3a; fig. S3B). Addition- BMS345541 Dexamethasone Nutlin3A 60 15
ally, we evaluated whether the number of cells 40 10
Transcription 20 5
recovered at each concentration could be used NF-kB NF-kB
GR p53 0 0
Factors
to infer toxicity akin to traditional screens. 0 0.10.5 1 5 10 50100 µM 0 0.10.5 1 5 10 50100 µM
0
05
25
50
5
5
5
25
0
05
25
5
5
5
25
50
0.
2.
0.
2.
0.
0.
0.
0.
Chromatin Dexamethasone
After fitting a response curve to the recovered Modifier
NF-kB
HDAC SAHA
cellular counts, we inferred a “viability score”
Cells (percent)
CDKN1A TP53I3
E 100
from sci-Plex data, a metric that was concor- 75 75
dant with “gold standard” measurements (Fig. 50
25
50
25
2F and fig. S2, G to I). 0 0
µM µM
25
25
25
5
.5
5
0
5
0
25
25
.5
25
5
0
2.
12
25
2.
12
25
B
12
12
1.
0.
0.
1.
Nutlin-3A
sci-Plex scales to thousands of samples and
4
Cells (percent)
Component 2
enables HTS Treatment BBC3 PMAIP1
BMS345541 30 40
To assess how sci-Plex scales for HTS, we per- 0 Dex 20 30
20
Nutlin3A 10 10
formed a screen of 188 compounds targeting a SAHA 0 0
µM
diverse range of enzymes and molecular path- µM
0
25
25
5
.5
25
5
0
0
25
25
5
.5
25
5
0
2.
12
25
2.
12
25
12
12
0.
1.
0.
1.
ways (Fig. 3A). Half of this panel was chosen Nutlin-3A
to target transcriptional and epigenetic regu- 0 5
lators. The other half was chosen to sample

Component 1 F BMS Dex
diverse mechanisms of action. We exposed C BMS345541 Dex

three well-characterized human cancer cell 4
lines, A549 (lung adenocarcinoma), K562 0
Dose
Cell counts
(chronic myelogenous leukemia), and MCF7
Component 2
10
00
05
50
10
0.
10
0.
(mammary adenocarcinoma), to each of these
Nutlin-3A SAHA
188 compounds at four doses (10 nM, 100 nM,
1 mM, and 10 mM) in duplicate, randomizing 4
compounds and doses across well positions Vehicle
0
in replicate culture plates (table S3). These
conditions, together with vehicle controls,
10
00
25
25
0
0 5 0 5
10
2.
25
10
0.
accounted for 4608 of 4992 independently Component 1 Log(Dose [µM])
treated cell populations in this experiment.
After treatment, we lysed cells to expose nu- Fig. 2. sci-Plex enables multiplex chemical transcriptomics at single-cell resolution. (A) Diagram
clei, hashed them with a specific combination depicting compounds and corresponding targets assayed within the pilot sci-Plex experiment. A549 lung
of two oligos (fig. S4A), and performed sci- adenocarcinoma cells were treated with either vehicle [dimethylsulfoxide (DMSO) or ethanol] or one
RNA-seq3 (21). After sequencing and filtering of four compounds (BMS345541, dexamethasone, nutlin-3a, or SAHA). (B) UMAP embedding of
based on hash purity (fig. S4, B to F), we ob- chemically perturbed A549 cells colored by drug treatment. (C) UMAP embedding of chemically
tained transcriptomes for 649,340 single cells, perturbed A549 cells faceted by treatment with cells colored by dose. (D and E) Expression of a
with median mRNA UMI counts of 1271, 1071, canonical (D) glucocorticoid receptor activated (ANGPTL4) and repressed (GDF15) target genes
and 2407 for A549, K562, and MCF7, respec- as a function of dexamethasone dose or (E) p53 target genes as a function of nutlin-3a dose. y-axes
tively (fig. S5A). The aggregate expression pro- indicate the percentage of cells with at least one read corresponding to the transcript. (F) Dose–response
files for each cell type were highly concordant viability estimates for BMS345541-, dexamethasone-, nutlin-3a-, and SAHA-treated A549 cells on the
between replicate wells (Pearson correlation = basis of the relative number of cells recovered at each dose.
0.99) (fig. S5B).
Visualizing sci-RNA-seq profiles separately
for each cell line revealed compound-specific A and epothilone B were associated with three sizes were diminished (fig. S9). Moreover,
transcriptional responses and patterns that such focal enrichments, each composed of cells sci-Plex dose-dependent effect profiles cor-
were common to multiple compounds. For each from both compounds at all four doses (fig. S7, related with compound-matched L1000 mea-
of the cell lines, UMAP projected most cells F and G). The cells in each focus were distinct surements (11) (fig. S10).
into a central mass, flanked by smaller clusters from one another, but transcriptionally simi- Genes associated with the cell cycle were
(Fig. 3B). These smaller clusters were largely lar to other treatments: a recently identified highly variable across individual cells, and
composed of cells treated with compounds microtubule destabilizer, rigosertib (27); the many drugs reduced the fraction of cells that
from only one or two compound classes (figs. SETD8 inhibitor UNC0397; or untreated pro- expressed proliferation marker genes (figs.
S6 and S7, A to C). For example, A549 cells liferating cells (fig. S7H). S11 and S12). In principle, scRNA-seq should
treated with triamcinolone acetonide, a syn- We next assessed the effects of each drug be able to distinguish shifts in the propor-
thetic glucocorticoid receptor agonist, were on the “population average” transcriptome of tion of cells in distinct transcriptional states
markedly enriched in one such small cluster, each cell line. In total, 6238 genes were dif- from gene-regulatory changes within those
comprising 95% of its cells [Fisher’s exact ferentially expressed in a dose-dependent man- states. By contrast, bulk transcriptome profiling
test, false discovery rate (FDR) < 1%; fig. S7, D ner in at least one cell line (FDR < 5%; fig. S8 would confound these two signals (fig. S13A)
and E]. Although many drugs were associated and tables S4 and S5). Bulk RNA-seq mea- (14). We therefore tested for dose-dependent
with a seemingly homogeneous transcription- surements collected for five compounds across differential expression on subsets of cells cor-
al response, we also identified cases in which four doses and vehicle agreed with averaged responding to the same drug but expressing
distinct transcriptional states were induced gene expression values and estimated effect high versus low levels of proliferation marker
by the same drug. For example, in A549, the sizes across identically treated single cells, genes (fig. S13B). Correlation between the dose-
microtubule-stabilizing compounds epothilone although correlations between small effect dependent effects on the two fractions of each

cell type varied across drug classes (fig. S13C), viability of 50% or more (Fig. 3C and fig. S5C). boring a constitutively active BCR-ABL fusion
with some frankly discordant effects for indi- Among the drugs that reduced viability, we kinase (28) and an observed increased sensi-
vidual compounds (fig. S13D). Viability analysis observed a higher sensitivity of K562 to the Src tivity of hematopoietic and lymphoid cancer
performed as in the pilot experiment revealed and Abl inhibitor bosutinib (Fig. 3C), a result cell lines to Abl inhibitors (29) (fig. S14B).
that after drug exposure at the highest dose, that we confirmed by cell counting (fig. S14A). To assess whether each compound elicited
only 52 (27%) compounds caused a decrease in This result is consistent with K562 cells har- similar responses across the three cell lines,
Fig. 3. sci-Plex enables global transcriptional profiling of thousands of compound or dose combinations that did not appreciably alter the corresponding
chemical perturbations in a single experiment. (A) Schematic of the large- cells’ distribution in UMAP space compared with vehicle controls (Fisher’s
scale sci-Plex experiment (sci-RNA-seq3). A total of 188 small molecules exact test, FDR < 1%). (C) Viability estimates obtained from hash-based counts
were tested for their effects on A549, K562, and MCF7 human cell lines, each at of nuclei at each dose of selected compounds (bosutinib is highlighted in
four doses and in biological replicate, after 24 hours of treatment. The plate red text). Rows represent compound doses increasing from top to bottom,
positions of doses and drugs were varied between replicates, and a median and columns represent individual compounds. Annotation bar at top depicts the
of 100 to 200 cells were recovered per condition. Colors demarcate cell line, broad cellular activity targeted by each compound. (D) UMAP embeddings
compound pathway, and dose. (B) UMAP embeddings of A549, K562, and MCF7 highlighted by treatment with the MEK inhibitor trametinib (red), an HSP90
cells in our screen with each cell colored by the pathway targeted by the inhibitor (purple), or vehicle control (gray). (E) HSP90AA1 expression levels
compound to which a given cell was exposed. To facilitate visualization of in cells exposed to increasing doses of trametinib. y-axes indicate the
significant molecular phenotypes, we added transparency to cells treated with percentage of cells with at least one read corresponding to the transcript.

we clustered compounds using the effect sizes cells and vehicle-treated cells from all three cell tion needed to drive a cell halfway across the
for dose-dependent genes as loadings in each lines using a mutual-nearest neighbor (MNN) HDAC inhibitor pseudodose trajectory (fig.
cell line (figs. S15 to S18). Joint analysis of matching approach (32) to produce a con- S27A and table S6). To compare the transcrip-
the three cell lines revealed common and cell- sensus HDAC inhibitor trajectory, which we tionally derived measures of potency with the
type–specific responses to different compounds call “pseudodose” [analogous to “pseudotime” biochemical properties of each compound, we
(figs. S19 and S20). For example, trametinib, a (33)] (Fig. 4A and fig. S21). We observed that collected published median inhibitory concen-
mitogen-activated protein kinase kinase (MEK) some HDAC inhibitors induced homogeneous tration (IC50) values for each compound from
inhibitor, induced a transcriptionally distinct responses, with nearly all cells localized to a in vitro assays performed on eight purified
response in MCF7 cells. Inspection of UMAP relatively narrow range of the HDAC inhibitor HDAC isoforms (table S7). With the excep-
projections revealed trametinib-treated MCF7 trajectory at each dose (e.g., pracinostat in tion of two relatively insoluble compounds,
cells interspersed among vehicle controls, re- A549), whereas other drugs induced much our calculated TC50 values increased as a func-
flecting limited effects. By contrast, trametinib- greater cellular heterogeneity (Fig. 4B and tion of compound IC50 values (Fig. 4C and
treated A549 and K562 cells, which harbor fig. S22). fig. S27, B and C).
activating KRAS and ABL mutations (30), re- Such heterogeneity could be explained by To assess the components of the HDAC
spectively, were tightly clustered, consistent cells executing a defined transcriptional pro- inhibitor trajectory, we performed differen-
with a strong, specific transcriptional response gram asynchronously, with the dose of drug tial expression analysis using pseudodose as a
to inhibition of MEK signaling by trametenib that the cells are exposed to modulating the continuous covariate. Of the 4308 genes that
(Fig. 3D). Further, we observed that these A549 rates of their progression through it. To test this were significantly differentially expressed
and K562 cells appeared proximal to clusters hypothesis, we sequenced the transcriptomes over this consensus trajectory, 2081 (48%) re-
enriched with inhibitors of HSP90, a key chap- of 64,440 A549 cells that were treated for sponded in a cell-type–dependent manner and
erone for protein folding (Fig. 3D). This obser- 72 hours with one of 48 compounds, includ- 942 (22%) exhibited the same pattern in all
vation was corroborated by concordant changes ing many of the HDAC inhibitors from the three cell lines (fig. S28, A and B, and table S8).
in HSP90AA1 expression in trametinib-treated large sci-Plex screen. Upon accounting for One prominent pattern shared by the three
cells (Fig. 3E). Analysis of Connectivity Map confluency-dependent cell-cycle effects and cell lines was an enrichment for genes and
data (11, 12) revealed further evidence that MEK MNN alignment (figs. S23 and S24), the co- pathways indicative of progression toward
inhibitors do indeed induce highly similar gene embedded UMAP projection revealed new cell-cycle arrest (figs. S28C and S29, A and B).
expression signatures to HSP90 perturbations focal concentrations of cells at 72 hours that DNA content staining and flow cytometry
(fig. S14C), especially in A549 but not in MCF7 were not evident at the 24-hour time point, confirmed that HDAC inhibition resulted in
(fig. S14, D and E). These results are concor- e.g., SRT1024 (fig. S25). However, for the the accumulation of cells in the G2/M phase
dant with previous observations of the reg- majority of HDAC inhibitors tested, we did of the cell cycle (34) (fig. S29, C and D).
ulation of HSP90AA1 downstream of MEK not observe that cells at a given dose moved The shared response to HDAC inhibition
signaling (31) and suggest that similarity in farther along an aligned HDAC trajectory at included not only cell-cycle arrest but also the
single-cell transcriptomes treated with distinct 72 hours (fig. S26). This suggests that the altered expression of genes involved in cellular
compounds can highlight drugs that target dose of many HDAC inhibitors governs the metabolism (fig. S28C). Histone acetyltrans-
convergent molecular pathways. magnitude of a cell’s response rather than its ferases and deacetylases regulate chromatin
rate of progression and that any observed accessibility and transcription factor activity
Inference of chemical and mechanistic heterogeneity cannot be attributed solely to through the addition or removal of charged
properties of HDAC inhibitors asynchrony (fig. S26). acetyl groups (35–37). Acetate, the product
For each of the three cell lines, the most prom- Next, we assessed whether a given HDAC of HDAC class I-, II-, and IV-mediated his-
inent compound response was composed of inhibitor’s target affinity explained its global tone deacetylation and a precursor to acetyl-
cells treated with one of 17 HDAC inhibitors transcriptional response to the compound. coenzyme A (acetyl-CoA), is required for histone
(Fig. 3B, dark blue, and table S6). To assess We used dose-response models to estimate acetylation but also has important roles in
the similarity of the dose–response trajectories each compound’s transcriptional median effec- metabolic homeostasis (23, 38, 39). Inhibi-
between cell lines, we aligned HDAC-treated tive concentration (TC50), i.e., the concentration of nuclear deacetylation limits recycling
Fig. 4. HDAC inhibitor trajectory captures cellular heterogeneity in drug response and biochemical affinity. (A) MNN alignment and UMAP embedding of
transcriptional profiles of cells treated with one of 17 HDAC inhibitors. Pseudodose root is displayed as a red dot. (B) Ridge plots displaying the distribution of cells along
pseudodose by dose shown for three HDAC inhibitors with varying biochemical affinities. (C) Relationship between TC50 and average log10(IC50) from in vitro
measurements. Asterisks indicate compounds with a solubility <200 mM (in DMSO) that were not included in the fit.

Fig. 5. HDAC inhibitors shared transcriptional response indicative of acetyl-CoA deprivation. (A) Heatmap of row-centered and z-scaled gene expression depicting
the up-regulation of pseudodose-dependent genes involved in cellular carbon metabolism. (B) Diagram of the roles of genes from (A) in cytoplasmic acetyl-CoA
regulation. Red circles indicate acetyl groups. Enzymes are shown in gray. Transporters are shown in green (FA, fatty acid; Ac-CoA, acetyl-CoA; C, citrate).
of chromatin-bound acetyl groups for both inhibitor trajectory, although no such shift to act through mechanisms directly involving
catabolic and anabolic processes (39). Accord- was observed in A549 (fig. S31, G, K, I, and chromatin regulation, our data support an
ingly, we observed that HDAC inhibition led M). Taken together, these results suggest that alternative model, albeit not a mutually exclu-
to sequestration of acetate in the form of a major feature of the response of cells to sive one, in which HDAC inhibitors impair
markedly increased acetylated lysine levels af- HDAC inhibitors, and possibly their associated growth and proliferation by interfering with
ter exposure to a 10 mM dose of the HDAC in- toxicity, is the induction of an acetyl-CoA– a cancer cell’s ability to draw acetate from
hibitors pracinostat and abexinostat (fig. S30). deprived state. chromatin (22, 23, 39). As such, variation in
Upon further inspection of pseudodose- cells’ acetate reservoirs is a potential expla-
dependent genes, we observed that enzymes Discussion nation for their heterogeneous responses to
critical for cytoplasmic acetyl-CoA synthesis Here, we present sci-Plex, a massively multiplex HDAC inhibitors.
from either citrate (ACLY) or acetate (ACSS2) platform for single-cell transcriptomics. sci-Plex As the cost of single-cell sequencing con-
were up-regulated (Fig. 5A). Genes involved in uses chemical fixation to cost-effectively and tinues to fall, the opportunities for leverag-
cytoplasmic citrate homeostasis (GLS, IDH1, irreversibly label nuclei with short, unmodified ing sci-Plex for basic and applied goals in
and ACO1), citrate cellular import (SLC13A3), ssDNA oligos. In the proof-of-concept exper- biomedicine may be substantial. The proof-of-
and mitochondrial citrate production and ex- iment described here, we applied sci-Plex to concept experiments described here, consist-
port (CS, SLC25A1) were also up-regulated. quantify the dose-dependent responses of can- ing of nearly 5000 independent treatments
Up-regulation of SIRT2, which deacetylates tu- cer cells to 188 compounds through an assay with transcriptional profiling of >100 single
bulin, was also observed in response to HDAC that is both high content (global transcription) cells per treatment, can potentially be scaled
inhibition. and high resolution (single cell). By profiling toward a comprehensive, high-resolution atlas
Together with increases in chromatin-bound several distinct cancer cell lines, we distin- of cellular responses to pharmacologic pertur-
acetate, these transcriptional responses sug- guished between shared and cell-line–specific bations (e.g., hundreds of cell lines or genetic
gest a metabolically consequential depletion of molecular responses to each compound. backgrounds, thousands of compounds, mul-
cellular acetyl-CoA reserves in HDAC-inhibited sci-Plex offers some distinctive advantages tichannel single-cell profiling, etc.). The ease
cells (Fig. 5B). To validate this further, we over conventional HTS: it can distinguish a and low cost of oligo hashing, coupled with
sought to shift the distribution of cells along compound’s distinct effects on cellular subsets the flexibility and exponential scalability of
the HDAC inhibitor trajectory by modulating (including complex in vitro systems such as single-cell combinatorial indexing, would facil-
cellular acetyl-CoA levels. We treated A549 and cellular reprogramming, organoids, and syn- itate this goal.
MCF7 cells with pracinostat in the presence thetic embryos); it can unmask heterogeneity
and absence of acetyl-CoA precursors (acetate, in cellular response to a perturbation; and REFERENCES AND NOTES
pyruvate, or citrate) or inhibitors of enzymes it can measure how drugs shift the relative 1. J. R. Broach, J. Thorner, Nature 384 (Suppl), 14–16
(ACLY, ACSS2, or PDH) involved in replen- proportions of transcriptionally distinct sub- (1996).
ishing acetyl-CoA pools. After treatment, cells sets of cells. Highlighting these features, our 2. D. A. Pereira, J. A. Williams, Br. J. Pharmacol. 152, 53–61
(2007).
were harvested and processed using sci-Plex study provides insight into the mechanism of 3. D. Shum et al., J. Enzyme Inhib. Med. Chem. 23, 931–945
and trajectories constructed for each cell line action of HDAC inhibitors. Specifically, we (2008).
(figs. S31 and S32). In both A549 and MCF7 find that the main transcriptional responses to 4. C. Yu et al., Nat. Biotechnol. 34, 419–423 (2016).
5. Z. E. Perlman et al., Science 306, 1194–1198 (2004).
cells, acetate, pyruvate, and citrate supplemen- HDAC inhibitors involve cell-cycle arrest and
6. Y. Futamura et al., Chem. Biol. 19, 1620–1630 (2012).
tation was capable of blocking pracinostat- marked shifts in genes related to acetyl-CoA 7. J. Kang et al., Nat. Biotechnol. 34, 70–77 (2016).
treated cells from reaching the end of the metabolism. For some HDAC inhibitors, we 8. K. L. Huss, P. E. Blonigen, R. M. Campbell, J. Biomol. Screen.
HDAC inhibitor trajectory (fig. S31, F, J, H, observed clear heterogeneity in responses 12, 578–584 (2007).
9. C. Ye et al., Nat. Commun. 9, 4307 (2018).
and L). In MCF7 cells, both ACLY and ACSS2 observed at the single-cell level. Although 10. E. C. Bush et al., Nat. Commun. 8, 105 (2017).
inhibition shifted cells farther along the HDAC HDAC inhibition is conventionally thought 11. A. Subramanian et al., Cell 171, 1437–1452.e17 (2017).

12. J. Lamb et al., Science 313, 1929–1935 (2006). MALARIA

13. M. B. Elowitz, A. J. Levine, E. D. Siggia, P. S. Swain,
Science 297, 1183–1186 (2002).
14. C. Trapnell, Genome Res. 25, 1491–1498 (2015). A Kelch13-defined endocytosis pathway mediates
15. S. M. Shaffer et al., Nature 546, 431–435 (2017).
16. S. L. Spencer, S. Gaudet, J. G. Albeck, J. M. Burke, P. K. Sorger,
Nature 459, 428–432 (2009).
artemisinin resistance in malaria parasites
17. M. Stoeckius et al., Genome Biol. 19, 224 (2018).
18. J. Gehring, J. H. Park, S. Chen, M. Thomson, L. Pachter, Highly Jakob Birnbaum1*†, Sarah Scharf1†, Sabine Schmidt1, Ernst Jonscher1,
multiplexed single-cell RNA-seq for defining cell population
Wieteke Anna Maria Hoeijmakers2, Sven Flemming1‡, Christa Geeke Toenhake2, Marius Schmitt1,
and transcriptional spaces. bioRxiv 315333 [Preprint]
5 May 2018. doi.org/10.1101/315333. Ricarda Sabitzki1, Bärbel Bergmann1, Ulrike Fröhlke1, Paolo Mesén-Ramírez1,
19. C. S. McGinnis et al., Nat. Methods 16, 619–626 (2019). Alexandra Blancke Soares1, Hendrik Herrmann1§, Richárd Bártfai2, Tobias Spielmann1¶
20. D. Shin, W. Lee, J. H. Lee, D. Bang, Sci. Adv. 5, eaav2249
(2019).
21. J. Cao et al., Nature 566, 496–502 (2019). Artemisinin and its derivatives (ARTs) are the frontline drugs against malaria, but resistance
22. M. A. McBrian et al., Mol. Cell 49, 310–321 (2013). is jeopardizing their effectiveness. ART resistance is mediated by mutations in the parasite’s Kelch13
23. S. A. Comerford et al., Cell 159, 1591–1602 (2014). protein, but Kelch13 function and its role in resistance remain unclear. In this study, we identified
24. D. A. Cusanovich et al., Science 348, 910–914 (2015).
25. J. Cao et al., Science 357, 661–667 (2017). proteins located at a Kelch13-defined compartment. Inactivation of eight of these proteins, including
26. L. McInnes, J. Healy, UMAP: Uniform Manifold Approximation Kelch13, rendered parasites resistant to ART, revealing a pathway critical for resistance. Functional
and Projection for Dimension Reduction. arXiv:1802.03426 analysis showed that these proteins are required for endocytosis of hemoglobin from the host cell.
[stat.ML] (9 February 2018).
27. M. Jost et al., Mol. Cell 68, 210–223.e6 (2017).
Parasites with inactivated Kelch13 or a resistance-conferring Kelch13 mutation displayed reduced
28. G. Grosveld et al., Mol. Cell. Biol. 6, 607–616 (1986). hemoglobin endocytosis. ARTs are activated by degradation products of hemoglobin. Hence, reduced
29. E. K. Greuber, P. Smith-Pearson, J. Wang, A. M. Pendergast, activity of Kelch13 and its interactors diminishes hemoglobin endocytosis and thereby ART activation,
Nat. Rev. Cancer 13, 559–571 (2013).
30. J. Barretina et al., Nature 483, 603–607 (2012).
resulting in parasite resistance.
31. C. Dai et al., J. Clin. Invest. 122, 3742–3754 (2012).
L
32. L. Haghverdi, A. T. L. Lun, M. D. Morgan, J. C. Marioni,
Nat. Biotechnol. 36, 421–427 (2018). ife-threatening malaria is caused by mechanism explaining the reduced suscepti-
33. C. Trapnell et al., Nat. Biotechnol. 32, 381–386 (2014).
34. W. Brazelle et al., PLOS ONE 5, e14335 (2010).
Plasmodium falciparum parasites con- bility to ART in resistant parasites.
35. J.-S. Roe, F. Mercan, K. Rivera, D. J. Pappin, C. R. Vakoc, tinuously multiplying in human red blood
Mol. Cell 58, 1028–1039 (2015). cells. Artemisinin and its derivatives The Kelch13 compartment and associated
36. J. E. Brownell et al., Cell 84, 843–851 (1996).
(ARTs) are the recommended first-line proteins defined by a novel BioID approach
37. J. Taunton, C. A. Hassig, S. L. Schreiber, Science 272, 408–411
(1996). drugs against malaria (1), but their usefulness We previously found endogenous Kelch13
38. S. K. Kurdistani, Curr. Opin. Genet. Dev. 26, 53–58 is threatened by “resistance” (2), defined as tagged with green fluorescent protein (GFP)
(2014). reduced clearance of parasites in ART-treated (i.e., expressed from its genomic locus) at foci
39. K. E. Wellen et al., Science 324, 1076–1080 (2009).
malaria patients (3). ART resistance mani- in close proximity to the parasite’s food vacuole
ACKN OW LEDG MEN TS fests as a decreased susceptibility of young (FV) (17), a lysosome-like compartment where
We thank members of the Shendure lab; the Trapnell lab; and ring-stage parasites to a short pulse of ART, parasites digest endocytosed host cell cytosol
others, particularly A. Adey, G. Booth, A. Hill, S. Henikoff, a property linked to delayed clearance of para- (predominantly hemoglobin). To better define
K. Cherukumilli, P. Selvaraj, and T. Zhou, for helpful suggestions,
discussion, and mentorship; and D. Prunkard and A. Leith for sites and recrudescence of disease in patients the cellular location of Kelch13, we colocalized
assistance in flow sorting. Funding: This work was funded by (4–6). This can be measured in vitro using a it with a series of markers (Fig. 1A and fig. S1).
grants from the NIH (DP1HG007811 and R01HG006283 to ring-stage survival assay (RSA) (4). This confirmed the proximity of Kelch13 foci
J.S.; DP2 HD088158 to C.T.), the W. M. Keck Foundation (to C.T.
and J.S.), the NSF (DGE-1258485 to S.R.S), and the Paul G. Allen ART resistance is primarily associated with to the FV. Although Kelch13 foci rarely over-
Frontiers Group (to J.S. and C.T.). J.S. is an Investigator of point mutations in the parasite’s Kelch pro- lapped with secretory pathway markers [endo-
the Howard Hughes Medical Institute. Author contributions: S.R.S., peller protein Kelch13 (7, 8). At present, there plasmic reticulum (ER) and Golgi], they were
J.L.M.-F., V.R., J.S., and C.T. conceived the project; S.R.S., J.L.M.-F.,
and V.R. designed experiments; F.Z., L.C., and F.S. provided
is no coherent picture of the mechanism of frequently in the vicinity of parts of the ER that
reagents and performed sequencing; S.R.S., J.L.M.-F., and C.T. ART resistance in the parasite, which, among were near the FV. Kelch13 foci were also found
analyzed the data; and S.R.S., J.L.M.-F., V.R., J.S., and C.T. wrote the others, has been connected to increased cell- close to the apicoplast, which itself is located
manuscript. Competing interests: L.C., F.Z., and F.S. declare
competing financial interests in the form of stock ownership and
ular stress, an activated unfolded protein rein proximity of the FV. Notably, the resistance-
paid employment by Illumina, Inc. One or more embodiments of one sponse, reduced protein translation, altered conferring form of Kelch13 (Kelch13C580Y;
or more patents and patent applications filed by Illumina and the DNA replication, and increased levels of phos- Cys580→Tyr) fully colocalized with episomally
University of Washington may encompass the methods, reagents,
phatidylinositol 3-phosphate (9–16). Condi- expressed wild-type Kelch13 (wtKelch13) (Fig.
and data disclosed in this manuscript. Data and materials
availability: Processed and raw data can be downloaded from tional inactivation of Kelch13 causes an arrest 1A, fig. S1, and table S1), indicating that re-
NCBI GEO (#GSE139944). Code used to perform analyses can of parasites in ring stages (17). However, the sistance does not involve an altered cellular
be accessed on Zenodo and https://github.com/cole-trapnell-lab/ cellular function of Kelch13 and its involvement distribution. Kelch13C580Y also colocalized with
sci-plex. All methods for making the transposase complexes
used in this paper are described in Cao et al. (25); however, in ART resistance remain enigmatic. Here, we episomally expressed hemagglutinin (HA)–
Illumina will provide transposase complexes in response to study Kelch13 and identify its interactors and tagged Kelch13 (fig. S1). Overall, this analysis
reasonable requests from the scientific community subject to a their function. This reveals an entire pathway suggests that Kelch13 is located at an unknown
material transfer agreement.
in ART resistance and a Kelch13-dependent compartment or cellular structure (henceforth,
Kelch13 compartment) not identified by gen-
SUPPLEMENTARY MATERIAL 1 eral markers of the secretory system.
Bernhard Nocht Institute for Tropical Medicine, Bernhard
science.sciencemag.org/content/367/6473/45/suppl/DC1 Nocht Str. 74, 20359 Hamburg, Germany. 2Department To gain insight into the Kelch13 compart-
Materials and Methods of Molecular Biology, Radboud University, Geert Grooteplein
Figs. S1 to S32 ment, we used a novel version of BioID (18),
26-28, 6525 GA Nijmegen, Netherlands.
Tables S1 to S8 *Present address: Medical Research Council Laboratory of which we call dimerization-induced quantita-
References (40–45) Molecular Biology, Francis Crick Avenue, Cambridge, UK. †These tive BioID (DiQ-BioID), to identify Kelch13 in-
authors contributed equally to this work. ‡Present address: teractors and compartment neighbors in the
8 April 2019; accepted 18 November 2019 Institute of Physics and Astronomy, Karl-Liebknecht-Straße 24/25,
Published online 5 December 2019 University of Potsdam, 14476 Potsdam, Germany. §Deceased living cell. In DiQ-BioID, the biotin ligase BirA*
10.1126/science.aax6234 ¶
Corresponding author. Email: spielmann@bnitm.de is not directly fused to the target protein but is

expressed separately (the biotinylizer) and con- titative mass spectrometry. We used this ap- parasites resulted in a list of proteins enriched
ditionally recruited to the target (using rapa- proach in parasites expressing FK506 binding with very high confidence over control (Fig. 1D,
log), where it biotinylates interaction partners protein (FKBP)–tagged Kelch13 from the ge- fig. S2, and data S1). Kelch13 was the top hit,
(Fig. 1B). The same parasite culture grown nomic locus (17). Induction of dimerization confirming that the biotinylizer was success-
without rapalog (i.e., BirA* not on the target) with rapalog efficiently recruited the biotinyl- fully recruited to Kelch13, and a band consistent
serves as a control (Fig. 1B), permitting highly izer to the Kelch13 foci (Fig. 1C), and quan- in size with the tagged Kelch13 was detected
specific identification of interactors by quantitative mass spectrometry of asynchronous in anti-biotin blots (fig. S2B). The other hits
Fig. 1. Kelch13 location and interactome. (A) Fluorescence microscopy images significant experiment. Asterisks indicate proteins previously suspected in ART
showing parasites with endogenously GFP-tagged Kelch13 (K13) (17) with resistance. Figure S2 shows full plot and replicas with a different biotinylizer.
coexpressed STVR-SDEL (ER), Sec13p (ER exit sites), P40 (phosphatidylinositol (E) Fluorescence microscopy of parasites with endogenously 2xFKBP-GFP-2xFKBP–
3-phosphate sensor: food vacuole), ACP (apicoplast), or endogenously GFP-tagged tagged Eps15 episomally expressing mCherry-Kelch13. (F) Immunoprecipitation
mutated Kelch13 (K13C580Y) with episomal mCherry-Kelch13wt. See fig. S1 for (IP) of episomally expressed Kelch13 in parasites endogenously expressing
all panels. (Bottom) Most common arrangement of markers in single-nucleus GFP-Eps15. I, input extract; P, post IP extract; W, last wash; E, eluate. Full blots
trophozoites. (B) Scheme of DiQ-BioID. (C) Rapalog-dependent recruitment of the and replicas in fig. S4. (G) Top-right quadrant of scatter plot of Eps15 DiQ-BioID,
biotinylizer (BirA*-FRB) to endogenously GFP- and FKBP-tagged Kelch13 (FKBP- as in (D). Figure S5 shows full plot and independent replicas. (H) Fluorescence
K13). (D) Top-right quadrant of scatter plot of Kelch13 DiQ-BioID, showing proteins microscopy in parasites with endogenously GFP-tagged DiQ-BioID hits (KIC1-10 or
enriched (log2 ratio) on rapalog (biotinylizer on target) compared with control short name) colocalized with episomal mCherry-Kelch13. Arrowheads indicate
(biotinylizer cytoplasmic). Significantly enriched proteins are indicated (short, unique Kelch13 foci overlapping with the KIC signal (all panels in fig. S6). Scale bars, 5 mm.
names or PlasmoDB identifiers). False discovery rate (FDR) indicated for the least Merge, merged green and red channel; DAPI, 4′,6-diamidino-2-phenylindole (nuclei).

[designated as Kelch13 interaction candidates To further validate the Kelch13 DiQ-BioID localizing proteins, eight also colocalized in
(KICs), unless already otherwise named] were results and the method per se, we carried out the resistance-relevant ring stage, whereas the
predominantly proteins of unknown function. a reverse DiQ-BioID with the endogenously other two (KIC8 and KIC9) were not detected
These included proteins that had previously tagged Eps15 (Fig. 1G and fig. S5). This resulted in that stage (fig. S6 and table S1). Thus DiQ-
been suspected in ART resistance or were in in a list of high-confidence hits that extensively BioID identifies interactors and compartment
that context highlighted in genome-wide as- overlapped with the Kelch13 DiQ-BioID (all 16 neighbors in living cells with high specificity
sociation studies—for example, an Eps15-like top hits were also found in the high-confidence and in this study defined a series of proteins
protein (PF3D7_1025000, originally annotated hits of the K13 DiQ-BioID), including Kelch13 proximal to the Kelch13-Eps15 complex.
as Formin2) (19); ubiquitin carboxyl-terminal and UBP1 (Figs. 1G and 2C and data S1). This
hydrolase 1 (UBP1) (19–23); KIC6 (PF3D7_ suggested that the hits are indeed interactors Kelch13-Eps15 marks an AP-2 compartment
0609700) (20); and MyosinC (MyoC) (19) (Fig. or compartment neighbors of the Kelch13- devoid of clathrin
1D, fig. S2, and data S1)—suggesting that the Eps15 complex and that the DiQ-BioID ex- Because Eps15 in other organisms is typically
interactome identified a pathway that may be periments reached a high depth of coverage. involved in endocytosis (24, 25), we hypothe-
relevant for ART resistance. Modification of This was confirmed by tagging the corre- sized that the Kelch13 compartment might also
the genomic locus to express the endogenous sponding endogenous genes with the sequence be involved in this process. In malaria blood
protein fused to GFP (fig. S3) showed that the encoding GFP (fig. S3), which showed that, be- stages, endocytosis mediates the large-scale
top hit of these previously highlighted pro- sides Eps15, 10 of 12 tested Kelch13 hits colo- uptake of host cell cytosol (consisting predom-
teins, the Eps15-like protein (henceforth Eps15), calize with Kelch13 compartment foci (Fig. 1H, inantly of hemoglobin), but the molecular
colocalized with Kelch13 (Fig. 1E), and coim- fig. S6, and table S1), demonstrating the power details of this process are poorly understood
munoprecipitation confirmed that Eps15 in- of the method. Only PFK9 and KIC10 did not (26). In model organisms, the most com-
teracts with Kelch13 (Fig. 1F and fig. S4). show an overlap with Kelch13. Of the 10 co- mon and best studied form of endocytosis is
Fig. 2. Kelch13 complex locates with AP-2m but is distinct to clathrin complex.
(A) Fluorescence microscopy images of parasites endogenously expressing 2xFKBP-GFP–tagged
clathrin heavy chain (CHC) or AP-2m with episomal mCherry-Kelch13. Arrows mark overlap,
and arrowheads mark similar region of Kelch13 foci. Scale bar, 5 mm. Merge, merged green
and red channel; DAPI (nuclei). (B) Top-right quadrant of scatter plot of CHC DiQ-BioID.
Figure S7 shows full plot and replicas with a different biotinylizer. (C) Heatmap of four Kelch13,
Eps15, and CHC DiQ-BioID experiments showing proteins enriched in at least three of four
experiments with FDR < 1%. K-means clustering into two clusters highlights overlap of
Kelch13 and Eps15 hits (dark green cluster) compared with CHC hits (light green clusters).
Gray blocks indicate not identified or no ratio due to missing label. Red font indicates DiQ-BioID
baits, and orange font indicates proteins further analyzed.

mediated by clathrin, which depends on the was supported by DiQ-BioID using the bio- indicate the existence of an unusual clathrin-
AP-2 adaptor complex (27, 28). Although tinylizer alone (fig. S8 and data S2), and a independent endocytosis pathway in malaria
P. falciparum Eps15 contains regions specific lack of colocalization with Kelch13 was experi- parasites.
to malaria parasites, it shows typical hallmarks mentally confirmed in the case of PFK9 (Fig.
of Eps15 proteins, including AP-2 binding sites 1G and fig. S6). Therefore, we did not analyze Kelch13 compartment proteins are involved
(fig. S5B). To test whether the Kelch13-Eps15 these proteins further. in endocytosis
complex marks an endocytic structure, we Overall, these findings show that Kelch13 To investigate a possible role of the Kelch13-
colocalized Kelch13 with endogenously GFP- defines proteins at a compartment that colo- Eps15 compartment in endocytosis, suggested
tagged clathrin heavy chain (CHC) or AP-2m calizes with the classical endocytosis adap- by the presence of Eps15 and colocation with
(fig. S3). AP-2m–GFP foci colocalized with (Fig. tor AP-2m but that is entirely distinct from Ap-2m, we first carried out correlative light and
2A, arrowhead) or were found in the same re- clathrin, which typically associates with the electron microscopy (CLEM) with the para-
gion as Kelch13 foci (Fig. 2A, arrow), indicat- AP-2 complex in other organisms. This may sites expressing endogenously GFP-tagged
ing an endocytosis role for the Kelch13-Eps15
compartment. Unexpectedly, CHC was present
in foci that did not overlap with Kelch13 (Fig.
2A). This showed that Kelch13 colocalizes with
a compartment that contains AP-2 but is de-
void of clathrin. Whereas clathrin typically
has additional functions besides endocytosis
where it associates with other adaptors, AP-2 is
generally associated with clathrin-dependent
endocytosis (27, 28). AP-2 independent of clath-
rin is a highly unusual configuration that,
to our knowledge, was so far only observed in
Aspergillus nidulans (29). Of note, AP-2m has
previously been suspected to be involved in
ART resistance (30, 31).
To probe this distinction between clathrin
and the Kelch13-Eps15 complex and further
assess the specificity of the DiQ-BioID proce-
dure, we performed DiQ-BioID with the CHC
(fig. S7). This resulted in a list of high-confidence
hits differing from the Kelch13 and Eps15
DiQ-BioIDs and included known CHC inter-
actors such as the putative clathrin light chain
(even more enriched than the CHC itself) and
subunits of the AP-1 adaptor complex (Fig. 2B,
fig. S7, and data S1). Unexpectedly, subunits of
AP-4—in other organisms considered a clathrin-
independent adaptor of the trans-Golgi net-
work (32, 33)—were also enriched, whereas
AP-3 was only very mildly enriched, not reach-
ing statistical significance (data S1). AP-4 is
not well studied to date, but it is known to
interact with tepsin (34, 35). P. falciparum
tepsin (34) was a prominent hit in the CHC
DiQ-BioID, validating the AP-4 hits. This sug-
gests that CHC associates with AP-4 in malaria
parasites, either indicating differences com-
pared to other organisms or that DiQ-BioID
uncovers interactions not retained using pre- Fig. 3. Kelch13 compartment proteins are involved in endocytosis of host cell cytosol. (A) CLEM
viously used approaches in other organisms. with a cell expressing endogenously tagged Eps15-GFP: confocal image (scale bar, 2 mm), EM section and
Hence, DiQ-BioID reveals a credible CHC magnification thereof (scale bars, 0.5 mm). Arrowheads indicate vesicles containing material of host cell
interactome that is clearly distinct from the density. N, nucleus; FV, food vacuole. Details and two more cells in fig. S9. (B) Summary of Kelch13
Kelch13-Eps15 interactomes. The CHC DiQ- colocalization and SLI-TGD screen of DiQ-BioID hits. nd, not done. (C) Growth curves of parasites after
BioID also detected a different Kelch protein induction of KS. One of three independent experiments. Figure S10 shows replicas and details on KS.
(PF3D7_1205400), potentially indicating that (D) Example DIC images showing bloated (control) and nonbloated (inhibited endocytosis) FVs (partly
different trafficking complexes harbor dis- highlighted: dashed line and arrows) after KS (rapalog) and E64 treatment. (E) Quantification of number
tinct Kelch proteins. Only PFK9, HSP90 (heat of cells with bloated FVs after E64 treatment for control (no rapalog) and KS (rapa) for 8 hours (n for
shock protein 90), and FKBP35 were hits over- control/rapa: UPB1, 33/33; KIC7, 33/33; Kelch13, 32/33; AP-2m, 33/32; Eps15, 33/33; 2102, 33/33; 3D7
lapping with the Kelch13 and Eps15 DiQ-BioID NaN3, 34/34). Fisher’s exact test. Replicas in fig. S10. (F) Growth (cell diameter) of cells scored for bloated
experiments, suggesting that these hits might FVs after KS compared with control, data pooled from all three (five for NaN3) independent experiments,
result from the method rather than being with a total of n = 99/97 cells for no rapalog/rapa with UBP1, 101/97 with KIC7, 134/134 with Eps15, 99/97 with
biologically meaningful (Fig. 2, B and C). This AP-2m, 100/100 with 2102, and 170/170 for 3D7 ± NaN3. Unpaired, two-tailed t test. ns, not significant.

Eps15. This revealed that Eps15 resides in be inferred that Kelch13, Eps15, AP-2, UBP1, cell cytosol. In support of this, we noted that in
proximity of host cell cytosol-filled membra- and the nine colocalizing KICs are present >90% of cells with vesicle-like structures visi-
nous structures in the parasite in three of three proximal to host cell cytosol-filled structures ble in differential interference contrast (DIC),
analyzed cells and in five of five foci in these in the parasite, congruent with a role of the Kelch13 foci were found at such structures (fig.
cells (Fig. 3A and fig. S9, A to C). Hence, it can Eps15-Kelch13 complex in endocytosis of host S9D), suggesting that these may represent the
Fig. 4. A Kelch13-defined pathway in endocytosis and resistance.

(A) Fluorescence microscopy images of saponin-released young ring stages
(examples of spherical, partial, and full amoeboid parasite) grown for 3 to
6 hours in red blood cells preloaded with fluorescent dextran (Alexa647) show
internalized host cell cytosol (arrowheads). Scale bar, 5mm. (B to D) Quantification
of dextran uptake into 3-to-6-hour rings after inactivation (rapa) of Kelch13
by KS in the 3xNLS line (17) (B) or KIC7 (C) compared with control (no rapa) or [values from (42)], clustered according to similarity for the proteins analyzed
in the K13mut parasites compared with wild type (3D7) (D). Each point shows the in this study (see also fig. S13). Note that KIC2 was not part of the dataset.
amount of fluorescence internalized into one cell (arbitrary fluorescence units). Bottom shows data of colocalization, RSA, and endocytosis experiments (table S1;
Data pooled from three independent experiments (individual experiments in Fig. 4, B, F, and H; and Fig. 3E; respectively) as heatmaps (color shading
fig. S12) with a total of n = 138, 140, 132, 130, 122, and 114 cells for Kelch13 KS indicated; gray, not applicable). Asterisk indicates ring-stage endocytosis assay.
control, rapa, KIC7 control, KIC7 rapa, 3D7, and K13mut, respectively. Unpaired, (J) RSA with parasites harboring a mutated ubp1 gene encoding a change
two-tailed t test. P values and percent reduction of the mean are indicated. from Arg to His at amino acid position 3138 (19). Resistance in RSA defined
(E) RSAs after inactivating Kelch13 by KS (rapa) compared with control with as mean survival above 1% [green line in (F), (H), and (J)], a previously set
Kelch13 3xNLS parasites (17). (F) RSAs with the TGD lines indicated. (G) Growth standard (4). [(B) to (H), and (J)] Error bars show SD (not shown if close to
rate of TGD lines. (H) RSAs after KS (rapa) to inactivate the proteins indicated mean). P values indicated [only selected P values in (F) and (G)]; ns, not
compared with control or in 3D7 after inhibiting growth using NaN3. (I) Heatmap significant. [(E) to (H), and (J)] Unpaired, two-tailed t test with Welch’s
of transcription levels in 1-hour intervals across asexual blood stage development correction. Each point indicates an independent experiment.

Fig. 5. Reduced Kelch13 activity leads

to ART resistance. (A) Heatmap of DIQ-BioID
data of the four Kelch13 and the two
Kelch13C580Y experiments [hits with FDR
< 1% in at least two of four (Kelch13) or
two (Kelch13C580Y) experiments]. Color
intensities: 2nd to 98th percentile of log2
rapalog/control normalized ratios of all
proteins in this analysis. NaN (data S3) values
in gray. Rows were clustered and gene
identifiers were color coded as in Fig. 2C.
(B) Immunoblot-based quantification of
wtKelch13 (K13wt) and Kelch13C580Y
(K13C580Y) levels in parasites expressing the
otherwise-identical GFP-fused protein from
the endogenous locus. (Left) An example band
pair with Kelch13 levels relative to the loading
control (aldolase) shown as bars below the
blot (axis shows intensity ratio). (Right)
K13C580Y levels compared with K13wt (n = 3
independent experiments; each experiment is
shown by a point that derives from the
average of a serial dilution of K13wt and K13C580Y run on the same gel and normalized relative to the respective aldolase control). (C and D) Quantification of Kelch13 foci
fluorescence intensity in trophozoites (n = 105, 102 foci) and rings (n = 47, 51 foci) in K13wt and K13C580Y parasites, respectively. Data pooled from three independent
experiments; percent reduction of the mean is indicated. Example fluorescence images of ring stages are shown. (E) Fluorescence microscopy images of parasites
with endogenous Kelch13C580Y episomally expressing the construct indicated (PSR, plasmodium-specific region; L, linker) and RSA with the respective cell lines (right).
Cell lines expressing mChe-K13wt and mChe-K13C580Y were imaged after smearing, mChe-K13DPSR live. Size bar, 5 mm. (F) Model of resistance and summary of results of this
study. Asterisk indicates only in ring stages; nd, not determined (see also table S1). Bars show SD, mean is indicated. Unpaired, two-tailed t test with Welch's correction. P values
are indicated [in (E), only for selected pairs]; ns, not significant.

host cell cytosol-filled structures observed by S11). Hence, while not lethal, partial inactiva- 4E). Thus, Kelch13 inactivation reduces endo-
CLEM and likely are cytostomes. tion of Kelch13 (3xNLS line) leads to a much cytosis of host cell material and increases re-
To more directly assess the function of KICs, milder, but similar, effect compared with full sistance. We conclude that early ring stages
we first carried out a selection-linked integra- inactivation of Kelch13 (1xNLS line). already endocytose hemoglobin, and that this
tion gene disruption (SLI-TGD) screen (17) to process is reduced in parasites with a resistance-
establish which of the Kelch13 compartment Kelch13 influences endocytosis and resistance conferring mutation in Kelch13 or in which
proteins are essential for parasite survival. Of in ring stages Kelch13 is inactivated, and that this correlates
the 13 tested genes, only the ones encoding Hemoglobin by-products activate ART, and with ART resistance.
Eps15, UBP1, and KIC7 were refractory to dis- decreased hemoglobin digestion reduces ART
ruption and hence are likely essential for para- susceptibility (37–39). Hence, our data might Kelch13 compartment proteins are involved
site survival (Fig. 3B and fig. S3). Knock sideways indicate that reduced levels of endocytic up- in ART resistance
(KS) (17) to conditionally inactivate these pro- take of host cell cytosol could, by controlling Having established that inactivation of Kelch13
teins, as well as AP-2m, confirmed their impor- the supply of hemoglobin available for diges- induces ART resistance, we next assessed wheth-
tance for parasite growth (Fig. 3, B and C, tion and ART activation, be the mechanism of er other Kelch13 compartment proteins also
and fig. S10). Note that the smaller effect on ART resistance. However, this would need to play a role in ART resistance. First, we used
growth after inactivating Eps15 likely is the re- take place in the resistance-relevant ring stage the cell lines with the disrupted, nonessential
sult of inefficient KS (Fig. 3C and fig. S10). To (4). As UBP1 and KIC7 affected endocytosis in KICs. Whereas parasites with disrupted KIC1,
test whether Kelch13, UBP1, KIC7, Eps15, and trophozoites and showed a similar phenotype 2, 3, 6, 8, and 9, as well as KIC10 (which is not
AP-2m are involved in endocytosis, we condi- in rings to Kelch13 (fig. S11), we reasoned that Kelch13 compartment–associated), did not
tionally inactivated them by KS (fig. S10) and Kelch13 might influence endocytosis specifi- show an altered response to ART, disruption of
assessed hemoglobin uptake into the parasite cally in ring stages. However, as it is not clear KIC4, KIC5, and MCA-2 led to a reduced sus-
over 8 hours using a previously established whether young ring stages endocytose hemo- ceptibility in a standard RSA (Fig. 4F). While
assay, which results in a bloated FV phenotype globin at all (39–41), we decided to first clarify the reduced susceptibility correlated with a low
if endocytosis is operational (36). Inactivation this. Because of the overabundance of hemo- growth rate of the KIC5 and MCA-2 disruption
of UBP1, KIC7, Eps15, and AP-2m significantly globin in the parasite culture, we used host lines (suggesting some importance of these
reduced transport of hemoglobin to the FV, cells loaded with fluorescent dextrans, a well- proteins for efficient parasite growth), there
whereas the controls or inactivation of Kelch13 established tool for studying endocytic uptake was little correlation overall between growth
or an unrelated essential protein [PF3D7_ in malaria parasites (36, 40). These experi- and ART resistance (Fig. 4G), excluding growth
0210200 (17), here called 2102] did not (Fig. ments demonstrated uptake of host cell mate- levels as a factor in resistance. For instance,
3D). Inactivation of KIC7, AP-2m, and to a lesser rial into young ring stages (Fig. 4A). the line with a disruption of KIC10, the only
extent Eps15, reduced growth over the 8-hour Next, we tested the role of Kelch13 in endo- non–Kelch13 compartment KIC (Fig. 1H), had
assay time. However, the growth defect itself cytosis in 0-to-6-hour young rings [the time the third lowest growth rate but did not show
was not the reason for the phenotype, as in- the drug pulse is applied in ART susceptibility resistance (Fig. 4, F and G).
ducing an unrelated growth defect using azide assays (4)] using the Kelch13 3xNLS parasites To assess the role of the essential proteins
did not impair endocytosis (Fig. 3E). Surpris- (the cell line leading to only partial inactiva- AP-2m, Eps15, UBP1, and KIC7 in ART resist-
ingly, inactivation of Kelch13 significantly tion of Kelch13, to avoid detrimental effects on ance, we partially inactivated them (using KS)
increased parasite size compared with the parasite growth). Inactivation of Kelch13 sig- before and during the ART drug pulse of the
control, and inactivation of UBP1 showed a nificantly reduced endocytic uptake in these RSA in a manner that did not kill the parasites.
clear reduction of endocytosis but no signif- rings (Fig. 4B and fig. S12). Inactivation of Inactivation of KIC7, AP-2m, Eps15, and UBP1
icant reduction in cell size over the 8-hour KIC7 similarly reduced endocytosis in rings induced ART resistance (Fig. 4H). In contrast,
assay time (Fig. 3, D to F). We conclude that (Fig. 4C and fig. S12), congruent with its sim- reducing parasite viability with azide or inacti-
proteins of the Kelch13 compartment play a ilar KS phenotype to Kelch13 in rings and its vation of an unrelated essential protein [2102
role in endocytic uptake of hemoglobin, but role in endocytosis in trophozoites (Fig. 3E). (17)] did not affect the outcome of the RSA,
Kelch13 itself does not appear to be necessary Hence, KIC7 affects endocytosis in both rings indicating that the effect on ART susceptibility
for this process in trophozoites. The lack of and trophozoites and, together with the sim- is specific for the function of Kelch13 and its
effect in trophozoites agrees with the previous ilar stage-specific phenotypes, suggests that compartment proteins (Fig. 4H). We conclude
finding that inactivation of Kelch13 specifi- the Kelch13 compartment proteins are needed that inactivation of more than half (7 of 13) of
cally affected the ring stage only (17). for endocytosis in all asexual blood stages, the tested Kelch13 compartment proteins,
A detailed comparison of the stage-specific with the exception of Kelch13, which is only as well as Kelch13 itself, reduces the respon-
essentiality of Kelch13, UBP1, and KIC7 re- needed for this process in rings. siveness to ART. Notably, all of the Kelch13
vealed that all three proteins caused a sim- To assess whether this could be the reason compartment proteins affecting endocytosis
ilar phenotype when inactivated in rings for resistance, we next tested hemoglobin up- (Fig. 3, D and E) also rendered parasites re-
(severe delay of ring stage with limited pro- take into rings of resistant parasites contain- sistant, establishing a strong link between he-
gression to the trophozoite stage). In contrast, ing a mutated genomic Kelch13C580Y (17). The moglobin uptake and ART resistance (table S1).
inactivation in early trophozoites prevented resistant parasites also displayed reduced endo- We conclude that Kelch13 compartment pro-
development into schizonts in KIC7 and UBP1, cytic uptake (Fig. 4D and fig. S12), suggesting teins are part of an ART resistance pathway,
but not in Kelch13 (fig. S11). This suggests a this as a reason for reduced susceptibility to and the correlation of reduced endocytosis and
similar function of these proteins but that ART. To further confirm this hypothesis, we resistance indicates that this is the mechanism
Kelch13 is not essential for this function in inactivated Kelch13 (again using the Kelch13 of this mode of resistance. To test whether the
trophozoites. A previously established KS cell 3xNLS line) and measured whether this can corresponding genes belong to a coregulated
line that only partially inactivates Kelch13 render parasites ART-resistant using a standard pathway, we assessed their stage-specific tran-
(Kelch13 3xNLS) and that does not noticeably RSA (4). When Kelch13 was inactivated, para- scription patterns across asexual blood stage
affect parasite growth in a 4-day growth assay sites showed levels of resistance comparable development (42). Except for mca2, all genes
(17) showed a mild delay of the ring stage (fig. to the resistant Kelch13C580Y parasite line (Fig. involved in resistance clustered into one group

when compared with (i) the other validated property held by wtKelch13 that is lacking wild-type parasites (45) and that removing func-
Kelch13 compartment hits (Fig. 4I), (ii) all in Kelch13 C580Y , we episomally expressed tions needed for amino acid access of the
significant Kelch13 interactome hits (fig. S13A), Kelch13C580Y on the resistant background parasite prolongs the ring stage (46). Reduced
and (iii) all hits significant in any of the BioIDs (resulting in parasites expressing episomal levels of amino acids could also account for
(fig. S13B). In contrast, the other hits showed Kelch13C580Y and endogenous Kelch13C580Y, in some of the changes observed in ART-resistant
widely differing transcription patterns, reflect- effect only raising the abundance of Kelch13C580Y). parasites, such as elevated cellular stress re-
ing the asynchronous nature of the parasites This also reverted the parasites from resistant sponses (9, 15). It also indicates that the fitness
used as input for the DiQ-BioID. Overall, the to sensitive again (Fig. 5E), demonstrating cost incurred by resistance-conferring Kelch13
coexpression lends independent support to that reduced protein levels (or reduced activ- mutations (47, 48) is a direct result of the re-
the idea that the genes experimentally iden- ity) alone explains resistance, congruent with sistance mechanism. Hence, there is a trade-
tified here belong to a coregulated pathway the finding that partially inactivating Kelch13 off between resistance and growth levels in
involved in hemoglobin endocytosis. has a similar effect (Fig. 4E). In contrast, Kelch13 ring stages (Fig. 5F), creating a pressure for
To show that the resistance pathway of the missing the Plasmodium-specific region (PSR), compensatory adaptations that may explain
Kelch13 interactors is clinically relevant, we a modification that led to a loss of the correct the importance of the genetic background in
chose a mutation in UBP1 (R3138H) that was location of this construct (Fig. 5E), did not re- parasite resistance and fitness (8, 48).
identified as a possible contributor to ART revert the resistance phenotype when added on Altered endocytosis as a mechanism of ART
sistance by genomic surveillance of parasite top of the kelch13C580Y background (Fig. 5E). resistance reconciles aspects of previous find-
field samples (19). We changed this position in This demonstrated the specificity of the effect ings relating to ART resistance, e.g., a role of
UBP1 in 3D7 parasites and found that this ren- with episomal Kelch13C580Y and indicated that phosphatidylinositol 3-phosphate (and its ge-
dered these parasites resistant to ART (Fig. 4J) the BTB and Kelch13 domain without the PSR nerating kinase) (12), a membrane signature
at a level comparable to the resistance demon- are not functional. specific for endosomes (49) that is also pre-
strated when UBP1 was inactivated by KS (Fig. Overall, these results show that reduced ac- sent on the endolysosomal system of malaria
4H). These findings demonstrate that the path- tivity of Kelch13 is the most likely mechanism parasites (36, 50), or of coronin (51) that could
way identified here can be clinically relevant of how Kelch13 mutation causes resistance influence endocytosis via actin. We also vali-
and that mutations in the UBP1 can modulate and that this reduced activity, at least in part, dated the suspected role of AP-2m in resistance
ART resistance. is due to decreased Kelch13 levels in resistant (30, 31) and show that, as in other organisms
parasites. A generally reduced function also (24), this protein is involved in endocytosis.
Reduced levels of Kelch13 explain explains why inactivation of several Kelch13 This indicates that Kelch13 marks a clathrin-
ART resistance compartment proteins renders parasites resist- independent endocytosis pathway that un-
To determine how mutated Kelch13 influences ant, as each will reduce the output of a common expectedly still contains the typical clathrin
resistance, we first assessed whether the inter- pathway that, based on our functional data, is adaptor AP-2m. Besides Eps15 and AP-2m, KIC4
action profile of mutated Kelch13 was altered endocytosis. shows homology to alpha adaptins [according
compared with wtKelch13. However, DiQ-BioID to HHPred (52)], adding an additional protein
with the Kelch13C580Y parasite line revealed Discussion typical for endocytosis in a generally highly
no marked differences to the DiQ-BioID with Here, we show that Kelch13 defines an endo- derived pathway.
wtKelch13 (Fig. 5A, fig. S14, and table S1). The cytosis pathway required for the uptake of In contrast to its interactors, inactivation of
only exception was KIC10, which was less en- host cell hemoglobin and that this pathway Kelch13 impaired endocytosis only in rings,
riched in the Kelch13C580Y DiQ-BioID, but this is critical for ART resistance. Although it is not in trophozoites. This likely is the reason
is the only KIC not colocalizing with Kelch13 well established that ARTs are activated by why Kelch13, among all the proteins influenc-
and has no role in resistance (Figs. 1H and 4F the parasite’s digestion products of hemoglo- ing endocytosis, is the one frequently found
and table S1). As these findings indicated that bin and that this is a prerequisite for ART mutated in resistant parasite isolates (53). Com-
there is no change in the interaction profile of action (37–39), this process was not consi- plete inactivation of Kelch13 arrests growth in
Kelch13C580Y compared with wtKelch13, we dered to be involved in resistance in parasites ring stages (17), and a similar phenotype was
hypothesized that mutating Kelch13 does not with Kelch13 mutations (15, 16). Instead, po- here observed with KIC7 and UBP1. The stage-
change a specific quality of its function but tential roles of Kelch13 downstream of ART specific effects of inactivating Kelch13 are
reduces its overall activity. This idea is also activation, e.g., in mitigating the ART-induced observed in a milder form in ART-resistant
suggested by our finding that reduced activ- cellular stress response, are the current target parasites that display a prolonged ring stage
ity of Kelch13 causes resistance (Fig. 4E). In of interest and central to hypotheses on the followed by an accelerated development in
addition, previous work indicated that resist- mechanism of resistance (15, 16). However, our later stages (9, 44), and this observation was
ant parasites harbor less Kelch13 protein (43). data indicate that Kelch13 and its compart- here recapitulated by partial inactivation
In agreement with this observation, we found ment proteins mediate resistance upstream of of Kelch13. This indicates that resistance-
~30 to 50% less Kelch13 in parasites with the both, drug activation and action. We propose conferring mutations partially reduce Kelch13
resistance-conferring form in Western blots a model where Kelch13 and its compartment function, a hypothesis supported by our find-
and by measuring Kelch13 levels directly in proteins control endocytosis levels, thereby ing that partial inactivation of Kelch13 in rings
living cells (Fig. 5, B to D). To test whether influencing the amount of hemoglobin avail- induces parasite resistance and that addi-
increased expression of Kelch13 can lead to able for degradation and hence the concentra- tional expression of Kelch13C580Y on a resistant
the loss of resistance, we episomally expressed tion of active drug (Fig. 5F). This mechanism background with the same mutation is suffi-
different versions of Kelch13 on the resistant explains the slowed development of ring stages cient to render parasites fully sensitive again.
genomic kelch13C580Y background. Adding observed in ART-resistant parasites (9, 44), as As resistant parasite field isolates harbor less
wtKelch13 (leading to expression of episomal the reduction in hemoglobin endocytosis in Kelch13 (43), which we also found in our re-
wtKelch13 and endogenous Kelch13C580Y) ren- rings diminishes the supply of amino acids. sistant laboratory line here, but transcript
dered these resistant parasites fully sensitive This agrees with the finding that ART-resistant levels appear to be unchanged (9), it can be
to dihydroartemisinin (DHA) again (Fig. 5E). parasites depend more on exogenous amino assumed that resistance-conferring mutations
To test whether this was due to an additional acids to mature from rings to trophozoites than alter Kelch13 protein stability.

RESEAR CH
We envisage that the mechanism of ART re- 44. A. Hott et al., Antimicrob. Agents Chemother. 59, 3156–3167 methods section. S.Scha., S.Schm., E.J., and S.F. designed and
sistance indicated by this work will aid in find- (2015). carried out experiments, analyzed data, and prepared figures.
45. D. Bunditvorapoom et al., Sci. Rep. 8, 12622 (2018). M.S., R.S., B.B., U.F., P.M.-R., and A.B.S. carried out experiments.
ing ways to antagonize it. It may also inform 46. P. Mesén-Ramírez et al., PLOS Biol. 17, e3000473 (2019). W.A.M.H. and C.G.T. conducted the mass spectrometry
the choice of ART partner drugs, particularly as 47. S. Nair et al., Antimicrob. Agents Chemother. 62, e00605-18 experiments, analyzed data, and prepared figures. R.B. supervised
(2018). the mass spectrometry experiments and analyzed data. T.S.
hemoglobin digestive processes are the target
48. J. Straimer et al., mBio 8, e00172-17 (2017). conceived of and supervised the project, designed experiments,
of existing drugs. Finally, the here-identified 49. T. Balla, Physiol. Rev. 93, 1019–1137 (2013). carried out imaging and image analysis, prepared figures, and
proteins of the resistance pathway are candi- 50. L. Tawk et al., Eukaryot. Cell 9, 1519–1530 (2010). wrote the paper with critical input from all authors. Competing
dates for novel parasite markers influencing 51. A. R. Demas et al., Proc. Natl. Acad. Sci. U.S.A. 115, interests: The authors declare no competing interests. Data and
12799–12804 (2018). materials availability: All data are available in the manuscript
ART resistance that now can be assessed in 52. L. Zimmermann et al., J. Mol. Biol. 430, 2237–2243 (2018). or the supplementary materials, and the Kelch13, Eps15, and CHC
population studies, as illustrated by the iden- 53. S. Takala-Harrison et al., J. Infect. Dis. 211, 670–679 (2015). mass spectrometry data are available via ProteomeXchange with
tification of a resistance-conferring mutation identifier PXD008219.
ACKN OWLED GMEN TS
in UBP1. It should, however, be noted that
This article is dedicated to Hendrik Herrmann. We thank L. Tilley SUPPLEMENTARY MATERIALS
resistance-causing mutations are most likely for suggestions of the timing of Kelch13 KS induction for RSAs.
to occur in proteins that have the least impact We thank Jacobus Pharmaceuticals for supplying WR99210. DSM1
Materials and Methods
on trophozoite survival, such as Kelch13. (MRA-1161) was received from MR4/BEI Resources, NIAID, NIH.
Fig. S1 to S14
Funding: E.J., S.F., A.B.S., and T.S. acknowledge funding by the
Table S1
RE FE RENCES AND N OT ES Research Training Group (GRK 1459) of the German Research
References (54–78)
Foundation (DFG). S.Schm. acknowledges funding by the German
1. WHO, “Guidelines for the treatment of malaria” (World Health Data S1 to S4
Center for Infection Research (TTU 03.806), M.S. thanks the
Organization, ed. 3, 2015).
Jürgen Manchot Stiftung and S.Scha. the Claussen-Simon Stiftung
2. A. M. Dondorp et al., N. Engl. J. Med. 365, 1073–1075 (2011).
for funding. W.A.M.H. obtained financial support from Netherlands
3. A. M. Dondorp et al., N. Engl. J. Med. 361, 455–467 (2009).
Organization for Scientific Research (NWO-VENI, 722.012.004). 26 March 2019; resubmitted 19 September 2019
4. B. Witkowski et al., Lancet Infect. Dis. 13, 1043–1049 (2013).
Author contributions: J.B. designed research, carried out Accepted 30 October 2019
5. N. Klonis et al., Proc. Natl. Acad. Sci. U.S.A. 110, 5157–5162
experiments, analyzed data, prepared figures, and wrote the 10.1126/science.aax4735
(2013).
6. S. Saralamba et al., Proc. Natl. Acad. Sci. U.S.A. 108, 397–402
(2011).
7. F. Ariey et al., Nature 505, 50–55 (2014).
8. J. Straimer et al., Science 347, 428–431 (2015). ◥
9. S. Mok et al., Science 347, 431–435 (2015).
10. F. Rocamora et al., PLOS Pathog. 14, e1006930 (2018).
REPORTS
11. J. Gibbons et al., BMC Genomics 19, 849 (2018).
12. A. Mbengue et al., Nature 520, 683–687 (2015). TOPOLOGICAL OPTICS
13. C. Dogovski et al., PLOS Biol. 13, e1002132 (2015).
14. M. Zhang et al., Cell Host Microbe 22, 766–776.e4 (2017).
15. L. Tilley, J. Straimer, N. F. Gnädig, S. A. Ralph, D. A. Fidock, A single photonic cavity with two independent
Trends Parasitol. 32, 682–696 (2016).
16. K. Haldar, S. Bhattacharjee, I. Safeukui, Nat. Rev. Microbiol. 16,
156–170 (2018).
physical synthetic dimensions
17. J. Birnbaum et al., Nat. Methods 14, 450–456 (2017).
18. K. J. Roux, D. I. Kim, M. Raida, B. Burke, J. Cell Biol. 196, Avik Dutt1, Qian Lin2, Luqi Yuan3*, Momchil Minkov1, Meng Xiao4, Shanhui Fan1*
801–810 (2012).
19. G. C. Cerqueira et al., Genome Biol. 18, 78 (2017).
20. P. Hunt et al., Mol. Microbiol. 65, 27–40 (2007).
The concept of synthetic dimensions has generated interest in many branches of science, ranging from
21. P. Hunt et al., BMC Genomics 11, 499 (2010). ultracold atomic physics to photonics, as it provides a versatile platform for realizing effective gauge
22. S. Borrmann et al., Sci. Rep. 3, 3318 (2013). potentials and topological physics. Previous experiments have augmented the real-space dimensionality
23. G. Henriques et al., J. Infect. Dis. 210, 2001–2008 (2014).
24. H. T. McMahon, E. Boucrot, Nat. Rev. Mol. Cell Biol. 12,
by one additional physical synthetic dimension. In this study, we endow a single ring resonator with two
517–533 (2011). independent physical synthetic dimensions. Our system consists of a temporally modulated ring
25. F. Tebar, T. Sorkina, A. Sorkin, M. Ericsson, T. Kirchhausen, resonator with spatial coupling between the clockwise and counterclockwise modes, creating a synthetic
J. Biol. Chem. 271, 28727–28730 (1996).
Hall ladder along the frequency and pseudospin degrees of freedom for photons propagating in the ring.
26. S. E. Francis, D. J. Sullivan Jr., D. E. Goldberg, Annu. Rev.
Microbiol. 51, 97–123 (1997). We observe a wide variety of physics, including effective spin-orbit coupling, magnetic fields, spin-
27. Z. Kadlecova et al., J. Cell Biol. 216, 167–179 (2017). momentum locking, a Meissner-to-vortex phase transition, and signatures of topological chiral one-way
28. M. Kaksonen, A. Roux, Nat. Rev. Mol. Cell Biol. 19, 313–326
edge currents, completely in synthetic dimensions. Our experiments demonstrate that higher-
(2018).
29. O. Martzoukou, S. Amillis, A. Zervakou, S. Christoforidis, dimensional physics can be studied in simple systems by leveraging the concept of multiple
G. Diallinas, eLife 6, e20083 (2017). simultaneous synthetic dimensions.
30. G. Henriques et al., Malar. J. 12, 118 (2013).
31. G. Henriques et al., Antimicrob. Agents Chemother. 59,
T
2540–2547 (2015).
32. J. Hirst, N. A. Bright, B. Rous, M. S. Robinson, Mol. Biol. Cell 10, here has been interest in creating syn- 1 (4, 5) or d = 0 (6–8). Although theoretical
2787–2802 (1999). thetic dimensions to study classical and proposals exist for creating two or more
33. E. C. Dell’Angelica, C. Mullins, J. S. Bonifacino, J. Biol. Chem. quantum dynamics (1) in systems with separate synthetic dimensions (9, 10), these
274, 7278–7285 (1999).
34. G. H. Borner et al., J. Cell Biol. 197, 141–160 (2012). extra dimensions beyond their real-space proposed phenomena have thus far eluded
35. A. K. Davies et al., Nat. Commun. 9, 3958 (2018). geometric dimensionality (2). Synthetic
36. E. Jonscher et al., Cell Host Microbe 25, 166–173.e5 (2019). dimensions can be formed by coupling atomic 1
37. N. Klonis, D. J. Creek, L. Tilley, Curr. Opin. Microbiol. 16, Ginzton Laboratory and Department of Electrical
722–727 (2013).
or photonic states with different internal de- Engineering, Stanford University, Stanford, CA 94305, USA.
38. N. Klonis et al., Proc. Natl. Acad. Sci. U.S.A. 108, 11405–11410 grees of freedom to form a lattice. These degrees 2
Department of Applied Physics, Stanford University,
(2011). of freedom may be based on the frequency, Stanford, CA 94305, USA. 3State Key Laboratory of
39. S. C. Xie et al., J. Cell Sci. 129, 406–416 (2016). Advanced Optical Communication Systems and Networks,
40. N. Abu Bakar, N. Klonis, E. Hanssen, C. Chan, L. Tilley,
spin, linear momentum, orbital angular mo-
School of Physics and Astronomy, Shanghai Jiao Tong
J. Cell Sci. 123, 441–450 (2010). mentum, spatial supermodes, or arrival time of University, Shanghai 200240, China. 4Key Laboratory of
41. L. H. Bannister, J. M. Hopkins, G. Margos, A. R. Dluzewski, light pulses (3). Previous experiments have pro- Artificial Micro- and Nano-structures of Ministry of Education
G. H. Mitchell, Microsc. Microanal. 10, 551–562 (2004). and School of Physics and Technology, Wuhan University,
vided demonstrations of (d + 1)-dimensional
42. Z. Bozdech et al., PLOS Biol. 1, e5 (2003). Wuhan 430072, China.
43. G. Siddiqui, A. Srivastava, A. S. Russell, D. J. Creek, J. Infect. Dis. physics on d-dimensional real-space lattices by *Corresponding author. Email: yuanluqi@sjtu.edu.cn (L.Y.);
215, 1435–1444 (2017). using one extra synthetic dimension, for d = shanhui@stanford.edu (S.F.)

RES EARCH | R E P O R T S
experimental observation. The realization of The tight-binding Hamiltonian describ- rections is therefore DfðwÞ ¼ f↓→↑ f↑→↓ ¼
two or more synthetic dimensions markedly ing our system is bðwÞ DL, where bðwÞ is the propagation con-
simplifies the experimental requirements for X X stant at frequency w for a mode in the connect-
studying a rich set of topologically nontrivial H¼ ½wm a†m;s am;s þ Jmm′ ðtÞa†m;s am′ ;s ing waveguides. Assuming that the connecting
phenomena—e.g., the high-dimensional quan- m;s
X m′ waveguides are the same as the waveguide of
tum Hall effect (11, 12)—without the need Ka†m;↑ am;↓ eimf0 þ H:c: the ring, and because bðwm Þ ¼ 2pm=L0 , the
for complex higher-dimensional structures m phase difference Df increases linearly with
ð1Þ
in real space. m: Dfðwm Þ ¼ 2pmDL=L0 ¼ 2mf0 .
We demonstrate a system exhibiting two where am;s and a†m;s
are the annihilation and To transform Eq. 1 into a time-independent
independent physical synthetic dimensions. creation operators for the mth longitudinal Hamiltonian, we define bm;↑ ¼ am;↑e imðWR tþf0=2Þ
Our system (Fig. 1A) consists of a ring res- cavity mode with frequency w m ¼ mW R and and bm;↓ ¼ am;↓ e imðW R t f0 =2Þ and use the
onator supporting a synthetic frequency di- with pseudospin s∈f↓; ↑g, and H.c. is the rotating-wave approximation to get
Hermitian conjugate. Jmm0 ðtÞ is the coupling
J X †
mension formed by the longitudinal cavity
modes, as well as a synthetic pseudospin along the synthetic frequency dimension H¼ bmþ1;↓ bm;↓ eif0 =2 þ
dimension formed by the clockwise (CW, ↑) (6, 13, 15, 16), produced by the electro-optic 2 m
and counterclockwise (CCW, ↓) modes at the modulation (7). Because a small portion of X
b†mþ1;↑bm;↑e if0=2 K b†m;↑bm;↓ þ H:c: ð2Þ
same frequency. The coupling along the fre- the ring is modulated, this coupling can be m
quency dimension is achieved with a modu- simplified as Jmm0 ðtÞ ¼ JcosWR t; that is, the
lator (13). The coupling along the pseudospin mode m can couple to all other modes of This Hamiltonian describes a two-legged lad-
dimension is achieved with a coupler in the the system, and the coupling strength is der pierced by a uniform magnetic field (a Hall
shape of a figure eight (hereafter, “8-shaped independent of the mode indices (7). Here, ladder) (18), as each plaquette is threaded by
coupler”), consisting of two-directional couplers WR is the free spectral range (FSR), corre- an effective magnetic flux f0 (Fig. 1, B and C).
connected by two nonintersecting waveguides. sponding to the separation between the lon- Thus, by choosing a nonzero DL, our structure
Our construction is different from methods of gitudinal modes. K in Eq. 1 is the strength in Fig. 1A naturally implements an effective
probing higher-dimensional phenomena using of the coupling between the two legs of the magnetic field. Large magnetic fluxes span-
topological pumps, for which the physics with ladder, created by the 8-shaped coupler com- ning the entire range in ½ p; p are achievable
two extra dimensions has been explored in prising two-directional
pffiffiffiffi couplers with splitting by choosing appropriate DL=L0 . Because a
recent experiments (11, 12). In these systems, amplitude K . This coupling has a frequency- purely one-dimensional (1D) lattice does not
a mathematical mapping between higher- and direction-dependent phase Tmf0 (Fig. permit magnetic field effects, our system cor-
dimensional lattices and lower-dimensional 1B), with f0 ∼ pDL=L0 (17), where DL is the responds to the simplest lattice model where
systems is achieved by varying some exter- length difference between the two connect- the physics emerging from effective magnetic
nal parameters of the lower-dimensional ing waveguides, and L0 is the length of the fields for photons can be observed.
system (2, 14). Although signatures of higher- ring. To explain how this phase Tmf0 is in- Instead of describing the system in Fig. 1 as
dimensional physics can be observed in such troduced, we note that the connecting wave- a two-legged ladder threaded by a uniform
topological pumping schemes, the full dy- guide depicted by the blue solid line in Fig. 1A magnetic field, we can alternatively derive the
namics are not captured because the external couples exclusively from the CW to the CCW physics of this system in terms of magnetic
parameters are not the dynamical variables of mode, whereas the connecting waveguide de- field–controlled spin-orbit coupling (SOC), with
the particles (3). In contrast, our approach picted by the dashed line couples only from the CW and CCW modes of each ring represent-
provides the ability to explore physical dy- the CCW to the CW mode. The phase dif- ing up and down spins. Going to the quasi-
namics in higher-dimensional space. ference between the coupling in the two di- momentum space (k-space), the Hamiltonian
Fig. 1. A modulated ring resonator A B C

with CW-CCW mode-coupling and
…
…
its corresponding lattice in syn-

thetic dimensions. (A) Schematic of
the ring of length L0 with electro-
CW CCW
optic modulation (EOM) and CW-CCW
coupling. The CW and CCW modes 1
form the pseudospin degree of free-
dom. The longitudinal modes of 2
the ring separated by the FSR WR
form the frequency degree of free-
dom. The two directional couplers are
connected into an 8-shaped coupler
by two connecting waveguides
of unequal lengths L1 and L2. By
varying DL ¼ L1 L2 , the phases of
couplings between CW and CCW EOM
modes [in (B) and (C)] can be varied,
…
and hence a controllable effective CW CCW

magnetic field penetrates the ladder.
The corresponding synthetic lattice is shown in two equivalent gauges: (B) a gauge with real inter-rung coupling J but complex inter-leg coupling (Eq. 1) and
(C) a translationally invariant gauge (Eq. 2) with real inter-leg coupling K and complex inter-rung coupling.

†
in Eq.p2ffiffiffiffiffiffiffiffiffiffiffi
becomes X H ¼ ∫dkbk HðkÞb T
k , with The Hall ladder has been formally shown to To directly detect the chiral modes of the
imWk
bk ¼ W=2p m
e ðb m;↑ bm;↓ Þ (where
; exactly reproduce the energies and eigenstates Hall ladder, we use time-resolved band struc-
T denotes the transpose), and of the topological chiral edge modes of a 2D ture spectroscopy (7). We can selectively excite
quantum Hall insulator (Fig. 2A) described by the CW or CCW pseudospin by exciting the
f
HðkÞ ¼ J 12 cos kW cos 0 þ the Hofstadter model (19). Even if the entire waveguide from the left or right, respectively,
2 bulk lattice sites are removed, the strip of and measure the transmitted signal to map out
f
sz sin kW sin 0 Ksx ð3Þ plaquettes forming the ladder retains the the band structure projected onto the corre-
2
chiral currents and spin-momentum locking, sponding spin (17). The results of these mea-
Here sx ; sy ; andsz are Pauli matrices. To make as can be seen by comparing Fig. 2B to Fig. 2A. surements (Fig. 2, E and H) were obtained by
the SOC explicit, we recast Eq. 3 into the form This attests to the topological robustness of using a setup consisting of a fiber ring with an
HðkÞ ¼ DðkÞ 1 þ BSOC ðkÞ s, where DðkÞ ¼ the 2D quantum Hall insulator. Such signa- embedded electro-optic modulator and an
Jcos kW cosðf0 =2Þ , BSOC ¼ ½K ; 0; J sin kW tures of topological chiral edge modes are 8-shaped coupler. The modulator is driven
sinðf0 =2Þ, and s ¼ ðsx ; sy ; sz Þ. The z com- evident in the theoretically calculated band at W ¼ 2WR ¼ 29:6 MHz [see (17) and (21) for
ponent of BSOC depends on the quasimomen- structure of HðkÞ (Fig. 2, D and G) along with details on the setup]. The measured band struc-
tum k, signifying SOC. The degree of SOC is the corresponding color-coded pseudospin ture agrees with that from the tight-binding
controlled by the effective magnetic flux f0. projections n↑ ¼ cos2 ðqB =2Þ and n↓ ¼ sin2 model (Fig. 2, D and G) and also with simula-
With the control of the magnetic flux, there- ðqB =2Þ, respectively. Here qB ¼ arctanfK = tions using a rigorous Floquet analysis (Fig. 2,
fore, our system can exhibit a rich set of ½J sin kW sinðf0 =2Þg represents the chiral F and I) (17). This constitutes a measurement
physics. Here we discuss three experimental Bloch angle of the eigenstate, and its k- of the dispersion of chiral one-way states in
observations of such physics, all controlled by dependence signifies chiral spin-momentum synthetic dimensions. It is analogous to direct
the magnetic gauge potential: spin-momentum locking (19, 20): In the lower band, positive methods of exploring surface-state disper-
locking in the band structure, chiral currents, (or negative) k states have predominantly CW sions in SOC topological insulators [using
and a Meissner-to-vortex phase transition. (or CCW) pseudospin character. angle-resolved photoemission spectroscopy
A B C in
2
CW
–2
D E F
2 CW
1
0.9
0
–2 0.1 0
G H I
2
CCW
0.9
0
–2 0.1
–1 0 1 –1 0 1 –1 0 1
Fig. 2. Chiral band structure and spin-momentum locking in the synthetic structure can be detected by exciting the waveguide from the left (right) and
Hall ladder. (A) Projected band structure of a 2D quantum Hall insulator infinite recording its transmission. (D and G) Theoretical band structures, with color-
along the vertical direction and finite along the horizontal direction (as shown coded pseudospin projections n↑ and n↓ for corresponding eigenstates. For the
in the inset), showing topological chiral edge states (CES) highlighted in blue and lower band, k > 0 states have predominantly CW pseudospin character, signifying
red between the bulk band gaps. f0 ¼ 2p=3. (B) Band structure of the two- spin-momentum locking. The dashed lines are band structures for the same
legged synthetic Hall ladder from the tight-binding Hamiltonian HðkÞ (Eqs. 2 and J but for K = 0. (E and H) Experimental time-resolved transmission through the
3) for J/K = 2. The bulk bands disappear but signatures of chiral edge states ring for CW excitation (E) and CCW excitation (H). Dw is the detuning of the input
are preserved (19). (C) Schematic setup to directly measure band structure by frequency win from the resonance frequency of the uncoupled CW and CCW
coupling an input-output waveguide to the ring in Fig. 1A. By varying win and modes. (F and I) Theoretical time-resolved transmission based on Floquet
detecting the time-resolved transmission through the ring, the band structure analysis (17). Experimental parameters: J/2p = 1.95 MHz and K/2p = 0.97 MHz.
can be directly read out in experiments. The CW (CCW) spin-resolved band f0 ≈3p=4. Cavity linewidth g=2p ¼ 480kHz.

(ARPES)] (22) or analyzing helical edge state we frequency-shift a portion of the input laser corresponds to exchanging the lengths L1 and
dispersions in real-space photonic crystals by dw = 500 MHz using an acousto-optic mod- L2 in our system in Fig. 1A. Whereas our ex-
(23). Spin-momentum locking is clearly seen ulator and interfere it with the cavity output. periments measured the steady-state chiral
in the experimental data (Fig. 2C), as the CW Here dw≫jm mL jW for all of the modes that current, we present theoretical simulations of
mode transmission predominantly peaks at we consider. A fast Fourier transform (FFT) of chiral one-way propagation and the reversal of
positive quasimomenta for the lower band. this interferogram directly yields P(m). Hetero- its direction with a switching of the magnetic
Additionally, we observe that the direction dyne detection (i.e., the use of a frequency flux in movies S1 to S3 (17). Our simulations
of spin-momentum locking switches for the shift, as mentioned above) is essential. If one show that such one-way propagation is re-
upper band. were to set dw = 0 in the experiment described silient to backscattering around corners in
The Hall ladder exhibits chiral currents—in above, one could not distinguish between the a finite synthetic lattice for nontrivial fluxes
our system, the CW (CCW) pseudospin evolves photon numbers at mL + m and mL − m f0 ≠ p; 0 but undergoes strong backreflec-
preferentially to higher- (lower-) frequency modes, because they produce beat notes at tion for trivial fluxes f0 ¼ p; 0 (movies S4
modes for the lower band. The direction of the same radio frequency mW . Figure 3A and S5) (17).
the current switches for the upper band. To shows the measured chiral current jC versus The Hall ladder in ultracold atomic sys-
quantify the direction of such spin- and band- the laser detuning Dw. For each Dw, jC is cal- tems has been predicted to exhibit a phase
dependent frequency evolution, we define the culated from the heterodyne FFT spectrum. transition on increasing J/K, from a phase
steady-state chiral current as An example of such a spectrum at Dw=K ¼ that has a single energy minimum in the
X X 0:67 is shown in Fig. 3B. In Fig. 3C, we show ground state (“Meissner” phase) at k = 0 to
jC ¼ Pðm; ↑Þ Pðm; ↑Þ ð4Þ such spectra for all Dw. In Fig. 3D, we show a a state that has a pair of energy minima at
m>mL m<mL theoretical computation of the same spectrum. degenerate k points (“vortex” phase) (19, 24).
The overall shape of the theoretical spectrum We demonstrate a similar transition in the
where mL is the order of the ring resonance agrees with the experiments. In both the theory band structure to illustrate the freedom in
closest to the input laser ðjwin mL Wj < and experimental results, the higher-frequency our system for shaping photonic bands. We
WR =2Þ and Pðm; ↑Þ is the steady-state photon modes have a larger occupation (jC > 0) in the adopt the same terminology to facilitate the
number of the CW mode at frequency mW. lower band. The sign of jC is switched for the comparison with existing literature. In our
To measure jC, we use frequency- and spin- upper band. Alternately, the sign of jC can be system, J can be easily tuned by changing the
resolved heterodyne detection of the modal switched by changing the direction of the modulation voltage while K remains constant.
photon numbers in the lattice (17). Specifically, effective magnetic field (Fig. 3, E and F), which For J=K ≪1, the system can be described as a
set of decoupled rungs of the ladder. In this
regime, the eigenstates are the standing-wave
A C E
symmetric and antisymmetric supermodes,
resulting in flat bands split by 2K (Fig. 4, A,
E, and left inset). Both bands have equal
contributions from the CW and CCW legs of
the ladder. For J=K ≫1, the two legs of the
ladder become decoupled, and we approach
the sinusoidal band structure of a 1D tight-
binding model with nearest-neighbor coupling
(7). In the intermediate regime, the competition
between synthetic SOC and effective magnetic
field causes a transition in the band structure
B D F from a single minimum at k = 0 (Fig. 4, A and
E) to two minima (Fig. 4, B to H) at (19)
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
f K2
kg;min ¼ Tarcsin sin2 0
2 J 2 tan2 ðf0 =2Þ
ð5Þ
The experimentally estimated band minima
positions agree with the theoretical prediction
within measurement uncertainties (Fig. 4J).
By measuring the time-averaged transmis-
Fig. 3. Direct measurements of chiral currents in the synthetic Hall ladder through heterodyne sion instead of the time-resolved transmission,
detection. (A) Chiral current jC (Eq. 4) versus laser-cavity detuning Dw measured by heterodyne mixing the we detect the spin-projected density of states
cavity output field with a frequency-shifted part of the input laser. The full heterodyne signal is shown in (C). (DOS) (Fig. 4I). For J≪K ; g, two peaks with
The lower band shows a positive jC for the CW mode. a.u., arbitrary units. (B) Steady-state–normalized Lorentzian lineshapes are seen, broadened by
photon number of the modes at frequencies mW in the lower band, at Dw=K ¼ 0:67 indicated by the purple the cavity photon decay rate g (Fig. 4I, blue
dashed line in (A). The asymmetric frequency mode occupation verifies that the CW mode predominantly curve). On increasing J, each of these peaks
evolves toward higher frequencies in the lower band. (C) Experimental heterodyne spectra mapping out the broadens owing to the increasing width of the
steady-state photon numbers for all Dw. (D) Theoretically calculated photon numbers based on a Floquet corresponding band structure (orange curve).
analysis. (E and F) Same as in (A) and (B), but with the direction of the effective magnetic field flipped, which Eventually, additional peaks are visible for
causes a change in the sign of jC. (A) and (C) also reveal a switching of the direction of chiral current on J > 2g (red and black curves), due to van
moving from the lower to the upper band. In (C) and (D), the strong signal in the excited mode (m − mL = 0) Hove singularities at the edges of both energy
has been suppressed to reveal the occupation of other modes clearly. bands (20, 25).

2 A B C D
–2
0
0.10 1 1
out
E F G H
2
–2
–0.5 0.0 0.5 –0.5 0.0 0.5 –0.5 0.0 0.5 –0.5 0.0 0.5
I J
1 0.4
CW
0.0
0 –0.4 CCW
–4 –2 0 2 4 0.0 0.5 1.0 1.5 2.0
Fig. 4. Observation of phase transition through spin-resolved band structure of the modulation signal. (I) Time-averaged transmission revealing the DOS.
measurements. (A to D) Theoretical band structure for f0 ¼ 2:38 ≈ 3p=4, Van Hove singularities due to a diverging DOS are also visible in the transmission,
for increasing J/K. (E to H) Corresponding experimentally measured time- smeared out by the cavity decay rate g=K ¼ 0:37. (J) Bifurcation of the
resolved transmission, showing good agreement with theory. The ladder energy minimum in k. Data points represent experimentally estimated splittings
insets on the left and right are indicative of the strengths in the pseudospin for band structures shown in (E) to (H), which agree with the solid lines
and frequency axes. J can be continuously tuned by varying the amplitude based on Eq. 5.
Although some aspects such as spin- eration with tailorable spectral envelopes, as only limited to frequency conversion but also
momentum locking, chiral currents, and van shown in fig. S5 (17). All of these features are including topologically protected manipula-
Hove singularities have been previously ob- achieved in a simple photonic structure con- tion of orbital angular momentum (9, 32) and
served in atomic systems (4, 20, 24, 25–29), sisting of a single modulated ring, completely transverse spatial modes (5). The advent of
several features are specific to our photonic based on the synthetic dimension concept. nanophotonic lithium niobate microring mod-
implementation. First, we are able to directly Several new possibilities and future appli- ulators with bandwidths exceeding the ring
measure the dispersion of the chiral one-way cations may be enabled using concepts dem- FSR shows promise for realizing synthetic
modes in synthetic space, owing to the time- onstrated in our experiments. One notable frequency dimensions on a chip (33). We an-
resolved band structure spectroscopy technique, possibility is the implementation of interact- ticipate that similar synthetic space concepts
as opposed to mapping of the density-of-states ing Hamiltonians in synthetic frequency di- could be extended to other frequency ranges,
in cold atom experiments (20, 25). Second, we mensions through the introduction of optical such as microwaves (34), or to real-space
have access to the entire band structure, in- nonlinearities (29, 30), which would permit photonic systems in which SOC (35), chiral
cluding the upper band, which allows us to the study of fundamental many-body physics quantum emission, and spin-momentum lock-
experimentally observe the chirality switching (31) and have applications in quantum sim- ing have been reported (36, 37). Additionally,
(Fig. 3) when going from the lower to the up- ulation (1) and quantum information process- CW-CCW mode excitation in microrings has
per band in a Hall ladder. Finally, our system ing. When combined with ideas from quantum been explored for the study of non-Hermitian
exhibits frequency conversion, which can have photonics, one can generate high-dimensional physics (38), counterpropagating solitons
applications in spectral manipulation of light. hyperentanglement in pseudospin and fre- (39, 40), and topological insulator lasers (41).
For example, the change in the dispersion re- quency axes by exciting the CW and CCW These ideas can be combined with our exper-
lations associated with the Meissner-to-vortex modes with entangled photon pairs (9). More- imentally demonstrated concepts of gauge po-
phase transition can be used for tunable fre- over, extensions of our setup can be used to tentials, effective magnetic fields, and SOC to
quency conversion and frequency comb gen- manipulate photonic degrees of freedom, not manipulate and control light in versatile ways.

RE FE RENCES AND N OT ES TOPOLOGICAL MATTER

1. O. Boada, A. Celi, J. I. Latorre, M. Lewenstein, Phys. Rev. Lett.
108, 133001 (2012).
2. L. Yuan, Q. Lin, M. Xiao, S. Fan, Optica 5, 1396–1405 (2018). Absence of evidence for chiral Majorana modes in
3. T. Ozawa, H. M. Price, Nat. Rev. Phys. 1, 349–357 (2019).
4. M. Mancini et al., Science 349, 1510–1513 (2015).
5. E. Lustig et al., Nature 567, 356–360 (2019).
quantum anomalous Hall-superconductor devices
6. B. A. Bell et al., Optica 4, 1433–1436 (2017).
7. A. Dutt et al., Nat. Commun. 10, 3122 (2019). Morteza Kayyalha1*, Di Xiao1*, Ruoxi Zhang1*, Jaeho Shin1, Jue Jiang1, Fei Wang1, Yi-Fan Zhao1,
8. A. Regensburger et al., Nature 488, 167–171 (2012). Run Xiao1, Ling Zhang1, Kajetan M. Fijalkowski2,3, Pankaj Mandal2,3, Martin Winnerlein2,3,
9. L. Yuan et al., Phys. Rev. Lett. 122, 083903 (2019).
10. I. Martin, G. Refael, B. Halperin, Phys. Rev. X 7, 041008 (2017).
Charles Gould2,3, Qi Li1, Laurens W. Molenkamp2,3, Moses H. W. Chan1†,
11. M. Lohse, C. Schweizer, H. M. Price, O. Zilberberg, I. Bloch, Nitin Samarth1†, Cui-Zu Chang1†
Nature 553, 55–58 (2018).
12. O. Zilberberg et al., Nature 553, 59–62 (2018).
A quantum anomalous Hall (QAH) insulator coupled to an s-wave superconductor is predicted to harbor
13. L. Yuan, Y. Shi, S. Fan, Opt. Lett. 41, 741–744 (2016).
14. M. Wimmer, H. M. Price, I. Carusotto, U. Peschel, Nat. Phys. 13, chiral Majorana modes. A recent experiment interprets the half-quantized two-terminal conductance
545–550 (2017). plateau as evidence for these modes in a millimeter-size QAH-niobium hybrid device. However,
15. T. Ozawa, H. M. Price, N. Goldman, O. Zilberberg, I. Carusotto, non-Majorana mechanisms can also generate similar signatures, especially in disordered samples. Here,
Phys. Rev. A 93, 043827 (2016).
16. C. Qin et al., Phys. Rev. Lett. 120, 133901 (2018). we studied similar hybrid devices with a well-controlled and transparent interface between the
17. See supplementary materials. superconductor and the QAH insulator. When the devices are in the QAH state with well-aligned
18. K. Fang, Z. Yu, S. Fan, Nat. Photonics 6, 782–787 (2012). magnetization, the two-terminal conductance is always half-quantized. Our experiment provides a
19. D. Hügel, B. Paredes, Phys. Rev. A 89, 023619 (2014).
20. S. Kolkowitz et al., Nature 542, 66–70 (2017). comprehensive understanding of the superconducting proximity effect observed in QAH-superconductor
21. A. Dutt et al., ACS Photonics 6, 162–169 (2019). hybrid devices and shows that the half-quantized conductance plateau is unlikely to be induced by
22. D. Hsieh et al., Nature 452, 970–974 (2008). chiral Majorana fermions in samples with a highly transparent interface.
23. N. Parappurath, F. Alpeggiani, L. Kuipers, E. Verhagen,
arXiv:1811.10739 [physics.optics] (26 November 2018).
T
24. M. Atala et al., Nat. Phys. 10, 588–593 (2014).
25. L. F. Livi et al., Phys. Rev. Lett. 117, 220401 (2016). opological superconductors (TSCs) are modes (CMEMs) (19). Alternative interpre-
26. L. W. Cheuk et al., Phys. Rev. Lett. 109, 095302 (2012). predicted to host Majorana fermions, tations, however, are also possible. For exam-
27. F. A. An, E. J. Meier, B. Gadway, Sci. Adv. 3, e1602685 (2017).
28. H. Cai et al., Phys. Rev. Lett. 122, 023601 (2019).
particles that are their own antiparticles ple, Huang et al. (24) and Ji and Wen (25)
29. M. E. Tai et al., Nature 546, 519–523 (2017). (1–5). These Majorana fermions obey theoretically discussed two different scenarios
30. L. Yuan, A. Dutt, M. Qin, S. Fan, X. Chen, arXiv:1909.12466 non-Abelian statistics and are promising in which a s1,2 = 0.5e2/h plateau (where e is
[physics.optics] (27 September 2019).
31. R. Ma et al., Nature 566, 51–57 (2019).
candidates to form a topological qubit, which the elementary charge and h is the Planck
32. X.-W. Luo et al., Nat. Commun. 6, 7704 (2015). is the basis for fault-tolerant topological quan- constant) can arise without invoking the
33. C. Reimer, Y. Hu, A. Shams-Ansari, M. Zhang, M. Loncar, tum computation (6–8). TSCs are predicted Majorana physics. Huang et al. considered
arXiv:1909.01303 [physics.optics] (3 September 2019).
to appear in a variety of condensed matter the percolation of QAH edges induced by
34. C. W. Peterson, W. A. Benalcazar, M. Lin, T. L. Hughes, G. Bahl,
Phys. Rev. Lett. 123, 063901 (2019). quantum systems including strong spin-orbit– magnetic disorder in the QAH insulator as
35. D. L. P. Vitullo et al., Phys. Rev. Lett. 118, 083601 (2017). coupled semiconductor-SC hybrid devices (9, 10), an alternative origin for the s1,2 = 0.5e2/h
36. S. Barik et al., Science 359, 666–668 (2018). fractional quantum Hall (QH) systems at fill- plateau. Ji and Wen argued that the s1,2 =
37. K. Y. Bliokh, D. Smirnova, F. Nori, Science 348, 1448–1451 (2015).
38. S. Malzard, H. Schomerus, Phys. Rev. A 98, 033807 (2018). ing factor n = 5/2 (11, 12), spinless px + ipy SCs 0.5e2/h plateau can arise if the SC layer pro-
39. Q.-F. Yang, X. Yi, K. Y. Yang, K. Vahala, Nat. Photonics 11, such as Sr2RuO4 (2, 13), hybrid topological vides good electrical contact to the chiral edge
560–564 (2017). insulator (TI)–SC devices (9), integer QH in- modes of the QAH insulator. In other words,
40. C. Joshi et al., Opt. Lett. 43, 547–550 (2018).
41. M. A. Bandres et al., Science 359, eaar4005 (2018). sulators covered by a conventional s-wave the local equilibrium between the chiral edge
SC (14), and thin films of transition metal modes of the QAH insulators and the SC strip
ACKN OW LEDG MEN TS dichalcogenides (15, 16). Theoretical work has ensures that the total resistance is the series
We thank D.A.B. Miller for initial discussions on the experiment and predicted a chiral TSC phase when a quan- resistance of two separate QAH regions, each
for providing lab space and equipment. Funding: This work is
supported by a Vannevar Bush Faculty Fellowship (grant N00014- tum anomalous Hall (QAH) insulator, a zero– with h/e2 resistance (26).
17-1-3030) from the U.S. Department of Defense and by MURI magnetic field manifestation of the integer Here, we studied the effect of contact trans-
grants from the U.S. Air Force Office of Scientific Research (grants QH effect (17, 18), is coupled to an s-wave SC parency in the appearance of the s1,2 =
FA9550-17-1-0002 and FA9550-18-1-0379). L.Y. acknowledges
support from the National Natural Science Foundation of China
(14, 19). 0.5e2/h plateau. To this end, we fabricated
(grant 11974245). M.M. acknowledges support from the Swiss The QAH effect has been experimentally magnetic TI-SC hybrid devices, an example of
National Science Foundation (grant P300P2_177721). Author demonstrated in thin films of magnetically which is shown in Fig. 1, A and B. Our device
contributions: Q.L. developed the idea, in conjunction with
L.Y. and A.D. A.D. designed, built, and characterized the setup
doped TI (18, 20–22). He et al. (23) recently consists of a superconducting Nb strip (width
and collected the data, in consultation with L.Y., Q.L., and S.F. reported a half-quantized plateau in the two- ~20 mm) covering the entire width of the QAH
M.M. and Q.L. contributed to theoretical analysis, simulations, terminal conductance s1,2 converted from re- layer on the left, a configuration similar to
and interpretations, with input from S.F., M.X., and L.Y. All authors
sistance measured across a millimeter-size that in (23), and a narrow Nb finger (width
contributed to discussion of the results and manuscript
writing. S.F. supervised the project. Competing interests: The QAH-Nb hybrid structure and interpreted ~200 nm) on the right (Fig. 1, A and B). The
authors declare no competing interests. Data and materials the half-quantized s1,2 plateau during mag- QAH sample in this device is a sandwich of
availability: All data are available in the manuscript or the netization reversal as a “distinct signature” 3QL Cr-doped (Bi, Sb)2Te3 / 5QL (Bi, Sb)2Te3 /
supplementary materials.
of one-dimensional chiral Majorana edge 3QL Cr-doped (Bi, Sb)2Te3, where QL stands
SUPPLEMENTARY MATERIALS for quintuple layer (27). Our device was de-
1
signed such that (i) the contact transparency
Materials and Methods Department of Physics, Pennsylvania State University,
between the magnetic TI and SC layers can
Figs. S1 to S5 University Park, PA 16802, USA. 2Faculty for Physics and
References (42–44) Astronomy (EP3), University of Würzburg, Am Hubland, be characterized using a differential conduct-
Movies S1 to S5 D-97074 Würzburg, Germany. 3Institute for Topological ance measurement on the QAH-Nb finger
Insulators, Am Hubland, D-97074 Würzburg, Germany. junction (28); (ii) the possible existence of the
29 August 2019; accepted 13 November 2019 *These authors contributed equally to this work.
Published online 28 November 2019 †Corresponding author. Email: mhc2@psu.edu (M.H.W.C.); CMEMs can be investigated by analyzing the
10.1126/science.aaz3071 nxs16@psu.edu (N.S.); cxc955@psu.edu (C.-Z.C.). two-terminal conductance s1,2 deduced from

Fig. 1. QAH-Nb device and characterizations. (A) Schematic of the device in the resistance of the Nb finger at T ~ 8.6 K is associated with a superconducting
consisting of a QAH insulator layer, a Nb strip, and a Nb finger. The Nb finger is transition of the Nb section with a larger width (~4 mm) in the device; see (B)
used to characterize the magnetic TI-Nb contact transparency, whereas the Nb and the inset of Fig. 2A. (D) m0H dependence of the resistance of the Nb finger
strip is used to study the two-terminal conductance s1,2 across the QAH-Nb and Nb strip. (E and F) The four-terminal longitudinal and Hall resistance
structure. (B) Optical microscope image of the device. (C) Temperature (rxx and ryx) (E) and their corresponding longitudinal and Hall conductance
dependence of the normalized resistance of the Nb finger and Nb strip. The drop (sxx and sxy) (F) as a function of m0H measured at Vg = Vg0 = +42 V and T = 30 mK.
resistance measured across the QAH-Nb strip leads contribute ~40 ohms to the total resist- state turns out to be the dominant contribu-
device (19, 23). Furthermore, our QAH film ance, which has been subtracted. Figure 1D tion in our samples (see fig. S3). On the other
(i.e., magnetic TI) can be tuned to the metallic plots the magnetic field (m0H) dependence hand, sU (sD) is a better probe of the AR/NR
state using the back-gate voltage Vg. This al- of the resistances of the Nb finger and the ratio when the magnetic TI is in its metallic
lows us to probe the Andreev reflection in- Nb strip. The Nb strip has an upper critical phase, as discussed below. To characterize
volved in the magnetic TI-SC hybrid device field m0Hc2,strip ~ 2.9 T. Shown in Fig. 1, E and the magnetic TI-Nb interface transparency,
through the entire phase diagram—that is, as F, are the m0H dependence of the longitudinal we applied a negative Vg = –50 V to reach the
a function of the chemical potential (tuned by resistance (conductance) rxx (sxx) and the metallic phase of the magnetic TI. At zero
Vg) and the external magnetic field. When the Hall resistance (conductance) ryx (sxy) at Vg = magnetic field, we observed an enhancement
QAH layer is tuned into the metallic phase, Vg0 = +42 V and T = 30 mK, where typical of the zero-bias conductance approaching
we observed a strong enhancement of the QAH characteristics, quantized ryx (sxy) ac- 180% of its high-temperature value, revealing
zero-bias electrical conductance, nearly twice companied by vanishing rxx (sxx), are ob- a highly transparent magnetic TI-SC interface.
(~180%) the normal-state conductance preserved. Because the rxx peak value during Remarkably, although the superconductivity
sumably induced by Andreev reflection. The magnetization reversal is comparable to the in the Nb finger is suppressed for m0H > 0.5 T
observation of Andreev reflection in our junc- quantized ryx value, the zero–Hall conduct- (Fig. 1B and fig. S1E), the magnetic TI-Nb
tion is strong evidence for the induced super- ance sxy = 0 plateau [i.e., Chern number C = 0 contact transparency is unaffected (Fig. 2,
conducting pair potential in the magnetic TI phase (30)] is not observable. The sxy = 0 A and B). For m0H larger than the coercive
layer and allows us to study the effect of a plateau is usually observed in thinner, uni- field (m0Hc ~ 0.06 T) of the magnetic TI layer
transparent interface on the two-terminal formly doped QAH samples with a larger rxx at T = 2 K, zero-bias sU is slightly reduced
conductance s1,2 in the QAH-SC hybrid struc- peak (31, 32). and zero-bias sD is slightly increased. The
ture. When the magnetic TI is in the QAH We characterized the interface transpar- reduction of sU and enhancement of sD are
regime, the differential conductance is domi- ency of the magnetic TI-Nb finger junction likely results of the magnetization reversal
nated by the density of state modulation (i.e., by measuring its differential conductance, in the magnetic TI layer around the m0Hc re-
breakdown) of the QAH effect. When the QAH which is related to the probabilities of the gime (see fig. S4).
and SC layers are strongly coupled, as demon- Andreev reflection (AR) and the normal reflec- In our experiment, it is difficult to extract
strated by our differential conductance data, s1,2 tion (NR) across the interface. Figure 2, A and accurately the voltage drop across the mag-
is always half-quantized when the magnetization B, shows the differential upstream conduct- netic TI-Nb junction because a large portion
is well aligned. Our conclusions are supported ance sU = dI6,8 /dV7,8 and the downstream of the voltage drop appears across the resist-
by measurements on ~30 devices (29). conductance sD = dI6,8 /dV9,8, where the sub- ive part of the magnetic TI layer (29). There-
Figure 1C shows the temperature depen- script numbers correspond to the electrodes fore, we plot the differential conductance as a
dence of the Nb finger and the Nb strip resist- shown in Fig. 1A, at different magnetic fields. function of the dc current, Idc, rather than the
ance. The Nb finger becomes superconducting sU and sD are normalized by their respective dc voltage, Vdc. Furthermore, the enhancement
below its critical temperature Tc,finger ~ 5 K. values at T > Tc,finger (i.e., s6K). For Vg = Vg0, in sU (sD) may be a result of the combined
The critical temperature of the Nb strip Tc,strip the differential conductance is determined by contribution of the AR process across the
is ~8.6 K. Because we are using a two-terminal an interplay between AR and NR at the in- magnetic TI-Nb interface and the metal-to-
technique to measure the resistance (between terface, as well as the breakdown of the QAH superconductor transition induced by the
electrodes 8a and 8b in Fig. 1A), the normal system (33–35); the breakdown of the QAH critical current of the Nb finger. We also note

strip were deposited onto the QAH devices

simultaneously, we expect the interface trans-
parency across the magnetic TI-Nb junction
to be similar for the strip and the finger. The
transparent interface and the chiral nature
of the edge modes in the QAH regime are
expected to ensure that an electron prop-
agating along the wide Nb strip will quickly
become an equal mixture of electrons and
holes (28).
Our QAH-SC hybrid device (minus the Nb
finger) shown in Fig. 1, A and B, is similar
to the device used in (23). A s1,2 ~ 0.5e2/h
plateau during the magnetization reversal
(~ m0Hc) followed by a s1,2 ~ e2/h plateau for
the m0Hc < |m0H|< m0Hc2,strip regime is reported
in (23). These measurements were interpreted
as induced by the presence of the CMEMs;
the transition from the s1,2 = e2/h plateau to
the s1,2 = 0.5e2/h plateau was attributed to a
topological phase transition in the TSC state
Fig. 2. Contact transparency in the magnetic TI-Nb finger device. (A and B) The differential upstream from N = 2 to N = 1, where N denotes the
conductance sU = dI6,8/dV7,8 (A) and the downstream conductance sD = dI6,8/dV9,8 (B) of the magnetic number of CMEMs (14, 19). In the same struc-
TI-Nb finger junction normalized by their high-temperature (T > Tc,finger) values s6K, measured at different ture, an extremely small two-terminal conduct-
values of m0H and T = 2 K. Inset of (A) shows a magnified optical image of the magnetic TI-Nb finger ance s1,3, measured between the Nb strip and
device. (C and D) The normalized sU (C) and sD (D) measured at different temperatures and zero magnetic the QAH sample, for m0Hc < |m0H| < m0Hc2,strip
field. The excitation current Iac is 10 nA. was also reported (23). The small value of s1,3
in this m0H range indicates that the Nb layer
is likely decoupled from the QAH sample,
and hence the s1,2 = e2/h plateau may be a
result of poor electrical contact between the
QAH insulator and the Nb layers; in that case,
there is no proximity-induced superconduc-
tivity and no AR at the QAH-Nb interface
(23, 25). We note that the observation of s1,2 =
0.5e2/h in the QAH insulator phase is not
unusual (24, 25). Indeed, a normal metal (e.g.,
gold) overlaying the two edges of the QAH
sample will give rise to such a quantization
in s1,2 (26).
Our results from the QAH-SC strip devices
Fig. 3. Two-terminal conductance s1,2 across the QAH-Nb strip device. (A) m0H dependence of s1,2 = dI13,6/dV1,2 can be explained without resorting to Majorana
measured at Vg = Vg0 = +42 V and T = 30 mK. s1,2 ~ 0.5e2/h for the entire m0H range when the magnetization physics. Figure 3A displays the m0H depen-
is well aligned. No change in s1,2 is observed when the Nb strip transitions from the superconducting state to dence of the two-terminal conductance s1,2
the normal state (m0H > m0Hc2,strip ~ 2.6 T). Inset magnifies the m0H axis during the magnetization reversal for Vg = Vg0 = +42 V of our device. In contrast
process. (B) m0H dependence of two-terminal conductance s1,3 = dI13,6/dV1,3. s1,3 approaches ~e2/h for |m0H| > to (23), we observed that s1,2 = dI13,6 /dV1,2 ~
m0Hc, indicating good contact transparency between the Nb strip and the QAH sample. The excitation current Iac is 0.5e2/h over the entire range of the magnetic
1 nA. Blue and red curves represent the process for decreasing and increasing m0H, respectively. field except in the m0H range, when the mag-
netization of the sample is being reversed near
m0Hc. In this range, s1,2 drops to ~0.21e2/h.
that the slight reduction of sU (sD) close to longer superconducting (Fig. 1C), and thus Specifically, no change in s1,2 is observed when
zero bias (i.e., the small zero-bias conduct- the differential conductance is a sum of the m0H is increased across the critical field of the
ance dip) observed in Fig. 2 is a result of the contributions from the NR at the interface Nb strip; that is, m0Hc2,strip ~ 2.9 T (Fig. 1D). We
interplay between the AR and NR at the inter- and the resistive part of the magnetic TI film. also measured s1,3 = dI13,6 /dV1,3, the conduct-
face, as predicted by the Blonder, Tinkham, Therefore, the zero-bias conductance at T = ance between the Nb strip and the QAH sam-
and Klapwijk (BTK) model for a normal metal- 6 K takes the same value as that of the high- ple (Fig. 3B). We found that s1,3 ~ e2/h in the
superconductor junction with a nonzero bar- bias regime for T ≤ 5 K, consistent with the entire |m0H| > m0Hc range, indicating that the
rier height (36). To confirm that the sharp AR picture for normal metal-superconductor Nb strip is strongly coupled to the QAH sam-
zero-bias conductance peak is indeed a result junctions (29, 36, 37). ple, leading to the equilibrium of chemical po-
of the AR process at the interface, we studied Our experimental observations reveal the tentials between chiral edge modes of the
the temperature dependence of sU (sD) versus presence of a highly transparent interface QAH sample and bulk Nb layer (25). This
Idc in Fig. 2, C and D, where we observed a between the magnetic TI and Nb finger behavior is what one would expect if a normal
featureless sU (sD) at T = 6 K > Tc,finger . We throughout the m0H range of interest (0 T < metal were used instead of the Nb strip. For
note that at T = 6 K, the Nb finger is no m0H < 1 T). Because the Nb finger and the Nb m0H > m0Hc2,strip, the Nb strip turns into the

8. C. Nayak, S. H. Simon, A. Stern, M. Freedman, S. Das Sarma,

Rev. Mod. Phys. 80, 1083–1159 (2008).
9. L. Fu, C. L. Kane, Phys. Rev. Lett. 100, 096407 (2008).
10. R. M. Lutchyn, J. D. Sau, S. Das Sarma, Phys. Rev. Lett. 105,
077001 (2010).
11. G. Moore, N. Read, Nucl. Phys. B 360, 362–396 (1991).
12. A. Stern, Nature 464, 187–193 (2010).
13. A. P. Mackenzie, Y. Maeno, Rev. Mod. Phys. 75, 657–712 (2003).
14. X. L. Qi, T. L. Hughes, S. C. Zhang, Phys. Rev. B 82, 184516 (2010).
15. N. F. Q. Yuan, K. F. Mak, K. T. Law, Phys. Rev. Lett. 113,
097001 (2014).
16. Y. T. Hsu, A. Vaezi, M. H. Fischer, E. A. Kim, Nat. Commun. 8,
14985 (2017).
17. F. D. M. Haldane, Phys. Rev. Lett. 61, 2015–2018 (1988).
18. C. Z. Chang et al., Science 340, 167–170 (2013).
19. J. Wang, Q. Zhou, B. Lian, S. C. Zhang, Phys. Rev. B 92,
064520 (2015).
Fig. 4. Two-terminal conductance s1,2 in 6QL uniformly doped QAH-Nb strip devices. (A) m0H dependence 20. C. Z. Chang et al., Nat. Mater. 14, 473–477 (2015).
of s1,2 measured across one Nb strip stacked on a 6QL (Bi, Sb)1.85Cr0.15Te3 QAH sample at Vg = Vg0 and T = 21. J. G. Checkelsky et al., Nat. Phys. 10, 731–736 (2014).
30 mK. (B) Same as (A) for two Nb strips. Insets show the corresponding device configurations. With increasing 22. X. Kou et al., Phys. Rev. Lett. 113, 137201 (2014).
23. Q. L. He et al., Science 357, 294–299 (2017).
the number (n) of Nb strips, the corresponding s1,2 plateau decreases as s1,2 ~ e2/(n + 1)h. The excitation current 24. Y. Y. Huang, F. Setiawan, J. D. Sau, Phys. Rev. B 97, 100501 (2018).
Iac is 1 nA. Blue and red curves represent the process for decreasing and increasing m0H, respectively. 25. W. Ji, X. G. Wen, Phys. Rev. Lett. 120, 107002 (2018).
26. C. Z. Chang et al., Phys. Rev. Lett. 115, 057206 (2015).
27. J. Jiang et al., arXiv 1901.07611 [cond-mat.mes-hall]
normal state, hence s1,2 remains half-quantized. tence of the zero–Hall conductance plateau (28 January 2019).
We have also studied 9QL V-doped TI sam- in QAH samples does not change our find- 28. G. H. Lee et al., Nat. Phys. 13, 693–698 (2017).
29. See supplementary materials.
ples, which were previously shown to exhibit ings; specifically, the s1,2 ~ 0.5e2/h plateau is 30. J. Wang, B. Lian, S. C. Zhang, Phys. Rev. B 89, 085106 (2014).
perfect QAH effect (35, 38–40) and signatures observed for the entire m0H region with well- 31. X. Kou et al., Nat. Commun. 6, 8474 (2015).
of axion electrodynamics (38). The devices aligned magnetization. We also studied the 32. Y. Feng et al., Phys. Rev. Lett. 115, 126801 (2015).
33. M. Kawamura et al., Phys. Rev. Lett. 119, 016803 (2017).
were patterned using an optical lithography V-doped TI/TI/Cr-doped TI QAH sandwich
34. E. J. Fox et al., Phys. Rev. B 98, 075145 (2018).
process and used MoRe as the SC strip. We samples, in which a well-established C = 0 35. M. Götz et al., Appl. Phys. Lett. 112, 072102 (2018).
again observed the s1,2 ~ 0.5e2/h plateau for insulating phase (i.e., the axion insulator state) 36. G. E. Blonder, M. Tinkham, T. M. Klapwijk, Phys. Rev. B 25,
4515–4532 (1982).
the entire m0H region with well-aligned mag- emerges (41, 42). Here, we also observed the 37. M. Tinkham, Introduction to Superconductivity (McGraw-Hill, 1996).
netization (see fig. S8). s1,2 ~ 0.5e2/h plateau for the entire m0H region 38. S. Grauer et al., Phys. Rev. Lett. 118, 246801 (2017).
The existence of the zero–Hall conductance with well-aligned magnetization (see fig. S7). 39. M. Winnerlein et al., Phy. Rev. Mater. 1, 011201 (2017).
40. S. Grauer et al., Phys. Rev. B 92, 201304 (2015).
plateau with the C = 0 phase in a QAH sam- To better understand the relation between s1,2
41. D. Xiao et al., Phys. Rev. Lett. 120, 056801 (2018).
ple was claimed as a prerequisite for the ob- and the coupling of the SC layer to the chiral 42. M. Mogi et al., Sci. Adv. 3, eaao1669 (2017).
servation of the N = 1 TSC phase (23). The edge modes, we measured s1,2 across a 6QL 43. C. Z. Chang, Data for “Absence of Evidence for Chiral Majorana
transition from the C = 0 (i.e., N = 0) phase Cr-doped (Bi, Sb)2Te3 QAH sample with two Modes in Quantum Anomalous Hall-Superconductor Devices”;
https://doi.org/10.7910/DVN/QWQEHT, Harvard Dataverse
to the C = 1 (i.e., N = 2) phase is given in Nb strips. We observed s1,2 ~ e2/3h for the (2019).
(23) [citing (19)] as the key evidence for the entire well-aligned m0H regimes (Fig. 4B). The
existence of the N = 1 TSC phase. We note, value of the s1,2 plateau decreases with an in- AC KNOWLED GME NTS
however, that the theoretical calculations in creasing number of Nb strips (n), specifically We thank C. X. Liu, K. T. Law, B. Lian, J. Wang, X. Dai, J. Jain,
H. Z. Lu, Z. Wang, B. H. Yan, G. H. Lee, Y. L. Chen, K. He, W. J. Ji,
(19) treated the superconductor strip merely s1,2 ~ e2/(n + 1)h, which indicates that the Q. K. Xue, and X. D. Xu for helpful discussions. Funding: Supported
as the “source” of the small energy gap while total two-terminal resistance r1,2 is a series by ONR grant N-000141512370 and Penn State 2DCC-MIP under
overlooking the fact that the superconductor resistance of (n + 1) QAH sections, each con- NSF grant DMR-1539916 (N.S.); DOE grant DE-FG01-08ER46531
strip also serves as an “electrical short” for the tributing h/e2 (26). (Q.L.); NSF grant DMR-1707340 (M.H.W.C.); NSF-CAREER award
DMR-1847811, ARO Young Investigator Program Award
QAH device. Our results demonstrate that if the SC layer W911NF1810198, and an Alfred P. Sloan Research Fellowship
Our results, on the other hand, show that is strongly coupled to the QAH sample, the two- (C.-Z.C.); and EU ERC-AG Programs (project 3-TOP and 4TOPS)
the s1,2 = e2/h plateau in the C = 1 phase is terminal conductance s1,2 is half-quantized (C.G. and L.W.M.). Support for transport measurements and data
analysis at Penn State is provided by DOE grant DE-SC0019064.
very likely a result of decoupling of the QAH throughout the magnetic field range where Author contributions: N.S., M.H.W.C., and C.-Z.C. conceived and
insulator from the Nb layer. Hence, it is not the magnetization is well aligned. The agree- designed the experiment; D.X. and R.X. grew the QAH samples.
predicated upon the existence of a TSC phase ment among the data obtained from the var- M.K. fabricated the devices; M.K., J.S., and R.Z. performed the
dilution refrigerator measurements; M.K., J.J., F.W., Y.-F.Z.,
with N = 2. To exclude the possibility that the ious QAH samples with different geometries and L.Z. carried out the PPMS transport measurements; C.G. and
s1,2 ~ 0.5e2/h plateau observed in our QAH demonstrates the robustness, reproducibility, L.W.M. conceived and designed the experiment done in Würzburg;
sandwich sample may be caused by the ab- and generality of the presented phenomena. M.W. grew the QAH samples in Würzburg; K.M.F. and P.M.
fabricated the devices and performed the dilution refrigerator
sence of the zero–Hall conductance plateau Therefore, we conclude that the observation measurements in Würzburg; M.K., C.G., N.S., M.H.W.C., and C.-Z.C.
(i.e., the C = 0 phase), we carried out mea- of s1,2 ~ 0.5e2/h plateau alone is not sufficient wrote the manuscript; all authors contributed to the analysis
surements on QAH samples with the C = 0 evidence for the existence of chiral Majorana of the data and the final editing of the manuscript. Competing
interests: The authors declare no competing interests. Data and
phase. We fabricated two 6QL Cr-doped (Bi, edge modes and the N = 1 TSC phase in the materials availability: All data in the main text and the
Sb)2Te3 samples similar to the ones used in millimeter-size QAH-SC hybrid structures. supplementary materials are available at (43).
(23). Next, we scratched both samples into
RE FERENCES AND NOTES
millimeter-size Hall bar structures (0.5 mm × SUPPLEMENTARY MATERIALS
1. E. Majorana, Nuovo Cim. 14, 171–184 (1937).
1 mm) and then sputtered Nb strips onto the science.sciencemag.org/content/367/6473/64/suppl/DC1
2. N. Read, D. Green, Phys. Rev. B 61, 10267–10297 (2000).
samples with a mask. We measured s1,2 across 3. C. W. J. Beenakker, Annu. Rev. Condens. Matter Phys. 4,
Supplementary Text
one and two Nb strips. The value of s1,2 for one 113–136 (2013).
Figs. S1 to S10
Nb strip sample (Fig. 4A) is similar to that 4. J. Alicea, Rep. Prog. Phys. 75, 076501 (2012).
5. X. L. Qi, S. C. Zhang, Rev. Mod. Phys. 83, 1057–1110 (2011).
measured in the QAH sandwich sample (Fig. 6. A. Y. Kitaev, Ann. Phys. 303, 2–30 (2003). 9 April 2019; accepted 7 November 2019
3A). Therefore, the existence or the nonexis- 7. F. Wilczek, Nat. Phys. 5, 614–618 (2009). 10.1126/science.aax6361

SUPERCONDUCTIVITY two types of jumps (Fig. 1B): small ones cor-

responding to the manipulation of a near-
Atomic manipulation of the gap in Bi2Sr2CaCu2O8+x surface oxygen dopant, which we can detect
through their signature in the differential con-
F. Massee1*, Y. K. Huang2, M. Aprili1 ductance (8, 9, 20), and big jumps correspond-
ing to the manipulation of a surface Bi atom
Single-atom manipulation within doped correlated electron systems could help disentangle the on the crest of the supermodulation that are
influence of dopants, structural defects, and crystallographic characteristics on local electronic observed directly in the constant current
states. Unfortunately, the high diffusion barrier in these materials prevents conventional images (Fig. 1C) [section 1 of (21)]. We can
manipulation techniques. Here, we demonstrate the possibility to reversibly manipulate select sites freeze-in any new surface-and-dopant config-
in the optimally doped high-temperature superconductor Bi2Sr2CaCu2O8+x using the local electric uration by switching back to low bias voltage.
field of the tip of a scanning tunneling microscope. We show that upon shifting individual Bi atoms at Because the spatial extent of the highest electric
the surface, the spectral gap associated with superconductivity is seen to reversibly change by field is determined by the size of the tip apex,
as much as 15 milli–electron volts (on average ~5% of the total gap size). Our toy model, which which is typically a few tens of nanometers
captures all observed characteristics, suggests that the electric field induces lateral movement of in diameter, in principle, several hundred Bi
local pairing potentials in the CuO2 plane. and oxygen dopants can be affected. However,
unlike in a homogeneous system where the
O
threshold field for manipulation is identical
ne of the challenges in the study of high- This is because the dopant atoms are buried for all field-affected atoms or molecules (19),
temperature superconductivity in the under the surface and the diffusion barrier the existence of intrinsic lattice distortions, the
cuprates (1) is their intrinsic inhomo- of the surface atoms themselves is too high. supermodulation, and numerous local dopant
geneous nature. This is exemplified by Alternatively, the electric field can be used to environments in Bi2212 (20) leads to a range
the archetypal system Bi2Sr2CaCu2O8+x manipulate a surface (17–19); however, be- of threshold fields, enabling selective and re-
(Bi2212), whose complicated crystal structure cause the field profile depends on the size versible manipulation by carefully tuning the
includes an incommensurate structural super- of the tip apex, this process is, in practice, tip position and the manipulation voltage.
modulation (2) and interstitial oxygen dopant difficult to control unless specific atoms are As expected for electric field–induced manip-
atoms. Particularly notable is the large varia- more sensitive to the field than others owing ulations, all manipulations we observe are
tion in spectral gap size over nanometer-scale to their charge or local environment. We dis- confined to a roughly circular area with a
distances (3–5), which has been shown to covered that this is exactly the case in Bi2212, diameter of a few tens of nanometers around
reflect local variations in the superconducting where we find two local environments that are the location of the tip during the application
transition temperature Tc (6, 7) and has been more readily influenced by the electric field of the manipulation voltage (fig. S5). Owing to
correlated to both oxygen dopants (8, 9) and than the rest of the system. the aforementioned intrinsic inhomogeneous
structural inhomogeneity (10). These correla- Figure 1A shows a schematic of the two nature of Bi2212, studying the influence of
tions, however, typically involve averages over environments that allow for field-induced single atoms on the local electronic structure
a large number of sites. To directly probe the atom manipulation: surface Bi atoms on the is normally impossible, and averaging over a
influence of dopants, structural defects, and crest of the periodic modulation of the bulk large number of sites is required instead. With
crystallographic characteristics on the local crystal structure (2), henceforth referred to as the ability to locally rearrange a select number
electronic states at the atomic scale, direct non- the supermodulation, and the recently dis- of atoms, however, we can effectively remove
invasive control of the dopant positions and the covered weakly coupled oxygen dopants (20). the inhomogeneous background information
inhomogeneous crystal structure they inhabit To manipulate the atoms, we position our tip by considering the difference in local electronic
is therefore desirable. Unfortunately, common above the surface and slowly increase the sam-
techniques for single-atom manipulation in- ple bias voltage, Vs. Upon reaching ~800 mV
volving short-range forces between the tip and at a tunnel current of ~100 pA, we start to 1
Laboratoire de Physique des Solides, CNRS UMR 8502,
the atom (11–13) and/or vibrational excita- observe jumps in the current, where each Bâtiment 510, Université Paris-Sud, Université
tion using the tunneling current (14–16) are jump corresponds to the manipulation of an Paris-Saclay, 91405 Orsay, France. 2Institute of Physics,
University of Amsterdam, 1098XH Amsterdam,
not well suited for controllable manipulation atom below the tip. For manipulation volt- Netherlands.
of atoms in single-crystal cuprate materials. ages Vs ≥ 1.2 V, the current typically shows *Corresponding author. Email: freek.massee@u-psud.fr
A B Vs = 1.3 V C
150 Bi
high
Current (pA)
E Bi 140
z(r)
2-
O2- 130 O
low
120
0 5 10 15 20 25 Time (s)
Fig. 1. Electric field–induced atom manipulation. (A) Illustration of the field-induced manipulation of surface Bi atoms and near-surface oxygen
dopants (Bi, blue; O, black or orange; Cu, red). (B) Current as a function of time at Vs = 1.3 V. Small jumps signal oxygen dopant manipulation (O2−),
large jumps signal surface Bi rearrangement (Bi). (C) Constant current images, z(r), (Eset = −100 meV; Iset = 100 pA) of reversible surface Bi atom
manipulation upon treatment with Vs > 1.2 V. The atoms in the dashed box are manipulated. Scale bars, 1 nm. All measurements throughout this
work were performed at temperature T ≤ 4.2 K.

states before and after manipulation. These in equal measure in the vicinity of the ma- nature of the manipulated oxygen atoms are
difference measurements will only show con- nipulated atoms. exemplified by the simulation shown in Fig.
trast where the density of states has been Using the topographies taken simultaneous- 2F. As discussed below, the simulation, which
altered by the atom manipulation, providing ly with the gap maps, we can next pinpoint is in good agreement with experiment, uses
a direct link between single atomic sites and where surface Bi atoms have been manipu- a model that takes into account only Bi
the local electronic structure. lated (fig. S8 and S9); the locations of ma- manipulations.
To generate a difference map of the spectral nipulated Bi atoms have been marked with Atomic scale difference images of two con-
gap, we record low-energy differential conduct- black dots on the two gap maps (Fig. 2, A and secutive manipulations of the same Bi atom
ance maps and extract a map of the peak-to- B) and in their difference map in Fig. 2D. are shown in Fig. 3A. Two observations stand
peak gap, Di(r), using identical settings before Similarly, from high energy differential con- out. First, whenever a surface modification
and after atom manipulation. Figure 2, A and ductance maps we can extract the position of reverts in a subsequent electric field treatment,
B, shows one example of a gap map taken all near-surface oxygen dopants, as well as the gap reverts back as well. Consequently,
before and after ~10 dopants and Bi surface determine which dopants have changed [sec- the difference images between the first and
atoms have been manipulated. Three subse- tion 1 of (21)]. The gap changes in Fig. 2D are third map, D3−1(r), is featureless (fig. S11),
quent electric field manipulation treatments clearly linked to the atomic manipulations, and D3−2(r) = −D2−1(r) (Fig. 3A, right and
are shown in section 4 of (21). Whereas the but in a rather unexpected manner: The al- left, respectively). For isolated manipulation
topographies and the gap maps are mostly tered Bi sites mark the boundary between re- events, even the full spectrum reverts to the
identical, excluding any change of the tip itself, gions of increasing and decreasing gaps, original configuration (fig. S12). Second, the
select locations show substantial changes, with whereas the gap on the sites itself is hardly direction along which the gap is altered ap-
gaps increasing and decreasing by >10 meV. affected. Whereas one would expect the ma- pears to be linked to the direction of the sur-
The two spectra in Fig. 2C, which are taken at nipulation of oxygen dopants to have a strong face Bi atom repositioning. As can be seen
the same location before and after the electric effect on the gap, the contribution to the gap from Fig. 3A and as shown schematically in
field–induced manipulation, highlight the modifications of the near-surface dopants we Fig. 3B, each time a Bi atom moves down be-
significance of these changes. The spectra are manipulate does not appear to be the dominant low (up into) the BiO plane, a neighboring atom
predominantly affected at the peak energies, one: The correlation between their location laterally shifts toward (away) from it, leading
whereas the low energy states, |E| < 20 meV, and where the gap changes is only moderate, to an enhancement of the spectral gap in the
are hardly modified, if at all, highlighting the and when we manipulate a single near-surface direction of this shift.
insensitivity of the latter to disorder (22) (fig. S7). dopant atom, the gaps in its vicinity shift by a To quantify the direction of the gap change
From the two gap maps, we then calculate the few milli–electron volts at most (fig. S6). This in more detail, we average the D(r) images
difference map, D2−1(r) = D2(r) − D1(r), which observation is in line with a previous study around each surface Bi manipulation after
shows directly where the gap has enhanced that found these dopants (i.e., those with a aligning their direction of maximum gap in-
(blue) or decreased (red) upon the electric field resonance at −1 eV < E < −0.4 eV) to have a crease [see section 4 of (21)]. As Fig. 4A shows,
manipulation (Fig. 2D). As the histograms of much weaker correlation to the intrinsic gap a clear dipole profile, centered at the sur-
the gap changes in Fig. 2E show, the average inhomogeneity than the ones closer to the face modification in an otherwise unaffected
gap size over the entire field of view is pre- CuO2 plane (9). The dominant contribution environment, is obtained. The two lobes of
served: The gap is both reduced and enhanced of manipulated Bi atoms and the secondary the dipole have opposite signs, are a few
12
A B C gap increased D
10
160
meV 8
g (nS)
Δ(r)
4
55
meV
2
before
1 nm 1 nm after 1 nm
0
-100 -50 0 50 100 -15 meV +15 meV
Vs > 1.2 V
Energy (meV) D2-1(r)
Fig. 2. Gap modification. Maps of the peak-to-peak gap, D, taken (A) before E i=1 F
12
number of pixels (x10 3)
and (B) after field-induced atom manipulation at 1.5 V and ~100 pA. Dots mark i=2
i=3
the locations where Bi atoms have been manipulated. (C) Spectrum taken i=4
before and after manipulation on the location indicated by an X in (A) and (B),
8
showing a clear change in D. (D) Difference of the gap maps taken (A) before
and (B) after manipulation: D2−1(r) = D2(r) – D1(r). Black dots mark the
same locations as in (A) and (B), the dashed box indicates the area of Figs. 1C
4
and 3A. (E) Histograms of D = Di+1 − Di for four consecutive manipulations;
i = 1 corresponds to the difference map in (D). For all maps, the average change
in the gap size is zero. (F) Simulation of the difference map in (D) using our 1 nm
0
model of a shifting Gaussian. The fitted dipole profile (Fig. 4B, bottom) is -10 0 10 -15 meV +15 meV
placed on every manipulated site (black dots) using the orientation extracted Δ i+1 - Δ i (meV) Dcalc(r)
from fits to the data (fig. S13) [see also section 4 of (21)].

Fig. 3. Atomic position A D2-1 = Δ2 - Δ1 D3-2 = Δ3 - Δ2 of ~50 meV, which is not unreasonable given the
dependence of gap changes. average peak-to-peak gap size of D = 96 meV.
(A) D2−1 = D2(r) – D1(r) image on the +15 Lastly, the width of the Gaussian is 1.7 nm,
same area as the constant current meV which is comparable to the superconducting
images in Fig. 1C. The manipulated Bi coherence length.
atoms are marked by circles, the A Gaussian-shaped gap distribution that
lateral movement upon manipulation falls off with length scales on the order of the
is indicated by an arrow. The right coherence length is highly suggestive of a lo-
-15 cal pairing potential originating from a point-
panel shows the opposite gap
meV
modification upon a reversion of like object. When we laterally move this entity
the atomic configuration after a using the electric field of the tip, its local
subsequent Vs = 1.5 V manipulation. pairing potential moves with it, leading to a
B
Scale bars, 1 nm. (B) Schematic dipole-shaped feature in the difference image.
Bi Bi The physical origin of the object is possibly
of the surface change and its effect
on the gap: One atom moves up Δ2 > Δ1 Δ2 < Δ1 Δ3 < Δ2 Δ3 > Δ2 related to the Bi atoms themselves, although
(down), another shifts laterally away the contribution of the Bi orbitals to the low
from (toward) it, resulting in gap enhancement in the direction of the shift. More details and examples energy density of states is limited (23). More
can be found in sections 1 and 4 of (21). likely, the apical oxygen atoms directly below
the Bi atoms shift concomitant with the Bi
manipulation, leading to the gap modifica-
tions. The importance of the apical oxygen
Fig. 4. Dipole profile and toy model. A 3 atoms in both the tunneling process (24–26)
(A) Average D(r) for all topographic and the gap size (10) have been stressed pre-
modifications after aligning their 2 viously. Our observations provide additional
orientation. The arrow indicates where
distance (nm)
1 input for further theoretical investigations—
the vertical line trace on the right is particularly those that take into account on-
taken for the experimental data 0 site correlations (27, 28)—into the origin of the
(red circles) and for the toy model data
spectral gap, and the tunneling process, in this
Gaussian -1
using a Gaussian (black dashed line) cuprate superconductor.
Lorentzian
and Lorentzian (blue dashed line) -2 An alternative source of the dipole-shaped
peak profile. (B) A 2D Gaussian difference image could be the appearance of
1 nm -3
(width = 1.7 nm; top left) is shifted topological defects that introduce a 2p-rotation
by a fraction of a nanometer (top right, 5 0 -5
of the phase of the order parameter (29). How-
shift exaggerated for clarity), leading D (meV)
ever, creation and annihilation of topological
to a difference (bottom) upon their B defects has to occur in pairs. Given our finite
shift ~ Å
subtraction with the same shape and
Δ (meV)
field of view, it may be that one-half of the pair

length scales as the experimental 100
Δ1 Δ2 is outside our measurement range, but given
data in (A). The inset shows how the the large field-of-view topographic before-and-
~nm
amplitude of the subtracted after comparison (fig. S5), and assuming that
10
Gaussians depends on the shift 0 Shift (nm) 1.0 both pairs will have a signature in topography,
distance: a 2-Å shift requires this is unlikely. Furthermore, in our optimally
an amplitude of D ~ 50 meV Vs > 1.2 V
Δ2 > Δ1 doped system we do not find a significant cor-
(dashed lines). y relation between the gap changes and topo-
D(r) logical defects in the smectic, or the d-form
= Δ2 - Δ1 x Δ2 < Δ1 factor density wave, reported for underdoped
~nm samples (30, 31). Extension of our work to
lower doping concentrations, where the vari-
ous charge- and spin-ordered states are more
nanometers in size, and are located at opposite observe in the experiment. For shifts smaller predominant, should give a more definite an-
ends, ~1.5 nm distance from the surface ma- than the width of the peak, the lateral extent swer to this issue. Additionally, the influence
nipulation. The horizontal line-cut on the of the resulting dipole will depend exclusively of dopants and the surface structure on these
right side of Fig. 4A shows this in more detail. on the width, whereas the magnitude of the (ordered) states themselves can be studied di-
The most straightforward way to create a dipole is a function of the size of the shift and rectly using the field-induced atom manipula-
dipole-shaped profile in a difference image is the peak amplitude (Fig. 4B, inset). Using this tion we introduce in this work.
by subtracting two identical [two-dimensional simple toy model, we can accurately repro- The observation of a profound influence on
(2D)] peaks that are laterally shifted with reduce both the size and shape of the observed the peak-to-peak gap in tunneling experiments
spect to each other (Fig. 4B). In the context of dipole when we use a Gaussian profile for the of subnanometer shifts in atomic positions
our experiment, this would correspond to a peak, as well as simulate the experimental highlights the importance of the lattice on
single peak that laterally shifted by the electric data (Fig. 2F). Conversely, a Lorentzian pro- the local electronic properties of the cuprates.
field manipulation, which upon generating file decays too slowly to properly fit the tails The spatial profile of the gap modification we
the difference map is thus subtracted from of the difference image, as can be seen from observe is highly suggestive of the field-induced
itself at a slightly different location. The max- the line-cut comparison in Fig. 4A. For a re- lateral movement of a local pairing potential in
ima of the subtraction, i.e., the lobes of the alistic lateral shift of 2 Å (see, for example, the CuO2 plane originating from a point-like
dipole, are in the direction of the shift, as we Fig. 1C), the Gaussian requires an amplitude object. This work demonstrates an avenue to

noninvasively and reversibly probe the influ- APPLIED PHYSICS

ence of the local lattice on the electronic states
of cuprate high-temperature superconductors
and related compounds.
Topological mechanics of knots and tangles
Vishal P. Patil1, Joseph D. Sandt2, Mathias Kolle2, Jörn Dunkel1*
RE FE RENCES AND N OT ES
1. B. Keimer, S. A. Kivelson, M. R. Norman, S. Uchida, J. Zaanen,
Nature 518, 179Ð186 (2015). Knots play a fundamental role in the dynamics of biological and physical systems, from DNA to
2. Y. Gao, P. Lee, P. Coppens, M. A. Subramania, A. W. Sleight, turbulent plasmas, as well as in climbing, weaving, sailing, and surgery. Despite having been studied
Science 241, 954Ð956 (1988). for centuries, the subtle interplay between topology and mechanics in elastic knots remains poorly
3. T. Cren, D. Roditchev, W. Sacks, J. Klein, EPL 54, 84Ð90
(2001). understood. Here, we combined optomechanical experiments with theory and simulations to
4. C. Howald, P. Fournier, A. Kapitulnik, Phys. Rev. B 64, 100504 analyze knotted fibers that change their color under mechanical deformations. Exploiting an analogy
(R) (2001). with long-range ferromagnetic spin systems, we identified simple topological counting rules to
5. S. H. Pan et al., Nature 413, 282Ð285 (2001).
6. K. K. Gomes et al., Nature 447, 569Ð572 (2007).
predict the relative mechanical stability of knots and tangles, in agreement with simulations and
7. A. N. Pasupathy et al., Science 320, 196Ð201 (2008). experiments for commonly used climbing and sailing bends. Our results highlight the importance of
8. K. McElroy et al., Science 309, 1048Ð1052 (2005). twist and writhe in unknotting processes, providing guidance for the control of systems with
9. I. Zeljkovic et al., Science 337, 320Ð323 (2012).
10. J. A. Slezak et al., Proc. Natl. Acad. Sci. U.S.A. 105, 3203Ð3208
complex entanglements.
(2008).
K
11. D. M. Eigler, E. Schweizer, Nature 344, 524Ð526 (1990).
12. L. Bartels, G. Meyer, K.-H. Rieder, Phys. Rev. Lett. 79, 697Ð700 nots are among the oldest, most endur- topological transformations are typically low
(1997).
13. M. Ternes, C. P. Lutz, C. F. Hirjibehedin, F. J. Giessibl,
ing human technologies, as valuable to in liquids and gases (17), they tend to become
A. J. Heinrich, Science 319, 1066Ð1069 (2008). ancient builders (1) and mariners (2) as prohibitively large in entangled solids (5, 6).
14. D. M. Eigler, C. P. Lutz, W. E. Rudge, Nature 352, 600Ð603 to modern engineers and surgeons (3). This fact has profound consequences for the
(1991).
Thought to predate the wheel (1), knotted stability and function of natural and engi-
15. T.-C. Shen et al., Science 268, 1590Ð1592 (1995).
16. T. W. Fishlock, A. Oral, R. G. Egdell, J. B. Pethica, Nature 404, structures owe their extraordinary longevity neered structures, from the microscopic knots
743Ð745 (2000). and widespread usage to an inherent mechan- in DNA (18, 19), proteins (20–22), and polymers
17. L. J. Whitman, J. A. Stroscio, R. A. Dragoset, R. J. Celotta, ical robustness that arises from the subtle (23) to knitted clothes (8) and macroscopic
Science 251, 1206Ð1210 (1991).
18. M. A. Rezaei, B. C. Stipe, W. Ho, J. Chem. Phys. 110, interplay of topology, elasticity, and friction. meshworks (24). Achieving a unified under-
4891Ð4896 (1999). Over the course of many centuries, sailors, standing of these various systems requires
19. M. Alemani et al., J. Am. Chem. Soc. 128, 14446Ð14447 weavers, climbers, and surgeons have acquired taking into account not only their topological
(2006).
20. F. Massee, Y. K. Huang, M. S. Golden, M. Aprili, Nat. Commun. a wealth of knowledge about the benefits and but also their elastic properties. Because key
10, 544 (2019). drawbacks of various types of knots (1, 2). concepts from topology and elasticity theory
21. Materials and methods are available as supplementary Experience has taught us that certain knots remain applicable over a wide range of scales,
materials.
22. M. C. Boyer et al., Nat. Phys. 3, 802Ð806 (2007).
are more stable than others, but we are still deciphering the topological principles (5, 6, 9)
23. M. S. Hybertsen, L. F. Mattheiss, Phys. Rev. Lett. 60, largely unable to predict the mechanical be- that determine the mechanical stability of
1661Ð1664 (1988). havior of knots and tangles from basic topo- knots promises insights into a broad spectrum
24. I. Martin, A. V. Balatsky, J. Zaanen, Phys. Rev. Lett. 88, 097003
(2002).
logical observables (4), such as the number of physically entangled structures. Therefore,
25. Y. He, T. S. Nunner, P. J. Hirschfeld, H.-P. Cheng, Phys. Rev. and relative ordering of crossings. Although our main goal is to identify generic topological
Lett. 96, 197002 (2006). recent experimental and theoretical research counting rules that enable us to estimate which
26. J. Nieminen, H. Lin, R. S. Markiewicz, A. Bansil, Phys. Rev. Lett.
has revealed important insights into the com- members of a given family of elastic knots are
102, 037001 (2009).
27. A. Kreisel et al., Phys. Rev. Lett. 114, 217002 (2015). petition between force transmission and fric- the most robust against untying. To this end,
28. P. Choubey, A. Kreisel, T. Berlijn, B. M. Andersen, tion in special classes of knots (5, 6), hitches we combined elements from mathematical and
P. J. Hirschfeld, Phys. Rev. B 96, 174523 (2017). (7), and fabrics (8), there currently exists no physical knot theory (9, 25) with optomechan-
29. A. Mesaros et al., Science 333, 426Ð430 (2011).
30. M. J. Lawler et al., Nature 466, 347Ð351 (2010). comprehensive mathematical theory (9) link- ical experiments and quantitative continuum
31. K. Fujita et al., Proc. Natl. Acad. Sci. U.S.A. 111, E3026ÐE3032 ing the topological and mechanical proper- modeling (Fig. 1).
(2014). ties of knotted elastic structures. We were interested in tying two lines to-
32. F. Massee, Supporting data for ÒAtomic manipulation of the
gap in Bi2Sr2CaCu2O8+xÓ by F. Massee et al., Version 1, Harvard Physical knots and their topology first gether so that they form a stable longer rope, a
Dataverse (2019); https://doi.org/10.7910/DVN/UUSS6R. assumed a central role in science with the task known as “tying a bend” among sailors (2).
introduction of Kelvin’s vortex-atom model Mathematically, this configuration describes
ACKN OW LEDG MEN TS
in the 1860s (10). Since then, the fundamen- an oriented 2-tangle, defined as the union of
We thank M. Civelli, A. Mesaros, and P. Simon for fruitful
discussions. Funding: F.M. acknowledges funding from H2020 tal importance of entangled structures has two oriented open curves embedded in space
Marie Skłodowska-Curie Actions (grant 659247) and the ANR become firmly established in a diverse range (4). Although an elegant mathematical formal-
(ANR-16-ACHN-0018-01). Author contributions: F.M. and M.A. of disciplines and contexts (9). In physics, for ism exists to describe certain simple families
conceived of the study and discussed and interpreted the results.
F.M. performed and analyzed all measurements. Y.K.H. grew the
example, interactions between knotted defect of 2-tangles (26), little is known theoretically
samples. F.M. wrote the manuscript with M.A. Competing lines are essential to understanding and con- about even the most basic bend knots used in
interests: The authors declare no competing interests. Data and trolling dynamics and mixing in classical and practice. We constructed a topological phase
materials availability: The data files for the results presented here
are available at (32).
complex fluids (11–14), including liquid crystals diagram that explains the relative stability
(15), plasmas (16), and quantum fluids (17). of a selection of bends that are commonly
SUPPLEMENTARY MATERIALS Whereas the energetic costs associated with used in the sailing and climbing commun-
science.sciencemag.org/content/367/6473/68/suppl/DC1 ities. To validate the underlying topologi-
Materials and Methods cal model, we compared its predictions with
Supplementary Text 1
Figs. S1 to S13
Department of Mathematics, Massachusetts Institute of simulations of an optomechanically verified
Technology, Cambridge, MA 02139, USA. 2Department continuum theory and with quantitative mea-
References (33Ð38)
of Mechanical Engineering, Massachusetts Institute of
8 February 2019; accepted 7 November 2019 Technology, Cambridge, MA 02139, USA. surements using laboratory “prisoner’s es-
10.1126/science.aaw7964 *Corresponding author. Email: dunkel@mit.edu cape” experiments.

Fig. 1. Experiments and simulations reveal mechanical properties of revealing the relative strength of bending and stretching strains along knots.
knots. (A and B) Color-changing mechanoresponsive fibers confirm the (D) and (E) show the evolution of these two complementary strain
stress patterns predicted by continuum simulations for the trefoil knot (A) contributions during tightening of the trefoil knot in (A) and the figure-of-
and the figure-of-eight knot (B) during the tightening process (movie S1). eight knot in (B). Pulling force is 0.02 N. The elastic moduli are given in
Fiber diameter is 0.4 mm. (C) Dependence of fiber color on strain (28). (F) Topology-preserving Reidemeister moves affect the elastic energy of
visualized as a trajectory in the CIE 1931 XYZ color space, where mean the underlying fibers differently. Move R1 induces strain and thus requires
positions (solid circles) lie within standard deviation ellipses (28). This strain higher energy than R2 and R3, highlighting that both topological and elastic
color coding is used in panels (A), (B), and (F). (D and E) Simulations properties determine the mechanical behavior of knots.
Our optomechanical experiments use re- ponsive photonic fibers (Fig. 1, A to C). Sim- ties of tangled fibers to their topology, one
cently developed color-changing photonic ulating the tightening process of a 1-tangle, must merge concepts from classical mathe-
fibers (27) that allow for the imaging of strain corresponding to a single knotted fiber pulled matical knot theory (4) with elasticity theory
in knots (Fig. 1, A and B). These fibers derive at both ends (Fig. 1, A and B, and movie S1), (5, 6, 9, 30).
structural color from a multilayer cladding reveals the relative strengths and localization Continuum simulations provide guidance
composed of alternating layers of transpar- of the bending and stretching strains (Fig. 1, for how one can complement bare topolog-
ent elastomers with distinct refractive indices D and E), which are not individually dis- ical knot diagrams (4) with coarse-grained
wrapped around an elastic core. Their color- cernible in our experiments. Furthermore, the mechanical information that is essential for
ation varies with the thickness of layers in the Kirchhoff model highlights why topological explaining why certain knots are more stable
periodic cladding, which changes upon elon- considerations (4) alone do not suffice to ex- than others (Fig. 2). In contrast to a 1-tangle,
gation or bending. As is typical of macroscopic plain the mechanical behaviors (2) of real- which is tightened by pulling diametrically at
materials at room temperature, the persistence world knots: Loosening or tightening a knot its two ends (Fig. 2A), each strand of a bend
lengths of the fibers used in our experiments transforms any of its planar projections ac- knot has one pulled and one free end (Fig. 2B).
are several orders of magnitude larger than cording to a sequence of three elementary Therefore, the local fiber velocity directions
the diameters of the tightened knots (28), with topology-preserving Reidemeister moves, R1, in the center-of-mass frame of the bend knot
empirical knowledge (2) indicating that es- R2, and R3 (Fig. 1F). Despite being topolog- define natural fiber orientations on the underly-
sential knot properties are only weakly depen- ically equivalent, the move R1 is energetically ing 2-tangle (Fig. 2B), thus establishing map-
dent on the elastic modulus. Theoretically, distinct as it involves substantial changes in ping between bends and oriented 2-tangles. At
we describe knotted fibers using a damped strain, whereas moves R2 and R3 are energet- each contact crossing, the fibers mutually gen-
Kirchhoff model (5, 28, 29) validated through ically favored soft modes (Fig. 1F), implying erate a frictional self-torque with well-defined
comparison with photographs depicting the that physical knots preferentially deform by handedness, depending on the relative velocity
strain-induced color changes in mechanores- R2 and R3. Thus, to link the physical proper- and ordering of the two fiber strands (Fig. 2, B

A 1-tangle (trefoil knot) B 2-tangle (reef knot) C

1
x 10-3 x 10-3
––
2
+8 +8
Self - torque density (N)
Self - torque density (N)

1
––
2
0 0
– – – 1
+–
2
+ 1
-8 -8 +–
+ + 2
+
Fig. 2. Topology and self-twisting in 1-tangles and 2-tangles. (A) Top: directions and fiber orientation permits the discretization of self-torque
A 1-tangle is tightened by pulling its two ends in opposite directions (large data over crossings by assigning twist charges qi = ±1 to each vertex i as
exterior arrows). The induced fiber velocity field (small interior arrows) in the described in the next panel. (C) Each individual fiber strand passing through
center-of-mass frame reverses its orientation near the fiber midpoint. vertex i induces a rotation in the other strand, thus contributing ±1/2
Bottom: As the velocity field is incompatible with any chosen global fiber to the vertex twist charge qi = ±1, with sign corresponding to rotation
orientation (black arrows), self-torque data cannot be consistently handedness. Blue–blue and red–red self-crossings found in more complex
assigned to a topological 1-tangle diagram. (B) Top: Because of the presence 2-tangles can be labeled accordingly. The sum of the qi defines the
of the two free ends, the pulling directions of a bend knot (large exterior total writhe Wr, providing a coarse-grained approximation of the total
arrows) define a canonical global orientation on each of the two fibers self-torque in 2-tangles; the reef knot has Wr = 0. Fiber diameter is
in the corresponding 2-tangle. Bottom: The alignment of local velocity 0.4 mm and pulling force is 15 N in (A) and (B).
and C). In analogy to the coarse-graining proce- where N is the crossing number, q ¼ ð1=N Þ perimeter in the tight limit, ensuring that each
P
dure underlying Ising-type spin models, we can i qi ¼ Wr=N is the average writhe, and t0 = face contributes a maximum of +1 to G.
associate a unit twist charge qi = ±1 with each 1 – 1/N can be interpreted as a ground-state The topological parameters N, t, and G allow
vertex i in the planar 2-tangle diagram, where energy density (28). Equation 1 has the form us to rationalize the stability of a large class
the sign of qi reflects the combined handedness of a ferromagnetic energy for an Ising-type of popular knots used by sailors and climbers
of the torques acting on the two intersecting spin model with long-range interactions, em- (Fig. 3G). These variables are easy to evaluate
X phasizing the concept of knots as strongly from knot diagrams (Fig. 3, D to F) and reflect
strands (Fig. 2C). The sum Wr ¼ qi , math-
coupled systems. topology-induced forces and torques through-
i
ematically known as the writhe, represents In addition to twist locking for large values out the knot. As such, the triplet (N, t, G) cap-
the total self-torque of a 2-tangle, establish- of t, knots can be stabilized when their inter- tures both essential topological and mechanical
ing a concrete link between topology and nal structure forces fiber segments to slide structure hidden within knots. The (N, t, G)
mechanics. tangentially against each other. For example, phase diagram explains existing empirical
A key puzzle of physical knot theory (9), the the reef knot and the thief knot both have t = 1, knowledge for simple knots (2), as well as
empirically observed mechanical difference but because their pulled ends differ, friction predictions of the Kirchhoff model about the
between the visually similar reef and granny makes the reef knot more stable (Fig. 3, D relative strength and stability of more com-
knots (Fig. 3, A and B), may be understood as a and G). At the coarse-grained level of planar plex 2-tangles (Fig. 3G). We verified these
consequence of this torque–writhe correspon- knot diagrams, these friction effects corre- predictions independently in experiments
dence in 2-tangles. The underlying mecha- spond to edge-to-edge interactions dominated by mimicking the prisoner’s escape problem
nism becomes evident by considering a pair by pairs of edges sharing a face and pulled in (Fig. 4A) with two thin Dyneema fibers made
of crossings as shown in Fig. 3C. Whereas opposite directions (Fig. 3, D to F). To formal- from ultra–high molecular weight polyethylene
equally directed torques lead to rolling, oppo- ize this notion, each edge around a face F is tied together (28). Of the two pulled ends for
site torques promote locking and thus stabilize assigned a weight of +1 or –1 if it winds around each knot, one is fixed in the experimental
a knot against untying. The overall stability of F in the anticlockwise or clockwise direction, apparatus and the other is perturbed while
2-tangles therefore depends on the self-torque respectively. Each face then contributes a fric- suspending incrementally higher masses
distribution along the fibers, as encoded by the tion energy given by the net circulation of the until the knot pulls through. Although the
vertex twist charges qi = ±1 in an untightened edges around the face, CF, normalized by the Kirchhoff model cannot account for surface
knot diagram (Fig. 3, D to F). The above argu- total number of edges eF. This yields the total contact details (30), the experimental data
ment suggests the following topological twist circulation energy: for the critical loads agree quantitatively
fluctuation energy per site: with the simulations for simple knots and,
X jC
Fj more importantly, confirm the predicted qua-
1
X G¼
eF
ð2Þ litative stability differences between various
t¼ ðqi Þ2
q F commonly used knots (Fig. 4B). Notably, both
N
i
2
X theory and experiments indicate that the
¼ t0 ðN Þ qi qj ð1Þ where the sum is taken over all faces of the Zeppelin knot is more secure against untying
N2 knot diagram. The normalization encodes than the popular alpine butterfly knot (Fig. 4,
i<j
the assumption that every face has the same B and C).

Fig. 3. Topology determines the mechanical stability of 2-tangles. (A) Small oriented by pulling direction correspond to a topological state defined as the triple
modifications in topology lead to substantial changes in the mechanical behavior of crossing number N, twist fluctuation t, and circulation G. These parameters
of 2-tangles, exemplified by the presence or absence of global rotation of the knot explain the relative stability of knots in the reef group (D) and the Carrick group (E).
body upon pulling (movie S2); fiber diameter is 0.4 mm and pulling force is 15 N. (F) The Zeppelin bend is more stable than the alpine butterfly bend, displaying
Knots are shown in order of least stable (grief knot) to most stable (reef knot). both higher twist fluctuation and higher circulation. (G) Topological state reveals the
(B) Simulated tight configurations of knots are validated with real knots tied in nylon underlying structure of bend knots and separates stable knots from unstable
rope (diameter, 20 mm) with horizontal ends being pulled. (C) Tight knots act knots. The dimensionless topological friction, obtained from simulation, is
on themselves by right-handed (positive) and left-handed (negative) torques. determined by the velocity response when the knot is pulled with a given force
Equally directed torques lead to rolling (top), whereas opposite torques promote and is a measure of the friction force caused by the knot (28). Labels in (G)
locking (bottom) and thus stabilize a knot against untying. (D to F) Knot diagrams correspond to those in (D), (E), and (F) and additional knots listed in fig. S3.

Fig. 4. Experiments for commonly used knots validate the theoretically the knot (28); black boxes indicate standard deviations of the individual experiments,
predicted phase diagram. (A) Our experimental setup (28) mimics the classical with N = 9 grief (Gf), N = 8 thief (Th), N = 12 granny (Gy), N = 16 reef (Re),
prisoner’s escape problem by determining the critical pulling force F = mg at N = 6 alpine butterfly (Ab), and N = 6 Zeppelin (Ze) knots. For complex knots with
which two lines untie. (B) Experiments measuring the critical mass m at which high crossing number, such as the Zeppelin bend, more sophisticated models
two Dyneema fibers untie confirm the ranking of knot stability predicted by accounting for material-specific friction forces and three-dimensional contact details
simulations. For simpler knots with crossing number ≤8, averages (horizontal lines) need to be developed in the future. Fiber diameter is 0.15 mm. (C) Nonetheless,
over individual experiments (small filled circles) agree quantitatively with the relative simulations of complex bends with generic friction (28) show good shape agreement
strength predicted from simulations measuring both the velocity-based friction with tight configurations of bends in nylon ropes (diameter, 20 mm) and reveal
(large empty circles) and the total compression force (large empty triangles) within the highly nonuniform strain distributions in such knots.
To conclude, the above analysis shows how 3. J. B. Trimbos, Obstet. Gynecol. 64, 274–280 (1984). 28. Materials and methods are available as supplementary
basic topological counting rules can be used 4. C. C. Adams, The Knot Book: An Elementary Introduction to the materials.
Mathematical Theory of Knots (American Mathematical Society, 29. M. Bergou, M. Wardetzky, S. Robinson, B. Audoly, E. Grinspun,
to estimate the relative stability of frequently 2004). ACM Trans. Graph. 27, 63 (2008).
encountered knots and tangles. From a theo- 5. B. Audoly, N. Clauvelin, S. Neukirch, Phys. Rev. Lett. 99, 164301 30. J. H. Maddocks, J. B. Keller, SIAM J. Appl. Math. 47, 1185–1200
retical perspective, the parallels with long- (2007). (1987).
6. M. K. Jawed, P. Dieleman, B. Audoly, P. M. Reis, Phys. Rev. Lett. 31. V. P. Patil, Knot simulation code. Zenodo (2019);
range coupled spin systems suggest that the https://doi.org/10.5281/zenodo.3528928.
115, 118302 (2015).
statistical mechanics (4, 21) of general knotted 7. B. F. Bayman, Am. J. Phys. 45, 185–190 (1977).
structures can be formulated within this frame- 8. P. B. Warren, R. C. Ball, R. E. Goldstein, Phys. Rev. Lett. 120, AC KNOWLED GME NTS
work. Tangled vortices (12, 17) in complex fluids 158001 (2018). J.D. thanks the Isaac Newton Institute for Mathematical Sciences for
9. J. A. Calvo, K. C. Millett, E. J. Rawdon, Eds., Physical Knots: support and hospitality during the program “The Mathematical Design
and defect loops in liquid crystals (15) may
Knotting, Linking, and Folding Geometric Objects in R3, vol. 304 of New Materials” (supported by EPSRC grant EP/R014604/1) when
permit similar statistical descriptions through of Contemporary Mathematics (American Mathematical work on this paper was undertaken. We thank J. Takagi for producing
reduction to topological crossing diagrams. In Society, 2001). Fig. 4A. Funding: This work was supported by an Alfred P. Sloan
10. W. Thomson II, Philos. Mag. 34, 15–24 (1867). Research Fellowship (J.D.), a Complex Systems Scholar Award from
elastic systems, joint experimental and theo- the James S. McDonnell Foundation (J.D.), the Brigham and
11. H. K. Moffatt, R. L. Ricca, “Helicity and the Călugăreanu
retical progress is needed to untangle long- invariant” in Knots And Applications, L. H. Kauffman, Ed. Women’s Hospital through a Stepping Strong Innovator Award
standing puzzles regarding the statistics of (World Scientific, 1995), pp. 251–269. (J.D.S. and M.K.), and the National Science Foundation through the
“Designing Materials to Revolutionize and Engineer our Future”
knots in DNA (18) and proteins (20, 21), where 12. D. Kleckner, W. T. Irvine, Nat. Phys. 9, 253–258 (2013).
13. M. W. Scheeler, W. M. van Rees, H. Kedia, D. Kleckner, program (DMREF-1922321 to J.D.S. and M.K.). Author contributions:
thermal effects induce a finite persistence V.P.P. and J.D. developed theory. V.P.P. performed simulations,
W. T. M. Irvine, Science 357, 487–491 (2017).
length, and other macroscopic structures 14. S. Kuei, A. M. Słowicka, M. L. Ekiel-Jeżewska, E. Wajnryb, for which J.D.S. and M.K. provided data and code for converting strain
(8, 24). In sailing, climbing, and many other H. A. Stone, New J. Phys. 17, 053009 (2015). into perceived knot color. J.D.S. and M.K. designed color-changing
15. U. Tkalec, M. Ravnik, S. Čopar, S. Žumer, I. Muševič, fibers and conceived optomechanical experiments. J.D.S. conducted
applications, nontopological material param- optomechanical experiments and provided the description of the
Science 333, 62–65 (2011).
eters and contact geometry (30) also play im- 16. J. B. Taylor, Phys. Rev. Lett. 33, 1139–1141 (1974). experiments in the supplementary materials. J.D.S. and V.P.P.
portant mechanical roles and must be included 17. D. Kleckner, L. H. Kauffman, W. T. Irvine, Nat. Phys. 12, performed in-lab prisoner’s escape experiments. V.P.P. and J.D. wrote
650–655 (2016). the first draft of the paper. All authors discussed and revised the
in more refined continuum models to quanti- manuscript. Competing interests: The authors declare no competing
18. R. Stolz et al., Sci. Rep. 7, 12420 (2017).
tatively describe practically relevant knotting 19. A. R. Klotz, B. W. Soh, P. S. Doyle, Phys. Rev. Lett. 120, 188003 interests. Data and materials availability: The code used for
phenomena. From a broader conceptual and (2018). numerical simulations is available on Zenodo (31). All data are available
20. P. Virnau, L. A. Mirny, M. Kardar, PLOS Comput. Biol. 2, e122 in the main text or the supplementary materials.
practical perspective, the above topological
(2006).
mechanics framework seems well suited for 21. R. C. Lua, A. Y. Grosberg, PLOS Comput. Biol. 2, e45 SUPPLEMENTARY MATERIALS
designing and exploring new classes of knots (2006).
with desired behaviors under applied load. 22. D. Goundaroulis et al., Polymers 9, 444 (2017).
23. R. C. Ball, M. Doi, S. F. Edwards, M. Warner, Polymer 22,
Supplementary Text
1010–1018 (1981).
RE FE RENCES AND N OT ES Figs. S1 to S6
24. C. Baek, A. O. Sageman-Furnas, M. K. Jawed, P. M. Reis,
Tables S1 and S2
1. J. C. Turner, P. van de Griend, Eds., History and Science of Proc. Natl. Acad. Sci. U.S.A. 115, 75–80 (2018).
Movies S1 and S2
Knots, vol. 11 of K & E Series on Knots and Everything 25. R. D. Kamien, Eur. Phys. J. B 1, 1–4 (1998).
(World Scientific, 1996). 26. J. R. Goldman, L. H. Kauffman, Adv. Appl. Math. 18, 300–332
2. C. W. Ashley, The Ashley Book of Knots (Doubleday Books, (1997). 5 August 2019; accepted 15 November 2019
1944). 27. M. Kolle et al., Adv. Mater. 25, 2239–2245 (2013). 10.1126/science.aaz0135

FLUORESCENT PROTEINS tition between fluorescence and isomerization

is the excited-state (ES) barrier for chromo-
Electrostatic control of photoisomerization phore bond rotation (Fig. 1B, process 2). Here
we present a systematic study investigating
pathways in proteins the contributions of sterics and electrostatics
to energetic features of the chromophore’s
Matthew G. Romei1*, Chi-Yun Lin1*, Irimpan I. Mathews2, Steven G. Boxer1† potential energy surface in both the ground
state (GS) and ES. To experimentally probe
Rotation around a specific bond after photoexcitation is central to vision and emerging opportunities in these effects, we introduced a diverse range of
optogenetics, super-resolution microscopy, and photoactive molecular devices. Competing roles for substituents on the chromophore’s P ring
steric and electrostatic effects that govern bond-specific photoisomerization have been widely using amber suppression (9) with substituted
discussed, the latter originating from chromophore charge transfer upon excitation. We systematically tyrosine residues (10), taking advantage of the
altered the electrostatic properties of the green fluorescent protein chromophore in a photoswitchable chromophore’s autocatalytic maturation pro-
variant, Dronpa2, using amber suppression to introduce electron-donating and electron-withdrawing cess (Fig. 2A). The electronic perturbation to
groups to the phenolate ring. Through analysis of the absorption (color), fluorescence quantum yield, the chromophore due to the substituent can
and energy barriers to ground- and excited-state isomerization, we evaluate the contributions of sterics be thought of as analogous to a perturbation
and electrostatics quantitatively and demonstrate how electrostatic effects bias the pathway of of the protein environment around the chro-
chromophore photoisomerization, leading to a generalized framework to guide protein design. mophore that alters the chromophore’s elec-
tronic properties, as suggested by past studies
P
on polymethine dyes (11). As a model system,
hotoisomerizable chromophores, such required for nonradiative decay, as demon- we chose the widely used photoswitchable FP
as those in rhodopsins, phytochromes, strated by studies involving chemically locked Dronpa2 [the Met159→Thr (M159T) mutant
photoactive yellow proteins, and fluo- or artificially confined chromophores (6). An of Dronpa] because of the balance between
rescent proteins (FPs), rotate around alternative hypothesis identifies the role of its moderately high FQY and photoisomer-
specific bonds after photoexcitation in electrostatics in modulating the FQY. After ization efficiency (12). We also include results
the protein environment, which is essential to a perturbation to either the chromophore’s from a nonphotoswitchable FP, a superfolder
converting light energy into molecular motion electronic state (e.g., by photon absorption) GFP construct, to generalize the scope of our
(1). To investigate the role of the protein envi- or nuclear coordinates (e.g., by isomerization), conclusions.
ronment on tuning bound chromophore and/or a redistribution of the chromophore’s electron We expressed wild-type and 10 Dronpa2 var-
ligand functionality, we chose to study FPs, a density occurs, which is usually described as iants with chromophores containing electron-
relatively simple model system consisting of charge transfer between the rings. Conse- donating and electron-withdrawing substituents
an autocatalytically formed chromophore con- quently, the electric field exerted by the envi- on the P ring (Fig. 2B). X-ray crystal struc-
tained in a b barrel (2). The chromophore’s ronment can either promote or hinder charge tures confirm that the P-ring substituent(s)
local environment can markedly alter its photo- transfer and thus could control whether fluo- occupies a single orientation, except for the
physical properties, leading to a wide range rescence or isomerization is more favorable 3-F variant, which has two orientations (sup-
of colors, fluorescence quantum yields (FQYs), after excitation (7). plementary text S1 and fig. S2). Introduction
and photoswitching characteristics (3). The In earlier work on split green fluorescent of the substituent(s) causes little to no struc-
chromophore’s FQY increases by three orders protein (GFP), we linked structure and function tural deviation compared with the wild type
of magnitude when contained in the protein with energetics (8) and showed that the dom- (supplementary text S2 and figs. S1 to S3).
scaffold compared with when it is free in so- inant energetic feature governing the compe- The absorption spectrum for each Dronpa2
lution (4). The dominant nonradiative decay
process that lowers the chromophore’s FQY is
isomerization about either the phenolate (P) A O
or imidazolinone (I) bonds, resulting in a P-ring P-bond rotation I-bond rotation
flip or cis-trans isomerization, respectively P I N R1
(Fig. 1A) (5). This nonradiative decay process
P
I phenolate N cis-trans P I
O
is enhanced in photoswitching FPs that are ring flip R2 isomerization
widely used for super-resolution microscopy
(3). Modulating the probability between ra- B
diative and nonradiative decay, and for the 2
latter, the propensity for P- or I-bond isomer- Fig. 1. Model for chromophore isomerization
S1
ization, epitomizes the essential features of in FPs. (A) Rotation can occur about either
protein control. the P or I bond, leading to a P-ring flip or cis-trans
Potential Energy
The most well-studied and intuitively ap- isomerization, respectively. R1 and R2 represent
pealing hypothesis for the chromophore’s residues Gly64 and Cys62, respectively, which
substantial increase in FQY in the protein sug- covalently link the chromophore to the rest of the
0-0
gests that steric confinement of the protein FP (Fig. 2A). (B) General potential energy diagram
TE
1
scaffold physically prevents the bond rotation along the isomerization reaction coordinate for a
photoisomerizable chromophore. 0-0 TE represents
3
the TE between the lowest vibrational state of the S0
1
Department of Chemistry, Stanford University, Stanford, CA ground and excited electronic states. Three features
94305, USA. 2Stanford Synchrotron Radiation Lightsource, studied in this work are emphasized: fluorescence
Menlo Park, CA 94025, USA.
*These authors contributed equally to this work. (1, green), ES barrier crossing (2, purple), and
†Corresponding author. Email: sboxer@stanford.edu GS barrier crossing (3, yellow). Isomerization Reaction Coordinate

variant directly reflects the electronic con- A SH Y63

tribution of the substituent: Electron-donating O autocatalytic
H G64 cis A state O cis B state O
groups red-shift, whereas electron-withdrawing N chromophore deprotonation X
maturation X
N N
groups blue-shift the absorption maximum H
O
H
O N R1 N R1
(Fig. 2C and fig. S4). Both the trend of elec- C62 N N
HO protonation O
trostatic color tuning and the direction of OH
R2 R2
charge transfer upon excitation can be under- X
stood through either a Hammett analysis
B Substituents
Electron-donating S uents
(supplementary text S3) or Olsen’s resonance
color theory (13) (supplementary text S4 and WT 3-CH3 3-OCH3
3
6NQJ
J 6NQVV 6NQS
fig. S8), agreeing with simulations that sug-
gest negative charge flows from the P ring to H 3C H 3CO
the I ring upon photon absorption (5). We use
the absorption peak maximum [an approxi- O O O
mation of the 0-0 transition energy (TE)] wing Substi
Electron-withdrawing S
Substituents
(supplementary text S5) as a scale to reflect the 3-F 3
3-Cl 3-Br 3-I
substituents’ electron-donating and electron- 6NQK 6NQL 6NQN 6NQO
withdrawing capabilities (supplementary text
F Cl Br I
S4) because the initial Franck-Condon excita-
tion is purely an electronic process.
O O O
We then sought to examine the influence of O
each chromophore’s steric and electronic prop- 2,3-F2 2,3,5-F3 3-NO2
3,5-F2 F
erties on ES processes. First, we measured each 6NQP F F 6NQQ 6NQR
F
variant’s FQY (Fig. 1B, process 1) and plotted F O2N
the values against the corresponding TE
O O
(Fig. 3A). The trend is nonmonotonic with a O
O F F
peaked shape; variants with red- and blue-
shifted TEs show positive and negative cor- C
relations with FQY, respectively. A variant of
superfolder GFP with the same series of sub-
stituted chromophores exhibits the same trend Variant 3-OCH3 3-CH3 WT 3-I 3-Br 3-Cl 3-F 2,3-F2 3,5-F2 2,3,5-F3 3-NO2
(fig. S7), suggesting that electrostatic sensitivity Absorption 497.7 486.3 481.5 480.3 478.2 477.3 478.5 474.9 469.8 460.5 454.8
is an intrinsic chromophore property. Maximum
(nm) Electron-donating Electron-withdrawing
To elucidate the underlying energetics, we
estimated the ES energy barrier (Fig. 1B,
process 2) for each Dronpa2 variant using Fig. 2. Incorporation of electron-donating and electron-withdrawing substituents into the Dronpa2
temperature-dependent fluorescence lifetime chromophore. (A) Scheme depicting incorporation of substituents (represented by a green “X”) through
measurements (supplementary texts S6 and amber suppression of Y63 and chromophore maturation in Dronpa2 variants (C, Cys; Y, Tyr; G, Gly).
S7 and figs. S10 to S15), which capture the com- (B) Dronpa2 amber suppression variants grouped by electron-donating and electron-withdrawing properties.
bined decay rate of all relaxation processes from The electron density maps (2mFo – DFc, 1s) from solved x-ray structures (except 3,5-F2, which could not be
the S1 minimum. As with FQY, the ES barrier crystallized; see supplementary text S2) show substituent orientation(s) (see fig. S2 for omit maps).
heights show a peaked trend when plotted Two conformations were necessary for modeling the chromophore of the 3-F variant (fig. S2). The legend
against TE (Fig. 3B). Linear fits to the electron- of fig. S2 includes the identity of the monomer displayed for each variant. WT, wild type. (C) Image of purified
donating and electron-withdrawing variants’ proteins and their corresponding 77 K absorption peak maxima.
data exhibit slopes with similar magnitude
but opposite sign (Fig. 3D), which describe between the GS barrier and the substituent’s transfer extent in Fig. 3D), suggesting that
the extent and direction of charge transfer electronic effects. Close examination of fig. S19 changes in the electronic properties of the
during the ES barrier crossing. A change of reveals that the substituent’s steric properties chromophore have a smaller, but still evident,
1 kcal/mol in TE in either direction corre- may also contribute to the observed trends. impact on thermal relaxation. In contrast to
sponds to a change of ~1.5 kcal/mol in ES For example, among the data points for the the GS barrier, the influence of sterics on the
barrier height, implying that ES barrier cross- 3-F, 3-Cl, 3-Br, and 3-I substituents, as high- ES barrier is minimal (Fig. 3D and fig. S20). If
ing, a nonradiative process, is more sensitive lighted by the gray box in fig. S19, the barrier sterics were the dominant factor, large sub-
to electronic effects than Franck-Condon height increases as a function of halogen size stituents would be expected to increase the
excitation. despite similar TEs, indicating that substitu- barrier to chromophore twisting in the ES and,
To investigate the role of steric and elec- ent size influences GS barrier height. To iso- consequently, FQY. However, electrostatics is
tronic effects on GS isomerization barrier height late the electrostatic contribution to GS barrier clearly the dominant factor for ES isomeriza-
(Fig. 1B, process 3), we determined the isomer- height, we created an isosteric substituent se- tion in a constant protein environment (Fig. 3,
ization rate constant through pH-dependent ries (defined in supplementary text S9) and A and B, and fig. S20).
thermal relaxation kinetics measurements after plotted the corresponding data for this sub- The observation of two approximately equal
photoexcitation to a cis-trans photostationary group, which monotonically decrease as a but opposite slopes between ES barrier height
state, assuming the validity of transition state function of TE (Fig. 3C). The extent of charge and TE (Fig. 3B) suggests a mechanism change
theory (supplementary text S8 and figs. S17 to transfer during GS barrier crossing reflected for barrier crossing that depends on the elec-
S18). A plot of the GS barrier height versus TE by the slope in Fig. 3C is approximately one- tronic properties of the chromophore (supple-
(fig. S19) appears to show a lack of correlation third of that in the ES (Fig. 3B and charge- mentary text S3). Charge transfer is coupled

A Wavelength (nm) B Wavelength (nm) C Wavelength (nm) with bond twisting in the ES of monomethine
dyes (5, 14). In the GFP chromophore, this
ES Barrier (kcal/mol)
500 490 480 470 460 500 490 480 470 460
GS Barrier (kcal/mol)
500 490 480 470 460
0.6 twisting can proceed about either the P or I
1 9 2 21 3 bond (Fig. 1A). Both P and I twisting in the
isolated anionic chromophore have compa-
FQY
0.4
7 20
rable barrier heights but opposite charge-
0.2 transfer directions, suggesting that small
5 19
perturbations to the chromophore or its envi-
0 3 18 ronment could influence which twisting pathway
57 58 59 60 61 62 63 57 58 59 60 61 62 63 57 58 59 60 61 62 63
is more energetically favorable (14). Our data
TE (kcal/mol) TE (kcal/mol) TE (kcal/mol)
reveal a decrease in FQY and ES barrier height
Donating Withdrawing Donating Withdrawing Donating Withdrawing compared with those from wild-type Dronpa2
regardless of the substituent’s electronic effect
3-OCH3 3-CH3 WT 3-I 3-Br 3-Cl 3-F 2,3-F2 3,5-F2 2,3,5-F3 3-NO2 (Fig. 3, A and B), indicating that the electronic
properties of the chromophore can tune the
relative ES barrier heights for P and I twisting,
D Summary of ground and excited state properties of Dronpa2 amber suppression variants thus biasing toward the pathway with the
Photophysical Charge Transfer
Electron Flow
Sensitivity lower barrier. By contrast, the GS chromo-
Process Extent a to Sterics phore follows the same isomerization mech-
Excitation +1 (by definition) no anism regardless of the substituent’s electronic
ES Withdrawing -1.5 minimal effect (Fig. 3C).
Barrier The observed trends in energetics can
Crossing Donating +1.6 minimal
be linked to structural intuition through a
GS Barrier Crossing -0.6 yes
valence-bond model depicting the deproto-
a Refers
to the degree of charge transfer of a given process, determined by the slopes in B and C, nated chromophore as an allylic anion (Fig. 4,
compared to that of the excitation process
A and B) (5, 14, 15). In Fig. 4C, we present an
Fig. 3. GS and ES properties of Dronpa2 amber suppression variants. Circled numbers refer to processes energetic model for chromophore isomeriza-
depicted in Fig. 1B. (A) FQY versus TE and (B) ES energy barrier height versus TE both exhibit a peaked tion guided by the adiabatic states in Fig. 4B.
shape (see fig. S7 for comparable FQY results from GFP variants). Linear fits to the electron-donating and The electronic properties of the chromophore,
electron-withdrawing substituent data in (B) are shown as dashed lines with positive and negative slope, governed in this work by the substituent on
respectively. (C) Isosteric series of the GS energy barrier height plotted against TE, with a dashed line the P ring, determine the ES barrier height
representing a linear fit. (D) Summary of GS and ES properties of Dronpa2 variants. difference between P and I twisting. For GS
-
A B - C
- - - P I
S2 - P I
P-twist
P I P I P I I-twist
P I S1
Potential Energy
Potential Energy
Potential Energy
-
P I S1 P I
0-0
P I
- P I TE
I
S0
P
P I
P I S0 P I ring flip trans

P I P I P I cis
P I
P-twist Planar I-twist
P I P I P I
P-twist Planar I-twist
90° 0° 90° 90° 0° 90° 180° 0° 180°
P I P I P I
Diabatic States for Mixed Adiabatic States for
Electron-withdrawing Substituents Electron-withdrawing Substituents
Fig. 4. Allylic anion model of isomerization for a chromophore containing chromophore at different bond rotation geometries (15). Negative charge
an electron-withdrawing substituent. Shades of gray represent relative transfer (CT) occurs from I to P for the I-twist pathway and from P to I for
magnitude of negative charge localized to the methine bridge, P ring, or I ring, the P-twist pathway, which agrees with a Hammett analysis (supplementary
and dots represent unpaired electrons. The color scheme is equivalent to text S3) and simulations of the free chromophore (5, 14). (C) Potential
that in Fig. 1A. (A) Three diabatic states of the chromophore in planar, I-twisted, energy diagram for FP chromophore isomerization with two competing bond
and P-twisted geometries, with energetic penalties required for breaking rotation pathways inspired by the mixed adiabatic states in (B). The GS
double bonds for rotation. Mixing of the coupled states (highlighted in green) cis chromophore is excited from S0 to S1 and relaxes to an S1 local minimum
leads to the adiabatic states shown in (B). For variants with electron-withdrawing (relaxation coordinate not shown) (13). From the S1 minimum, the chromo-
substituents, the I-twist pathway is more energetically downhill, and thus phore rotates about either the P or the I bond, depending on the relative
preferred, compared with the P-twist pathway. Electron-donating substituents ES barrier heights of the competing processes. The diagram represents
would have the opposite energetic effect and favor the P-twist pathway (not Dronpa2 variants with electron-withdrawing substituents; variants with electron-
shown for clarity). Although the relative energy levels of this allylic anion model donating substituents would have an inverted barrier height ratio between the
are qualitative, they are consistent with high-level calculations on the free two competing twisting pathways.

relaxation from trans to cis, we are limited also introduced into the cell that is capable of charging the was supported by a Center for Molecular Analysis and Design
by what we can measure spectroscopically, non-native tRNA with the noncanonical amino acid. The graduate fellowship. C.-Y.L. was supported by a Kenneth and
tRNA then recognizes the amber stop codon on the mRNA Nina Tai Stanford Graduate Fellowship and the Taiwanese Ministry
namely the photochromic cis-trans isomer- within the translation machinery of the cell, allowing for the of Education. This work was supported, in part, by the NIH
ization. We can therefore observe only one incorporation of the noncanonical amino acid into the growing (grant GM118044 to S.G.B.). Part of this work was performed at
GS twisting mechanism (I twist), which ex- polypeptide chain. the Stanford Nano Shared Facilities (SNSF) and supported by the
10. A. Dumas, L. Lercher, C. D. Spicer, B. G. Davis, Chem. Sci. 6, National Science Foundation (award ECCS-1542152). Use of
plains the monotonic trend in GS barrier 50–69 (2015). the Stanford Synchrotron Radiation Lightsource, SLAC National
height as a function of TE (Fig. 3C). In the 11. S. R. Marder et al., Science 265, 632–635 (1994). Accelerator Laboratory, is supported by the U.S. Department of
potential energy diagram (Fig. 4C), the tran- 12. A. C. Stiel et al., Biochem. J. 402, 35–42 (2007). Energy (DOE), Office of Science, Office of Basic Energy Sciences
13. C.-Y. Lin, M. G. Romei, L. M. Oltrogge, I. I. Mathews, S. G. Boxer, (contract no. DE-AC02-76SF00515). The SSRL Structural
sition state for either the P- or I-twist path- J. Am. Chem. Soc. 141, 15250–15265 (2019). Molecular Biology Program is supported by the DOE Office of
way in S1 (ES) lies closer to planarity than the 14. S. Olsen, K. Lamothe, T. J. Martínez, J. Am. Chem. Soc. 132, Biological and Environmental Research and by the National
corresponding transition state in S0 (GS). As a 1192–1193 (2010). Institutes of Health (NIH), National Institute of General Medical
15. S. Olsen, R. H. McKenzie, J. Chem. Phys. 130, 184302 (2009). Sciences (NIGMS) (including P41GM103393). The contents
result, reaching the transition state on the GS
16. S. Schenkl, F. van Mourik, G. van der Zwan, S. Haacke, of this publication are solely the responsibility of the authors and
surface requires greater bond rotation, ex- M. Chergui, Science 309, 917–920 (2005). do not necessarily represent the official views of the NIGMS
plaining the enhanced steric sensitivity ob- 17. C. Ko, A. M. Virshup, T. J. Martínez, Chem. Phys. Lett. 460, or NIH. Author contributions: M.G.R. and C.-Y.L. designed the
served for GS isomerization. 272–277 (2008). experiments, expressed the proteins, performed the experiments,
18. L. S. Wolfe et al., Proc. Natl. Acad. Sci. U.S.A. 107, interpreted the results, and wrote the manuscript. I.I.M. assisted
By engineering chromophore variants using 16863–16868 (2010). with protein crystallization, collected x-ray diffraction data, and
amber suppression, we have systematically 19. S. Gozem, I. Schapiro, N. Ferré, M. Olivucci, Science 337, assisted with data refinement. S.G.B. discussed results and wrote
elucidated the role of electrostatics on chro- 1225–1228 (2012). the manuscript. Competing interests: The authors declare no
20. G. Bassolino et al., J. Am. Chem. Soc. 136, 2650–2658 (2014). competing interests. Data and materials availability: All x-ray
mophore color and isomerization in an FP 21. S. Gozem, H. L. Luk, I. Schapiro, M. Olivucci, Chem. Rev. 117, density maps and atomic models for Dronpa2 variants have been
environment. The electrostatic sensitivities of 13502–13565 (2017). deposited in the Protein Data Bank (wild type: 6NQJ; 3-F: 6NQK;
the chromophore stem from the intrinsic di- 22. D. Smyrnova, M. d. C. Marín, M. Olivucci, A. Ceulemans, 3-Cl: 6NQL; 3-Br: 6NQN; 3-I: 6NQO; 2,3-F2: 6NQP; 2,3,5-F3: 6NQQ;
J. Chem. Theory Comput. 14, 3163–3172 (2018). 3-NO2: 6NQR; 3-CH3: 6NQV; 3-OCH3: 6NQS). All other data are
rection of charge transfer during electronic 23. M. d. C. Marín et al., J. Am. Chem. Soc. 141, 262–271 (2019). presented in the main text or supplementary materials.
transitions and photoisomerizable bond rota- 24. C. Punwong, S. Hannongbua, T. J. Martínez, J. Phys. Chem. B
tions, which is ubiquitous in other photo- 123, 4850–4857 (2019).
SUPPLEMENTARY MATERIALS
isomerizable systems (8, 16–24). By tuning ACKN OWLED GMEN TS science.sciencemag.org/content/367/6473/76/suppl/DC1
the environment of the chromophore in these We dedicate this manuscript to the memory of Seth Olsen, whose
protein systems, with an emphasis on the Supplementary Texts S1 to S9
theoretical studies of the GFP chromophore motivated much of the
Figs. S1 to S27
often-overlooked electrostatic component, it analysis of this work. We thank S. Lynch of the Stanford NMR
Tables S1 to S11
Facility for assistance with NMR data collection and interpretation.
may be possible to finely control properties of R. A. Mehl of the Unnatural Protein Facility was instrumental in
interest, such as regioselective isomerization, providing an aminoacyl-tRNA synthetase for 3-methyltyrosine
because of distinctive charge redistributions incorporation. S. H. Schneider helped develop the MATLAB code
for statistical analysis of the fluorescence lifetime data. We thank 1 March 2019; resubmitted 4 June 2019
as different bonds are rotated. On the basis of
J. I. Brauman, T. J. Martínez, N. H. List, S. D. Fried, and Accepted 31 October 2019
our results in FPs, introducing hydrogen-bond– L. M. Oltrogge for discussions regarding this work. Funding: M.G.R. 10.1126/science.aax1898
donating residues around the P ring of the
chromophore would bias toward the I-twist
photoisomerization pathway (13). In the photo-
isomerizable retinal chromophore in rhodop- OPTICS
sins, theoretical studies have suggested that
different bond-specific photoisomerization On-chip integrated laser-driven particle accelerator
intermediates have different electronic dis-
tributions (21), allowing for similar targeted Neil V. Sapra1*, Ki Youl Yang1, Dries Vercruysse1, Kenneth J. Leedle1, Dylan S. Black1, R. Joel England2,
environmental modifications to bias bond Logan Su1, Rahul Trivedi1, Yu Miao1, Olav Solgaard1, Robert L. Byer1, Jelena Vučković 1
rotation pathways. As such, our conclusions
provide an initial, generalizable framework Particle accelerators represent an indispensable tool in science and industry. However, the size and cost
to incorporate electrostatic and steric effects of conventional radio-frequency accelerators limit the utility and reach of this technology. Dielectric
into the design of other photoisomerizable laser accelerators (DLAs) provide a compact and cost-effective solution to this problem by driving
systems to help develop improved variants accelerator nanostructures with visible or near-infrared pulsed lasers, resulting in a 104 reduction of
and new functionalities in optogenetics and scale. Current implementations of DLAs rely on free-space lasers directly incident on the accelerating
imaging (1). structures, limiting the scalability and integrability of this technology. We present an experimental
demonstration of a waveguide-integrated DLA that was designed using a photonic inverse-design
RE FE RENCES AND N OT ES approach. By comparing the measured electron energy spectra with particle-tracking simulations, we
1. C. P. O’Banion, D. S. Lawrence, ChemBioChem 19, 1201–1216 infer a maximum energy gain of 0.915 kilo–electron volts over 30 micrometers, corresponding to an
(2018).
2. R. Y. Tsien, Annu. Rev. Biochem. 67, 509–544 (1998).
acceleration gradient of 30.5 mega–electron volts per meter. On-chip acceleration provides the
3. D. Bourgeois, V. Adam, IUBMB Life 64, 482–491 (2012). possibility for a completely integrated mega–electron volt-scale DLA.
4. A. Acharya et al., Chem. Rev. 117, 758–795 (2017).
5. M. E. Martin, F. Negri, M. Olivucci, J. Am. Chem. Soc. 126,
D
5452–5464 (2004).
6. C. L. Walker et al., Curr. Opin. Chem. Biol. 27, 64–74 (2015). ielectric laser accelerators (DLAs) have that produce them. Together, these advantages
7. J. W. Park, Y. M. Rhee, J. Am. Chem. Soc. 138, 13619–13629 (2016). emerged as a promising alternative allow DLAs to make an impact in the develop-
8. C.-Y. Lin, J. Both, K. Do, S. G. Boxer, Proc. Natl. Acad.
to conventional radio-frequency accel- ment of applications requiring mega–electron
Sci. U.S.A. 114, E2146–E2155 (2017).
9. Amber suppression allows for the site-specific incorporation of erators because of the large damage volt energy beams of nanoampere currents, such
noncanonical amino acids into recombinant proteins. threshold of dielectric materials (1, 2);
The amino acid site of interest on the relevant gene is mutated the commercial availability of powerful, near- 1
to the amber stop codon, TAG. Non-native tRNA containing E. L. Ginzton Laboratory, Stanford University, Stanford, CA, USA.
an anticodon recognizing the amber stop codon is encoded on infrared femtosecond pulsed lasers; and the 2
SLAC National Accelerator Laboratory, Menlo Park, CA, USA.
a separate plasmid. A non-native aminoacyl tRNA synthetase is low-cost, high-yield nanofabrication processes *Corresponding author. Email: nvsapra@stanford.edu

as tabletop free-electron lasers, targeted cancer and exciting a waveguide mode that acts as We expressed the acceleration gradient, Gz(Ei),
therapies, and compact imaging sources (3–7). the source for the accelerator (Fig. 1A). the integrated field that the electron expe-
DLAs are designed by choosing an appro- To meet the phase-matching condition, the riences as it travels through one period of
priate pitch and depth of a periodic structure periodicity of the accelerating structure, L, is the accelerator, in the frequency domain (29).
such that the near fields are phase matched set by L = bl, where b = v/c is the ratio of the The second term, Gy(Ei), corresponds to the
to electrons of a specific velocity (8, 9). These velocities of the incident electrons to the speed deflecting transverse gradients, which we pe-
structures, together with focusing elements, of light and l is the center wavelength of the nalized. The fields were subjected to Maxwell’s
integrated electron sources, and microbunch- pump laser (23). To match experimental pa- equations and the permittivity of the device,
ing structures, form the building blocks to rameters, we designed for a center pump e(p), was parameterized by a vector of design
achieve mega–electron volt-scale energy gain wavelength of 2 mm and an input electron variables, p (30). This vector describes the
through cascaded stages of acceleration (10–13). velocity of v = 0.5c, resulting in an accelera- permittivity of the device during the first,
Previous demonstrations of DLAs have relied tor period of L = 1 mm. Fig. 1B captures the continuous stage of optimization and a level-
on free-space lasers directly incident on the geometry of the optimization problem. Using set function that defines the boundaries of
accelerating structure, often pillars or gratings an in-house inverse-design software suite the device in the second, discrete stage of op-
made of fused silica or silicon (14–20). How- (24–27), the design of the accelerator was timization. To have good spectral overlap with
ever, free-space excitation requires bulky optics; optimized over a 3-mm region, ensuring the the broadband input pulsed laser spectrum,
therefore, integration with photonic circuits preservation of a 250-nm center channel for each objective function evaluation is the sum
would enable increased scalability, robustness, electron propagation. The accelerator was of m = 3 simulations, each with a different in-
and impact of this technology. simulated with a fully-3D finite-difference put source frequency, wi. The three simulations
Integration with photonic waveguides rep- frequency-domain (FDFD) solver meshed with uniformly sample a 30-nm total bandwidth
resents a design challenge because of dif- a uniform grid spacing of 30 nm. Periodic around 2 mm. During the final optimization
ficulties in accounting for scattering and boundary conditions were applied in the direc- stage, an additional constraint was introduced
reflections of the waveguide mode from sub- tion of electron propagation (z-axis) to enforce to enforce a minimum fabricable feature size of
wavelength features. Although tuning the geo- the accelerator period, and perfectly matched 80 nm. Further details regarding the design of
metric parameters and location of a few etched layers were used in the remaining axes (28). the accelerator and a time-lapse movie of the
holes in the waveguide is possible (21), this The structure was excited with the fundamen- optimization can be found in the supplementary
requires brute-force optimization of only a tal slab waveguide mode and the following 3D materials (31). A scanning electron microscope
small subset of the design space. Instead, we optimization problem was solved: (SEM) image of a fabricated optimized acce-
used an inverse-design approach to develop X
m lerator is shown with a frame from simu-
a waveguide-integrated DLA on a 500-nm maximize jGz ðE i Þj jGy ðE i Þj lated time-domain fields overlaid (Fig. 1C).
p;E 1 ;E 2 ;…;E m
device layer silicon-on-insulator (SOI) plat- i¼1 As the optimization was performed with pe-
form, which allows for expansion of the de- 1 riodic boundary conditions, the performance
subject to ∇ ∇ E i w2i DðpÞE i ¼ iwi J i ;
sign space (22). This on-chip accelerator is m0 of a finite-length 30-period accelerator struc-
demonstrated by coupling light from a pulsed i ¼ 1; 2; …; m ture was verified in a 3D finite-difference time-
laser through a broadband grating coupler ð1Þ domain (FDTD) simulation (32). The frequency
Fig. 1. Inverse design of on-chip particle accelerator.

(A) Schematic (not to scale) depicting components
of the on-chip accelerator. An inverse-designed
grating couples light from a normally incident
free-space beam into the fundamental mode of
a slab waveguide (inset 1). The excited waveguide
mode then acts as the excitation source for the
accelerating structure. The accelerator structure,
also created through inverse design, produces
near fields that are phase matched to an input
electron beam with initial energy of 83.4 keV. Inset 2
depicts the phase-matched fields and electron
at half an optical cycle (t/2) apart. (B) Geometry
of the optimization problem. We designed on a 500-nm
silicon (gray), 3-mm buried oxide layer (light-blue)
SOI material stack. Periodic boundary conditions
(green) are applied in the z-direction, with a period
of L = 1 mm, and perfectly matched layers were used
in the remaining directions (orange). We optimized
the device over a 3-mm design region (yellow) with
an input source of the fundamental TE0 mode. During
the optimization, a 250-nm channel for the electron
beam to travel in is maintained. (C) SEM image of
the final accelerator design obtained from the
inverse-design method. A frame from a time-domain
simulation of the accelerating fields, Ez, is overlaid.

response of the grating coupler and accelera- where t0 is the delay between the time of pler to excite the fundamental TE0 waveguide
tor were computed and cascaded to determine source injection and the electron entering mode of the slab waveguide [for grating cou-
the frequency-domain acceleration gradient the accelerator channel and L = 30 mm is the pler design, see the materials and methods (31)].
(31). This complete acceleration gradient spec- length of the accelerator. The accelerating A custom-built scanning transmission elec-
trum is shown in Fig. 2. The spectrum peaks at and deflecting gradients down the center of tron microscope was used as the source for
l = 1.964 mm (Fig. 2A), indicating a shift from the channel are evaluated at t0, which maxi- the electron beam, which travels through the
the design parameters caused by the finite mizes the acceleration gradient, Gz(t0) (Fig. channel in the accelerator structure with an
length of the structure and numerical dis- 2B). As the time-domain gradients are nor- initial energy of 83.4 keV (v = 0.51c). Electrons
persion (33). With knowledge of the peak malized to the peak incident pulse amplitude, that passed through the accelerator were sep-
operating wavelength, the time-domain char- Fig. 2B also provides the simulated structure arated by energy in a magnetic spectrometer
acteristics were modeled by propagating a factors, the ratio of acceleration gradient to before terminating at a microchannel plate
300-fs unchirped Gaussian pulse, centered at incident field. Although we obtained good detector to image the energy distribution [see
1.964 mm, through the grating coupler, wave- suppression of the deflecting gradients, one the materials and methods section for addi-
guide, and accelerator (31). The time-domain can also operate at another time delay, t0′, tional details (31)].
acceleration gradient (Gz) and deflecting gra- such that the deflecting gradients, Gy(t0′) and The electron energy spectra (Fig. 4A) showed
dients (Gy, Gx) are given by: Gx(t0′), are further minimized in the center of that electrons had been successfully accel-
the channel. erated by our structure. The blue curve depicts
1 L
Gk ðt0 Þ ¼ ∫ Ek ðz; t0 þ z=bc0 Þdz
L 0
ð2Þ A 30-period accelerator, waveguides, and the energy spectrum of the electrons passing
grating couplers were fabricated on a 500-nm- through the accelerator structure with the
thick SOI wafer using electron beam lithogra- laser off, and the red curve shows the energy
phy and reactive ion etching. The input grating spectrum when the laser (3 mW average power,
coupler was separated by 50 mm of wave- 335 MV/m peak field, at 1.94 mm) was inci-
guide from the accelerator structure, and the dent on the grating coupler. Because the bunch
output coupler was separated by 30 mm of length was larger than the optical cycle, we
waveguide from the accelerator. The entire observed symmetric broadening of the energy
structure had a width of 30 mm. To provide spectrum, resulting in electrons being accel-
clearance for the electron beam, the area sur- erated and decelerated. To characterize the
rounding the accelerator was etched with an broadening of the laser-on spectra, we intro-
additional photolithography step to form a duced an energy spectrum width metric, x,
“mesa” (Fig. 3). Complete fabrication details which we define as the first trailing energy
can be found in the materials and methods at which the difference between the laser-on
section (31). spectra F(e) and the laser-off spectra f(e) was
The experimental setup was adapted from <0.01: F(x) – f(x) ≤ 0.01. For the spectra shown
previous direct-incidence pillar experiments to in Fig. 4A, centered at e0 = 83.4 keV, this cor-
support normal incidence on a grating coupler responds to a value of x = 84.31 keV. The dotted
(13, 18). Light, polarized in the direction of red curve depicts simulated performance of the
electron propagation (z), generated from a accelerator based on a commercial particle-
300-fs FWHM pulse-length, 100-kHz repeti- tracking code to propagate a distribution of
tion rate optical parametric amplifier was fo- particles consistent with experimental param-
cused to a 40-mm, 1/e2-diameter spot. The beam eters through the 3D electromagnetic field
is normally incident on the input grating cou- map of Fig. 1C, providing agreement with the
Fig. 3. Fabricated
single-stage accelerator.
SEM image of a single-
stage accelerator of
30 periods fabricated on a
Fig. 2. Simulated performance of optimized 500-nm SOI stack.
accelerators. (A) Acceleration gradient The accelerator sits
spectrum for a finite-length accelerator on a 25-mm-tall mesa
composed of 30 periods, including frequency structure to provide
response of grating coupler. The gradient is clearance for the input
normalized to the maximum frequency-domain electron beam.
amplitude of the incident Gaussian beam. Dashed
line indicates optimal operation wavelength
of the simulated structure, l = 1.964 mm.
(B) Time-domain accelerating gradients and
transverse deflecting gradients as a function of
input electron energy from simulated fields.
Gradients were evaluated at time-delay, t0, which
maximizes the acceleration gradient, Gz. Fields
normalized to the peak electric field of the pulse
incident on the grating coupler.

Fig. 4. Experimental verification of accelerator.

(A) Electron energy spectrum (log-scale)
without laser incident (blue curve) and with
laser incident (3.0 mW, 335 MV/m peak field, at
l = 1.94 mm; red curve) on the grating coupler.
Simulated spectrum is based on particle-tracking
simulations shown in the dotted red curve.
On the spectra, e0 denotes the center energy
of the distribution and x provides an energy
spectral width metric that marks the energy at
which the difference between the laser-on and
laser-off spectra is below 0.01. (B) Energy
spectral width broadening, Dx = x – e0, (blue,
left axis), and peak depletion (green, right axis)
for a fixed power at 2.75 mW, 321 MV/m peak
field, as a function of varying the wavelength
of the pump laser. (C) Measured energy
spectral width, Dx, at a fixed wavelength of
1.94 mm as a function of input power, with
simulation from the tracking code superimposed
as a dashed curve.
experimental spectrum (31). Because of the the power, the measured values of the spectral gies capitalizing on the inherent scalability of
spread in energy and phase of the input elec- width, Dx, compared favorably with those photonic circuits.
tron spectrum, the maximal energy gain is a obtained from the simulated particle-tracking
quantity not directly measurable from the spectra indicated by the dashed curve in REFERENCES AND NOTES
laser-on spectrum. Instead, we could obtain Fig. 4C (31). 1. B. C. Stuart, M. D. Feit, A. M. Rubenchik, B. W. Shore,
this value from the particle-tracking simula- Although nonlinear dephasing has been M. D. Perry, Phys. Rev. Lett. 74, 2248–2251 (1995).
tions (31). From these simulations, we inferred observed in other DLA experiments (34, 35), 2. A.-C. Tien, S. Backus, H. Kapteyn, M. Murnane, G. Mourou,
Phys. Rev. Lett. 82, 3883–3886 (1999).
a maximal energy gain of 0.915 keV over 30 mm, the short waveguide distances (50 mm) in this 3. P. G. O’Shea, H. P. Freund, Science 292, 1853–1858
providing a gradient of 30.5 MeV/m and a experiment were much smaller than the hun- (2001).
structure factor (ratio of acceleration gradient dreds of micrometers of propagation distance 4. R. W. Hamm, M. E. Hamm, Industrial Accelerators and Their
Applications (World Scientific, 2012).
to incident field) of 0.09. required to introduce nonlinear dephasing
5. S. Hanna, RF Linear Accelerators for Medical and Industrial
To determine the operating wavelength of (36). Additionally, coupling into higher-order Applications (Artech House, 2012).
our accelerator, the average power of the inci- modes of the slab waveguide, specifically the 6. K. Wootton, J. McNeur, K. Leedle, Reviews of Accelerator
dent laser pulses on the grating coupler were TE2 waveguide mode, can result in dephasing. Science and Technology: Volume 9: Technology and
Applications of Advanced Accelerator Concepts
fixed to be 2.75 mW (321 MV/m peak field) and However, with 95% of the total power in the (World Scientific, 2016), pp. 105–126.
the wavelength was swept (Fig. 4B). A peak TE0 mode and only 3.8% of power expected 7. T. Plettner, R. Byer, Phys. Rev. Spec. Top. Accel. Beams 11,
was observed in the energy spectrum width, to couple into TE2, this negative contribu- 030704 (2008).
Dx = x – e0, at 1.94 mm. Moreover, the ratio tion would be minimal (31). Although not 8. R. J. England et al., Rev. Mod. Phys. 86, 1337–1389
(2014).
of laser-on to laser-off counts at the center catastrophic to operation of the accelera- 9. T. Plettner, P. Lu, R. Byer, Phys. Rev. Spec. Top. Accel. Beams
energy, referred to as “peak depletion,” was tor, postexperiment SEM imaging revealed 9, 111301 (2006).
optimal at 1.94 mm, consistent with an increase laser-induced damage at the input grating 10. J. McNeur et al., Optica 5, 687 (2018).
11. U. Niedermayer, T. Egenolf, O. Boine-Frankenheim,
in the number of modulated electrons at this coupler. Additional characterization iden-
P. Hommelhoff, Phys. Rev. Lett. 121, 214801 (2018).
wavelength. The greatest broadening of the tified this damage to occur after 3 to 4 mW of 12. B. Naranjo, A. Valloni, S. Putterman, J. B. Rosenzweig,
energy spectra and dip in peak depletion sug- input power. System-level analysis of an SOI Phys. Rev. Lett. 109, 164803 (2012).
gested an operating wavelength of 1.94 mm. integrated accelerator such as the one pre- 13. D. S. Black et al., Phys. Rev. Lett. 122, 104801 (2019).
14. E. A. Peralta et al., Nature 503, 91–94 (2013).
This wavelength was blue-shifted from the sim- sented here predicts acceleration gradients
15. J. Breuer, P. Hommelhoff, Phys. Rev. Lett. 111, 134803
ulated operating value of 1.964 mm. Because of of 45.3 MeV/m (36). This suggests that work (2013).
the cavity-like nature of this accelerator, we toward higher-efficiency grating couplers 16. K. J. Leedle et al., Opt. Lett. 40, 4344–4347 (2015).
attribute this spectral shift and flattening of with greater resiliency to high-field hotspots 17. K. P. Wootton et al., Opt. Lett. 41, 2696–2699 (2016).
18. K. J. Leedle et al., Opt. Lett. 43, 2181–2184 (2018).
the gradient spectrum to fabrication imper- provides an achievable path to this estimated 19. D. Cesar et al., Opt. Express 26, 29216–29224 (2018).
fections. Additionally, as a consequence of value. 20. Z. Chen, K. Koyama, M. Uesaka, M. Yoshida, R. Zhang,
the blue shift, the L = bl design condition The fabricated devices accelerate electrons of Appl. Phys. Lett. 112, 034102 (2018).
21. C.-M. Chang, O. Solgaard, Appl. Phys. Lett. 104, 184102
was no longer satisfied exactly and so some an initial energy of 83.4 keV by an inferred maxi-
(2014).
dephasing was to be expected, contributing mum energy gain of 0.915 keV over 30 mm, dem- 22. S. Molesky et al., Nat. Photonics 12, 659–670 (2018).
to the diminished structure factor. Fixing the onstrating acceleration gradients of 30.5 MeV/m. 23. J. Breuer, R. Graf, A. Apolonski, P. Hommelhoff, Phys. Rev.
wavelength to 1.94 mm (Fig. 4C), we conducted In this integrated form, these devices, alongside Spec. Top. Accel. Beams 17, 021301 (2014).
24. J. Vuckovic et al., SPINS-B: Inverse Design Software for
a sweep of the input power from 0.5 to 5 mW focusing and bunching elements, can be cas- Nanophotonic Structures, Stanford OTL Reference S18-012
(137 to 433 MV/m peak fields). As we increased caded to reach mega–electron volt-scale ener- (2018); https://github.com/stanfordnqp/spins-b.

25. A. Y. Piggott et al., Nat. Photonics 9, 374–377 (2015). ACKN OWLED GMEN TS the particle-tracking simulations. L.S. provided the grating coupler
26. A. Y. Piggott, J. Petykiewicz, L. Su, J. Vučković, Sci. Rep. 7, Funding: We thank all the members of the Accelerator on a Chip design code. R.T. assisted in simulation analysis. J.V., R.L.B.,
1786 (2017). International Program for discussion and collaboration. This and O.S. organized the collaboration and supervised the experiments.
27. L. Su, A. Y. Piggott, N. V. Sapra, J. Petykiewicz, J. Vuckovic, work was supported by the Gordon and Betty Moore Foundation All authors participated in the discussion and interpretation of
ACS Photonics 5, 301–305 (2017). (grant no. GBMF4744) and the U.S. Department of Energy, the results. Competing interests: The authors declare no competing
28. W. Shin, S. Fan, J. Comput. Phys. 231, 3406–3431 interests. Data and materials availability: All data needed to
Office of Science (grant nos. DE-AC02-76SF00515 and
(2012). evaluate the conclusions in the paper are available in the main text
DE-SC0009914). K.Y.Y. acknowledges funding from a Nano- and
29. T. Hughes, G. Veronis, K. P. Wootton, R. Joel England, S. Fan, or the supplementary materials.
Quantum Science and Engineering Postdoctoral Fellowship.
Opt. Express 25, 15414–15427 (2017).
D.V. acknowledges funding from FWO and the European Union
30. D. Vercruysse, N. V. Sapra, L. Su, R. Trivedi, J. Vučković,
Horizon 2020 Research and Innovation Program (under SUPPLEMENTARY MATERIALS
Sci. Rep. 9, 8999 (2019).
31. See the supplementary materials for additional information. Marie Sklodowska-Curie grant no. 665501). R.T. acknowledges a
32. FDTD: 3D Electromagnetic Simulator (Lumerical Inc., 2019); Kailath Graduate Fellowship. Part of this work was performed at
https://www.lumerical.com/products/. the Stanford Nano Shared Facilities (SNSF)/Stanford
Supplementary Text
33. A. Taflove, S. C. Hagness, Computational Electrodynamics: Nanofabrication Facility (SNF), which is supported by the
Figs. S1 to S4
The Finite-Difference Time-Domain Method (Artech House, National Science Foundation under award no. ECCS-1542152.
2005). Author contributions: N.V.S. performed and led the design,
Movies S1 and S2
34. D. Cesar et al., Commun. Phys. 1, 46 (2018). simulation, and fabrication of the accelerator. K.Y.Y. and Y.M. assisted
35. S. Tan et al., Opt. Lett. 44, 335–338 (2019). with fabrication. D.V. assisted with design. K.J.L. and D.S.B. 5 July 2019; accepted 26 November 2019
36. T. W. Hughes et al., Phys. Rev. Appl. 9, 054017 (2018). conducted the electron acceleration experiment. R.J.E. performed 10.1126/science.aay5734
NEUROSCIENCE shown previously in rodent L2/3 pyramidal

neurons (4). However, these high-frequency
Dendritic action potentials and computation in signals were substantially attenuated at distal
tuft dendrites (fig. S2). Furthermore, Ca2+ influx
human layer 2/3 cortical neurons in spines was similar to that in the nearby
dendritic branches, regardless of the somatic
Albert Gidon1, Timothy Adam Zolnik1, Pawel Fidzinski2,3, Felix Bolduan4, Athanasia Papoutsi5, AP frequency (fig. S2D).
Panayiota Poirazi5, Martin Holtkamp2, Imre Vida3,4, Matthew Evan Larkum1,3* We next examined whether human L2/3
dendrites have intrinsic mechanisms to com-
The active electrical properties of dendrites shape neuronal input and output and are fundamental to pensate for the large dendritic attenuation.
brain function. However, our knowledge of active dendrites has been almost entirely acquired from We injected a current step into the dendrite
studies of rodents. In this work, we investigated the dendrites of layer 2 and 3 (L2/3) pyramidal neurons (Idend) and recorded the membrane poten-
of the human cerebral cortex ex vivo. In these neurons, we discovered a class of calcium-mediated tials at both the dendrite and at the soma. At
dendritic action potentials (dCaAPs) whose waveform and effects on neuronal output have not been the soma and at the proximal dendritic sites
previously described. In contrast to typical all-or-none action potentials, dCaAPs were graded; their (170 mm from the soma, on average), a supra-
amplitudes were maximal for threshold-level stimuli but dampened for stronger stimuli. These dCaAPs threshold current readily evoked somatic APs,
enabled the dendrites of individual human neocortical pyramidal neurons to classify linearly nonseparable which back-propagated into the dendrite (Fig. 1,
inputs—a computation conventionally thought to require multilayered networks. A and B, and fig. S5H). However, when the
dendritic electrode was positioned more dis-
T
tally, suprathreshold stimuli often evoked trains
he expansion of the human brain during drites in acute slices from surgically resected of repetitive APs that were initiated exclusively
evolution led to an extraordinarily thick brain tissue of the human neocortex from epi- in the dendrite (Fig. 1D; for transient stimulus,
cortex (~3 mm), which is disproportion- lepsy and tumor patients. Subthreshold (steady- see fig. S10). These results imply that L2/3
ately thickened in layers 2 and 3 (L2/3) state) potentials attenuated from the dendrite dendrites in human cortical pyramidal neu-
(1). Consequently, human cortical neu- to the soma with a length constant (lsteady) of rons are distinctly more excitable than the
rons of L2/3 constitute large and elaborate 195 mm (fig. S1; n = 23 cells). In the opposite homolog dendrites in rodents, where similar
dendritic trees (2, 3), decorated by numer- direction, the back-propagating action poten- steady currents evoke, at most, only a single
ous synaptic inputs (1). The active electrical tials (bAPs) attenuated from the soma to the dendritic AP at the beginning of the voltage
properties of these dendrites largely deter- dendrite with a lbAP of 290 mm (Fig. 1, A to C; response (5). In contrast to L2/3 pyramidal
mine the repertoire of transformations of the n = 31 cells). Both lbAP and lsteady were shorter neurons, layer 5 pyramidal neurons of the
synaptic inputs to axonal action potentials than the length of the apical dendrite (the human neocortex were recently reported to
(APs) at the output. Thus, they constitute a key somata of these cells were located ~850 mm have reduced dendritic excitability compared
element of the neuron’s computational power. below the pia mater, on average, and the apical with their homolog neurons in rodents (6).
We used dual somatodendritic patch clamp dendrite extended up to layer 1), which implies High-frequency dendritic APs (>200 Hz)
recordings and two-photon imaging to directly that strong attenuation governs the electrical that were uncoupled from somatic firing have
investigate the active properties of L2/3 den- activity to and from most synapses located on been observed in rodent dendrites in vivo
the apical dendrite. (7, 8). The authors of these studies have at-
We filled cells with the calcium indicator tributed these spikes to dendritic voltage-gated
1
Institute for Biology, Humboldt-Universität zu Berlin, Berlin, Oregon-green BAPTA-1 (100 mM) and mea- Na+ channels and/or N-methyl-D-aspartate
Germany. 2Epilepsy-Center Berlin-Brandenburg, Department sured the change in fluorescence (DF/F) under (NMDA) receptors. The dendritic APs in hu-
of Neurology, Charité - Universitätsmedizin Berlin, Berlin, a two-photon microscope while triggering APs man L2/3 neurons were not blocked by the
Germany. 3NeuroCure Cluster, Charité - Universitätsmedizin
Berlin, Berlin, Germany. 4Institute of Integrative at the soma. Trains of somatic APs (10 APs) at sodium channel blocker tetrodotoxin (1 mM;
Neuroanatomy, Charité-Universitätsmedizin Berlin, Berlin, 50 Hz failed to cause Ca2+ influx at distal apical n = 4 cells; fig. S3), but they were abolished
Germany. 5Institute of Molecular Biology and Biotechnology, dendrites (fig. S2). AP trains with a higher fre- by the Ca2+ channel blocker Cd2+ (200 mM;
Foundation for Research and Technology - Hellas (IMBB-
FORTH), Crete, Greece. quency (10 APs at 200 Hz) did invade most of n = 5 cells; fig. S3). The dendritic Ca2+ APs
*Corresponding author. Email: matthew.larkum@hu-berlin.de the apical dendrite, similarly to what has been that we observed in human L2/3 neurons have

A Bi ii iii C
bAP delay (ms)

3
bAP amp. (mV)

100
pia Vdend 0
40 mV 50 0 600
L1 dist. (µm)
2 nA
L2/3 200 ms
bAP 0
200 µm
Vsoma 0 300 600 900 1200
312 µm
Isoma distance from soma (µm)
soma 10 mV 20 mV
10 ms
Di ii iii 10 ms E
dCaAP amp. (mV)

L4 100
Vdend
50
Idend dCaAP
40 mV
2 nA 0
200 ms 200 350 500
Vsoma
distance from soma (µ m)
F coupled simple dCaAP G coupled complex dCaAP H uncoupled simple dCaAPs

dCaAP
20 mV
dCaAP bAP
20 mV
bAP 200 ms
Vdend
Vdend
200 ms Vdend
500 pA
40 mV
Idend Vdend
20 mV
Vsoma
20 mV
200 ms
20 mV
50 ms
Vdend
Vsoma 10 ms Vdend
50 ms
Vsoma Vsoma
Fig. 1. bAPs and dCaAPs in human dendrites of L2/3 neurons. blue) triggered by a square current injected at the dendrite (Idend) and the
(A) Experimental setting: L2/3 neuron at a depth of 1130 mm below resulting somatic AP (black) from the cell in (A). (ii) Somatic AP (in black)
the pial surface, with a somatic electrode and a dendritic electrode placed and a dCaAP (in blue) magnified from (i). The slow rising dCaAP (blue arrow)
at 312 mm from the soma. (B) (i) Recordings from the cell in (A). bAPs in precedes the somatic AP. (iii) Initial dCaAP in each recording at threshold
green (Vdend) and corresponding somatic APs in black (Vsoma) triggered in the same 16 dendrites (gray) in (iii) of (B) and their average trace (blue)
by somatic current injection (Isoma) are shown. (ii) Somatic AP preceded the aligned to their peak. (E) dCaAP amplitude is independent of the distance from
bAP [magnified from the frame in (i)]. (iii) bAPs in 16 dendrites (gray) and the soma (n = 28 cells). Linear fit is shown with the dashed line (r2 = 0.0009).
their average (green) aligned to their peak. (C) bAP amplitude (green dots) (F) (Left) Coupled and simple dCaAPs (blue trace) and somatic APs (purple
and exponential fit (length constant lbAP = 290 mm; n = 31 cells; dashed line) trace) triggered by Idend. (Right) Magnified dCaAP (in blue) and a somatic AP
against distance from the soma. Gray area indicates the putative tuft region (in purple) framed in the traces on the left are shown. (G) (Top) Two coupled and
in layer 1 for the longest dendrite. (Inset) Delay of the bAP peak against complex dCaAPs (in red) triggered delayed somatic APs [in purple, magnified
distance from the soma with linear fit (r2 = 0.78, where r2 is the coefficient at (bottom)]. (H) Burst of simple and uncoupled dCaAPs in blue (top) with
of determination). amp., amplitude; dist., distance. (D) (i) dCaAPs (Vdend; somatic APs (bottom).
not been described in the cortical neurons of also in neurons from other neocortical areas of In the remaining 12 dendrites, a single dCaAP
other mammalian species. Dendritic APs that tumor patients (n = 4 cells from 3 patients; was triggered immediately after the beginning
are mediated (or are assumed to be mediated) fig. S4). This suggests that dCaAPs are neither of the stimulus. Unlike the bAP (Fig. 1C), the
by sodium currents in rodents’ neurons have regionally confined nor related to pathology. amplitude of the dCaAPs (Fig. 1E) and their
been variously named dendritic spikes (9), The waveform of dCaAPs was stereotypical upstroke (fig. S5) were not dependent on the
prepotentials (10), Na-dSpikes (11), and den- and easily distinguished from that of bAPs. distance from the soma (average dCaAP am-
dritic action potentials (DAPs) (8). To distin- dCaAPs were typically wider than bAPs (with plitude 43.8 ± 13.8 mV, ranging between 13.0
guish the dendritic APs that we found in the widths of 4.4 ± 1.4 ms, ranging between 2.6 and and 67.0 mV; n = 32 cells, measured at thresh-
human dendrites from those described pre- 8.0 ms; n = 32 cells), they were slow rising, and old). This is consistent with both variability
viously, we refer to them as dendritic Ca2+ they did not have a kink at onset (7) (Fig. 1D). of the dCaAP initiation site and variability
APs (dCaAPs). The majority of the cells (27 of 39) showed a of dCaAP properties (for further details, see
dCaAPs were present not only in neurons train of (two or more) dCaAPs with a mean figs. S5 and S11). We never detected high-
from the temporal lobe of epilepsy patients but firing rate of 4.6 ± 1.7 Hz (dCaAPs per second). amplitude, long-duration, Ca2+ mediated plateau

A Bi ii iii Vdend iv
Idend Vsoma
305 pA
60 pA
245 pA
417 µm
200 µm
100 pA 30 mV 30 mV
250 ms 250 ms 40 mV
200 ms 10 ms
Ci I soma
ii Vsoma iii Vdend iv
soma
150 pA
350 pA
200 pA
Di ii E dCaAP rheobase F
dCaAP amp. (norm.)
dCaAP amp. (mV)

increasing stim. intensity 60
200 pA
1
200 ms
40
amp.
20 mV
5 ms 20
-30 mV
dCaAPs 0 0
0.5 1 1.5 2 200 400 600
Idend / Irhe Idend (pA)
Gi ii H AP rheobase I
increasing stim. intensity 1
200 pA AP amp. (norm.)
sharply tuned dCaAP
200 ms
amp.
0.5 (I / Irhe)
-35 mV 0 threshold for somatic AP

20 mV 0.5 1 1.5 2
2 ms AP Isoma / Irhe
Fig. 2. dCaAPs are sharply tuned to the stimulus intensity. (A) L2/3 the dCaAPs’ amplitude (ii); vertical tick on each trace marks 50 ms after Idend
pyramidal neuron with soma 886 mm below the pia. The somatic and onset. stim., stimulation; norm., normalized. (E) Amplitude of the first dCaAP
dendritic electrodes are shown in black and blue, respectively. Recordings in each trace against Idend normalized by rheobase (Irhe) for uncoupled dCaAPs
from this cell are shown in (B) and (C). (B) Dendritic current (Idend) injected (12 dendrites) and exponential fit (dashed line), with a decay constant (tdCaAP)
417 mm from the soma (i) and corresponding somatic (ii) and dendritic traces (iii). of 0.39 (median 0.38) in units of rheobase. (F) dCaAP amplitudes as in (E) but
(ii) Idend of 260 and 275 pA, but neither smaller nor larger current, resulted not normalized by Irhe. Dots in different colors represent dCaAP amplitudes
in somatic APs. (iii) dCaAP amplitudes were maximal for Idend of 260 and from different cell (12 dendrites) with exponential fit (dashed lines). (G) As in (D)
275 pA, whereas Idend > 275 pA dampened them. (iv) dCaAP (in blue) precedes but for somatic APs. Isoma, (i) and the resulting somatic APs (ii). (H) AP
the somatic AP (in gray); traces are magnified from the framed APs in amplitude plotted against the normalized somatic input current strength
(ii) and (iii). (C) Somatic current injection, Isoma (i), somatic AP trains (ii), (Isoma/Irhe). The amplitude of the somatic AP was fixed and did not depend
and bAP (iii) for similar ranges of current intensity as those shown in (i) of (B). on Isoma for a range of stimuli strengths as in (ii) of (G) (exponential fit with
(iv) Somatic AP (in gray) precedes the dendritic bAP (in green); traces are tAP = 82, units of somatic Irhe). (I) Dendritic and somatic activation functions for
magnified from the framed APs in (ii) and (iii). (D) Increase in Idend (i) dampened dCaAPs (blue curve) and for somatic APs (black curve).
potentials, which are common in the apical in 11 out of 17 coupled cells; Fig. 1, F and G, and Typically, uncoupled dCaAPs were observed
dendrites of L5 neurons in rodents. fig. S6). Coupled dCaAPs that triggered somatic in more distal dendritic recording sites (335 ±
The impact of dCaAPs on the soma was var- APs with a delay were classified as complex. 113 mm from the soma) than the coupled
iable. In some of the cells (17 of 37), the dCaAPs Without exception, each coupled dCaAP trig- dCaAPs that triggered somatic APs (265 ±
were coupled with somatic APs (coupled dCaAPs; gered a single somatic AP, which implies that, 71 mm from the soma), but the distance differ-
e.g., Fig. 1F). Unlike forward-propagating den- unlike calcium APs in the dendrites of other ence was not statistically significant (Wilcoxon
dritic APs in other pyramidal neurons (12–14), neurons (15, 16), dCaAPs did not induce bursts rank sum test, P = 0.077). Additionally, one
coupled dCaAPs triggered somatic APs imme- of somatic APs. In the other 20 cells, dCaAPs coupled and three uncoupled cells fired bursts
diately and/or with a delay ranging between were uncoupled. They were confined to the of three or more dCaAPs at the beginning
21.6 and 116.9 ms (53.8 ± 26.8 ms, on average, apical dendrite, unable to evoke somatic APs. of the stimulus (28 to 73 Hz). In fig. S7, we

A B pathway X C pathway Y
0 mV
200 µm
10 mV
40 ms
20 mV
290 pA 1 sec
220 pA D pathways X + Y E pathways X + Y + inhib.
background ex. syns.

ex. syns. pathway X
ex. syns. pathway Y
inhib. syns.
Fig. 3. Anti-coincidence in L2/3 of the human cortex. (A) (Left) L2/3 F G apical tuft
neuron modeled with passive membrane and dCaAP mechanism at the AND
ND
apical dendrite, demarcated by the blue circle (550 mm from the soma). 1 O + A
100 background excitatory synapses (ex. syns.) (AMPA) indicated by gray dots Y
AN
0 O +
AN
were randomly distributed over the entire dendritic tree and were activated in
D
dCaAP
amp.
simulations (B) to (E). Pathways X and Y with 25 excitatory synapses each (red
apical
and blue dots) modeled by AMPA and NMDA conductances (33) targeted a XOR
0 1
subregion of the apical dendrite in addition to 20 GABAergic inhibitory (inhib.) X
synapses (yellow dots). For model details see materials and methods. (Right)
The modeled dCaAP amplitude depended on the stimulation current intensity
(Idend) with decay constant (tdCaAP) of 0.3. The dCaAP threshold was set to soma
–36 mV with 220 pA current step. (B to C) dCaAP at the dendrite during activity 1 O + AND/OR
D
of either pathway X (B) or pathway Y (C). (D to E) dCaAPs diminished when both Y
AN
AN
+
D
0 O
pathway X and Y were active together (D) but recovered with the addition of
amp.
AP
AN
inhibition (E). (F) (Top) Solution for XOR classification problem using the D
AND
activation function of dCaAP (above the abscissa). X and Y inputs to the apical
0 1 basal dendrites
dendrites triggered dCaAPs with high amplitude for (X, Y) input pairs of (1, 0) X
and (0, 1), marked by blue circle and red cross, but not for (0, 0) and (1, 1),
marked by red circle and blue cross. (Bottom) Solution for OR classification. Somatic AP was triggered for (X, Y) input pairs of (1, 1), (0, 1), and (1, 0), but not for
(0, 0). (G) Schematic model of a L2/3 pyramidal neuron with somatic compartment (green) presented as logical AND/OR gate with activation function of somatic AP,
apical dendrite compartment as logical XOR gate, and basal and tuft dendritic braches, in gray background, as logical gate AND due to the NMDA spikes (33).
summarize the classification of the dCaAPs properties of dCaAPs. dCaAPs evoked by the mammalian neocortex—namely, NMDA spikes
on the basis of their ability to trigger APs at dendritic electrode triggered somatic APs near (18) and dendritic Ca2+ APs in layer 5 pyram-
the soma (i.e., coupled versus uncoupled) and threshold but were suppressed by further in- idal neurons (15, 19–22)—that were previously
their complexity (i.e., complex and/or simple). crease in the stimulus intensity (Fig. 2B). shown to increase with the stimulus strength,
Most of the dendrites with complex dCaAPs The dendritic activation function (namely, the activation function of dCaAPs in L2/3
also triggered simple dCaAPs, suggesting that the amplitude of dCaAPs as a function of the neurons was sharply tuned to a specific input
their behavior might be activity- or input- intensity of the current injection in the den- strength (Fig. 2I).
dependent and/or modulated by other fac- drite, Idend) reached its maximal value at the We used a compartmental model of a L2/3
tors (17). rheobase (i.e., for Idend = Irhe where Irhe is the pyramidal neuron that replicated the pheno-
dCaAPs affected the input–output transfor- threshold current for triggering a dCaAP) and menology of the dCaAP behavior in the den-
mation of the cells. Typically, somatic AP firing decayed for stronger Idend (Fig. 2, D to F; 12 un- drite to investigate the functional outcome of
increases with the input current intensity in- coupled dCaAPs). The mean width of the den- the dCaAP activation function (for a biophys-
jected to the soma. In contrast, in 4 cells (out dritic activation function (defined here as the ical model of dCaAPs, see fig. S12). L2/3 py-
of 12 cells that had repetitive and coupled decay constant of a single exponential fit) was ramidal neuron morphology was digitally
dCaAPs) our recordings revealed an inverse 0.39 (0.38 median; in units of Irhe), which in- reconstructed and modeled in the NEURON
behavior where increasing the intensity of den- dicates that dCaAPs are sharply tuned (highly (23) simulation environment (Fig. 3A). The
dritic (rather than somatic) current injection selective) to a particular input strength. Addi- dCaAP’s threshold, width, and amplitude as a
resulted in decreased somatic firing. For exam- tionally, L2/3 dendrites were heterogeneous function of the input strength were simulated
ple, in Fig. 2, A and B, the dendritic electrode in their activation function threshold and width by the sum of current sources with a sigmoidal
evoked one or two somatic APs with current (Fig. 2F). In contrast, in a similar range of input shape (for details, see materials and methods
near threshold but failed to evoke APs for intensities, somatic APs (Fig. 2, G to H) showed and Fig. 3A, right panel). To simulate two dis-
higher current intensity. In contrast, at the a typical threshold activation function; once a tinct classes of inputs, pathways X and Y, we
soma of the same cells, AP output increased somatic AP was triggered, its amplitude was used 25 excitatory synapses for each pathway
with the input’s strength (Fig. 2C). These virtually independent of the input intensity (Fig. 3A), targeting a subregion of the apical
results are explained by the unusual active (Fig. 2H). Unlike other dendritic APs in the dendrite (blue and red dots in Fig. 3A). Each of

these pathways was able to trigger dCaAPs by 7. S. L. Smith, I. T. Smith, T. Branco, M. Häusser, Nature 503, Meeting Planner, Society for Neuroscience, San Diego, CA, 3 to
itself (Fig. 3, B and C). Because of the activa- 115–120 (2013). 7 November 2018.
8. J. J. Moore et al., Science 355, eaaj1497 (2017). 36. A. Gidon et al., Dendritic action potentials and computation
tion function of the dCaAPs in our simulation, 9. B. Sivyer, S. R. Williams, Nat. Neurosci. 16, 1848–1856 in human layer 2/3 cortical neurons. Zenodo (2020);
coincident activation of two synaptic input (2013). https://doi.org/10.5281/zenodo.3530043.
pathways diminished the dCaAP amplitude 10. S. Crochet, P. Fuentealba, I. Timofeev, M. Steriade,
Cereb. Cortex 14, 1110–1121 (2004). AC KNOWLED GME NTS
(Fig. 3D) in contrast to other dendritic APs 11. Y. Kim, C.-L. Hsu, M. S. Cembrowski, B. D. Mensh, N. Spruston, We thank U. Schneider for providing the human tissue; L. Kraus,
that amplify coincident dendritic inputs (24) eLife 4, e06414 (2015). A. Ragot, O. Kruchik, and I. Wolter for assisting with human tissue
[e.g., in layer 5 pyramidal neurons in the 12. M. E. Larkum, J. J. Zhu, B. Sakmann, J. Physiol. 533, 447–466 processing; and S. Grosser and F. J. Barreda Tomás for assisting
(2001). with confocal imaging. Funding: This work was supported by
rodent neocortex (25) or in CA1 neurons of 13. N. L. Golding, N. Spruston, Neuron 21, 1189–1200 (1998). Deutsche Forschungsgemeinschaft DFG: 2112280105 (to T.A.Z.),
the rodent hippocampus (26)]. Our simula- 14. L. M. Palmer et al., Nat. Neurosci. 17, 383–390 (2014). EXC 257 (to I.V., P.P., and M.E.L.), FOR 2143 (to I.V.), EXC 2049
tion is therefore a simple and explicit dem- 15. M. E. Larkum, J. J. Zhu, B. Sakmann, Nature 398, 338–341 (to P.F.), LA 3442/3-1 (to M.E.L.), SPP1665 (to M.E.L.), and
(1999).
onstration of how the dendritic mechanism SFB1078 B2 (to M.E.L.); 7FP Health-F2-602531-2013 DESIRE
16. N. L. Golding, H. Y. Jung, T. Mickus, N. Spruston, J. Neurosci. (to M.H.); Hellenic Foundation for Research and Innovation HFRI
observed in human L2/3 pyramidal neurons 19, 8789–8798 (1999). and the General Secretariat for Research and Technology GSRT
computes an anticoincident function for mul- 17. C. Labarrera et al., Cell Rep. 23, 1034–1044 (2018). 1357 (to A.P.); Humboldt Foundation Friedrich Wilhelm Bessel
tiple input pathways, limiting the number 18. J. Schiller, G. Major, H. J. Koester, Y. Schiller, Nature 404, Research Award (to P.P.); H2020 European Research Council ERC
285–289 (2000). STG 311435 (to P.P.); H2020 Research and Innovation Programme
and/or the strength of inputs integrated in 19. M. E. Larkum, K. M. M. Kaiser, B. Sakmann, Proc. Natl. Acad. 720270/HBP SGA1, 785907/HBP SGA2, and 670118/ERC
the dendrite (for impact on the cell body, see Sci. U.S.A. 96, 14600–14604 (1999). ActiveCortex (to M.E.L.); and EMBO ALTF 1193–2015 (to A.G.).
fig. S9). Inhibition (27, 28) placed at the same 20. M. E. Larkum, T. Nevian, M. Sandler, A. Polsky, J. Schiller, Author contributions: M.E.L. and A.G. conceptualized and
Science 325, 756–760 (2009). performed the experiments and analysis and wrote the original
dendritic subregion (20 GABAergic synapses), 21. J. J. Zhu, J. Physiol. 526, 571–587 (2000). draft. T.A.Z. performed the 2-p experiments. P.F. and M.H.
in addition to the two excitatory pathways, 22. J. Schiller, Y. Schiller, G. Stuart, B. Sakmann, J. Physiol. 505, managed human tissue ethical aspects, delivery, and quality
repolarized the membrane and recovered the 605–616 (1997). optimization. A.G., A.P., and P.P. conceptualized and created the
amplitude of the dCaAPs [Fig. 3E; (29)]. These 23. M. L. Hines, N. T. Carnevale, Neural Comput. 9, 1179–1209 models. F.B. and I.V. performed visualization, imaging, and
(1997). morphological reconstruction of recorded neurons. All the authors
results suggest that the precise balance be- 24. M. London, M. Häusser, Annu. Rev. Neurosci. 28, 503–532 participated in writing, reviewing, and editing the manuscript.
tween excitation and inhibition is essential (2005). Competing interests: The authors declare no competing interests.
for the generation of dCaAPs and indicate a 25. A. Polsky, B. W. Mel, J. Schiller, Nat. Neurosci. 7, 621–627 Data and materials availability: NEURON simulation files are
(2004). available at https://modeldb.yale.edu/254217 for Fig. 3 and fig. S9
counterintuitive role for inhibition in enhancing and at https://modeldb.yale.edu/260178 for fig. S12.
26. A. Losonczy, J. C. Magee, Neuron 50, 291–307 (2006).
the excitability of the dendrite (see also fig. S9, 27. E. Boldog et al., Nat. Neurosci. 21, 1185–1195 (2018). Reconstructed neurons and all summary graphs with
C and D). 28. A. Gidon, I. Segev, Neuron 75, 330–341 (2012). corresponding data are stored at Zenodo (36).
It has long been assumed that the summa- 29. A. T. Gulledge, G. J. Stuart, Neuron 37, 299–309 (2003).
30. G. M. Shepherd, R. K. Brayton, Neuroscience 21, 151–165 (1987).
tion of excitatory synaptic inputs at the den- SUPPLEMENTARY MATERIALS
31. M. Minsky, S. A. Papert, Perceptrons: An Introduction to
drite and the output at the axon can only Computational Geometry (MIT Press, 1969). science.sciencemag.org/content/367/6473/83/suppl/DC1
instantiate logical operations such as AND 32. F. Rosenblatt, The Perceptron, a Perceiving and Recognizing Materials and Methods
Automaton: (Project Para) (Cornell Aeronautical Laboratory, Figs. S1 to S12
and OR (30). Traditionally, the XOR opera- Tables S1 and S2
1957).
tion has been thought to require a network 33. G. Eyal et al., Front. Cell. Neurosci. 12, 181 (2018). References (37–51)
solution (31, 32). We found that the dCaAPs’ 34. P. Poirazi, T. Brannon, B. W. Mel, Neuron 37, 989–999 (2003).
activation function allowed them to effectively 35. G. Testa-Silva, S. Honnuraiah, C. French, J. King,
compute the XOR operation in the dendrite K. Drummond, L. M. Palmer, G. J. Stuart, “NMDA spikes in 30 April 2019; accepted 22 November 2019
human neocortex,” program no. 463.12, 2018 Neuroscience 10.1126/science.aax6239
by suppressing the amplitude of the dCaAP
when the input is above the optimal strength
(Fig. 2). Thus, on the basis of our results and
those of previous studies (30, 33), we consider
a model that portrays the somatic and den- ARCHAEOLOGY
dritic compartments of L2/3 neurons as a
network of coupled logical operators and Cooked starchy rhizomes in Africa
corresponding activation functions (Fig. 3,
F and G). In this model, the XOR operation 170 thousand years ago
is performed in the dendrites with dCaAPs,
whereas AND/OR operations are performed Lyn Wadley1*, Lucinda Backwell1,2, Francesco d’Errico3,4, Christine Sievers5
at the soma and at tuft and basal dendrites
with sodium and NMDA spikes, respectively Plant carbohydrates were undoubtedly consumed in antiquity, yet starchy geophytes were seldom
(20, 25, 34, 35). Our findings provide insights preserved archaeologically. We report evidence for geophyte exploitation by early humans from at least
into the physiological building blocks that 170,000 years ago. Charred rhizomes from Border Cave, South Africa, were identified to the genus
constitute the algorithms of cellular func- Hypoxis L. by comparing the morphology and anatomy of ancient and modern rhizomes. Hypoxis
tion, which ultimately give rise to the cortical angustifolia Lam., the likely taxon, proliferates in relatively well-watered areas of sub-Saharan Africa and
network behavior. in Yemen, Arabia. In those areas and possibly farther north during moist periods, Hypoxis rhizomes
would have provided reliable and familiar carbohydrate sources for mobile groups.
A
1. J. DeFelipe, L. Alonso-Nanclares, J. I. Arellano, J. Neurocytol.
31, 299–316 (2002).
2. H. Mohan et al., Cereb. Cortex 25, 4839–4853 (2015).
ncient hunting strategies receive more security in the past, as reported for hunter-
3. Y. Deitcher et al., Cereb. Cortex 27, 5398–5414 (2017). attention from archaeologists than plant gatherers in Africa during the last century (1).
4. J. Waters, M. Larkum, B. Sakmann, F. Helmchen, J. Neurosci. collecting, because plant preservation African venison is especially lean in the dry
23, 8558–8567 (2003).
is often poor in archaeological sites. A season; thus, human populations able to sup-
5. M. E. Larkum, J. Waters, B. Sakmann, F. Helmchen, J. Neurosci.
27, 8999–9008 (2007). plant diet, though sometimes invisible, plement meat diets with carbohydrate or fat
6. L. Beaulieu-Laroche et al., Cell 175, 643–651.e14 (2018). must have contributed substantially to food avoided malnourishment (2, 3). Geophytes

(corms, bulbs, tubers, and rhizomes) store near the base of the member. Only humans comes from amylase gene analysis results,
starch in their underground organs, and could have transported whole rhizomes from which indicate that a high starch diet, pos-
these underground portions become sources the field to the cave. The Border Cave speci- sibly involving processing and/or cooking of
of carbohydrate for humans and those animals mens were preserved because they were charred carbohydrate-rich geophytes by early humans,
able to excavate them. Modern collecting of and presumably because they were lost while was already in place by the Middle Pleistocene
edible geophytes in South Africa demonstrates roasting in the ashes, from which they were (24). Cooking enables dietary diversity (25),
that a gatherer’s daily caloric requirement can recovered archaeologically (Fig. 2). Nothing and transporting geophytes to a home base
be met within 2 hours (4), and such foods may indicates that the rhizomes were pulverized like Border Cave facilitates both food process-
become dietary staples. Cooking increases di- (23), but circumstantial evidence for cooking ing and sharing.
gestibility of meat and plant food, reduces is compelling. The spatial context of the rhi- The early Middle Stone Age occupations at
toxicity, and, in the case of geophytes, has a zomes in ash rather than adjacent sediment Border Cave were dated by electron spin reso-
considerable softening effect (5), eases peeling is significant. Further support for cooking nance on tooth enamel to 161 ± 10 and 144 ±
and chewing, and enhances glucose availability
(6, 7).
Before the use of fire, hominins may have
Fig. 1. Border Cave
eaten geophytes raw, especially Cyperaceae
entrance and
and aquatic plants, though some of these first
stratigraphy in
required pulverizing (8). No geophyte remains
Members 5 BS and
have been recovered from early, pre-fire sites.
4 WA. (A) Border
Instead, circumstantial evidence is from sources
Cave perched on a
such as isotopes, DNA, or dental calculus
cliff in the Lebombo
(9–14). Seeds from several edible geophytes
Mountains. (B) Border
and aquatic plants were found in samples
Cave stratigraphy,
from 780 thousand years (ky) ago at Gesher
excavated from 2015
Benot Ya‘aqov in Israel (15); the samples con-
to 2018, from Mem-
tained seeds rather than the geophytes them-
bers 5 BS and 4 WA
selves, perhaps because no trace remains after
in Squares N108 E113
consumption of geophytes. At Klasies River,
and N109 E113. Note
South Africa, 120-ky-old charred parenchyma
the stacked combus-
fragments from unidentified geophytes imply
tion features in 4 WA.
cooking (16). Holocene layers of Bushman
Scale bar, 30 cm.
Rock Shelter and Melkhoutboom, South Africa,
yielded geophytes including Hypoxis spp.
(17, 18). Here, we present earlier evidence, from
at least 170 ky ago, for the cooking of identified
edible rhizomes from Border Cave, northern
KwaZulu-Natal, South Africa (19) (fig. S1).
Border Cave formed in a Lebombo Moun-
tain cliff facing west across eSwatini (Fig. 1A).
The cave was extensively excavated from early
in the 1900s (20), but little attention was given
to its botanical remains, apart from prelimi-
nary studies of 40-ky-old seeds and leaves (21)
and chemical identifications of plant poison
and resin (22). The cave has alternating brown
sand (BS) and white ash (WA) stratigraphic
members (21), and rhizomes were recovered
from the 5 Brown Sand (5 BS) and 4 White Ash
(4 WA) samples. Discrete layers are discern-
ible in each member (Fig. 1B). The majority of
identified 4 WA rhizomes come from White 8
to 5 (Table 1), which are combustion features Table 1. Border Cave whole rhizome frequencies in Members 5 BS and 4 WA.
1
Evolutionary Studies Institute, University of the Member Layer Frequency of whole rhizomes
Witwatersrand, Johannesburg, South Africa. 2Instituto
Superior de Estudios Sociales, ISES-CONICET, Tucumán,
4 WA Top and Pinkish Grey 6
.....................................................................................................................................................................................................................
Argentina. 3Centre National de la Recherche Scientifique, White 1, 2, and 3 8
.....................................................................................................................................................................................................................
UMR 5199—PACEA, Université de Bordeaux, Bâtiment White 5 and 6 14
.....................................................................................................................................................................................................................
B2, Allée Geoffroy Saint Hilaire, CS 50023, 33615 Pessac,
White 7 and 8 13
France. 4SFF Centre for Early Sapiens Behaviour .....................................................................................................................................................................................................................
(SapienCE), University of Bergen, Øysteinsgate 3, Reddish Brown Ian, Dark Brown pit 3
.....................................................................................................................................................................................................................
Postboks 7805, 5020 Bergen, Norway. 5School of 5.....................................................................................................................................................................................................................
BS Very Dark Grey and Slump 6
Geography, Archaeology and Environmental Studies,
Dark Greyish Brown 5
.....................................................................................................................................................................................................................
University of the Witwatersrand, Johannesburg, South
Africa. Total 55
.....................................................................................................................................................................................................................
*Corresponding author. Email: lyn.wadley@wits.ac.za

11 ky ago for 5 BS and 168 ± 5, 115 ± 8, and 113 ± prise 55 complete, charred specimens (Table 1) tures, notwithstanding the deterioration caused
5 ky ago for 4 WA (means ± standard errors) that seem to belong to the same taxon, as well by charring and mechanical abrasion by cave
(26). Bayesian analysis to obtain 95.4% proba- as fragments of charred parenchyma that can- sediment (supplementary text). Vascular bun-
bility ranges for boundaries between members not be securely identified (table S1). They are dles mostly comprise elongated clusters of xylem
indicated that Member 5 BS occupations may typically globose with a convex or slightly coni- vessels (12 to >20) with thickened walls of
have commenced around 177 ky ago, while cal base, a depressed center on the proximal scalariform tissue (Fig. 3D). Parenchyma cells,
those in 4 WA began about 150 ky ago and surface, radial splits in the ground tissue, rings when not fused into a solid carbon mass, have
terminated no later than 99 ky ago (27). from leaf scars or fiber on the outer circum- geometric or oval shapes. The outer ground
The term rhizome is used here (supplemen- ference (Fig. 2), and root traces emerging from tissue (cortex) contains many bundles of needle-
tary text), following the terminology of Wiland- fiber sheaths within the cortex (Fig. 3B). High like raphides of calcium oxalate (Fig. 3F)
Szymańska and Adamski (28) and Singh (29). magnification using scanning electron micros- that are also present in lower frequencies
The Border Cave 5 BS and 4 WA rhizomes com- copy (SEM) reveals distinctive anatomical fea- in the inner ground tissue (pith). All of these
Fig. 2. Two whole Border Cave charred

rhizomes. (A) Three views of Border
Cave rhizome BC 17 from Square N108
E114, Member 4 WA, layer White 6.1 Idaho
(#9577). The proximal end (top left)
has been abraded, exposing radial splits
caused by charring the fresh organ.
Scale bar, 2 mm. (B) Three views of
Border Cave rhizome BC 6 from Square
N108 E114, Member 4 WA, layer White 6
Idaho (#9433). The rhizome proximal
end (left) has a marked central depression
and cavities, some of which are root
cavities on the circumference. The rhizome
profile (center) has a few raised ring
scars, and the distal end (right) has a
wrinkled, convex base. Scale bar, 2 mm.
Fig. 3. Selected anatomical

features of modern
charred Hypoxis rhizomes
compared with Border
Cave rhizomes. (A) Modern
charred Hypoxis rhizome
#55 with rhizodermal sheath.
Transverse section (TS) scale,
500 mm. (B) Border Cave
rhizome BC 17 with rhizoder-
mal sheath. TS scale, 200 mm.
(C) Modern charred Hypoxis
rhizome #27, cluster of xylem
vessels with walls of scalari-
form tissue. TS scale, 30 mm.
(D) Border Cave BC 6, cluster
of xylem vessels with walls
of scalariform tissue. TS scale,
30 mm. (E) Modern, charred
Hypoxis rhizome #27 with
raphide bundle. TS scale,
30 mm. (F) Border Cave
rhizome BC 30, raphide
bundle. TS scale, 30 mm.
X, xylem; P, parenchyma;
RB, raphide bundle; RS,
rhizodermal sheath.

Fig. 4. Modern Hypoxis angustifolia field

appearance, morphology, anatomy, and
distribution. (A) H. angustifolia habit.
(B) H. angustifolia fresh rhizome, cut
transversely to reveal white flesh. Scale
bar, 500 mm. (C) H. angustifolia mature fresh
rhizome (left) with developing offshoot (right).
Scale bar, 10 mm. (D) Simplified distribution of
H. angustifolia through Africa, Yemen, and Indian
Ocean islands. (E) H. angustifolia fresh rhizome
cut transversely to expose parenchyma cells
filled with starch grains. Scale bar, 200 mm.
morphological and anatomical features match (Fig. 4C), and retains visibility year-round, unlike 2. P. J. Butterworth, P. R. Ellis, M. Wollstonecroft, in Wild
those of modern Hypoxis L. spp. rhizomes (Fig. deciduous Hypoxis species. It thrives in a variety Harvest: Plants in the Hominin and Pre-agrarian Human Worlds,
K. Hardy, Kubiak-Martens, Eds. (Oxbow, Oxford, 2016),
3, A, C, and E; figs. S2 and S3). Attributes of of modern habitats and is thus likely to have pp. 31–54.
geophytes from other families appropriate to had a wide distribution in the past as it does 3. J. D. Speth, K. A. Spielmann, J. Anthropol. Archaeol. 2, 1–31
the area did not match the Border Cave speci- today. It occurs in sub-Saharan Africa, Sudan (1983).
4. E. Singels et al., J. Hum. Evol. 101, 79–89 (2016).
mens (supplementary text; table S2). (to about 13°N), some Indian Ocean islands, 5. N. J. Dominy, E. R. Vogel, J. D. Yeakel, P. Constantino,
The morphology of the Border Cave rhizomes and as far afield as Yemen (Fig. 4D) (28, 31). Its P. W. Lucas, Evol. Biol. 35, 159–175 (2008).
6. R. Wrangham, Curr. Anthropol. 58 (S16), S303–S313
is suggestive of Hypoxis angustifolia Lam. modern occurrence in Yemen may indicate
(2017).
(Fig. 4, A, B, and C) rather than the tiny wider distribution of the rhizome during pre- 7. L. Wandsnider, J. Anthropol. Archaeol. 16, 1–48
H. filiformis Baker or large H. hemerocallidea vious periods of humid conditions. (1997).
8. M. M. Wollstonecroft, P. R. Ellis, G. C. Hillman, D. Q. Fuller,
Fisch., C.A.Mey. & Avé-Lall. The relatively The Border Cave discovery is early evidence of Veg. Hist. Archaeobot. 17, 19–27 (2008).
small, slender-leafed Hypoxis taxa, such as cooked starchy plant food. The wide distribu- 9. M. Sponheimer, J. A. Lee-Thorp, Science 283, 368–370
H. angustifolia, H. argentea Harv. ex Baker, tion of Hypoxis, particularly the small, palatable (1999).
10. C. R. Peters, J. C. Vogel, J. Hum. Evol. 48, 219–236
and H. filiformis, have rhizomes with white Hypoxis angustifolia rhizome that grows gre-
(2005).
(Fig. 4) or pale yellow flesh. They are more gariously in many habitats, implies that it could 11. J. Lee-Thorp et al., Proc. Natl. Acad. Sci. U.S.A. 109,
palatable than the orange-fleshed rhizomes have provided a reliable, familiar staple food 20369–20372 (2012).
12. K. Hardy, J. Brand-Miller, K. D. Brown, M. G. Thomas,
from plants such as H. hemerocallidea (29) and source for early humans moving within or out
L. Copeland, Q. Rev. Biol. 90, 251–268 (2015).
are therefore favored as food. While they are of Africa. Monocotyledons also provided food 13. A. G. Henry, A. S. Brooks, D. R. Piperno, J. Hum. Evol. 69,
edible raw, Hypoxis rhizomes have high frac- security when people moved through south- 44–54 (2014).
ture toughness until they are cooked (5). The ern Europe during the Palaeolithic (32). Cer- 14. L. S. Weyrich et al., Nature 544, 357–361 (2017).
15. Y. Melamed, M. E. Kislev, E. Geffen, S. Lev-Yadun,
rhizomes are nutritious with an energy value tain raw materials found in the Middle Stone N. Goren-Inbar, Proc. Natl. Acad. Sci. U.S.A. 113, 14674–14679
of approximately 500 KJ/100 g, supplemented Age site of Olorgesaile, East Africa, had dis- (2016).
by essential vitamins and minerals (17). tant origins by ≥295 to ~320 ky ago, implying 16. C. Larbey, S. M. Mentzer, B. Ligouis, S. Wurz, M. K. Jones,
J. Hum. Evol. 131, 210–227 (2019).
H. hemerocallidea rhizomes have constant that mobile networks were longstanding (33). 17. L. Wadley, Hunters and Gatherers of the Later Stone Age,
carbohydrate composition year-round, though Border Cave is remote from dispersal routes Southern Transvaal (Cambridge Monographs in African
soluble sugars are slightly reduced during the out of Africa, yet the site contributes data (19) Archaeology 25, Oxford, 1987).
18. H. J. Deacon, Where Hunters Gathered (South African
resting season (30). for assessing the ease with which early Homo
Archaeological Society Monograph Series 1, 1976).
H. angustifolia, H. argentea, H. filiformis, and sapiens could travel within the continent. 19. Materials and methods are available as supplementary
H. hemerocallidea grow in the Border Cave materials.
20. L. R. Backwell et al., J. Field Archaeol. 43, 417–436
area, but recent overexploitation has reduced RE FERENCES AND NOTES (2018).
their populations. H. angustifolia is gregari- 1. G. B. Silberbauer, Hunter and Habitat in the Central Kalahari 21. P. B. Beaumont, thesis, University of Cape Town, South Africa
ous, propagates readily from rhizome side shoots Desert (Cambridge Univ. Press, 1981). (1978).

22. F. d’Errico et al., Proc. Natl. Acad. Sci. U.S.A. 109, 13214–13219 J.-J. Hublin, M. P. Richards, Eds. (Springer, 2009), was by L.W. with contributions from C.S., L.B., and F.D.
(2012). pp. 171–180. All authors reviewed and edited the manuscript. Competing
23. L. González Carretero, M. Wollstonecroft, D. Q. Fuller, 33. A. S. Brooks et al., Science 360, 90–94 (2018). interests: The authors declare no competing interests. Data
Veg. Hist. Archaeobot. 26, 415–432 (2017). and materials availability: All data and materials are included
24. C. E. Inchley et al., Sci. Rep. 6, 37198 (2016). ACKN OWLED GMEN TS in the text and supplementary materials. The Border Cave
25. G. Hillman, M. Wollstonecroft, in Archaeology of African We thank M. Bamford, N. Crouch, S. Simm, H. Glen, R. Glen, and excavated material (by L.B.) is housed in the Evolutionary
Plant Use, C. J. Stephens, S. Nixon, M. A. Murray, L. Kubiak-Martens for comments and botanical assistance and Studies Institute, University of the Witwatersrand,
D. Q. Fuller, Eds. (Left Coast Press, Walnut Creek, 2014), two anonymous reviewers for useful suggestions. Border Cave Johannesburg, South Africa (Border Cave 2015/2019).
pp. 37–49. was excavated with permit #SAH 15/7645 (to L.B.) from Amafa.
26. R. Grün, P. Beaumont, P. V. Tobias, S. Eggins, J. Hum. Evol. 45, The comparative plant collection was made with permit
155–167 (2003). #OP4367/2017 (to C.S.) from Ezemvelo KZN Wildlife. G. Pettit
27. A. R. Millard, Quat. Geochronol. 1, 159–166 (2006). supplied some geophytes. The Evolutionary Studies Institute science.sciencemag.org/content/367/6473/87/suppl/DC1
28. J. Wiland-Szymańska, Z. Adamski, J. Wiland-Szymanska, Novon provided resources. A. Kruger produced the drone image. Materials and Methods
12, 142–151 (2002). A. Millard provided data from his 2006 Bayesian analysis of Supplementary Text
29. Y. Singh, thesis, University of Pretoria, South Africa ESR dates. Funding: L.B. received funding from the National Figs. S1 to S3
(2000). Geographic Society (NGS-54810R-19) and Centre of Excellence Tables S1 and S2
30. J. W. Bews, J. E. Vanderplank, Ann. Bot. os-44, 689–719 (OP2015/04LB). The Research Council of Norway Centres of References (34–59)
(1930). Excellence, SFF Centre for Early Sapiens Behavior (SapienCE)
31. J. Wiland-Szymańska, Biodivers. Res. Conserv. 14, 1–129 (262618) partly funded F.D. Author contributions: The
(2009). excavation was conceived by L.B. and F.D. and largely funded
32. M. Jones, in The Evolution of Hominin Diets: Integrating by L.B. Botanical investigation, formal analysis, and project 22 September 2019; accepted 5 November 2019
Approaches to the Study of Palaeolithic Subsistence, conceptualization were undertaken by L.W. and C.S. Writing 10.1126/science.aaz5926
PLANT SCIENCE stages, organ primordia are dome-shaped

(Fig. 1G). On the basis of morphology alone, it
Evolution of carnivorous traps from planar leaves is unclear at this stage whether these primor-
dia will become leaflets or traps. At later
through simple shifts in gene expression stages, leaflet primordia form tapering cylin-
ders that grow to be slightly wider than thick
Christopher D. Whitewoods1*, Beatriz Gonçalves1*, Jie Cheng1,2,3*, Minlong Cui4, Richard Kennaway1, and curve longitudinally toward the apex
Karen Lee1, Claire Bushell1, Man Yu1, Chunlan Piao4, Enrico Coen1† (Fig. 1, H to J). By contrast, trap primordia
are curved in both longitudinal and trans-
Leaves vary from planar sheets and needle-like structures to elaborate cup-shaped traps. Here, verse sections and consist of three cell layers
we show that in the carnivorous plant Utricularia gibba, the upper leaf (adaxial) domain is restricted (Fig. 1K). The inner layer is positioned adaxially
to a small region of the primordium that gives rise to the trap’s inner layer. This restriction is (facing the spiral apex, the right-hand side of
necessary for trap formation, because ectopic adaxial activity at early stages gives radialized leaves the images in Fig. 1). As development pro-
and no traps. We present a model that accounts for the formation of both planar and nonplanar gresses, the trap grows to a near-spherical
leaves through adaxial-abaxial domains of gene activity establishing a polarity field that orients shape with a closed mouth (Fig. 1, L to M,
growth. In combination with an orthogonal proximodistal polarity field, this system can generate white arrowheads), and a two-cell-thick trap
diverse leaf forms and account for the multiple evolutionary origins of cup-shaped leaves through door grows out near the dorsal lip (Fig. 1, N
simple shifts in gene expression. and O, orange arrowheads). Over a 20-fold
increase in trap length (about 400-fold in-
L
crease in area), lamina thickness only doubles
eaves come in many shapes and sizes. orients the plane of division or the plane of (Fig. 1, K to O, and fig. S1), resulting in a curved
Most consist of planar sheets of cells division orients growth. Finally, it is unclear sheet.
that harvest light for photosynthesis. how the system for planar leaf development To define the ad-ab domains in U. gibba
Formation of these leaves depends on has been modified to generate nonplanar (9, 10), we identified homologs (named with
adaxial and abaxial domains of gene leaves, such as filiform (needle-like) leaves Ug prefix) of the adaxially expressed PHV and
activity in leaf primordia (1, 2). However, the and cup-shaped leaves of carnivorous plants, PHB genes and abaxially expressed FIL and
mechanism by which these domains generate which have evolved multiple times indepen- KAN genes (1, 11). Before trap and leaflet primor-
sheet-like development is unclear. It is un- dently (3). dia morphologies clearly diverged, UgPHV1
known whether growth is oriented by the Computational models for formation of flat was expressed on the adaxial side (black arrow-
adaxial-abaxial (ad-ab) boundary throughout or cup-shaped leaves have been proposed based head in Fig. 2A) and was more restricted in
the leaf or solely at the epidermis. It is also on cell divisions being induced by the epider- some primordia (yellow arrowhead in Fig. 2A).
unclear how orientations of growth and cell mal ad-ab boundary, with the plane of division In leaflet primordia at later stages, UgPHV1
division are specified and whether growth orienting growth (4, 5). However, these models and UgFIL1 were expressed on the adaxial and
are not easily reconciled with observations that abaxial sides, respectively (Fig. 2, B to E, and
1
cell divisions occur throughout the leaf lamina, fig. S3, A to J). In trap primordia, UgPHV1 was
Department of Cell and Developmental Biology, John Innes
Centre, Norwich Research Park, Colney Lane, Norwich NR4
not solely at the margin (6–8). Here, we sug- expressed in the innermost cell layer, extend-
7UH, UK. 2State Key Laboratory of Systematic and gest an alternative mechanism based on the ing into the inner side of the trap door (Fig. 2,
Evolutionary Botany, CAS Center for Excellence in Molecular analysis of ad-ab genes in trap and filiform F to I, and fig. S3, K to P). UgFIL1 and UgKAN1
Plant Sciences, Institute of Botany, Chinese Academy of leaf development of the humped bladderwort, were expressed in the outer layers (Fig. 2, J to
Sciences, Beijing 100093, China. 3College of Life Sciences,
University of Chinese Academy of Sciences, Beijing 100049, Utricularia gibba (Lentibulariaceae), an Q, and fig. S3, Q to V), although only UgKAN1
China. 4College of Agriculture and Food Science, Zhejiang aquatic carnivorous plant (Fig. 1, A to F). was expressed in the outer ventral region
Agriculture and Forestry University, Linan 311300, Zhejiang, Each U. gibba leaf consists of several fili- (yellow arrowheads in Fig. 2, N to Q). Thus,
China.
*These authors contributed equally to this work. form leaflets (Fig. 1, B, E and F) and may bear a the adaxial and abaxial domains of a planar
†Corresponding author. Email: enrico.coen@jic.ac.uk trap (Fig. 1, D and F). At early developmental leaf broadly correspond to the inner and outer

Fig. 1. U. gibba morphology. (A) Stolon with

circinnate apex (red arrowhead) bearing leaves and
traps (black arrowheads) oriented adaxially (black
arrows). (B) Schematic of (A) showing stolons
(black) and leaves (gray). L, leaflet-bearing leaf;
LT, trap-bearing leaf; AM, axillary meristem;
M, apical meristem. (C) Magnification of apex
shown in (A). (D) Trap showing dorsal (black line)
and ventral (white line) sides and mouth (white
arrowhead). (E) Diagram of a leaflet-bearing
leaf in (B), with one leaflet in red. (F) Diagram
of a trap-bearing leaf in (B), with a trap in green.
(G and G′) Confocal micrographs of early
primordium, in longitudinal (G) and transverse
section viewed obliquely (G′). (H to J ) Confocal
micrographs of a developmental series of leaflets in
longitudinal [(H) to (J)] and transverse sections
viewed obliquely [(H′) to (J′)]. (K to O ) Confocal
micrographs of a developmental series of traps in
longitudinal [(K) to (O)] and transverse sections
viewed obliquely [(K′) to (O′)]. In (G) to (O′),
green lines mark primordium and leaf length; yellow
lines mark primordium, leaf, and trap thickness; red
lines mark primordium, leaf, and trap width; and
blue lines mark trap length. Measurements are
given in the same colors as the lines (see fig. S1
for additional data). In (L) to (O), orange
arrowheads mark the trap door, and white arrow-
heads mark the trap mouth. In (D) and (H) to (O′),
organs are shown with the adaxial side to the right,
marked “ad” in (H) and (K). Scale bars are 5 mm
in (A), 1 mm in (C), 500 mm in (D), 50 mm in
[(G), (G′), (H), (H′), (I), (I′), (J′), and (K) to (N′)],
and 100 mm in [(J), (O), and (O′)]. The image in (N)
was previously published in (26).
regions of the trap, respectively. Similar find- induced ectopic expression of microRNA- tion confirmed ectopic induction throughout
ings were reported for the Sarracenia purpurea resistant UgPHV1 under the control of the the tissue (fig. S4).
trap, although KAN expression was not de- 35S promoter, using a Cre-Lox system (HS- To determine how ectopic UgPHV1 affected
tected (5). UgPHV1 plants, see methods for details). development, tissues were imaged daily after
To determine whether the observed expres- After extended heat shock, green fluorescent induction (Fig. 3, A to D). At 7 days post-
sion patterns have functional importance, we protein fluorescence and in situ hybridiza- induction (Fig. 3D), the main axis could be

Fig. 2. UgPHV1 and UgFIL1 Young Primordia Leaflet

expression patterns in U. gibba
leaflet and trap development.
A B C
(A) UgPHV1 expression in young
primordia. Black and yellow arrow-
heads mark primordia that show UgPHV1
extended and restricted domains (ad) ad
of the adaxial marker, respectively.
(B to E) UgPHV1 [(B) and (C)] and
UgFIL1 [(D) and (E)] in leaflet
UgPHV1
primordia. (F to I) UgPHV1 during (ad)
trap development, in longitudinal D E
[(F), (H), and (I)] and cross (G)
sections. (J to M) UgFIL1 expression
during trap development in longitu- UgFIL1 ad
dinal [(J), (L), and (M)] and cross (ab)
(K) sections. (N to Q) UgKAN1
expression during trap development
in longitudinal sections. The dashed
red lines in (B), (D), (F), and (J)
indicate the angle of cross Trap
sections shown in (C), (E), (G),
and (K), respectively. Trap mouth F G H I
(black arrowheads), trap door (red
arrowheads), and ventral regions
(yellow arrowheads) are indicated. UgPHV1
In all panels, the adaxial side is (ad)
shown to the right, marked “ad” in
(B) and (D). Scale bars are 50 mm.
See an expanded version of this
figure in fig. S3.
J K L M
UgFIL1
(ab)
N O P Q
UgKAN1
(ab)
divided into three regions. (i) An upper re- (Fig. 3D, blue) derived from leaves bearing (Fig. 3, E and F) and generation of radialized
gion, encompassing the apex and leaves (Fig. small traps (80 to 200 mm long) at the time of organs, similar to the effect of ectopic PHV
3D, red). The normal spiral organization of induction (Fig. 3C). This region had normal and PHB expression in Arabidopsis (11–13).
the apex had been replaced by an open linear leaves bearing traps up to 300 mm long (Fig. Ectopic UgPHV1 expression in later trap
structure (Fig. 3, H to M), containing no trap 3D, white arrowheads, and fig. S6B), with primordia, but before traps are 200 mm long,
primordia and only radially symmetrical leaf- thick walls and malformed trap doors (fig. S6, leads to aberrant trap development and growth
lets (fig. S5). The leaves below the apex bore no F to H). (iii) A lower region (Fig. 3D, black) arrest (Fig. 3G and fig. S6). Ectopic UgPHV1
traps or bore small malformed traps (Fig. 3D derived from leaves bearing traps that were expression after this stage has no effect, al-
and fig. S6, C to E). Tracing this upper region longer than 200 mm at the time of induction though this may be due to inefficiency of
back through the sequence of daily images (Fig. 3C). This region contained normal leaves induction in older traps (fig. S4). These results
showed it derived from primordia located and traps (Fig. 3D, black arrowheads, and fig. S6B). indicate that restricted UgPHV1 expression is
within the spiral apex at the time of induction Thus, ectopic expression of UgPHV1 in early necessary for initiation and maintenance of
(Fig. 3C and fig. S6B). (ii) A middle region primordia leads to loss of trap development trap development.

Fig. 3. Induction of UgPHV1 prevents trap formation. (A to F) Uninduced plants (control) produce traps (black arrowheads) at regular intervals [(A) and
(B)], whereas induction of UgPHV1 prevented new trap formation [(C), (D), and (E)] and trap growth (white arrowheads) but not leaf formation (F).
(G) Upon induction, traps smaller than 200 mm did not grow. (H to J) Apex of control plants showing circinate structure in confocal images [(I) and (J)].
(K to M) Apex of induced plants showing open structure and straight leaves in confocal images [(L) and (M)]. Scale bars are 5 mm in (A) to (D), 1 mm
in (H) and (K), 100 mm in (I) and (J), and 50 mm in (L) and (M).
To explore how domains of ad-ab identity the base toward the tip. In reality, such fields (black arrows in Fig. 4D). This field was
may control leaf morphogenesis, we modeled most likely reflect cellular polarity rather than implemented by taking the gradient of a
primordia as mechanically connected visco- a continuous field (15, 16). Specifying a higher morphogen diffusing from the outer surface
elastic volumes of material, with growth ori- growth rate parallel to, rather than perpen- of the primordium toward the junction be-
ented by a polarity field (14). We began with a dicular to, the polarity generated a tapering tween the adaxial and abaxial domains through-
hemispherical ground state, representing a cylinder, corresponding to the phenotype out the midplane. Evidence for a midplane
radialized leaf primordium without a history of radialized mutants (Fig. 4, A to C, and domain playing a role in lamina formation
of ad-ab activity, with polarity pointing prox- movie S1). comes from analysis of the WOX-PRS mod-
imodistally (red arrows in Fig. 4A). For sim- To model formation of a planar leaf, we ule (17–20). Three growth rates could then
plicity, the field was implemented by taking invoked a second polarity field, orthogonal to be specified: parallel to proximodistal polar-
the gradient of a morphogen diffusing from the first, termed the orthoplanar polarity field ity (Kpd), parallel to the orthoplanar polarity

Fig. 4. Modeling of leaf and trap develop-

ment. Oblique, cutaway, and section
views. (A to C) Generation of tapering
cylinder. Initial [(A) and top of (C)]
and final [(B) and bottom of (C)] states
are shown. Proximodistal polarity
(red arrows) runs from organizers at
the base (magenta) to the tip (yellow).
(D to G) Generation of flat sheet.
Initial [(D) and top of (G)] and final
[(F) and bottom of (G)] states are shown.
Adaxial (blue), abaxial (brown), and
midplane (green) domains are shown.
Orthoplanar polarity (black arrows) runs
from surface to midplane. Proximodistal
and orthoplanar polarity are shown
only on outer and cutaway surfaces,
respectively. Three growth rates
(Kpd, Kop, and Kper) are specified by
two polarity fields (E). (H and I) Generation
of an elliptic cylinder. Final state [(H) and
bottom of (I)] and initial state [top of (I)].
(J to L) Generation of a cup. Red arrowhead
indicates shifted position of sink for
proximodistal polarity. Initial [(J) and top
of (L)] and final [(K) and bottom of
(L)] states. (M to P) Generation of cup
with ridge. Initial [(M) and top of (O) and
(P)] and final [(N) and bottom of (O)
and (P)] states. (Q to S) 2D models showing
cell division patterns corresponding to
section levels shown in (M) and (N). Initial
(Q), intermediate (R), and final (S) states.
Periclinal division walls in ridge are indicated
by the arrow in (R). Levels of transverse
sections are indicated by color-coded
rectangles. Scale bars represent the same
length in all panels and are in arbitrary units.
(Kop), and perpendicular to both (Kper) (14) restricted the domain to a small region on one base of the primordium as a narrow ventral
(Fig. 4E). side of the primordium while keeping the strip, which grows out to form a ridge (5).
Setting Kop to be much lower than Kpd and same values of Kper, Kop, and Kpd as in the Incorporation of such a strip within the above
Kper generated a flattened sheet, correspond- planar leaf model (Fig. 4, J and L). Running model led to the formation of a cup with a ridge,
ing to a planar leaf (Fig. 4, F and G, and this model generated a cup-shaped sheet similar to the form observed in S. purpurea
movie S2). Similar results were obtained with the adaxial domain on the inner surface (Fig. 4, M to P, and movie S5). In S. purpurea,
using elliptical or elongated initial primor- (Fig. 4, K and L, and movie S4). This result cell divisions were observed to be preferen-
dium shapes, illustrating the robustness of suggests that those primordia showing re- tially periclinal (new cell walls parallel to the
the model (fig. S7). Low Kper as well as Kop stricted PHV1 expression at early stages (Fig. outer surface of the primordium) in the ridge-
generated a tapering elliptic cylinder, corre- 2A) are fated to form traps. The model also forming region but not in the more distal
sponding to a leaflet of U. gibba (Fig. 4, H and predicts that expressing the adaxial domain cup-forming region (5). Fukushima et al. in-
I, and movie S3). Thus, the formation of a throughout the primordium eliminates the corporated these observations into a two-
planar leaf can be accounted for by having orthoplanar polarity field, generating a taper- dimensional (2D) model of a transverse section
relatively low Kop, and the formation of a ing cylinder (Fig. 4B), consistent with the of the primordium, with small outgrowths
filiform leaf by both low Kop and low Kper. results of ectopic PHV1 expression (Fig. 3 and prespecified at the ad-ab boundary. Assum-
To determine the effect of a more confined fig. S6). ing divisions are induced near the epider-
adaxial domain, as observed in some U. gibba In contrast to U. gibba, the adaxial domain mal ad-ab boundary and that the planes of
primordia (yellow arrowhead in Fig. 2A), we in trap primordia of S. purpurea extends to the division determine the orientation of growth,

this model could generate the observed 2D 13. K. R. Siegfried et al., Development 126, 4117–4128 University of Chinese Academy of Sciences Joint Ph.D. Training
shapes (5). (1999). Program and a National Natural Science Foundation of
14. R. Kennaway, E. Coen, Open Biol. 9, 190057 (2019). China grant (91631308). Author contributions: C.D.W. and
To determine whether our model would 15. K. Abley et al., Development 140, 2061–2074 (2013). B.G. performed biological experiments and data analysis
give similar division patterns, we modeled 16. C. Mansfield et al., Curr. Biol. 28, 2638–2646.e4 and contributed to conceptualization; J.C. performed
transverse sections of the primordium con- (2018). computational modeling and data analysis and contributed
17. H. Lin et al., Proc. Natl. Acad. Sci. U.S.A. 110, 366–371 to conceptualization; M.C. and C.P. performed plant
taining virtual cells (Fig. 4Q), with specified (2013). transformation; R.K. performed software development
growth oriented relative to an orthoplanar 18. M. Nakata et al., Plant Cell 24, 519–535 (2012). and computational modeling; K.L. developed U. gibba resources;
polarity field. The plane of cell division was 19. M. Vandenbussche et al., Plant Cell 21, 2269–2283 C.B. performed RNA sequencing and data analysis; M.Y.
(2009). performed biological experiments and data analysis; and
set by taking the shortest path through the E.C. supervised the research, performed funding acquisition,
20. M. J. Scanlon, R. G. Schneeberger, M. Freeling, Development
cell center (6, 21). Some cells were marked in 122, 1683–1691 (1996). and contributed to conceptualization. Competing interests:
white to allow clones to be visualized. Running 21. L. Errera, C. R. Acad. Sci. 103, 822–824 (1886). The authors declare no competing interests. Data and
22. T. Blein et al., Science 322, 1835–1839 (2008). materials availability: All data are available in the main
this model recapitulated the observed patterns
23. A. Hay, M. Tsiantis, Nat. Genet. 38, 942–947 (2006). text or the supplementary materials. Code is available at
of division (Fig. 4, R and S, and movies S6 and the following websites: https://github.com/JIC-Enrico-Coen/
24. E. E. Kuchen et al., Science 335, 1092–1096 (2012).
S7), with divisions in the ridge-forming region 25. A. Richardson, A. B. Rebocho, E. Coen, Plant Cell 28, GFtbox_CarnivorousTraps_2019 and http://cmpdartsvr3.cmp.
being preferentially periclinal (black arrow in 2079–2096 (2016). uea.ac.uk/wiki/BanghamLab/index.php/Software.
Fig. 4R). Thus, our model accounts for both 26. C. D. Whitewoods, E. Coen, Curr. Biol. 27, R910–R918
(2017).
trap morphogenesis and observed planes of di- science.sciencemag.org/content/367/6473/91/suppl/DC1
vision in S. purpurea. ACKN OWLED GMEN TS Materials and Methods
Figs. S1 to S10
Our findings provide a simple mechanistic We would like to thank B. and P. Steward at The Fly Trap Plants
Tables S1 to S6
explanation for the generation of planar, fili- and T. Bailey from the Carnivorous Plant Society for plants,
seeds, and advice. Funding: This work was supported by a
form, and cup-shaped leaves, through shifts in European Research Council grant (323028-CarnoMorph) and
Movies S1 to S10
expression domains or their effects on growth. Data S1
Biotechnology and Biological Sciences Research Council grants
Unlike previous models, growth is oriented by (BBS/E/J/000PR9787, BB/M023117/1, and BB/L008920/1)
awarded to E.C. C.B. was funded by grant 1243441 from the
a polarity field, anchored by ad-ab domains Biotechnology and Biological Sciences Research Council. B.G. 28 June 2019; accepted 12 November 2019
acting throughout the leaf (not just at the epi- was supported by a Long-Term Fellowship from the Federation Published online 21 November 2019
dermal boundary), consistent with observed of European Biochemical Societies. J.C. was supported by the 10.1126/science.aay5433
division patterns. The planes of division are
a result (emergent property) of, rather than
the cause of, oriented growth. Moreover, our
model does not depend on the primordium EVOLUTIONARY BIOLOGY
already having outgrowths in the regions
that form the lamina, showing that it can
break morphological symmetry rather than
Protein-coding changes preceded cis-regulatory gains
simply elaborating it.
Additional structures, such as petioles, can
in a newly evolved transcription circuit
be generated by introducing further domains Candace S. Britton1,2, Trevor R. Sorrells1,2*, Alexander D. Johnson1†
into the model (figs. S8 to S10 and movies S8 to
S10). Diverse shapes and patterns of dissection Changes in both the coding sequence of transcriptional regulators and in the cis-regulatory
in the outline of planar leaves may also be gen- sequences recognized by these regulators have been implicated in the evolution of
erated through modulation of growth oriented transcriptional circuits. However, little is known about how they evolved in concert. We
by a proximodistal polarity field (22–25). Thus, describe an evolutionary pathway in fungi where a new transcriptional circuit (a-specific
a system in which regional identities modify gene repression by the homeodomain protein Mata2) evolved by coding changes in this
growth rates oriented by two orthogonal po- ancient regulator, followed millions of years later by cis-regulatory sequence changes
larity fields provides developmental flexibility in the genes of its future regulon. By analyzing a group of species that has acquired the
and can account for how cup-shaped forms coding changes but not the cis-regulatory sites, we show that the coding changes became
evolved multiple times independently from necessary for the regulator’s deeply conserved function, thereby poising the regulator to
species with planar leaves. jump-start formation of the new circuit.
C
1. J. R. McConnell, M. K. Barton, Development 125, 2935–2942 hanges in transcriptional circuits over though changes in cis-regulatory sequences
(1998). evolutionary time are an important are often stressed as sources of novelty that
2. R. Waites, A. Hudson, Development 121, 2143–2154 source of organismal novelty. Such cir- avoid extensive pleiotropy, it is clear that
(1995).
3. A. M. Ellison, N. J. Gotelli, J. Exp. Bot. 60, 19–42
cuits are typically composed of one or coding changes in the transcriptional regu-
(2009). more transcriptional regulators (sequence- latory proteins are also of key importance
4. Y. Hayakawa, M. Tachikawa, A. Mochizuki, J. Theor. Biol. 404, specific DNA binding proteins) and their di- (1–6). Some well-documented changes in tran-
206–214 (2016).
5. K. Fukushima et al., Nat. Commun. 6, 6450 (2015).
rect target genes, which contain cis-regulatory scriptional circuitry require concerted changes
6. S. Fox et al., PLOS Biol. 16, e2005952 (2018). sequences recognized by the regulators. Al- in both elements (7, 8). Although such con-
7. K. J. I. Lee et al., PLOS Biol. 17, e3000427 (2019). certed changes are likely to be widespread, we
8. R. S. Poethig, I. M. Sussex, Planta 165, 170–184
1 know little about how they occur.
(1985). Department of Microbiology and Immunology, University
9. E. Ibarra-Laclette et al., Nature 498, 94–98 (2013). of California, San Francisco, CA 94158, USA. 2Tetrad In this work, we study a case in the fungal
10. T. Lan et al., Proc. Natl. Acad. Sci. U.S.A. 114, E4435–E4441 Graduate Program, University of California, San Francisco, lineage where gains in cis-regulatory sequences
(2017). CA 94158, USA. and coding changes in the transcriptional reg-
11. J. R. McConnell et al., Nature 411, 709–713 *Present address: Laboratory of Neurogenetics and Behavior, The
(2001). Rockefeller University, New York, NY 10065, USA. ulator were both required for a new circuit to
12. J. F. Emery et al., Curr. Biol. 13, 1768–1774 (2003). †Corresponding author. Email: ajohnson@cgl.ucsf.edu have evolved. Specifically, we addressed which

A a cells cells a/ cells

asgs asgs asgs
sgs sgs sgs

Mata2
a cells
hsgs hsgs hsgs Mcm1
asgs Mata1 Mat 2
a/ cells
Mat 2
hsgs
B
cells
Mcm1
asgs
Mata2
a cells
Mcm1
asgs Mata1 Mat 2 Saccharomyces cerevisiaea
a/ cells
E Kluyveromyces lactis S. cerevisiae clade
D C Kluyveromyces wickerhamii (Saccharomycetaceae)
hsgs B ? Lachancea kluyveri
cells
Wickerhamomyces anomalus
Wickerhamomyces ciferrii W. anomalus clade
asgs Cyberlindnera jadinii (Phaffomycetaceae)
A Cyberlindnera fabianii
Candida albicans C. albicans clade
A Ancestral state (Pichiaceae &
B Gain of Mat 2-Mcm1 cis-regulatory sites in asg promoters Pichia membranifaciences
Yarrowia lypolytica Debaryomycetaceae)
C Gain of Mat 2's Mcm1-interaction region
Lipomyces starkeyi
D Gain of Mat 2's Tup1-interaction region
E Earliest known ancestor in which Mat 2 represses asgs ~300 mya Present
Fig. 1. Cell type–specific gene expression in the Saccharomycotina yeast. species. In the last common ancestor of the Saccharomycotina yeast (see circled
(A) Across the Saccharomycotina clade, a and a cells each express a set of A in the figure), transcription of the a-specific genes was activated by Mata2,
genes specific to that cell type (a- and a-specific genes, or asgs and asgs, a protein produced only in a cells, which binds directly to the regulatory region of
respectively), as well as a shared set of haploid-specific genes (hsgs). a and a each a-specific gene (10, 23). Much later in evolutionary time (see circled E in
cells can mate to form a/a cells, which do not express the a-, a-, or haploid- the figure), repression of the a-specific genes by direct binding by Mata2 evolved.
specific genes (22). Wavy arrows represent active transcription. (B) The Still later, the Mata2-positive form of control was lost in some species (including
mechanism underlying the expression of a-specific genes is different among S. cerevisiae), leaving only the Mata2-negative form. mya, million years ago.
came first: the changes in the regulatory pro- the S. cerevisiae clade) but after the divergence (12, 13). We chose the species Wickerhamomyces
tein or the changes in the cis-regulatory se- of this clade and that containing Candida anomalus, and we were able to optimize
quences of its 5 to 10 target genes. The system albicans and Pichia membrifaciens (formally relatively simple procedures to alter it ge-
we analyzed consists of an ancient regulator, known as the Pichiaceae and Debaryomece- netically (14).
the homeodomain protein Mata2, and the taceae, here called the C. albicans clade) (Fig. We examined the W. anomalus Mata2 pro-
changes—both in the protein itself and in the 1B). Three events must have occurred for the tein sequence to determine whether it is
regulatory regions of the genes it controls—that newer (repression) scheme to have evolved: more similar to the ancestral (represented by
occurred across the Saccharomycotina clade of (i) Mata2 acquired the ability to contact the C. albicans) or the derived (represented by
fungi, which spans roughly 300 million years. Tup1-Ssn6 co-repressor, bringing it to DNA S. cerevisiae) form of Mata2. Alignment of
[In terms of protein diversity, this represents to carry out the repression function; (ii) Mata2 the Mata2 coding sequences across many
roughly the range between humans and sea acquired the ability to bind to DNA coopera- species indicated that, of the five functional
sponges (9)]. Throughout this time, Mata2 tively (through a direct protein-protein con- regions described for the S. cerevisiae protein
has maintained its ancient function: It binds tact) with Mcm1; and (iii) the a-specific genes (Fig. 2A and fig. S1), the W. anomalus protein
cooperatively to DNA with a second homeo- (numbering between 5 and 10, depending on shares all of them. In particular, it has a
domain protein, Mata1, to repress a group the species) each acquired a new cis-regulatory similar Tup1-interacting region (region 1, Fig.
of genes called the haploid-specific genes site for the Mata2-Mcm1 combination (Fig. 1B). 2A) and Mcm1-interacting region (region 3,
(Fig. 1). More recently, Mata2 formed an To determine the order of these events, Fig. 2A); these regions are missing in out-
additional circuit, which is present in only a we studied Mata2 and the regulation of the group proteins and are needed to repress
subset of the Saccharomycotina: It binds a-specific genes in a clade that branched the a-specific genes in S. cerevisiae (11, 15). By
DNA cooperatively with the MADS box pro- from the ancestor before the occurrences of swapping these W. anomalus regions into the
tein Mcm1 to repress the a-specific genes (Fig. all three of these events. We reasoned that S. cerevisiae protein, we confirmed that they
1). Before this time, the a-specific genes were this group of species might have acquired are functional in repressing the a-specific genes
regulated by a different mechanism—positive some, but not all, of the changes needed to (Fig. 2B). In the course of these experiments,
control by the HMG-domain protein Mata2 form the new circuit, and it therefore might we found that the homeodomain of the
(10, 11). provide clues to the evolutionary history. This W. anomalus protein contained mutations
The switch between the two mechanisms of approach was made possible by the genome that prevented its binding to the a-specific
controlling the a-specific genes occurred some- sequencing of a monophyletic group of spe- gene cis-regulatory sequence in S. cerevisiae,
time before the divergence of Saccharomyces cies that branches before the last common a derived change within this clade alone (Fig.
cerevisiae and Kluyveromyces lactis (formally ancestor of the S. cerevisiae clade (formal- 2B and fig. S1). Similar results were obtained
known as the Saccharomycetaceae, here called ly known as the Phaffomycetaceae) (Fig. 1B) with the Mata2 protein from two additional

Fig. 2. W. anomalus A S. cerevisiae Mat 2 D W. anomalus mRNA-seq

Mata2 has func- Regions: 1 2 3 4 5 MAT vs. MAT mat 2-Δ
tional Tup1- and
Mcm1-interacting
MAT mat 2-Δ (tpm + 1)

regions but does not
Interactions: Tup1 Mata1 Mcm1 DNA Mata1
repress the
a-specific genes.
B Expression of PCYC1-GFP reporter asgs
(A) The five modules
with asg cis-regulatory site MAT 2
of the S. cerevisiae
mat 2-Δ Nat
Mata2 protein. Struc-
tural domains are S. cer Mat 2
shown as globular, and W. ano Mat 2
unstructured regions
Mat 2 chimeras
S. cer/ W. ano
are shown as wavy
lines. (B) Expression
of an a-specific E
gene reporter in 10000
Transcript quantity (Nanostring Counts)

MATa
the presence of
MATa mat a2- Δ
S. cerevisiae (S. cer) 8000
00
0
00
00
00
Mata2 (purple),
50
10
15
20
W. anomalus (W. ano) GFP fluorescence, arbitrary units
6000
Mata2 (green), and
hybrid proteins (purple C Mating efficiency (%)
and green). Means Genotype of 4000
limiting parent Experiment 1 Experiment 2
and SDs of three
independent genetic MAT 76 94 2000
isolates, grown and MAT mat 2-Δ1 96 90
tested in parallel, are MAT mat 2-Δ2 86 100
0
a 2 E2
shown. GFP, green
E6 XL1 FA 1 7 E3 TE4
MATa 100 71
fluorescent protein. MATa mata2-Δ <0.14 <0.16 AT ST ST A M BAR SG ST S
M A
(C) In W. anomalus,
Mata2, but not Mata2,
is required for a cells to mate (see supplementary text for details). (D) mRNA given in fig. S3. (E) a-specific gene expression levels in a wild-type W. anomalus a
sequencing (mRNA-seq) (tpm, transcripts per million) of wild-type W. anomalus cells (MATa) compared with a cells with MATa2 deleted (MATa2 mata2-D),
a cells (MATa) compared with a cells with MATa2 deleted (MATa2 mata2-D). measured by the NanoString nCounter system (24). For comparison, expression
a-specific genes STE2, AXL1, ASG7, BAR1, STE6, and MATa2 are shown in levels of the a-specific gene STE3 and the haploid-specific gene STE4 are
green. Expression of MATa2 and the marker used to delete it (Nat) are shown also given. Means and SDs of two cultures per genotype, grown and tested in
in pink and opaque black, respectively. Data from independent replicates are parallel, are shown.
species that branch with W. anomalus, in- a bioinformatic analysis showing that the only in the W. anomalus clade. To test this
dicating that these two conclusions—that a-specific genes possess Mata2-Mcm1, but idea, we analyzed the requirements for haploid-
W. anomalus clade Mata2 bears functional not Mata2-Mcm1 cis-regulatory sequences specific gene repression in W. anomalus. We
protein-protein interactions but cannot bind (fig. S4B). These results argue against the deleted MATa2 and MATa1 in a/a cells and
the S. cerevisiae a-specific genes—are charac- possibility that direct, a-specific gene repres- found that they are both necessary for haploid-
teristic of the W. anomalus clade rather than sion by Mata2 existed in an ancestor of specific gene repression, a conclusion con-
of a single species (fig. S1D). W. anomalus but was subsequently lost, as firmed by chromatin immunoprecipitation
The observation that the W. anomalus this would have required the independent (Fig. 3A and figs. S5 and S6C). However,
Mata2 protein acquired the necessary coding loss of Mata2 binding sites from all of the unlike in species outside the W. anomalus
changes to interact with Tup1 and Mcm1 but a-specific genes across numerous species. clade, the Tup1-interaction region and the
could not bind to the S. cerevisiae a-specific Our experiments up to this point demon- Mcm1-interaction region of Mata2 are nec-
gene control region raised the question of strate that Mata2 had acquired the coding essary for repression of the haploid-specific
whether it has any role in regulating the a- changes needed to repress the a-specific genes genes within the clade (Fig. 2A and fig. S6B).
specific genes in W. anomalus. A series of millions of years before its cis-regulatory se- Finally, an Mcm1 cis-regulatory site is also re-
otherwise-isogenic strains was constructed quences appeared in the a-specific genes. quired for the repression of the W. anomalus
with Mata2 (and Mata2) deleted, and the We next addressed how these changes in haploid-specific gene RME1 (Fig. 3C and
results show that, in this species, Mata2 does the Mata2 protein could have been main- fig. S6). Taken together, these experiments
not regulate the a-specific genes; they are tained in the absence of their usefulness in show that Mata2, Mata1, and Mcm1 are all
instead regulated by Mata2 (Fig. 2, C to E, repressing the a-specific genes. One hypoth- required for haploid-specific gene repression
and fig. S3). Thus, despite the changes in esis focuses on Mata2’s ancient function— in W. anomalus, and that the portions of
Mata2, W. anomalus retains the ancestral repressing the haploid-specific genes with Mata2 that interact with Mcm1 and Tup1 are
form of a-specific gene regulation and activa- Mata1—and holds that the Mata2 coding also required. This three-part recognition of
tion by Mata2. This conclusion is supported by changes became required for this function the haploid-specific genes in the W. anomalus

Fig. 3. Mata1, Mata2, and A W. anomalus mRNA-seq C

MATa/MAT vs. MATa/MAT mat 2-Δ Cell
an Mcm1 cis-regulatory PRME1 reporter type
MATa/MAT mat 2-Δ (tpm + 1)

sequence are all required
for haploid-specific gene a/ GFP
GFP
repression in W. anomalus. a RME1
(A) mRNA-seq of a wild-type a/
W. anomalus a/a cell hsgs GFP
a
(MATa/MATa) compared with sgs a/
an a/a cell with MATa2 asgs GFP
a
deleted (MATa/MATa MAT 2
mata2-D). The a-specific a/
Nat GFP
genes are shown in green, the a
haploid-specific genes in a/
GFP
orange, and the a-specific a
genes in blue. Data from one MATa/MAT (tpm + 1)
a/
GFP
culture of each genotype B a
are plotted here, and data W. anomalus PRME1
from replicates, grown and 0 1 2 3
RME1
prepared in parallel, and Transcript quantity relative to TBP1
-500 +1
similar results obtained by
Predicted Mcm1 binding site
deleting Mata1 are shown in Predicted Mata1/Mat 2 binding site
fig. S5. (B) Diagram of the
sequence upstream of the RME1 coding sequence indicating presumptive Mata1-Mata2 (green) and Mcm1 (blue) binding sites. Arrow indicates the transcription start
site. (C) Expression levels of endogenous RME1 transcript (which serves as a control) and various PRME1-GFP reporter constructs in W. anomalus a and a/a cells
measured by reverse transcription quantitative polymerase chain reaction. Quantities are means and SDs of two cultures grown and measured in parallel,
normalized to expression of the housekeeping gene TBP1. Independent replicates are given in fig. S6.
Mat 2 in extant species

Fig. 4. Order of evolutionary events leading to
repression of the a-specific genes by Mata2. The hsg regulation asg regulation
three-protein solution for repressing the haploid-
specific genes remains in the W. anomalus clade, but S. cerevisiae hsgs asgs
in the S. cerevisiae lineage it was partitioned into C
Mata1-Mat 2 Mat 2-Mcm1
a-specific gene regulation (which uses only two
B
proteins, Mcm1 and Mata2) and repression of the
haploid-specific genes (which requires Mata2 and W. anomalus hsgs No role
Mata1). The three-protein intermediate explains how A
Mata1-Mat 2-Mcm1
the necessary changes in the regulatory protein
Mata2 could have been maintained for millions of
years before being co-opted for the new circuit. C. albicans hsgs No role
A Ancestral state: Mat 2 regulates hsgs
B Gain of Mat 2's Mcm1- and Tup1-interaction regions Mata1-Mat 2
C Gain of Mat 2-Mcm1 binding sites at asgs
clade was not anticipated from studies of when transcriptional regulators acquire as being compatible with the ancestral func-
other species. Even in the S. cerevisiae clade, new functions? The modular structure of tion of Mata2.
where Mcm1 and Mata2 are known to inter- Mata2 is evident from the protein domain Second, is the evolutionary pathway we de-
act, this interaction is not required for haploid- swap experiments (Fig. 2B and fig. S6B), show- scribe in this paper compatible with the con-
specific gene repression (11). These results ing that the derived regions of the protein cept of constructive neutral evolution, or the
explain the observation that the key changes (Tup1- and Mcm1-interaction regions) can be idea that new functions can evolve through
in Mata2 needed for the new a-specific gene transplanted to a variety of outgroup Mata2 evolutionary transitions of approximately equal
circuit were already in place in the last common proteins and that they endow the ancestral fitness (16–18)? Before the results presented
ancestor of S. cerevisiae and W. anomalus, long proteins with the new functions without com- here were obtained, it was difficult to un-
before the circuit came into play (Fig. 4). An promising the existing functions (11). How- derstand how the derived circuit represented
alternative scenario—in which the Mata2 pro- ever, there is a second, more subtle way that by S. cerevisiae (repression of the a-specific
tein gained the Mcm1-interaction region twice, extensive pleiotropy was avoided in the case genes by Mata2 in a cells) could have evolved
once in the S. cerevisiae clade and once in the studied in this work. In the shift between the because it required changes in both the Mata2
W. anomalus clade—is unlikely because the different ways of controlling the haploid- coding region and in the cis-regulatory se-
same seven amino acids would have had to be specific genes, pleiotropy was avoided auto- quences controlling the 5 to 10 a-specific
gained in exactly the same position in the matically; even before the new a-specific gene genes. We propose that the prior changes to
protein (fig. S1). circuit was formed, the Mata2-Mcm1 combi- Mata2 represent an example of constructive
This study helps to illuminate several long- nation (which forms the basis of the new neutral evolution, in the sense that the neu-
standing issues. First, how is pleiotropy avoided circuit) had been “vetted” for millions of years tral sampling of different ways to repress the

haploid-specific genes over evolutionary time plasmid used to genetically modify W. anomalus. Funding: The SUPPLEMENTARY MATERIALS
led to changes in Mata2 that, millions of years work was supported by NIH grant R01 GM037049 (to A.D.J.), an science.sciencemag.org/content/367/6473/96/suppl/DC1
ARCS Scholarship (to C.S.B.), and an NSF Graduate Fellowship Materials and Methods
later through exaptation, formed the basis of (to T.R.S.). Author contributions: C.S.B., T.R.S., and A.D.J. Supplementary Text
the new circuit. Although we cannot rule out designed and interpreted experiments and wrote and edited the Figs. S1 to S6
the possibility that the differences in the way manuscript. Competing interests: The authors declare References (25–45)
no competing interests. Data and materials availability: Data S1
that the haploid-specific genes were repressed Plasmid pCS.DLig4 can be obtained from Evonik Industries
were somehow adaptive, it seems more likely under a material transfer agreement. mRNA-seq data have
that they occurred neutrally—an explanation been deposited at the National Center for Biotechnology
Information Gene Expression Omnibus under accession 1 April 2019; accepted 5 November 2019
consistent with a wide variety of theoretical number GSE133191. 10.1126/science.aax5217
work (16–19). In any case, there is no obvious
adaptive explanation, and neutral evolution is
an appropriate default hypothesis.
Third, is there an inherent logic to the TUBULIN
mechanisms underlying a given transcription
circuit? In this paper, we show that some
clades regulate the haploid-specific genes
TTC5 mediates autoregulation of tubulin via
with a combination of three proteins, whereas
others use only two of the proteins, even though
mRNA degradation
the third is present. Nonetheless, the overall Zhewang Lin1, Ivana Gasic2*, Viswanathan Chandrasekaran1*, Niklas Peters1†, Sichen Shao3,
pattern of haploid-specific gene expression Timothy J. Mitchison2, Ramanujan S. Hegde1‡
is the same. If there is any overriding design
logic to the different mechanisms of regulat- Tubulins play crucial roles in cell division, intracellular traffic, and cell shape. Tubulin
ing these genes, it is difficult to discern (20). concentration is autoregulated by feedback control of messenger RNA (mRNA) degradation
More broadly, the work presented here illus- via an unknown mechanism. We identified tetratricopeptide protein 5 (TTC5) as a tubulin-specific
trates that a given transcription circuit is best ribosome-associating factor that triggers cotranslational degradation of tubulin mRNAs in
understood as one of several possible inter- response to excess soluble tubulin. Structural analysis revealed that TTC5 binds near the
changeable, mechanistic solutions rather than ribosome exit tunnel and engages the amino terminus of nascent tubulins. TTC5 mutants
as a finished, optimized design (21). incapable of ribosome or nascent tubulin interaction abolished tubulin autoregulation and
showed chromosome segregation defects during mitosis. Our findings show how a subset of
RE FE RENCES AND N OT ES mRNAs can be targeted for coordinated degradation by a specificity factor that recognizes the
1. V. J. Lynch, G. P. Wagner, Evolution 62, 2131–2154 nascent polypeptides they encode.
(2008).
A
2. A. M. Cheatle Jarvela, V. F. Hinman, EvoDevo 6, 3
(2015).
3. D. L. Stern, V. Orgogozo, Evolution 62, 2155–2177 lpha and beta tubulins form obligate het- sion of the MREI motif (12), a factor is thought
(2008). erodimers (hereafter ab-tubulin) that to engage this sequence on nascent tubulin
4. G. A. Wray, Nat. Rev. Genet. 8, 206–216 (2007). reversibly and dynamically polymerize to initiate degradation of the mRNA being
5. P. J. Wittkopp, G. Kalay, Nat. Rev. Genet. 13, 59–69
(2011).
into microtubules—cytoskeletal elements translated.
6. H. Li, A. D. Johnson, Curr. Biol. 20, R746–R753 (2010). that regulate cell shape, drive mitosis, We used a site-specific photo–cross-linking
7. C. Sayou et al., Science 343, 645–648 (2014). provide platforms for intracellular transport, strategy (Fig. 1A and fig. S1) to detect cyto-
8. C. R. Baker, B. B. Tuch, A. D. Johnson, Proc. Natl. Acad.
Sci. U.S.A. 108, 7493–7498 (2011).
and mediate cell movement (1). Microtubule solic factors that specifically recognize the
9. X.-X. Shen et al., Cell 175, 1533–1545.e20 (2018). dynamics, and the various processes that de- N-terminal autoregulatory motif (MREI) of
10. A. E. Tsong, M. G. Miller, R. M. Raisner, A. D. Johnson, Cell 115, pend on it (2, 3), is strongly influenced by the nascent b-tubulin early during its transla-
389–399 (2003).
concentration of soluble (i.e., nonpolymerized) tion. A ribosome–nascent chain complex
11. C. R. Baker, L. N. Booth, T. R. Sorrells, A. D. Johnson, Cell 151,
80–95 (2012). ab-tubulin (4). When cells detect an increase (RNC) displaying the first 94 amino acids of
12. X.-X. Shen et al., G3 6, 3927–3939 (2016). in soluble ab-tubulin concentration, they trig- [35S]methionine-labeled human b-tubulin con-
13. R. Riley et al., Proc. Natl. Acad. Sci. U.S.A. 113, 9882–9887 ger degradation of tubulin mRNAs via a process taining the ultraviolet light (UV)–activated cross-
(2016).
14. C. Kurtzman, J. W. Fell, Eds., The Yeasts - A Taxonomic Study termed tubulin autoregulation (5–7). linking amino acid p-benzoyl-L-phenylalanine
(Elsevier, 1998). Autoregulation requires translation, indicat- (Bpa) was produced by in vitro translation in
15. K. Komachi, A. D. Johnson, Mol. Cell. Biol. 17, 6023–6028 ing that ribosome-engaged tubulin mRNAs rabbit reticulocyte lysate. Irradiation of these
(1997).
16. A. Stoltzfus, J. Mol. Evol. 49, 169–181 (1999). are selectively targeted for degradation (8, 9). RNCs with UV light generated nascent chain
17. J. Lukeš, J. M. Archibald, P. J. Keeling, W. F. Doolittle, Analysis of b-tubulin autoregulation in mam- cross-links to various proteins, only one of
M. W. Gray, IUBMB Life 63, 528–537 (2011). malian cells indicates a critical role for the first which was sensitive to mutation of residues 2,
18. M. W. Gray, J. Lukes, J. M. Archibald, P. J. Keeling,
W. F. Doolittle, Science 330, 920–921 (2010).
four residues (Met-Arg-Glu-Ile, or MREI) com- 3, and 4 of the MREI motif (Fig. 1B). This
19. A. Wagner, FEBS Lett. 579, 1772–1778 (2005). mon to all b-tubulin isoforms (10, 11). Because MREI-specific interaction partner was iden-
20. C. K. Dalal, A. D. Johnson, Genes Dev. 31, 1397–1405 autoregulation is prevented by physical occlu- tified by quantitative mass spectrometry to
(2017).
21. T. R. Sorrells, A. D. Johnson, Cell 161, 714–723 (2015).
be TTC5 (Fig. 1C), a highly conserved protein
22. I. Herskowitz, Nature 342, 749–757 (1989). found widely across eukaryotes (fig. S2). TTC5
23. A. E. Tsong, B. B. Tuch, H. Li, A. D. Johnson, Nature 443, 1
MRC Laboratory of Molecular Biology, Cambridge CB2 0QH, engaged the MREC motif at the N terminus
415–420 (2006). UK. 2Department of Systems Biology, Blavatnik Institute,
24. G. K. Geiss et al., Nat. Biotechnol. 26, 317–325 (2008). of nascent a-tubulin comparably to the MREI
Harvard Medical School, Boston, MA 02115, USA.
3
Department of Cell Biology, Blavatnik Institute, Harvard motif on b-tubulin (Fig. 1D), consistent with
ACKN OW LEDG MEN TS Medical School, Boston, MA 02115, USA. position 4 being less critical than positions 2
We thank L. Noiman, M. Lohse, C. Dalal, K. Fowler, and *These authors contributed equally to this work. †Present address: or 3 (Fig. 1B) (11). Thus, TTC5 is a nascent
L. Booth for comments on the manuscript and C. Baker, Center for Molecular Biology of Heidelberg University (ZMBH),
I. Nocedal, N. Ziv, and B. Heineke for advice. We thank 69120 Heidelberg, Germany. polypeptide binding protein specific for the
C. Schorsch of Evonik Industries for providing us with the ‡Corresponding author. Email: rhegde@mrc-lmb.cam.ac.uk N termini of a- and b-tubulins.

Fig. 1. TTC5 interacts with the

N termini of nascent tubulins.
(A) Experimental strategy to detect
interaction partners close to the
N terminus of nascent tubulin. The
UV-activated cross-linking amino acid
p-benzoyl-L-phenylalanine (Bpa) is
introduced site-specifically at position 7
using amber suppression (see fig. S1).
IVT, in vitro translation; XL, cross-linking;
PAGE, polyacrylamide gel electrophoresis.
(B) Photo–cross-linking analysis of
35
S-labeled 94–amino acid ribosome–
nascent chain complexes (RNCs) of
human b-tubulin and mutants (indicated
in red) in the N-terminal MREI motif.
The positions of non–cross-linked tRNA-
associated nascent chain (NC-tRNA)
and a cross-linking partner specific to
wild-type tubulin (red asterisk) are
indicated. Other nascent chain cross-links
agnostic to the MREI motif are indicated
by black asterisks. (C) Quantitative
mass spectrometry of proteins copurified
with wild-type (WT) versus MHQV
mutant (mut) b-tubulin RNCs plotted
by molecular weight. (D) Photo–cross-
linking and immunoprecipitation (IP)
analysis of 35S-labeled 94–amino acid RNCs of a- or b-tubulin compared to the indicated mutants. Amino acid abbreviations in this or later figures: A, Ala; C, Cys; D, Asp;
E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; M, Met; Q, Gln; R, Arg; V, Val.
Fig. 2. Mechanism of ribosome–nascent

chain engagement by TTC5. (A) Overview
of the cryo-EM–derived structure of the
complex between TTC5 and a ribosome
containing the first 64 amino acids of b-tubulin.
(B) Close-up view of the TTC5 interaction
with 28S rRNA. Three conserved Lys residues
in TTC5 within electrostatic interaction distance
of the rRNA backbone are indicated. (C) The
surface of TTC5 that interacts with ribosomal
protein uL24 is indicated in orange. The
28S-interacting residues from (B) are shown
in yellow. The N-terminal eight amino acids
of the b-tubulin nascent chain are shown in red
within the chain’s binding groove of TTC5.
(D) Close-up view of the N-terminal eight amino
acids of nascent b-tubulin (MREIVHIQ) within
TTC5. Yellow spheres denote C-b atoms for the
indicated side chains (not modeled). Arg147 is
within salt-bridge distance of Glu3, and Asp225
and Glu259 are within salt-bridge distance of
Arg2. Phe194 on the “floor” of the binding
groove, shown in Fig. 3A to cross-link with
nascent b-tubulin, is indicated.

Fig. 3. Avidity-based RNC binding imparts specificity

to TTC5. (A) Photo–cross-linking analysis of
35
S-labeled 94–amino acid RNCs of human b-tubulin
and N-terminal mutants (indicated in red) with
recombinant StrepII-tagged TTC5 containing the
photo–cross-linking residue Bpa at position Phe194.
The nascent chain cross-link to TTC5 is indicated
(TTC5-XL) and verified by pulldown via the StrepII tag
(bottom panel). (B) Photo–cross-linking analysis
using 35S-labeled 64–amino acid RNCs of human
b-tubulin or the N-terminal MRQI mutant. Wild-type
or mutant recombinant StrepII-tagged TTC5 was
included in the assay as indicated. The photo–cross-
linking residue Bpa is at position 7 of the b-tubulin
nascent chain. An aliquot of the total translation
reaction was analyzed to verify equal levels of nascent
chain (NC) synthesis by autoradiography and equal levels
of recombinant TTC5 (SII-TTC5) by immunoblotting
for the StrepII tag. The remainder was UV-irradiated,
and TTC5 cross-links were recovered via the
StrepII tag and visualized by autoradiography
(bottom panel). (C) Summary of interaction analysis
between the indicated recombinant TTC5 proteins
and the indicated synthetic peptides in a thermal shift
denaturation assay (see fig. S7). (D) Wild-type or
mutant StrepII-tagged TTC5 was included during in
vitro translation of wild-type or mutant 64–amino
acid b-tubulin RNCs as indicated. Equal translation of
35
S-labeled nascent chain synthesis was verified
(NC total). The remainder of each translation was
affinity-purified via the StrepII tag and analyzed by
staining of total proteins (bottom panel).
To understand how TTC5 engages its sub- consistent with photo–cross-linking results (Fig. 3C and fig. S7) verified these findings and
strates on the ribosome, we purified nascent (fig. S6). additionally showed that Met1 of nascent
tubulin RNCs in complex with TTC5 (fig. S3) The structural model allowed us to deduce tubulin is critical for TTC5 binding and must
and determined the structure of this complex likely interactions between the MREI motif strictly be at the N terminus. Thus, the struc-
by single-particle cryo–electron microscopy and conserved side chains lining the TTC5 tural analysis rationalizes all earlier b-tubulin
(cryo-EM). The TTC5-RNC reconstruction groove (Fig. 2D). Depending on its orientation, mutagenesis studies on autoregulation require-
(figs. S4 and S5) showed the ribosome with a Arg2 of nascent tubulin is within salt-bridge ment (11) and reveals the mechanistic basis of
peptidyl-tRNA, a nascent b-tubulin polypeptide distance of Glu259 and Asp225 of TTC5. Glu3 the exquisite specificity of autoregulation for
within the ribosome exit tunnel, and TTC5 in nascent tubulin would likely interact with a- and b-tubulins (5) that uniquely contain an
bound at the mouth of the tunnel (Fig. 2A). The Arg147 in TTC5. Ile4 faces a moderately hy- MREI or MREC motif at the N terminus (17).
heterodimeric nascent polypeptide–associated drophobic surface that could accommodate Mutating the ribosome-interacting residues
complex (NAC) was observed at its previous- a cysteine (as in a-tubulins) or possibly other Lys285 and Lys287 of TTC5 to glutamic acid
ly established binding site (13, 14) opposite amino acids, consistent with earlier muta- (KK-EE) completely abolished b-tubulin RNC
the exit tunnel from TTC5 (see fig. S4). NAC genesis (11). Collectively, the structure shows binding in the cross-linking assay (Fig. 3B),
is not specific to tubulin RNCs (15, 16), does how TTC5 binds near the ribosome exit tun- despite unperturbed binding of TTC5 to syn-
not contact TTC5 in the structure, and is not nel with its peptide-binding groove positioned thetic tubulin autoregulatory peptide in a ther-
discussed further. to engage nascent tubulins shortly after they mal shift assay (Fig. 3C and fig. S7). Affinity
TTC5 was seen to make two contacts with emerge from the ribosome. purification of recombinant TTC5 from in vitro
the ribosome. The first contact involves three Recombinant TTC5 containing Bpa at posi- translation reactions of nascent b-tubulin RNCs
highly conserved lysine side chains in the tion 194 in the “floor” of the peptide binding showed that no ribosomes were recovered with
oligonucleotide-binding domain of TTC5 groove (Fig. 2D) efficiently cross-linked with either TTC5(KK-EE) or the peptide-binding
making electrostatic interactions with phos- MREI-containing nascent chains, weakly cross- mutant TTC5(R147A), in contrast to wild-type
phates of the 28S rRNA backbone (Fig. 2B). linked with MREV-containing nascent chains, TTC5 (Fig. 3D). Thus, the avidity of bipartite
The second contact involves ribosomal protein and did not form cross-links with any other binding to the ribosome and nascent tubulin
uL24 and buries ~500 Å2 of TTC5 adjacent to mutants (Fig. 3A). Analysis of RNC cross-linking imparts high affinity and specificity to the
a deep groove formed by the tetratricopeptide with various TTC5 mutants (Fig. 3B) validated TTC5-RNC interaction.
repeat domain of TTC5 (Fig. 2C). The groove Arg147, Asp225, and Glu259 as key residues within CRISPR-mediated disruption of TTC5 ex-
faces the mouth of the exit tunnel and con- the groove that likely interact with Arg2 and Glu3 pression in multiple cell lines completely abol-
tains cryo-EM density that we assigned to the of nascent tubulin (see Fig. 2D). Binding assays ished the decay of a- and b-tubulin mRNAs in
first eight amino acids of b-tubulin (fig. S5), with purified TTC5 and synthetic peptides response to acute microtubule destabilization

Fig. 4. TTC5 is required for tubulin autoregulation and accurate mitosis. the indicated times with colchicine and used to prepare lysates. One of the
(A) The indicated human embryonic kidney (HEK) 293 cell lines were either control samples included colchicine added after cell lysis (indicated as 0*). The
left untreated or treated for 3 hours with colchicine. The relative amounts of the products recovered by binding to recombinant TTC5 were analyzed for b-tubulin
indicated mRNAs or pre-mRNAs were quantified by reverse transcription mRNA by RT-qPCR. The relative recoveries are plotted (means ± SD from
quantitative polymerase chain reaction (RT-qPCR) and normalized to a control three replicates). Similar results were seen for a-tubulin and when nocodazole
ribosomal RNA. Data are means ± SD from three replicates. Similar results was used instead of colchicine to trigger autoregulation (fig. S11). (D) Diagram
are seen in HeLa cells and with different microtubule-destabilizing agents (left) and examples (right) of accurate (top) and erroneous (bottom)
(fig. S8). (B) Preformed RNC-TTC5 complexes (see fig. S10) were mixed with chromosome alignment and segregation visualized with SirDNA dye during
buffer or cytosol from TTC5 knockout cells that had been pretreated (+col) mitosis in HeLa cell lines (see fig. S13). Scale bar, 5 mm. (E) Quantification
or not (–col) with colchicine for 3 hours. All samples were subjected to UV of errors in chromosome alignment and segregation in the indicated HeLa
cross-linking to monitor the nascent chain interactions. The positions of cell lines. Data are means ± SEM from four to six independent biological
the non–cross-linked tRNA linked nascent chain (NC-tRNA) and TTC5 cross-link replicates (dots) with 200 to 400 analyzed cells per replicate. **P < 0.001
(TTC5-XL) are indicated. (C) TTC5-knockout HEK293 cells were pretreated for (paired Student t test); n.s., not significant.
(Fig. 4A and fig. S8). Pulse labeling of wild- bulin at the ribosome is strictly required for vated when cells perceive excess ab-tubulin,
type cells with [35S]methionine showed that of tubulin mRNA degradation when cells initiate freeing TTC5 to engage tubulin RNCs and trig-
the major proteins visualized, tubulins were autoregulation. Access of TTC5 to the ribosome ger tubulin mRNA degradation. TTC5’s access
selectively reduced in their synthesis when cells proved to be a regulated event. to RNCs only during autoregulation explains
are pretreated with microtubule-destabilizing The TTC5-RNC complex was found to be why normally growing TTC5 knockout cells
agents (fig. S9). Selective reduction in tubulin disrupted by a cytosolic factor whose activity did not show notably elevated tubulin mRNA
protein synthesis was completely lost in TTC5 was lost when cells were pretreated with col- and protein (fig. S12). Because overexpressed
knockout cells, consistent with the failure to chicine to initiate autoregulation (Fig. 4B and TTC5 in the rescue cell lines did not trigger
degrade tubulin mRNAs. Tubulin autoregula- fig. S10). Loss of this inhibitory activity under tubulin mRNA degradation until cells perceived
tion, as judged by both mature mRNA versus autoregulation conditions was accompanied excess ab-tubulin (fig. S12), it seems that the
pre-mRNA levels (Fig. 4A) and rates of protein by increased capacity of TTC5 to engage tu- inhibitor is not easily saturated. Further work
synthesis (fig. S9), could be restored to TTC5 bulin RNCs as measured by recovery of tubulin will be necessary to identify the inhibitor and
knockout cells by reexpression of wild-type mRNAs (Fig. 4C and fig. S11). These results its mechanism of regulation.
TTC5 but not the peptide-binding mutant indicate that cells ordinarily contain an inhib- Chromosome alignment and segregation
R147A or the ribosome-binding mutant KK- itory factor that prevents TTC5 engagement of during mitosis are sensitive to altered micro-
EE. Thus, TTC5 engagement of nascent tu- tubulin RNCs. This TTC5 inhibitor is inacti- tubule dynamics (18–21), motivating us to

monitor these parameters in cells impaired in 14. R. D. Wegrzyn et al., J. Biol. Chem. 281, 2847–2857 (2006). Harvard Medical School for help with light microscopy.
tubulin autoregulation (Fig. 4D). TTC5 knock- 15. B. Wiedmann, H. Sakai, T. A. Davis, M. Wiedmann, Nature 370, Funding: Supported by UK Medical Research Council grant
434–440 (1994). MC_UP_A022_1007 (R.S.H.), NIH grant P50 107618 (T.J.M.),
out HeLa cells showed a higher rate of chro- 16. M. Gamerdinger, M. A. Hanebuth, T. Frickey, E. Deuerling, Harvard Medical School and Vallee Scholars Program (S.S.),
mosome alignment errors in metaphase (by a Science 348, 201–207 (2015). and a Human Frontier Science Program postdoctoral fellowship
factor of ~6.5) (Fig. 4E and fig. S13), a higher 17. UniProt Consortium, Nucleic Acids Res. 46, 2699 (2018). (Z.L.). I.G. is a Merck Fellow of the Damon Runyon Cancer
18. J. J. Vicente, L. Wordeman, Curr. Opin. Cell Biol. 60, 36–43 Research Foundation (DRG:2279-16). V.C. was supported
rate of segregation errors during anaphase (2019). by V. Ramakrishnan whose funding was from the MRC
(by a factor of ~2.4) (Fig. 4E and fig. S13), 19. S. Petry, Annu. Rev. Biochem. 85, 659–683 (2016). (MC_U105184332), the Wellcome Trust (WT096570), the
and a subtle but highly reproducible increase 20. S. L. Prosser, L. Pelletier, Nat. Rev. Mol. Cell Biol. 18, 187–201 Agouron Institute, and the Louis-Jeantet Foundation. Author
(2017). contributions: Z.L. discovered TTC5 and performed all
of mitotic duration (fig. S14). These pheno- 21. S. L. Kline-Smith, C. E. Walczak, Mol. Cell 15, 317–327 (2004). biochemical analyses; I.G. and S.S. generated and validated HeLa
types in TTC5 knockout cells were rescued by 22. J. T. Lynch, T. D. D. Somerville, G. J. Spencer, X. Huang, cell lines; I.G. designed and performed phenotypic analysis of
reexpression of wild-type TTC5 but not the T. C. P. Somervaille, Cell Death Dis. 4, e573 (2013). mitosis; V.C. performed structural analysis of the TTC5-ribosome
23. C. Demonacos, M. Krstic-Demonacos, N. B. La Thangue, complex; N.P. set up and characterized the in vitro photo–
peptide-binding or ribosome-binding mutants
Mol. Cell 8, 71–84 (2001). cross-linking system; S.S., T.J.M., and R.S.H. supervised different
of TTC5 (Fig. 4E and figs. S13 and S14). Al- 24. X. Hu, R. D. Mullins, J. Cell Biol. 218, 251–266 (2019). aspects of the project; R.S.H. and Z.L. conceived the project,
though the specific basis of mitotic defects in 25. N. Sonenberg, A. G. Hinnebusch, Cell 136, 731–745 (2009). oversaw its implementation, and wrote the manuscript. All
TTC5 knockout cells remains to be determined, 26. J. C. Darnell, E. Klann, Nat. Neurosci. 16, 1530–1536 authors contributed to manuscript editing. Competing interests:
(2013). The authors declare no competing interests. Data and materials
we can ascribe the phenotypes to autoregula- 27. O. Brandman, R. S. Hegde, Nat. Struct. Mol. Biol. 23, 7–15 availability: The cryo-EM map has been deposited to the EMDB
tion, and not another TTC5 function (22–24), (2016). (EMD-10380) and atomic coordinates have been deposited to the
because the effects were not rescued by two 28. C. J. Shoemaker, R. Green, Nat. Struct. Mol. Biol. 19, 594–601 Protein Data Bank (PDB 6T59). All other data are available in the
(2012). manuscript or the supplementary materials.
unrelated point mutants of TTC5 that perturb 29. W. F. Marzluff, K. P. Koreski, Trends Genet. 33, 745–759
autoregulation by different mechanisms. (2017). SUPPLEMENTARY MATERIALS
TTC5 represents a highly selective and reg- 30. B. J. DiDomenico, G. E. Bugaisky, S. Lindquist, Cell 31,
593–603 (1982).
ulated ribosome-associating factor that only Materials and Methods
Figs. S1 to S15
engages the ~2 to 3% of a cell’s ribosomes that ACKN OWLED GMEN TS
Table S1
actively synthesize a- and b-tubulins. By mark- We thank V. Ramakrishnan for support and advice; S.-Y. Peak-Chew
and M. Skehel for mass spectrometry analysis; J. Grimmett
ing tubulin-translating ribosomes, TTC5 is and T. Darling for advice, data storage, and high-performance
ideally positioned to recruit yet unidentified computing; P. Emsley for advice; the MRC Laboratory of
downstream effectors to this site that trigger Molecular Biology EM Facility for microscopy support, sample 9 September 2019; accepted 31 October 2019
preparation, and data collection; B. Raught and W. Harper for Published online 14 November 2019
mRNA decay. More generally, the translating
Flp-In TRex HeLa cells; and the Nikon Imaging Center at 10.1126/science.aaz4352
ribosome represents a platform from which to
effect abundance control of key cellular pro-
teins because translation initiation (25), elon-
gation (26), polypeptide fate (27), and mRNA TOPOLOGICAL MATTER
stability (28) can all be locally regulated from
this site. Specificity for particular substrates
would be imparted by recognition of the nascent
Evidence for dispersing 1D Majorana channels in an
protein emerging from the ribosome exit
tunnel. Thus, cells may contain a family of
iron-based superconductor
substrate-specific ribosome-associating fac- Zhenyu Wang1,2, Jorge Olivares Rodriguez1, Lin Jiao1, Sean Howard1, Martin Graham3, G. D. Gu4,
tors analogous to TTC5 that dynamically tune Taylor L. Hughes5, Dirk K. Morr3, Vidya Madhavan1*
the abundance of key proteins such as his-
tones (29) and chaperones (30). The methods The possible realization of Majorana fermions as quasiparticle excitations in condensed-matter
and paradigm of cotranslational abundance physics has created much excitement. Most studies have focused on Majorana bound states;
control established here should provide a however, propagating Majorana states with linear dispersion have also been predicted. Here, we
framework for studying this mode of cellular report scanning tunneling spectroscopic measurements of crystalline domain walls (DWs) in
regulation. FeSe0.45Te0.55. We located DWs across which the lattice structure shifts by half a unit cell.
These DWs have a finite, flat density of states inside the superconducting gap, which is a
RE FE RENCES AND N OT ES hallmark of linearly dispersing modes in one dimension. This signature is absent in DWs in the
1. G. Borisy et al., Nat. Rev. Mol. Cell Biol. 17, 322–328 (2016). related superconductor, FeSe, which is not in the topological phase. Our combined data are
2. A. Desai, T. J. Mitchison, Annu. Rev. Cell Dev. Biol. 13, 83–117
(1997).
consistent with the observation of dispersing Majorana states at a p-phase shift DW in
3. G. J. Brouhard, L. M. Rice, Nat. Rev. Mol. Cell Biol. 19, 451–463 a proximitized topological material.
(2018).
M
4. R. A. Walker et al., J. Cell Biol. 107, 1437–1448 (1988).
5. I. Gasic, S. A. Boswell, T. J. Mitchison, PLOS Biol. 17, e3000225
(2019).
ajorana fermions are putative elemen- (10–14), ferromagnetic atomic chains (15–17),
6. D. W. Cleveland, M. A. Lopata, P. Sherline, M. W. Kirschner, tary particles that are their own anti- and topological insulators that are proximity-
Cell 25, 537–546 (1981). particles (1). Emergent analogs of these coupled with s-wave superconductors (18, 19);
7. D. W. Cleveland, Trends Biochem. Sci. 13, 339–343 (1988).
8. D. A. Gay, S. S. Sisodia, D. W. Cleveland, Proc. Natl. Acad. Sci.
fermions have been argued to exist as in all of these cases, the MBS were spectroscop-
U.S.A. 86, 5763–5767 (1989). quasiparticle excitations in condensed- ically identified as zero-energy conductance
9. J. S. Pachter, T. J. Yen, D. W. Cleveland, Cell 51, 283–292 matter systems (2–7) and have attracted much anomalies. In addition to the localized MBS,
(1987).
10. T. J. Yen, P. S. Machlin, D. W. Cleveland, Nature 334, 580–585
attention as possible building blocks of fault- however, theoretical predictions show that dis-
(1988). tolerant quantum computation (8, 9). So far, persing Majorana states may also be realized
11. C. J. Bachurski, N. G. Theodorakis, R. M. Coulson, various predictions and realizations of lo- as quasiparticles in condensed-matter systems
D. W. Cleveland, Mol. Cell. Biol. 14, 4076–4086 (1994).
calized Majorana bound states (MBS) have (18, 20–22). These quasiparticles are of fun-
12. N. G. Theodorakis, D. W. Cleveland, Mol. Cell. Biol. 12, 791–799
(1992). been reported. The platforms include strong damental interest and may be harnessed for
13. M. Gamerdinger et al., Mol. Cell 75, 996–1006.e8 (2019). spin-orbit–coupled semiconductor nanowires quantum computing. Dispersive Majorana

modes have been proposed as edge states As with most iron-based superconductors, probe with which to measure gap values on
of chiral p-wave superconductors, in hybrid the Fermi surface of FeSe0.45Te0.55 is composed different bands with high accuracy. How-
systems that combine superconductors with a of two hole pockets (Fig. 1A, a′ in red and b in ever, because of the inhomogeneity caused by
quantum anomalous Hall insulator (23), or green) around the G-point and two electron doping in Fe(Se,Te), the gap values and thus
two-dimensional (2D) magnetic Fe islands pockets (g in blue) at the Brillouin zone corner the number of gaps seen in any one spectrum
(24, 25). However, these platforms are diffi- (M-point). According to theory, Te substitu- show spatial variations (Fig. 1D). To obtain
cult to fabricate; moreover, most of them are tion into FeSe shifts the bulk pz band [found statistical information on the gap values and
stable only at low temperatures. This makes above the Fermi energy (EF) in FeSe] downward distribution, we recorded tunneling spectra
future applications highly challenging in these toward the Fermi level (29). This band then dI(r, V)/dV on a densely spaced grid (240 by
systems. hybridizes with the dxz band (a band) to create 240) over a 100- by 100-nm field of view (FOV).
Iron-based superconductors provide an alter- a topological band inversion that pushes the The gap values were extracted through our
native pathway for pursuing Majorana modes a band ~14 meV below EF. In the resulting multigap-finding algorithm, which finds the
at higher temperatures. FeSexTe1-x [Fe(Se,Te)] band gap, a topological Dirac surface state position of peak and shoulder features in each
is the simplest compound in the Fe-based su- emerges, centered at the G-point on the (001) dI/dV spectrum and accepts them as coher-
perconductor family, with an optimum critical surface (Fig. 1B). Below Tc, superconducting ence peaks if they are particle-hole symmetric
temperature (Tc) of 14.5 K. This family of mate- gaps are expected to open on both the sur- (37). We classify the results by the number of
rials is highly attractive owing to its versatility face and bulk bands (32). gaps found for each spectrum and show a
and tunability. The materials grow well in thin In Fig. 1C, we show an atomically resolved color-coded 2D map (gap map) in Fig. 2A.
film form (26), and their Tc can be substantially STM topographic image of the chalcogen We found spectra with one, two, or three
enhanced through doping, pressure, and strain (Se/Te) surface layer, where the Te atoms ap- gaps in the energy range (–3.5 meV, 3.5 meV).
(27). Through density functional theory, it has pear brighter than Se because of their more To visualize the evolution of the spectra as a
been found that for a range of concentrations extended electronic orbitals. There is a marked function of position, a spectral line cut tra-
around 50% Se, Fe(Se,Te) possesses helical absence of interstitial Fe atoms on the surface, versing the three regions (white line on the
Dirac surface states owing to band inversion usually observed as bright protrusions in the gap map) is shown in Fig. 2B. One can see the
along the G-Z direction (28–31). In accord- topography (35, 36). This, in combination with spectra evolving from displaying two gaps to
ance with the Fu and Kane model (18), when the sharp superconducting transition (fig. three and then to a single gap. The statistical
an s-wave superconducting gap opens in the S1) (37), confirms the high quality of these analysis of the gap magnitudes divided by
Dirac surface states (because of proximity samples. We measured a series of differential category (colored histograms), as well as the
to s-wave superconductivity in the bulk), it conductance (dI/dV) spectra along the line overall results, are shown in Fig. 2C. For ~20%
provides the ideal conditions for hosting shown in Fig. 1C and present them in Fig. 1D. of the spectra taken in this FOV, we can dis-
MBS. There are multiple pieces of support- The data show that the spectral weight is com- tinguish only one gap centered around 1.4 meV;
ive evidence for this scenario in Fe(Se,Te): pletely suppressed to zero over a finite energy the two-gap spectra cover ~57% of the FOV
High-resolution angle-resolved photoemis- range ~±1 meV, and sharp peaks appear near area, with mean gap values around 1.4 and
sion spectroscopy (ARPES) data reveal helical the gap edge. These observations strongly 2.4 meV; in the remaining area, three gaps can
surface states that exhibit an s-wave gap be- suggest that there is no nodal structure in be detected simultaneously, with mean values
low Tc, and a sharp zero-bias peak has already the gap function of FeSe0.45Te0.55, and the gap of 1.4, 1.9, and 2.4 meV. These data suggest
been observed inside vortex cores (32–34). minima, if anisotropy exists, should be larger that multiple gaps exist at all points of the
s-Wave–proximitized topological surface than 1 meV. sample, but statistical variations in their mag-
states can also host time-reversed pairs of dis- There is an ongoing controversy regarding nitude sometimes prevent us from resolving
persing 1D Majorana states along domain the gap values for each band reported by var- them individually. The largest gap value ex-
walls (DWs) separating regions in which the ious ARPES and optical conductivity studies tracted from ARPES is around 4 to 5 meV (40).
superconducting order parameter is phase- on similar materials (38–41). STM is the ideal This larger gap is only seen as a hump-like
shifted by p (18). These modes possess a lin-
ear dispersion (E = ±vky) with momentum
A C D
parallel to the DW. This linear dispersion in γ
one dimension implies a constant density of
ky β
states (DOS) for energies below the super-
Γ dI/dV (a.u.)
conducting gap—one of the key experimental
signatures of dispersing Majorana states.
In this work, we used scanning tunneling kx M
microscopy (STM) to interrogate crystalline B
Γ
DWs in the proximitized Dirac surface states BC
B EF
Bulk SC M
of FeSe0.45Te0.55 in a search for signatures of
1D dispersing Majorana modes.
8nm
1 B 0
Department of Physics and Frederick Seitz Materials BV TSC High Low -6 -4 -2 0 2 4 6
Research Laboratory, University of Illinois Urbana- ce
rfa Energy (meV)
Champaign, Urbana, IL 61801, USA. 2Department of Γ Su
M
Physics, University of Science and Technology of China,
Hefei, Anhui 230026, China. 3Department of Physics,
Fig. 1. Band structure and superconductivity in FeSe0.45Te0.55. (A) Sketch of bulk Fermi surfaces of
University of Illinois at Chicago, Chicago, IL 60607, USA.
4
Condensed Matter Physics and Materials Science Fe(Se,Te) at momentum kz = 0. (B) Cartoon image showing superconductivity in the bulk and proximitized
Department, Brookhaven National Laboratory, Upton, NY superconductivity in the topological surface state (31). (C) Topographic image in a 25- by 25-nm field
11973, USA. 5Department of Physics and Institute for of view (bias voltage VS = 40 mV, tunneling current It = 100 pA). (D) Scanning tunneling spectroscopy (STS)
Condensed Matter Theory, University of Illinois at
Urbana-Champaign, Urbana, IL 61801, USA. data taken along the line shown in (C) at 0.3 K [VS = 6 mV, It = 300 pA, modulation voltage (Vmod) = 58 mV].
*Corresponding author. Email: vm1@illinois.edu The spectra are vertically offset for clarity.

feature in our spectra and was not picked A 1 C8 Overall

with all the essential ingredients necessary
up in our gap map because of the suppressed for hosting dispersing Majorana modes.
2 4
intensity of the corresponding peaks. How- To distinguish this possibility from other
3 0
ever, this feature can be seen in line cuts (fig. scenarios, we studied other extended defects
1 Gap Area
S2) (37). If we assign this hump-like feature (figs. S9 and S10) that are not expected to
counts (10 )
1
3
(around ±4.5 meV) to be the coherence peaks give rise to a p-phase shift and found that the
40nm
arising from the large superconducting gap on 0 flat DOS signature is absent (37); 1D defects
the g sheet as shown in (40), then, by compar- B 2 Gaps Area without the half-unit-cell shift have the effect
10
4
ison with ARPES data, the mean gap values of of decreasing the gap magnitude (fig. S9),
dI/dV (a. u.)

1.4 and 2.4 meV may be assigned to the smaller 0 whereas step edges, which are strong poten-
gaps on the a′ and b bands, respectively. This 3 Gaps Area tial scatterers, induce resonant bound states
2
suggests that consistent with recent ARPES inside the superconducting gap (fig. S10).
data, the 1.9-meV gap may be assigned to the 0
Furthermore, the half-unit-cell shift DW is
0
topological surface state, indicating the topo- -2 0 2 1 1.5 2 2.5 3 spectroscopically similar to the rest of the
logical nature of these samples (32). Energy (meV) Δ (meV) sample above Tc. As shown in fig. S8, at 14 K
Recent STM measurements in Fe(Se,Te) have the spectra both on and away from the DW
Fig. 2. Statistical analysis of superconducting
reported the existence of zero-bias conductance are almost identical (37). This indicates that
gaps. (A) 100- by 100-nm map depicting the
peaks inside vortex cores and near interstitial the DW does not have a noticeable effect on
distribution of superconducting gaps in FeSe0.45Te0.55.
Fe atoms (33, 34, 36), which have been pro- the local electronic structure above Tc, and its
Blue, orange, and green colors indicate whether a
posed to be signatures of zero-dimensional effects become prominent only below Tc.
single gap, two gaps, or three gaps were found
MBS. We observed similar spectral line shapes One might wonder whether the experimen-
at each pixel, respectively. The gap values at each pixel
inside several vortex cores and near–atomic- tally observed DW modes could also possess a
were obtained through a multipeak-finding algorithm
scale defects (figs. S3 and S4) in our samples, topologically trivial origin, unrelated to exis-
(37). (B) STS spectra at 0.3 K obtained along the white
all of which are consistent with a topologically tence of a topological surface state. On the
line shown in (A), starting (top to bottom) with a
nontrivial surface state (37). Here, we report basis of previous studies, the superconduct-
two-gap region (orange), which transitions to a three-
the existence of 1D dispersing Majorana modes ing order parameter in Fe(Se,Te) is expected
gap region (green) and ends in a single-gap region
near a domain-wall defect. This defect was dis- to be a sign-changing s± state (45, 46). In such
(blue). The spectra are vertically offset for clarity.
covered with atomically resolved topography a state, defects, regardless of their magnetic
(C) Histogram of the gap values in the one-, two-, and
as a 1D feature on the surface represented properties, would induce impurity states inside
three-gap regions. The dark curves show Gaussian
by a bright line (Fig. 3A). A zoomed-in view the superconducting gap. The experimentally
fits to the gap distribution, with mean gap values
(Fig. 3D) reveals that this bright line sepa- observed DW representing a 1D defect could
of 1.4, 1.9, and 2.4 meV.
rates two crystal domains where the lattice therefore lead to the emergence of an impurity
shows a relative phase shift. This shift is re- band inside the superconducting gap even in a
flected as a split in reciprocal-space Bragg the presence of linearly dispersing Majorana topologically trivial phase. To investigate this
peaks of the Fourier transform of the image states at the DW because it would naturally possibility, we used a theoretical model for a
(Fig. 3B and fig. S5) (37). The magnitude of the give rise to a constant dI/dV in one dimension. topologically trivial superconducting state of
split in reciprocal space corresponds to a spa- According to the Fu-Kane model (18), real- Fe(Se,Te) (47) and represented the DW as a
tial scale of 12 nm (half of the FOV of Fig. 3A), izing 1D dispersing Majorana states requires line of potential scatterers (fig. S11) (37). We
which is consistent with the domain size in three ingredients: nontrivial topological sur- found, as expected, that the DW gives rise to
this FOV. To determine the magnitude of the face states, s-wave superconductivity that gaps impurity states inside the superconducting
lattice shift between the two sides of the DW, the surface states, and a p-phase shift in the gap. However, these states do not in general
we carried out a displacement analysis (fig. superconducting order parameter across the traverse the superconducting gap (only for
S6), which maps the relative phase of the DW. Our detailed gap maps already indicate fine-tuned values of the scattering potential
lattice on either side (37). The analysis indicates the presence of proximity-gapped Dirac sur- do impurity states near zero energy emerge).
that the lattice undergoes a half-unit-cell shift face states, thus satisfying the first two criteria. Moreover, such states are not uniformly dis-
across such a DW. This half-unit-cell shift, as This leaves us with the question of how to tributed in energy inside the gap and cannot
we will show later, is essential for the real- generate a superconducting phase shift. For result in the observed constant density of
ization of dispersing Majorana modes. the pairing symmetries allowed in Fe(Se,Te), states. The same conclusion holds if the DW-
Differential conductance spectra obtained it is possible to have an interplay between the separated, p-phase–shifted superconducting
along three distinct paths traversing the DW crystal structure and the phase of the super- regions are present in an otherwise topologi-
(Fig. 3F) reveal an intriguing evolution. As conducting order parameter. One possibility cally trivial phase (fig. S12) (37). These findings
one approaches the DW, the superconduct- comes from the predicted odd parity s-wave are further confirmed by our experimental
ing coherence peaks in the dI/dV spectra are pairing in iron-based superconductors, which study of twin DWs in the topologically trivial
suppressed, and new electronic states emerge encodes pairing between next-nearest-neighbor but related superconducting compound, FeSe
inside the gap, resulting in a nearly featureless, sites (42, 43) in the 2-Fe unit cell. In this case, (fig. S13) (37). Although such DWs give rise
constant finite dI/dV inside the gap at the DW the order parameters on the two Fe-sublattices to a suppression of the superconducting gap,
center as indicated with the highlighted lines have a p-phase difference. A half-unit-cell shift they do not result in a constant DOS. These
in Fig. 3F and the inset (fig. S7) (37). Despite of the lattice in such a system would natural- theoretical and experimental findings taken
the constant DOS inside the superconducting ly create a p-phase shift across the domains together make it unlikely that the observed
gap, the DOS at the DW still exhibits super- (Fig. 3E). It has also been argued that the s± constant DOS near the DW can arise in a topo-
conducting signatures (Fig. 3F, inset, and fig. pairing in iron-based superconductors can logically trivial superconducting phase, which
S8H) , which indicates that the flat DOS is not generate a p-phase shift at the intersections of further emphasizes the important role played
simply caused by a region of normal metal crystal terminations with different orientations by nontrivial topology in creating the observed
(37). One explanation for this observation is (44). Thus, Fe(Se,Te) is an excellent candidate, DW modes.

A B D The topological nature of the DW modes

0.25 necessarily dictates a specific spatial and en-
L1 ergy dependence of the DW states. In par-
ticular, with increasing energy, the Majorana
nm
L2 E DW mode must continuously evolve from being
-0.2 localized at the DW at zero energy to being
delocalized as their energy reaches the gap
C edge. In other words, the modes’ localization
30mV
length increases with increasing energy, lead-
Z (Å)
2 100mV
L3
0 Y ing to an increase in the modes’ spatial extent
Se/Te top Se/Te bottom Fe
from the DW. To visualize this evolution, we
8nm 0 5 10 X
r (nm) obtained spatial differential conductance maps
dI/dV (a. u.)

(dI/dV maps) in the vicinity of the DW ranging
F 8 L1 L2 L3
in energies from EF up to 1.5 mV, where the
-3 0 3 first set of coherent peaks is located (Fig. 4, A
dI/dV (a. u.)
E (meV)
to F). At EF, the in-gap states are confined
4 within an ~3-nm width. These states begin
to expand in real space with increasing en-
ergy and become far less visible at 0.85 mV be-
0 cause of a lack of contrast in the intensity with
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3 respect to the rest of the area. The dI/dV maps
Energy (meV) Energy (meV) Energy (meV) show that the DW states are present at all
energies inside the gap. However, the spatial
Fig. 3. Signature of dispersing 1D Majorana modes at a DW. (A) 25- by 25-nm topographic image
extent of the states grows with increasing en-
showing a DW (bright line) (VS = 4 mV, It = 250 pA). (B) 2D fast Fourier transform of (A), showing a
ergy, which is consistent with a topological
splitting of the Bragg peaks, which indicates the presence of domains in this image. (Inset) Zoom-in near
origin for these modes.
one of the Bragg peaks. (C) Height scans taken at different bias voltages along the yellow dashed line
FeSe0.45Te0.55 may have provided the first
in (A). (D) Zoom-in of the DW. The white and red lines track the atomic lattice on both sides of
glimpse into linearly dispersing 1D Majorana
the DW. A half-unit-cell shift can be observed between one side and the other. (E) Schematic of the
modes. Possible future experiments include the
half-unit-cell shift across the DW. The schematic also depicts how one might obtain a p-phase shift in
measurement of the fractional (4p) Josephson
the superconducting order parameter across such a DW. Superimposed on the lattice are red and
effect using a flux loop or depositing magnetic
green bars, which denote the parity of next-nearest-neighbor pairing (42, 43). As an example, tracking
layers to generate chiral Majorana modes or
the atoms inside the dashed box, one can see that the parity shifts from red on the left of the DW to
localized MBS. Beyond the Majorana fermion
green on the right. This creates a p-phase shift in the superconducting order parameter. (F) Line-cut
context, our experimental results have two
profiles of dI/dV spectra along the three blue lines in (A), which cross the DW (VS = 4 mV, It = 250 pA,
important implications. First, our observa-
Vmod = 58 mV). The spectra shapes obtained right on the DW [at position of dots in (A)] are highlighted
tions provide supporting evidence for the ex-
with a dark blue color. For clarity, the spectra are vertically offset from each other by 0.06 (6 nS). (Inset)
istence of topological surface states and a
A direct comparison of the spectra taken on the DW (orange) and far away (black). All data were
Fu-Kane proximitized superconducting state
obtained at 0.3 K.
in FeSe0.45Te0.55. Second, the connection be-
tween the crystal DW and the superconduct-
ing p-phase shift provides evidence in support
of a superconducting order parameter with a
A B C D real-space sign change within a unit cell (42).
1.2
dI/dV (a.u.)
REFERENCES AND NOTES

1. E. Majorana, Nuovo Cim. 14, 171–184 (1937).
2. N. Read, D. Green, Phys. Rev. B 61, 10267–10297
(2000).
0mV 8nm 0.3mV 0.6mV 0.85mV 0.1 3. A. Kitaev, Ann. Phys. 321, 2–111 (2006).
4. D. A. Ivanov, Phys. Rev. Lett. 86, 268–271 (2001).
E F G 5. J. Alicea, Rep. Prog. Phys. 75, 076501 (2012).
6. C. W. J. Beenakker, Annu. Rev. Condens. Matter Phys. 4,
113–136 (2013).
1.6
dI/dV (a. u.)
7. S. R. Elliott, M. Franz, Rev. Mod. Phys. 87, 137–163 (2015).

0.85mV 8. A. Kitaev, Ann. Phys. 303, 2–30 (2003).
9. C. Nayak, S. H. Simon, A. Stern, M. Freedman, S. Das Sarma,
Rev. Mod. Phys. 80, 1083–1159 (2008).
10. R. M. Lutchyn, J. D. Sau, S. Das Sarma, Phys. Rev. Lett. 105,
0mV 077001 (2010).
0
1.1mV 1.5mV 0.15
11. Y. Oreg, G. Refael, F. von Oppen, Phys. Rev. Lett. 105, 177002
-10 0 10 (2010).
r (nm)
12. V. Mourik et al., Science 336, 1003–1007 (2012).
13. A. Das et al., Nat. Phys. 8, 887–895 (2012).
Fig. 4. Spatial distribution of the 1D Majorana mode at a DW with increasing energy. (A to F) dI/dV 14. S. M. Albrecht et al., Nature 531, 206–209 (2016).
maps from 0 to 1.5 meV at 0.3 K. The maps are 25 by 25 nm in size, and spectra were obtained on a 130- by 15. S. Nadj-Perge et al., Science 346, 602–607 (2014).
130-pixel grid. DW states are present at all energies inside the gap up to 1 meV, when the states merge 16. M. Ruby et al., Phys. Rev. Lett. 115, 197204 (2015).
17. H. Kim et al., Sci. Adv. 4, eaar5251 (2018).
into the coherence peaks. However, the spatial extent of the states grows with increasing energy. (G) DOS 18. L. Fu, C. L. Kane, Phys. Rev. Lett. 100, 096407 (2010).
profiles measured at different energies along the white line perpendicular to the DW. 19. H. H. Sun et al., Phys. Rev. Lett. 116, 257003 (2016).

20. A. P. Schnyder, S. Ryu, A. Furusaki, A. W. W. Ludwig, 38. B. Zeng et al., Nat. Commun. 1, 112 (2010). Foundation under the Materials Research Science and Engineering
Phys. Rev. B 78, 195125 (2008). 39. K. Okazaki et al., Phys. Rev. Lett. 109, 237011 (2012). Center program under NSF award DMR-1720633 (SuperSEED) for
21. X. L. Qi, T. L. Hughes, S. Raghu, S.-C. Zhang, Phys. Rev. Lett. 40. H. Miao et al., Phys. Rev. B 85, 094506 (2012). support. M.G. and D.K.M. acknowledge support from the U.S.
102, 187001 (2009). 41. C. C. Homes et al., J. Phys. Chem. Solids 72, 505–510 (2011). Department of Energy, Office of Science, Basic Energy Sciences,
22. Y. Tanaka, M. Sato, N. Nagaosa, J. Phys. Soc. Jpn. 81, 011013 42. J. Hu, Phys. Rev. X 3, 031004 (2013). under award DE-FG02-05ER46225. Author contributions:
(2012). 43. N. Hao, J. Hu, Phys. Rev. B 89, 045144 (2014). V.M. and Z.W. conceived the experiments; Z.W., J.O.R., and
23. Q. L. He et al., Science 357, 294–299 (2017). 44. R. X. Zhang, W. S. Cole, S. Das Sarma, Phys. Rev. Lett. 122, L.J. performed the experiments; J.O.R. and S.H. wrote software for
24. S. Rachel, E. Mascot, S. Cocklin, M. Vojta, D. K. Morr, 187001 (2019). data analysis; Z.W., J.O.R., and S.H. carried out data analysis;
Phys. Rev. B 96, 205131 (2017). 45. T. Hanaguri, S. Niitaka, K. Kuroki, H. Takagi, Science 328, G.D.G. made the samples; M.G., T.L.H., and D.K.M. carried out the
25. A. Palacio-Morales et al., Sci. Adv. 5, eaav6600 (2019). 474–476 (2010). theory; and V.M. and Z.W. wrote the paper, with contributions from
26. P. Mele, Sci. Technol. Adv. Mater. 13, 054301 (2012). 46. P. J. Hirschfeld, M. M. Korshunov, I. Mazin, Rep. Prog. Phys. 74, all authors. Competing interests: The authors declare no competing
27. K. Horigane, N. Takeshita, C.-H. Lee, H. Hiraka, K. Yamada,
124508 (2011). interests. Data and materials availability: The experimental
J. Phys. Soc. Jpn. 78, 063705 (2010).
47. S. Sarkar et al., Phys. Rev. B 96, 060504 (2018). data and theory code of this study are available at (48).
28. N. Hao, J. Hu, Phys. Rev. X 4, 031053 (2014).
48. Z. Wang et al., Replication data for: Evidence for dispersing 1D
29. Z. Wang et al., Phys. Rev. B 92, 115119 (2015).
30. X. Wu, S. Qin, Y. Liang, H. Fan, J. Hu, Phys. Rev. B 93, 115129 Majorana channels in an iron-based superconductor.
(2016). Zenodo (2019); 10.5281/zenodo.3529111
31. G. Xu, B. Lian, P. Tang, X.-L. Qi, S.-C. Zhang, Phys. Rev. Lett. Materials and Methods
ACKN OWLED GMEN TS
117, 047001 (2016). Supplementary Text
32. P. Zhang et al., Science 360, 182–186 (2018). The authors thank H. Ding, S. Kivelson, and C. Kane for useful
Figs. S1 to S13
33. D. Wang et al., Science 362, 333–335 (2018). conversations. Funding: The work in Brookhaven is supported by
Tables S1 and S2
34. T. Machida et al., Nat. Mater. 18, 811–815 (2019). the Office of Science, U.S. Department of Energy, under contract
References (49, 50)
35. M. Chen et al., Nat. Commun. 9, 970 (2018). DE-SC0012704. V.M. gratefully acknowledges support from
36. J. Yin et al., Nat. Phys. 11, 543–546 (2015). U.S. Department of Energy, Office of Science, Basic Energy 29 January 2019; resubmitted 20 July 2019
37. Materials and methods are available as supplementary Sciences, under award DE-SC0014335 (STM studies) and NSF Accepted 14 November 2019
materials. award DMR-1610143. T.L.H. thanks the U.S. National Science 10.1126/science.aaw8419

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
0103Product.indd 109 12/20/19 11:43 AM

LIFE SCIENCE TECHNOLOGIES Produced by the Science/AAAS Custom Publishing Office
new pro d uc ts

Compact in design, FLIR Systems

A8200sc and A8300sc midwave
.31

- 𰀯𰁒𰁑𰁝𰁄
rate temperature measurements. /*)""'%))#(&

-
!

$ %)# $ %)# *)%##
$ %)# (%# *)&##

"2

54 - 0
5.0++

1, "
$ 5
-
/
&
'

!-,
.0+.0+

0+ !
30+

2-32-3
"
+

54 .0++ 2-3

+
+,"'"" +

2-3
1$"

( %$$$"#!! "
,&('
+ '+*% www.origene.com
"!# &(-
&()% !,

'+* &( *)'

! - -)'

$ -

! &
-

+,
& -
0 ,%#$ +'* )+##

+
www.bd.com )'

- 5 2 /-
-
- / -,$$!+*&!($$$
6

(
"
"

6

!
4425 !

-
!/-

-

"4

!

0 ,+## '%' )*%& "4 023
! ! "

,$$. "4

143
2 .14-
3 0 ) $$$"4425 .1

23 "4

&"%1#

!5 "

/ -((!$!%+*'!&*&*&*

- " "#
!. .14

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.
110 3 JANUARY 2020 • VOL 367 ISSUE 6473 sciencemag.org/custom-publishing SCIENCE
0103Product.indd 110 12/23/19 11:18 AM

online @sciencecareers.org
Bacaner Endowed Chair in THE UNIVERSITY Academic Gastroenterologist –
Physiology OF CHICAGO Hepatology
Biological Sciences Division:
The department of Integrative Biology and
Department of Medicine
Physiology at the University of Minnesota
Medical School is recruiting a tenured faculty Position Description
member for the Bacaner Endowed Chair in Physiology. We seek exceptional The University of Chicago’s Department of Medicine, Section of Gastroenterology,
faculty candidates with an established research program in integrative systems Hepatology and Nutrition, is searching for a full-time faculty member at any
physiology, spanning from gene to whole organism, with specifc interests in rank to join our rapidly growing Center for Liver Diseases. The appointee will
the cardiovascular system, metabolism (diabetes/obesity), muscle physiology, have a mix of outpatient and inpatient clinical responsibilities, including general
stress physiology and related felds. Substantial resources, with state-of-the-art hepatology and liver transplant service and consultation at multiple sites in the
research space, is committed to this effort. Successful candidates will have an Chicago area. Other duties will include teaching and supervision of trainees and
established and innovative research program that embraces biological complexity students, and scholarly activity. We especially welcome applicants with advanced
and physiology from molecular building blocks to the living organism and human training in transplant hepatology. Academic rank and compensation (including a
disease. Applicants must have a strong record of research accomplishments, as generous package of fringe benefts) are dependent upon qualifcations.
documented by publications in leading peer-reviewed journals along with an Prior to the start of employment, quali.ed applicants must: (1) have an medical
outstanding record of extramural funding. Minimum requirements are a PhD, doctorate or equivalent, (2) hold or be eligible for medical licensure in the State
MD or MD/PhD with tenure at an academic research institution. The Department of Illinois, (3) be Board Certi.ed or Eligible in Gastroenterology, and (4) have
is a central part of the newly established Biomedical Discovery District at the expertise in clinical and research (basic or translational) aspects of nonalcoholic
UMN and hosts a state-of-the-art Core facility for cardio-metabolic phenotyping fatty liver disease (NAFLD).
(IBP Phenotyping Core). The Department is an integral component of Institutes/ To be considered, those interested must apply through The University of
Centers including the Institute for Diabetes, Obesity and Metabolism, Lillehei Chicago’s Academic Recruitment job board, which uses Interfolio to accept
Heart Institute, Institute for the Biology of Aging and Metabolism, Masonic applications: http://apply.interfolio.com/72324. Applicants must upload: a
Cancer Center, Center for Immunology, with exceptional on-site research core CV including bibliography and cover letter. Review of applications ends when
facilities. There are also several active and relevant training grant (T32) programs the position is .lled.
that offer support for training students and fellows. Additional information about
the department can be found at https://www.physiology.umn.edu. Equal Employment Opportunity Statement
We seek a diverse pool of applicants who wish to join an academic community
To apply, please upload a letter of interest, curriculum vitae, statement of that places the highest value on rigorous inquiry and encourages diverse
programmatic and research goals in response to job opening ID 334734 at http:// perspectives, experiences, groups of individuals, and ideas to inform and
www1.umn.edu/ohr/employment. Please also arrange to have three (3) letters stimulate intellectual challenge, engagement, and exchange.
of recommendation sent to jotto@umn.edu or Search Committee, Department of
The University of Chicago is an Af.rmative Action/Equal Opportunity/Disabled/
Integrative Biology and Physiology, Medical School, 6-125 Jackson Hall, 321
Veterans Employer and does not discriminate on the basis of race, color, religion,
Church Street S.E., Minneapolis, MN 55455. Review of applications will begin
sex, sexual orientation, gender identity, national or ethnic origin, age, status as
February 3, 2020 and continue until the position is 2lled.
an individual with a disability, protected veteran status, genetic information, or
The University of Minnesota provides equal access to and opportunity in its other protected classes under the law. For additional information please see the
programs, facilities, and employment without regard to race, color, creed, University’s Notice of Nondiscrimination.
religion, national origin, gender, age, marital status, disability, public Job seekers in need of a reasonable accommodation to complete the application
assistance status, veteran status, sexual orientation, gender identity, or gender process should call 773-702-1032 or email equalopportunity@uchicago.edu
expression. To learn more about diversity at the U: http://diversity.umn.edu. with their request.
Where Science The Department of Neurobiology and Anatomy

at Wake Forest Medical School announces open
positions for
Gets Social. Postdoctoral Training in

Multisensory Processes
We seek strong candidates for postdoctoral
training funded by an NIH T32 Training
Grant. The training program provides a rich
AAAS.ORG/COMMUNITY collaborative research environment that fosters
interdisciplinary approaches to understanding
how the brain integrates information from
multiple senses to produce perception and
adaptive behavior. Candidates with direct
experience as well as those in related felds are
encouraged to apply. Trainees will have access to
any of 9 laboratories using human subjects and/
or a variety of animal models rodents-primates
with approaches spanning molecular/cellular to
perceptual/behavioral. Fellowships are awarded
on a competitive basis.
Applications including a current curriculum
vitae or nominations should be sent to the
Training Grant Director: Dr. Barry E. Stein
bestein@wakehealth.edu, or to its Co-Directors:
Dr. Terrence Stanford (stanford@wakehealth.
edu and Dr. Dwayne Godwin dgodwin@
wakehealth.edu. A description of the faculty
and the program can be accessed via the
website: http://graduate.wfu.edu/admissions/
t32/training_tpmp.html. Please also submit your
application and CV to either job opening 38704
or 38705 at www.wakehealth.edu/jobssearch/.
EOE/AA: Minorities/Females/Disabled/Vets.
0103Recruitment_SC.indd 111 12/27/19 9:59 AM

online @sciencecareers.org
Confused about your next

career move?
Download Free Career
Advice Booklets!
ScienceCareers.org/booklets

Whoʼs the Top Employer for 2019?
Science Careersʼ annual survey reveals the top
companies in biotech & pharma voted on by
Science readers.
Read the article and employer profles and listen
to podcasts at sciencecareers.org/topemployers

WORKING LIFE
By Desiree Dickerson
Strange dreams
F
our years into my Ph.D. program, my hands started to go numb each night as I slept. I didn’t
think much of it at the time; it was just uncomfortable and weird. Then one night, while sound
asleep, I somehow mistook my numb arm for a snake. I threw myself out of bed and woke in a
panic—panting, trembling, heart racing—on my bedroom floor, poised to defend myself against
this imaginary snake. A few nights later, my hand was a spider. Once again, I launched myself
out of bed in utter panic. I later discovered that these nighttime events weren’t simply strange
dreams; they were a direct result of grad school stress.
At the time, I was working toward When virtually all your peers are
my Ph.D. and training to be a clini- exhausted, stressed, and working
cal psychologist. I spent half my time overtime, how can you see your own
sitting in a dark room listening to experience as a cause for concern?
the “pop pop” of neurons firing as There were some exceptions—
rats explored mazes and the other peers who had better work-life
half seeing patients, helping every- balance and didn’t appear to be rid-
day people work through anxiety, dled with anxiety—but my friends
depression, stress, and burnout. and I persuaded ourselves that their
Juggling the Ph.D. and the clini- more balanced approach was a fail-
cal program caused the perfection- ing, a lack of wanting. They clearly
ist in me to run riot. There was weren’t on the same career path.
never enough time to immerse When I figured out why my
myself fully in either world. I oscil- hands were going numb, I turned to
lated between striving to be the running to relieve my stress. That
best and, when I couldn’t, strug- helped somewhat and my strange
gling to find motivation to work
at all. The clinical exam loomed
“I lived with a constant dreams went away, but it was a
Band-Aid solution to what should
large in my mind and played into sense of impending doom.” have struck me as a more serious
my worst fear: being exposed problem. A psychologist could have
in front of a panel of experts as the fraud I believed I helped me manage my stressors more effectively and push
was. I lived with a constant sense of impending doom. back on the mounting workload.
During lunch, I’d sit around the table in the department I can’t go back and change my approach to grad school.
staff room with Ph.D. student and postdoc friends, joking But I’m now in a position to show others an alternative
about my nighttime attacks. “You thought your hand was a path. I specialize in helping academics navigate similar
snake?” they’d say. Then we’d laugh, swapping stories about challenges and take a healthier approach to dealing with
the sorry state we were in. At no point did I think I should the pressures of academia. I’ve worked with academics at
see a doctor. It simply made for a funny story. all career stages, and I’ve seen firsthand how many of them
Later that year, while discussing a patient’s symptoms harbor a chorus of inner voices demanding perfection and
with a neurologist, I realized that my anxiety could be the telling them that the sacrifices are necessary for success.
direct cause of the numbness in my hands. The perpetual My message to those I work with is that the stereotype
stress was triggering tension in my shoulders, and that, in of the overworked, stressed-out academic is unhealthy
turn, was putting pressure on my nerves and cutting off the and outdated. Productivity and well-being are not mutu-

circulation to my extremities. ally exclusive. In fact, prioritizing your well-being can
There was a certain irony in that realization. I’d spent improve not only your productivity, but also your motiva-
years learning about psychological tools for dealing with tion, insight, creativity, and enjoyment. So, when your inter-
stress and anxiety. But until then it never occurred to me nal alarm bells send you a warning signal—or when snakes
that I, too, was experiencing an abnormal level of stress and attack in the night—don’t laugh; take action. j
anxiety—or that I might benefit from the tools I was using to
help others. I had never seen my problems as anything other Desiree Dickerson is a neuroscientist and clinical psychologist
than reasonable responses to the pressures of academic life. based in Valencia, Spain.
0103WorkingLife.indd 114 12/23/19 9:10 AM

CALL FO R PA PERS
BioDesign Research is a Science Partner Journal published in afliation with Nanjing Agricultural University (NAU)
and distributed by the American Association for the Advancement of Science (AAAS). BioDesign Research
publishes high quality breakthrough research, reviews, editorials, and perspectives focusing on in silico biosystems
design, genetic or epigenetic modifcations, and genome writing or rewriting in any organism.
Submit your research to BioDesign Research today!

Learn more at spj.sciencemag.org/bdr
The Science Partner Journals (SPJ) program was established by the American Association for the Advancement of Science (AAAS), the
non-proft publisher of the Science family of journals. The SPJ program features high quality, online-only, editorially independent open-
access publications produced in collaboration with international research institutions, foundations, funders and societies. Through these
collaborations, AAAS expands its eforts to communicate science broadly and for the beneft of all people by providing a top-tier international
research organization with the technology, visibility, and publishing expertise that AAAS is uniquely positioned to ofer as the world’s largest
general science membership society.
Learn more at spj.sciencemag.org
@SPJournals @SPJournals
A R T I C L E P R O C E S S I N G C H A R G E S WA I V E D U N T I L 2 0 2 2
0103Product.indd 115 12/20/19 8:48 AM

Custom Antibodies for
Research & Immunotherapy
Antibody Development and Production Services
Fab
Fab
Fc
ProMab Biotechnologies has two decades of experience generating highly

specific and potent monoclonal antibodies
Mouse / Rat Monoclonal Antibody Recombinant Antibody Antibody Sequencing

Rabbit Monoclonal Antibody Bispecific Antibody Stable Cell Line Generation
Human Monoclonal Antibody Antibody Humanization Antibody Production
All products are for research only
Discovermore
Discover more| www.promab.com
| www.ProMab.com
2600 Hilltop Dr, Building B, Suite C320, Richmond, CA 94806

1.866.339.0871 | info@promab.com
0103Product.indd 116 12/26/19 9:13 AM

Science20200103 DL

Uploaded by

Copyright:

Available Formats

Science20200103 DL

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

Science20200103 DL

Uploaded by

Copyright:

Available Formats

What’s in store in 2020?

A sci-f legend’s legacy Earliest cooked root

Apply by 1/24/20 at www.sciencemag.org/noster

0103Product.indd 2 12/20/19 8:48 AM

11 Study pushes emergence of measles

SCIENCE sciencemag.org 3 JANUARY 2020 • VOL 367 ISSUE 6473 3

0103TOC.indd 3 12/27/19 3:48 PM

PHOTO: MICHAEL ROBINSON CHAVEZ/THE WASHINGTON POST VIA GETTY IMAGES

4 3 JANUARY 2020 • VOL 367 ISSUE 6473 sciencemag.org SCIENCE

0103TOC.indd 4 12/27/19 3:48 PM

SCIENCE sciencemag.org 3 JANUARY 2020 • VOL 367 ISSUE 6473 5

0103Editorial.indd 5 12/20/19 5:55 PM

PHOTO: TONY HEALD/MINDEN PICTURES

6 3 JANUARY 2020 • VOL 367 ISSUE 6473 sciencemag.org SCIENCE

0103NewsInBrief.indd 6 12/27/19 11:32 AM

CRISPR’s big clinical tests

SCIENCE sciencemag.org 3 JANUARY 2020 • VOL 367 ISSUE 6473 7

0103NewsInBrief.indd 7 12/27/19 11:32 AM

OCEAN CONSERVATION The United

PHOTO: MATTHEW KAPUST/SANFORD UNDERGROUND RESEARCH FACILITY

STEM CELL FUNDING California voters will

8 3 JANUARY 2020 • VOL 367 ISSUE 6473 sciencemag.org SCIENCE

0103NewsInBrief.indd 8 12/27/19 11:32 AM

Aurorae were seen widely

U.S. military tests radiation belt cleanup in space

SCIENCE sciencemag.org 3 JANUARY 2020 • VOL 367 ISSUE 6473 9

0103NewsInDepth.indd 9 12/27/19 3:50 PM

the Air Force Research Laboratory.

PHOTO: RODRIGO GARRIDO/REUTERS/NEWSCOM

10 3 JANUARY 2020 • VOL 367 ISSUE 6473 sciencemag.org SCIENCE

0103NewsInDepth.indd 10 12/27/19 3:50 PM

SCIENCE sciencemag.org 3 JANUARY 2020 • VOL 367 ISSUE 6473 11

0103NewsInDepth.indd 11 12/27/19 3:50 PM

proteins and other large molecules, includ-

PHOTO: ARIZONA STATE UNIVERSITY

12 3 JANUARY 2020 • VOL 367 ISSUE 6473 sciencemag.org SCIENCE

0103NewsInDepth.indd 12 12/27/19 3:50 PM

Panchanathan, attended IIT Madras in the

to be a lot of uncertainty in the next year or 0

SCIENCE sciencemag.org 3 JANUARY 2020 • VOL 367 ISSUE 6473 13

0103NewsInDepth.indd 13 12/27/19 3:50 PM

14 3 JANUARY 2020 • VOL 367 ISSUE 6473 sciencemag.org SCIENCE

0103NewsInDepth.indd 14 12/27/19 3:50 PM

SCIENCE sciencemag.org 3 JANUARY 2020 • VOL 367 ISSUE 6473 15

0103NewsInDepth.indd 15 12/27/19 3:50 PM

16 3 JANUARY 2020 • VOL 367 ISSUE 6473 sciencemag.org SCIENCE

SCIENCE sciencemag.org 3 JANUARY 2020 • VOL 367 ISSUE 6473 17

18 3 JANUARY 2020 • VOL 367 ISSUE 6473 sciencemag.org SCIENCE

you go on social media you can 90

writer and molecular biologist

SCIENCE sciencemag.org 3 JANUARY 2020 • VOL 367 ISSUE 6473 19

INSIGHTS SCIENCE FICTION

PHOTO: COURTESY OF THE RALPH MCQUARRIE ARCHIVES

This pensive android appeared on the cover

20 3 JANUARY 2020 • VOL 367 ISSUE 6473 sciencemag.org SCIENCE

0103Books.indd 20 12/20/19 5:32 PM

SCIENCE sciencemag.org 3 JANUARY 2020 • VOL 367 ISSUE 6473 21

0103Books.indd 21 12/20/19 5:32 PM

Uncovering the ART of antimalarial resistance